上下文感知推荐

本文提出了一种基于社交网络的上下文感知推荐算法，它系统的结合了上下文信息和社交网络信息来提高推荐质量。该算法首先应用随机决策树算法基于各种上下文因素分割初始评分 ...

情景(context)，也称上下文，是指任何可以用来描述实体情形和特征的信息，其实体 ... 情景感知推荐对于现有推荐系统模型和方法的完善和扩展，以及泛在环境下个性 ...

上下文摘要智能体，对匹配商品进行摘要，提取核心亮点和推荐理由。该模块能够根据用户特征和上下文，生成个性化的商品描述，突出商品与用户兴趣的匹配点。例如，对于关注环保的 ...

陈芸. 基于协同过滤的上下文感知推荐算法的研究[D]: [硕士学位论文]. 武汉: 武汉理工大学, 2014.

上下文信息的捕捉在POI推荐中起着至关重要的作用。对于POI推荐中上下文信息的捕捉主要分为基于RNN、基于LSTM、基于GRU三种捕捉方式。 2.1. 基于RNN的POI推荐方法. Liu ...

会话推荐旨在基于匿名用户的历史行为序列预测下一个最有可能交互的项目。现有的大多数基于超图解决会话推荐模型没有考虑超边同构，忽略了会话内可能存在的多种依赖关系， ...

基于此，本文提出基于方向性偏好的个性化序列推荐模型，通过推荐符合用户偏好方向的物品，综合用户的偏好和需求进行推荐从而提高了推荐的准确性。本文以电影评论数据集为背景， ...

本文提出一种基于共现关系重构的超图增强会话推荐模型(CR-HGNN)，在充分考虑项目间共现关系的基础上，将会话序列建模成全局超图和局部图。该模型使用超图卷积神经网络捕获 ...

[结论]：最终实验表明，基于本文所构建的旅游场景知识图谱设计的LTTE推荐模型，在旅游目的地推荐实验中表现的效果明显优于基准算法TF-IDF。

综上所述，我们提出了一种基于强化学习的页面式推荐系统，该系统不仅能够生成一组多样且互补的推荐项目，还能优化它们在二维页面上的显示策略，从而最大化用户的整体回报。相比 ...

本文采用两个并行的卷积神经网络(CNN)分别处理用户和物品的评论与描述，并通过词级、短语级和句子级的注意力机制来提取不同层次的语义信息，实现评论文本的深度融合，从而 ...

相比传统的推荐方式,AI虚拟助手推荐通过语音或文字输入来与用户进行交互,建立类似人际对话的场景,其不仅考虑了用户的历史行为,还能够在一定程度上理解用户当前的情感和意图 ...

本文基于传统的协同过滤算法，通过融合社交关系和兴趣偏好特征建立用户相似性矩阵，并基于形如指数函数的时间衰减函数对用户兴趣偏好进行动态加权，最后通过改进的矩阵分解 ...

本方法通过在用户的自适应音乐推荐系统中结合情绪情境的推理能力，将面部表情融合到音乐推荐中去，结合用户的社交网络状态建立用户情绪与音乐类型的关联，计算用户在某种情绪 ...

推荐系统是现代在线平台中缓解信息过载、提升用户体验的关键技术[1]，其核心在于对用户与物品之间的交互关系进行建模，从而实现个性化推荐。然而，传统协同过滤方法高度 ...

在推荐系统中设置少量跨品类内容,避免用户陷入信息同质化,挖掘潜在需求.同时结合用户活跃时段推送场景化内容——通勤时段侧重便携商品,晚间推荐家居好物,周末主推休闲娱乐 ...

亚马逊的个性化推荐系统已经渗透到其平台的各个层面，包括用户画像生成、产品推荐、广告推送等方面。生成式AI在这些应用中的引入，使得推荐系统能够更加精准地理解和预测用户 ...

算法推荐是一种基于用户数据和机器学习技术的个性化信息推送系统.该系统通过分析用户行为模式,历史偏好及实时反馈,预测并推荐用户潜在感兴趣的商品,内容或服务.

推荐系统作为一种信息过滤技术，用于提供个性化推荐服务，帮助用户在“信息超载”的互联网中找到感兴趣的物品，提高用户选择决策的质量[1] 。个性化推荐系统作为解决当前大数据 ...

本文旨在基于从抖音平台采集的真实电商短视频数据，提出一种多模态融合的分析框架，综合视频的视觉、音频、文本及上下文模态信息，构建最优时长预测模型。通过对多维度数据 ...

(3) 支持多模态内容协同建模：面对短视频、动图、图文融合等内容形式，CTR模型逐步扩展至多模态建模维度，结合图像卷积网络(CNN)与文本嵌入技术，实现语义理解与场景适配[8]。

本文设计了一种基于多模态特征记忆库的视频语音检索模型，该模型主要分为三个模块，分别为特征提取模块，多模态特征映射融合模块和特征记忆库模块。在特征提取模块中，我们分别 ...

Recommender systems are essential for providing personalized content across various platforms. However, traditional systems often struggle with limited information, known as the cold start problem, and with accurately interpreting a user's comprehensive preferences, referred to as context. The proposed study, CD-LLMCARS (Cross-Domain fine-tuned Large Language Model for Context-Aware Recommender Systems), presents a novel approach to addressing these issues. CD-LLMCARS leverages the substantial capabilities of the Large Language Model (LLM) Llama 2. Fine-tuning Llama 2 with information from multiple domains can enhance the generation of contextually relevant recommendations that align with a user's preferences in areas such as movies, music, books, and CDs. Techniques such as Low-Rank Adaptation (LoRA) and Half Precision Training (FP16) are both effective and resource-efficient, allowing CD-LLMCARS to perform optimally in cold start scenarios. Extensive testing of CD-LLMCARS indicates outstanding accuracy, particularly in challenging scenarios characterized by limited user history data relevant to the cold start problem. CD-LLMCARS offers precise and pertinent recommendations to users, effectively mitigating the limitations of traditional recommender systems.

In the domain of context‐aware recommender systems, understanding and leveraging feature interactions is crucial for enhancing recommendation quality. Feature interactions delve into the complex interdependencies among user characteristics, item attributes, and contextual factors like time and location. Traditional models often struggle to effectively combine these diverse features, potentially leading to suboptimal recommendations. To tackle this issue, we propose enhancing context‐aware recommender systems through deep feature interaction learning. Our model, which combines BiLSTM and Hybrid Attention mechanisms, offers a sophisticated architecture designed to exploit deep feature interactions effectively. This approach ensures that our system captures essential contextual dynamics, thereby improving the effectiveness of the recommendation process. Experimental results across multiple datasets validate the efficacy of our approach, showing significant improvements in key metrics such as and compared to traditional and contemporary models. These achievements underscore our model's ability to deliver nuanced and adaptively tailored recommendations, marking a valuable contribution to the field of recommender systems.

The ubiquity of the Internet of Things (IoT) has unlocked the potential for recommender systems to become truly context-aware, moving beyond historical preferences to understand a user's immediate, dynamic situation. However, a primary challenge remains: dynamic context modeling—specifically, the acquisition, representation, and fusion of multimodal data streams generated by IoT sensors. While direct research into multimodal data fusion for IoT-based Context-Aware Recommender Systems CARS is still nascent, a wealth of established techniques exists in adjacent “smart systems” domains such as healthcare, activity monitoring, and smart manufacturing. This review article bridges this critical gap. We first establish a taxonomy of multimodal context relevant to recommendations, sourcing data from environmental, physiological, and kinematic IoT sensors. We then review data fusion architectures—from early and late fusion to modern hybrid approaches such as attention mechanisms and graph-based models—drawing on concrete examples from established literature in smart health, activity recognition, and industrial IoT. Finally, we synthesize these findings to propose a structured roadmap for Multimodal IoTbased CARS. This work culminates in a conceptual framework that maps established fusion techniques to recommendation tasks and outlines a research agenda to address key future challenges, including privacy and real-time performance.

While the benefits of product recommender systems (RS) are prominent, due to the complexity of recommendation algorithms and data models, it is difficult for businesses to deploy such systems on their e-stores. Related works that tackle RS development complexity do not entirely abstract the technical details from developers; thus, margins for improvement exist. Moreover, these works do not offer solutions that eliminate the need to write code by developers. These are the motivations for the Ubiquitous Context-Aware Recommender Systems (UbiCARS) Framework proposed in the current work. UbiCARS utilize user feedback acquisition techniques from both e-stores and physical stores to offer recommendations. The framework aims to reduce development complexity, abstract technical details and expedite the development of UbiCARS (facilitating both e-stores and physical stores) by non-RS experts. This is achieved through a Model-Driven Development methodology, that uses a model-based configuration process where models of recommender systems drive the dynamic configuration of UbiCARS on e-stores. The framework was evaluated with developers and experts via the survey method. The evaluation results show the framework’s potential.

Contextual situations, such as having dinner at a restaurant on Friday with the spouse, became a useful mechanism to represent context in context-aware recommender systems (CARS). Prior research has shown important advantages of using latent embedding representation approaches to model contextual information in the Euclidean space leading to better recommendations. However, these traditional approaches have major challenges with the construction of proper embeddings of hierarchical structures of contextual information, as well as with interpretations of the obtained representations. To address these problems, we propose the HyperCARS method that models hierarchical contextual situations in the latent hyperbolic space. HyperCARS combines hyperbolic embeddings with hierarchical clustering to construct contextual situations, which allows loose coupling of the contextual modeling component with recommendation algorithms and, therefore, provides flexibility to use a broad range of previously developed recommendation algorithms. We demonstrate empirically that HyperCARS better captures and interprets hierarchical contextual representations, leading to better context-aware recommendations. Because hyperbolic embeddings can also be used in many other applications besides CARS, we also propose the latent embeddings representation framework that systematically classifies prior work on embeddings and identifies novel research streams for hyperbolic embeddings across information systems applications.

This paper proposes an enhanced framework for user-based context-aware collaborative filtering (CACF) by integrating energy distance as a robust metric to capture distributional discrepancies in user- item interactions, alongside a post-filtering mechanism that adjusts recommendations based on contextual features. Unlike traditional CACF methods that often struggle to incorporate context effectively, the proposed approach models user preferences with greater fidelity by leveraging both statistical distance and contextual relevance. Experiments conducted on three benchmark datasets-MovieLens, Amazon, and Yelp-demonstrate that our model consistently achieves superior performance in terms of prediction accuracy and adaptability to context, outperforming both standard CACF and Energy-Based Models (EBM). These results underscore the effectiveness of combining distribution-aware similarity with postfiltering contextual refinement, offering a promising direction for building more accurate, flexible, and context-sensitive recommender systems in real-world applications.

No abstract available

Recommendation systems have become a ubiquitous presence across various domains such as e-commerce, entertainment (movies and music), tourism, news, advertising, stock markets and social networks. The integration of Ontology and Large Language Models (LLMs) presents a transformative approach for enhancing recommender systems by improving contextual intelligence, personalization, and explainability. Traditional recommendation algorithms often suffer from data sparsity, cold-start problems, and limited semantic understanding, which hinder their ability to deliver highly relevant and interpretable suggestions. This research provides an exhaustive exploration of how ontology-driven knowledge representation can be seamlessly combined with LLMs to refine user-item interactions, enrich embeddings, and improve reasoning in recommendation pipelines. We propose a hybrid framework where ontological structures encode domain-specific relationships while LLMs process, infer, and enhance recommendations through contextual embeddings.This synergy enables semantic-aware recommendations, explainable item suggestions, and adaptive learning mechanisms that adjust based on user behavior and external context. Furthermore, we investigate novel ontology injection techniques, including knowledge graph-based reinforcement, to improve long-tail item exposure and enhance recommendation diversity. Through rigorous evaluation on real-world datasets, we demonstrated how ontology-LLM fusion significantly outperforms baseline models in accuracy, novelty, and interpretability. This study establishes a foundation for the next generation of context-aware, knowledge-enhanced recommender systems that leverage LLMs for reasoning and ontologies for structured knowledge representation, paving the way for more intelligent and user-centric recommendations.

Contextual information has been widely recognized as an important modeling dimension in social sciences and in computing. In particular, the role of context has been recognized in enhancing recommendation results and retrieval performance. While a substantial amount of existing research has focused on context-aware recommender systems (CARS), many interesting problems remain under-explored. The CARS 2024 workshop provides a venue for presenting and discussing the important features of the next generation of CARS and application domains that may require the use of novel types of contextual information and cope with their dynamic properties in group recommendations and in online environments.

Recommender systems are extensively employed across various domains to mitigate information overload by providing personalized content. Despite their widespread use in sectors such as streaming, e-commerce, and social networks, utilizing them for personal information assistance is a comparatively novel application. This emerging application aims to develop intelligent systems capable of proactively providing knowledge workers with the most relevant information based on their context to enhance productivity. In this paper, we explore this innovative application by first defining the scope of our study, outlining the key objectives, and introducing the main challenges. We then present our current results and progress, including a comprehensive literature review, the proposal of a framework, the collection of a pioneering dataset, and the establishment of a benchmark for evaluating a recommendation scenario on our published dataset. We also discuss our ongoing efforts and future research directions.

Recommender systems have become a cornerstone of personalized user experiences across various domains, from e-commerce to entertainment. While traditional models focus on user-item interactions, they often overlook critical contextual information that could significantly enhance prediction accuracy. In this paper, we explore the concept of context-aware recommender systems, which integrate temporal, geospatial, and social context to improve recommendation quality. Temporal context considers factors such as time of day, seasonality, or the user's activity history, while geospatial context leverages the user's location or proximity to items, enhancing the relevance of recommendations. Additionally, social context incorporates user interactions with social networks or communities, enriching the understanding of preferences through peer influence and shared interests. We examine various techniques for incorporating these contexts into recommendation algorithms, including hybrid models, deep learning approaches, and context-sensitive matrix factorization. Furthermore, we address the challenges in balancing the complexity of these models with the need for real-time recommendations and scalability. Finally, we present empirical evaluations on real-world datasets, demonstrating that context-aware models significantly outperform traditional recommender systems in terms of prediction accuracy, diversity, and user satisfaction. This paper aims to provide a comprehensive framework for developing context-aware recommender systems and outline key areas for future research, such as integrating emerging contextual dimensions like sentiment and emotional state.

Contextual information is a prerequisite for timely offering of personalized decision support and recommendation. Yet, research on context-aware recommender systems (CARS) does not appear to be thriving, and finding public datasets containing context factors is a challenging task. We can make various assumptions about why this drop in research interest happened – be it ethical considerations or the popularity of opaque deep learning models that merely consider context in an implicit way. This is an unwelcome development. We argue that continued effort must be put on the creation of suitable datasets. Furthermore, we see significant opportunities in the development of next-generation CARS in the space of interactive AI assistants powered by Large Language Models.

No abstract available

Privacy protection is one of the key concerns of users in recommender system-based consumer markets. Popular recommendation frameworks such as collaborative filtering (CF) suffer from several privacy issues. Federated learning has emerged as an optimistic approach for collaborative and privacy-preserved learning. Users in a federated learning environment train a local model on a self-maintained item log and collaboratively train a global model by exchanging model parameters instead of personalized preferences. In this research, we proposed a federated learning-based privacy-preserving CF model for context-aware recommender systems that work with a user-defined collaboration protocol to ensure users’ privacy. Instead of crawling users’ personal information into a central server, the whole data are divided into two disjoint parts, i.e. user data and sharable item information. The inbuilt power of federated architecture ensures the users’ privacy concerns while providing considerably accurate recommendations. We evaluated the performance of the proposed algorithm with two publicly available datasets through both the prediction and ranking perspectives. Despite the federated cost and lack of open collaboration, the overall performance achieved through the proposed technique is comparable with popular recommendation models and satisfactory while providing significant privacy guarantees.

No abstract available

No abstract available

No abstract available

Conversational recommender systems enable natural language conversations and thus lead to a more engaging and effective recommendation scenario. As the conversations for recommender systems usually contain limited contextual information, many existing conversational recommender systems incorporate external sources to enrich the contextual information. However, how to combine different types of contextual information is still a challenge. In this paper, we propose a multi-type context-aware conversational recommender system, called MCCRS, effectively fusing multi-type contextual information via mixture-of-experts to improve conversational recommender systems. MCCRS incorporates both structured information and unstructured information, including the structured knowledge graph, unstructured conversation history, and unstructured item reviews. It consists of several experts, with each expert specialized in a particular domain (i.e., one specific contextual information). Multiple experts are then coordinated by a ChairBot to generate the final results. Our proposed MCCRS model takes advantage of different contextual information and the specialization of different experts followed by a ChairBot breaks the model bottleneck on a single contextual information. Experimental results demonstrate that our proposed MCCRS method achieves significantly higher performance compared to existing baselines.

Abstract Massive Open Online Courses (MOOCs) have witnessed a fast emergence in recent years thanks to their open and massive nature. A huge number of learners with different profiles, knowledge and learning objectives follow the same online courses with different personalized experiences. With this rapid development, course recommendation based on traits and characteristics becomes a necessity during the learning process to determine potentially suitable learning objects while analyzing user’s behaviours. Context-Aware Recommender Systems (CARS) have been shown to be effective in recommending items according to user’s interests and affect. In this paper, we propose a novel approach for Context-Aware Recommender Systems based on Factorization Machines (FMs). We propose a new algorithm called Unsupervised Graph Predictor Factorization Machines (UGPFMs) that models feature interactions during the recommendation process. UGPFMs apply Convolutional Neural Networks (CNNs) in the graph predictor to model interactions between context features in the rating matrice, then it employs the Factorization Machine (FM) to select the suitable items for the recommendation. UGPFM is a generic model that extends FM to improve the accuracy of recommendation, it outperforms state-of-the-art feature interactions-based FM techniques using MSE, RMSE, MAE and R-Squared metrics. We applied UGPFMs on MOOC while considering the sentiment, the cognition and the confusion as contexts for the recommendation. We compared different context combinations’ impact on the recommendation accuracy, then we computed each context-to-context feature interactions to better understand each context efficiency on e-learning recommendation.

Recommender systems assist users by providing recommendations based on some filtration criteria to reduce information overload. Embedding context-awareness allows recommender systems to use context information around the user, situation, and system to adapt and provide more efficient, relevant, and personalized recommendations. However, embedding context-awareness into recommender systems inherently limits the users‘ control over the systems due to reduced interactivity from automatic adaptations. This may potentially impact users’ use and perception of the systems. Control can be purposefully designed to be given to the user in context-aware recommender systems at different levels. Our work investigates the effects of different levels of user control on the effectiveness and understandability of context-aware recommender systems (CARS) within the scenario of learning through web-based search (called ‘Search-As-Learning’). To enable our study, we implemented a CARS that supports web-based search by recommending users a link using context such as browsing history. Our study found that participants used more recommendations from the CARS with high control compared to no control and some control. In conclusion, higher control in a recommender system for web-based search is preferred by the user despite control manipulation taking more time possibly due to explicit user needs.

Intelligent decision-making in dynamic recommender systems is crucial for capturing temporal user preferences and optimizing long-term user satisfaction. Traditional recommender systems often rely on static modeling, neglecting the temporal dynamics of user-item interactions. To address this limitation, we propose a novel framework, Temporal Interpretability Graph Neural Network with Reinforcement Learning (TIGNN-RL), which integrates dynamic graph neural networks (DGNNs) and Proximal Policy Optimization (PPO) to optimize personalized recommendations. Specifically, our method models user-item interactions as dynamic graphs and utilizes temporal interpretability modules to encode both temporal features and node-specific static features. The temporal interpretability module assigns time-aware and interactions weights to user-item, enabling more time-sensitive and explainable dynamic embeddings. This TIGNN dynamic graph sequential embedding is processed by some LSTM modules to be used as the state of the deep reinforcement learning agent and states. We take a joint approach to training, earn graph embeddings that enable better PPO policy. To evaluate the proposed framework, we conduct experiments on three benchmark datasets: Last.fm 1K, MovieLens 1M, and Amazon Product Review. Results show that TIGNN-RL outperforms state-of-the-art baselines, which use GNNs for augmenting DRL-based RS, in terms of accuracy (NDCG@K) and diversity (ILD@K@K), demonstrating its effectiveness in dynamic and interpretable recommendation scenarios. In this research, some biomechanics knowledge is integrated to further enhance the understanding and application of the proposed framework in scenarios where user behavior is influenced by physical factors.

Context-aware recommender systems (CARS) are specially designed to consider contextual conditions in which users interact with items, aiming to generate enhanced recommendations. A well-recognized challenge when constructing recommender systems is how to resolve sparse data problem. Although the utilization of contextual information offers detailed signals for the recommendation process, it further exacerbates data sparsity and introduces increased computational complexity. In this paper, we present a method for making context-aware recommendations, which is less sensitive to data sparseness and allows full integration of context information. In which, context integration is done by splitting users based on contextual conditions and the sparse data problem is solved through a transfer learning-based method. We experimentally evaluated the proposed method on some contextually-tagged data sets. The results show that our method outperforms several baselines and state-of-the-art context-aware recommendation methods.

No abstract available

No abstract available

As an essential task for recommender systems, the rating prediction problem over several contexts has attracted more attention over the recent years. The traditional approaches ignore the contexts and thus fail to predict the ratings for the unseen data in the rating tensor for varying contextual scenarios. Matrix factorization is preferred over decomposing the rating tensor for avoiding the burden of very high computational complexity while learning the interaction of users’ and items’ latent features. In this work, we propose a novel kernel loss function for optimizing the objective function of matrix factorization in a non-linearly projected rating space under multiple contexts in an optimum manner and also incorporate the implicit feedback of items in the learning process. Further, the optimization is regularized by applying different weights for each regularization term depending on the users’ and items’ participation. Extensive experimental evaluation on five benchmark context-aware datasets indicates the superiority of the proposed work for capturing the non-linearity and predicting the ratings of unseen items for users under varying contexts over the existing and baseline methods. The proposed kernel loss function is also shown to be resistant against shilling attacks in the recommender system. A detailed ablation study demonstrates the validity of the proposed work and the results are shown to be statistically significantly better with RMSE improvement in the range of 3% to 11% over the baseline methods.

No abstract available

Context-Aware Recommender Systems (CARS) refer to recommender systems that can incorporate side information regarding to users, items and ratings. In the present study, we are concerned with CARS, where the side information is provided in the form of item-attribute matrix with entries indicating whether an item has an attribute. We propose to multiply this matrix with user-item rating matrix to represent the the users in the attribute space of the items. We then apply a popular metric learning method, specifically Mahalanobis Metric Learning (MMC), in the attribute space to calculate the distances between the users and their favorite items as less as possible. We recommend the n items that are closest to the users based on these calculations. We verify the effectiveness of the proposed method on two famous MovieLens datasets that differ in size showing that using metric learning increases the success of CARS up to 7% in comparison with using the traditional cosine similarity.

Personalization has remained a very crucial aspect of present-day business strategies as corporations begin to increasingly use the data and machine learning to personalize experiences towards individual users. This paper explores designing a food recommendation system catering to each user's personalized preferences. The system assimilates various inputs gathered from different users, incorporating dietary restrictions, time, and prevailing weather conditions, while facilitating context-aware recommendations coherent with the meal choices adopted by the users. These two methodologies are core to this system: content-based filtering and collaborative filtering. The content-based filtering methodology suggests items based on the characteristics of food items, which might be similar to previously enjoyed dishes by the user. The algorithm suggests meals based on the specific features of the user's taste profile, thus ensuring that suggestions are highly relevant to his established preferences. In contrast, collaborative filtering is an even broader approach to analyze the patterns and selection habits of other users within the system. It identifies individuals who have the same taste profile and uses their selections to make recommendations for food. Based on the group of collective preferences among all the users, collaborative filtering will expose the new user to dishes they might never have discovered but will most probably enjoy since they are derived from common tastes. To make the personalization of food recommendations even more specific, the system integrates k-nearest neighbors, commonly known as KNN. It measures the similarity between the users based on their previous interactions, ratings for their meals, and preferences and selects the “k” users whose tastes are the closest to those of the current user, thus making a cluster of like-minded people. It checks out the favorite food items of those neighboring users to produce suggestions that resonate with this user's palate. In this way, the technique adds another layer of precision to the suggestions so that it becomes more aligned with the specific tastes of the user. This technique combined helps deliver an extremely personalized food recommendation experience. With content-based filtering's focus on the characteristics of individual dishes, insights into user behavior by collaborative filtering, and precise similarity analysis results through KNN, the recommendation produced will be more specific to a person's preference while at the same time adjusted to time and weather, making this holistic approach improve the experience of the users as it enhances their interaction and satisfaction levels because the users get the most appropriate meal recommendations to their taste and circumstance.

Contextual bandits serve as a fundamental algorithmic framework for optimizing recommendation decisions online. Though extensive attention has been paid to tailoring contextual bandits for recommendation applications, the"herding effects"in user feedback have been ignored. These herding effects bias user feedback toward historical ratings, breaking down the assumption of unbiased feedback inherent in contextual bandits. This paper develops a novel variant of the contextual bandit that is tailored to address the feedback bias caused by the herding effects. A user feedback model is formulated to capture this feedback bias. We design the TS-Conf (Thompson Sampling under Conformity) algorithm, which employs posterior sampling to balance the exploration and exploitation tradeoff. We prove an upper bound for the regret of the algorithm, revealing the impact of herding effects on learning speed. Extensive experiments on datasets demonstrate that TS-Conf outperforms four benchmark algorithms. Analysis reveals that TS-Conf effectively mitigates the negative impact of herding effects, resulting in faster learning and improved recommendation accuracy.

Contextual multi-armed bandit (CMAB) algorithms have become a cornerstone of modern recommendation systems owing to their ability to effectively manage the exploration-exploitation trade-off in dynamic and uncertain environments. In this study, we conducted a comprehensive empirical comparison of three advanced CMAB algorithms: Double Explore-Then-Commit (DETC), Feel-Good Thompson Sampling (FG-TS), and Multinomial Logit Upper Confidence Bound (MNL-UCB). Leveraging the MovieLens 1M dataset, we constructed a realistic experimental setting by encoding detailed user profile features, including demographic and behavioral attributes, and generating low-dimensional movie embeddings using truncated singular value decomposition (SVD). We employed a logistic regression framework that captures the probabilistic nature of user preferences. The algorithms behaved markedly differently under long-term (10,000 rounds) and short-term (200 rounds) recommendation scenarios. It indicates that MNL-UCB achieves the lowest cumulative regret and shows strong performance stability across varying contexts in the long-term experiment. And FG-TS demonstrates robust adaptability in highly dynamic environments, making it particularly effective in scenarios with unpredictable behavior. However, DETC tends to underperform in complex contextual settings because of its lack of adaptability, leading to increased regret and fluctuating performance.

The rapid advancement of web technologies has revolutionized English language learning, prompting the design of an innovative English Course Learning System (ECLS). This study addresses key challenges such as lack of personalized learning, low engagement, and inefficient progress tracking in traditional systems. To overcome these, we propose a hybrid recommendation algorithm integrating Collaborative Filtering (CF), Natural Language Processing (NLP), and a BERT-based deep learning model for adaptive content generation and proficiency assessment. The system leverages the Duolingo English Test dataset alongside fluent-speech-corpus to ensure diverse linguistic inputs. The BERT model enhances contextual understanding, enabling dynamic adjustments to learning paths based on real-time student interactions. Novelty emerges through BERT-powered semantic analysis, personalized feedback generation, and automated difficulty scaling, achieving $\sim 95.3 \%$ accuracy in proficiency prediction and $\sim 20 \%$ higher engagement compared to rule-based systems. Performance is evaluated using precision, recall, F1-score, demonstrating significant improvements in retention rates and satisfaction scores. This research bridges AI and language education, offering a scalable, transformer-driven framework for future e-learning systems.

The recommendation method of integrating com-ment text attempts to use user comments as auxiliary data sources to recommend to users. This article proposes a deep learning model Context2Rec to improve the interactivity between products and meet the needs of user multi feature learning. This model utilizes comment text to discover important word order and contextual information of words in the text, and improves semantic representation through pre-training the model. It inno-vatively combines comment text features and sequence features to enhance the performance of recommendation systems. The model effectively captures dynamic changes in user preferences by using BERT for semantic understanding and combining item2vec for association feature learning. Tested on different datasets, the experimental results show that compared with other algorithms, this model effectively improves the recommendation performance, providing a novel method for personalized recommendation services.

Recommender systems in the tourism domain are gaining increasing attention, yet the development of diverse recommendation tasks remains limited, largely due to the scarcity of public datasets. This paper introduces Context Trails, a novel dataset addressing this gap. Context Trails distinguishes itself by including not only user interactions with touristic venues, but also the itineraries (trails or routes) followed by users. Furthermore, it enriches existing item features (e.g., category, coordinates) with contextual attributes related to the interaction moment (e.g., weather) and the venue itself (e.g., opening hours). Beyond a detailed description of the dataset’s characteristics, we evaluate the performance of several baseline algorithms across three distinct recommendation tasks: classical recommendation, route recommendation, and contextual recommendation. We believe this dataset will foster further research and development of advanced recommender systems within the tourism domain. Dataset is available at https://zenodo.org/records/15855966; further code available at https://github.com/pablosanchezp/ContextTrailsExperiments.

In the context of smart tourism, recommender systems play a pivotal role in enhancing the personalization and quality of travel experiences. Tourists often face challenges in decision-making due to information overload. While context-aware recommender systems provide promising solutions by utilizing dynamic contextual data such as time, weather, and location, they struggle to adapt to real-time changes and to balance multiple objectives effectively. To address these challenges, this paper introduces a novel multi-objective contextual multi-armed bandit (MOC-MAB)-based recommender system. This approach integrates the strengths of contextual bandit algorithms with multi-objective optimization to provide personalized recommendations while simultaneously considering relevance and fairness. The proposed system dynamically learns from user feedback to optimize multi-objective recommendations. Extensive experiments conducted on a designed dataset simulating real-world scenarios and the TripAdvisor dataset demonstrate the approach’s superior performance in terms of cumulative reward, click-through rate, and regret minimization when compared to baseline methods. This study also illustrates its practical application in the smart tourism context of Marrakesh, showcasing its potential to enhance tourism experiences in smart cities.

In Yelp merchant recommendation, a semantic-enhanced Contextual Multi-armed slot machine model (SE-Contextual MAB) was proposed to solve the balance problem of “exploring unknown merchants” and “exploiting known high-quality merchants”, as well as the shortcomings of traditional recommendation algorithms that ignore the fine-grained semantic preferences of reviews and the coarse-grained contextual features. This model firstly extracted from Yelp data set user - merchant context features (such as time, location, merchants property) and comment on semantic characteristics (such as theme emotion, user preferences keywords), realize double dimension through attention weighted feature fusion; Secondly, based on the Linear Upper Confidence Bound (LinUCB) framework, a dynamic exploration strategy was designed to increase the exploration weight of new merchants to alleviate the cold start problem. Finally, a semantic consistency penalty was introduced to optimize the reward function to improve the recommendation matching accuracy. The experiment is based on Yelp Academic Dataset, and the results show that: With random recommendations, collaborative filtering (CF), the traditional LinUCB compared to the baseline models such as the accumulation of SE - Contextual MAB reduced by $23.6 \% \sim 41.2 \%$, the new merchant click-through rates rise by 35.8%, the average user semantic matching degree is 0.72. The effectiveness of the model in exploration-exploitation balance and personalized recommendation is verified.

With the advent of the information explosion era, personalized news recommendation faces critical challenges including cold start problems, real-time changes in user preferences, and information filter bubbles. Traditional collaborative filtering methods rely heavily on historical data and struggle to adapt to the rapid update characteristics of news content. This paper proposes a news recommendation solution based on Multi-Armed Bandit (MAB) algorithms, addressing these challenges by balancing exploration and exploitation. The study implements four core algorithms: -greedy algorithm balances exploration and exploitation through probability mechanisms; Upper Confidence Bound (UCB) algorithm employs optimistic estimation using confidence upper bounds; Thompson sampling adopts probability adaptation based on Bayesian framework; and Contextual Linear Bandit (LinUCB) integrates user and news features for personalized recommendations. Experiments Youdaoplaceholder0 on the MIND large-scale news dataset (containing 160,000 news articles, 1 million users) and 15 million click interactions) demonstrate that contextual bandit algorithms outperform traditional methods in click-through rate, dwell time, and recommendation diversity. Thompson sampling shows outstanding performance in click-through rates, while LinUCB excels in convergence speed and recommendation diversity. The experiments confirm that MAB algorithms can effectively adapt to dynamic changes in user preferences, providing a viable solution for real-time news recommendation systems.

The COVID-19 pandemic has intensified the urgency for effective and accessible mental health interventions in people’s daily lives. Mobile Health (mHealth) solutions, such as AI Chatbots and Mindfulness Apps, have gained traction as they expand beyond traditional clinical settings to support daily life. However, the effectiveness of current mHealth solutions is impeded by the lack of context-awareness, personalization, and modularity to foster their reusability. This paper introduces CAREForMe, a contextual multi-armed bandit (CMAB) recommendation framework for mental health. Designed with context-awareness, personalization, and modularity at its core, CAREForMe harnesses mobile sensing and integrates online learning algorithms with user clustering capability to deliver timely, personalized recommendations. With its modular design, CAREForMe serves as both a customizable recommendation framework to guide future research, and a collaborative platform to facilitate interdisciplinary contributions in mHealth research. We showcase CAREForMe’s versatility through its implementation across various platforms (e.g., Discord, Telegram) and its customization to diverse recommendation features.

The growing volume of bibliographic metadata, such as books, and journal articles, has made it increasingly difficult for users to locate relevant resources efficiently. Traditional search engines often return an overwhelming number of results, many of which are irrelevant, leading to frustration and wasted time. Ontologies help structure and organize complex bibliographic data by defining relationships between concepts, while contextual search algorithms enhance the system’s ability to understand user intent. This paper presents the implementation of an ontology-based contextual search and recommendation system within the GujCat portal, specifically focusing on the search functionality. Our system uses ontology mapping for linking knowledge frameworks for consistency and SPARQL Protocol and RDF Query Language (SPARQL) to enhance search capabilities. This solution helps users easily locate specific resources and discover related materials that might otherwise be overlooked. This work underscores the potential of ontology-driven frameworks in academic and research environments, offering a robust solution for metadata management and resource accessibility.

We provide an overview of the approach used as team FeatureSalad for the ACM RecSys Challenge 2024, organized by Ekstra Bladet. The competition addressed the problem of News Recommendation, where the goal is to predict which article a user will click on given the list of articles that are shown to them. Our solution is based on a stacking ensemble of consolidated algorithms, such as gradient boosting for decision trees and neural networks. It relies on numerous features, which model the interest of a user and the lifecycle of an article. The proposed solution allowed our team to rank first among the academic teams, and sixth overall.

Compared with the traditional knowledge graph-enhanced recommendation method, this paper introduces a multi-task learning module to alternately train knowledge graphs and recommendations to alleviate the data sparsity and cold start problems in traditional recommendation methods. Specifically, in the multi-task learning module, the item features and contextual content features are taken, and the features after feature interaction are obtained using the interactive attention network, as a way to learn finer-grained features, and then the gating mechanism processes the item features and entity features that fuse the contextual content, which can filter the unimportant features and obtain the important potential features, and can capture the implicit higher-order feature interaction more effectively. Optimized for multi-task learning tasks. The validity of our model was verified on three publicly available datasets.

The Multi-Armed Bandit (MAB) algorithm holds significant prominence as a recommendation system technique, effectively presenting user-centric content preferences based on the analysis of collected data. However, the application of the basic MAB algorithm in real-world recommendation systems is not without challenges, including issues related to data volume and data processing accuracy. Therefore, the optimization algorithm based on the MAB algorithm is more widely used in the recommendation system. This paper briefly introduces the multi-armed bandit algorithm, that is, the use of MAB in the recommendation system and the problems of the basic MAB algorithm. Aiming at the problems of the basic MAB algorithm, it introduces the MAB-based optimization algorithm used in the recommendation system. At the same time, this paper also analyzes and summarizes such algorithms. This paper introduces two different MAB-based optimization algorithms, namely The Details of Dynamic clustering based contextual combinatorial multi-armed bandit (DC3MAB) and Binary Upper Confidence Bound (BiUCB). In addition, this paper also introduces the application of algorithm in recommended system. Finally, this paper summarizes the introduced algorithms and proposes the future prospects for MAB optimization algorithms.

In the dynamic landscape of online businesses, recommender systems are pivotal in enhancing user experiences. While traditional approaches have relied on static supervised learning, the quest for adaptive, user-centric recommendations has led to the emergence of the formulation of contextual bandits. This tutorial investigates the contextual bandits as a powerful framework for personalized recommendations. We delve into the challenges, advanced algorithms and theories, collaborative strategies, and open challenges and future prospects within this field. Different from existing related tutorials, (1) we focus on the exploration perspective of contextual bandits to alleviate the "Matthew Effect'' in the recommender systems, i.e., the rich get richer and the poor get poorer, concerning the popularity of items; (2) in addition to the conventional linear contextual bandits, we will also dedicated to neural contextual bandits which have emerged as an important branch in recent years, to investigate how neural networks benefit contextual bandits for personalized recommendation both empirically and theoretically; (3) we will cover the latest topic, collaborative neural contextual bandits, to incorporate both user heterogeneity and user correlations customized for recommender system; (4) we will provide and discuss the new emerging challenges and open questions for neural contextual bandits with applications in the personalized recommendation, especially for large neural models. Compared with other greedy personalized recommendation approaches, Contextual Bandits techniques provide distinct ways of modeling user preferences. We believe this tutorial can benefit researchers and practitioners by appreciating the power of exploration and the performance guarantee brought by neural contextual bandits, as well as rethinking the challenges caused by the increasing complexity of neural models and the magnitude of data.

Context-aware recommender systems are intended primarily to consider the circumstances under which a user encounters an item to provide better-personalized recommendations. Users acquire point-of-interest, movies, products, and various online resources as suggestions. Classical collaborative filtering algorithms are shown to be satisfactory in a variety of recommendation activities processes, but cannot often capture complicated interactions between item and user, along with sparsity and cold start constraints. Hence it becomes a surge to apply a deep learning-based recommender model owing to its dynamic modeling potential and sustained success in other fields of application. In this work, a trust-based attentive contextual denoising autoencoder (TACDA) for enhanced Top-N context-aware recommendation is proposed. Specifically, the TCADA model takes the sparse preference of the user that is integrated with trust data as input into the autoencoder to prevail over the cold start and sparsity obstacle and efficiently accumulates the context condition into the model via attention framework. Thereby, the attention technique is used to encode context features into a latent space of the user's trust data that is integrated with their preferences, which interconnects personalized context circumstances with the active user's choice to deliver recommendations suited to that active user. Experiments conducted on Epinions, Caio, and LibraryThing datasets make it obvious the efficiency of the TACDA model persistently outperforms the state-of-the-art methods.

Recommender systems are widely used in many Web applications to recommend items which are relevant to a user’s preferences. However, focusing on exploiting user preferences while ignoring exploration will lead to biased feedback and hurt the user’s experience in the long term. The Mutli-Armed Bandit (MAB) is introduced to balance the tradeoff between exploitation and exploration. By utilizing context information in the reward function, contextual bandit algorithms lead to better performance compared to context-free bandit algorithms. However, existing contextual bandit algorithms either assume a linear relation between the expected reward and context features, whose representation power gets limited, or use a deep neural network in the reward function which is impractical in implementation. In this paper, we propose a new contextual bandit algorithm, DeepLinUCB, which leverages the representation power of deep neural network to transform the raw context features in the reward function. Specifically, this deep neural network is dedicated to the recommender system, which is efficient and practical in real-world applications. Furthermore, we conduct extensive experiments in our online recommender system using requests from real-world scenarios and show that DeepLinUCB is efficient and outperforms other bandit algorithms.

Recent awareness of privacy protection and compliance requirement resulted in a controversial view of recommendation system due to personal data usage. Therefore, privacy-protected recommendation emerges as a novel research direction. In this paper, we first formulate this problem as a vertical federated learning problem, i.e., features are vertically distributed over different departments. We study a contextual bandit learning problem for recommendation in the vertical federated setting. To this end, we carefully design a customized encryption scheme named orthogonal matrix-based mask mechanism (O3M). O3M mechanism, a tailored component for contextual bandits by carefully exploiting their shared structure, can ensure privacy protection while avoiding expensive conventional cryptographic techniques. We further apply the mechanism to two commonly-used bandit algorithms, LinUCB and LinTS, and instantiate two practical protocols for online recommendation. The proposed protocols can perfectly recover the service quality of centralized bandit algorithms while achieving a satisfactory runtime efficiency, which is theoretically proved and analysed in this paper. By conducting extensive experiments on both synthetic and real-world datasets, we show the superiority of the proposed method in terms of privacy protection and recommendation performance.

Modeling contextual information is a vital part of developing effective recommender systems. Still, existing work on recommendation algorithms has generally put limited focus on the effective treatment of contextual information. Moreover, adding context to recommendation models is challenging since it increases the dimensionality and complexity of the model. Therefore, an efficient learning method is required to extract an association and inter-relationship between user/item features and contextual features for preference-driven modeling. The engineering of features through the exploration of adjacent correlations between the user/item and their context, and their further learning through a distance-based metric, is critical for effective personalization. Motivated by this, we introduce a context-aware recommendation strategy using a ‘contextual grid triplet network.’ This strategy uses a contextual grid topology to capture robust semantic representations of users, items, and contextual data. We present a learning methodology that merges a triplet network with a convolutional neural network. This fusion enables the exploration of associations both ‘within’ the contextual grid, such as between users or items, and ‘between’ different contextual grids, like between a user and items of input. Moreover, we present a variant of a hinge loss function using a triplet network for improved performance and fast convergence. In this work, we study how these aspects boost the quality of top-N recommendations. Furthermore, We show through extensive ablation-based experiments that the proposed method outperforms existing state-of-the-art techniques, demonstrating its robustness and feasibility.

No abstract available

Emojis are becoming increasingly prevalent in everyday online communication such as messaging, email, and social networking. To enhance the user experience of expressing emotions and conveying information through emojis, various techniques have been developed. Our proposed system aims to analyze chat conversations, identify different emotions or topics of discussion, and suggest emojis that align with the context of the conversation. Our model considers contextual and personal information derived from the user’s chat conversations to predict appropriate emojis. The emoji recommendation system is designed with several key considerations in mind. Firstly, our model considers entire conversations, rather than relying on individual sentences to predict emojis. To suggest suitable emojis based on the chat context, our emoji prediction system should take into account multiple previous messages. This includes not only the text and emojis exchanged between participants but also the identity of the speaker and the sequence of sentences within the chat. Secondly, our model seeks to capture different conversational contexts. Conversations can convey a broad range of emotions, information, and feelings. As such, the selection of emojis should be based on the specific context of the ongoing chat, and our recommendation model should offer users a range of emojis that align with different contexts.

ABSTRACT In recent years, recommendation systems have started to gain significant attention and popularity. A recommendation system plays a significant role in various applications and services such as e-commerce, video streaming websites, etc. A critical task for a recommendation system is to model users’ preferences so that it can attain the capability to suggest personalized items for each user. The personalized list suggested by a suitable recommendation system should contain items highly relevant to the user. However, many a times, the traditional recommendation systems do not have enough data about the user or its peers because the model faces the cold-start problem. This work compares the existing three MAB algorithms: LinUCB, Hybrid-LinUCB, and CoLin based on evaluating regret. These algorithms are first tested on the synthetic data and then used on the real-world datasets from different areas: Yahoo Front Page Today Module, Lastfm, and MovieLens20M. The experiment results show that CoLin outperforms Hybrid-LinUBC and LinUCB, reporting cumulated regret of 8.950 for LastFm and 60.34 for MovieLens20M and 34.10 for Yahoo FrontPage Today Module.

Recommender systems can recommend products by analyzing the interests and habits of users. To make more efficient recommendation, the contextual information should be collected in recommendation algorithms. In the restaurant recommendation, the location and the current time of customers should also be considered to facilitate restaurants to find potential customers and give accurate and timely recommendations. However, the existing recommendation approaches often lack the consideration of the influence of time and location. Besides, the data sparsity is an inherent problem in the collaborative filtering algorithm. To address these problems, this paper proposes a recommendation approach which combines the contextual information including time, price and location. Instead of constructing the user-restaurant scoring matrix, the proposed approach clusters price tags and generates the user-price scoring matrix to alleviate the sparsity of data. The experiment on Foursquare dataset shows that the proposed approach has a better performance than traditional ones.

We consider the following variant of contextual linear bandits motivated by routing applications in navigational engines and recommendation systems. We wish to learn a hidden d -dimensional value w ∗ . Every round, we are presented with a subset X t ⊆ R d of possible actions. If we choose (i.e. recommend to the user) action x t , we obtain utility (cid:104) x t , w ∗ (cid:105) but only learn the identity of the best action arg max x ∈X t (cid:104) x, w ∗ (cid:105) . We design algorithms for this problem which achieve regret O ( d log T ) and exp( O ( d log d )) . To accomplish this, we design novel cutting-plane algorithms with low “regret” – the total distance between the true point w ∗ and the hyperplanes the separation oracle returns. We also consider the variant where we are allowed to provide a list of several recommendations. In this variant, we give an algorithm with O ( d 2 log d ) regret and list size poly( d ) . Finally, we construct nearly tight algorithms for a weaker variant of this problem where the learner only learns the identity of an action that is better than the recommendation. Our results rely on new algorithmic techniques in convex geometry (including a variant of Steiner’s formula for the centroid of a convex set) which may be of independent interest.

Point-of-interest (POI) recommendation is a crucial task in location-based social networks, especially for enhancing personalized travel experiences in smart tourism. Recently, large language models (LLMs) have demonstrated significant potential in this domain. Unlike classical deep learning-based methods, which focus on capturing various user preferences, LLM-based approaches can further analyze candidate POIs using common sense and provide corresponding reasons. However, existing methods often fail to fully capture user preferences due to limited contextual inputs and insufficient incorporation of cooperative signals. Additionally, most methods inadequately address target temporal information, which is essential for planning travel itineraries. To address these limitations, we propose PSLM4ST, a novel framework that enables synergistic interaction between LLMs and a lightweight temporal knowledge graph reasoning model. This plugin model enhances the input to LLMs by making adjustments and additions, guiding them to focus on reasoning processes related to fine-grained preferences and temporal information. Extensive experiments on three real-world datasets demonstrate the efficacy of PSLM4ST.

In contextual multi-armed bandits, the relationship between contextual information and rewards is typically unknown, complicating the trade-off between exploration and exploitation. A common approach to address this challenge is the Upper Confidence Bound (UCB) method, which constructs confidence intervals to guide exploration. However, the UCB method becomes computationally expensive in environments with numerous arms and dynamic contexts. This paper presents an adaptive noise exploration framework to reduce computational complexity and introduces two novel algorithms: EAD (Exploring Adaptive Noise in Decision-Making Processes) and EAP (Exploring Adaptive Noise in Parameter Spaces). EAD injects adaptive noise into the reward signals based on arm selection frequency, while EAP adds adaptive noise to the hidden layer of the neural network for more stable exploration. Experimental results on recommendation and classification tasks show that both algorithms significantly surpass traditional linear and neural methods in computational efficiency and overall performance.

This work addresses correlation bias and causal effect confounding in advertising recommendation systems and presents a causal learning–based recommendation framework. We first examine the limitations of conventional recommendation algorithms in complex advertising environments, where confounding variables and exposure bias often prevent models from capturing users’ true preferences. To tackle these issues, we design a unified embedding architecture that jointly represents user, advertisement, and contextual features, and incorporates a structural causal graph to explicitly model dependencies among variables. During model training, causal consistency regularization and inverse propensity weighting are integrated to mitigate the impact of biased exposure mechanisms and non-uniform sampling. A joint optimization objective is further formulated to couple click-through rate prediction with causal consistency estimation, enabling robust causal effect learning without sacrificing predictive accuracy. Extensive experiments on large-scale advertising datasets demonstrate that the proposed approach consistently outperforms several representative baselines in terms of Precision@10, Recall@10, NDCG@10, and MAP, while exhibiting strong robustness under multi-dimensional sensitivity analysis. Overall, this study highlights the practical value of causal modeling and consistency-aware learning in advertising recommendation and offers a computationally grounded approach for improving both interpretability and fairness in recommendation systems.

The principal drug recommendation algorithms are now founded on historical electronic health records. This data insufficiently represents patients’ present health situation and neglects their immediate health requirements, leading to diminished suggestion efficacy. This work introduces a pharmacological recommendation algorithm that integrates online discussions and disease data to address this issue. This approach utilizes a discussion framework and a graph attention network centered on a patient-oriented device. Grey relational analysis integrates graph attention networks to construct connections between nodes. A novel association-aware graph structure is introduced to address the limitations of traditional graph networks in recording node associations. The advanced graph attention network develops a hierarchical dialogue encoder that encodes utterances and dialogue representations. Two categories of relational graph structures are created to illustrate discourse frameworks that incorporate contextual semantics and to comprehend the adjacency relationships among nodes. It utilizes discourse representations to improve drug prediction and recommendation, leveraging knowledge graphs and advanced graph network learning for disease representations. The F1 and Jaccard scores of the proposed technique improved by 1.8% and 3.5%, respectively, in comparison to the leading baseline DDN. This indicates that the algorithm can effectively enhance suggestion efficacy.

The intersection of generative artificial intelligence (GenAI) and financial technology (fintech) is redefining how financial services are conceptualized, delivered, and experienced. As consumer expectations shift toward hyper-personalization, traditional recommendation systems—rooted in rule-based algorithms and shallow learning paradigms—fall short in addressing the dynamic, contextual, and human-centric nature of financial decision-making. This research introduces a novel framework that harnesses the capabilities of GenAI, specifically large language models (LLMs) and multimodal learning, to generate personalized financial product recommendations based on real-time transactional data, behavioral signals, and inferred user intent. This approach fuses techniques from natural language processing, reinforcement learning, and time-series modeling to continuously learn from user interactions, adapting recommendations across life stages and financial contexts. Furthermore, the framework is designed with ethical AI principles at its core, embedding differential privacy, fairness-aware modeling, and explainability layers to ensure regulatory compliance and build user trust. We conduct a robust evaluation using synthetic yet realistic financial datasets, benchmarking against collaborative filtering, matrix factorization, and neural recommender baselines. Results show up to 30% improvement in recommendation relevance, a 25% increase in user engagement, and a notable enhancement in adaptability and interpretability metrics. The proposed GenAI-powered system sets a new direction for intelligent, responsible, and adaptive financial ecosystems in the era of open banking and AI-driven digital transformation.

The recommendation of ideological and political course resources plays a main character in cultivating students' moral values, political awareness, and critical thinking in higher education. Traditional recommendation methods, such as collaborative filtering and content-based models, face challenges including lack of contextual understanding, limited adaptability to dynamic learner needs, and poor personalization in resource delivery. To overcome this challenge, this research paper proposes the intelligent recommendation systems using deep gated and recurrent reinforcement learning networks. The suggested model evaluates the student's engagement behaviour for the ideological and political course in the academic institutions and Campuses. The experimentation wide range is carried out with EROLP datasets and performance metrics such as recall, accuracy, and precision, are calculated for recommending the feedbacks in accordance to the students' engagements. The comprehensive results demonstrate the suggested model has effectively enhanced than other traditional deep learning architectures by producing the recall of 0.9865, accuracy of 0.999, and precision of 0.987 respectively. The deployment of the deep learning-based recommendation systems has shown the brighter path in educating the impactful ideological and political younger generation.

Traditional recommender systems (RecSys) primarily infer user preferences from implicit signals (such as clicks, watches, and purchases), often neglecting the rich explicit contextual feedback users provide through verbal text, like comments and reviews. This explicit context feedback captures the nuanced reasons behind user decisions regarding their preferences. In addition, it offers critical heterogeneous information for user preference alignment and more explainable recommendations. Overlooking such signals can lead to misaligned user preferences and further reinforce filter bubbles, as algorithms fail to understand the "semantic context" behind user choices. Recent advances in Large Language Models (LLMs) present new opportunities to harness user-generated content for more accurate and diverse recommendations, yet current LLM-based recommendations still focus on using item meta-data and underutilize this resource. In this paper, we advocate for prioritizing explicit context feedback in the next generation of LLM-based RecSys. We review the evolution of recommendation paradigms, highlight the value of context-rich feedback, call for new benchmarks and metrics, and introduce frameworks for integrating explicit user signals into scalable LLM-driven RecSys. Centering on user-preference modeling, we aim to foster more personalized, transparent, and explainable RecSys online platforms.

Modern movie recommendation systems face challenges such as dynamic personalization and real-time adaptability. Traditional methods like collaborative filtering and content-based recommendations struggle with dynamic user preferences and cold-start problems. Contextual multi-armed bandit (CMAB) algorithms offer a promising solution by balancing the exploration and exploitation trade-off while incorporating contextual information. This paper evaluates the performance of CMAB algorithms in dynamic movie recommendation scenarios using the MovieLens Beliefs Dataset 2024. The study employs Linear Upper Confidence Bound (LinUCB) and Contextual Thompson Sampling (CTS) algorithms, incorporating user historical data and expected ratings as context. Results show that CMAB algorithms significantly outperform traditional multi-armed bandits (MAB) in terms of cumulative regret and rating accuracy. Specifically, LinUCB achieves rating accuracies of 0.712, compared to 0.623 for the Upper Confidence Bound (UCB) algorithm, indicating a 10% improvement. The enhanced context with user expectation ratings further improves recommendation performance. This research demonstrates the effectiveness of CMAB algorithms in dynamic environments and provides insights for future recommendation system designs.

Contextual bandit algorithms are crucial in various decision-making applications, such as personalized content recommendation, online advertising, and e-commerce banner placement. Despite their successful applications in various domains, contextual bandit algorithms still face significant challenges with exploration efficiency compared to non-contextual bandit algorithms due to exploration in feature spaces. To overcome this issue, model selection policies such as MetaEXP and MetaCORRAL have been proposed to interactively explore base policies. In this paper, we introduce a novel counterfactual approach to address the model selection problem in contextual bandits. Unlike previous methods, our approach leverages unbiased Off-Policy Evaluation (OPE) to dynamically select base policies, making it more robust to model misspecification. We present two new algorithms, MetaEXP-OPE and MetaGreedy-OPE, which utilize OPE for model selection policy. We also provide theoretical analysis on regret bounds and evaluate the impact of different OPE estimators. We evaluated our model on synthetic data and a semi-synthetic simulator using a real-world dataset, and the results show that MetaEXP-OPE and MetaGreedy-OPE significantly outperform existing policies, including MetaEXP and MetaCORRAL.

Objectives: The purpose of the present investigation is to build a contextual recommendation framework based on user history and proven machine learning algorithms to improve user experience. New OTT entrants can adopt the findings to create their recommendation system or monetize it for other providers. Methods: It follows a hybrid approach where standard, proven algorithms used by large OTT providers and other researchers are evaluated using open-source technologies on the MovieLens-1M dataset. The proposed model describes additional factors like user feedback, contextual search and key performance indicators to evaluate the recommendation framework for improved user experience. Findings: As per the findings, a hybrid framework of Content-Based Filtering, Collaborative Filtering (CF), SVD++ and RBM can provide 80% of the tray requirements of modern OTT platforms. By using open-source libraries and various optimization techniques not only we can achieve a lower error rate, but we can improve user experience by considering other factors like novelty, diversity etc. Novelty and application: The manuscript provides the minimal set of algorithms that can provide the maximum tray requirement of any OTT platform, along with key performance indicators. It also describes contextual time-based tags generated from a sample video, which can be searched by the user in near real time. The proposed service-based framework once implemented, can also be passed to other providers for monetization. Keywords: Recommendation, OTT, Contextual Search, Content Based Filtering, CF, SVD

With the rapid growth of digital music services, personalized music recommendation has become an essential part of enhancing user experience. Traditional recommendation algorithms, primarily relying on collaborative filtering, often struggle to effectively capture complex relationships between users and music, especially when considering dynamic factors like time and context. To address these limitations, this paper introduces the MR-CTCFI (Multi-Relation Collaborative Temporal Contextual Feature Interaction) algorithm. The MR-CTCFI algorithm integrates multi-relational data, including user-item interactions and item-item similarities, with temporal and contextual features, enabling the model to better understand evolving user preferences. By leveraging collaborative filtering alongside temporal context and feature interactions, the algorithm improves the precision and adaptability of recommendations over time. Experimental results on diverse music datasets demonstrate that the MR-CTCFI-based method outperforms traditional approaches, offering more accurate, context-aware, and dynamic recommendations. This work contributes to the field of music recommendation systems by providing a novel framework that incorporates complex data interactions and temporal dynamics to enhance user satisfaction and recommendation quality.

There is a growing interest in using social networks to model users’ dependencies and information-sharing behaviors in modern applications. However, the existence of anomalies could significantly distort system performance and decisionmaking. To tackle this problem, it is imperative to design efficient online learning algorithms that robustly learn users’ preferences in the presence of anomalies. In this paper, we study the problem of anomaly detection in attributed networks within an online learning framework by characterizing the users’ preferences and residuals of feature information. We propose a novel bandit algorithm that simultaneously provides personalized recommendation strategies for individual users and detects anomalies in real time. Experiments demonstrate that our proposed approach obtains better performance compared to state-of-the-art algorithms.

The traditional single-context recommendation algorithm is constrained by the sparse connection between users and items and a user/item cold start problem. Based on the review, the text-based cross-context recommendation algorithm extracts user/item comment information in the auxiliary context to alleviate the target context’s data sparsity problem and improve recommendation accuracy. A cross-context recommendation strategy based on self-attention-based federated learning (SAFL) is suggested in this work. In contrast to current algorithms, SAFL fully mixes the target and auxiliary background information. Federated learning addresses crucial challenges including data privacy, data security, data access rights, and access to heterogeneous data by allowing several players to develop a single, strong machine learning model without sharing data. To begin, self-attention mechanism is introduced to model the user’s preference information; then, the information from one field is used to improve the recommendation accuracy of another area; and finally, the information from the two contexts is integrated into the knowledge fusion module and the score prediction module to predict the score. Experiments on the Amazon dataset indicate that MAE and MSE values of SAFL are higher when compared to the current cross-context recommendation model, MAE values rose by 8.4%, 13.2%, and 19.4% across three cross-context datasets.

Code recommendation with programming context is to use the contextual code surrounding the missing code to automatically find that which of the code snippets would be useful to assist in the completion of the program. In this way, the developers need not take time to formulate explicit queries or write descriptions. Existing work only treats code as textual documents and use information retrieval techniques to retrieve relevant code snippets, which is difficult to capture the semantics of code adequately. The self-attention mechanism have achieved promising progress in various natural language processing tasks, especially for extracting deep semantic information from long sequences. Inspired from this, we propose a novel code recommendation with programming context based on self-attention mechanism (CRAM). The proposed approach first builds a small-scale candidate set from codebase. Then, it utilizes self-attention networks in the abstract syntax tree to capture the deep semantics of code, and finally recommend the relevant code to developers. We conduct several experiments to evaluate our approach in a large-scale codebase containing 741 148 code snippets. The experimental results show that CRAM can effectively recommend code and outperforms related work in recall, precision, and NDCG.

This paper introduces an innovative online resource recommendation system tailored for English text and video content, leveraging the power of attention mechanisms and graph neural networks. Given the exponential growth of online learning resources, a crucial challenge lies in delivering personalized and efficient recommendations to users. Our study strives to optimize both the accuracy and efficiency of these recommendations by harnessing the synergistic effects of attention mechanisms and GNNs. By collecting and analyzing a large amount of user behavior data, we build a user-resource interaction graph. This graph not only contains the interaction information between users and resources, but also incorporates the association information between resources, providing a rich context for subsequent recommendations. We introduce an attention mechanism to handle node and edge information in graphs. By assessing the significance of various nodes and edges in the recommendation process, we are able to capture users’ interests and preferences with greater precision. According to experimental data, the integration of an attention mechanism has led to a notable improvement in the system’s recommendation accuracy, achieving an increase of approximately 15%. This significant enhancement underscores the effectiveness of the attention mechanism in effectively capturing user interests. Additionally, we leverage graph neural networks to model the intricate structural information within the graph. With graph convolution operations, we are able to capture potential relationships between resources and use these relationships in the recommendation process. Experimental results show that combined with GNN, the recommendation coverage of the system has increased by about 20%, providing users with more diverse recommendation results. The proposed online resource recommendation system for English text and video based on attention mechanism and GNN has achieved significant improvements in both accuracy and diversity of recommendations. In the future, we will further explore more optimization methods to provide more personalized and efficient online learning resource recommendation services.

Many methods of TPL recommendation are developed to help app developers find available TPLs, but current methods encounter two common limitations when evaluating the match degree between the app and the candidate TPL. First, they mainly complete the evaluation task based on the TPL context, but ignore the function information of the app. In fact, since TPLs are primarily used to support the implementation of various app functions, this information is critical for making an informed decision. Second, they focus on the relationship between the candidate TPL and the entire TPL context. However, it is crucial to emphasize that a candidate TPL is worth considering as long as it can collaborate with certain TPLs in the TPL context. In this study, we propose a novel model called Atten-TPL that can evaluate the match degree between the app and the candidate TPL by mining the relationships among the candidate TPL, TPL context, and app functions. By using an attention mechanism, Atten-TPL can pay more attention to the TPLs in the TPL context that belong to the same task domain as the candidate TPL, thus mitigating the second limitation of current methods. The experiments indicate that Atten-TPL outperforms prevalent methods of recommendation.

Graph Neural Network (GNN)-based collaborative filtering has emerged as a powerful approach for recommendation systems by leveraging user-item interaction graphs to capture higher-order relationships. While existing methods exploit graph topology, they often overlook the heterogeneous importance of neighboring nodes and suffer from data sparsity, leading to sub optimal representations. To address these issues, we propose a novel GNN-based model that synergistically integrates efficient simple dot-product attention with multi-view contrastive learning. Specifically, the attention module assigns context-aware weights to neighbors based on spectral graph theory, dynamically filtering noisy interactions. Concurrently, structural contrastive learning (via edge dropout) and semantic contrastive learning (via feature augmentation) generate hard negative samples to enhance representation discriminability. The unified objective function combines Bayesian Personalized Ranking loss with contrastive losses, jointly optimizing recommendation accuracy and robustness. Extensive experiments on three benchmarks (MovieLens-1M, Yelp, and Amazon-Books) demonstrate consistent improvements of $\mathbf{1 2. 3 \%}$ in Recall@20 and 9.7% in NDCG@20 over state-of-the-art baselines, particularly under sparse interactions. Ablation studies further validate the necessity of contrastive components.

In recent years, the integration of artificial intelligence in education has improved the efficiency of learning resource recommendations, especially for ideological and political courses. However, traditional recommendation systems often struggle to understand the deep relationships between learning materials, students' cognitive levels, and ideological context, resulting in repetitive or irrelevant suggestions. To solve these challenges, this research proposes a Cognitive Graph Convolutional Network (CGCN) with Attention Mechanismbased recommendation model that combines cognitive behavior analysis with graph learning. The model captures both the semantic connections among teaching materials and the cognitive progression of learners, enabling more accurate and personalized recommendations. The system integrates two datasets — the Manifesto Project Database (MPD) for ideological and political text data and EdNet for student learning behavior and engagement patterns. By combining these sources, the model can recommend personalized ideological and political learning materials based on both content relevance and cognitive understanding. The proposed system is evaluated using accuracy, precision, recall, F1-score as performance metrics. Experimental results show that the CGCN-based model achieves 94.8% accuracy, outperforming conventional neural recommendation models.

Tensor factorization is an effective tool that has been successfully applied in the field of context-aware recommendation. However, most existing factorization models assume a multilinear relationship between recommendation rating entries and their corresponding factors, whereas in reality, real-world tensors often contain more complex interactions. In addition, recommendation data usually exhibits sparsity, which limits the amount of information that can be learned. In order to solve the above problems, this paper proposes a new nonlinear tensor factorization model called Convolution and Attention based Tensor Factorization (CoATF). First, we introduce a more generalized implicit feedback to comprehensively represent user preference. Next, a two-layer convolutional neural network is used to model the interactions between tensor factors. Finally, the attention mechanism is utilized to weight the features and improve the robustness of the model. The results of extensive experiments on multiple context-aware recommendation tensors show that the CoATF model significantly outperforms linear and nonlinear state-of-the-art tensor decomposition correlation models with superior recommendation performance.

Tourism experiences are shaped by rapidly changing conditions such as weather, local events, and visitor flows, yet most recommendation systems assume stable patterns, limiting their ability to adapt in real time. This study introduces a robust context‑aware tourism recommendation framework that integrates a Hypergraph Neural Network, an Echo State Network reservoir tuned to operate at the edge of chaos, and a transformer with Lyapunov‑adaptive attention. The hypergraph encoder models complex, multi‑entity relationships among users, destinations, and contextual factors; the reservoir captures evolving context signals with high sensitivity; and the Lyapunov‑adaptive attention mechanism adjusts focus based on online estimates of the largest Lyapunov exponent, enabling the system to detect and respond to sudden regime shifts. The framework is trained and evaluated on the publicly available Travel Recommendation Dataset from IEEE DataPort, enriched with historical weather records and local event schedules. Comparative experiments against strong context‑aware, graph‑based, and sequence‑based baselines show consistent improvements in accuracy, measured by hit rate and normalized discounted cumulative gain, and in diversity, measured by intra‑list diversity and serendipity, particularly under simulated disruptions such as abrupt weather changes. These results demonstrate that combining graph learning, recurrent dynamics, and chaos‑aware attention can substantially increase the resilience of personalization in volatile environments, paving the way for recommendation systems that remain both relevant and exploratory despite unpredictable shifts in user context.

Modeling user intention with limited evidence in short-term historical sequences is a major challenge in session recommendation. In this domain, research exploration extends from traditional methods to deep learning. However, most of them solely concentrate on the sequential dependence or pairwise relations within the session, disregarding the inherent consistency among items. Additionally, there is a lack of research on context adaptation in session intention learning. To this end, we propose a novel session-based model named C-HAN, which consists of two parallel modules: the context-embedded hypergraph attention network and self-attention. These modules are designed to capture the inherent consistency and sequential dependencies between items. In the hypergraph attention network module, the different types of interaction contexts are introduced to enhance the model’s contextual awareness. Finally, the soft-attention mechanism efficiently integrates the two types of information, collaboratively constructing the representation of the session. Experimental validation on three real-world datasets demonstrates the superior performance of C-HAN compared to state-of-the-art methods. The results show that C-HAN achieves an average improvement of 6.55%, 5.91%, and 6.17% over the runner-up baseline method on Precision@K, Recall@K, and MRR evaluation metrics, respectively.

No abstract available

The role of group recommender systems is pivotal in recommending content to groups of users across various information systems. In this paper, we propose, a deep reinforcement learning based Group Recommendation system with Multi-head Attention mechanism (GRMA), to address the need for dynamic preference aggregation in group recommender systems. By incorporating multi-head attention, GRMA captures varying weights of group members and their interactions with different items, resulting in a comprehensive modeling of member-item interactions. The multi-head attention mechanism enhances the system's adaptability to varying group dynamics and preferences, enabling personalized and context-aware recommendations. Evaluations on the MovieLens-Rand dataset, a random group dataset generated from MovieLens 1M, using two baselines (DRGR and AGREE), demonstrate that GRMA outperforms DRGR by enabling a more comprehensive modeling of member-item interactions. Moreover, GRMA achieves a better NDCG@K score compared to AGREE as GRMA using a multi-head attention mechanism improves ranking quality by placing more pertinent items at the top of the recommendation list.

No abstract available

Session-based recommendation is a crucial task aiming to predict users’ interested items based only on anonymous user behaviors. Most recent solutions for session-based recommendation comprehensively consider the interactive information of all sessions but bring the problem of imbalanced positive and negative samples on model training. In this paper, we propose a novel approach, named Attention-enhanced Graph Neural Networks with Global Context for Session-based Recommendation (AGNN-GC), to learn and merge item transitions of all sessions in a cleverer way to enhance the recommendation effects. AGNN-GC first constructs global and local graphs based on all training sequences. Next, it uses graph convolutional networks with a session-aware attention mechanism to learn global-level item embedding in all sessions. Then it employs a graph attention networks module to learn local-level item embedding in the current sessions. After that, it fuses the learned two-level item embedding to enhance the feature presentations of items in the current session by a novel attention mechanism. Finally, applying the focal loss to balance positive and negative samples on model training accomplishes the prediction. Our experiments on three real-world datasets consistently show the superior performance of AGNN-GC over state-of-the-art methods.

POI(point of interest) recommendation is a very necessary research field in both academic and commerce, however, predicting users’ potential points of interest is always faced with the problems of data sparsity and context semantics. Some studies have shown that graph embedding technology alleviates the problem of data sparsity to a certain extent. However, neither graph embedding techniques nor unsupervised learning models can adaptively learn the different effects of multiple relations between users and POIs, respectively. In view of this, we leverage the contextual information of users and POIs to build the multi-view affinity graphs(e.g. User-User, POI-POI and User-POI), and learn the latent representations of users and POIs based on the Graph Embedding technology and Attention mechanism, namely the GEA model. In particular, we first construct multi-view affinity graphs by using user’s social relationship, geographical distance and check-in behaviour, and embed them into a low dimensional shared space to learn the latent representation of users and POIs. Afterwards, in order to take advantage of the different effects of multiple relationships in the final recommendation task, we exploit the attention mechanism to obtain the fused latent representation and make recommendation according users’ potential preferences. Finally, we design a multi-task objective function for joint optimization to obtain more accurate recommendation results. Extensive experiments on Gowalla have verified the effectiveness of our model.

Recently, with the triumph of deep learning, attention mechanism, and graph convolutional networks in their respective fields, using new representation learning techniques or introducing auxiliary information to improve the representation ability of embedding has become the core content of the recommendation algorithm research. Generally, most existing GNN-based recommendation methods recursively propagate embedding information on the graph structure and capture collaborative signals by exploring the high-level connectivity between users and items. Despite the great success, those methods do not consider the influence of temporal context on user preferences embedding information propagation, nor do they distinguish the contribution of different neighbor node information to the target node. In order to address the two problems, we propose a graph collaborative filtering model TAGCF combing time factors and attention based on the existing method. The model uses the time factor to integrate temporal information into the process of embedding information propagation and uses the attention mechanism to distinguish the influence of embedding information from different neighbors. The effectiveness of TAGCF, time information, and attention mechanism are verified through comparative experiments with multiple baseline methods on the two recommendation system datasets, MovieLens and Amazon-books.

In the dynamic discipline of clever tourism, personalized journey pointers have grow to be critical for catering to the numerous options of modern travelers. This research introduces DeepRouteRecommendation (DRR), an innovative deep learning-based framework designed to craft context-conscious and consumer-centered tour itineraries. DRR stands out by using incorporating a wide array of user statistics, along with demographic facts, beyond tour behaviors, user preferences, and subtle comments, to create information of user decision-making strategies. The framework employs a complex multi-layer neural network architecture, complemented by sequence modeling strategies, to maintain the spatial and temporal coherence of the factors of interest (POIs). To strengthen methodological clarity, the abstract now explicitly specifies that DRR uses an LSTM-based sequence encoder combined with an attention mechanism to align user intent with POI characteristics. Additionally, a constraint-conscious optimization module ensures that the generated itineraries are realistic, taking into consideration factors such as budget constraints, time availability, and the accessibility of POIs. The evaluation was performed on a clearly defined hybrid dataset comprising real-world POI data (TripAdvisor + OpenStreetMap) integrated with synthetically generated user profiles to ensure diverse behavioral patterns. The outcomes verified that DRR drastically outperformed conventional recommendation structures, as well as Collaborative Filtering (CF), Content-Based Filtering (CBF), and Reinforcement Learning-based Route (RL-Route) techniques. Specifically, DRR outperformed the strongest baseline (RL-Route) by 12.3% in Recall, 11.7% in Precision, and 10.8% in Diversity, achieving a Recall of 82.4%, a Precision of 79.6%, and a Diversity score of 81.2%.

Learning resources in online learning systems typically adhere to uniform formats and settings, lacking flexibility and personalization to meet diverse learning needs and preferences. This inability to meet individualized learning needs and preferences has spurred research interest in personalized learning path recommendations. Many researchers have explored recommending learning path by leveraging user historical learning resource sequence to model personalized characteristics. However, these methods overlook the time information in the learning process and fail to interpret the dynamic shifts in learning preferences during recommendation. Therefore, we propose a method, termed TA-RL, for learning path recommendation, based on time-aware attention mechanism and reinforcement learning. First, we propose a novel time-aware attention mechanism to trace the evolving learning preferences of user, in which attention weights are computed using a context-aware time distance measure and the similarity between history learning resources. Then, we employ a Monte Carlo policy gradient reinforcement learning method to generate learning path recommendation based on learning preferences. We validate the effectiveness of our proposed method by comprehensive experiments on two real-world datasets.

From the past few years, personalized news recommendation systems have gained significant attention by helping several users to access relevant information from large amount of online content. Several recommendation systems for news have been proposed that incorporates collaborative filtering and content-based filtering. However, the existing Double layer residual and Double Multi head self-attention mechanism (DDM) based recommendation system failed to include positional context awareness and temporal evolution of user interests. This research overcomes these limitations through a Contextual-Temporal Enhanced Double layer residual and Double Multi head self-attention mechanism (CTDDM) to achieve adaptive personalization. Firstly, data related to news and user queries is collected from Microsoft News Dataset (MIND) and preprocessed through Term Frequency Inverse Document Frequency (TF-IDF) for text weighing, handling missing values using mean substitution and min max normalization for temporal scaling. Next, the encoded news titles are processed through positional encoding integrating with DDM block and changing user interest are captured by time aware attention mechanism. Finally, a dot similarity function is used to calculate final click prediction. Experimental demonstrates that the proposed CT-DDM achieved Area Under Curver (AUC) of 0.7018 and Mean Reciprocal Rank (MRR) of 0.3377 demonstrating improved personalized recommendation.

No abstract available

Session-based recommendation (SBR) aims to provide personalized recommendations based on anonymous user click sequences. Although existing methods have achieved notable progress, most focus solely on user preferences within a single session, overlooking item transitions across sessions, which limits their ability to model complex behavior patterns. To address this, we propose GCACL-Rec, a model that enhances dynamic modeling by incorporating global item transition information. It constructs a multi-scale graph structure using Multi-scale graph neural networks (MSGNN) and introduces a relative multi-head attention mechanism (RMA) to enhance cross-session dependency modeling. In addition, a multi-view contrastive-adversarial joint learning strategy (MPACL) is adopted to distinguish better relevant from irrelevant information and extract user intent more effectively. During prediction, we use a hybrid structure that combines a neural decision forest (NDF) with the softmax function to enable fine-grained decision optimization and improve feature discrimination and accuracy. Experiments on the Diginetica, Tmall and RetailRocket benchmark datasets show that GCACL-Rec outperforms existing methods, demonstrating clear advantages in cross-session recommendation tasks.

In recent years, there has been a growing interest in developing next point-of-interest (POI) recommendation systems in both industry and academia. However, current POI recommendation strategies suffer from the lack of sufficient mixing of details of the features related to individual users and their corresponding contexts. To overcome this issue, we propose a deep learning model based on an attention mechanism in this study. The suggested technique employs an attention mechanism that focuses on the pattern’s friendship, which is responsible for concentrating on the relevant features related to individual users. To compute context-aware similarities among diverse users, our model employs six features of each user as inputs, including user ID, hour, month, day, minute, and second of visiting time, which explore the influences of both spatial and temporal features for the users. In addition, we incorporate geographical information into our attention mechanism by creating an eccentricity score. Specifically, we map the trajectory of each user to a shape, such as a circle, triangle, or rectangle, each of which has a different eccentricity value. This attention-based mechanism is evaluated on two widely used datasets, and experimental outcomes prove a noteworthy improvement of our model over the state-of-the-art strategies for POI recommendation.

Social recommendation utilizes social relations to extract auxiliary collaborative signals, effectively mitigating data sparsity issues. However, existing approaches predominantly focus on static influence from social friends while neglecting two critical aspects: the dynamic contextual patterns in user behaviors and the potential of collaborative users. To address these limitations and further alleviate data sparsity, we propose a context-aware dual graph attention network (CDGA) that simultaneously captures users’ static and dynamic interests through social relations and interaction records. The proposed CDGA model introduces a dynamic activation mechanism to simulate contextual influences, generating dynamic embeddings for users and items. Furthermore, we develop an adaptive fusion mechanism that integrates interaction channels across static and dynamic embeddings for interaction prediction. Extensive experiments on three benchmark datasets demonstrate that CDGA consistently outperforms state-of-the-art social recommendation methods, confirming its effectiveness.

Industrial cyber-physical systems are smart systems, which amalgamate the physical processes with computational capabilities to seamlessly capture, monitor and control the entities and scenarios in industrial environments. Among them, event-based industrial cyber-physical systems (EICPSs), such as Meetup and Plancast, have gained rapid developments. EICPSs provide event recommendation service for groups, which alleviates the information overload problem. However, existing group recommendation models in EICPSs focus on how to aggregate the preferences of group members, failing to model the complex and deep influence of contexts on groups. In this article, we propose an attention-based context-aware group event recommendation model (ACGER) in EICPSs. ACGER models the deep, nonlinear influence of contexts on users, groups, and events through multilayer neural networks. Especially, a novel attention mechanism is designed to enable the influence weights of contexts on users/groups change dynamically with the events concerned. Considering that groups may have completely different behavior patterns from group members, we acquire the preference of a group from two perspectives: indirect preference and direct preference. To obtain the indirect preference, we propose a method of aggregating preferences based on attention mechanism. Compared with existing predefined strategies, this method can flexibly adapt the strategy according to the events concerned by the group. To obtain the direct preference, we employ neural networks to learn it from group-event interactions. Furthermore, to make full use of rich user-event interactions in EICPSs, we integrate the context-aware individual recommendation task into ACGER, which enhances the accuracy of learning of user embeddings and event embeddings. Extensive experiments on three real datasets from Meetup and Douban event show that our model ACGER significantly outperforms the state-of-the-art models.

Sequential Recommendation (SR) navigates users' dynamic preferences through modeling their historical interactions. The incorporation of the popular Transformer framework, which captures long relationships through pairwise dot products, has notably benefited SR. However, prevailing research in this domain faces three significant challenges: (i) Existing studies directly adopt the primary component of Transformer (i.e., the self-attention mechanism), without a clear explanation or tailored definition for its specific role in SR; (ii) The predominant focus on pairwise computations overlooks the global context or relative prevalence of item pairs within the overall sequence; (iii) Transformer primarily pursues relevance-dominated relationships, neglecting another essential objective in recommendation, i.e., diversity. In response, this work introduces a fresh perspective to elucidate the attention mechanism in SR. Here, attention is defined as dependency interactions among items, quantitatively determined under a global probabilistic model by observing the probabilities of corresponding item subsets. This viewpoint offers a precise and context-specific definition of attention, leading to the design of a distinctive attention mechanism tailored for SR. Specifically, we transmute the well-formulated global, repulsive interactions in Determinantal Point Processes (DPPs) to effectively model dependency interactions. Guided by the repulsive interactions, a theoretically and practically feasible DPP kernel is designed, enabling our attention mechanism to directly consider category/topic distribution for enhancing diversity. Consequently, the Probabilistic Attention mechanism (PAtt) for sequential recommendation is developed. Experimental results demonstrate the excellent scalability and adaptability of our attention mechanism, which significantly improves recommendation performance in terms of both relevance and diversity.

No abstract available

Sequential models have achieved admirable success in recommendation systems. However, most sequential models typically only consider the chronological order of items through timestamps and ignore the relative distances in the sequence, which weakens the temporal relationships between items. To address this issue, we propose a temporal recommendation system using the Gaussian distribution and attention mechanism, which considers the sequentiality and interaction among items. Technically, we first deploy the word vector space along the time dimension as sequence features. Then, we use the Gaussian process to effectively represent the duration influence of items and the context interaction between items as high-level features. Finally, an innovative attention mechanism is used to capture the hidden correlation relationships between representation subspaces of different levels of features. Experiments conducted on two widely used real public datasets show that our model outperforms the state-of-the-art recommendation systems.

No abstract available

In the context of patients with complex conditions, the capacity to assist physicians in making appropriate prescribing decisions is of paramount importance. The field of prescription recommendation has attracted a growing body of research interest, given its significant clinical value. However, existing research has not taken into account the patients’ intention to visit the doctor and the similarities between patients’ visit intentions. In this study, we employ medical order data in both prescription intention modeling and patient similarity analysis, subsequently integrating these to generate more personalized and accurate prescription recommendations. The cross-attention mechanism is first employed to focus the representation of the patient’s health status on the patient’s chief complaints, with the objective of predicting treatment intention and, subsequently, the prescriptions. Subsequently, when we learn from the successful treatment experiences of similar patients, we focus more on the similarities of their chief complaints. The experimental results on the MIMIC-III dataset demonstrate that our method achieves optimal performance in terms of Jaccard score, PRAUC, F1 score and average precision, with relative improvements of 7%, 5%, 4%, and 8% respectively, in comparison to state-of-the-art approaches.

This paper proposes a novel recommendation system model, the Attentive Recurrent Recommender Network (ARRN), that addresses the challenge of incorporating demographic information into recommendations. ARRN leverages user-item interaction data along with age information from the data set to deliver personalized recommendations specifically tailored to different age groups. The approach utilizes embedding techniques and semantic analysis to capture user preferences and behaviors associated with their age. An attention mechanism prioritizes relevant features based on user age groups, enabling ARRN to dynamically adapt recommendations for users with limited interaction history. The paper presents a comprehensive evaluation of ARRN’s performance compared to existing state-of-the-art recommendation algorithms. The results demonstrate that ARRN outperforms existing approaches, particularly for users with limited interaction history, by effectively mitigating the cold-start problem in age-sensitive product domains.

Sequential Recommendation (SR) has been widely used in various Internet platforms, such as content streaming platforms and e-commerce. While the Self-Attention (SA) mechanism has been shown to be effective in capturing global dependencies, its quadratic complexity makes it less efficient. Mamba can improve efficiency, but its unidirectional architecture makes it difficult to capture the full context of a sequence, preventing further development.To address these issues, this work proposes a sequential recommendation model called F2MSA2 This model combines Mamba and SA for hybrid modelling. A modified version of the Mamba block, flipped Mamba (FMA), is first introduced, which flips the interaction sequences and then serves as an input to the Mamba block to enhance the model's ability to acquire and utilize future information. Secondly frequency-domain SA (FSA) transforms the SA from the time domain to the frequency domain in order to acquire both low-frequency and high-frequency information, where the low-frequency information usually reflects the user's long-term stable preferences while the high-frequency information reveals the user's short-term fast-changing interests. And several frequency-domain feed-forward networks are constructed to learn external factors and mitigate their effects on user preferences.

Recommender systems (RS) have been increasingly applied to food and health. However, challenges still remain, including the effective incorporation of heterogeneous information and the discovery of meaningful relationships among entities in the context of food and health recommendations. To address these challenges, we propose a novel framework, the Health-aware Food Recommendation System with Dual Attention in Heterogeneous Graphs (HFRS-DA), for unsupervised representation learning on heterogeneous graph-structured data. HFRS-DA utilizes an attention technique to reconstruct node features and edges and employs a dual hierarchical attention mechanism for enhanced unsupervised learning of attributed graph representations. HFRS-DA addresses the challenge of effectively leveraging the heterogeneous information in the graph and discovering meaningful semantic relationships between entities. The framework analyses recipe components and their neighbours in the heterogeneous graph and can discover popular and healthy recipes, thereby promoting healthy eating habits. We compare HFRS-DA using the Allrecipes dataset and find that it outperforms all the related methods from the literature. Our study demonstrates that HFRS-DA enhances the unsupervised learning of attributed graph representations, which is important in scenarios where labelled data is scarce or unavailable. HFRS-DA can generate node embeddings for unused data effectively, enabling both inductive and transductive learning.

Aiming at the problems resulting from the fact that the existing point of interest (POI) recommendation methods cannot effectively consider the personalized differences of users' mobile behavior in space and time, the author proposes a personalized POI recommendation method using attention-based time sequence and distance contexts gated recurrent unit (ATSD-GRU). First, the author combined the time sequence and distance context with the GRU to extract useful information from users, effectively alleviating the data sparsity. Second, inspired by the attention mechanism, the author introduced the attention model further into the neural network to capture the user's main mobile behavior intention. Finally, the author studied the ATSD-GRU and trained through Bayesian personalized sorting framework and back propagation algorithm. Experiments imply that the proposed method outperforms the comparison method in terms of the F1 index for any recommended number. When the recommendation list length is 15, the proposed algorithm exhibits an accuracy of 9.23% and a recall rate of 14.65%, both higher than the compared algorithm.

PurposeSession-based recommendation aims to predict the user's next preference based on the user's recent activities. Although most existing studies consider the global characteristics of items, they only learn the global characteristics of items based on a single connection relationship, which cannot fully capture the complex transformation relationship between items. We believe that multiple relationships between items in learning sessions can improve the performance of session recommendation tasks and the scalability of recommendation models. At the same time, high-quality global features of the item help to explore the potential common preferences of users.Design/methodology/approachThis work proposes a session-based recommendation method with a multi-relation global context–enhanced network to capture this global transition relationship. Specifically, we construct a multi-relation global item graph based on a group of sessions, use a graded attention mechanism to learn different types of connection relations independently and obtain the global feature of the item according to the multi-relation weight.FindingsWe did related experiments on three benchmark datasets. The experimental results show that our proposed model is superior to the existing state-of-the-art methods, which verifies the effectiveness of our model.Originality/valueFirst, we construct a multi-relation global item graph to learn the complex transition relations of the global context of the item and effectively mine the potential association of items between different sessions. Second, our model effectively improves the scalability of the model by obtaining high-quality item global features and enables some previously unconsidered items to make it onto the candidate list.

Recent studies in recommender systems focus on addressing data sparsity and cold-start problems by utilizing side information, such as tags, images, and testimonials. Among these, user-written testimonials (purchase reviews) are precious for analyzing personal preferences, and many methods have been developed based on this context. Generally, existing methods apply 2D text convolution followed by selecting important words using the attention mechanism. However, the text convolution scheme inevitably suffers from information loss since the number of words in reviews commonly exceeds hundreds. To address this limitation, we focus on the Large Language Model (LLM), which has shown promising results in various fields, including search engines, natural language processing, and healthcare. In particular, LLM has demonstrated excellent performance in text summarization and QA tasks, leading to the development of text-based recommender systems. Nevertheless, LLM alone struggles to perform collaborative filtering, which is essential in a recommender system. Thus, we propose LLM-based text summarization before applying 2D convolution, followed by the widely used collaborative filtering mechanism. This approach can improve recommendation quality by removing unnecessary words in advance, reducing the smoothing effect while capturing the rich user-item interactions. Our method is integrated with recent text-based recommendation algorithms, which have proven to improve the quality of all baselines by about 16.9 % on average. We conduct experiments and ablation studies using benchmark datasets, demonstrating that our method is scalable and efficient.

Personalized news recommendation is a crucial technology for helping users discover news articles tailored to their interests. Key challenges in this field include modeling user preferences based on implicit behaviors, accounting for the influence of the news agenda on user interests, and managing the rapid decay of news items. The ACM RecSys Challenge 2024, organized by Ekstra Bladet, provide a large-scale news dataset for benchmarking news recommendation research. In this paper, we present our solution to the challenge. We propose real-time feedback learning mechanisms to capture users’ immediate interests and explore generative sequence modeling techniques to learn impression-level user behaviors. Furthermore, we develop an ensemble method to combine tree models and deep models to improve recommendation accuracy. Based on this solution, our team ("hrec") achieved an impressive AUC score of 0.8667 on the final test set, securing the fifth place in the competition.

Generative recommendation has emerged as a transformative paradigm in recommender systems, enabling modeling user behavior autoregressively without explicit target conditioning. While this approach eliminates the need for target signals, it necessitates compressing extensive historical interactions-potentially spanning lifelong sequences-into coherent interest representations. Conventional methods for handling long sequences typically rely on target-guided search mechanisms (e.g., SIM) to efficiently filter and compress behaviors. However, this strategy is incompatible with generative frameworks due to their target-agnostic nature. To address these challenges, we propose a novel encoder-decoder model named HiCoGen (Hierarchical Compression-based Session-wise Generative Model), which efficiently models long-term interests in generative models. In the encoder, HiCoGen compresses behavior sequences using hierarchical content similarity clustering and employs a hierarchical attention architecture to reduce sequence length while preserving information integrity. In the decoder, HiCoGen uses session-wise generation instead of point-wise generation to better align with industrial short-video applications. To enhance the stability of session-wise generation, we introduce an auxiliary Hierarchical Multi-Token Prediction module. Extensive experiments on public and industrial datasets show significant performance gains over state-of-the-art methods (21.2% in ML-1M and 35.6% in industrial datasets on NDCG@3). We also conducted visualization and performance analysis to explore the advantages of long sequence modeling.

Generative models have recently demonstrated strong potential in multi-behavior recommendation systems, leveraging the expressive power of transformers and tokenization to generate personalized item sequences. However, their adoption is hindered by (1) the lack of explicit information for token reasoning, (2) high computational costs due to quadratic attention complexity and dense sequence representations after tokenization, and (3) limited multi-scale modeling over user history. In this work, we propose GRACE (Generative Recommendation via journey-aware sparse Attention on Chain-of-thought tokEnization), a novel generative framework for multi-behavior sequential recommendation. GRACE introduces a hybrid Chain-of-Thought (CoT) tokenization method that encodes user-item interactions with explicit attributes from product knowledge graphs (e.g., category, brand, price) over semantic tokenization, enabling interpretable and behavior-aligned generation. To address the inefficiency of standard attention, we design a Journey-Aware Sparse Attention (JSA) mechanism, which selectively attends to compressed, intra-, inter-, and current-context segments in the tokenized sequence. Experiments on two real-world datasets show that GRACE significantly outperforms state-of-the-art baselines, achieving up to +106.9% HR@10 and +106.7% NDCG@10 improvement over the state-of-the-art baseline on the Home domain, and +22.1% HR@10 on the Electronics domain. GRACE also reduces attention computation by up to 48% with long sequences.

Contemporary recommendation systems are designed to meet users' needs by delivering tailored lists of items that align with their specific demands or interests. In a multi-stage recommendation system, reranking plays a crucial role by modeling the intra-list correlations among items. The key challenge of reranking lies in the exploration of optimal sequences within the combinatorial space of permutations. Recent research proposes a generator-evaluator learning paradigm, where the generator generates multiple feasible sequences and the evaluator picks out the best sequence based on the estimated listwise score. The generator is of vital importance, and generative models are well-suited for the generator function. Current generative models employ an autoregressive strategy for sequence generation. However, deploying autoregressive models in real-time industrial systems is challenging. Firstly, the generator can only generate the target items one by one and hence suffers from slow inference. Secondly, the discrepancy between training and inference brings an error accumulation. Lastly, the left-to-right generation overlooks information from succeeding items, leading to suboptimal performance. To address these issues, we propose a Non-AutoRegressive generative model for reranking Recommendation (NAR4Rec) designed to enhance efficiency and effectiveness. To tackle challenges such as sparse training samples and dynamic candidates, we introduce a matching model. Considering the diverse nature of user feedback, we employ a sequence-level unlikelihood training objective to differentiate feasible sequences from unfeasible ones. Additionally, to overcome the lack of dependency modeling in non-autoregressive models regarding target items, we introduce contrastive decoding to capture correlations among these items. Extensive offline experiments validate the superior performance of NAR4Rec over state-of-the-art reranking methods. Online A/B tests reveal that NAR4Rec significantly enhances the user experience. Furthermore, NAR4Rec has been fully deployed in a popular video app Kuaishou with over 300 million daily active users.

Traditional recommender systems such as matrix factorization methods have primarily focused on learning a shared dense embedding space to represent both items and user preferences. Subsequently, sequence models such as RNN, GRUs, and, recently, Transformers have emerged and excelled in the task of sequential recommendation. This task requires understanding the sequential structure present in users’ historical interactions to predict the next item they may like. Building upon the success of Large Language Models (LLMs) in a variety of tasks, researchers have recently explored using LLMs that are pretrained on vast corpora of text for sequential recommendation. To use LLMs for sequential recommendation, both the history of user interactions and the model’s prediction of the next item are expressed in text form. We propose CALRec, a two-stage LLM finetuning framework that finetunes a pretrained LLM in a two-tower fashion using a mixture of two contrastive losses and a language modeling loss: the LLM is first finetuned on a data mixture from multiple domains followed by another round of target domain finetuning. Our model significantly outperforms many state-of-the-art baselines (+37% in Recall@1 and +24% in NDCG@10) and our systematic ablation studies reveal that (i) both stages of finetuning are crucial, and, when combined, we achieve improved performance, and (ii) contrastive alignment is effective among the target domains explored in our experiments.

Deep Learning Recommendation Models (DLRMs) often rely on extensive manual feature engineering to improve accuracy and user experience, which increases system complexity and limits scalability of model performance with respect to computational resources. Recently, Meta introduced a generative ranking paradigm based on HSTU block that enables end-to-end learning from raw user behavior sequences and demonstrates scaling law on large datasets that can be regarded as the state-of-the-art (SOTA). However, splitting user behaviors into interleaved item and action information significantly increases the input sequence length, which adversely affects both training and inference efficiency. To address this issue, we propose the Twin-Flow Generative Ranking Network (TFGR), that employs a Twin-flow mechanism to optimize interaction modeling, ensuring efficient training and inference through end-to-end token processing. TFGR duplicates the original user behavior sequence into a real flow and a fake flow based on the authenticity of the action information, and then defines a novel interaction method between the real flow and the fake flow within the QKV module of the self-attention mechanism. This design reduces computational overhead and improves both training efficiency and inference performance compared to Meta's HSTU-based model. Experiments on both open-source and real industrial datasets show that TFGR outperforms DLRM, which serves as the industrial online baseline with extensive feature engineering, as well as Meta's HSTU and other common recommendation models such as DIN, DCN, DIEN, and DeepFM. Furthermore, we investigate optimal parameter allocation strategies under computational constraints, establishing TFGR as an efficient and effective next-generation generative ranking paradigm.

Watch time prediction (WTP) has emerged as a pivotal task in short video recommendation systems, designed to quantify user engagement through continuous interaction modeling. Predicting users' watch times on videos often encounters fundamental challenges, including wide value ranges and imbalanced data distributions, which can lead to significant estimation bias when directly applying regression techniques. Recent studies have attempted to address these issues by converting the continuous watch time estimation into an ordinal regression task. While these methods demonstrate partial effectiveness, they exhibit notable limitations: (1) the discretization process frequently relies on bucket partitioning, inherently reducing prediction flexibility and accuracy and (2) the interdependencies among different partition intervals remain underutilized, missing opportunities for effective error correction. Inspired by language modeling paradigms, we propose a novel Generative Regression (GR) framework that reformulates WTP as a sequence generation task. Our approach employs \textit{structural discretization} to enable nearly lossless value reconstruction while maintaining prediction fidelity. Through carefully designed vocabulary construction and label encoding schemes, each watch time is bijectively mapped to a token sequence. To mitigate the training-inference discrepancy caused by teacher-forcing, we introduce a \textit{curriculum learning with embedding mixup} strategy that gradually transitions from guided to free-generation modes. We evaluate our method against state-of-the-art approaches on two public datasets and one industrial dataset. We also perform online A/B testing on the Kuaishou App to confirm the real-world effectiveness. The results conclusively show that GR outperforms existing techniques significantly.

Recent advancements in autoregressive Large Language Models (LLMs) have achieved remarkable progress, largely driven by their scalability—commonly formalized as the scaling law. Inspired by these successes, there has been growing interest in adapting LLMs to recommendation systems (RecSys) by reformulating recommendation tasks as generative sequence modeling problems. However, existing End-to-End Generative Recommendation (E2E-GR) methods often sacrifice the practical advantages of traditional Deep Learning-based Recommendation Models (DLRMs)—including mature feature engineering, modular architectures, and production-grade optimization practices. This trade-off introduces critical challenges that hinder the effective application of scaling laws in industrial RecSys. In this paper, we present Large User Model (LUM), a scalable and production-aware framework that bridges the gap between generative modeling and industrial recommendation requirements. LUM addresses these limitations through a principled three-step paradigm, designed to preserve the flexibility of autoregressive generation while maintaining compatibility with real-world deployment constraints. Extensive experiments show that LUM outperforms state-of-the-art DLRMs and E2E-GR approaches across multiple benchmarks. Notably, LUM exhibits strong scalability: performance improves consistently as the model scales up to 7 billion parameters. Furthermore, LUM has been successfully deployed in a large-scale industrial application, where it delivered statistically significant gains in a live A/B test, demonstrating both its effectiveness and practical viability.

Session-based recommendation aims to suggest items to users mainly by modeling sequential dependencies. However, existing methods mainly focus on resorting either temporal interests with time-series data or inherent interests by leveraging static user behaviors, ignoring benefits of the dynamic aggregation from short- and long-term preferences, resulting in unfavorable performance in various sequence recommendation scenarios. In this paper, we attempt to combine users' short interests with long preferences synthetically, and tackle the sequential recommendation problem from new views of the causal (short) and non-causal (long) perspectives. We propose a dynamic gap-complementary generative framework named DGRec for session-based Recommendation. Specifically, we fuse two practical neural network models: the dilated convolutional neural network (DCNN) and the self-attention block (SAB). DCNN mainly captures users' short-term interests while SAB is responsible for obtaining users' long-term inherent behaviors. Besides, we devise a gap-complementary strategy to dynamically learn item sequence representation, effectively enhancing the robustness and anti-interference ability of model. Extensive experimental results over five public datasets demonstrate that DGRec gains significant improvements against the advanced sequential recommendation methods.

To address the limitations of existing learning path recommendation methods—such as poor adaptability, weak personalization, and difficulties in processing long sequences of student behavior and interest data—this paper proposes a personalized learning path recommendation system for the second classroom based on large language model (LLM) technology, with a focus on integrating the pre-trained model GPT-4. The goal is to improve recommendation accuracy and personalization by leveraging GPT-4’s strong long-sequence modeling capability. The system fuses students’ multimodal data (e.g., physiological signals, facial expressions, activity levels, and emotional states), extracts deep features using GPT-4, and generates tailored learning paths based on individual feature vectors. It also incorporates incremental learning and self-attention mechanisms to enable real-time feedback and dynamic adjustments. A generative adversarial network (GAN) is introduced to enhance diversity and innovation in recommendations. The experimental results show that the system achieves a personalized recommendation accuracy of over 92%, with coverage and recall rates exceeding 91% and 93%, respectively. Feedback adjustment time remains within 1.5 s, outperforming mainstream models. This study provides a novel and effective technical framework for personalized learning in the second classroom, promoting both efficient resource utilization and student development.

Large-scale recommendation systems are pivotal to process an immense volume of daily user interactions, requiring the effective modeling of high cardinality and heterogeneous features to ensure accurate predictions. In prior work, we introduced Hierarchical Sequential Transducers (HSTU), an attention-based architecture for modeling high cardinality, non-stationary streaming recommendation data, providing good scaling law in the generative recommender framework (GR). Recent studies and experiments demonstrate that attending to longer user history sequences yields significant metric improvements. However, scaling sequence length is activation-heavy, necessitating parallelism solutions to effectively shard activation memory. In transformer-based LLMs, context parallelism (CP) is a commonly used technique that distributes computation along the sequence-length dimension across multiple GPUs, effectively reducing memory usage from attention activations. In contrast, production ranking models typically utilize jagged input tensors to represent user interaction features, introducing unique CP implementation challenges. In this work, we introduce context parallelism with jagged tensor support for HSTU attention, establishing foundational capabilities for scaling up sequence dimensions. Our approach enables a 5.3× increase in supported user interaction sequence length, while achieving a 1.55× scaling factor when combined with Distributed Data Parallelism (DDP).

Sequential recommendation aims to capture the temporal dependencies of items in a user's historical interactions and make recommendations based on this. Previous generative methods addressed the issue of data not directly reflecting user preference uncertainty by modeling the distribution of latent item representations. Diffusion model (DM)-based methods have achieved significant success due to their high-quality generation and stable training. However, they lack satisfactory user sequence representations to guide the generation process, impacting recommendation performance. Moreover, these methods overlook the drawback of slow inference speed, severely limiting their practical value. To obtain effective generative guidance signals and accelerate the recommendation process, we propose DAE4Rec. In this approach, a Graph Auto-Encoder (GAE) is used to obtain interpretable item node representations, revealing global transitions of items that previous methods struggled to uncover. Then, we use it to construct a generative guidance signal with lower coupling and variance for the diffusion model. Additionally, by employing a non-Markov chain derived from the forward diffusion process, it is the first to implement a 'skip-step' reverse process in diffusion model-based methods. And a creatively designed compensator is used to bridge the performance gap caused by 'skip-step'. Extensive experiments on three real-world datasets demonstrate that DAE4Rec outperforms other state-of-the-art generative sequential recommenders.

The growing adoption of vector databases has revolutionized AI-driven applications across domains like recommendation systems, natural language processing (NLP), and generative AI. However, a critical challenge remains in effectively handling temporal and sequential patterns within vectorized data. This paper addresses the integration of temporal awareness into vector databases, proposing novel methodologies for the storage, retrieval, and analysis of time-sensitive embeddings. By extending traditional vector search models to incorporate the time dimension, we enable more accurate, context-aware decisionmaking and predictions across dynamic applications such as time-series forecasting, real-time event detection, and user behavior modeling. Through a detailed review of current techniques—ranging from dynamic embedding models, recurrent neural networks (RNNs), and transformers for sequence learning, to hierarchical indexing and sliding window approaches for temporal query optimization—this paper provides a comprehensive framework for managing evolving data streams. We also explore key challenges, including scalability, query optimization, and the integration of realtime data streams, highlighting the need for robust solutions that balance accuracy and efficiency in large-scale, time-sensitive environments. The findings aim to pave the way for future advancements in AI systems, enhancing their ability to manage complex, dynamic data while unlocking new opportunities in predictive analytics, personalized recommendations, and anomaly detection.

The goal of modern sequential recommender systems is often formulated in terms of next-item prediction. In this paper, we explore the applicability of transformer-based generative models for the Top-K sequential recommendation task, where the goal is to predict items that a user is likely to interact with in the “near future.” This goal aligns with real-world applications of such models in an offline scenario or as a part of multi-stage recommender pipelines. We explore commonly used autoregressive generation strategies, including greedy decoding, beam search, and temperature sampling, to evaluate their performance for the Top-K sequential recommendation task. In addition, we propose novel Reciprocal Rank Aggregation (RRA) and Relevance Aggregation (RA) generation strategies based on multi-sequence generation with temperature sampling and subsequent aggregation. Experiments on diverse datasets give valuable insights regarding the applicability of commonly used strategies and show that the suggested approaches improve performance on longer time horizons compared to the widely used Top-K prediction approach and single-sequence autoregressive generation strategies.

Sequential recommendation is a classic task in the field of recommendation, which aims to predict the next user-preferred item based on their historical interactions. However, in practical scenarios, users' needs are dynamically evolving in a short period and exhibit a chain-like structure. Consequently, recommending only the next single item does not fully meet user demands and limits the potential for increasing business traffic on platforms. To overcome this limitation, we propose a new recommendation paradigm, Next Chain Prediction, which requires the model to predict a chain of items. Due to the advantages of generative recommendation models on user preference representation and scalability, we design a generative recommendation model for next chain prediction. The generative model extracts long-term interests and short-term demands within a unified framework. By designing a Sequence-Chain Attention mechanism, the model performs self-attention learning across dual dimensions. Additionally, we design a generative loss function to balance the hit rate and diversity of the recommended items in the chain. We conduct experiments across three datasets and the experimental results show that our method achieves at least 1.22% improvement in HR@10 across three datasets for recommending multi-item chains. Furthermore, our method improves the diversity of recommended items and also offers the flexibility to adjust the size of predicted chains, maintaining state-of-the-art performance even when limited to predicting a single item.

Although generative recommenders demonstrate improved performance with longer sequences, their real-time deployment is hindered by substantial computational costs. To address this challenge, we propose a simple yet effective method for compressing long-term user histories by leveraging inherent item categorical features, thereby preserving user interests while enhancing efficiency. Experiments on two large-scale datasets demonstrate that, compared to the influential HSTU model, our approach achieves up to a 6× reduction in computational cost and up to 39% higher accuracy at comparable cost (i.e., similar sequence length). The source code will be available at https://github.com/Genemmender/CAUSE.

Leveraging Large Language Models (LLMs) for generative recommendation has attracted significant research interest, where item tokenization is a critical step. It involves assigning item identifiers for LLMs to encode user history and generate the next item. Existing approaches leverage either token-sequence identifiers, representing items as discrete token sequences, or single-token identifiers, using ID or semantic embeddings. Token-sequence identifiers face issues such as the local optima problem in beam search and low generation efficiency due to step-by-step generation. In contrast, single-token identifiers fail to capture rich semantics or encode Collaborative Filtering (CF) information, resulting in suboptimal performance. To address these issues, we propose two fundamental principles for item identifier design: 1) integrating both CF and semantic information to fully capture multi-dimensional item information, and 2) designing order-agnostic identifiers without token dependency, mitigating the local optima issue and achieving simultaneous generation for generation efficiency. Accordingly, we introduce a novel set identifier paradigm for LLM-based generative recommendation, representing each item as a set of order-agnostic tokens. To implement this paradigm, we propose SETRec, which leverages CF and semantic tokenizers to obtain order-agnostic multi-dimensional tokens. To eliminate token dependency, SETRec uses a sparse attention mask for user history encoding and a query-guided generation mechanism for simultaneous token generation. We instantiate SETRec on T5 and Qwen (from 1.5B to 7B). Extensive experiments on four datasets demonstrate its effectiveness across various scenarios (e.g., full ranking, warm- and cold-start ranking, and various item popularity groups). Moreover, results validate SETRec's superior efficiency and show promising scalability on cold-start items as model sizes increase.

Generative recommendation (GR) is an emerging paradigm where user actions are tokenized into discrete token patterns and autoregressively generated as predictions. However, existing GR models tokenize each action independently, assigning the same fixed tokens to identical actions across all sequences without considering contextual relationships. This lack of context-awareness can lead to suboptimal performance, as the same action may hold different meanings depending on its surrounding context. To address this issue, we propose ActionPiece to explicitly incorporate context when tokenizing action sequences. In ActionPiece, each action is represented as a set of item features. Given the action sequence corpora, we construct the vocabulary by merging feature patterns as new tokens, based on their co-occurrence frequency both within individual sets and across adjacent sets. Considering the unordered nature of feature sets, we further introduce set permutation regularization, which produces multiple segmentations of action sequences with the same semantics. Our code is available at: https://github.com/google-deepmind/action_piece.

In recommendation systems, the traditional multi-stage paradigm, which includes retrieval and ranking, often suffers from information loss between stages and diminishes performance. Recent advances in generative models, inspired by natural language processing, suggest the potential for unifying these stages to mitigate such loss. This paper presents the Unified Generative Recommendation Framework (UniGRF), a novel approach that integrates retrieval and ranking into a single generative model. By treating both stages as sequence generation tasks, UniGRF enables sufficient information sharing without additional computational costs, while remaining model-agnostic. To enhance inter-stage collaboration, UniGRF introduces a ranking-driven enhancer module that leverages the precision of the ranking stage to refine retrieval processes, creating an enhancement loop. Besides, a gradient-guided adaptive weighter is incorporated to dynamically balance the optimization of retrieval and ranking, ensuring synchronized performance improvements. Extensive experiments demonstrate that UniGRF significantly outperforms existing models on benchmark datasets, confirming its effectiveness in facilitating information transfer. Ablation studies and further experiments reveal that UniGRF not only promotes efficient collaboration between stages but also achieves synchronized optimization. UniGRF provides an effective, scalable, and compatible framework for generative recommendation systems.

No abstract available

Generative recommendation (GR) has gained increasing attention for its promising performance compared to traditional models. A key factor contributing to the success of GR is the semantic ID (SID), which converts continuous semantic representations (e.g., from large language models) into discrete ID sequences. However, varied modeling techniques, hyper-parameters, and experimental setups in existing literature make direct comparisons between GR proposals challenging. Furthermore, the absence of an open-source, unified framework hinders systematic benchmarking and extension, slowing model iteration. To address this challenge, our work introduces and open-sources a framework for Generative Recommendation with semantic ID, namely GRID, specifically designed for modularity to facilitate easy component swapping and accelerate idea iteration. Using GRID, we systematically experiment with and ablate different components of GR models with SIDs on public benchmarks. Our comprehensive experiments with GRID reveal that many overlooked architectural components in GR models with SIDs substantially impact performance. This offers both novel insights and validates the utility of an open-source platform for robust benchmarking and GR research advancement. GRID is open-sourced at https://github.com/snap-research/GRID.

Point-of-interest (POI) recommendation systems aim to predict the next destinations of user based on their preferences and historical check-ins. Existing generative POI recommendation methods usually employ random numeric IDs for POIs, limiting the ability to model semantic relationships between similar locations. In this paper, we propose Generative Next POI Recommendation with Semantic ID (GNPR-SID), an LLM-based POI recommendation model with a novel semantic POI ID (SID) representation method that enhances the semantic understanding of POI modeling. There are two key components in our GNPR-SID: (1) a Semantic ID Construction module that generates semantically rich POI IDs based on semantic and collaborative features, and (2) a Generative POI Recommendation module that fine-tunes LLMs to predict the next POI using these semantic IDs. By incorporating user interaction patterns and POI semantic features into the semantic ID generation, our method improves the recommendation accuracy and generalization of the model. To construct semantically related SIDs, we propose a POI quantization method based on residual quantized variational autoencoder, which maps POIs into a discrete semantic space. We also propose a diversity loss to ensure that SIDs are uniformly distributed across the semantic space. Extensive experiments on three benchmark datasets demonstrate that GNPR-SID substantially outperforms state-of-the-art methods, achieving up to 16% improvement in recommendation accuracy.

Sequential Recommender Systems (SRSs) aim to predict the next item that users will consume, by modeling the user interests within their item sequences. While most existing SRSs focus on a single type of user behavior, only a few pay attention to multi-behavior sequences, although they are very common in real-world scenarios. It is challenging to effectively capture the user interests within multi-behavior sequences, because the information about user interests is entangled throughout the sequences in complex relationships. To this end, we first address the characteristics of multi-behavior sequences that should be considered in SRSs, and then propose novel methods for Dynamic Multi-behavior Sequence modeling named DyMuS, which is a light version, and DyMuS+, which is an improved version, considering the characteristics. DyMuS first encodes each behavior sequence independently, and then combines the encoded sequences using dynamic routing, which dynamically integrates information required in the final result from among many candidates, based on correlations between the sequences. DyMuS+, furthermore, applies the dynamic routing even to encoding each behavior sequence to further capture the correlations at item-level. Moreover, we release a new, large and up-to-date dataset for multi-behavior recommendation. Our experiments on DyMuS and DyMuS+ show their superiority and the significance of capturing the characteristics of multi-behavior sequences.

Large Language Model-based generative recommendation (LLMRec) has achieved notable success, but it suffers from high inference latency due to massive computational overhead and memory pressure of KV Cache. Existing KV Cache reduction methods face critical limitations: cache compression offers marginal acceleration given recommendation tasks' short decoding steps, while prompt compression risks discarding vital interaction history. Through systematic analysis of attention patterns in LLMRec, we uncover two pivotal insights: 1) layer-wise attention sparsity inversion where early layers retain dense informative patterns while later layers exhibit high redundancy, and 2) dual attention sinks phenomenon where attention scores concentrate on both head and tail tokens of input sequences. Motivated by these insights, we propose EARN, an efficient inference framework that leverages the early layers to compress information into register tokens placed at the input sequence boundaries, then focuses solely on these tokens in the subsequent layers. Extensive experiments on three datasets, two LLMRec methods and two LLM architectures demonstrate EARN's superiority, achieving up to 3.79x speedup and 80.8% KV Cache reduction with better accuracy than the general finetuning approach. Our work bridges the efficiency-effectiveness gap in LLMRec, offering practical deployment advantages for industrial scenarios.

Developing a universal model that can effectively harness heterogeneous resources and respond to a wide range of personalized needs has been a longstanding community aspiration. Our daily choices, especially in domains like fashion and retail, are substantially shaped by multi-modal data, such as pictures and textual descriptions. These modalities not only offer intuitive guidance but also cater to personalized user preferences. However, the predominant personalization approaches mainly focus on the ID or text-based recommendation problem, failing to comprehend the information spanning various tasks or modalities. In this paper, our goal is to establish a Unified paradigm for Multi-modal Personalization systems (UniMP), which effectively leverages multi-modal data while eliminating the complexities associated with task- and modality-specific customization. We argue that the advancements in foundational generative modeling have provided the flexibility and effectiveness necessary to achieve the objective. In light of this, we develop a generic and extensible personalization generative framework, that can handle a wide range of personalized needs including item recommendation, product search, preference prediction, explanation generation, and further user-guided image generation. Our methodology enhances the capabilities of foundational language models for personalized tasks by seamlessly ingesting interleaved cross-modal user history information, ensuring a more precise and customized experience for users. To train and evaluate the proposed multi-modal personalized tasks, we also introduce a novel and comprehensive benchmark covering a variety of user requirements. Our experiments on the real-world benchmark showcase the model's potential, outperforming competitive methods specialized for each task.

Generative recommendation aims to learn the underlying generative process over the entire item set to produce recommendations for users. Although it leverages non-linear probabilistic models to surpass the limited modeling capacity of linear factor models, it is often constrained by a trade-off between representation ability and tractability. With the rise of a new generation of generative methods based on pre-trained language models (LMs), incorporating LMs into general recommendation with implicit feedback has gained considerable attention. However, adapting them to generative recommendation remains challenging. The core reason lies in the mismatch between the input-output formats and semantics of generative models and LMs, making it challenging to achieve optimal alignment in the feature space. This work addresses this issue by proposing a model-agnostic generative recommendation framework called DMRec, which introduces a probabilistic meta-network to bridge the outputs of LMs with user interactions, thereby enabling an equivalent probabilistic modeling process. Subsequently, we design three cross-space distribution matching processes aimed at maximizing shared information while preserving the unique semantics of each space and filtering out irrelevant information. We apply DMRec to three different types of generative recommendation methods and conduct extensive experiments on three public datasets. The experimental results demonstrate that DMRec can effectively enhance the recommendation performance of these generative models, and it shows significant advantages over mainstream LM-enhanced recommendation methods.

Scaling law has recently been validated in the recommendation system, adopting generative recommendation strategies to achieve scalability. However, these generative approaches require abandoning the meticulously constructed cross features of traditional recommendation models,leading to a significant decline in model performance. To address this challenge, we propose Meituan Generative Recommendation, which is based on the HSTU architecture and is capable of retaining the original deep learning recommendation model (DLRM) features, including cross features. Additionally, MTGR achieves training and inference acceleration through user-level compression to ensure efficient scaling. We also propose Group-Layer Normalization (GLN) to enhance the performance of encoding within different semantic spaces and the dynamic masking strategy to avoid information leakage. We further optimize the training frameworks, enabling support for our models with 10 to 100 times computational complexity compared to the DLRM, without significant cost increases. MTGR achieved 65x FLOPs for single-sample forward inference compared to the DLRM model, resulting in the largest gain in nearly two years both offline and online. This breakthrough was successfully deployed on Meituan, the world's largest food delivery platform, where it has been handling the main traffic.

The task of multi-behavioral sequential recommendation (MBSR) has grown in importance in personalized recommender systems, aiming to incorporate behavior types of interactions for better recommendations. Existing approaches focus on the next-item prediction objective, neglecting the value of integrating the target behavior type into the learning objective. In this paper, we propose MBGen, a novel Multi-Behavioral sequential Generative recommendation framework. We model the MBSR task into a consecutive two-step process: (1) given item sequences, MBGen first predicts the next behavior type to frame the user intention, (2) given item sequences and a target behavior type, MBGen then predicts the next items. To model such a two-step process, we tokenize both behaviors and items into tokens and construct one single token sequence with both behaviors and items placed interleaved. Furthermore, we design a unified generative recommendation paradigm that learns to autoregressive generate next behavior and item tokens, naturally enabling a multi-task capability. Additionally, we exploit the heterogeneous nature of token sequences in the generative recommendation and propose a position-routed sparse architecture to efficiently scale up models under the generative recommendation paradigm. Extensive experiments on real-world public datasets demonstrate that MBGen significantly outperforms existing MBSR models across multiple tasks.

Generative recommendation models often struggle with two key challenges: (1) the superficial integration of collaborative signals, and (2) the decoupled fusion of multimodal features. These limitations hinder the creation of a truly holistic item representation. To overcome this, we propose CEMG, a novel Collaborative-Enhaned Multimodal Generative Recommendation framework. Our approach features a Multimodal Fusion Layer that dynamically integrates visual and textual features under the guidance of collaborative signals. Subsequently, a Unified Modality Tokenization stage employs a Residual Quantization VAE (RQ-VAE) to convert this fused representation into discrete semantic codes. Finally, in the End-to-End Generative Recommendation stage, a large language model is fine-tuned to autoregressively generate these item codes. Extensive experiments demonstrate that CEMG significantly outperforms state-of-the-art baselines.

Embedding-based retrieval serves as a dominant approach to candidate item matching for industrial recommender systems. With the success of generative AI, generative retrieval has recently emerged as a new retrieval paradigm for recommendation, which casts item retrieval as a generation problem. Its model consists of two stages: semantic tokenization and autoregressive generation. The first stage involves item tokenization that constructs discrete semantic tokens to index items, while the second stage autoregressively generates semantic tokens of candidate items. Therefore, semantic tokenization serves as a crucial preliminary step for training generative recommendation models. Existing research usually employs a vector quantizier with reconstruction loss (e.g., RQ-VAE) to obtain semantic tokens of items, but this method fails to capture the essential neighborhood relationships that are vital for effective item modeling in recommender systems. In this paper, we propose a contrastive quantization-based semantic tokenization approach, named CoST, which harnesses both item relationships and semantic information to learn semantic tokens. Our experimental results highlight the significant impact of semantic tokenization on generative recommendation performance, with CoST achieving up to a 43% improvement in Recall@5 and 44% improvement in NDCG@5 on the MIND dataset over previous baselines.

Generative recommendation systems have recently seen a surge in interest, largely due to the promising advancements in generative AI. As a competitive solution for multi-behavior sequence recommendations, much of the recent research has concentrated on predicting the next item a user will likely interact with using a generative approach. However, these methods often 1). assign multiple residual quantization layers to obtain item codes, which leads to extra storage costs of more codebooks. And 2). explicitly utilize behavior sequences leading to longer sequences, potentially increasing the training time as well as inference time compared with original sequences. In response to these challenges, we introduce the Implicit Multi-Behavior Generative recommendation with a mixture of quantization (IMBGen) approach in this paper. Specifically, we have devised a Mixture of Quantization (MoQ) that combines the merits of both residual and parallel quantization for a more effective tokenization process. Additionally, we propose an Implicit Behavior Modeling (IBM) framework, allowing for more efficient integration of users’ behaviors into the interacted items. Finally, we conducted extensive experiments on two widely used benchmark datasets and further confirmed our findings with an online A/B test. The results consistently demonstrate the advantages of our approach over other baseline methods.

Generative Recommendation (GR) has emerged as a new paradigm in recommender systems. This approach relies on quantized representations to discretize item features, modeling users’ historical interactions as sequences of discrete tokens. Based on these tokenized sequences, GR predicts the next item by employing next-token prediction methods. The challenges of GR lie in constructing high-quality semantic identifiers (IDs) that are hierarchically organized, minimally conflicting, and conducive to effective generative model training. However, current approaches remain limited in their ability to harness multimodal information and to capture the deep and intricate interactions among diverse modalities, both of which are essential for learning high-quality semantic IDs and for effectively training GR models. To address this, we propose Multi-Aspect Cross-modal quantization for generative Recommendation (MACRec), which introduces multimodal information and incorporates it into both semantic ID learning and generative model training from different aspects. Specifically, we first introduce cross-modal quantization during the ID learning process, which effectively reduces conflict rates and thus improves codebook usability through the complementary integration of multimodal information. In addition, to further enhance the generative ability of our GR model, we incorporate multi-aspect cross-modal alignments, including the implicit and explicit alignments. Finally, we conduct extensive experiments on three well-known recommendation datasets to demonstrate the effectiveness of our proposed method.

Recommender systems based on collaborative filtering (CF) effectively model user-item interactions but struggle with data sparsity and cold-start issues. On the other hand, large language models (LLMs) offer strong semantic understanding but fail to capture structured user-item relationships. To bridge this gap, we propose AlignGenRec, a framework that integrates collaborative and textual knowledge for generative recommendation. Specifically, we introduce an embedding alignment mechanism that aligns item embeddings from a pre-trained CF model with text embeddings from item descriptions. These aligned embeddings are then transferred to the LLM without requiring fine-tuning, enabling structured knowledge integration while preserving generative capabilities. Additionally, constrained sequence decoding ensures that generated recommendations correspond to valid items, improving recommendation accuracy. Experimental results demonstrate that AlignGenRec outperforms both CF-based and LLM-based baselines, particularly in cold-start scenarios, with performance improvements of 7 - 11% across three benchmark datasets. Beyond recommendation tasks, AlignGenRec also supports preference prediction and user profiling, highlighting its versatility in real-world applications.

Next basket recommendation aims at predicting the next set of items that a user would likely purchase together, which plays an important role in e-commerce platforms. Unlike conventional item recommendation, the next basket recommendation focuses on capturing item correlations among baskets and learning the user’s temporal interest from the past purchasing basket sequence. In practice, most users interact with items in various kinds of behaviors. The multi-behavior data sheds light on user’s potential purchasing intention and resolves noisy signals from accidentally purchased items. In this article, we conduct an empirical study on real datasets to exploit the characteristics of multi-behavior data and confirm its positive effects on next basket recommendation. We develop a novel Multi-Behavior Network (MBN) model that captures item correlations and acquires meta-knowledge from multi-behavior basket sequences effectively. MBN employs the meta multi-behavior sequence encoder to model temporal dependencies of each individual behavior and extract meta-knowledge across different behaviors. Furthermore, we design the recurring-item-aware predictor in MBN to realize the high degree of the repeated occurrences of items, leading to better recommendation performance. We conduct extensive experiments to evaluate the performance of our proposed MBN model using real-world multi-behavior data. The results demonstrate the superior recommendation performance of MBN compared with various state-of-the-art methods.

No abstract available

The task of point-of-interest (POI) recommendation aims to recommend locations to users in location-based applications. Among them, the task of time-aware POI recommendation aims to capture the user’s preferences that change dynamically over time, so as to make more accurate recommendations to users at a specific time. While existing works take into account the spatial, temporal and category context of POIs, they cannot capture user preferences that are more fine-grained than the category granularity. Additionally, RNN-based methods suffer from the problem of long-term dependency when capturing a user’s check-in patterns. To address these challenges, we propose a novel model with POI multi-grained grouping method which captures the user’s co-visit patterns and weekly patterns, to obtain finer-grained POI groups. The model also utilizes the transformer model to capture the user’s check-in preference patterns. We evaluate our model on two real-world datasets, and the experimental results demonstrate the effectiveness of our proposed model.

No abstract available

Integrating geolocation recommendation engines into local libraries is an achievement that improves user interactivity, access, and resource utilization. This research examines the application of geolocation technologies and customized recommendation systems to suggest real-time books, resources, or events to patrons based on their locations inside or outside the library. Libraries can create responsive user experiences using data like borrowing history, preferences, and location which helps in the discovery of materials that are not highly sought after. Such technologies also enable patrons, including those with disabilities, to receive guided search aids to specific shelves or sections which makes complex libraries easier to navigate.Moreover, these systems can assist in advertising localized events such as workshops, book readings, or community gatherings to patrons within a determined range. This paper analyzes the applicable literature addressed with Bluetooth beacons, RFID, GPS, and mobile applications with machine learning models. The critique analyzes privacy, data security, infrastructure expenses, and proffered recommendations for socially responsible implementation.This study found that location-aware recommendation systems transform library services and meet the shifting demands of contemporary users, transforming libraries into interactive, agile, and integrative spaces. Such automated customization can greatly enhance the public's interest in libraries as enduring resources for learning within the community.

No abstract available

This study aims to develop an innovative recommendation system that provides a tailored café experience by integrating users’ nuanced preferences with a real-time environmental context. With the recent surge in domestic coffee consumption, the café market has reached a saturation point, leading consumers to seek spaces that match a specific atmosphere or purpose beyond merely consuming beverages. Existing recommendation systems, which primarily rely on past ratings or static information, have shown limitations in meeting these dynamic and multidimensional demands. To overcome these limitations, this paper proposes a novel framework that fuses hyper-personalization, context-aware recommendation, acoustic scene classification (ASC), passive crowd density estimation, and location-based augmented reality (AR). The proposed system utilizes a large language model (LLM) to extract abstract atmospheric characteristics such as “cozy,” “vibrant,” or “suitable for work” from unstructured text data collected from social media and review platforms. Simultaneously, it quantifies the actual environment of a space by analyzing real-time data collected through in-store sensors (or user devices). Specifically, ASC technology identifies the qualitative characteristics of sound – such as conversations, background music, and machine noise – going beyond simple noise levels. Passive detection of smartphone Wi-Fi probe signals accurately estimates indoor crowd density without infringing on personal privacy. This multi-modal data is combined with user profiles to generate a list of recommendations optimized for each individual. Finally, users can have an immersive exploration experience through a location-based AR interface, visually confirming recommended cafés, friends’ reviews, and personalized notes overlaid on their real-world surroundings. Through the synergistic combination of these advanced technologies, this research presents the design and implementation potential of a system that shifts the paradigm of physical space recommendation and provides a truly experience-centric service.

Extensive user check-in data incorporating user preferences for location is collected through Internet of Things (IoT) devices, including cell phones and other sensing devices in location-based social network. It can help traveling enterprises intelligently predict users' interests and preferences, provide them with scientific tourism paths, and increase the enterprises income. Thus, successive point-of-interest (POI) recommendation has become a hot research topic in augmented Intelligence of Things (AIoT). Presently, various methods have been applied to successive POI recommendations. Among them, the recurrent neural network-based approaches are committed to mining the sequence relationship between POIs, but ignore the high-order relationship between users and POIs. The graph neural network-based methods can capture the high-order connectivity, but it does not take the dynamic timeliness of POIs into account. Therefore, we propose an Interaction-enhanced and Time-aware Graph Convolution Network (ITGCN) for successive POI recommendation. Specifically, we design an improved graph convolution network for learning the dynamic representation of users and POIs. We also designed a self-attention aggregator to embed high-order connectivity into the node representation selectively. The enterprise management systems can predict the preferences of users, which is helpful for future planning and development. Finally, experimental results prove that ITGCN brings better results compared to the existing methods.

No abstract available

: Point-of-Interest (POI) recommendation is a fundamental service for guiding users to relevant venues among a large number of suggestions. In light of this, academia and industry have extensively studied influential factors to improve recommendation efficiency in Location-Based Social Networks (LBSN). While existing approaches leverage spatial distance, social relationships, and POI categories, they often insufficiently consider the dynamic nature of user behavior. This paper introduces Time-Aware Auto-SCS , an enhanced deep learning-based approach for POI recommendation. It serves as a significant extension to the original Auto-SCS framework by focusing on overcoming its static limitations. The overall architecture comprises three phases, namely, feature extraction, multi-aspect fusion, and recommendation generation. The first phase includes the spatial proximity and social trust steps from the Auto-SCS model, along with a new step called time-aware categorical behavior. This step is a novel Convolutional Neural Network (CNN)-based method that processes check-in sequences. It generates for each user a dense embedding that captures their time-sensitive preferences. The multi-aspect fusion phase utilizes a deep autoencoder. It combines the three embeddings from the previous outputs to generate the recommendation task. Extensive experiments on real-world LBSN datasets demonstrate that Time-Aware Auto-SCS outperforms the state-of-the-art baselines.

Integrating session frequency, temporal, and spatial data is critical for recommendation systems. This paper proposes a novel GNN-based session recommendation algorithm using an improved gated graph convolutional network to capture item relationships and evolving user interests. Firstly, a multidimensional information-aware session graph is constructed incorporating frequency, temporal, and spatial data. Secondly, item features from GNN layers are concatenated with multidimensional features, with weights calculated via a fully connected layer and Softmax. Residual connections add original item features to preliminary fused features for enhanced representations. The model uses cross-entropy loss and Adam optimizer for training, recommending top-K items with highest probabilities. Experimental comparisons with GCE-GNN, SEDGNN, and other algorithms on Diginetica, Yoochoose, and Nowlaying datasets validate the superiority of the proposed FTLAGNN model.

As the popularity of online travel platforms increases, users tend to make ad-hoc decisions on places to visit rather than preparing the detailed tour plans in advance. Under the situation of timeliness and uncertainty of users’ demand, how to integrate real-time context into dynamic and personalized recommendations have become a key issue in travel recommender system. In this article, by integrating the users’ historical preferences and real-time context, a location-aware recommender system called TRACE ( T ravel R einforcement Recommendations Based on Location- A ware C ontext E xtraction) is proposed. It captures users’ features based on location-aware context learning model, and makes dynamic recommendations based on reinforcement learning. Specifically, this research: (1) designs a travel reinforcing recommender system based on an Actor-Critic framework, which can dynamically track the user preference shifts and optimize the recommender system performance; (2) proposes a location-aware context learning model, which aims at extracting user context from real-time location and then calculating the impacts of nearby attractions on users’ preferences; and (3) conducts both offline and online experiments. Our proposed model achieves the best performance in both of the two experiments, which demonstrates that tracking the users’ preference shifts based on real-time location is valuable for improving the recommendation results.

Motivated by the collaboration with Fliggy1, a leading Online Travel Platform (OTP), we investigate an important but less explored research topic about optimizing the quality of hotel supply, namely selecting potential profitable hotels in advance to build up adequate room inventory. We formulate a WWW problem, i.e., within a specific time period (When) and potential travel area (Where), which hotels should be recommended to a certain group of users with similar travel intentions (Why). We identify three critical challenges in solving the WWW problem: user groups generation, travel data sparsity and utilization of hotel recommendation information (e.g., period, location and intention). To this end, we propose LINet, a Location and Intention-aware neural Network for hotel group recommendation. Specifically, LINet first identifies user travel intentions for user groups generalization, and then characterizes the group preferences by jointly considering historical user-hotel interaction and spatio-temporal features of hotels. For data sparsity, we develop a graph neural network, which employs long-term data, and further design an auxiliary loss function of location that efficiently exploits data within the same and across different locations. Both offline and online experiments demonstrate the effectiveness of LINet when compared with state-of-the-art methods. LINet has been successfully deployed on Fliggy to retrieve high quality hotels for business development, serving hundreds of hotel operation scenarios and thousands of hotel operators.

No abstract available

No abstract available

Mobile edge computing (MEC) is important in location-based social networks (LBSNs). It puts services near users to cut delays. Edge service recommendation needs to mix context details with user privacy. Data sparsity makes this hard. Traditional methods have trouble with little data. They miss small context details or hurt privacy with central systems. This paper introduces FedCAFE, a federated learning system for edge service recommendation with context awareness. FedCAFE uses three main parts. It has a denoising autoencoder to get strong user and service features from small data. This tool learns patterns by fixing noisy information. It helps when user-service interactions are few. FedCAFE also uses a new adaptive fuzzy clustering method to group users and services by context matches. This part looks at things like time and place. It changes how it groups based on different situations. FedCAFE applies federated learning to keep privacy safe. It trains on user devices. It sends only model updates, not personal data. This stops private stuff like location from leaving the device. We tested FedCAFE on the WSDream dataset with real service information. FedCAFE beats other methods in these tests.

POI (point-of-interest) recommendation as an important type of location-based services has received increasing attention with the rise of location-based social networks. Although significant efforts have been dedicated to learning and recommending users' next POIs based on their historical mobility traces, there still lacks consideration of the discrepancy of users' check-in time preferences and the inherent relationships between POIs and check-in times. To fill this gap, this paper proposes a novel recommendation method which applies multi-task learning over historical user mobility traces known to be sparse. Specifically, we design a cross-graph neural network to obtain time-aware user modeling and control how much information flows across different semantic spaces, which makes up the inadequate representation of existing user modeling methods. In addition, we design a check-in time prediction task to learn users' activities from a time perspective and learn internal patterns between POIs and their check-in times, aiming to reduce the search space to overcome the data sparsity problem. Comprehensive experiments on two real-world public datasets demonstrate that our proposed method outperforms several representative POI recommendation methods with 8.93% to 20.21 % improvement on Recall@1, 5, 10, and 9.25% to 17.56% improvement on Mean Reciprocal Rank.

No abstract available

Location based social network develops and gets widely concern along with the population and widespread use of mobile. Point of interest(POI) recommendation become one of the most widely application among location-based service. To get better POI recommendation performance, a fuzzy clustering based collaborative filtering algorithm (FCCF) for time-aware POI recommendation is proposed in this paper. It first constructs the user feature vector from users’ check-in behaviours. Individual’s check-in behaviour can be under the influence of location region and time slots, so user’s feature consists of two parts. One is the vising frequency of each user in different location regions, and the other is the vising frequency of each user in different time slots. Next fuzzy c-means is adopted due to its simplicity to group users according to user feature vector. Then the user similarity computation can be limited in the similar small user groups. In the end, a collaborative filtering algorithm is applied to recommend a number of top-N POIs at a given time for the target user. Some experiments are conducted and the comparative results on Foursquare and Gowalla show that FCCF has higher precision and recall value than the comparative algorithms.

ABSTRACT The rising prosperity of Location-based Social Networks (LBSNs) witnessed an explosion in the availability of geo-tagged social media data, which enables tremendous location-aware online services, especially next point of interest (POI) recommendation. However, previous next POI recommendation studies usually adopt fix-length time windows for user check-in sequence modeling, leading to a limited capacity in capturing fine-grained user temporal preferences that easily change over time. Besides, existing methods often directly leverage multi-modal contexts as auxiliary to alleviate the data sparsity issue, which fails to fully exploit the sequential patterns of contextual information for inferring user interest drift. To address the above challenges, we propose a novel framework named iTourSPOT which extends traditional collaborative filtering methods with a context-aware POI embedding architecture. For enhancing temporal interests modeling capacity, we associate the context feature extraction with varying-length sessions and incorporate check-in frequencies of POIs as prior knowledge to instruct the session representation learning of our model. Moreover, a collaborative sequence transduction model is designed for joint context sequence modeling and session-based POI recommendation. Experimental results on a real-world geo-tagged photo dataset clearly demonstrate the effectiveness of the proposed framework when compared with state-of-the-art baseline methods, especially in both sparse and cold-start scenarios.

Providing real-time product recommendations based on consumer profiles and purchase history is a successful marketing strategy in online retailing. However, brick-and-mortar (BAM) retailers have yet to utilize this important promotional strategy because it is difficult to predict consumer preferences as they travel in a physical space but remain anonymous and unidentifiable until checkout. In this paper, we develop such a recommender approach by leveraging the consumer shopping path information generated by radio frequency identification technologies. The system relies on spatial-temporal pattern discovery that measures the similarity between paths and recommends products based on measured similarity. We use a real-world retail data set to demonstrate the feasibility of this real-time recommender system and show that our approach outperforms benchmark methods in key recommendation metrics. Conceptually, this research provides generalizable insights on the correlation between spatial movement and consumer preference. It makes a strong case that the emerging location and path data and the spatial-temporal pattern discovery methods can be effectively utilized for implementable marketing strategies. Managerially, it provides one of the first real-time recommender systems for BAM retailers. Our approach can potentially become the core of the next-generation intelligent shopping environment in which the stores customize marketing efforts to provide real-time, location-aware recommendations.

This paper proposes a unique approach to emotion recognition and employs Vision Transformer (ViT) model for the sake of giving personalised suggestions of music selections and environments. ViT is able to determine the user emotion based on his facial expression detection technology. Those emotions will provide the critical inputs that will allow recommendations to be modified to context-aware standards the mood you're in the time of year, the specifics of the month. Proprietary technology creates music playlists suitable to the mood of the moment as it suggests the best locations for experiences based on an emotional context and other external factors such as the season and time of month. As we move deeper into the future, this merging of emotion detection technology and customized recommendations can deliver an unparalleled user experience by offering a user exactly what they would like in their location of the area they are you engaged in. It just ensures that the location to suggest should not only be ideal to the user mood but must also be practical as well by taking into account the locations weather and local events. The whole concept of this is to create a more rounded and complete approach to curating a fitness routine tailored to the individual, taking into account their mental health as well as external issues.

Recommender systems have become essential in large-scale e-commerce and content platforms. While user preferences are crucial in generating recommendations, the context in which recommendations are made—such as time, location, and occasion—also plays a significant role. Over the past decades, context-aware models have been developed to address this. However, the use of context information in negative sampling remains less expored, despite the well-known impact of sampling strategies on recommendation performance. In this study, we first quantitatively demonstrate that item distributions vary significantly across different contexts. Based on this observation, we propose two novel negative sampling methods: context-aware hard negative item sampling and negative context sampling. These methods enhance recommendation diversity by more accurately reflecting context-dependent item distributions during training. To validate the effectiveness of the proposed methods, we apply them to a sequential recommendation model leveraging temporal context, where time information plays a critical role. The proposed methods are designed to be easily integrated into any sampling-based sequential recommendation model in a plug-and-play manner, and extensive experimental results on 10 real-world datasets demonstrate that the proposed methods significantly improve recommendation diversity, with an average increase of 20.65%, while incurring only a minor accuracy loss of 1.89%.

Smartphone services are growing rapidly in number and complexity, making traditional graphical user interfaces increasingly inefficient for service access. To improve accessibility, we propose Nexa, a unified, self-evolving context-aware assistant that integrates multimodal user input (such as voice commands, camera input, screen content and motion gestures) and contextual information (such as time, location, network and notification) to recommend and execute smartphone services. For service recommendation, the assistant employs a dual-layered prediction module: a rule-based System 1 for fast matching, complemented by an LLM-based System 2 that performs reasoning when no rules apply. The system continuously learns from user interactions; when recommendations fail, it logs the user's subsequent GUI interactions and updates its prediction model by analyzing causal relationships among multimodal inputs, context, and the final chosen service. Additionally, macro scripts are extracted from logged GUI traces and added to the service list. We implement a prototype on Android with a cloud backend. We assume that Nexa can reduce the effort required to access services and adapt to individual preferences over time, thereby enhancing user experience.

With the increasing complexity of user behaviour and the rising cost of customer acquisition, digital platforms face significant challenges in sustaining long-term user engagement—particularly during the early stages marked by cold-start conditions. Traditional churn prediction models often fall short in providing actionable strategies for personalized retention, necessitating more adaptive and user-centric solutions. This study proposes PRISM (Personalized Retention-Integrated Strategy Model), a modular architecture designed to bridge behavioural prediction with intelligent task recommendation, ensuring both immediate engagement and sustained user retention. PRISM integrates several core modules: the Retention-Oriented Influence Model (ROIM) captures dynamic social propagation patterns; the Retention-Aware Engagement Model (RAEM) evaluates contextual factors such as location, time, reward relevance, and user interest to estimate task acceptance; the Fuzzy Retention Prediction Model (FRPM) leverages fuzzy logic to interpret engagement stimuli; and the Retention-Oriented Behaviour Estimation (ROBE) forecasts user interaction trends. These components work cohesively within the Personalized Fuzzy Engagement Recommendation (PFER) framework to allocate tasks tailored for maximum retention impact. The proposed system is validated across three benchmark datasets—IBM Telco, iQIYI, and MovieLens—using comprehensive evaluation metrics including BLEU, ROUGE, NDCG@10, HR@10 for prediction accuracy, and MB-URS, SB-URS, IUR, NRC for retention performance. Experimental results demonstrate that PRISM consistently surpasses state-of-the-art baselines, establishing a robust, explainable, and domain-neutral strategy for retention-oriented task recommendation and user engagement.

Abstract Next point-of-interest (POI) recommendation has gained growing attention in recent years due to the emergence of location-based social networks (LBSN) services. Most existing approaches focus on learning user’s preferences to POIs from check-in records and recommend a POI to visit next given his/her previously visited POIs. However, the user’s visiting behavior is not only driven by user preferences in real-world scenarios. The real-time demand is another crucial factor to determine the user’s visiting behaviors, which is usually neglected in established approaches. In this paper, we propose a new next point-of-interest (POI) recommendation method, called DSPR, by exploring user’s preferences and real-time demand simultaneously. To model the real-time demand, different kinds of contextual information are exploited, such as absolute time, POI–POI transition time/distance, and the types of POIs. By incorporating user’s preferences, these contextual factors are further modeled and learned automatically with an attention-based recurrent neural network model to support the final next POI recommendation. Experiments on three real-world check-in datasets show that DSPR has better recommendation performance compared with many state-of-the-art methods.

Compared to the traditional recommender systems, context-aware recommender systems are more in line with actual application contexts. However, the existing researches are mostly focused on single context-aware recommendation, such as time-aware recommendation or location-aware recommendation, and lack of in-depth research on multi-context-aware recommendation. Therefore, we proposed a recommendation method of high-order tensor factorization based on multi-context-aware. First, on the basis of analyzing the influence of context on users’ interest preferences, the sensitivity of users to multiple contexts was detected using statistical methods. For context-sensitive users, four-dimensional tensors and feature matrices used to solve data sparsity were constructed based on rating matrix and situational information. And then the stochastic gradient descent algorithm was used for iterative calculation to fill in missing data values and carry out parameter optimization. For context-insensitive users, we used matrix factorization to predict users’ interest preferences. Finally, we tested and validated our method on a multi-context-aware movie dataset, and the experimental results show that the proposed method could effectively reduce the prediction error and improve the recommendation quality.

Recommender Systems (RSs) are a subclass of information filtering systems. RSs assist users in choosing interesting items from an extensive collection of items. This article addresses two research topics in RS, namely cross-domain RSs (CDRSs) and the context-aware RSs (CARSs). CDRSs were developed to improve the quality of recommendations in a target domain using the source domain information. Moreover, CDRSs look to limit the spread of fake information through RSs. CARSs are designed to utilize contextual information, such as location, time, companions, and others, in the recommendation as user interests change with context. In this work, CDRSs and CARSs are implemented in an integrated manner to construct a more specific RS that offers both these systems’ advantages. For including contextual information in data, contextual prefiltering is applied. These approaches recommend items more accurately, overcoming cold start, sparsity, and scalability issues, and provide a more personalized, novel, and diversified recommendation. The developed system, cross-domain recommendation using context-aware sequences (CDRec-CAS), is evaluated in terms of accuracy achieved in recommending preferred item sequences and the next preferred item. In recommending preferred item sequences, it is found that it improves recommendation accuracy that varied from approximately 7.85%–9.74% (considering the single context) and 4.41%–8.17% (considering dual-context) when compared with existing noncontextual RS. In recommending the next preferred item, it is found that it improves recommendation accuracy that varied from approximately 3.81%–9.81% (considering the single context) −2.24%–9.21% (considering dual-context) when compared with existing noncontextual RS. The results obtained by implementing CDRec-CAS are compared with existing approaches, proving that recommendations can be enhanced using cross-domain and contextual information.

An efficient data management system is essential in the field of student housing to guarantee the availability of high-quality data that can make it easier to find appropriate hostels. Although recommender systems concerning hotels and tourism points of interest (POI) have been extensively researched, little attention has been paid to students housing in Higher Education Institutions (HEIs). In this paper, we present GeoHostel, a novel location-based recommender system that combines real-time hostel availability data with Geographic Information System (GIS) technology. In contrast to traditional methods, GeoHostel is customized to meet the housing needs of Ghanaian students by fusing context-aware recommendation filtering with spatial features (distance to campus, safety, accessibility) to enhance decision-making. Evaluation metrics involving Precision, Recall, F-measure, MAE, and NMAE were used to compare GeoHostel with contemporary baseline methods to verify its efficacy through scientific benchmarking experiments. Evaluation results show that GeoHostel not only performs better than previous approaches but also offers a scalable, domain-specific, and socially significant solution for student housing in Ghana.

When users interact with Recommender Systems (RecSys), current situations, such as time, location, and environment, significantly influence their preferences. Situations serve as the background for interactions, where relationships between users and items evolve with situation changes. However, existing RecSys treat situations, users, and items on the same level. They can only model the relations between situations and users/items respectively, rather than the dynamic impact of situations on user-item associations (i.e., user preferences). In this paper, we provide a new perspective that takes situations as the preconditions for users' interactions. This perspective allows us to separate situations from user/item representations, and capture situations' influences over the user-item relationship, offering a more comprehensive understanding of situations. Based on it, we propose a novel Situation-Aware Recommender Enhancer (SARE), a pluggable module to integrate situations into various existing RecSys. Since users' perception of situations and situations' impact on preferences are both personalized, SARE includes a Personalized Situation Fusion (PSF) and a User-Conditioned Preference Encoder (UCPE) to model the perception and impact of situations, respectively. We conduct experiments of applying SARE on seven backbones in various settings on two real-world datasets. Experimental results indicate that SARE improves the recommendation performances significantly compared with backbones and SOTA situation-aware baselines.

In the digital commerce era, the significance of customized promotions for individual users has reached new heights. This research paper introduces an innovative application designed to transform personalized marketing strategies by incorporating a Location-Based Offer Recommendation System (LBORS). The proposed system not only leverages user location data to dynamically recommend offers from nearby shops but also ensures that customers receive discounts, establishing a win-win situation for both customers and marketers. This approach aims to create a seamless and context-aware shopping experience, enhancing the overall value proposition for users. The system employs a sophisticated algorithm that combines geospatial analytics, user preferences, and historical data to accurately identify and present relevant offers to users based on their current geographical location. By integrating real-time location information, the application ensures that users receive timely and contextually appropriate promotions, thereby maximizing the likelihood of engagement and conversion.

Trip planning, which targets at planning a trip consisting of several ordered Points of Interest (POIs) under user-provided constraints, has long been treated as an important application for location-based services. The goal of trip planning is to maximize the chance that the users will follow the planned trip while it is difficult to directly quantify and optimize the chance. Conventional methods either leverage statistical analysis to rank POIs to form a trip or generate trips following pre-defined objectives based on constraint programming to bypass such a problem. However, these methods may fail to reflect the complex latent patterns hidden in the human mobility data. On the other hand, though there are a few deep learning-based trip recommendation methods, these methods still cannot handle the time budget constraint so far. To this end, we propose a TIme-aware Neural Trip Planning (TINT) framework to tackle the above challenges. First of all, we devise a novel attention-based encoder-decoder trip generator that can learn the correlations among POIs and generate trips under given constraints. Then, we propose a specially-designed reinforcement learning (RL) paradigm to directly optimize the objective to obtain an optimal trip generator. For this purpose, we introduce a discriminator, which distinguishes the generated trips from real-life trips taken by users, to provide reward signals to optimize the generator. Subsequently, to ensure the feedback from the discriminator is always instructive, we integrate an adversarial learning strategy into the RL paradigm to update the trip generator and the discriminator alternately. Moreover, we devise a novel pre-training schema to speed up the convergence for an efficient training process. Extensive experiments on four real-world datasets validate the effectiveness and efficiency of our framework, which shows that TINT could remarkably outperform the state-of-the-art baselines within short response time.

Next-POI recommendation aims to explore from user check-in sequence to predict the next possible location to be visited. Existing methods are often difficult to model the implicit association of multi-modal data with user choices. Moreover, traditional methods struggle to fully explore the variation of user preferences at variable time intervals. To tackle these limitations, we propose a Multi-Modal Temporal Knowledge Graph-aware Sub-graph Embedding approach (Mandari). We first construct a novel Multi-Modal Temporal Knowledge Graph. Based on the proposed knowledge graph, we integrate multi-modal information and leverage the graph attention network to calculate sub-graph prediction probability. Next, we implement a temporal knowledge mining method to model the segmentation and periodicity of user check-in and obtain temporal prediction probability. Finally, we fuse temporal prediction probability with the previous sub-graph prediction probability to obtain the final result. Extensive experiments demonstrate that our approach outperforms existing state-of-the-art methods.

Point-of-Interest (POI ) recommendation systems have gained popularity for their unique ability to suggest geographical destinations with the incorporation of contextual information such as time, location, and user-item interaction. Existing recommendation frameworks lack the contextual fusion required for POI systems. This paper presents CAPRI, a novel POI recommendation framework that effectively integrates context-aware models, such as GeoSoCa, LORE, and USG, and introduces a novel strategy for the efficient merging of contextual information. CAPRI integrates an evaluation module that expands the evaluation scope beyond accuracy to include novelty, personalization, diversity, and fairness. With an aim to establish a new industry standard for reproducible results in the realm of POI recommendation systems, we have made CAPRI openly accessible on GitHub, facilitating easy access and contribution to the continued development and refinement of this innovative framework.

As interest in eco-friendly transportation increases, the demand for electric vehicles (EVs) is also growing. With this trend, establishing an efficient charging infrastructure in urban environments has become essential. However, despite the presence of charging station infrastructure, its utilization remains low. In this study, we introduce a context-aware refining approach that incorporates location proximity and charging type preference into a Graph Neural Network (GNN)-based recommendation algorithm. For the EV charging station data, we collect and manage the large-scale real-time data. Additionally, to overcome the limitations posed by the absence or restricted access to driver profile data–an essential component for developing a recommendation system–we develop a simulator that analyzes and replicates real drivers’ charging patterns and behavioral characteristics. This enables us to generate realistic and reliable driver profile data. We employ the state-of-the-art GNN-based recommendation model for the base collaborative filtering. Then, we incorporate the user’s location proximity to the charging station and charging type preference (i.e., slow/fast) as context-aware refinement factors for the final recommendation results. To further enhance recommendation performance, we go beyond simple location proximity utilization by applying a clustering-based approach that reflects the actual spatial distribution, considering both charging stations and users. Through extensive experiments, we demonstrate that incorporating our contextual refinements consistently improves recommendation quality compared to the baseline GNN approach across different collaborative filtering models. In particular, leveraging charging type information and spatial clustering leads to substantial and stable performance gains, and their combined use yields the most robust results. These findings highlight the importance of jointly modeling functional preferences and geographical context for effective EV charging station recommendations.

As a class of context-aware systems, context-aware service recommendation (CASR) aims to bind high-quality services to users, w.r.t. their context requirements (e.g., invocation time, location, social profiles, connectivity). However, current CASR lacks a rich context modelling and does not allow for multi-relational interactions between users and services in different contexts. We propose a context-sensitive service recommendation, by constructing a contextual service knowledge graph (C-SKG), which we translated into a low-dimensional vector space to facilitate its processing. Dilated Recurrent Neural Networks are applied to allow a context-aware C-SKG embedding, based on the principles of subgraph-aware proximity. A recommendation algorithm, finally, returns the top-rated services w.r.t. the target user’s context and the proximity degrees.

Location-based recommendation has become a significant method to help people locate fascinating and appealing points of interest (POIs) with the rapid popularity of smart mobile devices and the prevalence of location-based social networks (LBSN). However, the sparsity of the user-POI matrix and the cold-start issue have generated serious challenges, resulting in a substantial decrease in collaborative filtering methods’ recommendation results. In reality, location-based recommendation demands spatiotemporal context awareness. In order to overcome these challenges, we develop an embedding model based on the heterogeneous graph attention network. Geographic influence, social relation and historical check-in influence are captured in a unified way by constructing a user-POI heterogeneous graph. Subsequently, we use the LSTM-based model to learn the category weight of the next POI to select. We are developing a score function to recommend the next POI for users by integrating category weights, user preferences and time impact. We conduct experiments on existing large-scale datasets to evaluate the performance of our model. The results demonstrate our proposal is superior to other rivals. Additionally, our method has been significantly improved compared with other competitive approaches in terms of recommending cold-start POI.

Over two decades, context awareness has been incorporated into recommender systems in order to provide, not only the top-rated items to consumers but also the ones that are suitable to the user context. As a class of context-aware systems, context-aware service recommendation (CASR) aims to bind high-quality services to users, while taking into account their context requirements, including invocation time, location, social profiles, connectivity, and so on. However, current CASR approaches are not scalable with the huge amount of service data (QoS and context information, users reviews and feedbacks). In addition, they lack a rich representation of contextual information, as they adopt a simple matrix view. Moreover, current CASR approaches adopt the traditional user-service relation and they do not allow for multi-relational interactions between users and services in different contexts. To offer a scalable and context-sensitive service recommendation with great analysis and learning capabilities, we provide a rich and multi-relational representation of the CASR knowledge, based on the concept of knowledge graph. The constructed context-aware service knowledge graph (C-SKG) is, then, transformed into a low-dimensional vector space to facilitate its processing. For this purpose, we adopt Dilated Recurrent Neural Networks to propose a context-aware knowledge graph embedding, based on the principles of first-order and subgraph-aware proximity. Finally, a recommendation algorithm is defined to deliver the top-rated services according to the target user's context. Experiments have proved the accuracy and scalability of our solution, compared to state-of-the-art CASR approaches.

No abstract available