时序知识图谱推理对于冷启动问题解决不充分

With the surge in Internet applications, personalized recommendation systems have become integral to platforms in e-commerce, social media, and digital content services. Despite significant advances, many sequence-based recommendation models still struggle with short-term interest modeling, cold-start robustness, and interpretability. This study proposes an innovative optimization method by integrating knowledge graphs with Long Short-Term Memory (LSTM) networks to address these challenges. By integrating the rich entity-level information in the knowledge graph, this model enhances the semantic representation of user behaviors. The sequence modeling based on LSTM effectively captures the temporal preferences of users, thereby improving the generalization ability of the model. Furthermore, to elevate model transparency, an attention mechanism is employed not merely to reweight inputs but to explicitly capture semantic alignments between users' evolving interests and candidate items. Experimental evaluations on multiple public datasets demonstrate that the proposed method consistently outperforms baseline models, particularly under cold-start scenarios, showcasing superior adaptability, accuracy, and interpretability.

No abstract available

Knowledge Graphs (KGs) have been incorporated into recommender systems as side information to solve the clas-sical data-sparsity and cold-start problems, with explanations for recommended items. Traditional embedding-based recommender systems generally utilize abundant information from KGs directly to enrich the representation of items or users, but the influences of spatial and temporal dependencies are usually ignored among them. In this paper, we propose a Spatio-Temporal Aware Knowledge Graph Embedding (STAKGE) for recommender systems, which incorporates spatio-temporal information with bias when propagating potential preferences of users in knowl-edge graph embedding. Moreover, we construct a multi-source KG-based recommender dataset - YelpST, containing spatio-temporal information. The experiments on YelpST dataset show that our proposed approach can capture comprehensive spatio-temporal correlations and improve the prediction performance as compared to various state-of-the-art baselines.

Temporal knowledge graph question answering (TKGQA) involves multi-hop reasoning over temporally constrained entity relationships in the knowledge graph to answer a given question. However, at each hop, large language models (LLMs) retrieve subgraphs with numerous temporally similar and semantically complex relations, increasing the risk of suboptimal decisions and error propagation. To address these challenges, we propose the multi-hop reasoning enhanced (MRE) framework, which enhances both forward and backward reasoning to improve the identification of globally optimal reasoning trajectories. Specifically, MRE begins with prompt engineering to guide LLM in generating diverse reasoning trajectories for the given question. Valid reasoning trajectories are then selected for supervised fine-tuning, serving as a cold-start strategy. Finally, we introduce Tree-Group Relative Policy Optimization (T-GRPO)—a recursive, tree-structured learning-by-exploration approach. At each hop, exploration establishes strong causal dependencies on the previous hop, while evaluation is informed by multi-path exploration feedback from subsequent hops. Experimental results on two TKGQA benchmarks indicate that the proposed MRE-based model consistently surpasses state-of-the-art (SOTA) approaches in handling complex multi-hop queries. Further analysis highlights improved interpretability and robustness to noisy temporal annotations.

Personalized recommendation method is an effective means to filter out the information users need from a large amount of information, which is rich in practical value. Personalized recommendation methods are maturing, and many e-commerce platforms have been using different forms of recommendation methods with great success. In the recommendation systems of large-scale e-commerce platforms, traditional recommendation algorithms represented by collaborative filtering are modeled only based on users' rating data, and sparse user-project interaction data and cold start are two inevitable problems. The introduction of knowledge graphs in recommendation systems can effectively solve these problems because of their rich knowledge content and powerful relationship processing capability. In this paper, we study the personalized recommendation algorithm based on knowledge graph as auxiliary information, and use the temporal information of user-item interaction in the graph to model users' interests over time at a finer granularity, taking into account the problem of high training time cost of the model due to frequent updates of the knowledge graph when recommending to users dynamically. The article proposes the Interactive Knowledge-Aware Attention Network Algorithmic Model for Recommendations (IKANAM) and conducts comparison experiments on public datasets. The results show that the IKANAM recommendation algorithm can effectively improve the effectiveness of personalized recommendation of products on Internet platforms.

No abstract available

No abstract available

Cross-domain sequential recommendation (CDSR) has emerged as an effective solution to address the challenges of data sparsity and cold-start issues in recommendation systems by leveraging shared knowledge across multiple domains. Nevertheless, existing methods face notable long-term challenges, such as ineffective knowledge transfer caused by distributional shifts and sparse overlapping users, insufficient modeling of temporal dynamics and intricate sequential patterns in user behavior, and suboptimal generalization across heterogeneous domains.To tackle these issues, we propose DiTAGInt, a novel diffusion-based generative network with augmented-graph embedding integration. DiTAGInt introduces a dynamic embedding fusion mechanism to harmonize domain-specific and shared-user representations, thereby enhancing generalization and alleviating rigid transfer constraints. Furthermore, it employs a diffusion-based generative module to effectively model temporal dynamics and capture complex sequential patterns in user behavior, facilitating precise user preference learning and significantly advancing recommendation accuracy. Extensive experiments conducted on three public datasets demonstrate the superiority of our method.

Graph neural networks (GNN) recently achieved huge success in collaborative filtering (CF) due to the useful graph structure information. However, users will continuously interact with items, which causes the user-item interaction graphs to change over time and well-trained GNN models to be out-of-date soon. Naive solutions such as periodic retraining lose important temporal information and are computationally expensive. Recent works that leverage recurrent neural networks to keep GNN up-to-date may suffer from the "catastrophic forgetting'' issue, and experience a cold start with new users and items. To this end, we propose the incremental graph convolutional network (IGCN) --- a pure graph convolutional network (GCN) based method to update GNN models when new user-item interactions are available. IGCN consists of two main components: 1) a historical feature generation layer, which generates the initial user/item embedding via model agnostic meta-learning and ensures good initial states and fast model adaptation; 2) a temporal feature learning layer, which first aggregates the features from local neighborhood to update the embedding of each user/item within each subgraph via graph convolutional network and then fuses the user/item embeddings from last subgraph and current subgraph via incremental temporal convolutional network. Experimental studies on real-world datasets show that IGCN can outperform state-of-the-art CF algorithms in sequential recommendation tasks.

With the rapid penetration of omni-channel retail scenarios, consumer behavior presents the complexity of dynamic interactions across channels. Traditional prediction models are difficult to capture the nonlinear temporal dependencies and dynamic decision-making feedback mechanisms between channels, resulting in fragmentation and static defects in behavior trajectory modeling. To this end, this paper proposes a hierarchical attention deep reinforcement learning model based on the proximal policy optimization (PPO) framework. First, we integrate multi-source heterogeneous data by building a knowledge graph of consumers' cross-channel behaviors. Second, we design a fusion architecture of the hierarchical attention mechanism and the PPO algorithm. Finally, we introduce a meta-reinforcement learning module to achieve rapid adaptation to the cold start scenario of new users. Experiments show that the proposed model has Top-1 and Top-3 accuracy rates of 77.5% and 90.3%, respectively in the behavior sequence prediction task; in the cold start adaptation experiment, the F1 value reaches 0.713, and the initial prediction accuracy is improved to 0.637; in the long-term strategy evaluation, the average cumulative reward is increased to 3.15, and the long-term return accounts for as much as 77%. The results verify the effectiveness and robustness of this model in complex behavior modeling, strategy migration, and long-term value guidance.

The rapid development of AI technology has addressed the issue of how to align academic curricula with the ever-changing demands of the industry. This has had a significant impact on vocational and technical education. This paper proposes an intelligent curriculum design framework to bridge the gap between academic courses and the ever-changing skill demands of the industry. Using artificial intelligence technology, the platform achieves dynamic and secure integration of industry education. The platform integrates a hybrid recommendation engine, LSTM-based demand forecasting, and privacy-preserving federated learning systems. The hybrid recommendation engine addresses the cold start problem and improves course matching accuracy through collaborative filtering and knowledge graph embedding. The demand forecasting engine achieves real-time course adjustments by analyzing temporal patterns in industry data to reduce delays. Federated learning can ensure that institutions collaborate securely without compromising data privacy. The VED Tech 2025 dataset, which includes 78,000 industry skill requirements and over 1.2 million student course records, shows that the platform's skill gap has decreased by 41%, course relevance has increased by 23.7%, and the median adaptation delay is 3.2 days. These findings highlight the platform's potential in enhancing graduates' employability, reducing corporate training costs, dynamically aligning academic services with industry needs, and providing a scalable, secure, and data-driven solution for integrating high-quality industry education.

The problem of community-level information pathway prediction (CLIPP) aims at predicting the transmission trajectory of content across online communities. A successful solution to CLIPP holds significance as it facilitates the distribution of valuable information to a larger audience and prevents the proliferation of misinfor- mation. Notably, solving CLIPP is non-trivial as inter-community relationships and influence are unknown, information spread is multi-modal, and new content and new communities appear over time. In this work, we address CLIPP by collecting large-scale, multi-modal datasets to examine the diffusion of online YouTube videos on Reddit. We analyze these datasets to construct community influence graphs (CIGs) and develop a novel dynamic graph frame- work, INPAC (Information Pathway Across Online Communities), which incorporates CIGs to capture the temporal variability and multi-modal nature of video propagation across communities. Ex- perimental results in both warm-start and cold-start scenarios show that INPAC outperforms seven baselines in CLIPP. Our code and datasets are available at https://github.com/claws-lab/INPAC

In today's data-driven world, recommender systems are essential for filtering information to deliver personalized content, with collaborative filtering (CF) being a popular technique for predicting user preferences based on past interactions. However, capturing the temporal dynamics of user behavior and handling cold-start scenarios remain significant challenges, as user preferences naturally evolve over time and CF often relies solely on historical data. To address these issues, we propose a novel approach that combines a self-attention-based spatiotemporal graph convolutional network with OPTICS clustering which forms adaptive, time-sensitive subgraphs. This enables the model to adapt to changing user preferences and prioritize the most relevant interactions dynamically. Evaluated on the MovieLens100k dataset, our model outperforms baseline methods, effectively generating embeddings for new user-item pairs and demonstrating an inductive capability that enhances both accuracy and adaptability in recommendations.

No abstract available

Internet financial platforms offering consumer credit services confront unique challenges compared to traditional financial institutions. Loan defaults are a primary risk associated with these credit products. Consequently, developing an effective credit risk prediction algorithm is critical for reducing losses and enhancing profits. Unlike the manual review process for loan or credit card applications in traditional banks, internet financial platforms typically grant credit limits based on insights derived from big data analysis. However, for new users, the lack of sufficient credit behavior data presents a significant challenge for default prediction. To overcome this cold-start problem and improve the accuracy of credit risk predictions, this study proposes an enhanced Temporal Graph Network(TGN) model incorporating node spatial information (SETGN) for credit risk prediction. It’s the first time to use TGN with a spatial-temporal attention mechanism in this field. Experimental results using real-world data from a major Chinese online lending platform show our method’s superiority compared to state-of-the-art baseline methods in identifying high-risk individuals.

No abstract available

Self-supervised knowledge-graph completion (KGC) relies on estimating a scoring model over (entity, relation, entity)-tuples, for example, by embedding an initial knowledge graph. Prediction quality can be improved by calibrating the scoring model, typically by adjusting the prediction thresholds using manually annotated examples. In this paper, we attempt for the first time cold-start calibration for KGC, where no annotated examples exist initially for calibration, and only a limited number of tuples can be selected for annotation.Our new method ACTC finds good per-relation thresholds efficiently based on a limited set of annotated tuples. Additionally to a few annotated tuples, ACTC also leverages unlabeled tuples by estimating their correctness with Logistic Regression or Gaussian Process classifiers. We also experiment with different methods for selecting candidate tuples for annotation: density-based and random selection. Experiments with five scoring models and an oracle annotator show an improvement of 7% points when using ACTC in the challenging setting with an annotation budget of only 10 tuples, and an average improvement of 4% points over different budgets.

Influence Maximization (IM) in temporal graphs focuses on identifying influential ``seeds'' that are pivotal for maximizing network expansion. We advocate defining these seeds through Influence Propagation Paths (IPPs), which is essential for scaling up the network. Our focus lies in efficiently labeling IPPs and accurately predicting these seeds, while addressing the often-overlooked cold-start issue prevalent in temporal networks. Our strategy introduces a motif-based labeling method and a tensorized Temporal Graph Network (TGN) tailored for multi-relational temporal graphs, bolstering prediction accuracy and computational efficiency. Moreover, we augment cold-start nodes with new neighbors from historical data sharing similar IPPs. The recommendation system within an online team-based gaming environment presents subtle impact on the social network, forming multi-relational (i.e., weak and strong) temporal graphs for our empirical IM study. We conduct offline experiments to assess prediction accuracy and model training efficiency, complemented by online A/B testing to validate practical network growth and the effectiveness in addressing the cold-start issue.

No abstract available

Knowledge graph embedding has attracted widespread attention in recent years, and since knowledge graphs are dynamically updated in nature, the temporal information embedded is essential. Most of the knowledge graph embedding focuses on static KGs, while temporal knowledge graphs have been poorly studied. In the real-world, much structured knowledge is valid only within a specific temporality, i.e., the development of facts follows a temporal order. Therefore, more and more research works start to incorporate temporal information into knowledge graph representation learning, and the embedding of temporal knowledge graphs focuses on how to embed temporal information into the vector space. Most of the existing temporal knowledge graph embedding models do not model the semantic hierarchy, not fully exploiting the semantic information in the temporal knowledge graph. In this paper, we propose a hierarchy-aware temporal knowledge graph embedding (HA-TKGE), which maps temporal information into a polar coordinate system. The HA-TKGE is mainly inspired by the HAKE model. Specifically, the purpose of radial coordinates is to model temporal information at different levels, where entities with smaller radius are indicated at higher levels, and angular coordinates are intended to represent temporal information at the same level, which has approximately the same radial coordinates and different angles. The HA-TKGE model uses the nature of the polar coordinate system to represent the semantic hierarchy of temporal knowledge graphs and proves its effectiveness in the temporal node prediction task. Experiments show that the HA-TKGE model can effectively model the semantic hierarchy of temporal information and outperforms existing methods overall on the benchmark dataset for the temporal node prediction task.

Reasoning over temporal knowledge graphs (TKGs) is a challenging task that requires models to infer future events based on past facts. Currently, subgraph-based methods have become the state-of-the-art (SOTA) techniques for this task due to their superior capability to explore local information in knowledge graphs (KGs). However, while previous methods have been effective in capturing semantic patterns in TKG, they are hard to capture more complex topological patterns. In contrast, path-based methods can efficiently capture relation paths between nodes and obtain relation patterns based on the order of relation connections. But subgraphs can retain much more information than a single path. Motivated by this observation, we propose a new subgraph-based approach to capture complex relational patterns. The method constructs candidate-oriented relational graphs to capture the local structure of TKGs and introduces a variant of a graph neural network model to learn the graph structure information between query-candidate pairs. In particular, we first design a prior directed temporal edge sampling method, which is starting from the query node and generating multiple candidate-oriented relational graphs simultaneously. Next, we propose a recursive propagation architecture that can encode all relational graphs in the local structures in parallel. Additionally, we introduce a self-attention mechanism in the propagation architecture to capture the query’s preference. Finally, we design a simple scoring function to calculate the candidate nodes’ scores and generate the model’s predictions. To validate our approach, we conduct extensive experiments on four benchmark datasets (ICEWS14, ICEWS18, ICEWS0515, and YAGO). Experiments on four benchmark datasets demonstrate that our proposed approach possesses stronger inference and faster convergence than the SOTA methods. In addition, our method provides a relational graph for each query-candidate pair, which offers interpretable evidence for TKG prediction results.

Temporal knowledge graph reasoning (TKGR) aims to predict future events by inferring missing entities with dynamic knowledge structures. Existing LLM-based reasoning methods prioritize contextual over structural relations, struggling to extract relevant subgraphs from dynamic graphs. This limits structural information understanding, leading to unstructured, hallucination-prone inferences especially with temporal inconsistencies. To address this problem, we propose IGETR (Integration of Graph and Editing-enhanced Temporal Reasoning), a hybrid reasoning framework that combines the structured temporal modeling capabilities of Graph Neural Networks (GNNs) with the contextual understanding of LLMs. IGETR operates through a three-stage pipeline. The first stage aims to ground the reasoning process in the actual data by identifying structurally and temporally coherent candidate paths through a temporal GNN, ensuring that inference starts from reliable graph-based evidence. The second stage introduces LLM-guided path editing to address logical and semantic inconsistencies, leveraging external knowledge to refine and enhance the initial paths. The final stage focuses on integrating the refined reasoning paths to produce predictions that are both accurate and interpretable. Experiments on standard TKG benchmarks show that IGETR achieves state-of-the-art performance, outperforming strong baselines with relative improvements of up to 5.6% on Hits@1 and 8.1% on Hits@3 on the challenging ICEWS datasets. Additionally, we execute ablation studies and additional analyses confirm the effectiveness of each component.