多源数据融合与情感

This paper proposes a design scheme for a public event public opinion monitoring and early warning system based on multi-source data fusion and sentiment analysis. The system collects multi-source data such as social media, news portals and user behavior, uses the BERT deep feature and statistical feature fusion method to perform sentiment classification, combines the LSTM time series model to predict the evolution trend of public opinion, and establishes a dynamic threshold mechanism to trigger intelligent early warning. Experimental results show that the designed system is superior to traditional methods in terms of sentiment analysis accuracy, trend prediction accuracy and early warning response speed, and has good real-time and stability. The research results provide effective technical support for improving the level of intelligent emergency management and social governance of public events.

To tackle the challenge of data diversity in sentiment analysis and improve the accuracy and generalization ability of sentiment analysis, this study first cleans, denoises, and standardizes multi-source data, and then proposes an improved sentiment analysis framework based on deep transfer learning technology. The swarm intelligence optimization algorithm is introduced to optimize the deep transfer learning architecture. The experiment outcomes indicate that the artificial bee colony-improved particle swarm optimization algorithm designed in the study has the minimum convergence value under single-peak and multi-peak test functions, with a standard deviation of less than 0.25 and a success rate of 100% in optimization. The solution set obtained by the algorithm performs well in terms of hyper-volume and inverse generation distance, and is closest to the true Pareto frontier. The diversity and convergence of the algorithm’s solutions outperform other algorithms, ensuring the optimization requirements of the sentiment analysis architecture. The deep transfer learning sentiment analysis model designed based on this achieves the best classification accuracy, with a minimum mean absolute error of 0.128, and good performance in accuracy recall and receiver operation characteristics. The recognition of emotional diversity has high accuracy, with an accuracy of 0.928 for 5 categories. The cross entropy loss and Matthews correlation coefficient confirm the reliability of this method for sentiment orientation recognition in multi-source data. The research explores the use of deep transfer learning in multi-source data fusion sentiment analysis, enriches and improves sentiment analysis techniques, and promotes innovation and development of sentiment analysis technology.

The field of financial sentiment analysis encounters substantial obstacles when attempting to leverage lexicon-based knowledge effectively, especially in developing comprehensive multi-level integration approaches and addressing inconsistencies between different sentiment resources. To address these challenges, this research introduces a novel XGS-FinBERT methodology that combines diverse financial lexicons with FinBERT via XGBoost-SHAP directed feature significance assessment. The approach involves creating an extensive financial sentiment lexicon through the integration of five established resources: SenticNet, FinSenticNet, the Loughran-McDonald vocabulary, Stock_lex, and the ML-based dictionary. An XGBoost-SHAP framework quantifies feature contributions, identifying five core dimensions: sentiment, score, negative, positive, and intensity. The model employs a parallel fusion architecture with dynamic gating mechanisms that adaptively weight FinBERT semantic features and dictionary-based sentiment features. Anti-forgetting mechanisms and differential learning rates preserve pre-trained knowledge while enhancing domain-specific performance. Experimental results on the FPB dataset demonstrate that XGS-FinBERT achieves $\mathbf{8 8. 6 \%}$ accuracy and $\mathbf{8 8. 7 \%} \mathbf{F 1}$ -score, improving by 2.2 and 2.2 percentage points respectively over the FinBERT baseline, validating the effectiveness of intelligent dictionary-model integration.

No abstract available

Sentiment analysis (SA)at the sentence, aspect, and document levels determines the sentiment of particular aspect phrases in a given sentence. Due to their capacity to extract sentiment information from text in aspect-level sentiment classification, neural networks (NNs) have achieved significant success. Generally speaking, sufficiently sizable training corpora are necessary for NNs to be effective. The performance of NN-based systems is reduced by the small size of the aspect-level corpora currently available. In this research, we suggest a gated bilateral recurrent neural network (G-Bi-RNN) as a foundation for multi-source data fusion, their system offers sentiment information that several sources. We develop a uniform architecture specifically to include information from sentimental lexicons, including aspect- and sentence-level corpora. To further provide aspect-specific phrase representations for SA, we use G-Bi-RNN, a deep bilateral Transformer-based pre-trained language model. We assess our methods using SemEval 2014 datasets for laptops and restaurants. According to experimental findings, our method consistently outperforms cutting-edge techniques on all datasets. We use a number of well-known aspect-level SA datasets to assess the efficacy of our model. Experiments show that when compared to baseline models, the suggested model can produce state-of-the-art results.

Abstract Neural networks have achieved great success in aspect-level sentiment classification due to their ability to learn sentiment knowledge from text. Generally, the effectiveness of neural networks relies on sufficiently large training corpora. However, existing aspect-level corpora are relatively small, which greatly limits the performance of neural network-based systems. In this paper, we propose a novel approach to aspect-level sentiment classification based on multi-source data fusion, which allows our system to learn sentiment knowledge from different types of resources. Specifically, we design a unified framework to integrate data from aspect-level corpora, sentence-level corpora, and word-level sentiment lexicons. Moreover, we take advantage of BERT, a pre-trained language model based on deep bidirectional Transformers, to generate aspect-specific sentence representations for sentiment classification. We evaluate our approach using laptop and restaurant datasets from SemEval 2014. Experimental results show that our approach consistently outperforms the state-of-the-art methods on all datasets.

Sentiment analysis of text data, such as reviews, can help users and merchants make more favorable decisions. It is difficult to use the popular supervised learning method to complete the sentiment classification task because marking data manually is time-consuming and laborious. Unsupervised sentiment classification methods are mostly based on sentiment lexicons. The existing sentiment lexicons are simply not capable of domain sentiment classification, it still requires to construct a domain sentiment lexicon. There are still many problems with the advanced domain sentiment lexicon construction methods, e.g., rely heavily on labeled data, poor accuracy. We propose a labeled data extension idea to reduce the dependence of supervised learning methods on labeled data. In order to solve the problems of domain sentiment lexicon construction, we proposed a novel framework based on multi-source information fusion (MSIF) for learning. We extracted four kinds of emotional information, which are lexicon emotional information, emotional word co-occurrence information, emotional word polarity information and polarity relationship information of emotional word pair. When extracting the co-occurrence information, a novel method based on the data extension idea is proposed to enhance its accuracy and coverage. In order to accelerate the solution of the fusion model, an optimization method based on the ADMM algorithm is applied. Experimental results on five Amazon product review datasets show that the sentiment dictionary constructed by the proposed method can significantly improve the performance of review sentiment classification compared with the current popular baseline and the state-of-the-art methods.

Recently, fine-tuning the large pre-trained language models on the labeled sentiment dataset achieves appealing performance. However, the obtained model may not generalize well to the other domains due to the domain shift, and it is expensive to update the entire parameters within the large models. Although some existing domain matching methods are proposed to alleviate the above issues, there are multiple relevant source domains in practice which makes the whole training more costly and complicated. To this end, we focus on the efficient unsupervised multi-source sentiment adaptation task which is more challenging and beneficial for real-world applications. Specifically, we propose to extract multi-layer features from the large pre-trained model, and design a dynamic parameters fusion module to exploit these features for both efficient and adaptive tuning. Furthermore, we propose a novel feature structure matching constraint, which enforces similar feature-wise correlations across different domains. Compared with the traditional domain matching methods which tend to pull all feature instances close, we show that the proposed feature structure matching is more robust and generalizable in the multi-source scenario. Extensive experiments on several multi-source sentiment analysis benchmarks demonstrate the effectiveness and superiority of our proposed framework.

Predicting stock movements in emerging markets is challenging due to high volatility and the influence of diverse factors. This study proposes a classification-based approach for forecasting next-day stock price direction on the Qatar Stock Exchange (QSE) using a modified Classification Temporal Fusion Transformer (TFT). The model integrates historical stock data, technical indicators, global market indices, commodity trends, and news sentiment. Experiments on four major QSE-listed companies from 2018 to 2024 show that combining technical and sentiment features improves predictive accuracy. The Classification TFT achieved up to 68% accuracy, outperforming traditional methods and highlighting the value of multi-source data for financial forecasting in emerging markets.

No abstract available

Determining the trend of the stock market is a complex task influenced by numerous factors like fundamental variables, company performance, investor behavior, sentiments expressed in social media, etc. Although machine learning models support predicting stock market trends using historical or social media data, reliance on a single data source poses a serious challenge. This study introduces a novel Explainable artificial intelligence (XAI) to address a binary classification problem wherein the objective is to predict the trend of the stock market, utilizing an integration of multiple data sources. The dataset includes trading data, news and Twitter sentiment, and technical indicators. Sentiment analysis and the Natural Language Toolkit are utilized to extract the qualitative information from social media data. Technical indicators, or quantitative characteristics, are therefore generated from trade data. The technical indicators are fused with the stock sentiment features to predict the future stock market trend. Finally, a machine learning model is employed for upward or downward stock trend predictions. The proposed model in this study incorporates XAI to interpret the results. The presented model is evaluated using five bank stocks, and the results are promising, outperforming other models by reporting a mean accuracy of 90.14%. Additionally, the proposed model is explainable, exposing the rationale behind the classifier and furnishing a complete set of interpretations for the attained outcomes.

Abstract Cross-domain sentiment classification uses knowledge from source domain tasks to enhance the sentiment classification of the target task. It can reduce the workload of data annotations in the new domain, and significantly improve the utilization of labeled resources in the source domains. Available approaches generally use knowledge from a single-source domain and hard parameter sharing methods, which are likely to ignore the differences among domain-specific features. We propose a novel framework with multi-source domain adaptation and joint learning for multi-source cross-domain sentiment classification tasks This framework uses bi-directional gated recurrent units and convolutional neural networks for deep feature extraction and soft parameter sharing for information transfer across tasks. Furthermore, it minimizes distance constraints for deep domain fusion. Multi-source domain adaptation involves multiple concurrent task learning, and the gradients are simultaneously back propagated. We validate the proposed framework on multi-source cross-domain sentiment classification datasets in Chinese and English. The experimental results demonstrate that the proposed method is more effective than state-of-the-art methods in improving accuracy and generalization capability.

No abstract available

Cryptocurrency markets are characterized by high volatility, nonlinear dependencies, and limited transparency, making short-term forecasting particularly challenging for both researchers and practitioners. To address these complexities, this study introduces a Temporal Fusion Transformer (TFT)-based forecasting framework that integrates on-chain and technical indicators to improve predictive performance and inform tactical trading decisions. By combining multi-source features—such as Spent Output Profit Ratio (SOPR), Total Value Locked (TVL), active addresses (AA), exchange net flow (ENF), Realized Cap HODL Waves, and the Crypto Fear and Greed Index—with classical signals like Relative Strength Index (RSI) and moving average convergence divergence (MACD), the model captures behavioral patterns, investor sentiment, and price dynamics in a unified structure. Five major cryptocurrencies—BTC, ETH, USDT, XRP, and BNB—serve as the empirical basis for evaluation. The proposed TFT model is benchmarked against LSTM, GRU, SVR, and XGBoost using standard regression metrics to assess forecasting accuracy. Beyond prediction, a signal-based trading strategy is developed by translating model outputs into daily buy, hold, or sell signals, with performance assessed through a comprehensive set of financial metrics. The results suggest that integrating attention-based deep learning with domain-informed indicators provides an effective and interpretable approach for multi-asset cryptocurrency forecasting and real-time portfolio strategy optimization.

Market sentiment plays a very significant role in influencing short-term movements of stock prices, but most existing sentiment analysis techniques are found on a single source of information, either social media or financial news. This research proposes a multi-source sentiment fusion method that incorporates leading financial news and real-time sentiments from social media (tweets) to offer a stronger and more stable stock sentiment indicator. The method leverages Natural Language Processing (NLP) techniques for preprocessing data, and then sentiment analysis using lexicon-based methods (VADER) and transformer models (FinBERT). To counter the nature of differences between news and tweets-credibility, linguistic tone, timeliness, this work suggests a weighted fusion framework, placing more weight on the news sources but leveraging the timeliness of Twitter for early market indications. Experiments on stock dataset coverages for popular tech stocks (e.g., Apple, Tesla) indicate that the aggregate sentiment score achieves higher accuracy and stronger correlation with short-term stock price movements compared to the baselines from single sources. The findings confirm the importance of multi-source fusion in finance sentiment analysis and pose the possibility of real-time decision aid systems for algorithmic trading, risk management, and market forecasting.

The growth of e-commerce has led to a rapid increase in user-generated content (UGC), attracting scholars’ attention as a new data source for investigating customer requirements. However, existing requirements analysis methods fail to integrate three critical requirement indicators: stated importance, derived importance, and performance. Using only one or two of these indicators inevitably has its limitations. This paper proposes a novel framework for analyzing and prioritizing customer requirements based on multi-source UGC. First, customer requirements are extracted from online reviews and questions & answers using non-negative matrix factorization. Next, aspect-level sentiment analysis and multi-source data fusion are employed to calculate dual importance and performance. Specifically, we developed an improved importance–performance analysis (IPA) model, named dual importance–performance analysis (Du-IPA), which integrates the three indicators to classify requirement types in a 3D cube with corresponding improvement strategies. Finally, by combining the three indicators, an improved prospect value and PROMETHEE-II are proposed using prospect theory to prioritize CRs for product service improvement. The effectiveness of the proposed method is demonstrated through a case study of fresh food in online retail.

No abstract available

Implicit sentiment identification is a significant classical task in text analysis. Graph neural networks (GNNs) have recently been successful in implicit sentiment identification, but the current approaches still suffer from two problems. On the one hand, there is a lack of structural information carried by the single-view graph structure of implicit sentiment texts to accurately capture obscure sentiment expressions. On the other hand, the predefined fixed graph structure may contain some noisy edges that cannot represent semantic information using an accurate topology, which can seriously impair the performance of implicit sentiment analysis. To address these problems, we introduce a knowledge-fusion-based iterative graph structure learning framework (KIG). Specifically, for the first problem, KIG constructs graph structures based on three views, namely, co-occurrence statistics, cosine similarity, and syntactic dependency trees through prior knowledge, which provides rich multi-source information for implicit sentiment analysis and facilitates the capture of implicit obscure sentiment expressions. To address the second problem, KIG innovatively iterates the three original graph structures and searches for their implicit graph structures to better fit the data themselves to optimize the downstream implicit sentiment analysis task. We compared our method with the mainstream implicit sentiment identification methods on two publicly available datasets, and ours outperformed both benchmark models. The accuracy, recall, and F1 values of KIG on the Pun of the Day dataset reached 89.2%, 93.7%, and 91.1%, respectively. Extensive experimental results demonstrate the superiority of our proposed method for the implicit sentiment identification task.

No abstract available

The grand challenge of cross-domain sentiment analysis is that classifiers trained in a specific domain are very sensitive to the discrepancy between domains. A sentiment classifier trained in the source domain usually have a poor performance in the target domain. One of the main strategies to solve this problem is the pivot-based strategy, which regards the feature representation as an important component. However, part-of-speech information was not considered to guide the learning of feature representation and feature mapping in previous pivot-based models. Therefore, we present a fused part-of-speech vectors and attention-based model (FAM). In our model, we fuse part-of-speech vectors and feature word embeddings as the representation of features, giving deep semantics to mapping features. And we adopt Multi-Head attention mechanism to train the cross-domain sentiment classifier to obtain the connection between different features. The results of 12 groups comparative experiments on the Amazon dataset demonstrate that our model outperforms all baseline models in this paper.