In response to the issues of noise and redundant information affecting the accuracy and robustness of existing multimodal sentiment analysis models, this paper suggests a multimodal sentiment analysis model based on TCN (Time-Causal Convolutional Network) and cross-modal interaction feedback networks. It is primarily composed of two modules: a modality fusion module and a feedback module. Firstly, the text, audio, and video feature vectors are encoded through TCN, followed by the generation of unimodal representations via dot-product attention. To capture interaction information across different modalities, a cross-modal fusion module comprising three symmetric multi-head attention mechanisms is introduced. To eliminate noise and redundant information generated from modality fusion, a feedback module consisting of TCN, fully connected layers, and sigmoid layers is designed to apply its generated masked features in the cross-modal fusion process. Finally, a fully connected layer is employed to integrate the concatenated vector of unimodal representations and cross-modal features, which is then fed into softmax for sentiment classification. Experimental results demonstrate that compared to existing models, this model exhibits superior performance in sentiment classification tasks.

Social platforms allow individuals to express their opinions and viewpoints using multiple information modes. The effective fusion of these various types of information can enhance the accuracy of predicting users’ emotional tendencies. However, existing multimodal sentiment analysis does not fully consider the emoticon information contained in the text and the semantic irrelevance between the text and the image, resulting in poor sentiment analysis. To address this problem, we propose an image-text multimodal emotion analysis model (ITMEA-FE) that incorporates emoji features and text features into feature vectors to enhance the utilization rate of features. The correlation between image information and text information is identified, reducing the influence of emoji information and image-text semantic irrelevance on sentiment analysis. Finally, sentiment analysis is performed through a network of multi-head attention mechanisms. Experimental results show that the proposed method achieves an accuracy rate of 75.32% and a Macro-F1 value of 75.11%, outperforming the benchmark model.

Aiming at the problem of a complex relationship and easy loss of time series length information in power load forecasting tasks. In this paper, a short-term power load forecasting model based on GLL-TCN is constructed. Firstly, the graph learning layer is taken to extract the unidirectional relationships between variables to form an adjacency matrix to obtain the asymmetric features of variables. Secondly, the graph convolutional network is improved, and the signals of each node are fused and propagated through the graph convolutional network, so that the characteristics of each node contain the load data information of its neighbor nodes. Finally, in the temporal convolutional network, the inflated convolutional module layer is used to process the load data of longer series, expand the receptive field, and generate a prediction model. Experiments are carried out on the power dataset and compared with GP, RNN-GRU, AR, LSTNet and TPA-LSTM models, the experimental results indicate that the accuracy of prediction of the GLL-TCN model is better than the base line model at different time intervals, thus proving the accuracy of the model in load forecasting.

Aiming at the problems of imperfect information management platform and Low quality of excavation, a combined algorithm of student behavior analysis based on clustering and association rules algorithm is proposed. Firstly, K-mediods algorithm is used to cluster the student behavior data, and the clustering results are discretized. Then, the association between student behavior and performance is analyzed by Eclat algorithm to extract important rules. Finally, combined with the results of the two algorithms, the behavior factors affecting students' performance are analyzed comprehensively. The analysis results show that the use of the above-mentioned combined algorithm not only improves the quality of the mining results, but also provides guidance and suggestions to teachers' teaching work and students' learning conditions through the mining results.