Multimodal sentiment analysis based on attention mechanism and graph convolutional neural network

• 1、Liaoning Technical University,School of Electronic and Information Engineering
• 2、Liaoning Technical University,School of Electronic and Information Engineering,Huludao 12510

Abstract：Multimodal sentiment analysis can make up for the shortcomings of unimodal sentiment analysis and enhance the performance of sentiment analysis by utilizing the complementary nature of different modalities to enhance the expressive ability of sentiment features. In solving the problem of direct cascade fusion of sentiment features, the attention mechanism is utilized to capture key semantic information in multimodal data for sentiment analysis. The attention mechanism is used to learn the importance of the interactions between unimodal features and the degree of association between different modalities to effectively obtain multimodal sentiment feature information. The accuracy rate reaches 82.71% after adding the attention mechanism. Adding graph convolutional neural network obtains more effective multimodal contextual emotional features and further improves the accuracy rate to 83.28%. The results show that incorporating the cited graph convolutional neural network can effectively improve the multimodal emotion feature information extraction ability.

Keywords： Multimodal sentiment analysis graph convolutional neural networks attention mechanisms.