Based on cross-modal interaction and multi-level fusion sentiment analysis

At present, in the research in the field of multimodal emotion analysis, there are prominent problems such as difficult to accurately and effectively extract the features of different modes (such as image and text), difficult to realize the interaction between cross-modal features, and the extracted features can be fully integrated. Therefore, this paper introduces a cross-modal attention mechanism in the process of different modal interaction, and constructs a multi-modal emotion analysis model through the process of multi-channel feature extraction and multi-level feature fusion, which has the characteristics of cross-modal interaction and multi-level fusion. First, this paper uses the bert model as the tool to extract text features, and the vet model and resnet model to extract global and local features respectively; next, the cross-modal attention mechanism is used to promote the information interaction between different modes; then, the attention mechanism and BILSTM are used to realize the integration of features; finally the integrated features are input into the Softmax classifier, thus obtaining the ultimate result of emotion analysis.