Multimodal Fusion Sentiment Analysis Based on Spatial Frequency Enhancement

In the multimodal domain, multimodal sentiment analysis (MMSA) aims to understand the user's true thoughts by combining image information and textual information to recognize specific aspects of feelings. However, if the features of different modalities are simply spliced together, the interaction and complementary information between modalities cannot be fully explored, and it is difficult to fully capture the more refined and frequently occurring information in the modalities, resulting in the accuracy of sentiment analysis being affected. In this regard, this paper proposes a multimodal sentiment analysis method based on spatial state enhancement with graphic fusion (MFSFE). The formula uses an attention mechanism to process image and text fusion features multiple times, and designs a spatial frequency enhancement module (SFEM) to perform adaptive enhancement processing of high and low frequency information of the image through an agent network. Finally, the features obtained from the processing are projected into a unified depth space. Experiments on TWITTER-15 and TWITTER17 and MASAD datasets show that the model improves the accuracy of sentiment analysis, exhibits excellent performance, and provides a design idea that can be used for subsequent models.