A Multimodal Sentiment Analysis Method Based on Interactive Attention

Abstract

Sentiment analysis, a crucial aspect of human-computer interactions and healthcare, faces challenges in multi-modal data representation and fusion. Traditional methods often fail to adequately capture the complexities of integrating diverse modalities. Addressing this, we introduce the Cross Attention Fusion (CAF) algorithm, a novel approach for effective multi-modal sentiment analysis. The CAF algorithm uniquely employs a dual attention mechanism—global and local—to enhance the representation and integration of features across modalities. Our extensive experiments, conducted on the CH-SIMS dataset, demonstrate that the CAF algorithm significantly outperforms existing unimodal and multimodal methods in accuracy. These results highlight the CAF algorithm's superiority in capturing nuanced sentiment expressions, marking a notable advancement in the field of sentiment analysis.