Sentiment Analysis Methods Based on Multi-Modal Fusion and Deep Learning
Jingdong Wang, Wenyan Zhao, Fanqi Meng, Guangqiang Qu
Abstract
Sentiment analysis is an important research topic in the field of artificial intelligence and occupies an important position in product evaluation, public opinion analysis, mental health analysis, risk assessment and other fields. However, traditional unimodal sentiment analysis is affected by factors such as missing data or noise, and suffers from shortcomings such as incomplete information, inconsistent information, lack of contextual information, and loss of multimodal features. Multimodal sentiment analysis can improve the accuracy, robustness and expressiveness of unimodal sentiment analysis by fusing data from multiple perceptual modalities, thus providing more comprehensive and accurate sentiment analysis results. Therefore, this paper addresses the shortcomings of unimodal sentiment analysis, describes in detail multiple feature fusion methods, and introduces two techniques in the field of deep learning, points out the current problems and future development direction of multimodal sentiment analysis, and provides a new theoretical approach for sentiment analysis.