Research on Public Opinion Sentiment Analysis Based on Multi-modal Feature Fusion

Abstract

Under the background of rapid development of Internet technology, social media provides diversified expression channels for the public, and users are more inclined to use a combination of text and pictures to post comments. However, most of the current sentiment analysis methods use single-modal analysis, resulting in limited accuracy. To overcome this problem, this paper constructs a multimodal sentiment analysis model based on InceptionClip-Bert. First, text sentiment features are extracted with the help of the Bert model, and the Clip model is improved to extract image sentiment features; then, the cosine similarity is used to calculate the correlation between graphic and text sentiment tendencies to realize feature fusion; finally, public opinion analysis is carried out in terms of word frequency, word cloud, IP of the information publisher, identity of the information publisher, and T-SNE image clustering. The experimental comparison results show that this method can significantly improve the accuracy of sentiment recognition and provides a new idea for the research of multimodal feature fusion for sentiment analysis of public opinion.