一种基于多模态的社交媒体情感分析方法

Sentiment analysis is of great significance and value in many fields such as economy and entertainment.In order to make fully use of the complementarity between the text and the image in social media and the correlation between images, this paper proposes a multimodal fusion method based on image and text.In this method, 3DCNN is used to extract the correlation between a group of images, Word2vec and LSTM are used to extract the features from the text, and a multimodal fusion method is designed to fuse them to get the result.The method was tested on the Weibo dataset.The experimental results show that the accuracy and other indicators of this method are improved compared with the existing sentiment analysis methods.