Research And Application Of Multi-modal Sentiment Analysis Methods In Chinese

Sentiment analysis technology can identify emotional information in massive data in the network,and plays an important role in marketing strategy formulation and network public opinion monitoring.Sina Weibo is one of the important data sources for network public opinion analysis in my country.At present,the sentiment analysis of Weibo mainly focuses on text sentiment analysis.However,with the diversification of data types on the Weibo platform,the combination of graphics and text It has become one of the important ways for people to express their opinions and emotions.Multimodal sentiment analysis research is a research demand generated with the development of the Internet.Multimodal sentiment analysis methods can integrate more comprehensive information to analyze the expresser’s emotion,so as to obtain analysis results that are closer to real emotions.This paper focuses on the multi-modal sentiment analysis task of microblog,and uses microblog text,emoji expressions and pictures to conduct sentiment analysis research.The main work is as follows:1.A pre-training task for microblog text characteristics is proposed,including topic matching task and emotional word masking task combined with emoji expressions.Use this pre-training task to continue pre-training on the BERT pre-training model and then use it for sentiment analysis of Weibo text.This task aims to help the model adapt to the characteristics of Weibo data,thereby improving the performance of the model on downstream Weibo sentiment analysis tasks.The experimental results on the n Co V-weibo Weibo epidemic sentiment analysis dataset show that this pre-training task has better experimental results than other commonly used pre-training tasks.Pre-training on the corpus,the accuracy of the method proposed in this paper has an increase of about 12.Propose a microblog multi-modal sentiment analysis method ABGN(Attentionbased and Gate-controlled Network).This method focuses on the sentiment analysis of text data,and uses the cross-modal attention mechanism to capture the interactive information between microblog pictures and texts.Aiming at the problem that there are multiple pictures in one text in microblog data,and each picture contributes differently to sentiment analysis,A picture sentiment analysis module based on the gating mechanism is designed by calculating the similarity of graphic and text features,and filtering non-key information helps the model perform better sentiment analysis.Experiments on the microblog multimodal data set show that ABGN has better performance than other baseline models.Compared with the sentiment analysis of plain text data,ABGN has improved the accuracy of the sentiment analysis task of microblog data by about 73.Build a prototype system for microblog multi-modal sentiment analysis results visualization and online sentiment analysis.It helps users to intuitively see the sentiment analysis results of multimodal data,and provides an online multimodal sentiment analysis function to verify the effectiveness of the multimodal sentiment analysis model in this study.