| In recent years,with the rapid development of Internet technology,social media platforms such as microblogs have become an integral part of people’s lives.As a platform for free expression of opinions and emotions,microblogs carry people’s emotions,ideas,experiences and feelings.This information not only reflects people’s real emotions and attitudes,but also has high reference value for various organizations such as enterprises and governments.Therefore,how to extract useful emotional information from these massive information and use it effectively has become the focus of research.This paper studies the sentiment analysis method of Chinese microblog text in the context of the new crown epidemic.We propose two methods based on sentiment dictionary and deep learning,and the research results show that these two methods have a good effect on the sentiment analysis of Chinese microblog,and both can obtain emotional information from Chinese microblog,providing valuable reference and support for users and organizations.The main research contents are as follows:(1)Construction of a comprehensive sentiment dictionary:Considering that the existing sentiment dictionaries cannot satisfy microblog sentiment analysis,this paper constructs a comprehensive sentiment dictionary,including basic sentiment dictionary,negative word dictionary,degree adverb dictionary and microblog new word sentiment dictionary.The key steps are to construct microblog new word sentiment dictionary based on statistical information method and SO-PMI algorithm.(2)Sentiment analysis of microblog text based on comprehensive sentiment dictionary and semantic rule set:According to the semantic characteristics of Chinese,this paper introduces a self-defined text sentiment polarity judgment rule and applies it to the sentiment analysis of Chinese microblogs.On the basis of constructing a comprehensive sentiment dictionary and semantic rule set,we use the method of splitting complex sentences into single sentences and then splitting single sentences into words for text sentiment calculation.(3)Sentiment analysis of microblog text based on deep learning:A model structure based on BERT+DPCNN is proposed,which continues to extract the features of sentences on the basis of the results of BERT model enconder to obtain more internal feature attributes of sentences and improve the text multiclassification effect,and at the same time conduct comparative experiments with other BERT+models in the experimental analysis part.The experimental results show that compared with a single emotional dictionary and the method without semantic rule set,the classification effect of the method based on the comprehensive emotional dictionary and the semantic rule set is improved to a certain extent.At the same time,considering the fine-grained classification of text sentiment,the proposed model structure based on BERT+DPCNN improves the classification accuracy by 2%-4%compared with other BERT+models. |
