Research On Sentiment Analysis For Short Tests With Improved Topic Model

With the development of the Web2.0 and social media,more and more people widely use the Internet to publish and share information.Most of information generated by users is short texts such as weibo,commodity comments.Though the length of short texts is short,their scale is very large with fast-changing.What's more,they contain lots of personal emotional information.It is of great significance for these fields,such as public opinion detection,customer analysis and commodity analysis,to discover short texts.However,it is not easy to mine topics with subjectivity from short texts.The content of short texts is quite sparse,is short of context information and usually contains a lot of wrong characters,new words.These cause topic models cannot discover high-quality topics with subjectivity.To solve the above problems,we propose two sentiment topic models to discover high-quality topics with subjectivity in short texts.In detail,the main contributions are as follows:(1)Propose a joint sentiment and topic model,Time-User Sentiment Latent Dirichlet Allocation(TUS-LDA),which leverages time and user information.TUS-LDA aggregates posts in the same timeslice or user as a pseudo-document to enrich contextual information,so that TUS-LDA can alleviate the context sparsity problem,and discovers high-quality topics with subjectivity.(2)Propose a joint sentiment and topic model,Weibo Sentiment Model(WSM),which leverages time,user and hashtag information.WSM extends TUS-LDA and further enriches contextual information by using hashtag which incorporates semantic knowledge.(3)Conduct multiple experiments in three real-world datasets to evaluate seven models in terms of discovering sentiment-aware topics and sentiment classification.TUS-LDA and WSM are better than other five models,WSM performs best.TUS-LDA and WSM are capable of discovering high-quality sentiment-aware topics,and benefit for sentiment analysis and the analysis of public opinion.(4)Design and implement the weibo sentiment analysis system WSAS based on WSM.