Research On Sentiment Analysis And Evolution Process For Hot Events Of Network Public Opinion

With the coming of web2.0 era and the maturity of Internet technology,the development of social networks sites,including mobile social network sites,is far beyond rapid.The social network platform also becomes an important tool for users to receive,publish,and disseminate information.However,the untouchability and rituality of the Internet,cannot fully guarantee the quality of micro blogging information,including its authenticity and quantity.In other words,the public opinion analysis on social network sites is faced with some difficulties and challenges,mainly in the following aspects.Firstly,the text is too short and pithy to recognize semantic.Texts described either in the way of word-bag representation or space vector model(SVM),will cause sparse matrixes.Then,the comments posted by web users are common to be short,but contains multiple themes.It is difficult to analyze sentiments adversely affected by the subjects' extraction.Lastly,the rapid change of the internet age makes the users take more dynamic and complex attitudes toward a certain information,but the relevant public opinion analysis software is not enough intelligence.Aiming at the existing problems of public opinion analysis in the current social network environment,this thesis proposes a solution for public opinion analysis of short text in social network sites based on topic models.The joint sentiment and topic models have been successfully applied in the fields of text mining,which has important theoretical and practical value in terms of topic evolution,personalized recommendation,public opinion analysis and so on.In view of this,the thesis presents a new model,TSTS(Time-aware Short-text Topic-Sentiment Model).Firstly,in order to solve the problem of sparse matrixes,maintain the correlation between the texts,and reduce the time complexity,the model limits the co-occur-words in a same document.Then,this thesis incorporates the emotional calculation method,following the original sentiment and topic models.That is,restrain the co-occur-words to obey the same emotion distribution.Finally,the texts on SNSs have obvious character of timeliness.According to the time information contained in the text itself,the TSTS model is also integrated the time factor,simulating the process curve of the changes of topic and sentiment.In order to verify the feasibility and accuracy of the model proposed,this thesis collected the data from micro blogging,which links to the hot topics,the case of Yunnan Lijiang,the case of Beijing wildlife park tiger,the case of football vs South Korea and the case of Shandong humiliation murder.The three experiments were designed to evaluate the TSTS model from the qualitative and quantitative angles respectively,in terms of the extraction of topic words,sentiment analysis and time variation.The experimental results show that the TSTS model has better performance than the traditional models.