Research On Emotion Analysis Method Based On Weibo Features

In the rapid spread of information today,Weibo has become an effective and authoritative carrier,carrying the dissemination of Internet news and gossip,most of life information,people can stay at home situation,through the Weibo hot search to understand today's hottest events,explore the new topic through Weibo comments.With the increasing popularity of Weibo,Weibo users are also climbing,and the analysis of Weibo text is becoming more and more critical.Weibo sentiment analysis arises at the historic moment.Through Weibo emotion analysis,we can monitor the network public opinion on emergencies,understand the negative emotions in the Internet,and resist the occurrence of adverse events through certain measures.It can also be used for marketing,film scoring and so on through emotional analysis.In this paper,we get the user Weibo from Sina Weibo official interface,and improve the way of feature selection and weight assignment by preprocessing under the condition of each Weibo emotional annotation.(1)Comparison of the relevant feature selection algorithm,through the literature reference,this paper uses CHI as the main research methods of feature selection,analysis on the inadequacy of CHI,considering the lack of frequency statistics,the frequency correction factor;considering the lack of distribution within class and between class and within class and introduced between class distribution factors,and ultimately the formation of LCHCHI feature selection method was improved,and a series of traditional feature selection methods to compare the effect of CHI.(2)In the feature weighting method,considering the shortcomings of the traditional TF-IDF method,we choose to use LCDCHI instead of IDF to form the final TF-LCDCHI feature weighting algorithm,and get a high accuracy through the comparison between the related algorithm experiments.When the feature dimension is 1000 dimension,twenty percent of the text is selected as the test set,and the accuracy rate is 88.12%.