Sentiment Analysis Of Microblog Users Based On Specific Events

With the rapid development of the mobile Internet,social media such as Microblog,Wechat,and news portals have play an increasingly important role in the spread of public opinion.Users in Microblog can freely express their opinions,and user opinion will lead to rapid propagation and spread of the event.Meanwhile,rapidly propagation of user's opinion over a particular event,a large number of negative views will cause the situation to be out of control,and mass incidents would be triggered.Therefore,the sentiment analysis of user comments is very necessary,through a comprehensive judgment of the sentiment polarity of all users on an event,to find the source of affecting the user's sentiment expression,can facilitate the relevant regulatory authorities to effectively Network supervision and grooming.Based on the above research background,this paper takes Sina Microblog as the research object,and the main research contents include the following three parts.Firstly,in view of the characteristics of large data amount,noisy data and many meaningless data in Microblog,data preprocess under the Hadoop platform is studied in the paper.Secondly,user vectorization and event vectorization are studied in this part.Different users tend to use the same words to express different emotional tendencies,at the same time,the same words are used to comment on different events and often bring about emotional differences.The vectorization of users and events is studied to better describe the information representation of different users and events.In user vectorization,the user's general features,behavioral features,and social features are used to characterize a user.In event vectorization,this article uses the word co-occurrence model to expand the event keyword,through the vectorization of event keywords to get the vectorization of the event.Thirdly,Two kind of different models of Microblog sentiment analysis based on deep learning are constructed.One is to classify emotion through residual network and attention mechanism,which is verified on the chinese hotel review dataset and the stanford sentiment dataset.And the other is a hierarchical network model based on users and events,which incorporates the information of users and events and achieves better sentiment analysis.