Research On Microblog Sentiment Analysis Based On Social Relations Among Users

With the rapid development of information technology,especially the emergence of social networking platforms such as Twitter and Sina-Weibo,Microblog Sentiment Analysis has become a popular research topic.However,microblogging messages are usually short,unstructured,less information and much noise,creating a significant challenge for the application of traditional contentbased methods.In fact,based on the link structures between online social networks nodes,users' sentiments are afffected by other nodes,unconsciously.Moreover,recent studies have pointed out that microblog sentiment analysis is not confine to traditional content-based analysis,considering the useful social network information could be helpful for MSA tasks.In web 2.0,interactions between users are more frequent,users' social relations are more complex,and the sentiment influence among users are more common.Meanwhile,according to homophily and social influence,users with influence relationships and similar relationships tend to express the same or similar sentiment on a particular topic.Therefore,MSA-USUI model that based on user similarity and influence social relations was proposed.According to online social network nodes similarity measurement and the characteristics of users' information,this paper proposes to measure the similarity relations between microblogging users from three aspects: network structure,basic information and user interest respectively.Meanwhile,based on the analysis of the influence of online social network nodes and the characteristics of users' behaviors,this paper proposes to measure the influence relations between microblogging users from the perspective of friend relationships and user interaction behaviors.Secondly,based on the content model,this paper incorporates user social relations into the modeling part by using user similarity information and user influence relations to build sentiment relationships between microblogging data and employ these microblog–microblog sentiment relations to further train the optimized sentiment classifier.Then,this paper deals with the over-fitting problem and gives the algorithm to solve the model.Finally,two Sina-Weibo datasets obtained by the Python crawler to verify the model.The experimental results show that the MSA-USUI model has a better sentiment classification accuracy and F1-score than the Support Vector Machine method and the state-of-the-art supervised model called SANT.In addition,it proven that the improvement brought by user influence relations is greater than the user similarity information,but the MSA-USUI model achieved the best performance when incorporating both two relations.