Microblog Sentiment Analysis And Influence Estimation Research

The emergence of new network technology in Web2.0has made the Internet rapidlydevelop. Surfing on the internet has changed from purely obtaining information from the internetto a new participation between the user and the net. Plenty of social media platforms haveemerged quickly from this mindset, among which microblog has undoubtedly become theoutstanding representative with the characteristics of the new age.Micoblog contains a large amount of subjective opinions of the public. These opinions canbe analyzed to be used in many practical applications, such as, judging consumers preference inpurchasing, predicting the tendency of voting, finding negative public opinions, and so on.Moreover, the study on the spread of microblogs and on the influential factors in this process isof great importance to public opinion supervision. this dissertation mainly researches onsentiment analysis and influence maximization for Chinese microblog. The achievements arelisted as follows:(1)The content and the context of the microblog is brought into sentiment analysis and anunsupervised method of sentiment orientation classification based on the tree structure isproposed. For the problems existing in microblog, which includes locking SV structure, writingat random, topic divergence, sentiment shifting, and so on, First, the tree structure of themicroblog is initialized. Then, dynamic self-adjustment is made to the tree structure. At last, thefinal tree sets is used to make tendency judge. The research results show that the proposedmethod effectively increases the classification accuracy and tackles the problem of sentimentshifting.(2) Extract summarization is difficult to apply to microblog because of the character limit.Therefore, a method of sentiment summarization based on the shortest path is presented. For theproblem existing in microblog, which includes huge data, too short content, big redundancy, andso no. First, the labels of the subtopics are extraced based on the shortest path. Then, thesentiment features are exploited. At last, the labels and the features are combined to form thesentiment summarization. The experiment results show that the new method proposed caneffectively reduce information redundancy and improve the readability.(3) State-of-art researchs on influence maximization estimation mainly focus on degree ofnode, considering the contributions of content information and sentiment, a method of influencemaximization estimation for microblog based on content and sentiment is put forwards. First, thetopic-hashtag is extracted and used to classify users. Then, user’s initial influence scores are calculated and initial user sets in different topic are picked up. At last, a progress to influencefoctor is made and the maxmum influence transmission range is searched. Experiment resultsshow proposed method can effectively improve the range of influence, which is more in linewith the actual situation.