Research On Weibo Sentiment Analysis Methods Based On Topic Model Of Multi-feature Fusion

With the increasing popularity of social platforms,Weibo has been popular by many netizens because of its short text and fast spread.The vast amount of information contained in Weibo text has brought great value from different aspects such as business,politics and entertainment.It has led to the in-depth study of scholars' sentiment analysis techniques on Weibo text.How to carry out accurate sentiment analysis on Weibo text under the premise of controlling cost and ensuring classification effect is a research hotspot now.Based on the unsupervised learning,this thesis combines the traditional topic model with the complex features of Weibo text,and proposes an effective topic sentiment model to analyze the sentiment of Weibo text.Firstly,for the problem of the lack of microblog features in joint sentiment/topic model in the field of Chinese microblog sentiment analysis,emoticons and user personality emotions are introduced into the model.The influence of emoticons is introduced into the model,and the distribution of emoticons in text is modeled.A modeling method of user personality emotional feature based on time according to the theory of “emotional consistency” is proposed.Then,the joint sentiment/topic model based on user character is proposed which includes the user personality emotions.The feasibility and effectiveness of the model are verified by experiments.Then,for the problem of that context semantics is ignored when the aspect and sentiment unification model is modeled in the field of Chinese microblog sentiment analysis,four kinds of inter-sentence relations “transition,hypothesis,progressive,causal” are introduced into the model.They are commonly used in Chinese semantic rules.The relationship vector reflects the emotional changes,so as to carry out emotional modeling.Microblog forwarding symbols are used to model the topic of Weibo,considering the topic relevance and potential semantic relevance of forwarding Weibo and original Weibo.On this basis,the aspect and sentiment unification model based on semantic rules is proposed based on the microblog forwarding relationship,and the results of experiments verify its effectiveness.Finally,we get the conclusion that the hypothesis of the joint sentiment/topic model “words in a text have different topic and sentiment” and the hypothesis of the aspect and sentiment unification model “words in a sentence have the same topic and sentiment” are too strong.So we combine the features of the joint sentiment/topic model based on user character with the aspect and sentiment unification model based on semantic rules and put forward the weak hypothesis that “the words in each clause have the same emotion and theme”.Based on this weak hypothesis,the weibo sentiment topic model based on muitiple features is proposed,which includes the typical Chinese microblog features.Finally,it is verified by experiments that the model has a better emotional classification effect.