Research On Sentiment Classification Technology For Chinese Weibo

With the rapid development of network technology, Weibo, as a new social platform, has gradually penetrated into every aspect of people’s lives. Weibo is full of users’ views and opinions towards products, entertainments, social events and so on, which contains a wealth of emotional information. Analyzing users’ emotional attitude during a period of time or towards a particular topic, classifying the emotions contain in Weibo effectively has great commercial value and social value. It not only allows businessmen to obtain users’ views instantly, but also allows government departments to keep abreast of social dynamics, listen to the voice of the people, which has a good monitoring role of public opinion.In this thesis, we will do deep research on Chinese Weibo sentiment classification, including sentiment polarity classification and sentiment fine-grained classification.(1) Sentiment polarity classification is studied. We classify the sentiment of Chinese Weibo into positive and negative, and improve some technology that involves. Firstly, information gain, the traditional method of feature selection does not consider the features’appearance in intra-class and inner-class, to solve the problem, we introduce two factors, concentration ratio and distributed ratio. Secondly, we consider Weibo’s characteristics when we calculate the feature weight, we combine the feature’s emotional information and location information with the traditional TF-IDF calculation method.(2) Sentiment fine-grained classification is studied. We classify the sentiment of Chinese Weibo into seven categories, including happiness, like, surprise, anger, sadness, fear and disgust. We analyze and improve the traditional methods. Firstly, we expends the existing multi-class emotional dictionary to compensate for its lack of words coverage. We make use of Weibo training corpus to generate candidate sentiment features, propose a sentiment feature selection TF-IDF method based on variance. Then calculate the feature’s category and emotional strength and add it into the multi-class emotional dictionary. Secondly, we calculate the Weibo fine-grained sentiment score based on the expanded emotional dictionary. In the process of Weibo sentiment fine-grained classification, we firstly classify the Weibo emotion into two polarities and then into fine-grained, finally propose a hierarchical-based algorithm for Weibo sentiment fine-grained classification.Experimental results show that the proposed sentiment polarity classification method and sentiment fine-grained classification method have a better result in accuracy, recall, and F Value than traditional method.