Semi-supervised Sentiment Classification Method Based on Weibo Social Relationship

Abstract

Sentiment classification of microblog text is one of the hotspots and important research issues in text sentiment analysis. Aiming at the problem that the existing researches mostly assume that the micro-blog sentiments are independent of each other and have strong dependence on the training set, a semi-supervised sentiment classification method based on Weibo social relationship is proposed. The method utilizes the user’s theme sentimental consistency and the approval of social relationships (like and repost) in Weibo to establish the sentimental relationship between microblogs to solve the problem that microblog sentiments are independent of each other. Semi-supervised sentimental classification model is constructed by establishing the sentimental relationship between labeled micro-blog and unlabeled micro-blog, which reduced the dependence on training set. Specifically, the semi-supervised sentiment classification method was constructed by constructing a microblog sentimental relationship matrix using the Laplacian matrix of the above microblog social relationship graph, and adding to the text content based classification model. Climbing the real dataset of Sina Weibo for experiment, the experimental results showed that the method is superior to other typical sentiment classification methods in terms of accuracy and recall rate. The validity of this method is verified and the dependence on training data set is reduced to a certain extent.