Unsupervised Weibo Sentiment Classification Based On Topic Sentiment Mixed Model

With the rise of the Internet,the rapid development of the social networking platform based on Weibo has been promoted.The number of Weibo comment texts has grown exponentially.Emotional information excavated from mass commentary texts is becoming more and more valuable in business planning and social application.At the same time,unsupervised microblog sentiment analysis relying on computers came into being.Different from traditional text sentiment classification,Weibo comment text has its unique characteristics in emotional expression.There are limits on the number of words in Weibo comments,so most of the comments are short texts and the number is large.Frequently,the text grammar is not standardized.New online words are easy to appear.Therefore,this paper studies the above characteristics of Weibo text,constructs a domain sentiment dictionary for microblog text,and proposes an unsupervised microblog sentiment classification model based on BTM theme model(W-BSTM).The sentiment dictionary constructed in this paper contains four parts:basic emotion dictionary,network language emotion dictionary,emoji emotion dictionary and domain expansion emotion dictionary.Based on the existing sentiment dictionary,a two-category basic sentiment dictionary is constructed.By observing and collating the microblog text,an emotional dictionary for network words and emojis in the text is constructed.An extended dictionary based on the HowNet semantic computing method is established to collect words that have no emotional tendency in traditional text but have emotional tendencies in the microblog text representation.The W-BSTM model is an unsupervised"sentiment-topic-term"three-layer Bayesian sentiment-topic model that adds an emotional layer to the BTM model and integrates the weight model.While retaining the original superiority of the BTM model,the model considers the importance of each feature word in sentiment classification and extracts the textual emotional information.Comparing the model with other thematic emotion models proves that the model has a good effect on short text sentiment classification.Finally,this paper uses Weibo crawler technology to get the comment text of Sina Weibo,and apply the emotional dictionary and W-BSTM model constructed in this paper to the micro-blog commentary sentiment analysis.The validity and feasibility of the unsupervised microblog sentiment classification method based on the topic sentiment hybrid model proposed in this paper is verified.