A Topic-Independent Hybrid Approach for Sentiment Analysis of Chinese Microblog

Abstract

People's attitude towards specific events is usually contained in their Internet speech. When monitoring public opinions on the Internet, the sentiments of social media users should be analyzed in real time. For example, the expression of target user should be analyzed to get his/her emotional changing trend. However, present literatures on text sentiment analysis are limited to specific domains and topics, because they usually employ machine learning method to get sentiment polarity, which is trained on one specific topic area. In this paper, our approach combines the lexicon-based with the similarity-based method to extract sentiment word, then utilize the semantic rules and emoticons to obtain the sentiment polarity of short text. The results show that the proposed approach can get higher accuracy than the SVM method on topic-independent corpus and can be applied to online sentiment analysis.