A Novel Feature-based Method for Opinion Mining in Chinese Product Reviews

Due to the rapid growth of web resources in the digital era, user comments have become ubiquitous in most websites. In China, the most prevalent websites include user-generated content site such as DOUBAN, review site like TAOBAO and JD, and various Blog sites. Existing research focuses on the opinion mining area using two main methods, including the supervised learning method and the semantic-based method, both of which have been proven as effective approaches. This paper reports our efforts in improving the work done by previous research. First, a few dependency relations are constructed to identify features and the opinion set. In order to extract features effectively, various combinations of digits and letters are also taken as features. Next, a novel distance method is utilized to identify features and opinion tuples. Finally, we build a new feature model using supervised learning method to detect the overall opinion polarity of a review. With the comparison to traditional methods, the experimental results indicate that our approach is more effective than the others.