Feature-level sentiment analysis for Chinese product reviews

The sentiment analysis for English product reviews has been widely researched in recent years, followed with many important achievements. Due to the special language traits of Chinese, the study on Chinese product reviews is much more difficult than the former. In this work, we focus on the finer-grained sentiment analysis for Chinese product reviews, that is feature-level based sentiment analysis. We propose a hybrid method which combines association rules and point-wise mutual information to extract the product features, and then take advantage of the sentiment dictionary - HowNet to analyze the opinion orientation expressed on the product features. The experiment result obtained shows the effectiveness and efficiency of our approach.