Sentiment Analysis for Product Features in Chinese Reviews Based on Semantic Association

Often there are more than one topic, more than one holder and more than one sentiment in an opinion sentence. But currently the studies on sentiment analysis are unable to identify exactly which sentiments a holder expresses on a given topic, so these are experimented based on the hypothesis, which is obviously unreasonable, that a review or a sentence have only a holder and a topic. The paper aims at addressing sentiment analysis for product features in product reviews written in Chinese by building the semantic associations between product features and sentiment words. The patterns, which identify semantic associations between sentiment words and features, and the approach, which builds the sentiment list of each feature, are proposed. The sentiment list of each feature is built up in the course of building semantic associations related to the feature. Using sentiment lists, the average sentiment expressed on each feature can be obtained. Our method is unsupervised. The result shows that our method is encouraging.