An Automatic Product Features Extracting Method in Chinese Customer Reviews

Product features extraction, as a subtask of the feature-level sentiment analysis, aims to discover features of products from text descriptions. In this paper, we propose a novel unsupervised approach together with some language rules to extract product features from Chinese customer reviews. The extraction system can automatically extract an initial seed list that serves as training data to get the relationship of product features and opinion words. This relationship combines both their distribution in corpus and their link information. According to it, new attributes of products can be extracted with high precision. The proposed method can be applied to different domains without any domain related training corpus or several domain related seed words. Our experiment results show that the algorithm gained good and stable performance in different domains.