Bootstrapping Both Product Features and Opinion Words from Chinese Customer Reviews with Cross-Inducing

Bootstrapping Both Product Features and Opinion Words from Chi- nese Customer Reviews with Cross-Inducing1 Bo Wang Institute of Computational Linguistics Peking University Beijing, 100871, China wangbo@pku.edu.cn Houfeng Wang Institute of Computational Linguistics Peking University Beijing, 100871, China wanghf@pku.edu.cn Abstract We consider the problem of 1 identifying product features and opinion words in a unified process from Chinese customer re- views when only a much small seed set of opinion words is available. In particular, we consider a problem setting motivated by the task of identifying product features with opinion words and learning opinion words through features alternately and it- eratively. In customer reviews, opinion words usually have a close relationship with product features, and the association between them is measured by a revised formula of mutual information in this paper. A bootstrapping iterative learning strategy is proposed to alternately both of them. A linguistic rule is adopted to identify low- frequent features and opinion words. Fur- thermore, a mapping function from opinion words to features is proposed to identify implicit features in sentence. Empirical re- sults on three kinds of product reviews in- dicate the effectiveness of our method. 1 Introduction With the rapid expansion of network application, more and more customer reviews are available on- line, which are beneficial for product merchants to track the viewpoint of old customers and to assist potential customers to purchase products. However, 1 Supported by National Natural Science Foundation of China under grant No.60675035 and Beijing Natural Science Foun- dation under grant No.4072012 it’s time-consuming to read all reviews in person. As a result, it’s significant to mine customer re- views automatically and to provide users with opinion summary. In reality, product features and opinion words play the most important role in mining opinions of customers. One customer review on some cell phone is given as follows: (a) “外型漂亮，屏幕大，拍照效果好。”(The appearance is beautiful, the screen is big and the photo effect is OK.) Product features are usually nouns such as “外 型” (appearance) and “屏幕” (screen) or noun phrases such as “拍照效果” (photo effect) express- ing which attributes the customers are mostly con- cerned. Opinion words (opword is short for “opin- ion word”) are generally adjectives used to express opinions of customers such as “漂亮” (beautiful), “大” (big) and “好” (well). As the core part of an opinion mining system, this paper is concentrated on identifying both product features and opinion words in Chinese customer reviews. There is much work on feature extraction and opinion word identification. Hu and Liu (2004) makes use of association rule mining (Agrawal and Srikant, 1994) to extract frequent features, the sur- rounding adjectives of any extracted feature are considered as opinion words. Popescu and Etzioni (2005) has utilized statistic-based point-wise mu- tual information (PMI) to extract product features. Based on the association of opinion words with product features, they take the advantage of the syntactic dependencies computed by the MINIPAR parser (Lin, 1998) to identify opinion words. Tur- 289 ney (2002) applied a specific unsupervised learn- ing technique …

Content truncated to 500 words