Extracting Service Aspects from Web Reviews

Experimental results show graph-based method outperforms FP-tree method and state-of-the-art machine learning methods such as CRFs and MLN (Markov Logic Network) performs well compared to other three methods.

Abstract

Web users have published huge amounts of opinions about services in blogs, Web forums and other review friendly social websites. Consumers form their judgements to service quality according to a variety of service aspects which may be mentioned in different Web reviews. The research challenge is how to extract service aspects from service related Web reviews for conducting automatic service quality evaluation. To address this problem, this paper proposes four different methods to extract service aspects. Two methods are unsupervised methods and the other two methods are supervised methods. In the first method, we use FP-tree to find frequent aspects. The second method is graph-based method. We employ state-of-the-art machine learning methods such as CRFs (Conditional Random Fields) and MLN (Markov Logic Network) to extract service aspects. Experimental results show graph-based method outperforms FP-tree method. We also find that MLN performs well compared to other three methods.