Abstract

Online review can help people getting more information about store and product. The potential customers tend to make decision according to it. However, driven by profit, spammers post spurious reviews to mislead the customers by promoting or demoting target store. Previous studies mainly utilize rating as indicator for the detection. However, these studies ignore an important problem that the rating will not necessarily represent the sentiment accurately. In this paper, we first incorporate the sentiment analysis techniques into review spam detection. The proposed method compute sentiment score from the natural language text by a shallow dependency parser. We further discuss the relationship between sentiment score and spam reviews. A series of discriminative rules are established through intuitive observation. In the end, this paper establishes a time series combined with discriminative rules to detect the spam store and spam review efficiently. Experimental results show that the proposed methods in this paper have good detection result and outperform existing methods.