GSCPM: CPM-Based Group Spamming Detection in Online Product Reviews

Abstract

Online product review is becoming one of important reference indicators for people shopping, but the current product review site contains a lot of fraudulent reviews. Group review spamming, which involves a group of fraudulent reviewers writing a lot of fraudulent reviews for one or more target products, becomes the main form of review spamming. However, solutions for group spammer detection are very limited, and due to lack of ground-truth review data, this problem has never been completely solved. In this paper, we propose a novel three-step method to detect group spammers based on Clique Percolation Method (CPM) in a completely unsupervised way, called GSCPM. First, it utilizes clues from behavioral data (timestamp, rating) and relational data (network) to construct a suspicious reviewer graph. Then, it breaks the whole suspicious reviewer graph into k-clique clusters based on CPM, and we consider such k-clique clusters as highly suspicious candidate group spammers. Finally, it ranks candidate groups by group-based and individual-based spam indicators. We use three real-world review datasets from Yelp.com to verify the performance of our proposed method. Experimental results show that our proposed method outperforms four compared methods in terms of prediction precision.