Monitoring online reviews for reputation fraud campaigns

Abstract Online reviews are critical for both purchasers and sellers in the era of E-commerce. Praiseful reviews and/or 5-star ratings can yield remarkable profit gains, on the other hand, a bad-mouth review or a low rating score often incurs sales decrease. Therefore, fake review detection has attracted lots of research interests in recent years. While most existing approaches detect fake reviews in an offline fashion, i.e., finding suspicious reviews from a large volume of historical data, few efforts have been made to detect fake reviews in an online fashion, i.e., detecting suspicious reviews in review data streams. Online detecting fake reviews has many more benefits than offline detection in that the damages of fake reviews can be significantly reduced by removing them as early as possible. In this paper, we propose a novel online monitoring technique for detecting reputation fraud campaigns in product reviews. The technique includes two phases. First, it monitors online reviews to generate the most abnormal review subsequences (MARSs), which can be considered as candidate reputation fraud campaigns. Second, conditional random fields are exploited to label each review in a MARS as fake or genuine. Experiments show that our proposed methods are highly effective and efficient, with many advantages compared with existing online detection approaches.