Topic Model Based Review Spammer Group Detection

With the rapid development of Internet,the electronic commerce has also been developed.When shopping on the Internet,most of the consumers in advance before decide to buy a product to read the comments of the product information,so the product reviews for consumers to purchase decisions has a guiding role.But there are a lot of individuals or organizations in order to profit and reputation,by hiring some "water army" to write some false comments or to some unfair allocation to promote a product or to denigrate goals.Although there have been much efforts in this field,the problem is still left open due to the difficulties in gathering ground-truth data.As more and more people are using Internet in everyday life,group review spamming,which involves a group of fraudsters writing hype-reviews(promote)or defaming-reviews(demote)for one or more target products,becomes the main form of opinion spamming.In this paper,we propose a LDA-based computing framework,namely GSLDA,for group spamming detection in product review data.GSLDA has three phases.First,we adapt LDA(Latent Dirichlet Allocation)to the product review context in order to bound the group spammers into a small reviewer cluster.Then,we concentrate on the abnormal reviewers in each cluster and use SCAN to extract high suspicious reviewer groups from these clusters.Finally,the group of candidates was sorted in descending order to find the actual cheating group.Experiments on three real-world datasets verified the effectiveness of GSLDA,which outperforms several state-of-the-art baselines,e.g.,FraudEagle,SPEagle,GSBP,GSBC,etc.