Research On Fake-review-groups Detection Based On Homogeneous And Heterogeneous Information Network

With the rapid development of the Internet,e-commerce platforms,have been integrated into the daily life of people and have changed people’s consumption pattern.All accounts organized to make false reviews and individual users who participate in the work of fake reviews are collectively referred to as the members of fake-review-groups in the review system of the e-commerce platform.Fake-review-groups use various methods such as manipulating a large number of e-commerce platform accounts or using rebates to instruct individual users to post fake reviews and profit from them,which greatly affects the fairness of the market and the authenticity of reviews.Domestic and foreign researchers and e-commerce platforms have conducted a series of research work on fake-review-texts and fake-review-users detection,and have achieved certain results.However,with the rapid development and changes in the mode of distributing fake reviews,research on the detection of fake-review-groups has become increasingly important.This paper conducts an in-depth study on the fake-review-groups detection mode and proposes two fake-review-groups detection algorithms based on homogeneous and heterogeneous information networks respectively.The algorithm based on the homogeneous information network constructs association features between users manually,designs association feature rules,uses these association features to build a homogeneous user information network with each user as a vertex,and finally obtains candidate groups by clustering.The other algorithm,which based on the heterogeneous information networks,makes full use of metadate while constructing user information networks.The multi-vertices type structure of heterogeneous information networks preserves the semantic information from the real world in the dataset to the maximum,hence,complex objects and their relationships can be captured in the process of obtaining candidate groups.By migrating and applying the spectral clustering method to the two kind of user information networks,we can mine the correlation similarity in homogeneous information network and heterogeneous information network,obtain multiple sub-candidate groups with higher relevance,and combined with the characteristics of fake-review-users and fake-review-groups,which are already existing and proposed by this paper,to further analyze the candidate and find fake-review-groups.This paper completed the construction of a homogeneous information network and a heterogeneous information network based on the Yelp dataset.The spectral clustering algorithm is used on the homogeneous information network,and the spectral clustering division is completed by solving the similarity matrix on the heterogeneous information network,so as to divide the two kinds of information networks,obtain candidate groups,and achieve fake-review-groups detection.The behavior characteristics were sorted out,and three fake-review-groups’ behavior characteristics,group extreme rating ratio,group repeated comment ratio and group rating deviation,were selected as clustering effect evaluation indicators.The algorithms proposed in this paper are compared with the existing fake-review-groups detection algorithms on the same dataset.The results verify the feasibility of the user homogeneous information network construction method proposed in this paper,and the accuracy of the two fake-review-groups detection algorithms proposed in this paper.