Research On Review Spammer Groups Detection Based On Heterogeneous Information Network

With the rapid development of Internet technology and e-commerce,which product reviews have become important reference information when people make decisions to purchase.Businesses in order to obtain considerable revenue,they usually hire multiple review spammers to make malicious reviews of their competitors' products to discredit the quality of their products or to make good reviews of their own products to increase product sales.These groups that are hired to make false comments are called review spammer groups.Compared with a individual spammer,review spammer groups are more hidden and make a greater threat to the e-commerce platform,which seriously affects the fairness of the e-commerce platform.In order to detect review spammer groups,a lot of detection methods have been proposed.However,these methods have the problems that they cannot make full use of the existing information and the detection accuracy is not high.This paper studies from the following two aspects to solve the above-mentioned problems.Firstly,to solve the problem of not being able to make full use of the existing information,this paper proposes an algorithm for detecting biased walk review spammer groups based on heterogeneous information networks.The algorithm establishes a heterogeneous information network with weight information based on user and product relationships;Use the method of biased walk on the meta path to generate low-dimensional vector.Obtain the candidate review spammer groups by using an improved K-Means algorithms.Calculate the suspicious degrees of review spammer groups based on the detection indicators to indentify review spammer groups.Secondly,to solve the problem that simple the existing nodes,this paper proposes a review spammer group detection algorithm based on the generative adversarial network.The algorithm extracts the purchase information in the data set and uses the adversarial network to perform negative sampling to generate nodes that are more similar to the original node;The clustering method is used to cluster the user node embeddings,and the candidate groups are sorted according to the indicators.Eventually obtain review spammer group.Finally,experiment on the algorithm proposed in this paper on Amazon dataset and Yelp dataset,and compare the experimental results with the existing detection group attack methods to verify the effectiveness of two detection method.