Research On Detection Of Review Spammer Group Based On Clique Percolation Method

The development of e-commerce has led to a sharp increase in the number of online reviews.Product review has become an important reference indicator for people's shopping and consumption.However,the number of spam reviews in the Internet has been increasing year by year,and its publishers are mostly professional review spammer groups.Although researchers at home and abroad have proposed many methods and models for detecting spam reviews and spammers,there are still many problems,such as low accuracy of manual annotation data,the failure of language features,low detection precision and detection lag.In order to solve the above problems,this thesis mainly studies the detection model of review spammer groups in online review websites.The main research contents include the following two aspects:1.Aiming at the low accuracy of manual annotation data and the failure of language features,an offline detection model of review spammer group based on Clique Percolation Method(CPM)is proposed,which uses a completely unsupervised way to detect review spammer groups.Firstly,build the review data into a reviewer network.Secondly,CPM is performed on the reviewer network,which innovatively treats each kclique cluster detected as a review spammer group.Then,construct spam indicators based on review behaviors to measure the suspiciousness of each review spammer group from two dimensions of individual and group,and generate a ranking-list of groups.Finally,comparison experiments are performed on three public datasets,and the results show that the detection precision and ranking quality of this model are higher than the four compared methods,and the larger the dataset,the better the performance of this model.2.Aiming at detection lag and low detection precision,an online detection model of review spammer group based on Incremental Clique Percolation Method is proposed.Firstly,detect review spammer groups in the initial static reviewer network based on classical CPM,and generate a ranking-list of groups based on behavioral spam indicators.Secondly,an Incremental Clique Percolation Method is proposed,which incrementally processes the dynamic changes of the network and updates the ranking-list in real time.Then,this model also provides a semi-supervised solution,which improves the detection precision by combining part of labeled review data.Finally,comparison experiments are performed on two public datasets,and the results show that the running time of this model is much shorter than the offline detection model.Although the detection precision and ranking quality are lower than that of the offline detection model,the detection precision and ranking quality can be significantly improved by combining with a few labeled review data.