Research On Review Spammer Groups Detection Algorithms Based On Label Propagation

Today,product reviews play a vital role in consumers' purchasing decisions.More praise can bring considerable profit for the businessman.Consumers' dependence on commentary information and business desire for interest catalyzes the emergence of fake reviews.In order to make false information much more credible,the publishers of fake reviews usually work together to oversell the target product.This group is called a false review group.The organizational form of this review spammer group is highly hidden and difficult to detect,and have a greater impact on consumer buying propensity.Since the low detection precision of the detection method on the review spammer group that cannot reflect the difference of different indicators,this paper has done some effective researches on these problems.Firstly,based on the comment time association and comment score association,we proposed an algorithm on the detection of review spammer group by the label propagation of core graph.According to the algorithm,we first extract the association between the time and score,and establishes a reviewer relationship graph.And then we find the core map on the reviewer relationship graph.Next,we discover the candidate group by using the label propagation rule with high intensity and automatic filtering mechanism;Finally,through multiple review spammer group detection indicators,we get suspicious scores of each group,and then we sort and identify it.Secondly,when it comes to multiple evaluation indicators,the previous detection method takes the average value as the group suspiciousness,which cannot reflect the difference of different indicators,to solve this problem,this paper proposes a review spammer group detection algorithm based on the index weight measurement.The algorithm will use the above algorithm to obtain the review spammer group candidate group.After sorting the group,the two methods of determining the multi-index weights,namely the entropy method and the analytic hierarchy process,are combined to construct the ranking algorithm of the metric weights.This algorithm weakens the excessive objectivity of the entropy method and the extreme subjectivity of the analytic hierarchy process.The candidate groups are sorted using this sorting algorithm to identify review spammer groups.Finally,experimenting on Amazon and Yelp datasets and comparing with some existing algorithms.