Collective Opinion Spammer Groups Detection In LBSN

In recent years,with the rapid development of mobile terminal locating technology and mobile Internet technology,location-based social network(LBSN)platform has achieved great success.Location feature in LBSN has built bridges between the virtual social space and the real behavior space.It could fuse the online relationship with the offline behavior.Users can rely on online network to post comments on spatial locations.In the meantime,they can rely on these reviews to find new places to explore or consume.However,there are a variety of fake reviews in massive amounts of information in LBSN.Unfortunately,the publishers of those reviews are mostly professional opinion spammer groups.Such groups change the reputation of the locations through the release of a number of fake comments,which affects users'decision-making.Further more,those groups seize illegal interests for the businesses and undermines the network environment,resulting in serious affect on user experience and network reputation.In view of the insufficient of current research on the opinion spammer groups detection in LBSN and the incomplete feature mining of opinion spammer groups,a novel opinion spammer groups detection model based on Markov random field is proposed in this paper.The model fuses abnormal characteristics in five aspects,including the group members,target locations,fake reviews,collusion between members and competition between locations.In addition,combined with the topological relation of LBSN,opinion spammer groups detection algorithm is designed.The aim is to detect the opinion spammer groups and their target locations more accurately and comprehensively.Firstly,we extracts multi-angle anomalous features of collective opinion spammer groups including group members,target locations,groups' reviews,the competitive relationships between locations and collective relationships between group members based on the multi-dimensional attributes of LBSN.Secondly,combining the multi-angle anomalous features and the network topology,we constructs a detection model based on Markov random field with differential relevance,and transform the detection problem into labeling problem.Besides,based on Markov random field detection model,collusion spammer detection algorithm is designed.Based on the topology structure,the fake reviewers and the suspicious locations in the groups are labeled.Finally,based on the abnormal feature of collective relationships between group members,the opinion spammer groups detection algorithm is designed to achieve the purpose of detecting the groups.In order to verify the performance of the proposed algorithm and analyze the effectiveness of the algorithm,two algorithms proposed in this paper are validated on the dataset of Phoenix in Yelp.Through the comparison and analysis of the experimental results,it can be concluded that:(a)detection algorithm proposed in this paper can detect spammers and the suspicious locations more effectively than other algorithms.(b)the features extracted in this paper play a role in the detection.(c)detection algorithm proposed in this paper can effectively explore potential opinion spammer groups.Therefore,for the detection of opinion spammer groups in LBSN,the algorithm proposed in this paper has the best overall effect.