Research On Shilling Attacks And Detection In Social Recommender Systems

Nowadays,social media is active in every corner of people's lives.The direction of social recommender by combining user social relations with traditional recommender methods has become a new research hotspot.Social recommender system uses social attribute information to improve the accuracy of recommender system,but social recommender system can be scored freely by outside users,so it is vulnerable to malicious attacks.This kind of behavior seriously affects the accuracy and fairness of social recommender systems,causing users to have a sense of distrust in recommender systems.Researchers at home and abroad have done a lot of research on scoring recommender system and social network attack detection,but little attention has been paid to the problem of trust attack detection for social recommender systems.Although some researchers have proposed detection methods for trust attack of social recommender systems,these methods lack of mining the relationship between user scoring behavior and user social relations.In view of these shortcomings,this paper studies the problem of toe attack detection in socialized recommender systems.Firstly,by studying the differences of rating behavior and social behavior between the attacker and the real user,this paper proposes a decision tree-based detection algorithm for the attacker.Firstly,the algorithm extracts user features according to user's social behavior,forms user feature space,and classifies users with C4.5 decision tree algorithm.Secondly,by analyzing the similarity of rating behavior among users and the social behavior of users,this paper proposes a token attack detection algorithm based on tag propagation.This algorithm improves the weight calculation rules of tag propagation,so that users with similar rating behavior can be divided into groups with the same tag in the process of tag propagation,and thus the token attack users can be detected.Finally,the experimental results on FilmTrust dataset show that the proposed detection algorithm can effectively detect mixed attacks against social recommender systems.