Research And Implementation Of Recommender Systems Security Detection Technology In Mobile Social Network

With the explosion of data,the personalized recommendation technology was brought into being.Meanwhile,the rapid development of e-commerce makes the safety of recommendation systems highly valued by industry and academia.Malicious users probably benefit from the behavior of injecting a great quantities of fake profiles into the recommendation system to reduce the recommended frequency of specific items.Therefore,in recent years,how to detect recommended attacks from the shilling attacks and make malicious out to ensure the reliability of the recommended service has become a hot research.This thesis analyzes the different types of recommendation system.By comparing the characteristics of the single attack model and the group attack model,we study the user profiles of several attack models,such as average attack.Based on the previous research,a new security detection algorithm and a new security detection model are proposed.The main work of this thesis is as follows:1.Different recommendation systems and two classical collaborative filtering recommendation systems are studied in detail.And then,the advantages and disadvantages of different recommendation systems and the bottleneck of recommendation system development are analyzed.And also,this thesis analyzes the attack problem of recommendation system,and studies the user profile structure of different attack patterns,and focuses on the group attack model.Based on the GSAGenl group attack model,the AGSA is proposed.Last,the influence of the AGSA model on the recommendation system is analyzed through experiments..2.Combining the extended Kalman filter(EKF)algorithm and the linear discriminant analysis(LDA)algorithm,a data tracing detection algorithm is proposed.It utilizes a large amount of effective data aggregation and applies it to big data environment.By comparison experiments,it proves that the data tracing detection algorithm can achieve more efficient and secure detection services.3.According to the time characteristics of user rating change,some new basic detection attribute are proposed.Based on the proposed basic detection properties,the particle filter algorithm is applied to the detection system of recommendation system for the first time.Based on the characteristics of anti-similarity with AGSA model,a detection model based on universal gravitation is proposed to eliminate the Pearson 's correlation coefficient to judge the attacking profile of injection.Furthermore,the detection algorithm can be used for on-line detection.The experimental results show that the detection model presented in this thesis has good performance on accuracy and the recall rate in the face of AGSA group attack model.It proves that the detection model has excellent detection capability for group attack.