Research On Security Problem Of Personalized Collaborative Filtering Recommender Algorithm

Personalized recommender system is an important research issue of E-commerce information technology. The collaborative filtering recommendation is by far the most widely used and the most successful personalized recommendation technology. However, the behavior that users inject a lot of malicious noises into the recommender system presents a challenge to the security issue of the system with the ever-growing of E-commerce systems. How to ensure the trustworthiness of collaborative filtering recommendation in the face of malicious noise has become an urgent problem to be solved. Based on the comprehensive analysis of the current research for this area, in this paper the security issue of personalized collaborative recommendation algorithm has been further studied.Firstly, based on the deep analysis for the attack data and attack detection algorithms of collaborative recommendation environment, aims at the problems such as higher computational cost and lower detection accuracy, the concept of time concentrated characteristic is introduced into the attack detection. A normal cloud model and time concentrated characteristic based approach to measure the suspicious ratings is proposed. On the basis of that, to improve the algorithm precision and recall rate, an improved PCA attack detection algorithm based on normal cloud model is presented, and the related experiments were carried out.Secondly, in current personalized recommender systems, the user trust calculation method emphasizes too much on the users'recommended history information and ignores the true extent of the user ratings, which affects the accuracy of the recommendation. In order to effectively solve the problem, in this article user trust is decomposed into user rating trust and user recommended trust, and the concept of time concentrated characteristic is introduced. Based on the outstanding characteristics of Cloud Model on the process of transforming a qualitative concept to a set of quantitative numerical values and beta distribution, a user rating trust calculation method based on attack detection process and a user recommended trust calculation method based on beta distribution are proposed. Their combination can make the results of the calculation more accurate.Finally, on the basis of traditional collaborative filtering algorithm, the novel user trust calculation method proposed in this paper is introduced into the recommendation algorithm,and a trust personalized recommender algorithm based on user trust is presented. This algorithm can improve the credibility of the algorithm and its anti-attack capability. In order to test the correctness and effectiveness of the proposed algorithm, the compared experiments with other algorithms are also carried out.