Font Size: a A A

Study Of Trust-Based Detecting Model In Secure Recommender Systems

Posted on:2010-09-15Degree:MasterType:Thesis
Country:ChinaCandidate:Q ZhangFull Text:PDF
GTID:2178360275470214Subject:Computer system architecture
Abstract/Summary:PDF Full Text Request
Traditional recommender systems based on collaborative filtering play an increasing role in filtering information in an overloaded information system. Recommender systems could be applied in grid environment to help grid users select more suitable services by making high quality personalized recommendations. Also, recommendation could be employed in the virtual machines managing platform to measure the performance and creditability of each virtual machine. The most popular types of algorithms for collaborative filtering (CF) are user-based algorithms and item-based algorithms. Besides, other factors such as trustworthiness of users are taken into consideration in guiding recommendation.However, such systems have been shown to be vulnerable to profile injection attacks (shilling attacks), attacks that involve the insertion of malicious profiles into the ratings database for the purpose of altering the system's recommendation behavior. A variety of attack models have been identified in prior work, from two basic attack models (random attack model, average attack model) to intricate attack models such as bandwagon attack model and segment attack model. Recent research has also proved that most popular algorithms employed in current CF applications can be rather easily manipulated through biased profiles, resulting in decreasing overall user satisfaction with the system.Recent efforts in recommender area have focused on reducing the impact of profile injection attacks. Through detecting and discounting suspected attack profiles, RDMA is an effective metric for analyzing rating patterns of malicious users and evaluated their potential for detecting such attacks. In this paper we also investigate the use of statistical metrics to reveal rating patterns of shilling attackers. Different from previous work, the most outstanding feature of our new introduced metrics is based on trust. We discuss the building process of trust, make some adjustments according to the feature of attackers and experimentally evaluate these metrics on average attack model. Also, we develop an algorithm that makes use of theses attributes to detect and isolate malicious attackers. Additionally, we describe how these new trust-based metrics can be incorporated into already existing detecting metrics such as RDMA at profile-level and item-level, respectively. In the end, for measuring our approach's performance, we use standard evaluation measurement to evaluate the effectiveness of our algorithms.
Keywords/Search Tags:Recommender Systems, Collaborative Filtering, Profile Injection Attack, Trust
PDF Full Text Request
Related items