基于数据非随机缺失机制的推荐系统托攻击探测

Detecting Shilling Attacks in Recommender Systems Based on Non-random-missing Mechanism

协同过滤推荐系统极易受到托攻击的侵害.开发托攻击探测技术已成为保障推荐系统可靠性与鲁棒性的关键.本文以数据非随机缺失机制为依托,对导致评分缺失的潜在因素进行解析,并在概率产生模型框架内将这些潜在因素与Dirichlet过程相融合,提出了用于托攻击探测的缺失评分潜在因素分析(Latent factor analysis for missing ratings,LFAMR)模型.实验表明,与现有探测技术相比,LFAMR具备更强的普适性和无监督性,即使缺乏系统相关先验知识,仍可有效探测各种常见托攻击.

Collaborative filtering recommender systems are highly vulnerable to shilling attacks. Developing detection techniques against shilling attacks has become the key to guaranteeing both the reliability and robustness of recommender systems. Through revealing the latent factors invoking missing ratings under the non-random-missing mechanism, and further combining these latent factors with Dirichlet process in the framework of probabilistic generative model, this paper proposes a latent factor analysis for missing ratings （LFAMR） model for attack detection. Experimental results show that comparing with the existing detection techniques, LFAMR is more universal and unsupervised, and that it can effectively detect shilling attacks of typical types even in lack of system-related prior knowledge.