基于用户声誉的鲁棒协同推荐算法

随着推荐系统在电子商务界的快速发展以及取得的巨大经济收益, 有目的性的托攻击是目前协同过滤系统面临的重大安全威胁, 研究一种可抵御攻击的鲁棒推荐技术已成为目前推荐系统领域的重要课题.本文利用历史记录得到用户声誉, 建立声誉推荐系统, 并结合协同过滤推荐领域内的隐语义模型, 提出基于用户声誉的隐语义模型鲁棒协同算法.本文提出的算法从人为攻击和自然噪声两个方面对系统的鲁棒性进行了改善.在真实的数据集 Movielens 1M 上的实验表明, 与现有的鲁棒性推荐算法相比, 这种算法具有形式简单、可解释性强、稳定的特点, 且在精度得到一定提升的情况下大大增强了系统抵御攻击的能力.

Robust Collaborative Recommendation Algorithm Based on User's Reputation

With the rapid development of recommender systems in e-commerce industry, such systems bring huge economic profits. As a consequence, shilling attacks pose a significant threat to the security of collaborative filtering recommender systems. Developing a kind of robust recommendation technology which can resist attacks has become an important issue in the field of the recommender system at present. In this paper, a reputation recommender system is built by user reputations which are obtained from the user historical records. Utilizing the latent factor model in the field of collaborative filtering recommendation, a novel robust collaborative recommendation algorithm based on user reputations is proposed. The algorithm improves the system''s robustness from two aspects of shilling attack and natural noise. Empirical results on Movielens 1M dataset demonstrate that compared with the existing robust recommendation, this algorithm is very effective. Characterized by simplicity, interpretability and stability, the algorithm has strong ability to resist the system attack along with the accuracy getting a certain improvement.