基于协同谱聚类的推荐系统托攻击防御算法

Shilling Attack Defense Algorithm for Recommender System Based on Spectral Co-Clustering

提出了一种基于协同谱聚类的推荐系统托攻击防御算法.该算法首先使用谱聚类方法对协同聚类算法进行改进,以在用户和项目2个维度上同时进行聚类;接着在聚类基础上结合分级偏离平均度对用户进行项目推荐.实验测试结果表明,在同等托攻击规模的情况下,该算法可以降低实施托攻击的用户和攻击数据对系统推荐结果的影响.

An algorithm for recommender system based on spectral co-clustering was proposed to defend shilling attacks. The proposed algorithm maintains spectral clustering and co-clustering priors and allows a mixed membership in user and item clusters. The rating deviations were used for mean agreement based on the co-clustering results to recommend for users. Experimental results demonstrated that under the same shilling attack dimensions,our algorithm could decrease the shilling attack affects to recommender systems apparently.