摘要
针对托攻击存在情况下推荐系统面临的数据稀疏性问题,提出一种融合k-距离和项目类别信息的鲁棒推荐算法.首先,根据离群点检测思想提出基于k-距离的用户可疑度计算方法,用来度量系统中每个用户是攻击用户的可疑程度大小;然后,将用户可疑度与项目类别信息相结合构建一种缺失值填充方法,对用户评分矩阵缺失评分进行填充;最后,基于填充后的评分矩阵,将用户相似度和可疑度进行加权组合,为目标用户选取可靠邻居,完成对目标用户的鲁棒推荐.在Movie Lens数据集上的实验结果表明,本文提出的方法能够有效解决推荐系统的数据稀疏性问题,提高推荐精度并具有较好的鲁棒性.
Under the condition of shilling attacks,the existing collaborative recommendation algorithms are facing the problem of data sparsity. To address this problem,in this paper we propose a robust recommendation method fusing k-distance and item category information. Firstly,according to the idea of outlier detection,we propose a k-distance-based method to compute user suspicion degree. The user suspicion degree can be used to measure the possibility of a user as an attacker. Secondly,we incorporate user suspicion degree with the item category to fill the missing values in the user rating matrix. Finally,based on the filled rating matrix,we combine the user similarity with the user suspicion degree to select reliable neighbors and make robust recommendations for target user. The experimental results on the M ovie Lens dataset show that the proposed method can solve the data sparsity effectively and outperforms the existing methods in term of both recommendation accuracy and robustness.
出处
《小型微型计算机系统》
CSCD
北大核心
2017年第11期2476-2481,共6页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61379116)资助
辽宁省教育厅科学研究项目(L2015240)资助
关键词
推荐系统
鲁棒推荐
托攻击
稀疏性
k-距离
recommender systems
robust recommendation
shilling attacks
sparsity
k-distance
