基于用户项目属性偏好的协同过滤推荐算法-《计算机技术与发展》

文章信息/Info

Title:: A Robust Collaborative Filtering Recommendation Algorithm Based on User Preference of Item Attributes

Author(s):: LYU Cheng-shu; School of Management Science and Engineering,Dongbei University of Finance and Economics,Dalian 116025,China

Keywords:: user preference of item attributes; user synthetical similarity; shilling attack; collaborative filtering; recommender system

摘要:: 协同过滤推荐系统是广泛应用的推荐技术之一,但是其面临着推荐精度低和托攻击问题。为了提高传统协同过滤推荐系统的推荐精度和抗攻击能力,提出了一种基于用户项目属性偏好的鲁棒协同过滤推荐算法。该算法在用户相似性计算过程中引入用户项目属性偏好相似性,并通过动态加权因子与传统的用户评分相似性进行组合,获得用户的综合相似性,在用户共同评分项匮乏的情况下也可以根据相同的项目属性偏好度量用户相似性,缓解评分数据稀疏性。在预测评分阶段,根据用户项目属性偏好类型条件过滤最近邻集合中的攻击概貌,消除攻击概貌对评分预测的不良影响,提高系统的抗攻击能力。实验结果表明,该算法可以有效提高推荐系统的推荐精度和抗攻击能力。

Abstract:: Collaborative filtering recommendation system is one of the most widely used recommendation technologies,but it owns the disadvantages of lower recommendation precision and shilling attack. In order to improve the recommendation precision and capability of attack resistance for traditional collaborative filtering recommendation system,we present a robust collaborative filtering recommendation algorithm based on user preference of item attributes. First,user preference similarity of item attributes is introduced in the calculation of
user similarity and synthesized with traditional item rating similarity through dynamic weighed factor to get users， synthetical similarity.The user similarity can be also measured according to the same item attribute preference in the case of lacking user，s common score,so as to alleviate the sparsity of the scoring data. While predicting the rating for the target user,the user preference of item attributes is used to filter shilling attack in nearest neighbor set,eliminating adverse impact of the shilling attack to predicted rating,improving the system ability against the shilling attack. The experiments show that the proposed algorithm can effectively enhance the recommendation precision and anti-attack capability of recommendation system.