基于K-medoids项目聚类的协同过滤推荐算法

针对传统协同过滤推荐算法通常针对整个评分矩阵进行计算,存在效率不高的问题,提出一种基于K-medoids项目聚类的协同过滤推荐算法.该算法根据项目的类别属性对项目进行聚类,构建用户的偏好领域,使用用户偏好领域内的评分矩阵进行用户间相似度的计算,得到目标用户的最近邻居集,并生成推荐结果.与常用的K-means聚类方法相比,采用K-medoids方法对项目类别属性进行聚类,不仅克服了评分聚类可靠性不高的问题,而且算法还具有更好的鲁棒性.实验结果表明,该算法能有效提高推荐质量.

Collaborative filtering recommendation algorithm based on K-medoids item clustering

In general, traditional collaborative filtering recommendation algorithms do the prediction computation based on the whole rating matrix, which leads to the low efficiency. To remedy this weakness, a collaborative filtering recommendation algorithm based on K-medoids item clustering is proposed. The proposed algorithm clustered the items according to the item category attributes, and then constructed the user preference domain. Only the rating matrix in the user preferences domain is used to calculate the user similarity and generates the nearest neighbor set of the target user and recommendation results. Different from the other K-means based clustering methods, the present K-medoids based clustering method focuses on the item category attributes, which overcomes the low reliability problem of using user ratings. Moreover, the present clustering method has better robustness. Experimental results show that the proposed algorithm improves the recommendation quality.