摘要
传统的协同过滤方法是一种有效的推荐方法,然而在应对冷启动用户或稀疏的评价矩阵时,系统的性能显著下降。为了提高推荐系统应对冷启动用户和稀疏数据的能力,本文提出了一种迭代式的基于核方法的协同过滤算法。首先根据用户对项的评价矩阵序列构建一个数据立方,并针对连续值、有序离散值和无序离散值三种类型定义3个核函数以及基于核函数的预测值估计器,最后根据相应的核函数估计器预测评价矩阵中的未知元素。实验结果表明,本文提出的算法与基于用户的和基于项的协同过滤方法相比误差小,并能更好的应对冷启动用户和稀疏的评价矩阵。
Collaborative filtering is an effective start users and sparse rating matrix, the performance method in recommender system. However, while dealing with cold of collaborative filtering decreases quickly. In order to improve the ability of handle cold start users and sparse rating matrix, this paper proposes an iterative collaborative filtering recommender algorithm based on kernel method. Firstly, builds a data cube according to the user-item rating matrixes, defines three kernel functions for continuous value, ordering discrete values and non-ordering discrete values, designs three kemel estimators based on these kernel functions, and finally, predicts unknown ratings with the proposed kernel estimators. The experiments show that, compared with user based and item based collaborative filtering methods, the proposed algorithm has less predicting error, and can handle cold start users and sparse rating matrix more effectively.
出处
《信息技术与信息化》
2014年第12期76-81,共6页
Information Technology and Informatization
关键词
协同过滤
核方法
推荐系统
Collaborative filtering Kernel method Recommender System Nearest neighbor
