摘要
为了克服"维灾"所带来的问题,提出一种基于主成分分析的维数约简方法,并在转换后的低维向量空间上进行K-means聚类算法,以减少目标用户的最近邻搜索范围,代替在超高维空间上逐一寻找最近邻的过程。实验结果证明了新算法的有效性,特别在目标用户的历史评价信息较少的情况下,也能有较好的预测精度。
To address the curse of dimensionality, this paper proposed a new hybrid recommendation model which imposed principal components analysis technique combined with K-means clustering. In the approach, the clusters generated from the relatively low dimension vector space transformed by PCA step, and then used for neighborhood selection in order to alternate the exiting K-nearest neighbor searching in high dimensions. The experiment results indicate that the proposed model can produce better prediction quality and higher efficiency. Especially, when the target visitor with few historic information comes, it performs more robust.
出处
《计算机应用研究》
CSCD
北大核心
2009年第10期3718-3720,3762,共4页
Application Research of Computers
基金
高等学校博士学科点专项科研基金资助项目(20020056047)
关键词
协同过滤
主成分分析
维数约简
K-MEANS聚类
collaborative filtering
principle components analysis(PCA)
dimension reduction
K-means clustering