摘要
针对推荐系统中协同过滤技术面临的数据稀疏性和推荐实时性难以保证的问题,提出一种基于主成分分析(Principle component analysis)和SOM(Self-organizing map)聚类的混合协同过滤模型.首先对原始评分数据进行全局降维,并在转换后的主成分空间上进行用户聚类,缩小了目标用户的最近邻搜索空间,减少了在线计算时间复杂度,最后对真实的电子政务门户网站Log日志数据进行了几种常用的推荐算法的比较,实验结果证明新的推荐模型具有较好的预测精度.
To alleviate data sparsity and scalability issues of collaborative filtering technique in recom-mendation systems,a new hybrid collaborative filtering model based on Principle Component Analysis and Self-Organizing Map cluster method was proposed.In our approach,dimension reduction technique was first performed on whole data space.The clusters were generated from relatively low dimension vector space transformed by the first step,and then used for neighborhood selection in stead of searching in the whole user space,which can reduce the computation complexity in online recommendation.The experiments were based on web log data from E-government portal web site,and the results indicate that the proposed algorithm can provide better prediction accuracy compared with some exiting collaborative filtering algorithms.
出处
《系统工程理论与实践》
EI
CSSCI
CSCD
北大核心
2010年第10期1850-1854,共5页
Systems Engineering-Theory & Practice
基金
高等学校博士学科点专项科研基金(20020056047)
关键词
推荐系统
协同过滤算法
主成分分析
自组织映射
聚类技术
recommendation system
collaborative filtering
principle component analysis
self-organizing map
clustering technique
