Abstract: According to the high dimensionality and sparsity of rating matrix in traditional collaborative filtering recommendation system, a new collaborative filtering recommendation model based on Local Principle Component Analysis(LPCA) is proposed which combines taxonomy technique and local principle component analysis method to make dimension reduction for different subject genre respectively, and remains the real interested users in one specific subject of the Web pages which accelerates the neighbor searching process. Experiment on real log data indicates the new model can improve the predication quality.

Key words: recommendation system, collaborative filtering algorithm, dimensionality reduction, Local Principle Component Analysis(LPCA)

摘要： 根据传统协同过滤算法中用户数据的高维稀疏特点，提出一种基于局部主成分分析协同过滤推荐模型，采用基于语义分类和主成分分析的二阶段降维技术，分别对各类主题页面进行局部降维处理，以保留对某类主题真正感兴趣的用户群，加速最近邻的搜索过程。通过对真实Web日志数据的测试，证明该模型具有较高的预测精度。

关键词: 推荐系统, 协同过滤算法, 维数约简, 局部主成分分析

CLC Number: 

• N945