摘要
协同过滤中的冷启动问题是个性化推荐系统的一个难点,本文提出了一种基于用户特征分解的解决算法。首先构建了用户表示矩阵同时为保证不同用户评价的可比性,把每个用户的评价向量进行标归一化处理得到标准用户评价矩阵然后将两者进行合成得到用户特征对资源的标准评价矩阵。当新用户表示为用户特征向量,与特征评价矩阵合成得到新用户的预测评价向量,从而进行个性化推荐。基于MovieLens数据集进行的实验表明,该算法可以一定程度上解决系统冷启动问题,提高系统推荐质量。该算法还可以很方便地推广解决新资源的冷启动问题。
Cold -Start Problem in Collaborative Filtering is a difficult problem in personalized recommendation system. The new method is put forward based on users＇characteristic decomposition. Firstly, users＇characteristic representing matrix is designed, standard users＇ evaluative matrix is builded by unitary processing, touser, eualluative on parabinty of different usersis promised. Secondly, combination of users＇ characteristic denotation matrix and standard users＇ evaluative matrix can calculate standard evaluative matrix of characteristic - resource. New user can be denoted user＇ evaluative vector, forecasted evaluative matrix of new user can calculate by combining characteristic evaluative matrix and personalized recommendation can carryd out. The experimental results based on MovieLens data set showed that the improved algorithm could solve the Cold -Start Problem, and it can improve the accuracy of system recommendation significantly. The method solves cold - start problem of new resource effectvely.
出处
《山东农业大学学报（自然科学版）》
CSCD
北大核心
2013年第4期616-619,共4页
Journal of Shandong Agricultural University：Natural Science Edition
基金
山东省高等学校科技计划项目(J12LN73)
山东省艺术科学重点课题(2012445)
关键词
用户特征
协同过滤
冷启动
评价矩阵
特征表示矩阵
Users＇ Characteristic
collaborative filtering
cold - Start, evaluative matrix
characteristic representation matrix
