摘要
提出一种改进的基于对分网络的评分预测方法,首先将用户对项目的行为记录利用对分网络来表示,利用对分网络的结构特征来设计算法。算法综合时间因素、评分差以及网络的路径信息,挖掘用户-项目对分网络顶点之间的关联性,计算用户之间的相似度,利用谱聚类算法建用户聚类为兴趣组,最后利用邻居用户的评分信息预测用户对未知项目的评分。在标准数据库上验证此方法的有效性,结果证明,方法的平均绝对误差低于对比方法达0.07以上。
Propose a improved prediction method based on bipartite network. Firstly use a bipartite network to represent the user＇s behavior records, then design algorithm using structural features of bipartite network. In this method, time parameter, the difference between ratings and path information of the network are used to mine the correlation between the vertices in the user-item network to calculate the similarity between users. Users are clustered into interest groups based on spectral clustering.Finally, predict the user＇s rating on unknown item by neighbor users＇ information. Verify the validity of the method on the standard data set movielens. The experimental results show that the mean absolute error of proposed method is lower than the contrast methods by more than 0.07.
出处
《电子技术（上海）》
2015年第8期72-77,共6页
Electronic Technology
基金
国家科技支撑计划课题:电视商务综合体新业态运营支撑系统开发(2012BAH73F01)
中国科学院"NGB有线无线融合应用"重点部署项目子课题:"面向NGB的互联网视频访问控制应用示范(KGZD-EW-103-5(5))"
关键词
对分网络
评分预测
谱聚类
兴趣组
bipartite network
rating prediction
spectral clustering
interest group
