摘要
信息推荐系统主要根据已有的用户历史信息来对未知信息进行预测。但用户的活跃度往往使得数据集本身过于稀疏,从而使相关算法产生过拟合问题。跨域推荐算法是为了解决在单域推荐中常遇到的数据稀疏性问题,然而大多数的推荐算法在考虑共享信息时并未考虑单个数据域的个性信息。本文通过矩阵聚类方法来提取矩阵的潜在因式,区别数据集合之间的共享信息和自身信息。通过这种方法来做跨域推荐预测,并在几个现实中的数据集上与现有的一些推荐算法进行比较。
Recommender system aims to predict the unknown information based on the existing user history information. However, the inacitivity of users leading the data set too sparse and causing the overift problem of certain algorithms.Cross-domain recommendations are proposed to solve the problem of sparsity.While most of them consider the sharing message between domains,they ignore the individual information of each data set. In this paper, we ifnd the latent factor model by clustering the matrix and distinguish the common information and the different.In this paper,we compare our model with several current recommender system on several real data sets.
出处
《软件》
2013年第12期142-147,共6页
Software
关键词
算法理论
跨域推荐
潜在因式
个性信息
algorithm theory
cross-domain recommendation
latent factor
individual information
