基于项目簇偏好的用户聚类算法  

A Clustering Users Algorithm Based on Users Preference for Item Sort

作　　者：刘芳先[1] 李国[2] 

机构地区：[1]广州城建职业学院机电与信息工程学院,广州510925 [2]湖北省农村信用社联合社信贷风险管理系统二期项目组,武汉430071

出　　处：《现代计算机》2013年第12期7-10,14,共5页Modern Computer

摘　　要：传统算法基于用户项目评分矩阵来进行推荐,存在冷开始、稀疏性等问题,邻居相似性只鉴于用户共同评价的项目,没有考虑项目本身的属性关系;在整个用户空间搜寻最近邻居,实时性差。针对这些问题,提出基于项目簇偏好的用户聚类算法,首先基于项目属性特征对项目进行聚类,然后再利用用户对项目簇的偏好对用户进行聚类,最后在和目标用户最相似的几个聚类中搜寻邻居用户,从而压缩搜寻空间,提高了搜寻速度。实验表明,该算法通过降低稀疏性、冷开始等问题,增强实时性,提高预测精度。The traditional collaborative filtering recommender algorithms are presently limited to the useritem rating matrix,which suffers from sparsity and cold-start problems.Neighbours＇ similarity only thinks about items which users evaluate together,but ignores the correlation of item attribute and user characteristic.In addition,the real-time nature is extremely bad,when they seek the nearest neighbors online in the whole matrix space.According to above problems,proposes a clustering users algorithm based on users preference for item sort.The algorithm firstly clusters items based on attributes,and gets users preference for item sort.Then it clusters users based on usrs preference.We can find the user＇s nearest neighbor from several nearest clusters to avoid the entire users base,and enhance the real-time response speed.Experimental results show that the optimized algorithm can increase prediction precision effectively,by way of reducing sparsity and cold-start problems,enhancing real-time nature.

关 键 词：个性化推荐 协同过滤 基于特征属性的项目聚类 基于项目簇偏好的用户聚类 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]