一种属性和评分的协同过滤混合推荐算法  

A Collaborative Filtering Hybrid Recommendation Algorithm for Attribute and Rating

作　　者：李克潮[1] 蓝冬梅[1] 

机构地区：[1]广西民族师范学院图书馆,广西崇左532200

出　　处：《计算机技术与发展》2013年第7期116-119,123,共5页Computer Technology and Development

基　　金：2012年广西教育科研项目(201204LX481);2011年广西民族师范学院科研项目(XYYB2011030)

摘　　要：传统协同过滤推荐算法仅仅根据稀疏的评分矩阵向用户推荐,存在推荐质量不高的问题。提出了一种属性和评分的协同过滤混合推荐算法。该算法由项目的类别属性计算项目之间基于属性的相似性,考虑到用户兴趣随时间的变化,构建评分时间权重的指数函数,并应用到项目之间的Pearson相关相似性中。通过权重因子加权项目之间基于属性的相似性和项目之间的Pearson相关相似性,然后计算基于项目属性的评分预测。描绘职业分类树,构建职业相似性模型,并与性别加权结合产生用户综合属性的相似性,得到基于用户属性的评分预测。最后,综合两者计算混合评分预测。在Movielens实验数据集下,实验结果表明提出的算法具有较好的平均绝对误差。Traditional collaborative filtering algorithm exists poor recommendation quality for recommending to the user based solely on sparse rating matrix. Propose a collaborative filtering hybrid recommendation algorithm for attribute and rating. The algorithm computes similarity based on attribute between item by category attributes of item, takes user＇ s interests change over time into account, builds expo- nential function based on weight of rating time, and applies to Pearson correlation similarity between item. Weighted similarity based on attribute between item and Pearson correlation similarity between item by weighting factor, then calculated rating prediction based on item attribute. Depict professional classification tree, builds profession similarity model, and gets similarity of user＇ s combined attribute by weighted sex, then obtains rating prediction based on user＇ s attribute. At last, compute combined rating prediction by integrated the two above. Under experimental data set of Movielen, experimental results show that the proposed algorithm has a better mean absolute error.

关 键 词：协同过滤 职业分类树 综合相似性 推荐算法 

分 类 号：C931.9[经济管理—管理学]