摘要
协同过滤是目前推荐系统中最为成功的一种方法,但面临稀疏数据特征时存在冷启动、稀疏性、可扩展性等问题。提出利用Web数据挖掘(WUM)获取隐性数据对显性用户评价矩阵进行补值,应用径向基函数(RBFN)对补值后的评价矩阵进一步进行平滑处理,得到消除稀疏性后的完全评价矩阵,基于完全评价矩阵利用协同过滤技术对相似用户进行聚类并实施推荐。实验评价结果表明该方法与传统协同过滤推荐方法相比,无论在推荐精度还是推荐相关性上都更为有效。
Collaborative filtering is the most successful method in recommender system, but it has cold start, sparsity, scalability and other issues when faced with sparse data. Web Usage Mining（WUM） is proposed to obtain implicit data which can fill explicit user rat- ing matrix, and Radial Basis Function Network（RBFN） is used to smoothen filled rating matrix to get a complete rating matrix. Collab- orative filtering is used to classify similar users based on smoothed complete rating matrix and generate reconunendation. Experimental results show that compared with traditional collaborative filtering methods, the proposed method is more effective both in the accuracy or relevance of recommendations.
出处
《计算机工程与应用》
CSCD
2012年第9期22-26,共5页
Computer Engineering and Applications
基金
国家教育部青年基金项目(No.10YJC790182)
天津市教委项目
天津市高等学校人文社会科学研究项目(No.20112125)
关键词
推荐系统
协同过滤
网络数据挖掘
径向基函数
recommender system
collaborative filtering
Web Usage Mining
radial basis function network
