结合LDA和用户特征的协同过滤算法  

Collaborative filtering algorithm combining LDA and user characteristics

作　　者：梁静 葛宇 LIANG Jing;GE Yu(Computer Engineering School,Chengdu Technological University,Chengdu 611730,China;College of Computer Science,Sichuan Normal University,Chengdu 610066,China)

机构地区：[1]成都工业学院计算机工程学院,四川成都611730 [2]四川师范大学计算机科学学院,四川成都610066

出　　处：《计算机工程与设计》2021年第9期2519-2525,共7页Computer Engineering and Design

基　　金：四川省教育厅基金项目(18ZB0033)。

摘　　要：针对协同过滤推荐算法的冷启动和数据稀疏问题,提出一种结合LDA和用户特征的协同过滤算法。利用基于吉布斯采样的LDA主题模型生成项目-主题隶属概率矩阵,通过矩阵运算构造用户-主题评分数据,设计结合夹角余弦法的用户相似性计算方案,从概率角度论证方案处理稀疏数据的有效性;针对用户特征信息结合海明距离进行编码,设计冷启动用户相似性评价方案,提高冷启动用户相似性评价的合理性。基于MovieLens数据集的实验结果表明,所提算法在面临数据稀疏和冷启动问题时均有较好推荐效果,在最近邻个数较少时有较好表现。Aiming at the problems of cold start and data sparsity in collaborative filtering recommendation algorithm,a collaborative filtering algorithm combining LDA and user characteristics was proposed.LDA topic model based on Gibbs sampling was carried out on the project type set,project-topic membership probability matrix was generated,the user-topic rating data were constructed by matrix operation.The user similarity calculation scheme combined with the angle cosine method was designed.And the effectiveness of the scheme to deal with sparse data was proved using probability method.The user characteristic information was encoded by Hamming distance,a cold start user similarity evaluation scheme was designed.The rationality of cold start user similarity evaluation was improved.Experimental results of MovieLens dataset show that the proposed algorithm has better performance in dealing with data sparsity and cold start users,and has better recommendation effects when the number of nearest neighbors is small.

关 键 词：协同过滤 数据稀疏 冷启动 LDA(隐含狄利克雷分布) 矩阵运算 海明距离 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]