基于隐式反馈LDA模型的协同推荐算法研究  

Research on Collaborative Recommendation Algorithm Based on Implicit Feedback LDA Model

作　　者：翟航天 汪学明[1] ZHAI Hang-tian;WANG Xue-ming(School of Computer Science and Technology,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学计算机科学与技术学院

出　　处：《计算机技术与发展》2019年第6期7-12,共6页Computer Technology and Development

基　　金：国家自然科学基金([2011]61163049);贵州省自然科学基金(黔科合J字[2014]7641)

摘　　要：传统基于用户和基于标签的协同过滤推荐算法仅采用用户评分显式信息进行分析,浪费掉大量的隐式反馈数据。为将隐式反馈数据加以利用,提出一种用户隐式反馈数据与资源标签相结合的协同过滤推荐算法。对资源-标签利用GibbsSampling算法进行采样分析,挖掘推荐系统中资源的主题并建立LatentDirichletAllocation(LDA)模型,将隐式反馈数据中的用户行为赋予主题标签以此获取用户标签偏好,并与资源标签计算出的资源相似度相结合,预测用户个性化偏好。在Retailrocket网站行为数据集上的实验结果表明,相较于传统基于隐式反馈和基于标签的协同过滤推荐算法,该算法能有效地解决用户标签模糊和资源主题分析存在偏差的问题,提高个性化推荐准确度。Traditional user-based and label-based collaborative filtering recommendation algorithms only use user ratings explicit information to analyze and waste a lot of implicit feedback data. In order to make use of implicit feedback data,we propose a collaborative filtering recommendation algorithm combining user implicit data with resource labels. The resource label is sampled and analyzed by Gibbs Sampling,the theme of the resource in the recommendation system is excavated and the Latent Dirichlet Allocation (LDA) model is established. The user behaviors in the implicit feedback data are given to the subject label to obtain the user label preference,and the resource similarity calculated by the resource label is combined to predict the user personalized preferences. The experiment on the Retailrocket website behavior data set shows that compared with the traditional implicit feedback and label-based collaborative filtering recommendation algorithm,the proposed algorithm can effectively solve the problem of user label blur and resource topic analysis deviation,and improve the accuracy of personalized recommendation.

关 键 词：隐式反馈 标签采样 LDA建模 协同过滤 个性化推荐 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]