张量分解的标签推荐及新用户标签推荐算法  

Tag Recommendation and New User Tag Recommendation Algorithms Based on Tensor Decomposition

作　　者：廖志芳[1] 王超群[1] 李小庆[2] 张苗 

机构地区：[1]中南大学软件学院,长沙410075 [2]中南大学信息科学与工程学院,长沙410083 [3]德克萨斯州立大学阿灵顿分校计算机科学与工程系,美国76019

出　　处：《小型微型计算机系统》2013年第11期2472-2476,共5页Journal of Chinese Computer Systems

基　　金：国家科技支撑项目(2012BAH08B01)资助;湖南省2012自然科学基金项目(12JJ3074)资助

摘　　要：标签推荐系统包含用户、资源和标签三个基本元素,由于标签推荐系统中数据存在大量稀疏值和缺失值,而张量分解方法对于稀疏数据处理有较好的性能,因此本文以传统的Tucker和ParaFac张量分解模型为基础,通过数据缺失值处理,进行局部最优求解获得标签推荐值,以此进行社会标签系统中的标签推荐预测.实验表明,基于张量分解的标签推荐算法的召回率和精确度得到提高.同时通过更新Tucker和ParaFac张量分解模型,提出对应的新用户标签推荐算法,实验表明,基于张量分解模型的新用户标签推荐算法可以为新用户提供良好的标签推荐.Tag recommendation system contains three basic elements of users, resources and tags. Because there are a large number of sparse datas and missing values in tag recommendation system, and tensor decomposition method could obtain good performance for sparse data processing, so in this paper, based on Tucker and ParaFac tensor decomposition model, we recommend the prediction tags in social tagging system by dealing with the missing value and computing the local optimal solution to obtain the tag reconunend val- ues. The experiments show that the tag recommendation algorithms based on tensor decomposition have greatly improved the precision and recali rate. Then by updating the original tensor decomposition model the thesis proposed the corresponding new user tag recom-mendation algorithms, experiments show that the proposed new user tag recommendation algorithms based on tensor decomposition models can provide good tag recommendation results for new users.

关 键 词：标签推荐 张量分解 新用户推荐 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]