User profiling by combining topic modeling and pointwise mutual information (TM-PMI) - 文章详情页

作者：

Wu, L. (Wu, L..) | Wang, D. (Wang, D..) | Guo, C. (Guo, C..) | Zhang, J. (Zhang, J..) | Chen, C.W. (Chen, C.W..)

收录：

Scopus

摘要：

User　profiling　is　one　of　the　key　issues　in　personalized　recommendation　systems.　A　content　curation　social　network　is　a　content-centric　network;　it　encourages　users　to　repin　items　from　other　users　and　other　websites.　It　further　permits　users　to　arrange　the　pins　according　to　their　interests.　It　is　therefore　possible　to　estimate　user　interest　from　the　pins.　In　this　paper,　we　propose　a　user　profiling　approach　to　combining　topic　model　and　pointwise　mutual　information　(TM-PMI).　We　first　extract　a　pin’s　description,　and　then　apply　latent　Dirichlet　allocation　(LDA,　one　of　the　topic　modeling　schemes).　A　three-layer　hierarchical　Bayesian　model　of　user-topic-word　is　thus　obtained.　Then,　a　personal　model　is　obtained　by　selecting　a　set　of　correlated　words　with　constraints　of　word　probability　and　PMI.　The　experimental　results　confirm　the　efficiency　of　the　proposed　approach.　©　Springer　International　Publishing　Switzerland　2016.

关键词：

Latent dirichlet allocation; Pointwise mutual information; Topic modeling; User profile

作者机构：

[ 1 ] [Wu, L.]School of Electronic Information and Control Engineering, Beijing University of Technology, Beijing, 100124, China
[ 2 ] [Wang, D.]School of Electronic Information and Control Engineering, Beijing University of Technology, Beijing, 100124, China
[ 3 ] [Guo, C.]School of Electronic Information and Control Engineering, Beijing University of Technology, Beijing, 100124, China
[ 4 ] [Zhang, J.]School of Electronic Information and Control Engineering, Beijing University of Technology, Beijing, 100124, China
[ 5 ] [Chen, C.W.]Department of Computer Science and Engineering, State University of New York at Buffalo, 316 Davis Hall, Buffalo, NY 14260-2500, United States

通讯作者信息：

[Wu, L.]School of Electronic Information and Control Engineering, Beijing University of TechnologyChina

电子邮件地址：

lfwu@bjut.edu.cn

查看成果更多字段

成果类型
所属机构

所有年份指定年份从至