Using time topic modeling for semantics-based dynamic research interest finding

作     者：Daud, Ali 

作者机构：Int Islamic Univ Sect H10 Dept Comp Sci Islamabad 44000 Pakistan 

出 版 物：《KNOWLEDGE-BASED SYSTEMS》 (Knowl Based Syst)

年 卷 期：2012年第26卷

页      面：154-163页

核心收录：

学科分类：08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：Higher Education Commission (HEC)  Islamabad  Pakistan 

主　　题：Dynamic research interests Exchangeability of topics Time topic modeling Social networks Unsupervised machine learning 

摘      要：Researchers interests finding has been an active area of investigation for different recommendation tasks. Previous approaches for finding researchers interests exploit writing styles and links connectivity by considering time of documents, while semantics-based intrinsic structure of words is ignored. Consequently, a topic model named Author-Topic model is proposed, which exploits semantics-based intrinsic structure of words present between the authors of research papers. It ignores simultaneous modeling of time factor which results in exchangeability of topics problem, which is important factor to deal with when finding dynamic research interests. For example, in many real world applications, like finding reviewers for papers and finding taggers in the social tagging systems one need to consider different time periods. In this paper, we present time topic modeling approach named Temporal-Author-Topic (TAT) which can simultaneously model text, researchers and time of research papers to overcome the exchangeability of topics problem. The mixture distribution over topics is influenced by both co-occurrences of words and timestamps of the research papers. Consequently, topics occurrence and their related researchers change over time, while the meaning of particular topic almost remains unchanged. Proposed approach is used to discover topically related researchers for different time periods. We also show how their interests and relationships change over a time period. Empirical results on large research papers corpus show the effectiveness of our proposed approach and dominance over Author-Topic (AT) model, by handling the exchangeability of topics problem, which enables it to obtain similar meaning of particular topic overtime. (C) 2011 Elsevier B.V. All rights reserved.