Understanding Topic Influence Based on Module Network

Jinlong Wang¹, Congfu Xu², Dou Shen³, Guojing Luo², and Xueyu Geng⁴

¹School of Computer Engineering, Qingdao Technological University, Qingdao, 266033, China
WangJinlong@gmail.com

²Institute of Artificial Intelligence, Zhejiang University, Hangzhou, 310027, China
xucongfu@cs.zju.edu.cn

³Microsoft adCenter Labs, Redmond WA 98052
doushen@microsoft.com

⁴Institute of Geotechnical Engineering Research, Zhejiang University, Hangzhou, 310027, China

Abstract. Topic detection and analysis is very important to understand academic document collections. By further modeling the influence among the topics, we can understand the evolution of research topics better. This problem has attracted much attention recently. Different from the existing works, this paper proposes a solution which discovers hidden topics as well as the relative change of their intensity as a first step and then uses them to construct a module network. Through this way, we can produce a generalization module among different topics. In order to eliminate the instability of topic intensity for analyzing topic changes, we adopt the piece-wise linear representation so that we can model the topic influence accurately. Some experiments on real data sets validate the effectiveness of our proposed method.

LNCS 4822, p. 391 ff.

Full article in PDF | BibTeX

lncs@springer.com
© Springer-Verlag Berlin Heidelberg 2007