Automatic topic discovery from hyperlinked text

Abstract

Topic discovery is an important means for marketing, e-Business, social science study and many other applications for various purposes, such as identifying a group with certain properties, observing the raise and diminishment of a certain group. The explosively growing of Internet makes automatic topic discovery a must for the task. In this paper, first we propose the TGM method to rank the eigenvectors of the Web hyperlinked matrix and their associated topics. Then we propose the ATD method, which combines a clustering algorithm with a conventional principal eigenvector computation method, to identify the topics relevant to a given query without manual examination. Our extensive experiments show the ATD method performs very well, and beats TGM in terms of computation time and topic discovery quality.