Improved hierarchical topic detection

Abstract

Due to characteristics of low centralization and the topic drift, how to analyze Internet news reports timely and effectively is more and more concerned. An improved hierarchical topic detection method is proposed with following two improvements. On the one hand, based on the traditional topic detection method of K-means, the new method improves the detection process by using a new parameter of news's contribution for topics to have better adaptability of hierarchical topics. The experimental result also presents that this new method has better detection performance, especially for those hierarchical topics. On the other hand, based on the above-mentioned method, an improved hierarchical clustering algorithm is further put forward. The result demonstrates that different aspects of hierarchical topics could be fully described with low time complexity. © 2013 Springer Science+Business Media.