Corpus-based Subtopic Segmentation Using Concept Segment Method

Abstract

Many studies on topic segmentation have been launched in the past, but these previous methods performed poorly on domain-specific short texts. This paper will propose a method which employs high-level concepts on concept hierarchy to predict the number and location of subtopics in a short text, and then determine the boundary between subtopics using similarity measures. Experimental results show that the proposed method can greatly increase the accuracy of subtopic segmentation for domain-specific short texts.