Research on topic link detection method based on semantic domain

Abstract

Topic Link Detection aims to detect whether a pair of random stories discuss the same topic, which is an important subtask of Topic Detection and Tracking. In previous works, statistical method and machine-learning approach are used more often than not, however, the semantic distribution of a story and the structure relationship of contents are ignored. A new method based on the semantic domain is proposed for the purpose of improved the precision. In this method, every story is divided some semantic domain through analyzing internal semantic distribution and structure relationships of contexts. The results of experiment proved that the proposed method can improve performance of system. © 2014 Springer International Publishing.