Applying Dynamic Co-occurrence in Story Link Detection

Abstract

Story link detection is part of a broader initiative called Topic Detection and Tracking, which is defined to be the task of determining whether two stories, such as news articles or radio broadcasts, are about the same event, or linked. In order to mine more information from the contents of the stories being compared and achieve a more high-powered system, motivated by the idea of the word co-occurrence analysis, we propose our dynamic co-occurrence, which is defined to be a pair of words that satisfy certain relation restriction. In this paper, relation restriction refers to a set of features. This paper evaluates three features: capital, location and distance. We use dynamic co-occurrence in the similarity computation when we apply it in the story link detection system. Experimental results show that the story link detection systems based on the dynamic co-occurrence perform very well, which testify the great capabilities of the dynamic co-occurrence. At the same time, we also find that relation restriction is critical to the performance of dynamic co-occurrence.