Enhancing the TDT Tracking Evaluation

Proceedings of the Second International Conference on Language Resources and Evaluation (LREC 2000)

Abstract

Topic Detection and Tracking (TDT) is a DARPA-sponsored initiative concerned with finding groups of stories on the same topic (tdt, 1998). The goal is to build systems that can segment, detect, and track incoming news stories (possibly from multiple continuous feeds) with respect to pre-defined topics. While the detection task detects the first story on a particular topic, the tracking task determines, for each story, which topic it is relevant to. This paper will discuss the algorithm currently used for evaluating systems for the tracking task, present some of its limitation, and propose a new algorithm that enhances the current evaluation.