Exploiting Topic Pragmatics for New Event Detection in TDT-2004
Ronald K. Braun, Ryan Kaneshiro · 2004
Stottler Henke participated for the first time in the New Event Detection (NED) track of TDT-2003 as a means of evaluating various prototyped components developed as part of a new story detection and topic tracking application. We combined a number of "pragmatics-based" classifiers in an ensemble-learning framework to identify the first story of a new topic to link subsequent stories together as they unfold across multiple news streams. We present an overview of our techniques and a preliminary characterization of their performances based on our experimental runs for the TDT-2003 Evaluation.