Dragon's Tracking and Detection Systems for the TDT2000 Evaluation

We describe the tracking and detection systems submitted by Dragon for the TDT2000 evaluation. Our research focus is on improvi ng the distance measure between story and story collection, a comp utation which is central to many of the TDT tasks. In our tracking engi ne, we improved the measure by strengthening our targeting proc edure, and introduced unsupervised adaptation on high-scoring te st s ories. Our detection engine uses a new measure, developed under tra cking experiments, that performs 15–20% better than the previous measure, and makes use of targeting in a computationally tracta ble way. In both tasks, we explore the effect of automatically segmen ting broadcast news stories and of introducing word stemming.