Statistical Models for Tracking and Detection

We explore the behavior of two different statistical models, one based on the beta-binomial distribution and another based on simple unigrams. We show that tracking systems based on these models have complementary strengths and weaknesses: the Beta-Binomial system yields high precision at high decision threshold, but performance quickly degrades as the threshold drops; the Unigram system is not as strong at high decision threshold, but is very good at suppressing false-alarms at lower threshold. We will describe the features of these systems that give rise to this behavior, and discuss ways that each system might be improved by borrowing from the other. We will also discuss our detection system, and how improvements in tracking can be carried over to improvements in detection. 1. OVERVIEW AND APPROACH The focus of our TDT research [1, 2, 3, 4, 5, 6, 7, 8] is the problem of modeling topical content, and our approach is to design, for a given topic, a statistical model of story generation within that topic. Such a model can be used to evaluate the probability that a test story is on the given topic.