On-Line New Event Detection Using Minimal New Sets

Abstract

On-line New Event Detection can be used to monitor multiple news streams and determine when a new newsworthy event that might warrant further attention occurs, thus helping the user not to miss important information while helping him sift through the vast amount of information available. In this thesis we explore a novel approach for On-line New Event Detection based on computing minimal new sets of small sizes from new stream documents with regard to previously seen documents. We present an algorithm that outputs sets of words likely to be indicative of new events which users easily can browse through and detail several different approaches to select the sets most likely to be informative of new events. In addition we examine the practicality of our approaches and compare them to the most prominent current techniques.