While considerable clustering research has been based on global clustering, (i.e. clustering stories over an entire data set), TDT clustering is done incrementally which means that for a given source file of stories, you can look ahead only by a specified number of 'days' before making a final decision.

Incremental clustering can be broken down into two phases: detecting when a new event is seen and putting stories that discuss previously seen events into appropriate clusters. Since the first phase is a difficult and interesting challenge, the First Story Detection evaluation task was instantiated to evaluate how well systems can detect the first story of an event.

To learn more about technology used to do TDT topic detection, or to learn how well these systems perform, there were several system description papers submitted for the 1999 TDT workshop. Also, the TDT Research Links contains pointers to related research.