C-MS-TDT - Robust Incremental Clustering for Multi-stream Topic Detection and Tracking

The vast majority of current TDT research makes two important assumptions about the nature of news: a) every news story discusses only one topic (in order to place a story to exactly one cluster); b) any hierarchical relationships between events should be ignored (to avoid the problem of events within events). The aim of this project is to devise some clustering technique that will allow us to remove (or significantly simplify) these two assumptions, as they limit considerably the power of TDT models.