On-line new event detection and clustering using the concepts of the cover coefficient-based clustering methodology

Date
2002
Instructor
Source Title
Print ISSN
Electronic ISSN
Publisher
Bilkent University
Volume
Issue
Pages
Language
English
Type
Thesis
Journal Title
Journal ISSN
Volume Title
Abstract

In this study, we use the concepts of the cover coefficient-based clustering methodology (C3 M) for on-line new event detection and event clustering. The main idea of the study is to use the seed selection process of the C3 M algorithm for the purpose of detecting new events. Since C3 M works in a retrospective manner, we modify the algorithm to work in an on-line environment. Furthermore, in order to prevent producing oversized event clusters, and to give equal chance to all documents to be the seed of a new event, we employ the window size concept. Since we desire to control the number of seed documents, we introduce a threshold concept to the event clustering algorithm. We also use the threshold concept, with a little modification, in the on-line event detection. In the experiments we use TDT1 corpus, which is also used in the original topic detection and tracking study. In event clustering and event detection, we use both binary and weighted versions of TDT1 corpus. With the binary implementation, we obtain better results. When we compare our on-line event detection results to the results of UMASS approach, we obtain better performance in terms of false alarm rates.

Course
Other identifiers
Book Title
Keywords
Clustering, on-line event clustering, on-line event detection
Citation
Published Version (Please cite this version)