News Clustering Track
RIRES: Russian Information Retrieval Evaluation Seminar

 News 
 About 
 Manifesto 
 Call for participation 
 General principles 
 Participation 
 Tracks 
 Participants 
 Test collections 
 Publications 
 Relevance tables 
 History 
 2004 
 2005 
 Forum 

По-русскиПо-русски
 

News Clustering Track

Overview

Splitting news feeds into event-based topics.

Definitions:

  • Event - something taking place at a certain time in a certain place along with all its causes and all its inevitable consequences.
  • Event-based topic - representation of the event in the news media (i.e. a set of news reports about the event). Topics can possibly be associatively related, see Further information for examples below.

Test Collection

The test collection used in this track is the ROMIP news collection that contains news feeds from several news agencies.

Task Description for Participating Systems

Participants have to structure/group an input stream of news reports from the collection into a set of event-based topics.

Evaluation Methodology

  • instructions for assessors: TBD
  • evaluation method: TBD
  • relevance scale: TBD
  • official metrics: TBD

Data Formats

Further information

Different event-based topics can still be associatively related (it can be a cause-effect or a space-time relation).
Examples:
  • "Mourning the death of Pope John Paul II" and "Altering the date of the next round of the Italian soccer championship because of the Pope's death" - these are two different topics but there is an associative relationship between them.
  • "NYC theatre co-founder Curt Dempster has died" and "Fulton Theatre presents 'The Mousetrap'" - there are neither cause-effect nor space-time relationship, i.e. no associative relationship.