Filtering Track
- collection: new Reuters corpus
- documents
- 810,000 news stories from August, 1996 - August, 1997
- each tagged with Reuters category codes
- topics
- 84 categories that occurred in 2-5% of training docs
- topic was code itself (hierarchy semantics) plus several word name
e.g.: C16 INSOLVENCY/LIQUIDITY
C1511 ANNUAL RESULTS
- judgments
- no new judgments made at NIST
- doc is relevant to a topic if it was assigned that code