Selection Procedure
- Eliminate stories longer than 15 minutes, commercials, and sport reports
from test pool
- Produce a balanced test pool by pruning the remainder of the
test pool with respect to the training data, taking into account that
stories longer than 15 minutes will be added back
- Random select from balanced test pool on section-by-section basis
until +-5% of time limit
- Add back the stories longer than 15 minutes to the test set
(first 15 minutes only)
- Merge fragmented sections