bag-of-words

This paper compares expert-assigned and automatically-assigned MeSH terms as features for systematic review screening classifiers, showing that evaluation design significantly affects measured performance gaps, with canonical designs showing larger gaps that attenuate under alternative designs.