Table 2.
Method | Experiment | Source domain | Target domain | Number of target domain labeled samples added to fine-tuning | DAPT corpus | TAPT corpus |
---|---|---|---|---|---|---|
In-domain section classification | FTtarget | Thyme | Thyme | All | Unlabeled notes in THYME corpus | Thyme training set |
Cross-domain section classification | FTsource | Discharge or progress | 0 | |||
Cross-domain section classification with continued pretraining | DAPT + FTsource | |||||
DAPT + TAPT + FTsource | ||||||
Cross-domain section classification with continued pretraining and target domain labeled data | FTsource + target | 10, 20, 30, 40, 50 | ||||
DAPT + FTsource + target | ||||||
DAPT + TAPT + FTsource + target | ||||||
Quantifying the value of domain adaptation methods | FTtarget | – | 10, 20, …, 190, 200 | – | – |