Table IV.

Performance of baselines

BaselinesPrecisionRecallF1
TF@5000.4020.5000.446
TF@10000.6000.7460.665
BT0.0990.6270.171
S-NER0.1310.0800.099
TermExtractor0.2020.1070.140
SK-CRF0.9140.8970.905
SSK-CRF0.8890.8250.856

Notes: SK-CRF and SSK-CRF use 30% data of subtitles for training; half of the training data as unlabeled for SSK-CRF; the italic data mean they are the best results among all the baselines

or Create an Account

Close subscription notice
Close access options