Table 2.

Summary of keyword extraction performance for pathology reports.

Methods	Precision			Recall			Exact Matching
	SPE	PRO	PAT	SPE	PRO	PAT	SPE	PRO	PAT
BERT	0.9951	0.9985	0.9961	0.9962	0.9990	0.9938	0.9839	0.9956	0.9795
LSTM	0.9871	0.9932	0.9438	0.9764	0.9919	0.9387	0.9327	0.9868	0.9151
Pre-trained LSTM	0.9940	0.9978	0.9924	0.9915	0.9979	0.9934	0.9646	0.9794	0.9631
CNN	0.9740	0.9769	0.9320	0.9716	0.9758	0.9204	0.9327	0.9502	0.8770
Pre-trained CNN	0.9947	0.9958	0.9855	0.9903	0.9964	0.9823	0.9631	0.9690	0.9218
Bayes Classifier	0.9300	0.9601	0.8956	0.8946	0.9775	0.8227	0.7130	0.9078	0.5168
Kea	0.7321	0.1154	0.3499	0.3751	0.1076	0.1198	0.1010	0.0981	0.0190
WINGNUS	0.6227	0.1786	0.1552	0.3904	0.1650	0.1017	0.1098	0.1552	0.0835

SPE represents specimen type, PRO represents procedure type, and PAT represents pathology type.