Table 4.

Detailed results of the best CRF-based NER system on admission and discharge summaries for each entity type

Entity	Admission notes		Discharge summaries
	Exact-match	Inexact-match	Exact-match	Inexact-match
Overall	93.52 (93.77/93.26)	94.69 (94.95/94.43)	89.23 (90.29/88.20)	91.00 (92.08/89.94)
Problems	93.96 (93.99/93.92)	95.35 (95.39/95.32)	90.19 (90.61/89.77)	92.20 (92.63/91.77)
Procedures	82.89 (85.44/80.48)	85.34 (87.97/82.86)	78.51 (82.80/74.64)	81.48 (85.93/77.46)
Tests	95.06 (95.22/94.91)	95.41 (95.56/95.26)	91.82 (92.22/91.42)	92.89 (93.30/92.49)
Medications	86.44 (88.18/84.76)	88.98 (90.78/87.26)	87.41 (90.82/84.24)	88.33 (91.78/85.13)

Values are F-measure (recall/precision) (%).

CRF, conditional random fields; NER, named entity recognition.