Recognition and Evaluation of Clinical Section Headings in Clinical Documents Using Token-Based Formulation with Conditional Random Fields

Table 5

Performance comparison among different methods.

Dataset	Configuration	(%)	(%)	(%)
Set 2	Dict. method 1 (SecTag)	19.9	79.31	31.82
	Dict. method 1 (set 1)	52.18	94.04	67.12
	Dict. method 1 (SecTag + set 1)	23.19	94.99	33.47
	Dict. method 2 (SecTag)	41.19	79.31	54.22
	Dict. method 2 (set 1)	75.5	94.04	83.76
	Dict. method 2 (SecTag + set 1)	45.33	94.99	61.37
	Sentence-based formulation (ME)	81.54	82.16	81.85
	Token-based formulation (CRF)	95.48	92.66	94.05
Test	Dict. method 1 (SecTag)	21.15	80.23	33.47
	Dict. method 1 (set 1 + set 2)	54.13	94.87	68.93
	Dict. method 1 (SecTag + set 1 + set 2)	24.38	95.48	38.84
	Dict. method 2 (SecTag)	41.72	80.23	54.89
	Dict. method 2 (set 1 + set 2)	76.37	94.84	84.6
	Dict. method 2 (SecTag + set 1 + set 2)	45.59	95.48	61.71
	Sentence-based formulation (ME)	85.46	85.54	85.5
	Token-based formulation (CRF)	96.04	92.4	94.19