Mining clinical relationships from patient narratives

BMC Bioinformatics

Table 4 Performance by feature set, non-syntactic features.

Relation	Metric	tok6+ atype	+dir	+dist	+str	+pos	+inter	+event	allgen	notok
has_finding	P	44	49	58	63	62	64	65	63	63
	R	39	63	78	80	80	81	81	82	82
	F1	39	54	66	70	69	71	72	71	71
has_indication	P	37	23	38	42	40	41	42	37	44
	R	14	14	46	44	44	47	47	45	47
	F1	18	16	39	39	38	41	42	38	41
has_location	P	36	36	50	68	71	72	72	73	73
	R	28	28	74	79	79	81	81	83	83
	F1	30	30	58	72	74	76	75	77	76
has_target	P	9	9	32	63	57	60	62	60	59
	R	11	11	51	68	67	67	66	68	68
	F1	9	9	38	64	60	63	63	63	62
laterality_modifies	P	21	38	73	84	83	84	84	86	86
	R	9	55	82	89	86	88	88	87	89
	F1	12	44	76	85	83	84	84	84	85
negation_modifies	P	19	54	85	81	80	79	79	77	81
	R	12	82	97	98	93	92	93	93	93
	F1	13	63	89	88	85	84	85	83	85
sub_location_modifies	P	2	2	55	88	86	86	88	88	87
	R	1	1	62	94	92	95	95	95	95
	F1	1	1	56	90	86	89	91	91	90
Overall	P	33	38	50	63	62	64	65	64	64
	R	22	36	70	74	73	75	75	76	76
	F1	26	37	58	68	67	69	69	69	70

Variation in performance by feature set, non-syntactic features. Features sets are abbreviated as in Table 3. For the first seven columns, features were added cumulatively to each other. The next two columns, allgen and notok, are as described in Table 3.

ISSN: 1471-2105