Skip to main content

View full-text article in PMC

. 2017 May 15;19(5):e162. doi: 10.2196/jmir.6887

Table 2.

Document-level classification results.

Metrics	BOW0^a	BOW1^b	BOW2^c	BOW3^d	Best system^e (95% CI)	Domain experts average
True positives	29	45	31	30	49	36
True negatives	207	200	210	209	205	206
False positives	6	13	3	4	8	7
False negatives	40	24	38	39	20	33
Sensitivity	.420	.652	.449	.435	.710 (.683-.737)	.527
Positive predictive value	.829	.776	.912	.882	.860 (.833-.886)	.848
F₁ measure	.556	.709	.602	.583	.778	.650
Specificity	.972	.939	.986	.981	.962 (.951-.974)	.966
Accuracy	.837	.869	.855	.847	.901 (.883-.918)	.862

^a BOW0: Initial bag-of-words.

^bBOW1: First refined bag-of-words.

^cBOW2: Second (more specific) refined bag-of-words.

^dBOW3: Third (most specific) refined bag-of-words.

^eBOW1 with refined dictionary.