Exploring semantic deep learning for building reliable and reusable one health knowledge from PubMed systematic reviews and veterinary clinical notes

Table 4 Performance of the short form detector with VetCN and PMSB datasets

Data set	Unique candidate terms (n-grams)	SF-U + SF-NU + SF	SF-I	SF-NF	n-grams with no clinically meaningful short forms	Detect n-grams with one or more clinically meaningful short forms			Detect n-grams with no clinically meaningful short forms
Data set	Unique candidate terms (n-grams)	SF-U + SF-NU + SF	SF-I	SF-NF	n-grams with no clinically meaningful short forms	P	R	F	P	R	F
VetCN	300	57	1	14	228	98.28	80.28	88.37	94.21	99.56	96.82
PMSB	333	75	2	0	256	97.40	100	98.68	100	99.22	99.61
						97.78	90.41	93.95	97.07	99.36	98.20

To assess the capability of the short form detector to identify candidate terms (n-grams) with one or more clinically meaningful short forms: the value of the column “SF-U + SF-NU + SF” is interpreted as TP; the value of the column “SF-I” is interpreted as FP; and the value of the column “SF-NF” is interpreted as FN. To assess the capability of the short form detector to identify candidate terms (n-grams) with no clinically meaningful short forms: the value of the column “n-grams with no clinically meaningful short forms” is interpreted as TP; the value of the column “SF-I” is interpreted as FN; and the value of the column “SF-NF” is interpreted as FP. The last row shows the micro-averaging values taking into account the total of 613 unique candidate terms (n-grams) for the 880 term pairs. Abbreviations: P = precision; R = recall; and F = F measure

ISSN: 2041-1480