De-identifying free text of Japanese electronic health records

Table 3 Detailed results for each privacy type in MedNLP-1 (De-identification task)

	<a > age			<x > sex			<t > time			<h > hospital name
	P	R	F	P	R	F	P	R	F	P	R	F
C3	90.32	87.5	88.89	100	100	100	87.16	91.49	89.27	97.30	94.74	96.00
B3	90.00	84.38	87.10	100	50.00	66.67	91.30	89.36	90.32	97.06	86.84	91.67
B1	93.33	87.5	90.32	100	100	100	90.65	89.36	90.00	89.47	89.47	89.47
B2	90.00	84.38	87.10	100	100	100	91.24	88.65	89.93	91.89	89.47	90.67
C1	96.67	90.62	93.55	100	50.00	66.67	91.18	87.94	89.53	93.55	76.32	84.06
A1	92.86	81.25	86.67	100	50.00	66.67	91.04	86.52	88.73	91.89	89.47	90.67
C2	96.67	90.62	93.55	100	50.00	66.67	89.13	87.23	88.17	96.77	78.95	86.96
A2	92.86	81.25	86.67	100	50.00	66.67	89.05	86.52	87.77	91.89	89.47	90.67
D1	92.31	75.00	82.76	100	50.00	66.67	82.84	78.72	80.73	96.15	65.79	78.12
G1	80.65	78.12	79.37	100	50.00	66.67	84.56	81.56	83.03	72.73	63.16	67.61
D3	88.89	75.00	81.36	100	50.00	66.67	83.08	76.60	79.70	96.15	65.79	78.12
D2	92.31	75.00	82.76	100	50.00	66.67	75.86	78.01	76.92	96.15	65.79	78.12
H2	83.87	81.25	82.54	100	100	100	73.79	75.89	74.83	77.78	73.68	75.68
H1	80.65	78.12	79.37	100	100	100	75.86	78.01	76.92	70.27	68.42	69.33
H3	83.87	81.25	82.54	100	100	100	73.79	75.89	74.83	70.27	68.42	69.33

P, R and F were calculated at the phrase level: P, precision; R, recall; F, F1-measure; and A, accuracy. A was calculated in the word level (the agreement ratio of B-*, I-* and O).
The first column stands for participants’ team names, where the first letter stands for a team ID and the second numerical value stands for a submission run ID

ISSN: 2041-1480