Care episode retrieval: distributional semantic models for information retrieval in the clinical domain

Table 9 Experiment 2: MAP scores for different IR models (rows) when using different models for measuring discharge summary similarity (columns).

Test set	Lucene	RI-Word	RI-Note	RI-ICD	RI-Index	W2V	W2V-ICD	Average	Rank
IR model
Lucene	0.084	0.046	0.041	0.050	0.030	0.062	0.071	0.055	4
RI-Word	0.041	0.049	0.029	0.036	0.016	0.048	0.051	0.039	7
RI-Note	0.048	0.041	0.063	0.061	0.024	0.050	0.074	0.052	5
RI-ICD	0.059	0.036	0.054	0.149	0.033	0.058	0.124	0.073	2
RI-Index	0.063	0.033	0.044	0.048	0.043	0.052	0.065	0.050	6
W2V	0.075	0.051	0.052	0.079	0.035	0.094	0.106	0.070	3
W2V-ICD	0.089	0.053	0.070	0.150	0.046	0.094	0.187	0.098	1

This table also illustrates the general approach to how the average scores are calculated for the results graphs for Experiment 2.

ISSN: 1472-6947