Using NLP in openEHR archetypes retrieval to promote interoperability: a feasibility study in China

Table 6 Retrieval performance comparison between expansion terms and combination terms with different similarity thresholds

Methods			AP	P@3	P@5
Low level set	Original search terms		0.050	0.050	0.050
	Expansion terms	Top3	0.963	0.950	0.975
		Top5	0.975	0.950	1.000
		Top10	0.963	0.950	0.975
		Mean	0.967	0.950	0.983
	Combination terms	Top3	0.512	0.500	0.525
		Top5	0.575	0.550	0.600
		Top10	0.6125	0.600	0.625
		Mean	0.567	0.550	0.583
Medium level set	Original search terms		0.137	0.125	0.15
	Expansion terms	Top3	0.888	0.850	0.925
		Top5	0.888	0.850	0.925
		Top10	0.875	0.850	0.900
		Mean	0.883	0.850	0.917
	Combination terms	Top3	0.200	0.200	0.200
		Top5	0.200	0.200	0.200
		Top10	0.250	0.250	0.250
		Mean	0.217	0.217	0.217
High level set	Original search terms		0.150	0.150	0.150
	Expansion terms	Top3	0.637	0.525	0.750
		Top5	0.612	0.550	0.675
		Top10	0.575	0.500	0.650
		Mean	0.608	0.525	0.692
	Combination terms	Top3	0.150	0.150	0.150
		Top5	0.175	0.175	0.175
		Top10	0.175	0.175	0.175
		Mean	0.167	0.167	0.167

Different similarity thresholds: in the process of synonym expansion, the first 3, 5, and 10 terms of similar results are used as expansion terms and then composed as combination terms
AP: average precision, P@3: precision at 3, P@5: precision at 5

ISSN: 1472-6947