Semantic similarity in the biomedical domain: an evaluation across knowledge sources

BMC Bioinformatics

Table 2 Comparison of correlations across measures and reference standards

Benchmark	Concept graph	Knowledge based							Distributional
		Path		Intrinsic IC			PPR
		Wu & Palmer	Path LCH	Lin	Path	LCH	Taxonomy	All	Lin	Vector
Pedersen Coders N=29	Pedersen 2006 [13]		0.51						0.75	0.75
	sct	0.66	0.66	0.61	0.60	0.61	0.61	0.66
	sct-umls	0.56	0.54	0.49	0.45	0.45	0.70	⁺0.26
	sct-msh	0.64	0.76	0.59	0.58	0.61	0.75	0.46
	umls	0.74	0.65	0.70	0.69	0.69	0.76	0.73
Pedersen Physicians N=29	Pedersen 2006		0.36						0.60	0.84
	sct	0.54	0.50	0.52	0.49	0.49	0.49	0.62
	sct-umls	0.44	0.38	0.41	⁺0.35	⁺0.35	0.56	⁺0.19
	sct-msh	0.57	0.62	0.53	0.52	0.53	0.60	0.43
	umls	0.66	0.60	0.72	0.69	0.69	0.67	0.63
Pedersen Combined N=29	Pedersen 2006		0.48						0.69	0.76
	sct	0.59	0.56	0.56	0.53	0.54	0.55	0.67
	sct-umls	0.49	0.44	0.45	0.38	0.38	0.63	⁺0.20
	sct-msh	0.62	0.69	0.57	0.56	0.57	0.66	0.45
	umls	0.70	0.61	0.72	0.70	0.70	0.69	0.68
Mayo N=101	Pakhomov 2011 [41]	0.30	0.29
	sct-umls	⁺0.05	⁺0.03	⁺0.09	⁺0.12	*0.30	⁺0.17	⁺0.00
	sct-msh	0.28	0.22	0.32	0.33	0.35	0.44	⁺0.13
	umls	0.38	0.30	0.39	0.41	0.44	0.46	0.21
UMN similarity N=566	Pakhomov 2010 [19]		0.14							0.02
	sct-umls	0.21	0.23	0.22	0.23	**0.36	0.23	⁺0.00
	sct-msh	0.30	0.30	0.32	0.32	**0.37	0.33	0.07
	umls	0.39	0.40	0.43	0.43	0.46	0.41	0.25
UMN relatedness N=587	Pakhomov 2010		0.10							−0.13
	sct-umls	0.14	0.17	0.16	0.16	**0.30	0.17	−0.01
	sct-msh	0.21	0.20	0.22	0.23	**0.31	0.23	⁺0.04
	umls	0.32	0.34	0.35	0.35	0.39	0.33	0.18
UMN relatedness subset N=430	Liu 2010 [32]									0.46
	sct-umls	0.13	0.17	0.16	0.16	**0.30	0.17	⁺0.03
	sct-msh	0.20	0.20	0.22	0.23	*0.32	0.23	⁺0.05
	umls	0.33	0.36	0.36	0.36	0.40	0.35	0.22

+Correlation not significant at 0.05 level. Significance of difference between Intrinsic LCH and Path Finding LCH **<0.05, *< 0.20. Abbreviations: LCH - Leacock & Chodorow, PPR – Personalized PageRank. Refer to Table 1 for concept graph descriptions.

ISSN: 1471-2105