Exploiting Semantic Annotations and -Learning for Constructing an Efficient Hierarchy/Graph Texts Organization

Table 5

The results of the comparison of the accuracy and acceptance rate between the proposed semantic relatedness measure and Islam and Inkpen [7].

MRPC dataset	Relatedness threshold	Human judgment (TP + FN)	Islam and Inkpen [7]		The proposed MFsSR
			Acc	AR	Acc	AR
Training subset (4,076)	0.1	2,753 true	0.67	1	0.68	1
	0.2		0.67	1	0.68	1
	0.3		0.67	1	0.68	1
	0.4		0.67	1	0.68	1
	0.5		0.69	0.98	0.68	1
	0.6		0.72	0.89	0.68	1
	0.7		0.68	0.78	0.70	0.98
	0.8		0.56	0.4	0.72	0.86
	0.9		0.37	0.09	0.60	0.49
	1		0.33	0	0.34	0.02
Test subset (1,725)	0.1	1,147 true	0.66	1	0.67	1
	0.2		0.66	1	0.67	1
	0.3		0.66	1	0.67	1
	0.4		0.66	1	0.67	1
	0.5		0.68	0.98	0.67	1
	0.6		0.72	0.89	0.66	1
	0.7		0.68	0.78	0.66	0.98
	0.8		0.56	0.4	0.64	0.85
	0.9		0.38	0.09	0.49	0.5
	1		0.33	0	0.34	0.03