Multiview Translation Learning for Knowledge Graph Embedding

Table 5

Experimental results of WN18 by mapping properties of relations.

Tasks (k = 35)		Predicting head				Predicting tail
		1-to-1	1-to-N	N-to-1	N-to-N	1-to-1	1-to-N	N-to-1	N-to-N
Global-view	Hits@10 (%)	90.5/90.5	91.9/92.7	56.0/87.6	85.7/89.7	92.9/93.2	54.5/86.9	94.1/94.2	84.6/88.6
(Raw/Filt)	Mean rank	12/12	158/158	228/198	217/216	35/35	258/228	132/132	227/225
Local-view	Hits@10 (%)	71.4/71.4	91.8/92.0	70.3/86.8	94.6/95.0	71.4/71.4	78.4/91.7	84.4/84.4	94.3/94.8
(Raw/Filt)	Mean rank	97/97	23/23	72/61	22/21	177/177	44/33	67/67	23/23
Multiview	Hits@10 (%)	90.5/90.5	94.2/94.3	76.1/94.2	95.0/95.4	92.9/93.9	79.8/93.9	94.1/94.2	94.8/95.2
(Raw/Filt)	Mean rank	8/8	12/11	47/34	39/38	20/19	44/31	14/14	23/22

The bold values stand for the best performance in the corresponding metric.