Representation Learning of Knowledge Graphs with Embedding Subspaces

Table 2

Prediction results with different training set sizes on FB15K.

		Mean rank		HITs@10 (%)
Training set size	Method	Raw	Filtered	Raw	Filtered
Train = 5000	TransE	5042	5030	10.4	10.7
	TransH	5125	5113	10.2	10.5
	ProjE	1340	1328	24.3	32.8
	Sub-ProjE	534	522	19.2	19.9
Train = 10000	TransE	3372	3358	16.0	16.6
	TransH	3388	3375	16.2	16.8
	ProjE	717	705	32.9	42.8
	Sub-ProjE	408	395	22.8	23.7
Train = all	TransE	165	98	53.7	70.9
	TransH	166	100	58.0	70.3
	ProjE	157	98	42.5	66.1
	Sub-ProjE	326	266	24.8	34.3