Ranking Support Vector Machine with Kernel Approximation

Table 2

Results of different RankSVM algorithms on the first fold of MQ2007 dataset. We take for the kernel approximation method.

Algorithm	Type	Loss			Mean-NDCG	Time (s)
RankSVM-TRON	linear	L1		—	0.5265	1.9
RankSVM-Struct	linear	L1		—	0.5268	2.2
RankSVM-Primal	linear	L2		—	0.5270	1.2
RankSVM-TRON	RBF	L1			0.5310	47463.5
RankNystöm	RBF	L2			0.5330	10.9
RankRandomFourier	RBF	L2			0.5336	16.1