I was studying RankSVM[1] and I was observing the ranking pairs listed in the example here. Formulation can be seen on page 4 of this paper.

One of the ordering is: $3B>3A, 3B>3D, 3A>3D$

My question is, if you have $3B>3A$ and $3A>3D$, the other relation: $3B>3D$ is implied. Do we have to put such relations in the training set? Will it affect accuracy if we do not put such examples in the training set?

There are other such orderings too:

$1A>1B, 1B>1C$ which implies $1A>1C$, which is still listed.

$1A>1B, 1B>1D$ which implies $1A>1D$, which is still listed.

$3C>3A, 3A>3D$ which implies $3C>3D$, which is still listed.

So this seems to be done purposely. However, are such examples redundant in the RankSVM formulation?

[1] Joachims, Thorsten. "Optimizing search engines using clickthrough data." Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 2002.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.