What is the best way to convert ordering to pairwise learning?

Suppose I have ordered objects for some query. Q1: o1,o2,o3,o4,o5

And I have learning to rank classifier which can be trained on data best from the list i,e, o1 is better then o2,o3,o4,o5. o2 is better then o3,o4,o5.

What is the best way to convert 1st problem to 2nd?

The are several different,

1. o1 vs rest
2. consequent pairs (o1,o2), (o2,o3), (o4, o5)
3. all pairs (o1,o2), (o1,o3), (o1, o5), (o2, o3) .... a lot of training data
4. one vs all below, (o1, o2-o5), (o2, o3-o5) ...

Which one is the best ?