Input format for rank:pairwise objective

Hello, I’m using Xgboost for ranking problem and is struggling with interpreting the prediction result. Usually, when you do pairwise approach, the input should be query + two documents and the output is probability between 0 and 1 (from which we know which one is better). But what i see are scores. So does that mean the input could still be query + one document and xgboost will do pairwise transformation automatically and then we can use its score to directly get a sense of the rank of document in each query? I’m very new to Xgboost. I will be grateful if someone can help answer my question.

1 Like

I have the same question…um…is there anyone can help us?

The scores (which are not probabilities) specifies the relative ordering between documents. See Evaluating XGBoost Ranking