From: Concept selection for phenotypes and diseases using learn to rank
Pairwise learn to rank | Listwise learn to rank | |
---|---|---|
Goal | Ranking by learning on object pairs | Ranking by learning on object lists |
Loss function | pairwise loss, e.g., hinge loss, exponential loss, logistic loss | listwise loss, e.g., cross entropy loss, cosine loss |
Advantages | Theoretical aspects are well studied | Considers the relationship among objects to their full extent |
Disadvantages | Considers only pairwise orders; May be biased towards lists with more objects | Theoretical aspects are less well studied |
Algorithms | ListNet [22] |