A review on ranking problems in statistical learning

[annarailton]

https://arxiv.org/abs/1909.02998

[annarailton]

Types of ranking problems

Label ranking

Have:

• a set of instances X
• finite set of labels Y

Preference given in the form:

[TODO insert image]

Training data: set of preference information Model task: find preference ranking for labels for any instance

Connection between classification (incl. multi-label) problems and label ranking problems.

Instance ranking

Have:

• a set of instances X
• finite set of labels Y
• each instance is associated with a label

Labels are defined to have a fixed order.

Training data: instances Model task: find ranking order for new set of instances

Object ranking

Have:

• a set of instances X
• NO labels

Training data: set of pairwise preference information for instances Model task: Find ranking order among instances

Subset ranking

Several searches are executed on disjoint subsets of a larger collection and the results are combined (?)

Rank aggregation

Methods for combining results of e.g. different search engines.