1. Type
Most current ranking algorithms can be divided into three main categories: pointwise, pairwise and listwise.
Pointwise ranking algorithms
This type of algorithms does not use any relative information between documents, instead attempting to design a score function, scores of which are then used to rank the documents.
Examples: PRank
Pairwise ranking algorithms
The goal of these algorithms is to minimize the relative pairwise misclassification error in ranking problem.
Examples: RankNet, LambdaRank, RankBoost
Listwise ranking algorithms
These algorithms use lists of ranked documents as “instances” during training, and learn a ranking model by minimizing some listwise loss function.
Examples: AdaRank, SoftRank, ListNet, C-CRF
Continue…