Reviews: Ranking Data with Continuous Labels through Oriented Recursive Partitions

The ms tries to solve a problem called continuous ranking, which is an extension of bipartite ranking. The idea of continuous ranking is to find a score function that increase or decrease with output y with highest probability. In bipartite ranking, each data point x is given a binary label y \in { 1, -1}. The goal is to find a score function s(x) such that the difference of p(s(x) t y 1) and p(s(x) t y -1) is maximal for all threshold t. This can be achieved by maximizing AUC of the score function. The ms extends the bipartite ranking idea for continuous output Y, by introducing a threshold y to separate the data into two ordered parts.