Label ranking is the supervised problem of learning a mapping from a general feature space to the space of full rankings. In this talk, I will briefly explain the key concepts of ranking data, and present two families of (non-parametric) methods we developed for label ranking. The first one [1] adapts well-known partition methods (k-nearest neighbor and tree-based methods) for ranking data. These predictive rules build a partition of the feature space from the data, and compute efficiently (approximate) Kemeny ranking aggregation at a local level. The second one [2] adopts the least square surrogate loss approach from structured prediction, that solves a supervised learning problem in two steps: a regression step in a well-chosen feature space and a pre-image (or decoding) step. We use specific feature maps/embeddings for ranking data, which convert any ranking/permutation into a vector representation.

[1] Ranking Median Regression: Learning to Order through Local Consensus. (ALT 2018) Stephan Clémençon, Anna Korba and Eric Sibony.

[2] A Structured Prediction Approach for Label Ranking. (NIPS 2018) Anna Korba, Alexandre Garcia, Florence D’Alché Buc.