CNRS - Centre National de la Recherche Scientifique : UMR8623, Inria Saclay - Ile de France, UP11 - Université Paris-Sud - Paris 11, LRI - Laboratoire de Recherche en Informatique
Abstract : This paper describes the ideas and methodologies that we used in the Yahoo learning-to- rank challenge1. Our technique is essentially pointwise with a listwise touch at the last combination step. The main ingredients of our approach are 1) preprocessing (querywise normalization) 2) multi-class AdaBoost.MH 3) regression calibration, and 4) an expo- nentially weighted forecaster for model combination. In post-challenge analysis we found that preprocessing and training AdaBoost with a wide variety of hyperparameters im- proved individual models significantly, the final listwise ensemble step was crucial, whereas calibration helped only in creating diversity.
https://hal.inria.fr/hal-00643001
Contributor : Balázs Kégl <>
Submitted on : Sunday, November 20, 2011 - 10:56:57 PM Last modification on : Wednesday, September 16, 2020 - 5:04:54 PM Long-term archiving on: : Friday, November 16, 2012 - 11:31:44 AM
Róbert Busa-Fekete, Balázs Kégl, Tamas Elteto, György Szarvas. Ranking by calibrated AdaBoost. Yahoo! Learning to Rank Challenge, Jun 2010, Haifa, Israel. ⟨hal-00643001⟩