Multileave Gradient Descent for Fast Online Learning to Rank

Anne Schuth and Harrie Oosterhuis and Shimon Whiteson and Maarten de Rijke. In Proceedings of WSDM'16, 2016.

Abstract

Modern search systems are based on dozens or even hundreds of ranking features. The dueling bandit gradient descent ( DBGD) algorithm has been shown to effectively learn combinations of these features solely from user interactions. DBGD explores the search space by comparing a possibly improved ranker to the current production ranker. To this end, it uses interleaved comparison methods, which can infer with high sensitivity a preference between two rankings based only on interaction data. A limiting factor is that it can compare only to a single exploratory ranker.

Links

Multileave Gradient Descent for Fast Online Learning to Rank
https://doi.org/10.1145/2835776.2835804

Bib

@inproceedings{schuth2016multileave,
  title = {Multileave Gradient Descent for Fast Online Learning to Rank},
  author = {Anne Schuth and Harrie Oosterhuis and Shimon Whiteson and Maarten de Rijke},
  year = {2016},
  booktitle = {Proceedings of WSDM'16},
  doi = {10.1145/2835776.2835804}
}