How to Kill Two Birds with One Stone: Learning to Rank with Multiple Objectives

In many practical applications, search relevance can be measured in multiple ways - for example, based on implicit user feedback or on manual assessment by annotators. This gives rise to multiple search quality metrics and it may be desirable to optimise all of them at once. Likewise, in the E-Commerce setting, both relevance and financial metrics need to be optimised together to meet business goals. Such situations naturally give rise to the problem of learning to rank with multiple objectives. In this talk, we discuss and compare several approaches to solving this problem paying special attention to the LambdaMART framework. We also present the results of experementation with those approaches in the context of catalog ranking at Zalando.

Alexey Kurennoy

Research Engineer - Zalando

Alexey Kurennoy is a Research Engineer working on ranking problems at Zalando. He has a Ph.D. in Numerical Optimisation from Moscow State University. His current interests lie in the field of Learning To Rank (LTR) including such topics as unbiased and multiobjective LTR