A cross-benchmark comparison of 87 learning to rank methods

作者：

Highlights：

•

• We propose a novel way to compare learning to rank methods.

• We perform a meta-analysis on a large set of papers that report ranking accuracy on a benchmark dataset.

• LRUF, FSMRank, FenchelRank, SmoothRank and ListNet are the most accurate, with increasing certainty.

摘要

•We propose a novel way to compare learning to rank methods.•We perform a meta-analysis on a large set of papers that report ranking accuracy on a benchmark dataset.•LRUF, FSMRank, FenchelRank, SmoothRank and ListNet are the most accurate, with increasing certainty.

论文关键词：Learning to rank,Information retrieval,Evaluation metric

论文评审过程：Received 1 October 2014, Revised 14 May 2015, Accepted 7 July 2015, Available online 22 August 2015, Version of Record 22 August 2015.

论文官网地址：https://doi.org/10.1016/j.ipm.2015.07.002