Pairwise Learning to Rank by Neural Networks Revisited: Reconstruction, Theoretical Analysis and Practical Performance

TL;DR

This paper introduces DirectRanker, a neural network-based pairwise learning to rank model that is mathematically sound and outperforms existing methods on benchmark datasets, offering a simpler yet effective approach.

Contribution

The paper revisits pairwise learning to rank with a neural network, providing theoretical properties and demonstrating superior practical performance over state-of-the-art methods.

Findings

Outperforms existing methods on LETOR and MQ datasets

Mathematically shown to be reflexive, antisymmetric, and transitive

Simpler structure with competitive or better accuracy

Abstract

We present a pairwise learning to rank approach based on a neural net, called DirectRanker, that generalizes the RankNet architecture. We show mathematically that our model is reflexive, antisymmetric, and transitive allowing for simplified training and improved performance. Experimental results on the LETOR MSLR-WEB10K, MQ2007 and MQ2008 datasets show that our model outperforms numerous state-of-the-art methods, while being inherently simpler in structure and using a pairwise approach only.