DenoiseRank: Learning to Rank by Diffusion Models

ACL ARR 2025 May Submission3614 Authors

Abstract: Learning to rank (LTR) is one of the core tasks in NLP by supervised algorithmic techniques trained on a dataset with queries and their corresponding labeled relevant items. LTR models have made great progress, but all of them implement the algorithms from discriminative perspective. In this paper, we aim at addressing LTR from a novel perspective, i.e., by a deep generative model. Specifically, we propose a novel denoise rank model, DenoiseRank, which is a denoising diffusion-based model, for the LTR task. Our DenoiseRank noises the relevant labels in the diffusion process and denoises them on the query documents in the reverse process to accurately predict their distribution. Our model is the first to address LTR from generative perspective and is a diffusion method for LTR. Extensive experiments were conducted on benchmark datasets and the results demonstrated the effectiveness of the proposed DenoiseRank model. DenoiseRank provides a benchmark for generative LTR model study.

Paper Type: Long

Research Area: Information Retrieval and Text Mining

Research Area Keywords: Information Retrieval, Re-ranking, Generative Models, Model Architectures

Contribution Types: Model analysis & interpretability

Languages Studied: English

Submission Number: 3614