DenoiseRank: Generative Diffusion Model for Learning to Rank
Researchers have introduced DenoiseRank, a novel approach to learning to rank (LTR) that reframes the task from a generative perspective using diffusion models. Traditional LTR methods are predominantly discriminative, but DenoiseRank applies a diffusion process to add noise to relevant labels and then denoises them on query documents in reverse, accurately predicting their distribution. This is the first diffusion-based generative model for LTR. Experiments on benchmark datasets demonstrate its effectiveness, establishing a new benchmark for generative LTR tasks. The paper is available on arXiv.
Key facts
- DenoiseRank is a diffusion model for learning to rank.
- It approaches LTR from a generative perspective, unlike traditional discriminative models.
- The model noises relevant labels in the diffusion process and denoises them on query documents.
- It is the first diffusion method for LTR.
- Experiments on benchmark datasets show effectiveness.
- The paper is published on arXiv under computer science information retrieval.
- The model provides a benchmark for generative LTR tasks.
- The approach uses deep generative modeling for ranking.
Entities
Institutions
- arXiv