REARANK: Reasoning Re-ranking Agent via Reinforcement Learning

TL;DR

REARANK is a reinforcement learning-based reranking agent built on LLMs that improves reasoning and performance in information retrieval tasks with minimal annotated data.

Contribution

It introduces a novel reasoning reranking approach using reinforcement learning and data augmentation, achieving high performance with limited annotated samples.

Findings

REARANK achieves performance comparable to GPT-4 on in-domain benchmarks.

REARANK surpasses GPT-4 on reasoning-intensive BRIGHT benchmarks.

Requires only 179 annotated samples for training.

Abstract

We present REARANK, a large language model (LLM)-based listwise reasoning reranking agent. REARANK explicitly reasons before reranking, significantly improving both performance and interpretability. Leveraging reinforcement learning and data augmentation, REARANK achieves substantial improvements over baseline models across popular information retrieval benchmarks, notably requiring only 179 annotated samples. Built on top of Qwen2.5-7B, our REARANK-7B demonstrates performance comparable to GPT-4 on both in-domain and out-of-domain benchmarks and even surpasses GPT-4 on reasoning-intensive BRIGHT benchmarks. These results underscore the effectiveness of our approach and highlight how reinforcement learning can enhance LLM reasoning capabilities in reranking.