ReasonRank: Empowering Passage Ranking with Strong Reasoning Ability

TL;DR

ReasonRank is a novel passage reranker that leverages automated reasoning data synthesis and a two-stage training process to significantly improve reasoning ability and ranking performance in complex scenarios.

Contribution

The paper introduces ReasonRank, a reasoning-intensive reranker trained with synthetic data and a two-stage approach, enhancing reasoning and ranking in complex passage retrieval tasks.

Findings

ReasonRank outperforms existing baselines significantly.

It achieves lower latency than pointwise rerankers.

The approach improves reasoning ability in passage ranking.

Abstract

Large Language Model (LLM) based listwise ranking has shown superior performance in many passage ranking tasks. With the development of Large Reasoning Models (LRMs), many studies have demonstrated that step-by-step reasoning during test-time helps improve listwise ranking performance. However, due to the scarcity of reasoning-intensive training data, existing rerankers perform poorly in many complex ranking scenarios, and the ranking ability of reasoning-intensive rerankers remains largely underdeveloped. In this paper, we first propose an automated reasoning-intensive training data synthesis framework, which sources training queries and passages from diverse domains and applies DeepSeek-R1 to generate high-quality training labels. To empower the listwise reranker with strong reasoning ability, we further propose a two-stage training approach, which includes a cold-start supervised fine-tuning (SFT) stage and a reinforcement learning (RL) stage. During the RL stage, we design a novel multi-view ranking reward tailored to the multi-turn nature of listwise ranking. Extensive experiments demonstrate that our trained reasoning-intensive reranker \textbf{ReasonRank} outperforms existing baselines significantly and also achieves much lower latency than the pointwise reranker. Our codes are available at https://github.com/8421BCD/ReasonRank.