ReasonRank: Empowering Passage Ranking with Strong Reasoning Ability

2025-08-09

AI Summary

Existing LLM-based re-ranking models underperform in complex ranking scenarios due to insufficient reasoning-intensive training data. Method: We propose ReasonRank—a novel framework integrating automated reasoning data synthesis, self-consistency filtering, and multi-view ranking reward reinforcement learning. Specifically, DeepSeek-R1 is leveraged to generate high-quality reasoning-aware ranking labels; the model is then trained via two stages—supervised fine-tuning followed by reinforcement learning with a multi-view ranking reward that jointly optimizes ranking consistency and logical coherence. Contribution/Results: ReasonRank achieves a new state-of-the-art score of 40.6 on the BRIGHT leaderboard, significantly outperforming prior methods, while maintaining lower inference latency than the pointwise model Rank1. Its core contribution is the first reasoning-enhanced training paradigm specifically designed for paragraph re-ranking, systematically improving both logical reasoning capability and ranking coordination.

Technology Category

• Machine Learning: Learning Preferences or Rankings
• Knowledge Representation and Reasoning: Common-Sense Reasoning
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Large Language Model (LLM) based listwise ranking has shown superior performance in many passage ranking tasks. With the development of Large Reasoning Models, 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. A self-consistency data filtering mechanism is designed to ensure the data quality. To empower the listwise reranker with strong reasoning ability, we further propose a two-stage post-training approach, which includes a cold-start supervised fine-tuning (SFT) stage for reasoning pattern learning and a reinforcement learning (RL) stage for further ranking ability enhancement. During the RL stage, based on the nature of listwise ranking, we design a multi-view ranking reward, which is more effective than a ranking metric-based reward. Extensive experiments demonstrate that our trained reasoning-intensive reranker extbf{ReasonRank} outperforms existing baselines significantly and also achieves much lower latency than pointwise reranker Rank1. extbf{Through further experiments, our ReasonRank has achieved state-of-the-art (SOTA) performance 40.6 on the BRIGHT leaderboardfootnote{https://brightbenchmark.github.io/}.} Our codes are available at https://github.com/8421BCD/ReasonRank.

Problem

• Addresses scarcity of reasoning-intensive training data for rerankers
• Improves reasoning ability in complex passage ranking scenarios
• Reduces latency while enhancing ranking performance

Innovation

• Automated reasoning-intensive training data synthesis
• Two-stage post-training with SFT and RL
• Multi-view ranking reward for RL enhancement