GR2 Technical Report

2026-06-30

AI Summary

Industrial recommendation systems have yet to effectively leverage large language models (LLMs) in the reranking stage, hindered by non-semantic item IDs, underutilized reasoning capabilities, and poorly designed reward mechanisms. This work proposes GR2, a novel framework that, for the first time, integrates reinforcement learning with a verifiable reward mechanism into LLM-based reranking. GR2 innovatively combines semantic ID mapping, reasoning trajectory distillation from a strong teacher model, context compression, and on-policy distillation to address training instability and deployment latency. Evaluated on real-world industrial traffic, GR2 significantly outperforms baseline methods, achieving relative improvements of 18.7% in R@1, 7.1% in R@3, and 9.6% in N@3, demonstrating its effectiveness and scalability.

Abstract

Industrial recommendation systems serve billions of users through a multi-stage funnel -- retrieval, early-stage ranking, and re-ranking -- where the final re-ranking step disproportionately shapes user engagement and downstream performance, particularly for carousel and grid display formats. Despite growing enthusiasm for Large Language Models (LLMs) in recommendation, three gaps hinder industrial adoption: (1) most efforts target retrieval and ranking, leaving re-ranking -- the stage closest to the final user experience -- largely underexplored; (2) LLMs are typically deployed zero-shot or via supervised fine-tuning, underutilizing the reasoning capabilities unlocked by reinforcement learning (RL) on verifiable rewards; (3) deployed catalogs index billions of items with non-semantic identifiers that lie outside any base-LLM vocabulary. We present GR2 (Generative Reasoning Re-Ranker), an end-to-end framework that combines (i) mid-training on semantic IDs produced by a tokenizer with >=99% uniqueness, (ii) reasoning-trace distilled from a stronger teacher via targeted prompting and rejection sampling, and (iii) RL with verifiable rewards purpose-built for re-ranking. To make GR2 resource-viable, we further (iv) introduce a context compressor that amortizes training cost, On-Policy Distillation (OPD) as a scalable alternative to SFT -- which we find collapses at industrial scale -- and reasoning distillation for low-latency serving. GR2 delivers +18.7% R@1, +7.1% R@3, and +9.6% N@3 over legacy baselines on industrial-scale traffic. We further find that reward design is critical in re-ranking: LLMs often hack rewards by preserving the incoming order or exploiting position bias, motivating conditional verifiable rewards as essential industrial components.

Problem

• re-ranking
• Large Language Models
• industrial recommendation systems
• non-semantic identifiers
• reinforcement learning

Innovation

• Generative Reasoning Re-Ranker
• Reinforcement Learning with Verifiable Rewards
• Semantic ID Tokenization
• On-Policy Distillation
• Reasoning Trace Distillation