PrLM: Learning Explicit Reasoning for Personalized RAG via Contrastive Reward Optimization

2025-08-10

AI Summary

Existing personalized RAG approaches rely on large language models (LLMs) to implicitly fuse user profiles with queries, making generated outputs highly sensitive to retrieval quality and prone to deviation from user preferences. This paper proposes an explicit user profile reasoning framework that models user preferences as interpretable intermediate representations and introduces a contrastive reward model—trained without human annotations—to optimize generation via reinforcement learning. Our core contributions are: (1) decoupling retrieval from reasoning to enforce explicit modeling of user-specific features; and (2) constructing fine-grained contrastive rewards grounded in multi-source user profiles to enhance preference alignment. Extensive experiments across three benchmark datasets demonstrate significant improvements over state-of-the-art methods, achieving +12.7% BLEU-4 and +9.3% ROUGE-L gains in personalized generation quality. Moreover, our method exhibits strong robustness under varying numbers of retrieved documents and across different retrievers.

Technology Category

Machine Learning: Learning Preferences or RankingsKnowledge Representation and Reasoning: PreferencesData Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

User Modeling, Personalization and Recommendation: Fairness-aware retrieval and rankingSearch and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAGSemantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Personalized retrieval-augmented generation (RAG) aims to produce user-tailored responses by incorporating retrieved user profiles alongside the input query. Existing methods primarily focus on improving retrieval and rely on large language models (LLMs) to implicitly integrate the retrieved context with the query. However, such models are often sensitive to retrieval quality and may generate responses that are misaligned with user preferences. To address this limitation, we propose PrLM, a reinforcement learning framework that trains LLMs to explicitly reason over retrieved user profiles. Guided by a contrastively trained personalization reward model, PrLM effectively learns from user responses without requiring annotated reasoning paths. Experiments on three personalized text generation datasets show that PrLM outperforms existing methods and remains robust across varying numbers of retrieved profiles and different retrievers.

Problem

• Improving personalized RAG by explicit reasoning over user profiles
• Reducing sensitivity to retrieval quality in personalized response generation
• Aligning generated responses with user preferences without annotated data

Innovation

• Reinforcement learning for explicit reasoning
• Contrastive personalization reward model
• Robust across varying retrieval conditions