R3-REC: Reasoning-Driven Recommendation via Retrieval-Augmented LLMs over Multi-Granular Interest Signals

2026-03-13

AI Summary

This work addresses the challenges of sparse evidence—such as cold-start scenarios, noisy interactions, and variable text lengths—and the opacity in modeling multi-dimensional long- and short-term user intents in sequential recommendation. To this end, the paper proposes R3-REC, a novel framework that, for the first time, integrates retrieval-augmented large language models (RAG-LLMs) into sequential recommendation. Through carefully designed prompt engineering, R3-REC unifies multi-level user intent reasoning, item semantic extraction, interest polarity modeling, and collaborative filtering enhancement within a single architecture. The approach enables explicit and interpretable inference over multi-granularity interest signals. Extensive experiments on ML-1M, Games, and Bundle datasets demonstrate that R3-REC significantly outperforms state-of-the-art neural and LLM-based baselines, achieving up to a 10.2% improvement in HR@1 and a 6.4% gain in HR@5, while maintaining controllable end-to-end latency.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Machine Learning: Large Multimodal Models (LMMs)
• Reasoning under Uncertainty: Sequential Decision Making

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

This paper addresses two persistent challenges in sequential recommendation: (i) evidence insufficiency-cold-start sparsity together with noisy, length-varying item texts; and (ii) opaque modeling of dynamic, multi-faceted intents across long/short horizons. We propose R3-REC (Reasoning-Retrieval-Recommendation), a prompt-centric, retrieval-augmented framework that unifies Multi-level User Intent Reasoning, Item Semantic Extraction, Long-Short Interest Polarity Mining, Similar User Collaborative Enhancement, and Reasoning-based Interest Matching and Scoring. Across ML-1M, Games, and Bundle, R3-REC consistently surpasses strong neural and LLM baselines, yielding improvements up to +10.2% (HR@1) and +6.4% (HR@5) with manageable end-to-end latency. Ablations corroborate complementary gains of all modules.

Problem

• sequential recommendation
• evidence insufficiency
• cold-start sparsity
• noisy item texts
• multi-faceted user intents

Innovation

• Retrieval-Augmented LLMs
• Multi-Granular Interest Signals
• Reasoning-Driven Recommendation
• Long-Short Interest Polarity
• Prompt-Centric Framework