[2603.21613] AgenticRec: End-to-End Tool-Integrated Policy Optimization for Ranking-Oriented Recommender Agents

Computer Science > Information Retrieval arXiv:2603.21613 (cs) [Submitted on 23 Mar 2026] Title:AgenticRec: End-to-End Tool-Integrated Policy Optimization for Ranking-Oriented Recommender Agents Authors:Tianyi Li, Zixuan Wang, Guidong Lei, Xiaodong Li, Hui Li View a PDF of the paper titled AgenticRec: End-to-End Tool-Integrated Policy Optimization for Ranking-Oriented Recommender Agents, by Tianyi Li and 4 other authors View PDF HTML (experimental) Abstract:Recommender agents built on Large Language Models offer a promising paradigm for recommendation. However, existing recommender agents typically suffer from a disconnect between intermediate reasoning and final ranking feedback, and are unable to capture fine-grained preferences. To address this, we present AgenticRec, a ranking-oriented agentic recommendation framework that optimizes the entire decision-making trajectory (including intermediate reasoning, tool invocation, and final ranking list generation) under sparse implicit feedback. Our approach makes three key contributions. First, we design a suite of recommendation-specific tools integrated into a ReAct loop to support evidence-grounded reasoning. Second, we propose theoretically unbiased List-Wise Group Relative Policy Optimization (list-wise GRPO) to maximize ranking utility, ensuring accurate credit assignment for complex tool-use trajectories. Third, we introduce Progressive Preference Refinement (PPR) to resolve fine-grained preference ambiguities. By minin...