AgenticRec: A Recommendation-Oriented Agentic Framework with Progressive Tool-Integrated Reasoning Optimization

Tianyi Li; Zixuan Wang; Guidong Lei; Xiaodong Li; Hui Li

cs.IRMar 23, 2026

AgenticRec: A Recommendation-Oriented Agentic Framework with Progressive Tool-Integrated Reasoning Optimization

Authors: Tianyi Li, Zixuan Wang, Guidong Lei, Xiaodong Li, Hui Li

Organizations: Xiamen University, Xiamen, China

Abstract

Recommender agents built on Large Language Models offer a promising paradigm for personalized recommendation. However, existing agents typically suffer from a misalignment between their tool-integrated reasoning trajectories and recommendation feedback, limiting their ability to distinguish fine-grained user preferences. To address these challenges, we propose AgenticRec, an agentic recommendation framework that formulates recommendation as a tool-integrated reasoning process over a recommendation-oriented tool suite. Built upon this framework, we further develop a dedicated two-stage training paradigm tailored for recommender agents. In the first stage, we introduce Recommendation-Oriented Trajectory Activation, optimize the agentic recommendation ability under implicit feedback. In the second stage, Progressive Preference Refinement further refines the agent through bidirectional preference reasoning over self-bootstrapped hard pairs, progressively sharpening preference boundaries. Theoretical analysis and extensive experiments demonstrate the effectiveness of AgenticRec. Our code is available at https://anonymous.4open.science/r/AgenticRec-FB16.

Figures & tables

Figure 1: Compared to existing training-free recommender agents like RecMind Wang et al. (2024b) and InteRecAgent Huang et al. (2025c) (left), AgenticRec (right) learns recommendation-driven tool use and reasoning from real-world interaction logs.

Conventional Sequential Recommenders
Model	CDs					Instruments					Office					Games
Model	H@1	H@5	N@5	H@10	N@10	H@1	H@5	N@5	H@10	N@10	H@1	H@5	N@5	H@10	N@10	H@1	H@5	N@5	H@10	N@10
Caser	0.0488	0.2526	0.1487	0.5089	0.2307	0.1825	0.4644	0.3275	0.6384	0.3828	0.1771	0.4374	0.3139	0.5541	0.3511	0.1535	0.3953	0.2813	0.5220	0.3217
GRU4Rec	0.0512	0.2348	0.1403	0.4851	0.2202	0.1960	0.4534	0.3291	0.6446	0.3907	0.1642	0.4324	0.3054	0.5705	0.3490	0.1708	0.3781	0.2817	0.5239	0.3288
SASRec	0.0536	0.2348	0.1413	0.4660	0.2145	0.1813	0.4546	0.3238	0.6323	0.3807	0.1662	0.4270	0.3018	0.5863	0.3525	0.1535	0.3915	0.2762	0.5374	0.3234
ReaRec	0.0357	0.1871	0.1082	0.4576	0.1942	0.1507	0.3786	0.2633	0.6188	0.3400	0.1716	0.4265	0.3049	0.5551	0.3457	0.1459	0.3723	0.2298	0.5873	0.2987

Table 1: Evaluation results among baselines and AgenticRec . The best performance score is denoted in bold .

Data	Setting	Variant	H@1	H@5	N@5
CDs	Frozen	R	0.1907	0.4696	0.3303
	Frozen	TIRR	0.2331	0.5476	0.3955
	Trained	R	0.2324	0.5399	0.3865
		TIRR (RTA)	0.2837	0.6162	0.4606
		TIRR (RTA+PPR)	0.2992	0.6472	0.4795
Instr.	Frozen	R	0.1973	0.5613	0.3818

Table 2: Ablation study of variants of AgenticRec . R: pure r easoning by LLM, TIRR: t ool- i ntegrated r easoning r ecommendation.

Figure 3: Performance changes during PPR training on the Games dataset, including the agent’s preference judgment accuracy on the PPR test set and H@1 on the final recommendation test set.

Figure 4: Dynamics of token count and average H@10 of rollout trajectories during first-stage training on the Games dataset.

Figure 5: Performance of AgenticRec with varying backbone sizes.

Figure 6: An example showing the effectiveness of AgenticRec . Textured highlights denote critical reasoning steps that most significantly influence the final ranking and illustrate the differences between methods.

Data	Avg. Latency (s)	Avg. Tool Time (s)
CDs	16.233	0.010
Games	18.260	0.022
Office	20.678	0.019
Instruments	17.074	0.013

Table 3: Average latency statistics for our method.

Appendix figures & tables8 assets

Supplementary material from the paper’s appendix.

Appendix

Dataset	#Users	#Items	#Inters.	Sparsity
CDs	5,437	8,785	13,826	99.71%
Instruments	7,593	5,279	15,746	99.61%
Office	27,130	11,511	47,333	99.85%
Games	6,251	3,003	11,457	99.39%

Appendix

Table 4: Statistics of the preprocessed datasets.

Panel of Figure 7 — Figure 7: Training statistics on the Office dataset: (a) tool usage statistics, where the blue line denotes the average number of tool calls per trajectory and the orange line indicates the percentage of positively rewarded trajectories that have tool invocation; (b) recommendation performance measured by H@10 during training.

Figure 8: Effect of group size in RTA

Figure 9: Full token cost statistics.

Panel of Figure 10 — Figure 10: An example of user profile tool invocation.

Panel of Figure 11 — Figure 11: An example of item information tool.

Panel of Figure 12 — Figure 12: An example of behavioral statistics tool.

Panel of Figure 13 — Figure 13: An example of collaborative information tool invocation.

Recommender Systems

Recommender systems predict and rank items for people using interaction histories, content, context, and feedback.

See all 516 papers

Personalized Recommendation Tool Learning via Autonomous Language Agents
Jul 22, 2026Mingdai Yang, Zhiwei Liu, Weizhi Zhang +3LLM Tool Use Recommender Systems
Univ. of Illinois Chicago Chicago, IL, USA · Microsoft Mountain View, USA · Beihang University, & Hangzhou Innovation Institute of BUAA Beijing & Hangzhou, China
AgenticRecTune: Multi-Agent with Self-Evolving Skillhub for Recommendation System Optimization
Apr 21, 2026Xidong Wu, Yue Zhuan, Ruoqiao Wei +7Multi-Agent LLM Systems Multi-Agent System Optimization
Google Mountain View, California, USA
Agentic Recommender System with Hierarchical Belief-State Memory
May 14, 2026Xiang Shen, Yuhang Zhou, Yifan Wu +8LLM Agent Memory Recommender Systems
Meta Recommendation Systems (MRS)

Jul 22, 2026J/K move · Enter open · S save

Personalized Recommendation Tool Learning via Autonomous Language Agents

Mingdai Yang, Zhiwei Liu, Weizhi Zhang, Yibo Wang+2

Univ. of Illinois Chicago Chicago, IL, USA · Microsoft Mountain View, USA · Beihang University, & Hangzhou Innovation Institute of BUAA Beijing & Hangzhou, China

LLM Tool Use Recommender Systems Large Language Model-Based Recommendation

Open paper AskPDF arXiv

Although large language models (LLMs) have recently gained traction in recommender systems due to their strong reasoning capabilities and extensive world knowledge, previous LLM-based agents suffer from hallucination and context-length limitations, and thus are not suitable for full-ranking recommendation tasks. To circumvent these limitations through architectural design rather than modifying the LLM itself, we propose an agent-based recommendation framework, memory-based

\textbf{P}

ersonalized

\textbf{R}

ecommendation

\textbf{T}

ool learning via autonomous language

\textbf{A}

gents (PRTA), in which an LLM acts as a central planner interacting with multiple recommendation models as tools. The LLM-based agent is responsible for high-level reasoning and personalized tool selection, while traditional recommendation models perform full-ranking scoring, leveraging their scalability in modeling behavioral patterns. To support personalized tool selection, we design reflection mechanisms that enable the agent to evaluate and compare tools for each user based on user profiles and candidate ranked lists. Extensive experiments across three public datasets demonstrate the superiority of \modelname over traditional recommendation and LLM-based baselines in improving full-ranking recommendation performance.

Abstract

Recommender Systems

Explore similar work

Personalized Recommendation Tool Learning via Autonomous Language Agents

AgenticRecTune: Multi-Agent with Self-Evolving Skillhub for Recommendation System Optimization

Agentic Recommender System with Hierarchical Belief-State Memory