arXiv cs.AI· Junxing Hu, Ai Han, Haolan Zhan, Pu Wei, Zhiqian Zhang, Yuhang Guo, Jiawei Lu, Zhen Chen, Haoran Li, Zicheng Zhang, Ling Li·· 13 小时前AI 评分33
HiMA-Ecom：面向分层多智能体电商助手的长程记忆基准
HiMA-Ecom: A Long-Horizon Memory Benchmark for Hierarchical Multi-Agent E-commerce Assistants
AI 导读
研究者提出 HiMA-Ecom，一个面向电商助手的分层多智能体长程记忆基准，含 22.8K 条实例，其中 17.7K 条在上下文中带有记忆条目。基于该基准的 HiMA-R1 采用 VR-GRPO 控制轨迹上下文预算，并用奖励驱动记忆机制将收敛所需更新步数减少约 55%。基于 3B 和 7B 开源模型构建的智能体平均追平 DeepSeek-R1，并超过 DeepSeek-V3 6.0 分。
正文 · 原文
Authors:Junxing Hu, Ai Han, Haolan Zhan, Pu Wei, Zhiqian Zhang, Yuhang Guo, Jiawei Lu, Zhen Chen, Haoran Li, Zicheng Zhang, Ling Li
Abstract:Long-horizon agentic systems become long-context systems by accumulation rather than through a single long document, since every step reassembles a prompt from role instructions, tool and sub-agent descriptions, recalled memories, dialogue history, and previous tool outputs, separately for every agent. This paper studies how such accumulated context can be trained, evaluated, and managed in hierarchical multi-agent assistants. We present HiMA-Ecom, a hierarchical multi-agent benchmark for e-commerce assistants with 22.8K instances, of which 17.7K carry memory entries in context, pairing agent-specific supervised fine-tuning data with system-level supervision for end-to-end multi-agent reinforcement learning. Building upon it, HiMA-R1 keeps the context budget of a trajectory under control through Variance-Reduction Group Relative Policy Optimization (VR-GRPO), which samples along an initial trajectory so that sampling cost grows additively with trajectory length and updates the agents with the largest reward variance, together with an efficiency reward on trajectory length and a memory evolution mechanism that repurposes GRPO rewards as cost-free signals for what stays in context. Agents built upon 3B and 7B open-source models match DeepSeek-R1 and surpass DeepSeek-V3 by 6.0 points on average, while reward-driven memory reduces the update steps needed for convergence by roughly 55%.
| Comments: | 33 pages, 9 figures. Accepted at the Workshop on Long-Context Foundation Models (LCFM), NeurIPS 2026 |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2506.19846 [cs.AI] |
| (or arXiv:2506.19846v3 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2506.19846 arXiv-issued DOI via DataCite |
Submission history
From: Junxing Hu [view email]
[v1]
Tue, 24 Jun 2025 17:59:31 UTC (7,799 KB)
[v2]
Wed, 1 Apr 2026 10:25:03 UTC (7,712 KB)
[v3]
Tue, 27 Oct 2026 09:05:05 UTC (6,259 KB)
来源：arXiv cs.AI · arxiv.org