This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
RCR-Router: Efficient Role-Aware Context Routing for Multi-Agent LLM Systems with Structured Memory
Created by
Haebom
Author
Jun Liu, Zhenglun Kong, Changdi Yang, Fan Yang, Tianqi Li, Peiyan Dong, Joannah Nanjekye, Hao Tang, Geng Yuan, Wei Niu, Wenbin Zhang, Pu Zhao, Xue Lin, Dong Huang, Yanzhi Wang
Outline
This paper proposes RCR-Router, a novel routing framework for efficient collaboration in multi-agent large-scale language model (LLM) systems. To overcome the limitations of existing static or global context routing strategies (excessive token consumption, unnecessary memory exposure, and lack of adaptability between interaction rounds), RCR-Router adopts a modular, role-aware approach that dynamically selects semantically relevant memory subsets based on each agent's role and task stage. A lightweight scoring policy guides memory selection, and agent outputs are iteratively merged into a shared memory store to incrementally improve the context. Furthermore, we propose the Answer Quality Score metric, which captures the explanations generated by LLMs beyond standard QA accuracy, to better evaluate model behavior. Experimental results on three multi-hop QA benchmarks (HotPotQA, MuSiQue, and 2WikiMultihop) demonstrate that RCR-Router maintains or improves answer quality while reducing token usage by up to 30%.
Takeaways, Limitations
•
Takeaways:
◦
A novel approach for efficient collaboration in multi-agent LLM systems (RCR-Router).
◦
Achieving reduced token usage and improved response quality through role-awareness and dynamic memory pathing.
◦
Proposing a new metric (Answer Quality Score) for qualitative evaluation of LLM creation explanations.
◦
Emphasizes the importance of establishing structured memory paths and evaluating output recognition.
•
Limitations:
◦
Further research is needed to determine the generalizability and objectivity of the proposed Answer Quality Score indicator.
◦
Validation of the generalization performance of RCR-Router for various types of multi-agent LLM tasks is needed.
◦
There is a need to evaluate the scalability of RCR-Router and its stability for complex interaction scenarios.
