.png){kind=icon}
.png){kind=icon}
Return
RoCoMAP: A Role-Based Collaborative Multi-Agent Framework for Patent Supply–Demand Matching
.png){kind=icon}
PRE
AI
DOI：10.1007/s11192-026-05777-w.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Effective supply–demand matching is essential for technology transfer, yet current online technology trading platforms are hindered by low-quality information and simplistic text-matching methods that fail to capture the complexity of real-world scenarios. Motivated by recent advances in large language model-based multi-agent systems, we propose RoCoMAP, a role-based collaborative multi-agent framework that simulates expert reasoning through coordinated agent interactions. RoCoMAP automates the full spectrum of the matching process, encompassing early-stage demand diagnosis, supply candidate identification, multi-dimensional matching, and final verification, thereby streamlining technology transfer workflows. We evaluate RoCoMAP on real-world data from the Innovation China platform, including large-scale technology demand records, curated patent supply databases with varying similarity characteristics, and a set of demand–supply pairs combining verified transaction cases and human-validated matches. Experimental results show that RoCoMAP delivers consistently strong performance across all core tasks. In demand diagnosis, it achieves substantially higher classification accuracy than machine learning, deep learning, LLM-based, and multi-criteria decision-making methods. In supply retrieval, the framework maintains stable top-k recall across supply databases with varying similarity characteristics. For supply–demand matching, RoCoMAP consistently outperforms representative pre-trained language model and multi-agent baselines by more than 10 percentage points, while maintaining robustness under heterogeneous supply settings. Moreover, the framework remains robust across different base LLMs. This research introduces an AI-powered paradigm for intelligent and automated technology supply–demand matching, facilitating a shift from traditional discriminative models with limited transparency toward dynamic and interpretable generative methodologies.
Keywords:
Technology supply–demand matching
Online technology trading platforms
Large language models
Multi-agent collaborative
Journal
IF:
3.5
Papers:
8.1K
Citations:
2.2W