《计算机应用》唯一官方网站 ›› 2026, Vol. 46 ›› Issue (3): 708-714.DOI: 10.11772/j.issn.1001-9081.2025030349

• 人工智能 • 上一篇    下一篇

多Agent协作的知识推理框架

王日龙1,2, 李振平1, 李晓松1, 高强1, 何亚3, 钟勇3, 赵英潇1()   

  1. 1.军事科学院 军事科学信息研究中心,北京 100142
    2.上海交通大学 电子信息与电气工程学院,上海 200240
    3.中国科学院 成都计算机应用研究所,成都 610213
  • 收稿日期:2025-04-03 修回日期:2025-07-01 接受日期:2025-07-03 发布日期:2025-08-06 出版日期:2026-03-10
  • 通讯作者: 赵英潇
  • 作者简介:王日龙(1996—),男,湖南益阳人,硕士研究生,主要研究方向:情报学、自然语言处理
    李振平(1990—),男,河南郑州人,博士研究生,主要研究方向:自然语言处理
    李晓松(1981—),安徽宿松人,研究员,博士,主要研究方向:数据分析
    高强(1987—),男,浙江长兴人,高级工程师,硕士,主要研究方向:数据科学、大语言模型
    何亚(1981—),男,四川成都人,高级工程师,硕士,主要研究方向:智慧城市、大数据、数据挖掘、机器学习
    钟勇(1966—),男,四川岳池人,研究员,博士,CCF会员,主要研究方向:大数据、人工智能、软件过程
  • 基金资助:
    国家社会科学基金重大项目(23ZDA119)

Multi-Agent collaborative knowledge reasoning framework

Rilong WANG1,2, Zhenping LI1, Xiaosong LI1, Qiang GAO1, Ya HE3, Yong ZHONG3, Yingxiao ZHAO1()   

  1. 1.Military Science Information Research Center,Academy of Military Science,Beijing 100142,China
    2.School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China
    3.Chengdu Institute of Computer Application,Chinese Academy of Sciences,Chengdu Sichuan 610213,China
  • Received:2025-04-03 Revised:2025-07-01 Accepted:2025-07-03 Online:2025-08-06 Published:2026-03-10
  • Contact: Yingxiao ZHAO
  • About author:WANG Rilong, born in 1996, M. S. candidate. His research interests include information science, natural language processing.
    LI Zhenping, born in 1990, Ph. D. candidate. His research interests include natural language processing.
    LI Xiaosong, born in 1981, Ph. D., research fellow. His research interests include data analysis.
    GAO Qiang, born in 1987, M. S., senior engineer. His research interests include data science, large language model.
    HE Ya, born in 1981, M. S., senior engineer. His research interests include smart city, big data, data mining, machine learning.
    ZHONG Yong, born in 1966, Ph. D., research fellow. His research interests include big data, artificial intelligence, software process.
  • Supported by:
    Major Project of National Social Science Foundation of China(23ZDA119)

摘要:

情报分析的核心任务是从海量数据中提取事件关联并推导因果关系。然而,当前基于大语言模型(LLM)的方法在处理长文本时,受限于上下文窗口和计算复杂度,难以有效捕捉事件间的因果关联,导致推理能力显著下降,这一现象在参数规模较小的语言模型中尤为突出。针对这一问题,提出一种多Agent协作的知识推理(MAKR)框架。该框架通过增量式构建实体关系图,显式建模实体间的复杂关联,从而辅助LLM实现更精准的因果推理。双塔结构的设计使图模型与语言模型能够分别独立处理图结构信息和文本信息,并通过融合机制增强模型对长文本中复杂逻辑关系的理解能力。此外,在语言预测任务的基础上,增加针对图中节点关系的预测任务,从而进一步优化语义对齐效果。实验结果表明,在有限的计算资源条件下,在GDELT数据集的安全事件分析和OpenSanctions数据集的制裁关联分析任务中,MAKR框架的事件预测和因果推断的性能均显著优于HetGNN(Heterogeneous Graph Neural Network)和HiGPT等对比方法,验证了该框架在计算资源受限的工业场景中的实用价值。

关键词: 图神经网络, 大语言模型, 图结构感知, 军事情报分析, 图结构统一表示

Abstract:

The core mission of intelligence analysis is to extract event associations and perform causal inference from massive data. However, the existing Large Language Model (LLM)-based methods are constrained by the context window and computational complexity in processing long texts, so as to have difficulty in effective capture of causal associations between events, leading to a significant decline in inference capability, particularly in language models with limited parameter sizes. To address this issue, a Multi-Agent collaborative Knowledge Reasoning (MAKR) framework based on a dual tower structure was proposed. In the framework, the complex associations among entities were modeled explicitly through incremental construction of an entity relationship graph, thus assisting LLMs to achieve more accurate causal inference. The dual tower structure was designed to enable the graph model and the language model to process graph structure information and textual information, respectively. The comprehension ability of the model for complex logical relationships within long texts was enhanced through a fusion mechanism. Additionally, a prediction task for node relations in the graph was added on the basis of the language prediction task, thereby further optimizing the semantic alignment effect. Experimental results demonstrate that under conditions of limited computational resources, MAKR framework achieves superior performance in both event prediction and causal inference compared to the existing methods such as HetGNN (Heterogeneous Graph Neural Network) and HiGPT in the tasks of security event analysis of the GDELT dataset and sanction associated analysis of the OpenSanctions dataset. In conclusion, the practical value of the framework in industrial scenarios with constrained computational resources was validated.

Key words: Graph Neural Network (GNN), Large Language Model (LLM), graph structure awareness, military intelligence analysis, unified graph structure representation

中图分类号: