|
| 基于智能体的水利科技报告形式审查系统构建研究 |
| Research on the Construction of a Water Resources Technology Report Formal Review System Based on Intelligent Agents |
| 投稿时间：2025-08-17 修订日期：2026-02-03 |
| DOI： |
| 中文关键词: 智能体 报告智能审查 大语言模型 Retrieval-Augmented Generation（RAG） LoRA微调 |
| 英文关键词: agent intelligent report review large language model Retrieval-Augmented Generation (RAG) LoRA fine-tuning |
| 基金项目:中国水科院基本科研业务费专项项目（JZ0145C072025）；水利部数字孪生流域重点实验室开放研究基金资助（Z0202042022） |
|
| 摘要点击次数: 34 |
| 全文下载次数: 0 |
| 中文摘要: |
| 报告审查作为项目质量控制的核心环节，传统人工审查方法存在一定的效率挑战和标准执行不一致等问题，现有通用审查系统难以适应水利科技报告结构复杂、审查任务多维与领域知识依赖的多重挑战。为此，本研究提出基于智能体的水利科技报告形式审查系统架构。通过智能体驱动的动态任务规划与协同审查机制，实现了对复杂报告结构的自适应解析与多任务编排；形成了基于LoRA微调和检索增强生成（RAG）技术的水利领域知识增强机制，构建了包含27,005条水利专业术语的术语知识库和计算关系库，形成10,358个样本的指令微调数据集，提升了系统在专业术语、计算逻辑及合同一致性等任务上的审查能力。以人工审查结果为基准，智能体系统对八类审查任务的F1均值在80%以上。实验结果表明：该系统实现了水利科技报告形式审查的全流程智能化处理，显著提升效率的同时保障了审查准确率。本研究构建的智能审查系统为水利项目文档质量管理提供了标准化工具支撑，为专业领域文本智能化审查提供了可参考的技术路径。 |
| 英文摘要: |
| Report review, as a core component of project quality control, faces efficiency challenges and inconsistent standard implementation with traditional manual review methods. Existing generic review systems struggle to adapt to the complex structural characteristics, multidimensional review tasks, and domain knowledge dependencies inherent in water resources science and technology reports. To address this, this study proposes an agent-based formal review system architecture for water resources science and technology reports. Through agent-driven dynamic task planning and collaborative review mechanisms, it achieves adaptive parsing of complex report structures and multi-task orchestration. It establishes a knowledge enhancement mechanism for the water resources domain based on LoRA fine-tuning and Retrieval-Augmented Generation (RAG) technology. A terminology knowledge base containing 27,005 specialized water resources terms and a computational relationship database are constructed, forming a command fine-tuning dataset with 10,358 samples. This enhances the system's review capabilities in tasks involving specialized terminology, computational logic, and contractual consistency. Using human review results as the benchmark, the agent system achieved an average F1 score above 80% across eight review tasks. Experimental results demonstrate that this system enables fully intelligent processing throughout the formal review workflow of water resources science and technology reports, significantly improving efficiency while ensuring review accuracy. The intelligent review system constructed in this study provides standardized tool support for water resources project document quality management and offers a reference technical pathway for intelligent text review in specialized domains. |
|
查看/发表评论 下载PDF阅读器 |
| 关闭 |