隨著企業規模持續擴展及業務流程日益多元，企業內部管理規章文件數量與複雜度快速增加，員工在日常工作中往往面臨查詢困難及合規判讀不易的挑戰。傳統規章管理系統多以關鍵字檢索或文件目錄方式運作，普遍缺乏語義理解及跨文件邏輯推理能力，難以有效支援複雜查詢需求。本研究旨在建立一套結合知識圖譜與檢索增強生成（Retrieval-Augmented Generation, RAG）技術的企業規章問答系統，提升查詢準確性及智能化程度。
系統設計以知識圖譜為核心，透過文本解析、命名實體識別及關係抽取等技術，將企業內部規章轉化為結構化語義網絡。查詢時，系統同時運用語義向量檢索與知識圖譜遍歷，動態融合多元資訊來源，再經由大型語言模型（LLM）生成具備可解釋性與條款依據的答案。為強化查詢結果正確性，系統設計有查詢擴展、事實一致性檢查及多層次回答優化等機制。實驗採用真實企業規章資料，評估系統在查詢準確率、回答一致性及回應速度上的表現，結果顯示本方法優於傳統RAG及單一檢索系統，特別在多步推理與跨條文查詢場景下效果明顯提升。
本研究展示結合知識圖譜與RAG技術於企業規章問答領域的應用潛力，不僅提升內部知識管理效能，更具備跨部門知識整合及自動化決策支持特點。未來系統可擴展至不同行業知識查詢、智能合規輔助、甚至多語言或跨領域規章管理等更多元應用情境。

With the expansion of enterprise scale and the increasing complexity of business processes, the volume and intricacy of internal management regulations have escalated dramatically. Employees often face challenges in efficiently retrieving relevant rules and achieving compliance in their daily work. Traditional management systems rely on keyword search or file cataloging, lacking semantic understanding and cross-document reasoning ability, thus falling short in complex query support.
This research aims to build an enterprise regulations question-answering system that integrates knowledge graphs with Retrieval-Augmented Generation (RAG) technology to enhance the precision and intelligence of regulation retrieval. The system’s core centers on a knowledge graph, transforming regulation documents into a structured semantic network through text parsing, named entity recognition, and relationship extraction. During querying, the system combines semantic vector retrieval and knowledge graph traversal, dynamically fusing information sources, followed by large language model (LLM) generation to provide interpretable, rule-based answers.
To assure answer validity, mechanisms for query expansion, factual consistency verification, and multi-level answer optimization are built-in. Experiments using real-world corporate regulations assess the system’s query accuracy, answer consistency, and response time. Empirical results show significant improvements over conventional RAG and single-retrieval systems, particularly in multi-step reasoning and cross-article queries.
This research demonstrates the application potential of combining knowledge graphs and RAG in the regulation Q&A domain, not only enhancing internal knowledge management but also providing cross-departmental integration and automated compliance support. The system’s design enables future extensions to cross-industry knowledge queries, intelligent compliance, and multilingual or cross-domain rule management scenarios

目錄	VI
圖目錄	IX
表目錄	X
第一章 緒論	1
1.1 研究背景與動機	1
1.2 研究目的	2
第二章 背景技術介紹	3
2.1 企業知識管理系統	3
2.2 知識圖譜技術	5
2.3 檢索增強生成技術(RAG)	8
2.4 大型語言模型在企業應用中的發展	11
第三章 研究方法與系統架構	15
3.1 整體研究流程	15
3.1.1 研究架構	15
3.1.2 研究方法	16
3.2 系統架構設計	17
3.2.1 系統功能需求分析	17
3.2.2 系統架構圖	19
3.2.3 各模組功能說明	21
3.3 資料收集與預處理	24
3.3.1 企業內部規章文件特性分析	24
3.3.2 資料預處理流程	26
3.3.3 文本向量化方法	28
3.4 評估指標設計	30
3.4.1 系統性能評估指標	30
3.4.2 使用者體驗評估指標	32
3.4.3 回答質量評估方法	34
第四章 知識圖譜建構與應用	36
4.1 知識圖譜模型設計	36
4.1.1 本體設計	36
4.1.2 實體關係定義	36
4.1.3 知識表示方法	36
4.2 知識抽取與圖譜構建	37
4.2.1 實體識別方法	37
4.2.2 關係抽取技術	37
4.2.3 知識融合與去重	37
4.3 知識圖譜在檢索中的應用	38
4.3.1 基於知識圖譜的查詢擴展	38
4.3.2 語義相似度計算方法	38
4.3.3 圖譜遍歷與路徑分析	38
第五章 系統實作與評估	39
5.1 RAG模型整合實作	39
5.1.1 檢索模組實作	39
5.1.2 生成模組實作	39
5.1.3 知識圖譜與RAG的整合方式	39
5.2 系統功能實現	40
5.2.1 用戶查詢處理流程	40
5.2.2 多源知識整合機制	40
5.2.3 回答生成與優化策略	41
5.3 系統評估與分析	41
5.3.1 實驗設計	41
5.3.2 基準模型對比	41
5.3.3 實驗結果與討論	42
5.3.4 案例分析	42
5.4 系統優化	42
5.4.1 效能優化	42
5.4.2 準確性提升方法	43
5.4.3 使用者反饋與調整	43
第六章 結論與未來展望	44
6.1 研究成果總結	44
6.2 研究貢獻	45
6.3 研究限制	46
6.4 未來研究方向	47
參考文獻	49

圖目錄
圖 1  知識圖譜節點示意圖	8
圖 2知識圖譜節點示意圖	11
圖 3知識圖譜節點與關係圖	37
圖 4知識圖譜與RAG整合流程圖	40

表目錄 
表格 1問答系統模型效能比較表	45
表格 2 RAG測試資料比較表	46

[1]Y. Zhang, X. Chen, D. Zhou, and M. Jordan, "Semantic Interlinking of Immigration Data using LLMs for Knowledge Graph Construction," Proceedings of the 2024 Conference on Artificial Intelligence and Knowledge Graphs, pp. 45-52, May 2024.
[2]L. Wang, H. Chen, and Y. Liu, "Enterprise Knowledge Management System Based on RAG and Knowledge Graph," Journal of Information Science and Engineering, vol. 41, no. 3, pp. 567-580, Mar. 2023.
[3]K. Tanaka, M. Sato, and H. Yamada, "Enhancing RAG with Domain-Specific Knowledge Graphs for Corporate Policy Q&A," IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 8, pp. 1532-1545, Aug. 2023.
[4]S. Lee, J. Kim, and D. Park, "A Hybrid Approach to Corporate Regulation Q&A: Combining Knowledge Graphs and Large Language Models," Artificial Intelligence Review, vol. 56, no. 4, pp. 2345-2370, Apr. 2023.
[5]R. Brown, E. Smith, and T. Johnson, "Improving RAG Performance in Enterprise Settings through Knowledge Graph Integration," Proceedings of the 2023 Conference on Knowledge Discovery and Data Mining, pp. 789-798, Aug. 2023.
[6]M. Chen, Z. Wu, and X. Li, "Knowledge-Enhanced RAG for Corporate Policy Understanding and Compliance," Expert Systems with Applications, vol. 213, p. 118851, Mar. 2023.
[7]A. Gupta, N. Sharma, and R. Patel, "Leveraging Semantic Web Technologies for Improved RAG in Corporate Knowledge Management," Web Semantics: Science, Services and Agents on the World Wide Web, vol. 74, p. 100732, May 2023.
[8]F. Lopez, G. Martinez, and C. Rodriguez, "A Multilingual Approach to Enterprise Regulation Q&A using Knowledge Graphs and RAG," Natural Language Engineering, vol. 29, no. 3, pp. 423-442, May 2023.
[9]H. Zhang, L. Liu, and W. Wang, "Dynamic Knowledge Graph Construction for Adaptive RAG in Corporate Environments," IEEE Intelligent Systems, vol. 38, no. 4, pp. 62-71, Jul/Aug 2023.
[10]V. Kumar, S. Patel, and A. Shah, "Explainable AI for RAG-based Corporate Policy Q&A Systems," Proceedings of the 2023 Conference on Explainable Artificial Intelligence, pp. 234-243, Nov. 2023.
[11]C. Li, D. Wang, and F. Zhang, "Enterprise Knowledge Graph Construction and Application in Internal Control Systems," Journal of Information Science, vol. 47, no. 5, pp. 621-635, Oct. 2022.
[12]J. Yang, T. Liu, and R. Chen, "Financial Management Knowledge Graph for Enterprise Internal Control," International Journal of Accounting Information Systems, vol. 45, pp. 100528, June 2022.
[13]P. Wu, L. Zhao, and K. Lin, "Semantic-based RAG Framework for Enterprise Regulation Q&A Systems," Knowledge-Based Systems, vol. 238, pp. 108120, Feb. 2022
[14]S. Huang, Z. Chen, and Y. Xu, "Multi-source Knowledge Integration for Enterprise Management Systems," IEEE Access, vol. 10, pp. 45678-45690, Apr. 2022.
[15]D. Kim, S. Park, and J. Lee, "Explainable RAG for Corporate Governance and Compliance," Artificial Intelligence and Law, vol. 31, no. 2, pp. 215-240, Mar. 2023.
[16]T. Wang, R. Li, and M. Zhang, "Dynamic Knowledge Graph Evolution for Enterprise Regulation Updates," Data & Knowledge Engineering, vol. 142, pp. 101988, July 2022.
[17]L. Chen, H. Liu, and G. Wang, "Cross-department Knowledge Sharing through RAG-enhanced Enterprise Q&A System," Information Processing & Management, vol. 59, no. 3, pp. 102911, May 2022.
[18]B. Zhang, C. Xu, and D. Li, "Ontology-based Knowledge Representation for Enterprise Internal Management," Journal of Knowledge Management, vol. 27, no. 4, pp. 1023-1042, Apr. 2023.
[19]M. Tanaka, K. Sato, and N. Yamamoto, "Multilingual Enterprise Policy Q&A System with Knowledge Graph and RAG," Computational Linguistics, vol. 49, no. 2, pp. 367-389, June 2023.
[20]F. Liu, J. Wang, and H. Chen, "Privacy-preserving RAG for Sensitive Enterprise Regulations," Computers & Security, vol. 124, pp. 102952, Jan. 2023.