數位轉型對於企業提升競爭力至關重要，但在管理數位化資源時也面臨多種挑戰，例如知識儲存分散與專家經驗難以傳承等。而大型語言模型（Large Language Model, LLM）雖然能有效提取並結構化知識，但其幻覺、安全性及缺乏可解釋性的問題也引發了擔憂。本研究提出DocuKMP系統，該系統結合知識圖譜（Knowledge Graph, KG）與LLM技術形成知識管理平台，系統內建知識圖譜建置工具鏈，解決了知識圖譜使用非結構化資料進行建置時的限制，並提供問答功能供使用者進行知識查詢，同時也設計使用者介面方便使用者與系統進行互動。而我們也分別針對知識圖譜建置工具鏈中的各個軟體工具進行實驗，並對DocuKMP系統所建置的知識圖譜進行四項品質標準評估，結果發現準確率為95.25%、一致性為100%、完整性為92.25%、冗餘度為99.95%，顯示知識圖譜具有相當高的品質。此外，DocuKMP系統在正確率上（82.35%）明顯優於RAG-Token和RAG-Sequence模型，且對硬體資源需求較低。綜上所述，DocuKMP系統在知識管理的正確性與資源效率方面均優於RAG模型，對企業數位轉型具有重要意義。;Digital transformation is crucial for enhancing the competitiveness of enterprises, yet managing digital resources poses various challenges such as scattered knowledge storage and difficulty in transferring expert experience. Although Large Language Models (LLMs) can effectively extract and structure knowledge, concerns about hallucinations, security, and lack of interpretability have been raised. This study proposes the DocuKMP system, which integrates Knowledge Graph (KG) and LLM technologies to form a knowledge management platform. The system includes a built-in knowledge graph construction toolchain that addresses the limitations of using unstructured data for building knowledge graphs, and it provides a Q&A function for users to query knowledge. Additionally, a user interface is designed to facilitate user interaction with the system. We conducted experiments on various software tools in the knowledge graph construction toolchain and evaluated the knowledge graph built by the DocuKMP system based on four quality standards. The results show an accuracy of 95.25%, consistency of 100%, completeness of 92.25%, and redundancy of 99.95%, indicating a high-quality knowledge graph. Moreover, the DocuKMP system outperforms the RAG-Token and RAG-Sequence models in terms of accuracy (82.35%) while requiring fewer hardware resources. In summary, the DocuKMP system excels in both accuracy and resource efficiency for knowledge management, making it significant for enterprise digital transformation.
