企業智慧財務決策支援系統：基於語意查詢與圖形檢索增強生成之雲端資源管理架構;Enterprise Intelligent Financial Decision Support System: A Cloud Resource Management Framework Based on Semantic Query and Graph Retrieval-Augmented Generation

NCUIR > School of Management at National Central University > Graduate Institute of Information Management > Electronic Thesis & Dissertation > Item 987654321/98249

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/98249

Title:	企業智慧財務決策支援系統：基於語意查詢與圖形檢索增強生成之雲端資源管理架構;Enterprise Intelligent Financial Decision Support System: A Cloud Resource Management Framework Based on Semantic Query and Graph Retrieval-Augmented Generation
Authors:	鍾舒渟;Shu-Ting, Chung
Contributors:	資訊管理學系
Keywords:	企業智慧財務決策支援系統;圖形檢索增強生成;語意查詢;檢索增強生成;雲端財務管理;多因子成本分析;Enterprise Financial Decision Support System;GraphRAG;Semantic Query;Retrieval-Augmented Generation;Cloud Financial Management;Multi-Factor Cost Analysis
Date:	2025-07-08
Issue Date:	2025-10-17 12:32:46 (UTC+8)
Publisher:	國立中央大學
Abstract:	隨著雲端運算環境日益多元，企業在資源配置與成本控管方面，面臨異質資料整合與語意理解的挑戰。傳統分析工具難以支援自然語言查詢與條件組合的語意解析，導致查詢效率不足、使用門檻高，限制了財務資訊的靈活運用。為解決此問題，本研究提出一套企業智慧財務決策支援系統，建構於圖形檢索增強生成(Graph-Based RAG, GraphRAG)架構，結合語意向量檢索與圖形資料關聯分析，提升查詢語意擷取準確性與回應結構一致性。本系統設計包含語意解析、資料轉換、語意嵌入、查詢節點展開與生成整合等模組，使用者可透過自然語言輸入多層次條件查詢，系統將語句轉換為語意向量，結合圖形資料庫處理欄位間語意邏輯關係，並由語言模型生成條件式回應內容，協助使用者完成雲端成本資訊的結構化查詢。本研究設計五項代表性查詢任務，分別為欄位定位、單日查詢、多日查詢、語意條件解析與成本加總，模擬企業雲端財務管理中常見需求。並以傳統 RAG 架構作為對照組進行實驗評估，觀察系統在查詢效率、語意擷取精度、語境一致性與資訊可追溯性等層面的表現。實驗結果顯示，GraphRAG 架構具備語意導向條件組合查詢的應用潛力，有助於強化企業財務查詢的結構一致性與操作便利性。綜上所述，本研究建構一套具備語意解析能力之智慧決策查詢系統，為雲端財務查詢應用提供實務基礎與後續應用參考。;As cloud computing environments diversify, enterprises face challenges in resource allocation and cost control due to heterogeneous data integration and semantic complexity. Traditional tools struggle to support natural language queries and conditional semantic parsing, limiting efficiency and usability. This study proposes an intelligent financial decision support system based on a Graph-Based Retrieval-Augmented Generation (GraphRAG) architecture, integrating vector-based semantic retrieval with graph-based relationship analysis to improve query accuracy and consistency. The system includes modules for semantic parsing, data transformation, embedding, node expansion, and query generation. Users can submit multi-level natural language queries without needing to understand data structures. The system maps semantic logic across fields using a graph database and generates condition-based responses via a language model. Five representative query tasks—field localization, single-day query, multi-day query, semantic condition parsing, and cost aggregation—simulate typical enterprise financial scenarios. A comparative experiment with traditional RAG evaluates performance in query efficiency, semantic precision, contextual consistency, and traceability. Results indicate that GraphRAG supports semantically guided conditional queries and enhances structural alignment in financial data access. This research presents a practical, semantic-aware support framework for optimizing cloud-based financial queries in enterprise environments.
Appears in Collections:	[Graduate Institute of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	10	View/Open

社群 sharing

Loading...