結合 RAG 與 LangGraph 架構提升 LLM 於台灣開放資料平台搜尋效率之研究;A Study on Enhancing Search Efficiency of Large Language Models on Taiwan Open Data Platforms via RAG and LangGraph Architectures

NCUIR > School of Management at National Central University > Graduate Institute of Information Management > Electronic Thesis & Dissertation > Item 987654321/98235

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/98235

Title:	結合 RAG 與 LangGraph 架構提升 LLM 於台灣開放資料平台搜尋效率之研究;A Study on Enhancing Search Efficiency of Large Language Models on Taiwan Open Data Platforms via RAG and LangGraph Architectures
Authors:	劉冠吟;LIU, GUAN-YIN
Contributors:	資訊管理學系
Keywords:	語意搜尋;開放資料;大型語言模型;RAG;LangGraph;Large Language Models;Semantic Retrieval;Retrieval-Augmented Generation Architecture;LangGraph;Open Data Query
Date:	2025-07-03
Issue Date:	2025-10-17 12:31:35 (UTC+8)
Publisher:	國立中央大學
Abstract:	隨著政府積極推動數位轉型與開放資料政策，台灣已建置多個涵蓋政策法規、統計資訊及公共服務的開放資料平台。然而，這些珍貴資料對一般使用者而言，常因查詢方式過於技術導向、語意支援不足或無法即時應對複雜需求，而形成「資料可用」與「資料能用」之間的落差。有鑑於此，研究導入語意檢索增強生成架構（RAG）、多代理任務鏈架構（LangGraph）與向量資料庫 ChromaDB，結合大型語言模型（LLM）之語意理解能力，開發一套具備語意比對與回應能力的智慧查詢系統。系統可透過自然語言輸入自動擷取開放資料內容，並以 ChatGPT-4.1 回應具上下文意義之查詢需求，前端則採用 Python 與 Streamlit 建構互動介面。初步測試結果顯示，系統具備良好的語意涵蓋率與回應準確性，並能有效支援政策查詢與公共資訊應用場景，展現其在智慧政府應用中的潛力。;This study presents a semantic search system that enhances query accuracy and user experience on Taiwan’s open data platforms. By integrating Retrieval-Augmented Generation (RAG), LangGraph-based workflow orchestration, and GPT-4.1 via OpenAI API, the system overcomes the limitations of traditional keyword search, enabling more context-aware and explainable responses. The architecture uses ChromaDB for vector-based semantic retrieval and Streamlit as the user interface. Multi-agent modules manage tasks such as intent classification, data retrieval, summarization, and answer generation. Real-world scenarios demonstrate the system’s ability to handle complex and ambiguous queries with high semantic precision. Evaluation results—using Success@1, Recall, F1-score, and user feedback via TAM and IS models—show improved effectiveness, user satisfaction, and semantic coverage. This work highlights the potential of combining LLMs with structured knowledge and modular workflows to support smarter, more accessible public data services. I am deeply grateful to my advisor for the invaluable support and guidance throughout this research journey.
Appears in Collections:	[Graduate Institute of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	19	View/Open

社群 sharing

Loading...