改變校園對話：基於具檢索增強生成的大語言模型聊天機器人

Abstract

大型語言模型(LLM)可謂是 2023 年人工智慧領域中最為熱門的名詞之一，而其在各種場景中的應用方法之一即為檢索增強生成(RAG)，因此受到廣泛關注。雖然在學術界受到高度重視，但目前在實際生活中的應用仍相對有限。為此，本論文特別選擇了一個與我身為學生緊密相關的場域 ── 學校，成功地將檢索增 強生成的大型語言模型應用於實際生活中。在語言模型資源相對較為匱乏的繁體 中文情境中，這篇作品整合了多項技術包含網頁爬蟲及網頁資料清理、嵌入檢索(embedding retrieval)、文檔切分最佳化、前後端、line 聊天機器人 UI 整合，最終取得了成功的應用。實現將技術應用到實際情境中的重要里程碑，檢索增強生成 的大語言模型朝著實現高品質、個人化支持及普及化的目標邁進。

Large Language Models (LLMs) can be considered one of the hottest terms in the field of artificial intelligence in 2023, and one of their application methods among various scenarios is Retrieval-Augmented Generation (RAG), which attracts widespread attention. Although highly regarded in the academia, their practical applications in real life are relatively limited. Therefore, this paper specifically chooses a field closely related to me as a student—school—and successfully applied the Retrieval-Augmented Generation of LLMs to real-life situations. In the context of Traditional Chinese where language model resources are relatively scarce, this work integrates multiple technologies, including web crawling and data cleaning, embedding retrieval, chunk optimization, front-end and back-end development, and Line chatbot UI integration, ultimately achieving successful application. This milestone in applying technology to real-world scenarios propels LLMs in Retrieval-Augmented Generation towards the goals of achieving high-quality, personalized support, and widespread use.