國立臺灣師範大學圖書館參考諮詢機器人建置與評估

Abstract

通訊軟體盛行，再加上自然語言處理的進步，人工智慧的應用又以聊天機器 人更為備受關注。有圖書館將聊天機器人導入，以提供新式的即時數位參考服務， 解決人工即時參考服務的問題。本研究目的在於建置國立臺灣師範大學(簡稱臺 師大)圖書館參考諮詢機器人並調查其使用評價。 本研究建置之參考諮詢機器人採用臺師大圖書館參考問答紀錄與常問問答 集作為問答語料，LINE Bot 作為開發平台，以 Python 程式語言編寫程式，並使 用 Jieba 進行中文斷詞、提取關鍵字，運用 Dice 係數計算問題與關鍵字相似度比 對，Beautiful Soup 協助擷取臺師大圖書館網站與館藏系統資訊，達成擁有館藏 查詢、開館查詢與問答語料庫回覆三大功能之參考諮詢機器人。 使用評估測試方面，請十名臺師大學生執行以問答語料中歸納之五個分類加 上館藏與開館查詢功能，設計而成的七項任務，依據機器人之回覆給予五分量表 滿意度。結果顯示，整體而言受測者對於參考諮詢機器人之回覆感到滿意，提問 之回覆正確率為 70%，系統中的回覆正確率則為 82.86%，滿意度較高的任務回 覆，其正確率也較高，唯獨任務五—針對「資源查詢指引」類別進行提問—滿意 度是七項任務中最低的，但其系統中回覆正確率卻有 90.00%。 未來研究可以擴增問答語料庫問答紀錄、增進問題比對的精確度、改進答案 呈現方式，以更加完善參考諮詢機器人之功能並解決任務五滿意度偏低之問題。Communication software is prevalent, and with the advancement of natural language processing, chatbots are a popular application of artificial intelligence. Some libraries have developed chatbots to provide a new type of real-time digital reference service to solve the problem of manual instant reference service. The purpose of this study is to develop a reference chatbot for National Taiwan Normal University Library (NTNU Library) and evaluate its performance from aspects of users. The reference chatbot developed in this study uses the reference question and answer set of NTNU Library and the FAQs as a question and answer corpus. The development platform is Line Bot, the programming language is Python, and the tool for performing Chinese word segmentation and keyword extraction is Jieba. Dice coefficient is used for the calculation of the similarity between a problem and keywords, and Beautiful Soup assists in crawling the information of the library website and OPAC of NTNU. The reference chatbot developed in this study comprises three functions: inquiry about library collection, inquiry about library opening hours and inquiry about Q&A corpus . In terms of the evaluation test, ten NTNU’s students were asked to perform seven tasks, and the five-level Likert scale were given according to the response of the chatbot. The results showed that the respondents were satisfied with the response of the reference chatbot. The correct response rate was 70%, and the system's response accuracy rate was 82.86%. The response with higher satisfaction was also correct. Only task five, inquiry about the library resources , receives the lowestsatisfaction, but the system's response accuracy rate was 90.00%. Future research can amplify the question and answer corpus, improve the accuracy of problem comparison, and improve the way in which answers are presented, so as to further improve the function of the reference chatbot and solve the problem of low satisfaction of the task five.