參考諮詢是圖書館提供讀者的重要服務,服務模式也從傳統的臨櫃諮詢,逐步轉變為線上同步或非同步模式,甚至發展到跨館合作線上聯合參考諮詢。隨著人工智慧、深度學習和自然語言技術的長足進步,對話機器人技術也漸臻成熟。許多即時互動軟體平臺紛紛釋出應用程式介面(application programming interface, API),方便開發者介接自家平臺提供對話諮詢服務。然而,在實際應用中,一些Chatbot由於無法準確辨識使用者的詢問意圖(intent),因此藉由限制使用者提問的方式來輔助系統正確理解使用者的詢問目的,從而搜尋對應的回覆結果。本文希望從蒐錄參考諮詢問答集語料進行訓練學習,擷取詞彙特徵重新建立群集,藉此能更正確的分析識別出讀者詢問意圖,未來可以提供讀者一個更精準預測詢問意圖與智慧化的參考諮詢機器人。
Reference consultation is an important service provided by libraries to their readers, and the service mode has gradually changed from traditional counter consultation to online synchronous or non-synchronous mode, and even developed to cross-library cooperation for online joint reference consultation. With the great progress of artificial intelligence, deep learning and natural language technology, the conversation robot technology is also getting mature. Many real-time interactive software platforms have released Application Programming Interfaces (APIs) to facilitate developers to connect to their own platforms to provide conversational consultation services. However, in practice, some Chatbots are unable to accurately recognize the user's intent, so they restrict the user's questions to help the system understand the purpose of the user's query and search for the corresponding response results. This paper endeavors to address this challenge by amassing a corpus of reference consulting question and answer sets for comprehensive training and learning purposes. By extracting vocabulary features and re-establishing clusters, our aim is to refine the system's ability to analyze and recognize the intent behind reader queries accurately. Ultimately, this will empower us to furnish readers with a more precise prediction of their query intent, paving the way for the development of an intelligent reference consultation robot poised to cater adeptly to their informational needs in the future.