透過您的圖書館登入
IP:18.218.54.178
  • 學位論文

基於問答集之文字客服機器人-以大學招生應用為例

An FAQ-based Text Bot Service for University Admissions

指導教授 : 魏世杰

摘要


隨著行動網路的普及,智慧客服近年來快速興起,改善了過往人工客服諸如耗時、人力成本高、不易長時間大量配置等缺點。為減輕大學招生部門的人力配置,本研究建置了基於問答集之文字客服機器人,即時提供問答服務,供有意入學的申請者參考。 為了計算用戶提問與問答集的相似度,本研究評估了3種句子向量化及比對方法,包含 TensorFlow Hub 平台提供的通用句子編碼器,問答型通用句子編碼器,和 BERT單一句子編碼器。3種方法皆經過實驗挑選合適的相似度門檻值。在經由人工分析問答記錄後,得知問答型通用句子編碼器表現最佳。在兩次不同對象下,採用科技接受模式構面之問卷調查後,得知問答型通用句子編碼器和BERT單一句子編碼器分別表現最佳。 另外,前台互動方面設計了LINE 與網頁兩種介面供評估。在採用使用者介面量表之問卷調查後,得知LINE介面於「認知負荷」構面及「資訊品質」構面上表現較佳;網站介面則於「系統態度」構面及「錯誤解決性」構面上表現較佳。

並列摘要


With the ubiquitous mobile networks, the intelligent customer services have seen a quick boom in recent years. Compared with traditional human customer services, the drawbacks of time-consuming, high labor cost, and difficulty in long hours of mass deployment have been thus improved. In order to reduce the manpower allocated for university admissions, the study has built an FAQ-based text chatbot to provide the real-time question answering service for prospective applicants. To compute the similarity between the user query and the frequently asked questions, three methods for sentence embedding and matching are evaluated. From TensorFlow Hub, they include the universal sentence encoder, the universal sentence encoder for question answering, and the BERT single-sentence encoder. Experiments have been conducted to set up appropriate similarity thresholds for the three methods. From log analysis based on human judgement, it is found that the universal sentence encoder for question answering performs the best in accuracy. In two questionnaire surveys targeting different subjects, it is found that based on factors in the technology acceptance model, the methods using the universal sentence encoder for question answering and the BERT single-sentence encoder perform the best separately. In addition, the front-end interactive system is designed with the LINE and web interfaces for assessment. From a questionnaire survey based on the user interface scale, it is found that the LINE interface performs better on the “cognitive load” and “information quality” dimensions while the web interface performs better on the “system attitude” and “error resolution” dimensions.

參考文獻


英文文獻
Adamopoulou, E., & Moussiades, L. (2020). An overview of chatbot technology. IFIP International Conference on Artificial Intelligence Applications and Innovations,
Bangor, A., Kortum, P., & Miller, J. (2009). Determining what individual SUS scores mean: Adding an adjective rating scale. Journal of usability studies, 4(3), 114-123.
Brooke, J. (1996). SUS-A quick and dirty usability scale. Usability evaluation in industry, 189(194), 4-7.
Bughin, J., Seong, J., Manyika, J., Chui, M., & Joshi, R. (2018). NOTES FROM THE AI FRONTIER MODELING THE IMPACT OF AI ON THE WORLD ECONOMY. Mckinsey&Company. Retrieved 02.05 from https://www.mckinsey.com/featured-insights/artificial-intelligence/notes-from-the-AI-frontier-modeling-the-impact-of-ai-on-the-world-economy#

延伸閱讀