客製化任務導向於電信場域之智慧文字客服;An Intelligent Chatbot with the Customized Task-based Sequence to Sequence Model

NCUIR > School of Management at National Central University > Executive Master of Information Management > Electronic Thesis & Dissertation > Item 987654321/86609

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/86609

Title:	客製化任務導向於電信場域之智慧文字客服;An Intelligent Chatbot with the Customized Task-based Sequence to Sequence Model
Authors:	章鴻琳;Chang, Hung-Ling
Contributors:	資訊管理學系在職專班
Keywords:	對話機器人;詞嵌入;遞歸神經網路模型;注意力序列對序列;客製化對話;dialogue robot;word embedding;GRU;attention sequence to sequence;customized dialogue
Date:	2021-08-04
Issue Date:	2021-12-07 13:01:22 (UTC+8)
Publisher:	國立中央大學
Abstract:	自然語言處理的技術 (Natural Language Processing , NLP) 在人工智慧應用上有很重要的地位，應用範圍也非常的廣泛包括語言的互相翻譯、文章給予適當的標籤、利用書寫方式辨識作者、對話機器人。一個有智慧的可以協助回答客戶問題的機器人，不但可以協助降低企業的成本，也可以解決7 X 24 小時不間斷的服務品質，但現今以條件式建構的知識庫應答機器人，僅能滿足六成的客戶問題，如何在提高應答率與準確率，也是本研究想要提出的一種模型。本研究以電信業為例，由於電信業的客服必須具備多種知識領域包括網路問題，手機問題，促案問題，帳單問題，催收問題，林林總總，本實驗利用兩階段對話學習框架，並在第二階段加入問題任務條件與客人屬性為回答內容，達成提升應答率與準確率。實驗的結果經過兩種文字生成評估方法驗證，加入問題分類與客戶屬性後訓練的模型是可以提升回應的準確率，藉此提高應答率。 ;Natural language processing technology (NLP) has a very important position in the application of artificial intelligence, and its application range is also very extensive, including mutual translation of languages, giving appropriate tags to articles, using writing methods to identify authors, and dialogue robots. A smart robot that can assist in answering customer questions cannot only help reduce the cost of the enterprise, but also solve the 7 X 24 hours uninterrupted service quality. However, the knowledge base answering robot constructed with conditional methods can only meet 60% of the requirements. How to improve the response rate and accuracy of the customer’s questions is also a model that this research wants to put forward. This study takes the telecommunications industry as an example. Since the customer service of the telecommunications industry must have a variety of knowledge areas including network issues, mobile phone issues, contract issues, billing issues, collection issues, and there are numerous problems, this experiment uses a two-stage dialogue learning framework, and In the second stage, the question task conditions and guest attributes are added as the answer content to achieve an improved response rate and accuracy rate. The results of the experiment have been verified by two text generation evaluation methods. The model trained after adding question classification and customer attributes can improve the accuracy of the response, thereby increasing the response rate.
Appears in Collections:	[Executive Master of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	161	View/Open

社群 sharing

Loading...