校務行政語音助理與智慧音箱應用

本論文以深度學習實作自然語言對話的訊息聊天機器人與Google Voice Kit 智慧音箱校園應用實作。由於校園行政人員與學生互動多半耗費在訊息提問題，這不僅增加行政作業人員的工作量，且無法解決學生的問題。本論文研究的目的是希望解決學生問題提問的分類問題，引導學生到負責問題的窗口，達到迎賓聊天機器人語音服務的目的。論文中的語音及文字訊息處理包含語音轉文字、中文斷詞、詞向量、深度學習模型建置與文字轉語音；其中深度學習模型使用卷積神經網路及長短期記憶神經網路與Google’s Dialogflow實現，並比較三者的差異。本論文收集學生赴行政大樓各處室詢問的語料庫共計372筆，使用卷積神經網路與長短期記憶神經網路模型的訓練正確率均達到96%以上，而使用Google’s Dialogflow正確率僅60%左右。實驗結果發現，使用訓練模型優於Dialogflow 結果。初步探究原因是Dialogflow建立意圖與關鍵實體定義是一個重要的關鍵。

關鍵字

自然語言處理；深度學習；聊天機器人

並列摘要

In this thesis, the campus administrative service using deep learning to be implemented on Linebot and Google Voice kit. The purpose of system is automatic voice or text response so that students can ask questions and get answers via voice or text. In this Chotbot system, the natural language understanding (NLP) is by training and testing to compare with the differences of the three methods: convolutional neural network (CNN), long short-term memory neural network (LSTM) and Google’s Dialogflow. Simulation result showed that the CNN and the LSTM training got more than 96%, and Google's Dialogflow was only about 60% correct rate. It is known that the training model using deep learning is better than the Google’s Dialogflow using intents and entities.

並列關鍵字

Natural Language Processing ； Deep Learning ； Chatbot ； Google Voice kit

參考文獻

[1]The Stanford NLP Group. Available at: https://nlp.stanford.edu/

Google Scholar

[2]Language Technology Platform, 2018. Available at: https://www.ltp-cloud.com/

Google Scholar

[3]Chinese Knowledge and Information Processing (CKIP). Available at: http://ckip.iis.sinica.edu.tw:8080/

Google Scholar

[4]Natural Language Toolkit (NLTK). Available at: https://github.com/nltk/nltk

Google Scholar

[5]T. Mikolov, K. Chen, G. Corrado, and Jeffrey Dean, “Efficient estimation of word representations in vector space,” In: Proceedings of the International Conference on Learning Representations (ICLR), pp. 1301-3781, 2013.

Google Scholar

國際替代計量

校務行政語音助理與智慧音箱應用

全文下載

主題瀏覽