跳到主要內容

臺灣博碩士論文加值系統

(44.213.60.33) 您好!臺灣時間:2024/07/20 05:40
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果 :::

詳目顯示

: 
twitterline
研究生:施冠成
研究生(外文):SHI, GUAN-CHENG
論文名稱:應用於機器人的情境式英文對話機器人
論文名稱(外文):Designing Situation-Based English Learning with a Dialogue Robot
指導教授:黃永廣黃永廣引用關係
指導教授(外文):WONG, WING-KWONG
口試委員:黃永廣熊博安謝文敏
口試委員(外文):WONG, WING-KWONGHSIUNG, PAO-ANNHSIEH, WEN-MIN
口試日期:2022-06-29
學位類別:碩士
校院名稱:國立雲林科技大學
系所名稱:電子工程系
學門:工程學門
學類:電資工程學類
論文種類:學術論文
論文出版年:2022
畢業學年度:110
語文別:中文
論文頁數:121
中文關鍵詞:自然語言處理機器人輔助語言學習意圖識別實體識別深度學習
外文關鍵詞:natural language processingrobot-assisted language learningintent recognitionentity recognitiondeep learning
相關次數:
  • 被引用被引用:1
  • 點閱點閱:275
  • 評分評分:
  • 下載下載:57
  • 收藏至我的研究室書目清單書目收藏:1
近年來,台灣英語水平在全國非英語母語國家逐年下滑,主要原因為大部份學
生很少主動學習英文,對學習英文產生焦慮感。本研究提出一個將自然語言處理應
用於機器人的情境式英文對話機器人系統,讓學生再練習英文對話的同時還可以
與虛擬和實體機器人互動,消除學生在與人進行英文對話的所帶來的緊張感、焦慮
感及不安達到提升孩童開口說英文的意願,這不僅能訓練孩童開口說英文,也可以
同時訓練閱讀及聽力,進而達到提升英文能力。系統架構分為 APP 端、自然語言
處理端、網頁顯示端、以及機器人端,透過圖文並茂的方式進行情境式英文對話。
然而,在過去語言機器人在語言模組精確度低且訓練上意圖與實體提取上錯誤過
多。本研究為解決以上問題,使用一套開源學習框架-Rasa,利用五種語言模型進
行預訓練,5 種模型如下: GPT2、BERT、DistilBERT、XLNet、RoBERTa,分別觀
察這五種模型的 Intent loss、Entity loss 以及正確率的表現狀況。經過實驗發現,意
圖與實體丟失過多的主要原因為在一個意圖中有標記的實體與無標記的實體相同
過多,這樣會導致語料模型裡的實體與有標記的實體會衝突到及語言模型實體無
法準確預測。經過實驗之後發現使用 DistilBERT 進行預訓練比其他四個模型精確
平均度還高,替整個文本意圖預測成功平均提升至 99.59%以上。

In recent years, Taiwan's English proficiency has been declining year by year in nonnative English-speaking countries. The main reason is that most students rarely actively
learn English and feel anxious about learning English. This research proposes a
situational English dialogue robot system that applies natural language processing to
robots, allowing students to practice English dialogue while interacting with virtual and
physical robots, eliminating the tension caused by students having English conversations
with humans It can not only train children to speak English, but also train reading and
listening at the same time, so as to improve their English ability. The system architecture
is divided into an APP side, a natural language processing side, a web page display side,
and a robot side, and contextual English conversations are conducted through pictures
and texts. However, in the past, language bots suffered from low accuracy of language
modules and too many errors in intent and entity extraction during training. In order to
solve the above problems, this study uses an open source learning framework-Rasa, and
uses five language models for pre-training. The five models are as follows: GPT2, BERT,
DistilBERT, XLNet, RoBERTa, and observe the Intent loss, Entity loss and the
performance of the correct rate. Through experiments, it is found that the main reason for
the excessive loss of intents and entities is that there are too many marked entities and
unmarked entities in one intent, which will cause the entities in the corpus model and the
marked entities to conflict with the language model entities. cannot be predicted
accurately. After experiments, it is found that using DistilBERT for pre-training is more
accurate than the other four models, and the success of the entire text intent prediction is
improved to more than 99.59% on average.
摘要 i
Abstract ii
誌謝 iii
目錄 v
表目錄 viii
圖目錄 ix
第一章 緒論 1
1.1. 研究背景 1
1.2. 研究動機與目的 2
第二章 文獻探討 3
第三章 系統與工具介紹 7
3.1. 應用程式端(APP) 7
3.1.1. Android Studio 7
3.1.2. Java 語言 7
3.1.3. 語音辨識 8
3.1.4. 硬體規格介紹 8
3.2. 自然語言處理端 9
3.2.1. Rasa背景介紹 9
3.2.2. Rasa目錄結構介紹 9
3.3. 伺服器後端 13
3.3.1. Python語言 13
3.3.2. Flask框架 13
3.3.3. MySQL語言 13
3.4. 網頁顯示端 15
3.4.1. HTML語言 15
3.4.2. JavaScript語言 15
3.4.3. jQuery庫 15
3.4.4. CSS語言 15
3.5. 機器人端 17
3.5.1. ROS框架 17
3.5.2. Yolo V3 17
3.5.3. OpenCV 17
3.5.4. DJI RobotMaster EP SDK 18
3.5.5. 硬體規格介紹 18
第四章 系統架構與研究方法 20
4.1. 系統框架 20
4.2. 應用程式端流程圖 21
4.3. 後端&RASA伺服器 22
4.4. 自然語言理解 23
4.4.1. Rasa NLU 23
4.4.2. 分詞器-WhitespaceTokenizer 24
4.4.3. 特徵化器 24
4.4.4. 稀疏特徵化器-CountVectorizerFeaturizers 24
4.4.5. 密集特徵化器-LanguageModelFeaturizers 25
4.4.6. BERT 27
4.4.7. GPT-2 30
4.4.8. XLNet 34
4.4.9. DistilBERT 39
4.4.10. RoBERTa 41
4.4.11. 意圖&實體分類器-DIETClassifiers 44
4.5. 對話管理 47
4.5.1. Rasa Core 47
4.5.2. Rasa Core策略訓練 49
4.6. 解決意圖與實體預測錯誤過高的問題以及實驗流程 51
4.7. 網頁顯示端 53
4.8. 設計聊天機器人實驗流程 58
4.9. 模型訓練資訊與電腦規格 59
第五章 研究結果與分析 60
5.1. 針對意圖與實體結果與分析 60
5.1.1. BERT (No NER)比較BERT (NER) 62
5.1.2. DistilBERT (No NER)比較DistilBERT (NER) 64
5.1.3. GPT2 (No NER)比較GPT2 (NER) 66
5.1.4. RoBERTa (No NER)比較RoBERTa (NER) 68
5.1.5. XLNet (No NER)比較XLNet (NER) 70
XLNet (NER) 70
5.1.6. 五種模型的實體與意圖F1-Score分析 72
5.1.7. 無模型與有模型之學習影響分析 74
5.2. 針對有無提升英文口說能力之分析 77
5.3. RASA弱點分析&自動化NER相似詞彙 82
5.4.保留無實體(NO NER)與刪除無實體(NER)詞彙模型訓練比較 86
5.5. 論文比較 89
第六章 結論與未來展望 91
6.1. 結論 91
6.2. 未來展望 91
參考文獻 93
附錄 97
附錄一、整體故事情節 97
附錄二、PARK故事情節 98
附錄三、SUPERMARKET故事情節 99
附錄四、RESTAURANT故事情節 100
附錄五、SCHOOL故事情節 101
附錄六、LIBARAY故事情節 102
附錄七、BERT(NO NER)與BERT(NER)混淆矩陣比較 103
附錄八、DISTILBERT(NO NER)與DISTILBERT (NER)混淆矩陣比較 104
附錄九、GPT2(NO NER)與GPT2 (NER)混淆矩陣比較 105
附錄十、ROBERTA(NO NER)與ROBERTA (NER)混淆矩陣比較 106
附錄十一、ROBERTA(NO NER)與ROBERTA (NER)混淆矩陣比較 107

[1]S. Punith, B. Chaitra, V. Kotagi and R. M. Chethana, “Research Paper on Chatbot for Student Admission Enquiry,” Journal of Advancement in Software Engineering and Testing”, vol. 3, no. 1, pp. 1-9, 2020.
[2]K. Deepika, V. Tilekya, J. Mamatha and T. Subetha, "Jollity Chatbot A contextual AI Assistant," 2020 Third International Conference on Smart Systems and Inventive Technology(ICSSIT),2020,pp.1196-1200,doi10.1109/ICSSIT48917.2020.9214076.
[3]S. Pawar, O. Rane, O. Wankhade and P. Mehta, “A Web Based College Enquiry Chatbot with Results,” International Journal of Innovative Research in Science, Engineering and Technology, vol. 7, no. 4, pp. 3874-3880, April 2018.
[4]H. Gawade, V. Patil, P. Vishe and S. Kolpe, “College Enquiry ChatBot System,” International Journal of Engineering Research & Technology, vol. 9, no. 9, pp. 741-742, September 2020, doi: http://dx.doi.org/10.17577/IJERTV9IS090396
[5]S. Meshram, N. Naik, M. VR, T. More and S. Kharche, "College Enquiry Chatbot using Rasa Framework," 2021 Asian Conference on Innovation in Technology (ASIANCON), 2021, pp. 1-8, doi: 10.1109/ASIANCON51346.2021.9544650.
[6]P. Srikanthan, R. Nizar, A. Ravikumar, K. Lalitharan, S. M. B. Harshanath and J. Alosius, "GLIB: Ameliorated English Skills Development with Artificial Intelligence," 2020 IEEE Bangalore Humanitarian Technology Conference (B-HTC), 2020, pp. 1-6, doi: 10.1109/B-HTC50970.2020.9297884.
[7]W. Astuti, D. P. I. Putri, A. P. Wibawa, Y. Salim, Purnawansyah and A. Ghosh, "Predicting Frequently Asked Questions (FAQs) on the COVID-19 Chatbot using the DIET Classifier," 2021 3rd East Indonesia Conference on Computer and Information Technology (EIConCIT), 2021, pp. 25-29, doi: 10.1109/EIConCIT50028.2021.9431913.
[8]H. Agus Santoso et al., "Dinus Intelligent Assistance (DINA) Chatbot for University Admission Services," 2018 International Seminar on Application for Technology of Information and Communication, 2018, pp. 417-423, doi: 10.1109/ISEMANTIC.2018.8549797.
[9]J. d. S. Oliveira, D. B. Espíndola, R. Barwaldt, L. M. Ribeiro and M. Pias, "IBM Watson Application as FAQ Assistant about Moodle," 2019 IEEE Frontiers in Education Conference (FIE), 2019, pp. 1-8, doi: 10.1109/FIE43999.2019.9028667.
[10]P. Yu and Y. Liu, "Roberta-based Encoder-decoder Model for Question Answering System," 2021 International Conference on Intelligent Computing, Automation and Applications (ICAA), 2021, pp. 344-349, doi: 10.1109/ICAA53760.2021.00070.
[11]F. Yu, J. Guo, W. Xi, Z. Yang, R. Jiang and C. Zhang, "Audio DistilBERT: A Distilled Audio BERT for Speech Representation Learning," 2021 International Joint Conference on Neural Networks (IJCNN), 2021, pp. 1-8, doi: 10.1109/IJCNN52387.2021.9533328.
[12]Geoffrey Hinton, Oriol Vinyals, Jeff Dean, "Distilling the Knowledge in a Neural Network", NIPS 2014 Deep Learning Workshop
[13]Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov, "RoBERTa: A Robustly Optimized BERT Pretraining Approach", 2019 Computation and Language
[14]Myle Ott, Sergey Edunov, David Grangier, Michael Auli, "Scaling Neural Machine Translation",arXiv:1806.00187,2018
[15]Tanja Bunk, Daksh Varshneya, Vladimir Vlasov, Alan Nichol, "DIET: Lightweight Language Understanding for Dialogue Systems",arXiv:2004.09936,2020
[16]D. Zheng, J. Wang and T. Zhao, "Japanese NER Post-Processing Based on Improved TBL Method," 2008 Fifth International Conference on Fuzzy Systems and Knowledge Discovery, 2008, pp. 161-165, doi: 10.1109/FSKD.2008.372.
[17]胡雅婷(2021)。輔助孩童學習英語之深度學習對話機器人。國立雲林科技大學電子工程系碩士論文,雲林縣。取自https://hdl.handle.net/11296/324esd
[18]Das, S., & Kumar, E. (2018, December). Determining accuracy of chatbot by applying algorithm design and defined process. In 2018 4th International Conference on Computing Communication and Automation (ICCCA) (pp. 1-6). IEEE.
[19]Jacob Devlin, Ming-Wei Chang, Kenton Lee, Kristina Toutanova. (2019), " BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding" , arXiv:1810.04805
[20]Zhilin Yang, Zihang Dai, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le, (2019)," XLNet: Generalized Autoregressive Pretraining for Language Understanding",arXiv:1906.08237
[21]Victor Sanh, Lysandre Debut, Julien Chaumond, Thomas Wolf, (2019), "DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter", arXiv:1910.01108
[22]Rasa NLU意圖和實體-分詞https://blog.csdn.net/csdn1e/article/details/117769141
[23]GPT2介紹
https://jalammar.github.io/illustrated-gpt2/
[24]XLNet介紹
https://medium.com/ai-academy-taiwan/2019-nlp最強模型-xlnet-ac728b400de3
[25]DistilBERT介紹
https://blog.csdn.net/fengzhou_/article/details/107211090
[26]Rasa Core介紹
https://www.twblogs.net/a/5d5f17c0bd9eee5327fde74c
[27]B. González, I. Ivagnes, J. Lejtreger, L. Chiruzzo and A. Rosá, "Application of Language Technologies to Assist English Teaching," 2021 40th International Conference of the Chilean Computer Science Society (SCCC), 2021, pp. 1-8, doi: 10.1109/SCCC54552.2021.9650395.
[28]A. F. Adoma, N. -M. Henry and W. Chen, "Comparative Analyses of Bert, Roberta, Distilbert, and Xlnet for Text-Based Emotion Recognition," 2020 17th International Computer Conference on Wavelet Active Media Technology and Information Processing(ICCWAMTIP),2020,pp.117121,doi:10.1109/ICCWAMTIP51612.2020.9317379.
[29]王衍 (2009)。影響學生英語能力之內部因素探討。臺中教育大學語文教育研界所助理教授論文,台中市。

QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top
無相關期刊