臺灣博碩士論文加值系統

疾病是伴隨人們一生的問題，病患可能因為自身的醫療知識不足或工作繁忙沒辦法隨意挪出時間看診導致心態恐慌甚至延誤最佳治療期。為了改善這種現象本研究建置了疾病聊天機器人，提供顧問的角色能夠全天候幫助使用者解決疑惑、給予相關建議。本研究提出一個整合聊天機器人、自然語言處理、機器學習以及雲端運算的整合自然語言處理與機器學習於疾病聊天機器人基於雲端運算架構(Integrating Natural Language Processing and Machine Learning in Disease Chatbot based on Cloud Computing Framework, INMDCCCF)，來解決以上議題。以此架構配合醫療疾病資訊建立疾病與機器學習資訊平台(Disease and Machine Learning Information Platform, DMIP)來測試INMDCCCF之可行性。本研究將不同技術架設於聊天機器人中，並在聊天機器人與使用者的交談中分析使用者的身體情況，使得使用者能夠對自己的疾病進行初步了解以便進行後續的就醫動作。
首先聊天機器人的部分，採用Line bot做為所有技術載體與使用者進行溝通。其次自然語言處理的部分，使用Rasa NLU建立模型用於理解使用者的對話內容；機器學習部分，採用邏輯回歸(Logistic Regression)、決策樹(Decision Tree)、隨機森林(Random Forests)三種演算法進行症狀文章、病因文章、非疾病類文章的三元分類訓練並將訓練結果以模型方式儲存。在評估階段使用邏輯回歸、決策樹、隨機森林三種演算法進行效能以及準確度的比較。最後決定採用決策樹演算法建置系統並在本系統的實際使用案例中決策樹模型對於症狀、病因分類的F1-Measure達到0.671及0.648。至於雲端運算部分，資訊量龐大及繁雜的問題，時常導致單機運算上的效率低下，如此一來勢必會拖慢運算時間。為了增加效率，本研究建置一個Spark叢集架構來改善執行速度。

Disease is a part of life. Patients may delay treatment in the optimal periods because they don't have enough medical knowledge or time. Hence, how to apply the disease information efficiently is discussed in this paper. In order to improve aforesaid problems, the disease chatbot was created to assist users to clarify doubts and provide suggestions around the clock. In this paper, the Integrating Natural Language Processing and Machine Learning in Disease Chatbot based on Cloud Computing Framework (INMDCCCF) was proposed to solve aforesaid problems. Disease and Machine learning Information Platform (DMIP) was created, combined with the medical knowledge to verify the feasibility of INMDCCCF. The users are analyzed with the dialogue when communicating to the chatbot. Firstly, the DMIP adopts Line bot as the main communication software. Secondly, we employed Rasa NLU to train a model for understanding the dialogue of users. We also utilized the three algorithms of Logistic Regression, Decision Tree, and Random Forest for multi-class classification. According to the system evaluation, the F1-Measure of the Decision Tree algorithm for symptoms and etiology is 0.671 and 0.648 in the actual use cases of the system. It means the multi-class classification can classify the medical data. Finally, the problem of a large amount of data cause computing time-consuming, so we used the Spark cluster architecture to improve the efficiency of the system.

摘要.........................................................i
Abstract....................................................ii
誌謝........................................................iii
目錄.........................................................iv
表目錄.......................................................vi
圖目錄.......................................................vii
第一章 簡介...................................................1
1.1 研究背景..................................................1
1.2 研究動機..................................................2
1.3 研究目的..................................................2
第二章 研究技術探討............................................3
2.1 SPARQL...................................................3
2.2 DBpedia..................................................3
2.3 Rasa NLU.................................................4
2.3.1 MITIE................................................7
2.3.2 Rasa NLU與Dialogflow比較..............................7
2.4 聊天機器人................................................8
2.5 TF-IDF...................................................9
2.6 潛在語意索引..............................................10
2.7 餘弦相似性................................................10
2.8 結巴斷詞..................................................11
2.9 多元分類..................................................11
2.9.1 決策樹(Decision Tree).................................12
2.9.2 隨機森林(Random Forest)...............................12
2.9.3 邏輯回歸(Logistic Regression).........................12
2.10 機器學習評估.............................................13
2.11 Hadoop..................................................14
2.12 Hadoop分散式檔案系統.....................................15
2.13 Hadoop YARN.............................................16
2.14 Spark...................................................17
2.15 彈性分散式資料集.........................................18
2.16 Spark叢集管理器..........................................18
2.17 相關研究.................................................19
第三章 系統設計與開發步驟......................................21
3.1 系統一般性架構............................................21
3.1.1 整合自然語言處理聊天機器人基於雲端運算架構...............21
3.2 疾病與機器學習資訊平台架構.................................23
3.2.1 疾病與機器學習資訊平台架構圖............................23
3.2.2 開發環境..............................................25
3.3 系統建置與開發............................................25
3.3.1 資料蒐集..............................................26
3.3.2 資料前處理............................................27
3.3.3 LSI模型建立..........................................28
3.3.4 多元分類機器學習.......................................34
3.3.5 Rasa NLU訓練.........................................38
3.3.6 聊天機器人整合.........................................43
第四章 系統應用場景與展示.......................................47
4.1 應用場景..................................................47
4.2 系統特色..................................................47
4.3 Line bot 功能畫面.........................................48
第五章 效能測試與準確性評估.....................................50
5.1 雲端運算環境架構...........................................50
5.1.1 測試環境..............................................50
5.1.2 測試環境架構..........................................51
5.2 效能測試..................................................53
5.2.1 Standalone模式.......................................53
5.2.2 YARN模式.............................................54
5.2.3 Standalone與YARN比較.................................55
5.3 準確性評估...............................................56
5.3.1 邏輯回歸.............................................57
5.3.2 隨機森林.............................................58
5.3.3 決策樹...............................................59
5.3.4 系統評估.............................................60
第六章 結論..................................................62
參考文獻.....................................................63

[1]衛生福利部, “衛生福利部統計”, [Online]. Available: http://www.nhi.gov.tw/Resource/webdata/Attach_17538_2_%E9%86%AB%E7%99%82%E6%B5%AA%E8%B2%BB%E6%B0%91%E7%94%9F10p.pdf.
[2]衛生福利部, “106年全民健康保險醫療統計”, [Online]. Available: https://www.mohw.gov.tw/dl-50575-c76c27ab-715e-4ca5-af1c-008d848cdcb4.html.
[3]D. Reinsel, J. Gantz, and J. Rydning, “The Digitization of the World”, [Online]. Available: https://www.seagate.com/tw/zh/our-story/data-age-2025/.
[4]L. Roffia, et al., “Dynamic linked data: A SPARQL event processing architecture”, Future Internet, vol.10, no.4, 2018.
[5]Wikipedia, “DBpedia”, [Online]. Available: https://zh.wikipedia.org/wiki/DBpedia.
[6]F.L. Keppmann, M. Maleshkova, and A. Harth, “DLUBM: A Benchmark for Distributed Linked Data Knowledge Base Systems”, On the Move to Meaningful Internet Systems. OTM 2017 Conferences, pp. 427-444, 2017.
[7]E.R. Kramer, et al., “Tell Your Robot What to Do: Evaluation of Natural Language Models for Robot Command Processing”, RoboCup 2019: Robot World Cup XXIII, pp. 255-267, 2019.
[8]T. Desot, et al., “Towards a French Smart-Home Voice Command Corpus: Design and NLU Experiments”, Text, Speech, and Dialogue, pp. 509-517, 2018.
[9]“Rasa NLU: Language Understanding for Chatbots and AI assistants”, [Online]. Available: https://rasa.com/docs/rasa/nlu/about/.
[10]“MITIE github”, [Online]. Available: https://github.com/mit-nlp/MITIE.
[11]A.S. Lokman and M.A. Ameedeen, “Modern Chatbot Systems: A Technical Review”, Proceedings of the Future Technologies Conference (FTC) 2018, pp. 1012-1023, 2019.
[12]L. Yao, Z. Pengzhou, and Z. Chi, “Research on News Keyword Extraction Technology Based on TF-IDF and TextRank”, 2019 IEEE/ACIS 18th International Conference on Computer and Information Science (ICIS), pp. 452-455, 2019.
[13]gensim, “TF-IDF”, [Online]. Available: https://radimrehurek.com/gensim/models/tfidfmodel.html.
[14]M.I. Jambak, et al., “The impacts of singular value decomposition algorithm toward indonesian language text documents clustering”, 3rd International Conference of Reliable Information and Communication Technology, IRICT 2018, vol.843, pp. 173-183, 2019.
[15]S. Adinugroho, et al., “Optimizing K-means text document clustering using latent semantic indexing and pillar algorithm”, 2017 5th International Symposium on Computational and Business Intelligence (ISCBI), pp. 81-85, 2017.
[16]Rohit and A.K. Singh, “Accuracy enhancement of collaborative filtering recommender system for blogs using latent semantic indexing”, 2017 Conference on Information and Communication Technology (CICT), pp. 1-4, 2017.
[17]Wikipedia, “Jieba”, [Online]. Available: https://github.com/ldkrsi/jieba-zh_TW.
[18]Wikipedia, “Multiclass classification”, [Online]. Available: https://en.wikipedia.org/wiki/Multiclass_classification.
[19]J. Hossen, M.Z.J. H, and S. Sayeed, “Modifying Cleaning Method in Big Data Analytics Process using Random Forest Classifier”, 2018 7th International Conference on Computer and Communication Engineering (ICCCE), pp. 208-213, 2018.
[20]L. Breiman, “Random Forests”, Machine Learning, vol.45, no.1, pp.5-32, 2001.
[21]X. Jinguo and X. Chen, “Application of decision tree method in economic statistical data processing”, 2011 International Conference on E-Business and E-Government (ICEE), pp. 1-4, 2011.
[22]M. Jeihouni, A. Toomanian, and A. Mansourian, “Decision Tree-Based Data Mining and Rule Induction for Identifying High Quality Groundwater Zones to Water Supply Management: a Novel Hybrid Use of Data Mining and GIS”, Water Resources Management, vol.34, no.1, pp.139-154, 2020.
[23]Y. Yu, et al., “Wrapper feature selection based multiple logistic regression model for determinants analysis of residential electricity consumption”, 2017 Asian Conference on Energy, Power and Transportation Electrification (ACEPT), pp. 1-8, 2017.
[24]G. Wang and J. Xu, “A Decision Tree Algorithm Based on Coordination Degree and Gini-Coefficient”, 2008 International Conference on MultiMedia and Information Technology, pp. 822-824, 2008.
[25]W. Hwang, Y. Chen, and T. Jiang, “Personalized Internet Advertisement Recommendation Service Based on Keyword Similarity”, 2015 IEEE 39th Annual Computer Software and Applications Conference, vol.1, pp. 29-33, 2015.
[26]Wikipedia, “Confusion matrix”, [Online]. Available: https://en.wikipedia.org/wiki/Confusion_matrix.
[27]A. Berger and S. Guda, “Threshold optimization for F measure of macro-averaged precision and recall”, Pattern Recognition, vol.102, pp.107250, 2020.
[28]“Hadoop HomePage”, [Online]. Available: https://hadoop.apache.org/.
[29]A.K. Karun and K. Chitharanjan, “A review on hadoop — HDFS infrastructure extensions”, 2013 IEEE Conference on Information & Communication Technologies, pp. 132-137, 2013.
[30]“Apache Hadoop YARN”, [Online]. Available: https://hadoop.apache.org/docs/current/hadoop-yarn/hadoop-yarn-site/YARN.html.
[31]“Apache Spark”, [Online]. Available: https://spark.apache.org/.
[32]J. Wang, et al., “Link Prediction Based on Community Information and Its Parallelization”, IEEE Access, vol.7, pp.62633-62645, 2019.
[33]R.B. Mathew, et al., “Chatbot for disease prediction and treatment recommendation using machine learning”, 3rd International Conference on Trends in Electronics and Informatics, ICOEI 2019, April 23, 2019 - April 25, 2019, pp. 851-856, 2019.
[34]A. Piau, et al., “A smartphone Chatbot application to optimize monitoring of older patients with cancer”, Int J Med Inform, vol.128, pp.18-23, 2019.
[35]M. Fischer and M. Lam, “From Books to Bots”, Proceedings of the First Workshop on IoT-enabled Healthcare and Wellness Technologies and Systems - IoT of Health '16, pp. 23-28, 2016.
[36]H. Honda and M. Hagiwara, “Question Answering Systems With Deep Learning-Based Symbolic Processing”, IEEE Access, vol.7, pp.152368-152378, 2019.
[37]F.S. Al-Anzi and D. AbuZeina, “Enhanced Search for Arabic Language Using Latent Semantic Indexing (LSI)”, 2018 International Conference on Intelligent and Innovative Computing Applications (ICONIC), pp. 1-4, 2018.
[38]M. Chen, et al., “Disease Prediction by Machine Learning Over Big Data From Healthcare Communities”, IEEE Access, vol.5, pp.8869-8879, 2017.
[39]Y.S. Hindistan, et al., “Alternative Credit Scoring and Classification Employing Machine Learning Techniques on a Big Data Platform”, 2019 4th International Conference on Computer Science and Engineering (UBMK), pp. 1-4, 2019.
[40]A.V. Hazarika, G.J.S.R. Ram, and E. Jain, “Performance comparision of Hadoop and spark engine”, 2017 International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC), pp. 671-674, 2017.
[41]R. Caruana and A. Niculescu-Mizil, “An empirical comparison of supervised learning algorithms”, 23rd International Conference on Machine Learning, ICML 2006, June 25, 2006 - June 29, 2006, vol.148, pp. 161-168, 2006.
[42]“Beautiful Soup Documentation”, [Online]. Available: https://www.crummy.com/software/BeautifulSoup/bs4/doc/.
[43]R. Boisvert, R. Pozo, and K. Remington, “The Matrix Market Exchange Formats: Initial Design”, 1997.
[44]Y. huang, “以 jieba 探索文本主題 (歌詞文本)”, [Online]. Available: https://github.com/youngmihuang/lyrics_application/blob/master/lyrics_2.ipynb.
[45]M.A. Omari, et al., “Sentiment Classifier: Logistic Regression for Arabic Services’ Reviews in Lebanon”, 2019 International Conference on Computer and Information Sciences (ICCIS), pp. 1-5, 2019.
[46]S. Regmi, B.K. Bal, and M. Kultsova, “Analyzing facts and opinions in Nepali subjective texts”, 2017 8th International Conference on Information, Intelligence, Systems & Applications (IISA), pp. 1-4, 2017.
[47]I. Kareem and S.M. Awan, “Pakistani Media Fake News Classification using Machine Learning Classifiers”, 2019 International Conference on Innovative Computing (ICIC), pp. 1-6, 2019.