近年來聊天機器人(Chat bot)的崛起，成為各領域廣泛運用的技術之一。其中，聊天機器人被應用在客服系統領域已經是不可或缺的技術。由於台灣的生育率下降，對於大專院校來說，招生已經成為重要課題之一。在大專院校的招生中，客服系統又為最重要的模型之一。
	由於，傳統客服系統不管在人力還是時間上，都需要極大的成本，因此本研究將聊天機器人的客服領域以及招生領域做結合，建構出一個以招生為領域的客服系統。本研究使用OpenAI所提出的GPT-2自然語言處理生成模型，建構招生領域的客服系統，並且使用BLEU生成評估模型，來評估招生客服系統所生成的文本與參考文本的相似度，來得知此模型所生成的文本好壞，以及使用User experience evaluation方式來尋找出不同篩選候選生成語句組合的表現，並且建構出最合適且優秀的系統。

In recent years, the rise of chat bots has become one of the technologies widely used in various fields. Among them, the application of chat bots in the field of customer service systems is already an indispensable technology. Due to the decline in Taiwan’s fertility rate, enrollment has become an important issue for colleges and universities. In the enrollment of colleges and universities, the customer service system is one of the most important models.
Since the traditional customer service system requires a huge cost in terms of manpower and time, this research combines the customer service field of chat robots with the enrollment field to construct a customer service system with enrollment as the field. This study uses the GPT-2 natural language processing generative model proposed by OpenAI to construct a customer service system in the enrollment field, and uses BLEU to generate an evaluation model to evaluate the similarity between the text generated by the enrollment customer service system and the reference text, so as to know this model The generated text is good or bad, and the User experience evaluation method is used to find the performance of different screening candidate generated sentence combinations, and construct the most suitable and excellent system.

目錄
第ㄧ章 緒論	1
1.1 研究背景	1
1.2 研究動機	2
1.3 研究目的	3
1.4 研究架構	3
第二章 文獻探討	6
2.1 聊天機器人 (CHAT BOT)	6
2.1.1 樣板式模型(Rule/Template Based Model)	9
2.1.2 檢索式模型(Information Retrieval Based Model)	10
2.1.3 生成式模型(Generative Based Model)	11
2.2 客戶服務系統(CUSTOMER SERVICE SYSTEM)	12
2.2.1 傳統客戶服務系統	12
2.2.2 智能客戶服務系統	15
2.3 人工智慧(ARTIFICIAL INTELLIGENCE)	19
2.4 自然語言處理(NATURAL LANGUAGE PROCESSING)	22
2.4.1 序列到序列(Sequence to Sequence)	23
2.4.2 注意力機制(Attention Mechanism)	24
2.4.3 自注意力機制(Transformer)	25
2.4.4 自然語言生成模型(GPT-2)	27
第三章 研究架構與方法	29
3.1 研究方法	29
3.2 研究架構	32
3.3 PRE-TRAINED LANGUAGE MODEL AND FINE-TUNING	33
3.3.1 語料庫來源	34
3.3.2 資料前處理	36
3.3.3 預訓練	41
3.3.4 微調	42
3.3.5 評估方式	44
3.4 RESPONSE RANKING MODEL	48
3.4.1 FAQ template	50
3.4.2 Maximum Mutual Information	51
3.4.3 評估方式	52
第四章 資料分析與結果	54
4.1 實驗環境	54
4.2 實驗參數	54
4.2.1 Pre-trained Language Model and Fine-tuning	54
4.2.2 Response Ranking Model	57
4.3 實驗結果	59
4.3.1 Pre-trained Language Model and Fine-tuning	59
4.3.2 Response Ranking Model	62
4.3.3 Further Experiment on Pre-trained Language Model and Fine-tuning	68
第五章 結論與意涵	72
5.1 結論	72
5.2 研究貢獻	73
5.3 管理實務意涵	74
參考文獻	75
附錄一 系統情境圖	82
一、RRM1使用者評分高的系統示意圖	82
二、RRM1使用者評分普通的系統示意圖	83
三、RRM1使用者評分低的系統示意圖	84
 
圖目錄
圖 1本研究之研究架構與流程	5
圖 2 RETRIEVING MODEL VS. GENERATIVE MODEL	9
圖 3 IF-THEN的規則集	9
圖 4文字資訊檢索的流程	11
圖 5AMAZON的FAQ網頁	13
圖 6 STRUCTURE OF THE AUTOMATIC CUSTOMER SERVICE SYSTEM	17
圖 7人工智能，深度學習和自然語言處理關係圖	22
圖 8序列到序列模型示意圖	24
圖 9注意力機制下的序列到序列模型	25
圖 10 TRANSFORMER 模型架構	26
圖 11 DECODER BLOCK處理文字示意圖	28
圖 12系統發展研究法生命週期循環圖	30
圖 13系統發展研究法系統開發流程圖	31
圖 14 GPT-2生成模型招生領域客服機器人系統架構	32
圖 15招生領域GPT-2生成模型的系統架構	33
圖 16百科類問答資料集範例	35
圖 17淡江大學招生資料集範例	36
圖 18資料前處理架構圖	37
圖 19語系轉換後範例	38
圖 20百科類問答資料集資料前處理後範例	40
圖 21淡江大學招生資料集資料前處理後範例	40
圖 22微調架構圖	42
圖 23 GPT-2 EXTRA LARGE、BERT、GPT2-ML BLOCK對比圖	44
圖 24 RESPONSE RANKING MODEL架構	48
圖 25 微調模型LOSS值	56
圖 26 MAXIMUM MUTUAL INFORMATION LOSS	58
圖 27第一面向(FLUENCY)盒鬚圖	65
圖 28第二面向(CORRELATION)盒鬚圖	66
圖 29第三面向(PRACTICALITY)盒鬚圖	68
圖 30無重複情況圈數盒鬚圖	71
 
表目錄
表 1 ELIZA和人之間的對話	8
表 2 THE DEFINITION OF FOUR CATEGORIES OF ARTIFICIAL INTELLIGENCE	20
表 3百科類問答資料集資料分析	35
表 4淡江大學招生資料分析	36
表 5本研究關鍵字擴展範例	39
表 6關鍵字擴展後淡江招生資料集資料分析	39
表 7資料前處理後資料分析	41
表 8 BLEU 1-GRAM情況範例	47
表 9預訓練模型參數	55
表 10微調模型參數	57
表 11 MAXIMUM MUTUAL INFORMATION模型參數	59
表 12 10-FOLD微調模型比較	60
表 13 GPT2-CHINESE MODEL生成範例	61
表 14 系統評估問卷問題	63
表 15 問卷總分	64
表 16第一面向(FLUENCY)分數	64
表 17第二面向(CORRELATION)分數	66
表 18第三面向(PRACTICALITY)分數	67
表 19 GPT2 CHINESE MODEL生成回覆有無重複比較表	69
表 20無重複回覆GPT2-CHINESE MODEL生成範例	70
表 21有無重複情況下迴圈執行圈數分析	71

Abraham, A. (2005). Rule‐Based expert systems. Handbook of measuring system design.
Bahdanau, D., Cho, K., & Bengio, Y. (2014). Neural machine translation by jointly learning to align and translate. arXiv preprint arXiv:1409.0473.
Brennan, K. (2006). The managed teacher: Emotional labour, education, and technology. Educational Insights, 10(2), 55-65.
Cambria, E., & White, B. (2014). Jumping NLP curves: A review of natural language processing research. IEEE Computational intelligence magazine, 9(2), 48-57.
Cava, W. (2011). U.S. Patent No. 7,962,463. Washington, DC: U.S. Patent and Trademark Office.
Ceeper, E. (1996). Improving FAQfinder’s performance: Setting parameters by genetic programming.
Cer, D., Yang, Y., Kong, S. Y., Hua, N., Limtiaco, N., John, R. S., ... & Kurzweil, R. (2018). Universal sentence encoder. arXiv preprint arXiv:1803.11175.
Chen, C., Zhang, X., Ju, S., Fu, C., Tang, C., Zhou, J., & Li, X. (2019, August). AntProphet: an Intention Mining System behind Alipay's Intelligent Customer Service Bot. In IJCAI (pp. 6497-6499).
Chowdhury, G. G. (2003). Natural language processing. Annual review of
Cimiano, P., Unger, C., & McCrae, J. (2014). Ontology-based interpretation of natural language. Synthesis Lectures on Human Language Technologies, 7(2), 1-178.
Colby, K. M., Weber, S., & Hilf, F. D. (1971). Artificial paranoia. Artificial Intelligence, 2(1), 1-25.
Cook, D. P., Goh, C. H., & Chung, C. H. (1999). Service typologies: a state of the art survey. Production and Operations management, 8(3), 318-338.
Crutzen, R., Peters, G. J. Y., Portugal, S. D., Fisser, E. M., & Grolleman, J. J. (2011). An artificially intelligent chat agent that answers adolescents' questions related to sex, drugs, and alcohol: an exploratory study. Journal of Adolescent Health, 48(5), 514-519.
Cui, L., Huang, S., Wei, F., Tan, C., Duan, C., & Zhou, M. (2017, July). Superagent: A customer service chatbot for e-commerce websites. In Proceedings of ACL 2017, System Demonstrations (pp. 97-102).
Daniel Cer, Yinfei Yang, Sheng-yi Kong, Nan Hua, Nicole Limtiaco, Rhomni St. John, Noah Constant, Mario Guajardo-Céspedes, Steve Yuan, Chris Tar, Yun-Hsuan Sung, Brian Strope, Ray Kurzweil. Universal Sentence Encoder . arXiv:1803.11175, 2018.
Fitzpatrick, K. K., Darcy, A., & Vierhile, M. (2017). Delivering cognitive behavior therapy to young adults with symptoms of depression and anxiety using a fully automated conversational agent (Woebot): a randomized controlled trial. JMIR mental health, 4(2), e19.
Forrester. 2018. The Six Factors that Separate Hype from Hope in Your Conversational AI Journey. Technical report, Forrester,
Friedman, B., Khan Jr, P. H., & Howe, D. C. (2000). Trust online. Communications of the ACM, 43(12), 34-40.
Hoch, R. (1994). Using IR techniques for text classification in document analysis. In SIGIR’94 (pp. 31-40). Springer, London.
https://www.forrester.com/report/The+Six+Factors+That+Separate+Hype+From+Hope+In+Your+Conversational+AI+Journey/-/E-RES143773
imcaspar. 2019. gpt2-ml https://github.com/imcaspar/gpt2-ml
information science and technology, 37(1), 51-89.
International Organization for Standardization (2009). Ergonomics of human-system interaction - Part 210: Human-centered design for interactive systems (formerly known as 13407). ISO F±DIS 9241-210:2009.
Jaakkola, T., & Haussler, D. (1999). Exploiting generative models in discriminative classifiers. In Advances in neural information processing systems (pp. 487-493).
Jurafsky, D., & Martin, J. H. (2014). Speech and language processing. Vol. 3.
Kaplan, A., & Haenlein, M. (2019). Siri, Siri, in my hand: Who’s the fairest in the land? On the interpretations, illustrations, and implications of artificial intelligence. Business Horizons, 62(1), 15-25.
Koole, G. (2004, September). Performance analysis and optimization in customer contact centers. In First International Conference on the Quantitative Evaluation of Systems, 2004. QEST 2004. Proceedings. (pp. 2-5). IEEE.
Kreer, J. G. (1957). "A question of terminology". IRE Transactions on Information Theory. 3 (3): 208.
Lee, R. S. (2020). Natural language processing. In Artificial Intelligence in Daily Life (pp. 157-192). Springer, Singapore.
Li, S., & Zhu, P. (2018). U.S. Patent Application No. 16/043,415.
Linsey Jepma,2019.Chatbots in 2019–5 trends https://chatbotsmagazine.com/chatbots-in-2019-5-trends-7c50d3050930
Luong, M. T., Pham, H., & Manning, C. D. (2015). Effective approaches to attention-based neural machine translation. arXiv preprint arXiv:1508.04025.
McCarthy, John. What Is Artificial Intelligence?. 2007-11-12.
Morizeyao. 2019. GPT2-Chinese https://github.com/Morizeyao/GPT2-Chinese
Morse, P. M. (7 September 1962). "Book review: Stochastic Service Systems by John Riordan". Science. 137 (3532): 742.
Nilsson, N. J. (1980). Principles of Artificial Inteligence. Morgan Kaufmann.
Osborne, R. J., & Wittrock, M. C. (1983). Learning science: A generative process. Science education, 67(4), 489-508.
Papineni, K., Roukos, S., Ward, T., & Zhu, W. J. (2002, July). BLEU: a method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics (pp. 311-318).
Pearce, J. (2011). The history of the call centre. Call Centre Helper. com http://www. callcentrehelper. com/the-history-of-the-call-centre-15085. htm. Retrieved October, 4, 2013.
Qiu, M., Li, F. L., Wang, S., Gao, X., Chen, Y., Zhao, W., ... & Chu, W. (2017, July). Alime chat: A sequence to sequence and rerank based chatbot engine. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) (pp. 498-503).
Quillian, M. R., & Minsky, M. (1968). Semantic information processing. M. Minsky (Ed.), 227-270.
Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., & Sutskever, I. (2019). Language models are unsupervised multitask learners. OpenAI Blog, 1(8), 9.
Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., & Sutskever, I. (2019). Language models are unsupervised multitask learners. OpenAI Blog, 1(8), 9.
Ritter, A., Cherry, C., & Dolan, W. B. (2011, July). Data-driven response generation in social media. In Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing (pp. 583-593).
Russell, S. J., & Norvig, P. (2016). Artificial intelligence: a modern approach: Malaysia; Pearson Education Limited.
Shannon, C. E. (1948). A mathematical theory of communication. The Bell system technical journal, 27(3), 379-423.
Song, S., Chen, H., & Shi, Z. (2017, December). Intension classification of user queries in intelligent customer service system. In 2017 International Conference on Asian Language Processing (IALP) (pp. 83-86). IEEE.
Sutskever, I., Vinyals, O., & Le, Q. V. (2014). Sequence to sequence learning with neural networks. In Advances in neural information processing systems (pp. 3104-3112).
thu-coai. 2020. CDial-GPT https://github.com/thu-coai/CDial-GPT
Tseng, J. C., & Hwang, G. J. (2007). Development of an automatic customer service system on the internet. Electronic commerce research and applications, 6(1), 19-28.
Tseng, J., Huang, Y. S., Cheng, L. C., & Hwang, G. J. (2004). Development of an Automatic Customer Service System on Computer Networks. PACIS 2004 Proceedings, 39.
TURING, I. B. A. (1950). Computing machinery and intelligence-AM Turing. Mind, 59(236), 433.
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., ... & Polosukhin, I. (2017). Attention is all you need. In Advances in neural information processing systems (pp. 5998-6008).
Vinyals, O., & Le, Q. (2015). A neural conversational model. arXiv preprint arXiv:1506.05869.
Wang, D., Li, T., Zhu, S., & Gong, Y. (2010). iHelp: An intelligent online helpdesk system. IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), 41(1), 173-182.
Wang, H., Lu, Z., Li, H., & Chen, E. (2013, October). A dataset for research on short-text conversations. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing (pp. 935-945).
Wang, W. M., & He, D. C. (2013). An intelligent customer service representative operating system. In Applied Mechanics and Materials (Vol. 373, pp. 1728-1734). Trans Tech Publications Ltd.
Wang, Y., Ke, P., Zheng, Y., Huang, K., Jiang, Y., Zhu, X., & Huang, M. (2020, October). A large-scale chinese short-text conversation dataset. In CCF International Conference on Natural Language Processing and Chinese Computing (pp. 91-103). Springer, Cham.
Weizenbaum, J. (1966). ELIZA—a computer program for the study of natural language communication between man and machine. Communications of the ACM, 9(1), 36-45.
World Population Review Total Fertility Rate 2020, https://worldpopulationreview.com/country-rankings/total-fertility-rate
Xu, A., Liu, Z., Guo, Y., Sinha, V., & Akkiraju, R. (2017, May). A new chatbot for customer service on social media. In Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems (pp. 3506-3510).
yangjianxin. 2019. GPT2-chitchat https://github.com/yangjianxin1/GPT2-chitchat
Young, S., Gašić, M., Keizer, S., Mairesse, F., Schatzmann, J., Thomson, B., & Yu, K. (2010). The hidden information state model: A practical framework for POMDP-based spoken dialogue management. Computer Speech & Language, 24(2), 150-174.
Zhang, H., Lan, Y., Guo, J., Xu, J., & Cheng, X. (2018, July). Reinforcing Coherence for Sequence to Sequence Model in Dialogue Generation. In IJCAI (pp. 4567-4573).
Zhang, Y., Sun, S., Galley, M., Chen, Y. C., Brockett, C., Gao, X., ... & Dolan, B. (2019). Dialogpt: Large-scale generative pre-training for conversational response generation. arXiv preprint arXiv:1911.00536.