大規模開放式線上課程 (MOOCs) 在近年來的教育策略上扮演非常重要的角色。MOOCs的出現，讓來自世界各地的學生們能夠不受時間與空間的限制，依照自己的學習偏好學習。然而，不同學生具有不同的學習行為，也導致截然不同的學習成效，因此一直是許多科學家致力研究的課題。人工智慧的出現在如何處理大量的學習數據，也就是學習者在學習線上課程時所產生的活動紀錄，在近年來引起了許多關注。另一方面，由於線上過多複雜的學習資源，線上教育出現"資訊超載"的問題也引起眾多矚目。因此在線上教育中，能夠提供個人化學習策略指引的人工智慧推薦系統，被視為可以讓學生在遇到資訊超載困難時，給予合適資源的助教角色。然而，即時推薦是常見推薦系統非常難達成的一件事，因為目前這些推薦系統通常建立於課程網站中，因此如果有學生從未登入網站學習，因為缺乏推薦媒介，推薦系統對他們束手無策無法進行任何推薦。另一方面，並不是所有學生都會按照規劃好的學習步驟來學習，因此制式化的課程安排並不能滿足所有人的需求，也無法滿足不同教育背景的學生。再者，常見的推薦策略通常是為了鼓勵學習，卻忽略該如何幫助學生評估自己的學習狀況，了解哪些方面需要加強。本論文提出一個基於LSTM的磨課師適性化學習路徑推薦系統，希望滿足每個學生在學習路徑的個人偏好。學習路徑由知識地圖中的不同知識概念組成，可以了解學生的完整學習過程。透過不同影片觀看特徵、學生分群結果和學習路徑，本論文建立一個基於LSTM的學習路徑預測模型，可以透過分析學習特徵，推薦最適合每個學生下一個學習的學習路徑。在關於模型的評估上，本論文利用了 F1-score來計算準確率並得到了約0.8的高分，也表示此模型具有一定水準之上的準確程度。本論文將完整的推薦流程實作於由國立清華大學開發之磨課師平台 - 清華雲的兩門"Python程式語言入門"課程。由實驗結果得知，學習路徑推薦可以幫助學生提高學習意願並持續學習，因此進度落後的學生們透過推薦可以變成進度中等甚至進度快速的學生們。另一方面，約五成在推薦前尚未開始學習的學生們，在學習路徑推薦後開始學習，這也顯示這樣的推薦策略可以讓學生有更高的學習動機。透過自我調節學習問卷的回饋可以發現，本系統不但可以提醒學生持續學習、有更高的學習效率，還可以幫助他們規劃最適合每個人的個人化學習步調。再者，本系統提供除了考試之後的自我評估方式，讓學生可以在學習過程中透過推薦隨時評估自己的學習狀況，因此問卷結果顯示這樣的推薦可以幫助學生們複習自己不熟悉的觀念，也可以了解自己的學習進度排名，在落後時可以努力追上其他學生。

Massive Open Online Courses, also known as MOOCs for short, has a great impact on nowadays educational strategies. MOOCs enable global learners to learn without time and space constraints, allowing distinct learning characteristics when participating in online courses. Despite this, the problem of how different learning behaviors may affect user's learning performance remains a popular issue. Therefore, to deal with an increasing number of learning data, refer to learner's online-activities records, Artificial Intelligence (AI) attracts much more attention in recent years. On the other hand, overwhelmed by complicated learning resources, a problem named "information overload" was widely discussed in online education. AI-based Recommender System, which is recognized as the powerful solution to improve resource acquisition via customized supply, has been regarded as an assistant in online learning by giving personalized learning strategies.

However, in-time recommendation is something hard to achieve in common Recommender Systems since they are established with course websites, so researchers have no idea if students have never participated in any course activities. Additionally, not all students follow the learning schedule prepared by teachers, so the well-defined schedule fails to satisfy various personal needs and different knowledge backgrounds. On the other hand, recommendations are usually for encouraging learning, but lose attention to help learners self-evaluate what they haven't understood yet.

In this thesis, a Personalized Learning Path Recommender System with LINE Bot in MOOCs based on LSTM is proposed to meet personal preferences on path of learning. Learning paths, which are composed based on knowledge concepts of the knowledge map, are constructed to realize the real learning process of learners. A LSTM Learning Path Prediction Model is built to consider video-watching preference features, clusters of students and learning paths to recommend personal learning path suitable for each student. From the evaluation part, F1-score of the proposed LSTM Learning Prediction Model is equal to 0.8, indicating this model has a certain degree of accuracy.

The proposed system is used in two courses of NTHU Cloud, a MOOC Platform developed by National Tsing-Hwa University, to give personalized learning path guidance. The experimental results demonstrate that learning path recommendations will help students have stronger learning willingness to keep learning, and move from Slow-progress Group to become Medium- or Fast-progress ones. In addition, around half of students who haven't started learning are encouraged to learn after recommendations, showing learning motivation is higher due to recommendations. The feedback from the self-regulated learning questionnaire shows that this proposed system no matter remind learners to keep learning and achieve higher learning efficiency, but help plan proper study steps to fulfill their own learning needs. On the other hand, this system provides another way except examinations to make judgements about one's learning status, and most responses agree that this is helpful to review unfamiliar concepts and catch up with others.

Abstract . . . . . . . . . . . . . . i
中文摘要. . . . . . . . . . . . . . iii
Contents . . . . . . . . . . . . . . v
List of Figures . . . . . . . . . . . . . . ix
List of Tables . . . . . . . . . . . . . . xi
Chapter 1 Introduction . . . . . . . . . . . . . . 1
Chapter 2 Background and Related Works . . . . . . . . . . . . 6
2.1 Recurrent Neural Network . . . . . . . . . . . . . . . . . 6
2.1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.2 Recurrent Neural Model . . . . . . . . . . . . . . . . . 7
2.1.2.1 Learning Method . . . . . . . . . . . . . . . . . . . 8
2.1.2.2 Common Types of RNN Architectures . . . . . . . 9
2.1.3 Long Short-Term Memory Network . . . . . . . . . . . . . . 11
2.1.3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.3.2 Architecture of LSTM . . . . . . . . . . . . . . . . 13
2.2 Knowledge Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.1 Knowledge Map in MOOCs . . . . . . . . . . . . . . . . . . 16
2.2.2 Learning Knowledge Map . . . . . . . . . . . . . . . . . . . 17
2.3 Recommender System . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3.2 Recommender Systems in MOOCs . . . . . . . . . . . . . . 19
2.3.2.1 Common Recommender Systems in MOOCs . . . 19
2.3.2.2 AI-based Recommender Systems in MOOCs . . . . 21
2.3.2.3 Learning Path Recommendation Systems . . . . . 22
Chapter 3 System Architecture . . . . . . . . . . . . . . 25
3.1 NTHU Cloud MOOC Platform . . . . . . . . . . . . . . . . . . . . 27
3.1.1 Student Video Activity Collection . . . . . . . . . . . . . . . 27
3.2 AITutor System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3 Learning Path Recommender System . . . . . . . . . . . . . . . . . 30
3.3.1 Clustering Module . . . . . . . . . . . . . . . . . . . . . . . 30
3.3.2 Learning Path Preprocessing Module . . . . . . . . . . . . . 31
3.3.3 LSTM Learning Path Prediction Model . . . . . . . . . . . 32
3.3.4 Recommender Module . . . . . . . . . . . . . . . . . . . . . 33
3.3.5 User Feedback Processing Module . . . . . . . . . . . . . . . 34
Chapter 4 System Implementation . . . . . . . . . . . . . . 36
4.1 Clustering Module . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.1.1 Video Features Extraction Module . . . . . . . . . . . . . . 36
4.1.2 User Clustering Module . . . . . . . . . . . . . . . . . . . . 42
4.1.2.1 User Clustering Features . . . . . . . . . . . . . . 42
4.1.2.2 Normalization . . . . . . . . . . . . . . . . . . . . 43
4.1.2.3 K-means Clustering Model . . . . . . . . . . . . . 43
4.2 Learning Path Preprocessing Module . . . . . . . . . . . . . . . . . 45
4.2.1 User Learning Path Extraction Module . . . . . . . . . . . . 45
4.2.1.1 Definition of Learning Path . . . . . . . . . . . . . 45
4.2.1.2 Student Learning Path Analysis . . . . . . . . . . 46
4.2.2 LSTM Model Features Preprocessing Module . . . . . . . . 47
4.3 LSTM Learning Path Prediction Model . . . . . . . . . . . . . . . . 50
4.3.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.3.2 LSTM Training Data Construction Module . . . . . . . . . 51
4.3.3 Next Learning Node Prediction Module . . . . . . . . . . . 54
4.4 Recommender Module . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4.1 Recommended Materials Generation Module . . . . . . . . . 56
4.4.2 User Grouping Module . . . . . . . . . . . . . . . . . . . . . 58
4.4.3 Recommendation Module . . . . . . . . . . . . . . . . . . . 60
4.5 User Feedback Processing Module . . . . . . . . . . . . . . . . . . . 62
Chapter 5 Experiment and Result . . . . . . . . . . . . . . 63
5.1 LSTM Learning Path Prediction Model Evaluation . . . . . . . . . 65
5.2 Learning Path Recommender System Evaluation . . . . . . . . . . . 66
5.2.1 2021: Introduction to Python Programming . . . . . . . . . 67
5.2.2 2021 AP: Introduction to Python Programming . . . . . . . 71
5.3 Questionnaire Results and Analysis . . . . . . . . . . . . . . . . . . 75
5.4 Comparison and Analysis . . . . . . . . . . . . . . . . . . . . . . . 83
Chapter 6 Conclusion and Future Work . . . . . . . . . . . . . . 85
6.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
6.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Bibliography . . . . . . . . . . . . . . 89

[1] C. Olah. Understanding lstm networks. [Online]. Available: https: //colah.github.io/posts/2015-08-Understanding-LSTMs/
[2] P. A. Otero Cano and E. C. Pedraza Alarcón, “Recommendation systems in education: A review of recommendation mechanisms in e-learning environments,” Revista Ingenierías, vol. 20, no. 38, pp. 147–158, Jul. 2020. [Online]. Available: https://revistas.udem.edu.co/index.php/ingenierias/ article/view/2914
[3] M. Khalil, B. Taraghi, and M. Ebner, “Engaging learning analytics in MOOCS: the good, the bad, and the ugly,” CoRR, vol. abs/1606.03776, 2016. [Online]. Available: http://arxiv.org/abs/1606.03776
[4] J. Wang, H. Xie, F. L. Wang, L.-K. Lee, and O. T. S. Au, “Top-n personalized recommendation with graph neural networks in moocs,” Computers and Education: Artificial Intelligence, vol. 2, p. 100010, 2021. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S2666920X21000047
[5] R. Kim, L. Olfman, T. Ryan, and E. Eryilmaz, “Leveraging a personalized system to improve self-directed learning in online educational environments,” 89 Computers Education, vol. 70, pp. 150–160, 2014. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0360131513002285
[6] Y. Zhou, C. Huang, Q. Hu, J. Zhu, and Y. Tang, “Personalized learning full-path recommendation model based on lstm neural networks,” Information Sciences, vol. 444, pp. 135–152, 2018. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0020025518301397
[7] L. Zhong, Y. Wei, H. Yao, W. Deng, Z. Wang, and M. Tong, “Review of deep learning-based personalized learning recommendation,” in Proceedings of the 2020 11th International Conference on E-Education, E-Business, E-Management, and E-Learning, ser. IC4E 2020. New York, NY, USA: Association for Computing Machinery, 2020, p. 145–149. [Online]. Available: https://doi.org/10.1145/3377571.3377587
[8] N.-F. Huang, C.-C. Chen, J.-W. Tzeng, T.-T. Fang, and C.-A. Lee, “Concept assessment system integrated with a knowledge map using deep learning,” in 2018 Learning With MOOCS (LWMOOCS), 2018, pp. 113–116.
[9] A. O’Donnell, D. Dansereau, and R. Hall, “Knowledge maps as scaffolds for cognitive processing,” Educational Psychology Review, vol. 14, pp. 71–86, 03 2002. [10] C.-C. Chen, “Opportunities and challenges of moocs: Perspectives from asia.” 2013, iFLA World Library and Information Congress: 79th IFLA General Conference and Assembly. 17-23 August 2013, Singapore. Bookmark or cite this item: http://library.ifla.org/157/1/098-chen-en.pdf. 90
[11] A. H. Nabizadeh, J. P. Leal, H. N. Rafsanjani, and R. R. Shah, “Learning path personalization and recommendation methods: A survey of the state-ofthe-art,” Expert Systems with Applications, p. 113596, 2020.
[12] C.-L. Tang, J. Liao, H.-C. Wang, C.-Y. Sung, Y.-R. Cao, and W.-C. Lin, “Supporting online video learning with concept map-based recommendation of learning path,” in Extended Abstracts of the 2020 CHI Conference on Human Factors in Computing Systems, ser. CHI EA ’20. New York, NY, USA: Association for Computing Machinery, 2020, p. 1–8. [Online]. Available: https://doi.org/10.1145/3334480.3382943
[13] Z. Liu, L. Dong, and C. Wu, “Research on personalized recommendations for students’learning paths based on big data,” International Journal of Emerging Technologies in Learning (iJET), vol. 15, no. 8, pp. 40–56, 2020.
[14] L. Breslow, D. Pritchard, J. DeBoer, G. Stump, A. Ho, and D. Seaton, “Studying learning in the worldwide classroom: Research into edx’s first mooc,” Research in Practice and Assessment, 06 2013.
[15] R.-S. Shaw, “A study of learning performance of e-learning materials design with knowledge maps,” Computers Education, vol. 54, pp. 253–264, 01 2010.
[16] K.-K. Chu, C.-I. Lee, and R.-S. Tsai, “Ontology technology to assist learners’navigation in the concept map learning system,” Expert Systems with Applications, vol. 38, no. 9, pp. 11 293–11 299, 2011. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0957417411004027
[17] P. J. Werbos, “Generalization of backpropagation with application to a recurrent gas market model,” Neural networks, vol. 1, no. 4, pp. 339–356, 1988. 91
[18] C. L. Giles, G. M. Kuhn, and R. J. Williams, “Dynamic recurrent neural networks: Theory and applications,” IEEE Transactions on Neural Networks, vol. 5, no. 2, pp. 153–156, 1994.
[19] B. A. Pearlmutter, “Learning state space trajectories in recurrent neural networks,” Neural Computation, vol. 1, no. 2, pp. 263–269, 1989.
[20] IBM. Recurrent neural networks. [Online]. Available: https://www.ibm.com/ cloud/learn/recurrent-neural-networks
[21] J. L. Elman, “Finding structure in time,” Cognitive science, vol. 14, no. 2, pp. 179–211, 1990.
[22] M. I. Jordan, “Serial order: A parallel distributed processing approach,” in Advances in psychology. Elsevier, 1997, vol. 121, pp. 471–495.
[23] C. Nicholson. A beginner’s guide to lstms and recurrent neural networks. [Online]. Available: https://wiki.pathmind.com/lstm#recurrent
[24] P. J. Werbos, “Generalization of backpropagation with application to a recurrent gas market model,” Neural networks, vol. 1, no. 4, pp. 339–356, 1988.
[25] Y. Wang, A. Sun, J. Han, Y. Liu, and X. Zhu, “Sentiment analysis by capsules,” in Proceedings of the 2018 World Wide Web Conference, ser. WWW ’18. Republic and Canton of Geneva, CHE: International World Wide Web Conferences Steering Committee, 2018, p. 1165–1174. [Online]. Available: https://doi.org/10.1145/3178876.3186015 92
[26] D. Li and J. Qian, “Text sentiment analysis based on long short-term memory,” in 2016 First IEEE International Conference on Computer Communication and the Internet (ICCCI). IEEE, 2016, pp. 471–475.
[27] S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural computation, vol. 9, no. 8, pp. 1735–1780, 1997.
[28] S. Kim, E. Suh, and H. Hwang, “Building the knowledge map: an industrial case study,” Journal of knowledge management, 2003.
[29] D. A. Wiegmann, D. F. Dansereau, E. C. McCagg, K. L. Rewey, and U. Pitre, “Effects of knowledge map characteristics on information processing,” Contemporary Educational Psychology, vol. 17, no. 2, pp. 136–155, 1992. [Online]. Available: https://www.sciencedirect.com/science/ article/pii/0361476X92900554
[30] P. Chen, Y. Lu, V. W. Zheng, X. Chen, and B. Yang, “Knowedu: a system to construct knowledge graph for education,” Ieee Access, vol. 6, pp. 31 553– 31 563, 2018.
[31] S. Khan, “Khan academy,” Retrieved June 05, 2021, from the World Wide Web:https://www.khanacademy.org/, 2006.
[32] Y. Qin, H. Cao, and L. Xue, “Research and application of knowledge graph in teaching: Take the database course as an example,” in Journal of Physics: Conference Series, vol. 1607, no. 1. IOP Publishing, 2020, p. 012127. 93
[33] R.-S. Shaw, “A study of learning performance of e-learning materials design with knowledge maps,” Computers & Education, vol. 54, no. 1, pp. 253–264, 2010.
[34] P. Resnick and H. R. Varian, “Recommender systems,” Communications of the ACM, vol. 40, no. 3, pp. 56–58, 1997.
[35] J. Lu, D. Wu, M. Mao, W. Wang, and G. Zhang, “Recommender system application developments: a survey,” Decision Support Systems, vol. 74, pp. 12–32, 2015.
[36] F. Bousbahi and H. Chorfi, “Mooc-rec: a case based recommender system for moocs,” Procedia-Social and Behavioral Sciences, vol. 195, pp. 1813–1822, 2015.
[37] S. Dwivedi and V. S. K. Roshni, “Recommender system for big data in education,” in 2017 5th National Conference on E-Learning E-Learning Technologies (ELELTECH), 2017, pp. 1–4.
[38] N. Manouselis, H. Drachsler, K. Verbert, and E. Duval, Recommender systems for learning. Springer Science & Business Media, 2012.
[39] D. Onah and J. Sinclair, “Collaborative filtering recommendation system: a framework in massive open online courses,” INTED2015 Proceedings, pp. 1249–1257, 2015.
[40] A. Kardan, A. Narimani, and F. Ataiefard, “A hybrid approach for thread recommendation in mooc forums,” International Journal of Social, Behav94 ioral, Educational, Economic, Business and Industrial Engineering, vol. 11, no. 10, pp. 2195–2201, 2017.
[41] J. Wang, H. Xie, O. T. S. Au, D. Zou, and F. L. Wang, “Attention-based cnn for personalized course recommendations for mooc learners,” in 2020 International Symposium on Educational Technology (ISET), 2020, pp. 180– 184.
[42] H. Zhang, T. Huang, Z. Lv, S. Liu, and H. Yang, “Moocrc: A highly accurate resource recommendation model for use in mooc environments,” Mobile Networks and Applications, vol. 24, no. 1, pp. 34–46, 2019.
[43] O. R. Zaiane, “Building a recommender agent for e-learning systems,” in International Conference on Computers in Education, 2002. Proceedings. IEEE, 2002, pp. 55–59.
[44] W. Xu and Y. Zhou, “Course video recommendation with multimodal information in online learning platforms: A deep learning framework,” British Journal of Educational Technology, vol. 51, no. 5, pp. 1734–1747, 2020.
[45] J. Gong, S. Wang, J. Wang, W. Feng, H. Peng, J. Tang, and P. S. Yu, “Attentional graph convolutional networks for knowledge concept recommendation in moocs in a heterogeneous view,” in Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, 2020, pp. 79–88.
[46] H. Zhu, F. Tian, K. Wu, N. Shah, Y. Chen, Y. Ni, X. Zhang, K.-M. Chao, and Q. Zheng, “A multi-constraint learning path recommendation algorithm 95 based on knowledge map,” Knowledge-Based Systems, vol. 143, pp. 102–114, 2018.
[47] C. Krauss, A. Salzmann, and A. Merceron, “Branched learning paths for the recommendation of personalized sequences of course items.” in DeLFI Workshops, 2018.
[48] Y. Zhou, C. Huang, Q. Hu, J. Zhu, and Y. Tang, “Personalized learning fullpath recommendation model based on lstm neural networks,” Information Sciences, vol. 444, pp. 135–152, 2018.
[49] T. Saito and Y. Watanobe, “Learning path recommender system based on recurrent neural network,” in 2018 9th International Conference on Awareness Science and Technology (iCAST). IEEE, 2018, pp. 324–329.
[50] N. T. H. University, “Nthu cloud mooc platform,” Web:https://mooc.nthu. edu.tw/, 2019.
[51] G. van Rossum, “Mongodb,” Retrieved June 29, 2018, from the World Wide Web:www.mongodb.com, 2009.
[52] Google Developers, “Youtube iframe player api,” Retrieved June, 2021, from the World Wide Web:https://developers.google.com/youtube/v3/docs, 2015.
[53] L. Buitinck, G. Louppe, M. Blondel, F. Pedregosa, A. Mueller, O. Grisel, V. Niculae, P. Prettenhofer, A. Gramfort, J. Grobler, R. Layton, J. VanderPlas, A. Joly, B. Holt, and G. Varoquaux, “API design for machine learning software: experiences from the scikit-learn project,” in ECML PKDD Workshop: Languages for Data Mining and Machine Learning, 2013, pp. 108–122. 96
[54] P. J. Guo, J. Kim, and R. Rubin, “How video production affects student engagement: An empirical study of mooc videos,” in Proceedings of the first ACM conference on Learning@ scale conference, 2014, pp. 41–50.
[55] National Tsinghua University, “2020 ap: Introduction to python programming,” Retrieved July, 2021, from the World Wide Web:https://mooc.nthu. edu.tw/classroom/chapter/16, 2020.
[56] ——, “2021 1-4: Introduction to python programming,” Retrieved July, 2021, from the World Wide Web:https://mooc.nthu.edu.tw/classroom/chapter/47, 2020. [57] ——, “2021 ap: Introduction to python programming,” Retrieved July, 2021, from the World Wide Web:https://mooc.nthu.edu.tw/classroom/chapter/71, 2020. [58] B. J. Zimmerman and D. H. Schunk, Self-regulated learning and academic achievement: Theoretical perspectives. Routledge, 2001.
[59] B. Zimmerman, “Becoming learner: Self-regulated overview,” Theory into Practice, vol. 41, no. 2, pp. 64–70, 2002.
[60] R. S. Jansen, A. van Leeuwen, J. Janssen, R. Conijn, and L. Kester, “Supporting learners’ self-regulated learning in massive open online courses,” Computers & Education, vol. 146, p. 103771, 2020.
[61] D. C. van Alten, C. Phielix, J. Janssen, and L. Kester, “Effects of selfregulated learning prompts in a flipped history classroom,” Computers in human behavior, vol. 108, p. 106318, 2020.