摘要


近年來因為系所調降必修學分，提升了大量的選修學分供學生去作選擇及規劃個人知識地圖於學生各階段的學習過程中，選修課程不但能幫助學生實現擴展知識領域，也能減少學生被限制於單一領域。隨著大學課程的數量不斷增加，如何提升各類型不同領域及專長的學生修習所需要的課程及知識，成為學生在選擇選修課程時需要面對的問題。
以清華大學為例，2017年合併了新竹教育大學之後在課程上除了數量的增加，也增加了許多不同領域的課程。學生因為在有限的時間下無法去了解到所有課程的內容，因此在選擇上往往都是以同儕間或者學長姐的推薦作為選課考量。因此可能也失去了一些知識擴展及學習到自己所需要的知識。在經過訪談之後歸納出學生在尋找課程希望有一方法可以探索未知但是自己本身需要的知識。因此為了縮減學生選擇課程上面的時間成本，此研究提出應用文件探勘技術於課程大綱與學生課程推薦之分析。
本研究以服務設計（Service Design）的雙鑽石理論（Double-Diamond Theory）為基礎進行使用者研究，進行以人為本的服務設計。在第一階段以訪談（Interview）收集來自五個不同學院的十位使用者資訊；於第二階段以親和圖法（Affinity Diagram）分析大量質性訪談資訊，歸結出五大焦點，並且在考量現有資源、技術、可測試場域等條件後，選擇其中兩項作為設計目標以得出使用者需要一個決策輔助系統。我們在本研究中設計一個以最低的人為介入的推薦系統雛型，並進行測試。過去多數的研究在Educational Data Mining領域上研究課程推薦系統時往往都是單獨以學生修課資料(sequential data)為依據去做關聯分析來進行課程推薦，但這可能忽略了學生修習這門課程不一定是因為自己想修而是被其他因素所干擾；例如，課程人數已滿無法加簽所以選擇替代課程，因此我們在實驗中以兩種推薦方式來模擬學生在接受到原本同儕或學長姊的推薦課程及以接受我們的推薦之後的修課改變的幅度。測試結果，使用者在運用本研究所發展的系統，獲得了平均40%的改變量。在質化訪談中，我們了解到了推薦系統給予使用者的價值，以及未呼應到使用者某些需求的弱點。
關鍵詞：推薦系統、教育資料探勘、文字探勘、協同過濾、內容推薦、服務設計。

Abstract

As a result of compulsory credits being revised down for the past few years, a large number of elective credits was increased to provide students with various choices and to plan on their personal knowledge map during learning process. Elective courses not only help students to expand the field of knowledge, but also decrease the restriction of single field. As university courses constantly increases, students have to face with how to raise courses and knowledge in need for students with different professional specialty and fields when choosing courses.
Take NTHU as an example, after merging NHCUE, besides the amount of course increased, various fields were added to courses. Due to lack of the time, instead of understanding courses by themselves, students often take suggestions from peers or seniors as references. This could cause the limited expansion of knowledge one needs. After interviews with students in campus, one conclusion was made that students wish to find a method to explore unknown but needed knowledge. To shorten time costs on choosing courses, this research applies a recommender system using text mining with course syllabi to recommend course for students .
This study applies user-based service design based on Double-Diamond Theory. In the first phase, Interview was used to collect ten user information from five different colleges; in the second phase, a large number of qualitative interviews were analyzed with the Affinity Diagram, which came up with five major foci. After considering existing resources, technology, and testable fields, we selected two of them as design goals to arrive at the user's need for a decision support system. In this study, we aimed to develop a recommendation system with a minimal human intervention. In the past, most researches in the field of educational data mining have often recommended the course recommendation system based on the student's course data, but this might ignore that students who take this course are not necessarily because they wish to take but are interfered by other factors (the capacity of a courses is too full to be signed in, so that they choose alternative courses; thus, in this study, we simulated the ways of students taking suggestions with two recommendation methods. The recommended course of the original peer or the senior manager and the amount of change after accepting our recommendation method. At the end, we got 40% of the average amount of change and understand the value of our system via the qualitative interviews, and did not respond to the disadvantage of the user.

Keywords: recommend system, educational data mining, text mining, collaborative filter, content-based recommendation, service design

Table of content
Chapter 1 Introduction 1
1.1 Research Background 1
1.2 Research Motivation 1
1.3 Research Objectives 3
Chapter 2 Literature Review 4
2.3 Application domains 8
2.3.1 Course Recommendation 8
2.3.2 Recommender system 8
2.4.1 Chinese word segmentation 12
2.4.2 TFIDF 12
2.4.3 Box-Cox transformation 14
Chapter 3 Research Method 16
3.1 Research Framework 16
3.2 Design Method 16
3.2.1 Interview – Used in the Discover phase 16
3.2.2 Affinity Diagram – Used in the Define phase 17
3.2.3 Personas 17
3.3 Develop phase 19
Chapter 4 Discover and Define Problem 20
4.1 Interview 20
4.2 Data analysis – Affinity Diagram 21
4.3 Results– Personas 23
4.4 Objective of Solutions 25
Chapter 5 System Implementation and Experimental Design 26
5.1 Data source 26
5.2 System Implementation 27
5.2.1 Words Segmentation 28
5.2.2 TFIDF 29
5.2.3 Cox-Box Transformation 32
5.2.4 Content-based recommendation 32
5.2.5 Collaborative recommendation 33
5.2.6 System interface 34
5.3 Experiment Design and System Outputs 35
5.3.1 Pilot validation 35
5.3.2 Experiment Design 36
5.3.3 System Outputs 40
Chapter 6 Evaluation Results 43
6.1 Evaluation Results 43
6.2 Evaluation by Delta 44
6.3 Evaluation by questionnaire 46
6.4 Qualitative Results 57
6.5 Discussions 60
6.6 Limitaion and Future work 61
6.6.1 Limitaions 61
6.6.2 Future work 62
Chapter 7 Conclusion 63
References 64
Appendix A 67
Affinity Diagram 67
Appendix B 68
Original Notes 68
Appendix C 70
Sequence Model 70
Appendix D 71
Persona 71
Appendix E 73
Descriptive statistics result of the questionnaire 73

 
List of Figures

Figure 2 1 The overview of general course structure in NTHU 5
Figure 2 2 The structure of the elective course and the self-learning course of elective courses. 6
Figure 2 3 Guidelines for electives in advanced/multi-area coursework in the general education program 6
Figure 2 4 Association rule example 9
Figure 2 5 Item-based CF example 10
Figure 2 6 Maximum likelihood estimation to get the marginal cost λ 15
Figure 4 1 Discover and Define process 20
Figure 5 1 Overview of the proposed system 26
Figure 5 2 Example for the structure of a syllabus 29
Figure 5 3 Example of the original probability density function of the normal distribution of a student’s preferred list 32
Figure 5 4 Example of the processed probability density function of the normal distribution of a student’s preferred list 32
Figure 5 5 Process of content-based recommendation 33
Figure 5 6 Process of item-based collaborative filter recommendation 33
Figure 5 7 The login interface of our system 34
Figure 5 8 The recommendation interface of our system 34
Figure 5 9 Interface of 5.3.1Pilot validation 35
Figure 5 10 Service blueprint of the experiment 39
Figure 5 11 Login interface 40
Figure 5 12 First step interface 41
Figure 5 13 Second step interface 41
Figure 5 14 Third step interface 42
Figure 5 15 Thank page 42
Figure 6 1 The three types users in our experiment 44
Figure 6 2 The distribution of participants by college 44
Figure 6 3 The distribution of participants by year 44
Figure 6 4 Delta in three type users 45

List of Tables

Table 2 1 Item-based CF example 10
Table 4 1 The demographic data of interviewees 21
Table 4 2 Personas identified in this research 23
Table 5 1 Example of segmentation 28
Table 5 2 POS tags weighting in our research 30
Table 5 3 Example for a student’s prefer list 31
Table 5 4 Questionnaire of pilot validation 35
Table 5 5 Questionnaires of evaluation session. 37
Table 6 1 Delta_P in type A and type B 48
Table 6 2 T-test of the equality of means in type A and type B 48
Table 6 3 Delta_P in type A and type C 48
Table 6 4 T-test of the equality of means in type A and type C 48
Table 6 5 Delta_N in type A and type B 50
Table 6 6 T-test of the equality of means in type A and type B - Hypothesis 2 50
Table 6 7 Delta_N in type A and type C 50
Table 6 8 T-test of the equality of means in type A and type C - Hypothesis 2 50
Table 6 9 Transformation of 5 point scale 51
Table 6 10 Information Content Satisfaction and User Interface Satisfaction in type A and type B 52
Table 6 11 T-test of the equality of means in type A and type B - Hypothesis 3 52
Table 6 12 Information Content Satisfaction and User Interface Satisfaction in type A and type C 53
Table 6 13 T-test of the equality of means in type A and type C - Hypothesis 3 53
Table 6 14 Triggering Serendipity in type A and type B 55
Table 6 15 T-test of the equality of means in type A and type B - Hypothesis 4 55
Table 6 16 Triggering Serendipity in type A and type C 55
Table 6 17 T-test of the equality of means in type A and type C - Hypothesis 4 55
Table 6 18 The results of quantitative evaluation 56

Chang, P. C., Lin, C. H., & Chen, M. H. (2016). A hybrid course recommendation system by integrating collaborative filtering and artificial immune systems. Algorithms, 9(3), 47.
Bydžovská, H. (2013). Course Enrolment Recommender System (Doctoral dissertation, Masarykova univerzita, Fakulta informatiky).
Pu, P., Chen, L., & Hu, R. (2011, October). A user-centric evaluation framework for recommender systems. In Proceedings of the fifth ACM conference on Recommender systems (pp. 157-164). ACM.
Bilgic, M., & Mooney, R. J. (2005, January). Explaining recommendations: Satisfaction vs. promotion. In Beyond Personalization Workshop, IUI (Vol. 5, p. 153).
Acilar, A. M., & Arslan, A. (2009). A collaborative filtering method based on artificial immune network. Expert Systems with Applications, 36(4), 8324-8332.
GODOY, D., & AMANDI, A. ISISTAN Research Institute, UNICEN University Campus Universitario, Paraje Arroyo Seco Tandil (7000), Bs. As., Argentina. TelÊFax: Î54 (2293) 440362 Also CONICET.
Ziegler, C. N. (2005). Towards decentralized recommender systems (Doctoral dissertation, Verlag nicht ermittelbar).
McNee, S. M., Riedl, J., & Konstan, J. A. (2006, April). Being accurate is not enough: how accuracy metrics have hurt recommender systems. In CHI'06 extended abstracts on Human factors in computing systems (pp. 1097-1101). ACM.
Ziegler, C. N., McNee, S. M., Konstan, J. A., & Lausen, G. (2005, May). Improving recommendation lists through topic diversification. In Proceedings of the 14th international conference on World Wide Web (pp. 22-32). ACM.
Castro, F., Vellido, A., Nebot, A., & Mugica, F. (2007). Applying data mining techniques to e-learning problems. In Evolution of teaching and learning paradigms in intelligent environment(pp. 183-221). Springer, Berlin, Heidelberg.
Klašnja-Milićević, A., Ivanović, M., & Nanopoulos, A. (2015). Recommender systems in e-learning environments: a survey of the state-of-the-art and possible extensions. Artificial Intelligence Review, 44(4), 571-604.
Bilgic, M., & Mooney, R. J. (2005, January). Explaining recommendations: Satisfaction vs. promotion. In Beyond Personalization Workshop, IUI (Vol. 5, p. 153).
Huang, D., Tong, D., & Luo, Y. (2010). HMM revises low marginal probability by CRF for Chinese word segmentation. In CIPS-SIGHAN Joint Conference on Chinese Language Processing.
Pham, D. D., Tran, G. B., & Pham, S. B. (2009, October). A hybrid approach to vietnamese word segmentation using part of speech tags. In 2009 International Conference on Knowledge and Systems Engineering (pp. 154-161). IEEE.
Lin, Q. X., Chang, C. H., & Chen, C. L. (2010). 結合長詞優先與序列標記之中文斷詞研究 (A Simple and Effective Closed Test for Chinese Word Segmentation Based on Sequence Labeling)[In Chinese]. International Journal of Computational Linguistics & Chinese Language Processing, Volume 15, Number 3-4, September/December 2010, 15(3-4).
Chen, X., Qiu, X., Zhu, C., Liu, P., & Huang, X. (2015). Long short-term memory neural networks for chinese word segmentation. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing (pp. 1197-1206).
Xu, J., & Sun, X. (2016). Dependency-based gated recursive neural network for chinese word segmentation. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) (Vol. 2, pp. 567-572).
Zhu, D., & Xiao, J. (2011, October). R-tfidf, a Variety of tf-idf Term Weighting Strategy in Document Categorization. In 2011 Seventh International Conference on Semantics, Knowledge and Grids (pp. 83-90). IEEE.
Paik, J. H. (2013, July). A novel TF-IDF weighting scheme for effective ranking. In Proceedings of the 36th international ACM SIGIR conference on Research and development in information retrieval (pp. 343-352). ACM.
Garcia, E. (2016). Local Term Weight Models from Power Transformations: Development of BM25IR: A Best Match Model based on Inverse Regression. arXiv preprint arXiv:1608.01573.
Peffers, K., Tuunanen, T., Rothenberger, M. A., & Chatterjee, S. (2007). A design science research methodology for information systems research. Journal of management information systems, 24(3), 45-77.
Lory, J. A., & Scharf, P. C. (2003). Yield goal versus delta yield for predicting fertilizer nitrogen need in corn. Agronomy Journal, 95(4), 994-999.
Wen-Shung Tai, D., Wu, H. J., & Li, P. H. (2008). Effective e-learning recommendation system based on self-organizing maps and association mining. the electronic library, 26(3), 329-344.
Pazzani, M. J., & Billsus, D. (2007). Content-based recommendation systems. In The adaptive web (pp. 325-341). Springer, Berlin, Heidelberg.
Council, D. (2005). The ‘double diamond’design process model. Design
Council.Kawakita, J. (1986). KJ method. Chuokoron-sha, Tokyo.
Fylan, F. (2005). Semi-structured interviewing. A handbook of research methods for clinical and health psychology, 5(2), 65-78.
Davis, F. D. (1985). A technology acceptance model for empirically testing new end-user information systems: Theory and results (Doctoral dissertation, Massachusetts Institute of Technology).
Liu, I. F., Chen, M. C., Sun, Y. S., Wible, D., & Kuo, C. H. (2010). Extending the TAM model to explore the factors that affect Intention to Use an Online Learning Community. Computers & education, 54(2), 600-610.
Fathema, N., Shannon, D., & Ross, M. (2015). Expanding the Technology Acceptance Model (TAM) to examine faculty use of Learning Management Systems (LMSs) in higher education institutions. Journal of Online Learning & Teaching, 11(2).
Park, S. Y. (2009). An analysis of the technology acceptance model in understanding university students' behavioral intention to use e-learning. Educational technology & society, 12(3), 150-162.
Chien C. L.(2018) Developing A Research Area Exploration System with Service Design Process and Text Mining Techniques.