應用資料視覺化與行為序列分析技術於影片學習行為之探討

Using Data visualization technique and sequential analysis for video learning analytics

王琪驊(Qi-Hua Wang)

指導教授 ： 謝尚賢

共同指導教授 ： 曾敬梅(Ching-Mei Tseng)

國立臺灣大學/工學院/土木工程學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201903514

摘要

大型開放式線上課程(Massive Open Online Courses, MOOCs)，或被稱為「磨課師」已是科技輔助學習的未來趨勢，具備了多人同時修課的特性，也讓學習者有更多自主學習的機會。在課程與教學上使用MOOCS有許多值得探究的問題，例如：高中離率(high dropout rate)與低完課率(low completion rate)。為了改善這些現象，「小規模限制性線上課程」(Small Private Open Courses, SPOCs)逐漸因應而生，這是一種利用MOOCs資源投入班級教學，經常與翻轉學習(flipped learning), 或混成學習(blended learning)做結合的教學模式。其中，翻轉學習模式下，透過適當的學習策略以及學生自主調節的能力，相較於傳統課堂教學的模式，會有更好的學習表現，因此，教學者如何輔導學生使用好的學習策略就格外重要。在MOOCs所提供的數據作為學生學習狀況的探討已行之多年，然而這些研究成果大多只呈現了以課程為導向的趨勢與統計，並不符合以了解個人學習情況做為輔助的SPOCs與翻轉學習的教學模式。延續著研究團隊的縱貫性研究，本研究以修習台灣大學土木工程學系必修的工程圖學課程的84位學生，在107學年度課程上，以觀看課程平台上的影片進行學習活動的行為紀錄進行分析，分別從三個面向輔助教師了解個別學生的學習情況以及輔導方向，擅用MOOCs儲存的資料，進行分析、視覺化與預測，進而輔助教學。
本研究完整呈現處理資料的流程：首先針對影片觀看行為定義名詞框架，利用此框架處理、結構化與編碼行為。接著以資料視覺化將行為圖像化，可以清楚地觀察個別行為樣態;使用學習分析方法中的滯後序列分析(Lag Sequential Analysis, LSA)證實了影片類型的不同在總體學生會有相異的行為模式，以及頻繁樣式探勘(Frequent Sequence Mining, FSM)進一步了解觀看不同類型的影片以及學習表現不同的學生在連續序列樣態的差異，最後經由統計後的編碼資料訓練一成績預警模型，再進而了解被模型誤判的學生在行為模式的特性，可能是學習低落的行為模式。藉由這一完整的分析流程與結果，提供實作學習分析的參考案例以及班級教學輔導的建議，觀看行為框架可提供後續分析者瞭解資料分析處理過程的依據以及分析方向。

關鍵字

學習分析 ； 資料分析 ； 影片觀看行為 ； 資料視覺化

並列摘要

Massive Open Online Courses(MOOCs) remains a growing trend and has generated enormous interest in the field of education, giving rise to a significant number of research studies in all kinds of aspects. While MOOCs obsesses the capacity to accommodate multiple learners at the same time, it also leads to high dropout rates and low completion rate. Many educators and researchers have begun to focus on a new development in hybrid MOOCs, for example, the Small Private Open Courses (SPOCs), both in their technological and pedagogical aspects and in placing students’ learning and academic results at the forefront of research. The use of more sophisticated research designs, for example, video learning analytics, is also recommended, paying more considerable attention to causal factors that promote student learning. The current paper-based on learning analytics of students who were enrolled in Engineering Graphics and their online video watching behaviors on NTU Cool. This course is a flip learning, or blended learning model, where a MOOC has been integrated with a traditional classroom. In the flip learning mode, through the appropriate learning strategies and the students’ self- regulation ability, compared with the conventional classroom teaching mode, better learning performance would be expected. In such a teaching environment, to teach students how to use appropriate learning strategies is particularly essential. Data provided by MOOCs have been discussed to depict to improve students' learning status for many years. However, most of the results are presenting only overall behavioral trends and statistics. They are not in line with the teaching of SPOCs and flip learning to understand individual learning mode. Our present study proposes three research goals: visual learning behavior, analysis of behavioral patterns, and predictive performance, with an ultimate goal to predict students learning outcomes, facilitate students learning, and assist teachers in providing appropriate learning strategies to students,
This study presents a series of process for processing data. First, the framework of terms for the viewing behavior of videos is defined. Secondly, we use this framework to process, structure and encode the behavior data. And then using the data to visualize the behavior, the visualization can observe the individual behavior patterns. In the analysis step, we use two learning analytics method to find the behavior patterns. The Lag Sequential Analysis(LSA) has been used to confirms that the differences in video types in the overall students will have different behavior patterns. The Frequent Sequence Mining(FSM) has been used to explore the differences in the continuous sequence of students watching different types of videos and students with diverse learning performance. And finally, through statistical coding for each person, data have been trained to be an early warning model, and then understand the model misjudgment may be learning frustration students. We would expect that our analysis process and results can be used as reference cases study for practical learning analysis. Timely predict students learning consequences and provide suggestions for class teaching guidance. The viewing behavior framework can provide follow-up analysts with an understanding of the data analysis process and the direction of study.

並列關鍵字

learning analytic ； data analysis ； video viewing behavior ； data visualization

參考文獻

Abeysekera, L., & Dawson, P. (2015). Motivation and cognitive load in the flipped classroom: definition, rationale and a call for research. Higher Education Research & Development, 34(1), 1-14.

Aguiar, E., Nagrecha, S., & Chawla, N. V. (2015). Predicting online video engagement using clickstreams. Paper presented at the 2015 IEEE International Conference on Data Science and Advanced Analytics (DSAA).

Allison, P. D., & Liker, J. K. (1982). Analyzing sequential categorical data on dyadic interaction: A comment on Gottman.

Anderson, A., Huttenlocher, D., Kleinberg, J., & Leskovec, J. (2014). Engaging with massive online courses. Paper presented at the Proceedings of the 23rd international conference on World wide web.

Bakeman, R., & Gottman, J. M. (1997). Observing interaction: An introduction to sequential analysis: Cambridge university press.