具主題式文本摘要萃取之線上討論工具發展與應用研究

A Topic Modeling Scheme with Abstract Extraction to Facilitate Asynchronous Online Discussion Performance

陳冠霖(Chen Kuan-Lin)

指導教授 ： 陳志銘

國立政治大學/文學院/圖書資訊與檔案學研究所/碩士(2021年)

https://doi.org/10.6814/NCCU202101446

摘要

為了解決線上討論中學習者常常需要耗費大量時間對討論內容進行理解，以及討論內容經處理分析後常出現資訊過於抽象、解釋性不足，因而導致影響學習者討論學習成效的問題，本研究採用文本探勘技術中的LDA (Latent Dirichlet Allocation)主題分析模型及摘要抽取技術，發展具摘要萃取之主題分析即時回饋系統(Topic Analysis Instant Feedback System with Abstract Extraction, TAIFS-AE)，改善Chen, Li, Chang 與 Chen (2021)所提出的主題分析即時回饋系統(Topic Analysis Instant Feedback System, TAIFS)，以降低TAIFS 採用LDA主題分析模型，並以幾個關鍵字代表所分析主題，仍難以讓學習者清楚解讀主題意涵的問題，以幫助學習者能更精確掌握整體討論的概要，以及議題討論的面向。
本實驗採真實驗研究法，透過網路招募各大專院校學生共29人為研究對象，將其中14位學生隨機分派為使用TAIFS-AE(提供主題摘要列表)輔以線上討論的實驗組，另外15位學生則分派為使用TAIFS(提供主題關鍵字)的控制組，進行「新冠肺炎防疫應變」之社會性科學議題(Socio-Scientific Issues, SSI)線上討論。以探討兩組學習者在討論學習成效與科技接受度上是否具有顯著的差異，並且以先備知識作為背景變項，探討不同先備知識之學習者，在學習成效與科技接受度上是否具有顯著差異。此外，也透過滯後序列分析(Lag Sequential Analysis,LSA)探討實驗組學習者之有效行為模式。
研究結果發現，使用TAIFS-AE與使用TAIFS的學習者在討論學習成效上沒有顯著的差異，而兩組學習者在科技接受度上亦無顯著的差異，但是兩組學習者的科技接受度均高於中位數，顯示其科技接受度良好。本研究進一步透過行為歷程分析的結果發現，採用TAIFS-AE學習者在摘要句點擊次數與整體學習成效以及多元觀點之分數具有顯著正相關。此外，在使用TAIFS-AE輔助線上討論的組別中，點擊摘要列表功能次數較多的學習者在討論學習成效中的總分及多元觀點面向上顯著優於較少點擊摘要列表功能的學習者，代表若學習者能充分運用TAIFS-AE中的主題摘要列表功能來輔助討論活動，則TAIFS-AE將能有效促進學習者進行線上討論時的表現。
基於研究結果，本研究提出TAIFS-AE教學與系統改善建議以及未來能夠延伸的研究方向。整體而言，本研究將討論區學習、自然語言處理與資料視覺化等技術進行整合所發展之TAIFS-AE，提供科技輔助線上討論之創新有效學習工具，對於促進數位學習之線上討論具有貢獻。

關鍵字

線上討論 ； 隱含狄利克雷分布主題模型 ； BM25 ； 自動摘要 ； 社會性科學議題 ； 學習行為歷程 ； 科技接受度

並列摘要

In online discussions, learners usually need to spend a lot of time to understand the content of the discussion, resulting in low learning effectiveness. Although the previous research has developed a Topic Analysis Instant Feedback System (TAIFS) (Chen, Li, Chang & Chen, 2021) that uses several keywords to represent the topic of discussion to solve this problem, it is still difficult for learners to comprehend the discussion content. Therefore, this study uses the topic model and abstract extraction technology of LDA (Latent Dirichlet Allocation) to develop Topic Analysis Instant Feedback System with Abstract Extraction (TAIFS-AE), try to decrease the time that learners need to spend to understand the discussion content in online discussions and support learners to comprehend the aspects of the overall discussion easier.
This experiment adopts the true-experimental design and recruits 29 college students through the internet as research objects, 14 of them are randomly assigned to the experimental group using TAIFS-AE supplemented by online discussion, the other 15 students are assigned to the control group using TAIFS supplemented by online discussion to conduct a discussion on the topic of COVID-19, explore whether there are significant differences between the two groups of learning effectiveness and technological acceptance. Furthermore, use prior knowledge as a background variable to explore whether learners with different prior knowledge have significant differences in learning effectiveness and technological acceptance. In addition, this research uses Lag Sequential Analysis (LSA) to explore the behavior patterns of learners in the experimental group.
The results of the study found that there was no significant difference between the learners who used TAIFS-AE and the learners who used TAIFS of learning effectiveness and technological acceptance. However, the technological acceptances of the two groups are higher than the median grade of the questionnaire, indicating that they have positive attitude toward technological acceptance. Moreover, this study found the results of learners’ operation record that the number of clicks on summary list function by TAIFS-AE has a significant positive correlation with the learning effectiveness of overall score and scores of perspectives.
In addition, the group that uses the TAIFS-AE to assist online discussion, learners who clicked on the summary list function more often had the significantly better overall score and scores of perspectives in the discussion of learning effectiveness than those who clicked on the summary list function less. Which means that if learners can make full use of the topic summary list function in TAIFS-AE to assist the discussion activities, then TAIFS-AE will promote learners’ performance in online discussions.
Based on the results, this research puts forward suggestions for the improvement of TAIFS-AE, as well as research directions that can be extended in the future. This research integrates online discussion, natural language processing, and data visualization technology to develop TAIFS-AE, and provides innovative and effective learning tools that assist online discussion with technology and contributes to the promotion of online discussions in digital learning.

並列關鍵字

Online Discussion ； Latent Dirichlet Allocation ； BM25 ； Automatic Summarization ； Socio-Scientific Issues ； Learning Behavior ； Technology Acceptance Model

參考文獻

AbuSeileek, A. F. (2012). The effect of computer-assisted cooperative learning methods and group size on the EFL learners’ achievement in communication skills. Computers & Education, 58(1), 231-239.

Adetimirin, A. (2015). An empirical study of online discussion forums by library and information science postgraduate students using technology acceptance model 3. Journal of Information Technology Education: Research, 14(1), 257-269.

Anderson, J. R. (1996). ACT: A simple theory of complex cognition. American psychologist, 51(4), 355.

Atapattu, T., Falkner, K., & Tarmazdi, H. (2016). Topic-Wise Classification of MOOC Discussions: A Visual Analytics Approach. International Educational Data Mining Society.

Bates, A. T. (2005). Technology, e-learning and distance education. Routledge.