經由反覆練習自動問題生成與評分系統提升學生的學習成效;Automatic Question Generation and Grading for Repeated Testing to Improve Student Learning Performance

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/86466

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/86466

Title:	經由反覆練習自動問題生成與評分系統提升學生的學習成效;Automatic Question Generation and Grading for Repeated Testing to Improve Student Learning Performance
Authors:	蔡之禮;Tsai, Chin-Li
Contributors:	資訊工程學系
Keywords:	自動問題生成;自動評分;反覆練習;BERT;關鍵字萃取;學習成效;Automatic Question Generation;Automatic Grading;Repeated Testing;BERT;Keyword Extraction;Learning Performance
Date:	2021-07-13
Issue Date:	2021-12-07 12:52:07 (UTC+8)
Publisher:	國立中央大學
Abstract:	近年來，教育各項資源逐漸數位化，數位教育平台也逐漸普及，我們透過機器學習過程，利用AI準確地協助人類執行日常任務。在教育中，我們可以在許多情況下利用AI的優勢，例如預測學生的學習成效與揭示學生的學習策略。儘管如此，大多數解決方法尚未利用現代AI的功能，例如自然語言處理。在本研究中，我們使用反覆練習(repeated testing)簡答題的學習策略，來加深學生對於課程中重要知識的長期記憶。然而，頻繁的編寫試題和批改，往往需要耗費大量的時間與人力成本，因此，本研究使用當今最先進的機器學習技術，透過自然語言處理來幫助教師自動產生簡答題問題與自動評分，以減少教師出考卷與批改試題的時間。此外，我們針對簡答題的主要原因是許多研究文獻證明簡答題練習可以增強學生的長期記憶，進而提高學生的學習成績。我們在研究中提出了一個自動問題生成系統，該系統結合了基於語義和基於語法的問題生成方法，為了證明該系統具有高度可用性，且可以提高學生的學習成效，我們對127名學生進行實驗，透過圖林測試和混淆矩陣的方法來評估機器問題生成與機器自動評分的品質。結果說明，(1) 與控制組相比，實驗組學生可以透過反覆練習的學習策略，有效地提高學生的學習成績。(2)學生在識別機器生成或是專家編寫問題的過程中，實驗組與控制組都無法正確區分，這意味著機器問題生成的品質接近於專家編寫的品質。(3)雖然在機器自動評分中存在一些缺陷，但實驗結果表明，機器評分可以在一定程度上代替專家評分。;In recent years, various educational resources have gradually become digitized, and digital education platforms have gradually become popular. We use AI to accurately assist humans in performing daily tasks through machine learning processes. In education, we can use the advantages of AI in many situations, such as predicting students′ learning effects and revealing students′ learning strategies. Nevertheless, most solutions have yet to take advantage of modern AI capabilities, such as natural language processing. This study adopts the learning strategy of repeated testing of short answer questions to deepen students′ long-term memory of important knowledge in the course. However, frequently writing and correcting test questions often takes a lot of time and labor costs. Therefore, this research uses the most advanced machine learning technology to help teachers automatically generate short answer questions and automatic grading through natural language processing, thereby reducing the time for teachers to write and modify test papers. In addition, the main reason we focus on short answer questions is that many research prove that short answer questions can enhance students′ long-term memory, thereby improving students′ learning performance. In this study, we proposed an automatic question generation system that combining semantics based and syntax based question generation methods. In order to prove that the system is highly usable and can improve the learning performance of students, we experimented with 127 students to evaluate the quality of machine question generation and machine automatic grading through Turing test and confusion matrix methods. The results show that (1) compared with the control group, the experimental group students can significantly improve their learning performance through repeated testing of learning strategies. (2) In the process of students identifying machine generated question or expert writing question, the experimental group and the control group cannot distinguish correctly, which means the quality of machine generated question is close to the quality of expert writing question. (3) Although the automatic machine grading has some defects, but the experimental results show that machine grading can replace expert scoring to a certain extent.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	48	View/Open

社群 sharing

Loading...