| 摘要: | 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，學生的學習活動歷程也得以數位化，在傳統教學現場，教師要了解學生的閱讀理解能力，往往透過小考或一些課程互動，而在今日的數位平台中，如何量測學生的閱讀理解能力，是學習分析領域中一項重要的議題。
隨著人工智慧快速發展，自然語言處理(NLP)領域在近年來得到顯著的突破，本文希望能利用當今最先進的自然語言處理技術，找出量測學生閱讀理解能力的最佳方法，此外，教師欲了解學生的閱讀理解能力，通常透過批改學生的小考來達成，然而教師在出題與批改上往往耗費大量的時間和人力成本，本文透過自然語言處理技術將這兩個步驟自動化，以利教師更快速地了解學生的閱讀理解能力。
在本文中，我們透過比對學生於電子書中畫的重點和教師畫的重點的一致性來判斷學生的閱讀理解能力，比較TextRank, RAKE, BERT三種方法的代理度量(Proxy measure)效能，透過語言生成模型GPT-2產生小考問答題，透過語言代表模型BERT自動批改學生答案，最後根據批改結果自動給予學生建議並將結果反饋給教師，以完成高度自動化的閱讀理解能力智慧化評量。
;In recent years, various educational resources have been gradually digitized, e-learning platforms have gradually become popular, and students’ learning activities have been digitized. At traditional teaching sites, teachers need to understand students’ reading comprehension, often interacting through quizzes or in-class activities. In today′s e-learning platforms, how to measure students′ reading comprehension is an important topic in the field of learning analytics.
With the rapid development of artificial intelligence, the field of natural language processing has made significant breakthroughs in recent years. This paper hopes to use state-of-the-art natural language processing technology to find the best way to measure students′ reading comprehension. In addition, teachers want to know students′ reading comprehension ability is usually achieved by marking students′ quizzes. However, teachers often spend a lot of time and labor on setting and marking exam papers. This paper uses natural language processing technology to automate these two steps to help teachers understand students′ reading comprehension more quickly.
In this paper, we measure the reading comprehension of students by comparing the consistency of the markers drawn by students in e-books and the markers drawn by teachers, then we compared the proxy measure performance of the three methods of TextRank, RAKE, and BERT. In quiz generation phase, we use GPT-2, a state-of-the-art language generation model, to generate quizzes by parsing materials. In the grading phase, we use BERT, a pre-trained language understanding model, to grade students’ answers automatically, and give them guiding according to grading results to complete a highly automated reading comprehension measurement framework. |
