近年來教育資源已逐漸數位化，有愈來越多的教師將課堂教材及作業藉由網路的方式讓學生能夠不受時空限制學習課堂知識，自動評分即是順應這股教學數位化的潮流而誕生的課題。與傳統的紙本作業人工批改有所不同，機器自動評分可以快速批改並且偵測到人工需要花費大量時間精力才能完成的作業項目，例如教材概念萃取或是抄襲偵測。只要設定好良好的評分標準，機器自動評分可以做為良好的評分依據。 有鑑於此，本研究透過基於語料庫(corpus-based)的文字探勘方法來實作自動評分系統。其中，有別於既有的基於語義(semantic-based)的評分方法，我們提出了基於概念(concept-based)的概念萃取評分方法，以萃取出課程教材中的關鍵概念來對學生作業進行評分。本研究實作了四種不同的自動評分方法，分別為基於語義方法中的潛在語義分析(Latent semantic analysis) 和顯示語義分析(Explicit semantic analysis) 以及基於概念方法中的TF-IDF方法和TextRank方法。 除了評分系統的實作外，我們也有透過比對學生作業和課程教材文本做抄襲檢測。自動評分框架由以下幾個步驟完成:學生作業文本透過文字前處理，並與課堂教材進行文本比對取得評分，最後再用機器的評分和人工評分的比對，以K-means和Spearman’s correlation來驗證評分準確度。本篇論文主要在探討在不同課程設計及文本類型之下，四種自動評分方法的評分效果。而在我們的實驗課程中，基於概念方法中的TextRank的評分效果是最好的。 ;In recent years, educational resources have gradually digitized. More and more teachers put materials and assignments on internet to enable students to learn classroom knowledge without being restricted by time and space. Automatic grading is the issue that is born in response to the trend of digital teaching. Differ from manual grading, automatic grading is faster and able to detect the problem that require a lot of time and effort to complete, such as textbook concept extraction or plagiarism detection. As long as a good grading standard is set, the machine automatic grading can be used as a good scoring basis. In view of this, this study implements an automatic grading system through corpus-based text mining method. Among them, unlike the existing semantic-based analysis and grading method, we propose a concept-based grading method to extract the key concepts in the course materials to grade student assignments. In this study, four different automatic scoring methods were implemented, namely the Latent semantic analysis (LSA) and Explicit semantic analysis (ESA) in semantic-based and the TF-IDF method and TextRank method in concept-based. In addition to the implementation of the scoring system, we also do plagiarism detection by comparing the texts of student assignments and course materials. The automatic scoring framework is completed in the following steps: the student′s homework text is pre-processed, and the text is compared with the course materials. Finally, the machine′s score and manual score are compared, and K-means and Spearman′s correlation are used to verify the accuracy of the score. This paper focuses on the grading effects of four automatic grading methods under different curriculum design and reports text category. In our experimental course, the TextRank grading method in the concept-based got the best result.
