臺灣大學: 資訊管理學研究所孫雅麗; 陳孟彰張筱珮Chang, Hsiao-PeiHsiao-PeiChang2013-03-222018-06-292013-03-222018-06-292011http://ntur.lib.ntu.edu.tw//handle/246246/251221過去自動出題系統相關研究的出題方式是窮盡所有可能考題的產生方法，僅有少數研究會針對題目的困難程度進行出題的考量，甚至鮮少考慮使用者程度以及文章本身閱讀困難度。因此本研究提出以文章的閱讀困難度與學習者的英語程度估計為基礎，為第二語言學習者設計了一個個人化的自動出題系統。在閱讀困難度估計方面，我們採用一些過去研究中較有意義及代表性的詞彙和語法特徵，然後再加入考量幾個不同特徵來分析一篇文章的可讀性，分別是語料庫中單字出現頻率、語言專家所制定的單字官方分及索引及從不同版本的高中教課書中整理出的文法模式 - 那些單字和文法的困難度代表第二語言學習者會學會該單字或文法的年級。在學生英語程度預測方面，我們從三個層面去估計一個學習者的能力，分別是字彙能力、文法能力，以及閱讀理解能力，然後再利用指數移動平均線去考慮他的歷史表現以確定學習者的英語程度水平。我們提供學生個人化的新聞閱讀與測驗活動，實驗結果顯示本研究所提供的閱讀困難度估計模型的結果優於其他的方法，且本研究所估計的文章閱讀困難度也接近專家的看法；並且學習者在使用我們的個人化閱讀和測驗系統後能顯著提升他們的英語程度且系統可準確預測出學習者的程度。A lot of research works have been done in the field of automatic quiz generation, however, almost all those studies generate all possible combination of quizzes and only few research consider difficulties of different quizzes. In this study, not only a quiz‘s difficulty but also the difference between learners and the reading difficulty of a document are taken into consideration. Therefore, we design a personalized automatic quiz generation system based on reading difficulty estimation scheme in a given document and proficiency level prediction for a second language learner. In the reading difficulty estimation scheme, we consult some meaningful lexical and grammatical features in early work, and then further consider several word frequency features from corpora, official grading indexes of vocabulary from language experts, and grammar patterns collected from textbooks — those which represent words and grammar patterns that the L2 learners have learned at various grade levels. In the proficiency level prediction, we estimate a learner‘s ability from three dimensions, which are vocabulary ability, grammar ability, and reading comprehension ability, and then further consider his historical performance to determine his proficiency level by weighted exponential moving average. A personalized news reading and testing experiment was conducted. The experimental results show that the proposed estimation outperforms the other estimations, and is close to the annotation of human experts. Moreover, It also shows that our system can increase learners‘ English proficiency, and provide a good prediction of learners‘ proficiency level.1196487 bytesapplication/pdfen-US可讀性閱讀困難度第二語言學習線性迴歸模型適性化測驗自動出題系統個人化指數移動平均線readabilityreading difficultysecond language learninglinear regression modelautomatic quiz generationpersonalizationexponential moving average(EMA)thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/251221/1/ntu-100-R98725037-1.pdf