隨著2019 年Covid-19 新冠肺炎疫情爆發，居家隔離或者是遠距工作已然 成為了不少人的日常，傳統的教學方式也因此次疫情受到了衝擊，此時大規模開放線上課程(MOOC)更彰顯出遠距教學與數位學習的重要。近年來人工智慧相關技術的發展以及各式各樣新穎的數據分析方法的誕生，推薦系統以及成效預測已經成為了一個重要的研究方向。大規模開放式線上課程 (Massive Open Online Courses，MOOCs)是現今不斷擴展的數位學習方式，將課程透過網路發送給學習者學習，這種線上學習的行為自主性強且不受時間和地點限制，對於有富有學習動機的人來說是絕佳的學習資源。 本研究使用由日本京都大學開發的BookRoll 線上電子書學習系統搭配國 立中央大學所開發的複習與答題系統，根據學生們使用教材學習的行為紀錄 (Log)經過文本轉換後透過無監督學習方式進行學習策略的歸納，並探討學習行為與學習成效之間的關聯性。 本文探討學生的學習行為足跡以此來了解學生對學習成效較佳的活動或行 為，提供參考進而改善學生們的學習成效。我們透過Bookroll 平台與各複習系統上所收集的學習歷程進行處理，生成學生們的學習動作序列並歸納出學習策略。本研究希望能從中找出學習策略與學習成效的關聯性，提供老師輔導學生的參考。我們使用中央大學109 學年度下學期的Python 程式設計課程在混合式教學場景中的學習歷程與成果來分析學生的學習策略，發現學生在Bookroll 平台上所留下的學習活動資料確實可以透過無監督學習的方式使用分群演算法來萃取其學習策略。 研究發現，學生們使用BookRoll 及其他練習系統的足跡使用基於神經網 路的文本表示方法後透過無監督學習所歸納出的所有學習策略皆與學習成效都達到顯著正相關，且使用基於神經網路的文本表示法運算速度非常快速，未來可應用於學習預警或推薦機制實現精準的教學干預。;With the outbreak of the Covid-19 in 2019, home quarantine or remote work has become a daily life for many people. Traditional teaching methods have also been affected by the epidemic. At this time, Massive Open Online Courses (MOOCs) have highlights the importance of distance teaching and digital learning. With the development of artificial intelligence-related technologies and the birth of various novel data analysis methods, recommendation systems and performance prediction have become an important research direction. Massive Open Online Courses (MOOCs) are digital learning methods that have been expanding in recent years. Courses are sent to learners through the Internet. This kind of online learning is highly autonomous and independent of time and location, and this is an excellent learning resource for those who are motivated to learn. It′s a very good learning resource for people with learning motivation. This research uses the BookRoll online e-book learning system developed by Kyoto University in Japan and the review and answer system developed by National Center University to analyze learning strategies based on the action logs of students using the Bookroll, and study the effects between learning behavior and learning effectiveness. This study explores the learning behavior logs of students to understand the activities or behaviors of students with better learning effectiveness, and provides teachers as a reference for counseling, thereby improving students′ learning effectiveness. We process the learning process data collected on the Bookroll platform and each practice system including Cloze, Short-Ans, OJ, Viscode and Assessment, and analyze the learning logs of students, summarize the learning strategies of students. This research hopes to find out the correlation between learning actions and learning effectiveness, and provide teachers with references for tutoring students. We use the learning logs of the Python programming course in the second semester of the 109 academic year of National Central University in the mixed teaching scenario to analyze the learning strategy using learning logs left by the students on the Bookroll can be summarized by the clustering algorithm Its learning strategy. The study found that all learning strategies summarized by students using BookRoll and other practice systems logs can using unsupervised learning clustering are significantly related to learning effectiveness, and can be used in learning early warning or recommendation mechanisms to achieve precise teaching interventions in the future.
