架構在SCORM相容環境下的使用者學習歷程分析與探勘

Abstract

隨著資訊科技的日新月異與網路技術的蓬勃發展，傳統的教學模式已漸漸發展成為不受時間與空間限制的網路學習(e-Learning)。SCORM 2004 提供了Sequencing and Navigation(SN)定義課程的次序行為，並使用活動樹用以描述階層性的課程結構，其根據不同的學習情況，而提供不同的學習者有不同的學習導引順序。所以，1.如何根據學習者個人特質的差異及能力的不同提供其客製化的學習活動，2.如何替不同的學習者新增、表示、管理個人化的學習樹成為兩個重要的議題。然而，由老師手動替每一位學習者打造個人化的學習樹是一項不可能的任務。學習者的學習歷程資訊有助於教師了解學習者其之所以得高分或是低分的原因。因此，提出了包含有四階段的學習歷程分析方法: 1.使用者模組定義階段:我們根據教育理論的需求定義了使用者的個人檔； 2.使用者學習行為萃取階段:使用循序樣式探勘(Sequential pattern mining)技術萃取學習者最常學習行為，並將之轉換成位元向量，隨之使用分群法(Clustering)將成效好的學習者分成適當數目的群體； 3.決策樹(Decision Tree)建構階段:使用2/3的學習者為訓練資料，及剩下的1/3學習者為測試資料，利用其個人檔案及前一階段的分群標籤建立決策樹； 4.活動樹(Activity Tree)建構階段:將產生的每一個群組使用萃取的學習行為以建構與SCORM相容的個人化活動樹。最後，為了評估以上四階段歷程分析的方法，研究者亦做了一個實驗以驗證其可行性。

With vigorous development of the Internet, e-learning system has become more and more popular. Sharable Content Object Reference Model (SCORM)2004 provides the Sequencing and Navigation (SN) to define the course sequencing behavior, control the sequencing, select and deliver of course, and organize the content into a hierarchical structure, namely Activity Tree (AT). Therefore, how to provide customized courses according to individual learning characteristics and capabilities, and how to create, represent and maintain the activity tree with appropriate associated sequencing definition for different learners become two important issues. However, it is almost impossible to design personalized learning activity trees for each learner manually. The information of learning behavior, called learning portfolio, can help teachers understand the reason why a learner gets high or low grade. A Learning Portfolio Mining (LPM) Approach is proposed includes of four phases: 1. User Model Definition Phase: define the learner profile based upon pedagogical theory. 2. Learning Pattern Extraction Phase: apply sequential pattern mining technique to extract the maximal frequent learning patterns from the learning sequence, transform original learning sequence into a bit vector, and then use distance based clustering approach to group learners with good learning performance into several clusters. 3. Decision Tree Construction Phase: use two third of the learner profiles with corresponding cluster labels as training data to create a decision tree, and the remaining are the testing data. 4. Activity Tree Generation Phase: use each created cluster including several learning patterns as sequencing rules to generate personalized activity tree with associated sequencing rules of SN. Finally, for evaluating our proposed approach of learning portfolio analysis, several experiments have been done and the research shows that generated personalized activity trees with sequencing rules are feasible for those learners.