- 學位論文
以知識地圖為基礎發展的遠距測驗選題策略
An Item-Selection Strategy of Internet Testing Based on Knowledge Map
摘要
網際網路與電腦科技的進步促進了網際網路遠距學習的發展，帶給學習者與教師截然不同的學習環境，讓師生可以在網際網路上無時空限制地進行各種合作學習互動。而在學習過程中與教學密不可分的測驗方法，也逐漸朝遠距測驗的模式進行，同時並結合電腦化適性測驗之技術，發展出遠距適性測驗的環境。 一個良好的測驗與診斷系統不但要能夠有效地評估學生的學習成效，更要能診斷學生的學習障礙與盲點，幫助學生突破學習障礙、改善學習成效。因此，關於遠距測驗之發展，近年來除了傳統的能力評估測驗模式外也逐漸發展出以迷思概念診斷為方向之測驗系統。但由於傳統電腦化適性測驗使用的試題反應理論基礎，係以能力值作為評估的標準，因此較不易適用於以概念診斷為主之測驗系統。 本研究提出了一個基於知識地圖的觀念，以概念階層為主的概念式選題策略，同時並使用隱含概念矩陣與雙向式累進標的概念向量來計算所選試題組合之內容效度，最後利用基因演算法來完成選題策略中試題組合最佳化的問題。若能透過與迷思概念診斷式測驗系統之整合，將可完成概念診斷式電腦化適性測驗之實現。 本論文中主要的評估指標是以資訊擷取中常用的Recall、Precision以及F-measure值來評估所選試題組合的內容效度，並透過英文、資料庫與網路工程三門課程之實際題庫資料作為實驗資料來源來驗證本研究中所提出之方法。透過實驗分析結果顯示此一方法的成效顯著，更加肯定本研究之價值。
並列摘要
The fast advancement of Internet and computer technology has promoted the evolution of Internet-distance learning in these several years. It provides learners and teachers an entirely different learning environment, which gives learners many kinds of cooperative learning on Internet timelessly and spacelessly. At the meanwhile, the testing methods, which are very important in the learning process have been turned into Internet testing gradually and become distance adaptive testing by using Computer Adaptive Testing (CAT) technology. A good testing system should not only evaluate the abilities of students, but also diagnose their misconceptions to help them overcome the learning obstacles and improve the learning results. Therefore, many scholars devoted themselves to the development of misconception diagnostic testing. By using Item Response Theory (IRT), traditional CAT can evaluate students’ abilities effectively, but it is hard to be applied to the misconception diagnostic testing system. In this thesis, an item-selection strategy of Internet testing, which is based on the concept hierarchy of knowledge map is being proposed. By integrating this item-selection strategy and misconception diagnostic testing system, we can realize a misconception diagnostic CAT. Moreover, the ideas of Embedded Concept Matrix and Bidirectional Cumulative Target Concept Vector are brought up to compute the content validity of selected item-combinations. At last, the items’ combinatorial optimization problem is solved via genetic algorithm. The evaluation criterion in this thesis adopts the values of precision, recall and F-measure, as often referred in information retrieval to verify the content validity of the selected item-combinations. The experiment uses real courses of three classes: English, Database and Network Engineering. With the result of retrieval performance evaluation, it can be concluded that the item-selection strategy has a high validity and is worth further studying.