- 學位論文
人工智慧方法於線上診斷測驗之應用～以國小五年級「表面積」為例
An Application of Artificial Intelligence Method in Computerized Diagnostic Test-Using The “Area of Surface” Unit of Grade 5 Math as an Example
摘要
本研究旨在研發電腦線上診斷測驗教材，使用於實際教學的實驗。研究範圍為國小五年級數學科領域「表面積」單元，以教育部頒訂之九年一貫課程能力指標「5-n-18」、「5-s-06」、「5-s-07」及「5-a-05」為參考依據，並與指導教授、專家等分析討論，整理出本單元之子技能及學生可能會產生的錯誤類型，並編製知識結構圖、貝氏網路圖來進行設計電腦適性測驗的數位教材，編製型式以選擇題及建構反應題為主。選定實驗對象後進行教學，教學完後一週給予實驗對象進行電腦線上診斷系統測驗，以測量教學及測驗之效果。 研究所得的結果分述如下 一、運用電腦線上診斷系統測驗，比起傳統紙筆測驗更能方便教師的閱卷效率。二、教師可以從電腦系統的數據中獲知學生的學習成效，在有效的時間內掌握學 生的迷思概念，給與學生最佳的補救教學，讓學生立即學習到正確的知識。 三、測驗結果選擇題子技能之平均辨識率為95.312％、錯誤類型之平均辨識率為 89.765％。；建構反應題型子技能之平均辨識率為96.138％、錯誤類型之平 均辨識率為90.909％。
並列摘要
The goal of this thesis is to develop online diagnostic testing software for use in realtime teaching. The area of application is fifth grade elementary school mathematics, for the units covering surface areas; in particular the competence indicators of the Nine Year Unified Curriculum, 5-n-18, 5-s-06, 5-s-07, and 5-a-05 were used for reference. After consultation with my advisor and peers, sub-skills for the unit, and possible student error types were also developed. The study then developed knowledge structure diagrams and Bayesian networks to design digital teaching materials for use in computerized adaptive testing. Testing formats developed include multiple choice and constructed response. Instruction began after test subjects were selected, and a week after instruction ended subjects were given an online diagnostic test to determine teaching and testing results. Results obtained were as follows: 1. Use of an online diagnostic testing system is more convenient and efficient for instructor grading than traditional pen and paper testing. 2. Instructors can determine student learning results from computer system data, grasp student misconceptions in an effective period of time, and provide students with optimal remedial instruction, allowing them to promptly acquire correct knowledge and methods. 3. Test results showed that the mean discrimination rate of sub-skills for multiple choice items was 95.312％, and for error types, 89.765％. The mean discrimination rate of sub-skills for constructed response items was 96.138％, and for error types, 90.909％.