- 學位論文
建構反應題與選擇題的評量效果比較－以國小五年級「體積」單元為例
The Comparison of the Evaluation Effect Between Multiple Choice and Constructed Response Tests –Using the “Volume” Unit of Fifth Grade Math as An Example
摘要
本研究主要以國小五年級「體積」單元為研究範圍，根據教學目標、子技能、錯誤類型等，以貝氏網路當作推論工具，建立一套以貝氏網路為基礎之國小數學科電腦診斷測驗，診斷出學生的錯誤類型，提供教師教學的參考。研究結果如下： (一)自動計分與分析錯誤類型的「體積」單元線上測驗系統 本研究開發一套以建構反應題為基礎可自動計分與分析錯誤類型的線上測驗系統，可透過系統記錄的結果得知學生的成績、作答反應及有哪些錯誤類型。 (二)電腦化建構反應題測驗系統的實施成效 本研究分析建構反應題的錯誤類型，此四題的平均辨識率為97.14％，因此本研究的自動化分析建構反應題模型擁有良好的辨識率。 (三)比較貝氏網路在傳統選擇題與加入建構反應題之成效評估 本研究結合貝氏網路、子技能及錯誤類型設計具有選擇題與建構反應題的試題，透過Matlab程式分析試題的辨識率，由研究結果顯示，加入建構反應題之貝氏網路之整體辨識率皆優於僅納入選擇題之辨識率。
並列摘要
The main scope of this research is about the unit “Volume” of the fifth grade in primary school. The theory of this research is to take the Bayesian networks as a reasoning tool. According to the teaching goals、sub skills and error types of the questions, Bayesian networks can be a reference tool applying in Math computer diagnostic tests in primary school. The error types of students would be diagnosed through this test and it can provide teachers with a reference tool in teaching. The results of the research are as the followings: 1. To build an automatic scoring and analyzing error types questions on-line computer diagnostic test. The goal of the research is to create an automatic scoring and analyzing error type questions on-line computer diagnostic test based on constructed response questions. We can realize the pupils’ test results、responses and error types through the test. 2. Effectiveness of computerized constructed response questions test. The research analyzes the error types of the constructed response questions. The average differential rate of the four questions is 97.14％. Therefore, the differential rate of the automated analysis of constructed response questions is excellent. 3. Evaluate the effectiveness of Bayesian network and compare with multiple-choices questions and add with the constructed response questions. The research combined with Bayesian network、sub skills and error types is to design a tool examining multiple-choices questions and constructed response questions. Through Matlab, it can analyze the differential rate. The outcome shows that Bayesian network works better in exams that with constructed response questions than only with choices.