傳統測驗大多限於紙筆測驗,不僅教師於施測前出題費時,施測後的批閱更勞心勞力,往往學生所犯的錯誤無法立即回饋,成效每下愈況,達不到測驗之目的,經年累月的結果,造成惡性循環,導致學生學習意願低落。藉由本研究所研發之線上診斷測驗,以電腦自動化迅速分析學生的錯誤類型,讓教師不再疲於奔命於批閱與補救教學。再者,如以純選擇題作為試題之結構,僅能以片面三個錯誤選項去推知預測,尚無法精確判斷學生的錯誤類型,甚至學生以猜題方式作答,所判斷的錯誤類型將無毫無意義,倘若輔以建構反應題,學生的解題歷程將完全記錄於電腦中,由電腦自動化分析判斷該生之錯誤類型,了解學生的解題策略,方便教師立即指正與回饋。 本研究結果概述如下: 一、利用貝氏網路將試題、錯誤類型、子技能結合,研究結果之建構反應題平均辨識率達到98.81%,成效極佳。 二、本研究將選擇題與建構反應題之貝氏網路結合之後,發現成效優於傳統純選擇題型之貝氏網路。 三、線上診斷測驗之結果顯示除了能迅速了解學生多元之錯誤類型,節省教師閱卷時間,更可於施測後馬上為學生做補救教學。
Conventional tests mainly take the form of paper tests. For teachers, constructing test items is quite a time consuming job, and marking those test papers after administering the test is labor intensive. Furthermore, there is no way for teachers to give their students immediate feedbacks regarding mistakes made in the test. This poor efficacy of paper tests hinders the purposes of tests. This situation worsens over time, forming a vicious cycle that reduces students’ learning intention. This study has developed an online diagnostic test that uses the computer to automatically and quickly analyze the error patterns of students. This feature saves teachers’ time on marking test papers and carrying out remedial instruction. Nevertheless, if a test comprises multiple-choice items only, the error pattern diagnosis will be built solely based on the three incorrect multiple-choice options. In that case, error patterns cannot be accurately identified. In addition, error pattern diagnosis can become meaningless if students adopt “guessing” as a strategy for answering the items. One way to solve the above-mentioned issues is to add constructed-response items to the test, record the question-solving procedure of students into the computer, and have the computer to automatically analyze students’ error patterns. By doing so, the teachers can gain insights into students’ question-solving strategies, correct their mistakes and give them feedbacks immediately. The study findings are summarized below: 1.Results from using Bayesian network to integrate test questions, error patterns and sub-skills suggest that the average recognition rate of constructed-response questions is nearly 98.88%, which is excellent. 2.Results from combining multiple-choice items with the Bayesian network of constructed-response items suggest that the outcome of this combination is superior to that of the Bayesian network of the conventional multiple-choice only type. 3.Results from conducting online diagnostic tests suggest that this developed system facilitates the identification of the pluralistic error patterns of students, saves test paper marking time, and provides students immediate remedial instruction after the test.