The study of test sheets composing optimization using evaluation algorithm

• Main Authors: Reen-feng Liu, 劉任峯
• Other Authors: none
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/42910970028867885071

碩士 === 國立屏東教育大學 === 數位學習教學碩士學位學程 === 99 === The progress of the information network technique makes distance learning feasible. The traditional written test, therefore, can be replaced by online real-time test. Whether in traditional learning or distance learning, test is one of important activities in the teaching and learning process. Test can help us to realize the appropriateness of teaching materials, as well as to evaluate the learning outcome. In the past, the quality of test questions, however, usually relies on the competence of the teacher. Nowadays, the role can be hosted by a computer-assisted testing system. A good testing system can compose each characteristic parameter of questions comprehensively and output a satisfactory test sheet. In recent years, most of the research on the testing system do not cover the ability indicator. They usually focus on how to control the number of questions, difficulty, and testing time, etc. Of these, exposure control has rarely been reported. The algorithm for optimal solution usually selects some specific questions to compose a test sheet after iterative operation, which results in the phenomenon that some questions are over-exposure whereas some are under-exposure. Consequently, exposure control is worthy of investigation. In the thesis, we discuss some characteristic parameters (e.g., difficulty, index of discrimination, ability indicator, and exposure rate) in a given question bank. The goal of a high-quality test sheet is set to have the highest index of discrimination. The constraints are the range of average degree of difficulty, fixed number of questions, full ability indicator coverage, and exposure control from the satisfactory test-sheet users. A genetic algorithm is proposed to find a near-optimal solution of the test sheet. The average index of discrimination and average exposure rate are evaluated in the simulation. The results reveal that the two parameters have good performance by using the algorithm. As a result, the modified genetic algorithm with addition of the exposure control parameter is a promising method for generating the test sheet.