The Optimization Of A-Stratified Method Of Computerized Adaptive Testing Systen Based On Item Response Theory

Computerized Adaptive Testing has many advantages that classical testing in papers and pens can't compare: the ability to provide dynamic testing service individually based on different testing status of each tester. Compared with Classical Testing Theory, Item Response Theory has such advantages like the independence of test ,repeatable test and so on. However, the algorithim is too complex ,so it has difficulty in the appliance of reality. Item Response Theory based Computerized Adaptive Testing exert the power of both.Its goal is to build the best test for each of the tester.It can accord to a theory model used to describe the reaction of IRT, selection the project which match the estimated ability level of the testers.Through exclude and control the subjects which does not match the ability of the tests,it reduces test time,improve the measurement accuracy,and reduced a variety of measurement error of tests caused by carelessness, fatigue, depression, or speculation . Selection strategy is an important part of the CAT, selection strategies have a direct impact on test security, accuracy and efficiency. The selection strategy which CAT currently used is a-stratified method. But there are some problems in its practical application to made us to improve it urgently.There are some difficulties and challenges a-stratified method shown as below: the number of levels, set the amount of information for each layer, the distribution of item difficulty parameters should be subject to and set the termination conditions of the test based on a-stratified method,etc. Based on the comparison and analysis on the existed Item Response Theory models, estimating algorithms and strategies of Computerized Adaptive Testing with the consideration of the actual requirement of the project, one executable designing proposal for the Item Response Theory based Computerized Adaptive Testing system combining with Logistic 2PM model, Maxium Likelihood Estimating is raised in the paper.Improving a-stratified method and the most appropriate condition for the termination of the test based on a-stratified method .The propose is that including the terminal condition of the test should be that the total information of measured project or the number of measured project has already meet predetermined criteria, proportion of the amount of information of every level should be increased gradually, distribution of item difficulty parameters should subject to uniform distribution,etc. They optimize a- stratified method and they significantly improve the accuracy and stability of CAT. Finally, design and implement a prototype adaptive testing system for e-commerce division based on the optimized a-stratified method and terminal condition of the test.These results have been tested for its related theory in the project named'E-Commerce Professional Development Project'which was developed and Commitmented both by Shanghai Jiaotong University and Shanghai Mei Hua System Co., Ltd. commissioned by Shanghai Vocational Skill Identification Center. The contents of this paper are organized as follows: Firstly, we summarize a- stratified method and terminal condition of the test; analysis and compare advantages and disadvantages of many selection strategies and terminal conditions of the test. Secondly, through analyzing the traits of a- stratified method and terminal condition of the test as well as the challenges encountered, we summarize the factors which affect a- stratified method. And then we propose improved a- stratified method and terminal condition of the test based on the improved a- stratified method. Finally, we design and implement a prototype adaptive testing system for e-commerce division and verify the feasibility of this improved method.