Research On Item Response Theory-Based System Of Computerized Adaptive Testing

In recent years, a newly developing psychometric and education test theory, named Item Response Theory, short for IRT, has attracted a lot of foreign and domestic researchers for its firmly theoretical basis and excellent generalization performance. The theory rapidly becomes one of the powerful theories for test. It develops based on criticizing on the limits of Classical Test Theory, short for CTT and with the development of statistical methods and the popular use of computer, IRT is rapidly used in many large-scale tests and computer-based testing (CBT). Almost all of the previous studies and applications are based on CTT-Item Bank.The paper firstly introduce the IRT and the Computer Adaptive Test present situation , analyzes the reason why IRT has not been in practice now, and discusses the basic IRT models and the different properties of IRT from CTT. It introduces the application of IRT in Computer Adaptive Test briefly and studies the different methods in main procedures of CAT . Some different methods are also analyzes and compares in this article.Secondly,we study the influence of the volume of the item bank on the test. We set three different volumes of item bank . Simulation method is used to find how many items are suited to the item pool for the system. The purpose of simulation for capacity is to find the minimum items suited to the adaptive algorithm.Then according to the fulfillment , present non-paper examination is analyzed by verifying the process dealing with the difficulty in the system, combined with the experts'experience and item-response theories. A new method of the difficulty verification is offered. The method is practical ,properly controlling the degree of difficulty,raising fair and rational examination with great progress in teaching and training.At last based on the IRT it designs a CAT system. According to each kind of topic's different characteristic,we select the different model to complete the test. A feasible strategy of item selection in CAT is designed for cognitive diagnosis. Design the methods for"Diagnosis first , Ability estimation second".The traditional method that using all the items done by examinee to estimate ability is thrown away.Some harmful items are removed. According to the different features of the items,it uses the different models to finish the CAT. The accuracy of ability estimation is improved accordingly. And this paper has performed Monte Carlo simulations to evaluate these three different structures of cognitive attributes and the results are favorable.