Research On Item Selection Strategy Under Maximum Priority Index

Item Response Theory (IRT) is developed on the basis of the latent trait theory.Compare with the traditional classical test theory (CTT), it has some advantages suchas the ability estimation independent of the sample, more accurate on item parameterestimation and so on. Computerized Adaptive Testing (CAT) is an importantapplication of IRT in artificial intelligence. It can be more intelligently on choosingitems, so that examinees answer the items which match their abilities. Then theirabilities were estimated through Expected a Posterior (EAP) estimation or otheralgorithms. We can see that the IRT in aid of the computer is faster and moreaccurately than CTT, also reduces the costs on human and material resources. In CAT,item selection strategies is a core part and also the most intelligent part, because itneeds to consider many factors, such as measurement accuracy, exposure rate and soon. How to trade off these conflicting factors is a subject for further research. IRT ismainly divided into binary scored model and polytomous scored model. Thepolytomous scored model mainly contains Graded Response Model (GRM) andGeneralized Partial Credit Model (GPCM). This study is based on GRM.In binary scored model, a-STR and exposure factor strategy are both good atcontrolling exposure rate, and the exposure factor strategy is better than the other.How to introduce exposure factor into polytomous scored model, how effective is thenew strategy and is it better than any other strategies are the focus of this study. Forthe balance control of the restrictions, there are c-STR and Maximum Priority Index(MPI) method. Then Pan Yirao found that MPI makes some restrictions cross theborder, and then put forward MMPI method. In the process of selecting items, MPIand MMPI will both easy to select the items which have large amounts of informationand contain more restrictions, so that these items appear frequently. The study putforward a new strategy named MMPI with exposure factor. The results show that thenew method will makes greatly improvement on exposure controlling. In addition, theMPI one-stage method may be difficult to avoid the cross-border situation. This studyanalyzes the reasons for the phenomenon, and then gives a revise. The experimentsshow that, the revised method is better. In addition, taking into account the lack ofexposure control in the revised method, the study combines the revised method with exposure factor. The ultimate method has advantages such as simple formula, lowexposure rate, accurate estimation and well controlling on restricts.