Improving the 3PLM-Based Computerized Adaptive Testing System with Multi-Agent Item Bank

A multi-agent-based computerized adaptive testing (CAT) system on IRT (item response theory) is introduced, which uses the three-parameter logistic model (3PLM) of IRT as the theoretical model. The paper explores and researches into the steps of constructing item bank , determining the beginning item, response, estimating ability, item selection and terminating condition, and a IRT-based item bank on multi-agent system is put forward here. Aiming at the limits of the most commonly used Newton-Raphson iterative algorithm in the ability estimating, median value algorithm is applied to calculate maximum likelihood estimation (MLE) instead of that, and utilizes a series of adjusting strategy to solve some other shortages in the logistic median value model. This CAT system has more superior function and lower calculation quantity, with fast convergence and the accuracy of ability estimating, which will be adapted further in modern test system.