A Content-Balanced Adaptive Testing Algorithm for Computer-Based Training Systems

摘要

There are two main obstacles that make use of adaptive testing in computer-based training difficult. One is the requirement of conducting a large-scale empirical study for item calibration. The other is the difficulty of generating content-balanced tests that meet the goal of the test administrators. In this research, we have developed a new adaptive testing algorithm, CBAT-2, to provide a solution for these problems and some other practical problems in adaptive testing. CBAT-2 generates questions based on the portion of the course curriculum that meets the goals of a test. It uses a simple machine learning procedure to determine the item parameter values.