Abstract
AI based e-testing has become popular as an educational assessment method. Using it, the examinee's abilities can be assessed without dependence on the test characteristics (difficulties, accuracies, and so on).To achieve this, e-testing needs parallel test forms, for which each test form has equivalent measurement accuracy but with a different set of items. In recent years, automated test assemblies for parallel test forms have been proposed and put to practical use. Automated test assemblies have a trade-off between the number of tests and the bias of measurement accuracies. Hybrid Maximum Clique Algorithm Using Parallel Integer Programming method is known to assemble the greatest number of tests with the highest measurement accuracy. However, this method does not control the bias of measurement accuracies. To control this bias, this study proposes a new method: Maximum Weight Clique Algorithm and Mixed-Integer Programming (MIP) with random sampling. Specifically, the proposed method minimizes the difference between the target measurement accuracy and the actual measurement accuracy using the Maximum Weighted Clique and MIP. Experimental results demonstrate that the proposed method decreases the bias of measurement accuracies without excessively decreasing the number of tests.
