Review of automated parallel test form assembly

This paper presents a review of advanced automated parallel test form assembly (ATA) methods for computer-based testing (CBT) using artificial intelligence approaches. Parallel test forms ensure equivalent measurement accuracy of examinees’ scores across different sets of question items and thereby enable assessments to be administered at any time and from any location. First, this paper presents a review of ATA methods, organizing them into four categories: Mixed-Integer Programming (MIP), multi-form MIP, metaheuristics, and form-maximization. Second, this paper describes a comparison of the characteristics of these methods in terms of measurement accuracy, accessibility, computational complexity, and feasible testing frequency. Moreover, based on these characteristics, the review presents an example decision framework for selecting ATA methods according to the requirements of CBT. Finally, numerical experiments were conducted for this study to compare representative ATA methods in terms of their relative strengths and weaknesses.