In educational context (online or face-to-face), with increasing cohort size and the need for individualization, the question of partial or full automation of assessments is growing. This paper deals with preliminary works that tackle the question of automatic generation of assessment Tests that could guarantee fairness and reasonable difference, by the content and the structure. A structural metric characterizing the distance between two given Tests is presented. This metric provides a dedicated fitness function that leads to define a Genetic Algorithm (GA) technique. The original use of GA allows optimizing this structural differentiation and thus guarantees the generation of collections of Tests with the largest distance possible while involving the smallest items source database. Preliminary experiments and results on the basis of multiple choice questions items are discussed.
Ciguené, R., Joiron, C., & Dequen, G. (2019). Automatically Generating Assessment Tests Within Higher Education Context Thanks to Genetic Approach. In Studies in Computational Intelligence (Vol. 774, pp. 269–282). Springer Verlag. https://doi.org/10.1007/978-3-319-95104-1_17