Acceso usuarios
[3]
;
Marini Othman
[1]
;
Tin Tin Ting
[2]
;
Yuhang Wang
[4]
Malasia
Australia
Automated Examination Paper Generation (EPG) has become a critical component of programming education, particularly in contexts where a common item bank is required to support multiple student cohorts characterised by varying difficulty levels, cognitive priorities, and skill-coverage needs. To establish a review-informed foundation for the practical design of deployable EPG systems in such environments, this study undertakes a PRISMA 2020-guided systematic literature review that conceptualises EPG as a multi-constraint combinatorial optimisation problem. By adopting a conservative evidence boundary together with a two-tier coding framework, the review integrates the ways in which earlier studies have operationalised structural requirements, difficulty levels, score allocation, skill coverage, Bloom’s taxonomy-aligned cognitive domains, and efficiency-related performance measures. In parallel, it critically examines whether these studies provide deployment-relevant details, including item-bank metadata, transparency in constraint-handling procedures, trade-off decision rules, benchmark comparisons, and evidence relating to runtime performance and scalability. Across the reviewed literature, solution approaches extend from basic heuristics and exact optimisation models to metaheuristic methods, hybrid GA-SA architectures, and learning-augmented techniques. The synthesis indicates that hybrid GA-SA is consistently presented as a credible engineering framework for addressing dense, multi-objective EPG problems. Nevertheless, the existing body of literature remains limited by incomplete methodological reporting, insufficient machine-readable formalisation of pedagogical objectives, and a lack of evidence relevant to governance and implementation. Rather than asserting the universal superiority of any single algorithm, this review converts diverse findings into a deployability-oriented synthesis grounded in MCDM principles. Four major design implications are identified for programming-focused EPG systems: clearly defined and auditable constraints, Bloom- and skill-aligned parameter specification, teacher-in-the-loop calibration mechanisms, and reproducible evaluation procedures. Collectively, these findings establish a defensible link between algorithmic selection and pedagogical application, while also providing a robust evidence base for future hybrid GA-SA development in cross-major programming assessment contexts.
