Research Framework Evaluates Adaptive Personalization of Educational Readings Using Simulated Learners

A new framework has been established for assessing the adaptive personalization of educational reading resources, utilizing theory-based simulated learners. This system creates an ontology of learning objectives and knowledge components derived from open textbooks, which is organized in a web-based Ontology Atlas. Segments of textbooks are tagged with ontology entities, and corresponding reading-assessment pairs are created. The simulated readers utilize a memory model inspired by Construction-Integration, incorporating DIME-style reader factors, KREC-style misconception correction, and an open New Dale-Chall readability indicator. Responses are generated through score-based option selection from the learner's explicit memory state, with adaptation guided by Bayesian Knowledge Tracing (BKT). In studies involving three subject ontologies with 50 simulated learners each, adaptive reading notably enhanced performance in computer science, while modest and inconclusive improvements were found in inorganic chemistry. This framework was published on arXiv under the identifier arXiv:2604.16744v1, classified as a cross announcement. The findings highlight the potential of simulated learners in evaluating educational personalization systems prior to their real-world application.