This study presents an automated systematic review of augmented reality research within educational neuroscience, utilizing PubMedBERT, a domain-specific natural language processing model, to extract and analyze research entities from fifty peer-reviewed studies. The analysis examined interventions, learning outcomes, cognitive constructs, neuroscience methods, and neural correlates to map current research patterns and methodological approaches. The corpus, derived from PubMed searches using the term “neuroeducation,” revealed electroencephalography as the most frequently employed neuroimaging technique and the anterior cingulate cortex as the most commonly investigated brain region. Learning outcomes were predominantly short-term and performance-oriented, with response time and accuracy emphasized over retention and transfer measures. The study integrates Cognitive Load Theory and Dual Coding Theory to interpret how augmented reality interventions may regulate cognitive load and engage dual-channel processing. Validation analysis of automated extraction demonstrated acceptable precision across entity types, with some variation in recall rates for complex relationships. The review identifies methodological patterns in augmented reality neuroscience research while acknowledging limitations in search scope and the descriptive nature of frequency-based analysis. The findings suggest that while augmented reality shows promise for investigating neurocognitive mechanisms of learning, the field requires more longitudinal designs, multimodal neuroimaging approaches, and explicit connections between neural measures and sustained learning outcomes to establish educational validity.
Çelik et al. (Tue,) studied this question.