The Multi-Level Regulation Engine: A Socially Adaptive Intelligent Tutoring System Architecture for Heterogeneous Collaborative Learning

<title>Abstract</title> <p>We introduce the Multi-Level Regulation Engine (MLRE), an Intelligent Tutoring System (ITS) architecture that regulates learning simultaneously at individual, dyadic, and group levels in heterogeneous collaborative groups, grounded in social learning theory. Most ITS adapt to learners as isolated individuals, leaving the quality of collaboration in diverse groups unregulated. Following a Design Science Research methodology, we validated two of the architecture’s four layers offline prior to any classroom deployment. We benchmarked individual learner modeling on the public ASSISTments dataset (area under the curve = 0.78, consistent with the literature). We tested the MLRE’s decision logic through a 1,000-session simulation (200 sessions × 5 scenarios), improving participation equity in every scenario, with the qualitative direction of this effect confirmed robust across a 135-combination parameter sweep. The group-interaction-modeling layer, identified as the architecture’s central novelty, remains to be validated on authentic multi-party data. We propose a quasi-experimental field study to evaluate learning performance, participation equity, and socio-metacognitive development. This work contributes a formally specified, partially validated, and ethically grounded artifact for inclusive, explainable artificial intelligence in education.</p>