#911: Measuring Cognitive Presence in Online Discussions: Interpretable AI and Instructional Insights from a MOOC Context

This study develops and validates an interpretable transformer model for detecting cognitive presence in MOOC discussions and examines its relation to learner outcomes. A RoBERTa model fine-tuned with theory-derived structural cues from the cognitive presence classified sentences into Triggering Event, Exploration, Integration, and Resolution. Using data from six Social Work MOOCs (4,260 learners and 227,863 sentences). The cue-augmented RoBERTa achieved strong accuracy (macro F1 = 0.70, κ = 0.65), outperforming baselines. Learner-level regressions linked integration to quiz success and motivation, while resolution predicted satisfaction and knowledge transfer. Overall, findings highlight both the potential and limitations of AI-supported cognitive presence detection for understanding and enhancing online collaborative learning. However, resolution was difficult to recall, suggesting that advanced inquiry depends heavily on context.