TutorTrace Dataset Helps AI Programming Tutors Adapt to Learner Behavior

Summary

TutorTrace is a dataset and behavioral pipeline built from roughly 180,000 IDE telemetry events collected from 480 learners in two introductory Python courses. Its taxonomy describes activity before and between AI queries and across full sessions. In a preliminary evaluation, behavior-aware prompts reduced query intervals without independent work from 50.0% to 20.7%. The system also predicted queries within 60 seconds and distinguished guided from dependent help-seeking, with AUROC scores of 0.726 and 0.717 respectively.