Tag
This arXiv paper examines how mastery thresholds interact with different knowledge tracing models, showing that the same cutoff produces very different advancement decisions across BKT and neural KT models, and that thresholds should be recalibrated per model and instructional priority.
The paper shows that study-strategy clusters from EdNet logs predict learner engagement but not mastery, highlighting that behavior-only profiling is insufficient for assessing knowledge gains.
This paper introduces the Unified Behavioral Prediction and Calibration Analysis Pipeline (UBP-CAP), an integrated framework linking student performance prediction, calibration error calculation, and variance decomposition of metacognitive misalignment. Evaluated on a dataset of 1,195 interaction records, the pipeline identifies key predictors of correctness and reveals that metacognitive calibration is primarily situational rather than a stable trait.
This paper benchmarks seven LLM feedback agents in propositional logic tutoring, finding they perform well on optimal steps but systematically fail to correctly diagnose valid suboptimal and incorrect solutions, highlighting limitations for adaptive tutoring.
This paper introduces engagement forecasting for intelligent tutoring systems, predicting weekly minutes practiced and new skills mastered using interaction logs from 425 middle-school students. Feature-based models reduce error by 22-33% over heuristic baselines, offering explainable patterns for tutor-learner goal setting.