The DRL policy produced a significantly different scaffolding distribution than Control and BKT, favoring Guided examples (60%) and avoiding Buggy examples (4%)¶

Claim · All claims
Evidence · 1 study · causal r2 · q3
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q3 i? The DRL group's training problem-type distribution differed significantly from both Control and BKT, with 35% PS, 4% Buggy, and 60% Guided examples. → Tithi 2026

Evidence¶

Tithi 2026¶

Tithi, S.D., Alam, N., Yasir, T., Shi, Y., Tian, X., Chi, M., Barnes, T. (2026). Adaptive Scaffolding for Cognitive Engagement in an Intelligent Tutoring System. arXiv. https://arxiv.org/abs/2602.07308 doi:10.48550/arxiv.2602.07308

q3 · i? · causal · r2

Chi-square tests of training problem-type distributions across the three conditions in the N=113 study (RQ1). The DRL group's distribution "encountered significantly different distribution from bothControl" and BKT, whereas Control and BKT did not differ significantly (χ2(2) = 1.77, p= 1.00).

"students inDRLgroup en- countered significantly different distribution from bothControl(χ2(2) = 131.2,p < .001), andBKT(χ 2(2) = 143.1,p < .001), with 35% PS, only 4% Buggy, and 60% Guided."

Discussion¶

Related Claims¶

• Adaptive scaffolding policies (BKT and DRL) significantly improve posttest performance over a non-adaptive control in a logic ITS — related
• BKT students completed posttest problems significantly faster than Control students, with DRL showing marginal advantages in time and solution optimality — related
• BKT scaffolding significantly benefits low prior knowledge students, who outperformed low prior knowledge controls — related
• DRL scaffolding significantly benefits high prior knowledge students while BKT does not, and condition-by-prior-knowledge interactions are not significant — related
• No significant differences in posttest rule accuracy across scaffolding conditions — related
• All training policies narrowed the prior-knowledge achievement gap from pretest to posttest, with BKT achieving the largest reduction (77.1%) — related
• In simulated online mastery learning, BKT gave the fewest problems, Streak the second fewest, Random the most, and DKT gave almost as many as Random in one problem type and the least in the other two — related