Intervention-BKT: a BKT extension that models the effect of instructional interventions on student knowledge states¶

Theory · All theories
Evidence · no claims cited

Description¶

Intervention-BKT (IBKT) extends Bayesian Knowledge Tracing by adding input nodes representing instructional interventions such as elicit and tell. The article describes it as "an extension of BKT, which incorporates the effect of multiple types of instructional interventions on student modeling": a student's knowledge state depends on the previous state and the current intervention, and performance depends on both the knowledge state and the intervention. It is a special case of an Input Output Hidden Markov Model using 1+4×K parameters, with per-intervention Learning Rate, Forget, Guess and Slip parameters.

Design Implications¶

Context¶

Requirements¶

• Requires logged sequences of instructional interventions paired with student performance observations, unlike conventional BKT which trains on performance alone

Constraints¶

• Observations on tell interventions are noisy because the exact duration a student spent reading a tell step is unknown; the authors used "fast and no performance observed" for all tells

Target Learners¶

• college students in introductory physics or probability intelligent tutoring systems

Target Learning Objectives¶

• domain knowledge in physics and probability as measured by post-test scores and learning gains

Claims¶

Related Theories¶

• Four-Phase Taxonomy of Knowledge Tracing Variants
• Bayesian Knowledge Tracing (Two-State Hidden Markov Model)

Examples¶

-

Key Sources¶

• Ye Mao, Chen Lin, and Min Chi. (2018). Deep Learning vs. Bayesian Knowledge Tracing: Student Models for Interventions. Journal of Educational Data Mining, Volume 10, No 2. https://jedm.educationaldatamining.org/index.php/JEDM/article/view/318