Compose LKT models by starting with a student-ability feature, adding KC/item difficulty intercepts, then adding and splitting dynamic learning features¶

Strategy · All strategies
Evidence · no claims cited

Description¶

The article recommends a stepwise procedure for building a learner model from scratch in LKT. Models meant to generalize to unseen students should begin with the propdec or logitdec features to capture student variability; initial KC or item difficulty is accounted for with fixed or random intercepts; then learning features such as lineafm are added, and dynamic features that work well can be split (e.g., lineafm into linesuc and linefail) to capture richer effects, with cross-validation confirming generalizability.

Design Implications¶

Context¶

Requirements¶

• A dataset with a sequence of learner events including subject and correctness columns
• Cross-validation and model visualizations to confirm found models are generalizable

Constraints¶

• Fixed effects may be inadequate if items are truly sampled from a distribution or if data are sparse

Target Learners¶

• students in adaptive learning technology systems

Target Learning Goals¶

• accurate prediction of student correctness to guide pedagogical decisions

Affordances¶

• Lkt Logistic Knowledge Tracing Framework

Related Strategies¶

Examples¶

-

Key Sources¶

• Philip I. Pavlik, Jr., Luke G. Eglington, and Leigh M. Harrell-Williams. (2021). Logistic Knowledge Tracing: A Constrained Framework for Learner Modeling. IEEE Transactions on Learning Technologies. https://doi.org/10.1109/TLT.2021.3128569