Research publication
Generalizing Expert Misconception Diagnoses through Common Wrong Answer Embedding
Mapped expert diagnoses onto neural embeddings of millions of Khan Academy answers to predict misconceptions on previously unseen questions.
This conference paper extended the earlier embedding work by combining millions of Khan Academy answer events with expert-written misconception diagnoses. Answer embeddings and cross-validated regression were used to test whether diagnostic language could generalize across educators, problem types, and previously unseen questions.
Kolb, J., Farrar, S., & Pardos, Z. A. (2019). Generalizing Expert Misconception Diagnoses through Common Wrong Answer Embedding. Proceedings of the 12th International Conference on Educational Data Mining, 342–347.