Back to Writing

Research publication

Distributed Representation of Misconceptions

Used neural embeddings and large-scale Khan Academy response data to surface shared mathematical misconceptions across varied problem instances.

This conference paper used answer sequences from three Khan Academy fraction exercises to learn high-dimensional representations of incorrect responses. The resulting clusters were compared with expert coding and interpreted through a constructivist account of how partial mathematical understandings develop.

Pardos, Z. A., Farrar, S., Kolb, J., Peh, G. X., & Lee, J. H. (2018). Distributed Representation of Misconceptions. Proceedings of the 13th International Conference of the Learning Sciences, 1791–1798.