Mining learner generated sketches holds significant potential for acquiring deep insight into learners’ mental models. Drawing has been shown to benefit both learning outcomes and engagement, and learners’ sketches offer a rich source of diagnostic information. Unfortunately, interpreting learners’ sketches—even sketches comprised of semantically grounded symbols—poses significant computational challenges. In this paper we describe SketchMiner, an educational sketch mining framework that automatically maps learners’ symbolic sketches to topology-based abstract representations that are then analyzed with graph similarity metrics to perform automated assessment and misconception discovery. SketchMiner has been used to mine a corpus of symbolic science sketches created by upper elementary students in inquiry-based drawing episodes as they interact with an intelligent science notebook in the domain of physical science. Results of a study with SketchMiner suggest that it can correctly assess learners’ symbolic sketches.
Smith, A., Wiebe, E. N., Mott, B. W., and Lester, J. C. (2014). SketchMiner: Mining Learner-Generated Science Drawings with Topological Abstraction. Proceedings of the Seventh International Conference on Educational Data Mining. London, England.