Matsuda, N., Chandrasekaran, S., & Stamper, J. (2016). How quickly can wheel spinning be detected? In T. Barnes, M. Chi & M. Feng (Eds.), Proceedings of the International Conference on Educational Data Mining (pp. 607-608).
Matsuda, N., Chandrasekaran, S., & Stamper, J. (2016). How quickly can wheel spinning be detected? In T. Barnes, M. Chi & M. Feng (Eds.), Proceedings of the International Conference on Educational Data Mining (pp. 607-608).
Abstract: We have developed a wheel spinning detector for cognitive tutors that uses a simplified method compared to existing wheel spinning detectors. The detector reads a sequence of the correctness of applying particular skill performed by a student using the cognitive tutor. The response sequence is first fed to Bayesian knowledge tracing to compute a sequence of probability of mastery at each time a skill was applied. The detector uses a neural-network model to make a binary classification for a response sequence into wheel- spinning and none-wheel spinning. To test the accuracy of the detector, we validated the detector using learning interaction data taken from a school study where students used a Geometry cognitive tutor. Human coders manually tagged the data to identify wheel spinning. The results show that the neural-network based detector has high recall (0.79) but relatively low precision (0.25) when combined with Bayesian knowledge tracing that detects mastery cases. The result suggests that the neural-network based detector is practical and has a potential for scalable use such as adaptive online course where cognitive tutors are embedded into online courseware.
PDF: download