Option Tracing: Beyond Binary Knowledge Tracing
This paper details our solutions to Tasks 1&2 of the NeurIPS 2020 Education Challenge.1 Knowledge tracing, a family of methods to estimate each student’s mastery levels on skills/knowledge components from their past responses to assessment questions, is useful for progress monitoring, personalization, and helping teachers to deliver personalized and targeted feedback to students to improve their learning outcomes. One key limitation of current knowledge tracing methods is that they can only estimate an overall knowledge level of a student since they analyze only the binary-valued correctness of student responses. We adapt a series of popular knowledge tracing methods to the task of option prediction in multiple choice questions. Experimental results show that our method performs well on both option prediction and correctness prediction.
PDF