Discovering the Pedagogical Resources that Assist Students to Answer Questions Correctly – A Machine Learning Approach

This paper describes preliminary results from a study in which we apply machine learning (ML) algorithms to the data from the introductory physics MOOC 8.MReV to discover which of the instructional resources are most beneficial for students. First, we mine the logs to build a dataset representing, for each question, the resources seen prior to each answer to this question; Second, we apply Support Vector Machines (SVMs) to these datasets to identify questions on which the resources were particularly helpful. Then, we use logistic regression to identify these resources and quantify their assistance value, defined as the increase in the odds of answering this question correctly after seeing the resource. The assistance value can be used to recommend resources to students that will help them learn more quickly. In addition, knowing the assistance value of the resources can guide efforts to improve these resources. Furthermore the order of presentation of the various topics can be optimized by first presenting those whose resources help on later topics. Thus, the contribution of this work is in two directions. The first is Personalized and Adaptive Learning, and the second is Pedagogical Design.