Costutor: An intelligent tutor for introductory computer science students

Abstract

In this thesis we describe the development of costutor, an automated, personalized tutor for students of Princeton University’s introductory CS Course, COS 126. We build off of work in Intelligent Tutoring Systems, with a focus on creating a tool that is intuitive and appealing to use and which draws upon research in educational psychology. We develop an intuitive markup language that helps exercise writers create dozens of variations on a programming problem very quickly, and a method of creating aptitude profiles from these exercises that can be analyzed to produce recommendations for students. Finally, we present the results of testing costutor on COS 126 students who used it to prepare for their final programming exam. These results suggest that costutor mostly succeeds in recommending exercises of the correct difficulty to students. We also note interesting and unexpected use patterns by students, which suggest places for future improvement. Areas for future work include improving costutor’s recommendation algorithm based on user feedback and applying costutor’s core recommendation algorithm to other domains of knowledge.