This project, at North Carolina State University, will explore ways to augment intelligent tutoring systems by using methods that use historical data from student work on assigned exercises, to enhance the tutoring system's ability to decide what to teach and how to teach it. This research will utilize both hint generation and worked examples. The PI team will begin by augmenting three existing learning environments, adding data-driven techniques for the automatic generation of next-step hints and for the automatic selection of learning activities. Subsequent studies will increase understanding of the benefits provided by hint mechanisms by comparing the effectiveness of sub-goal hints with that of next-step hints. This will then lead to empirical evaluations of the learning impact of such data-driven student support.
The project team hypothesizes that existing logic and probability tutors will produce significant learning gains when enhanced by data-driven hint generation coupled with data-driven pedagogical strategy induction. The project will compare logic, probability, and programming learning with and without data-driven hints and data-driven pedagogies, measuring quantitative and qualitative impact on student success. The research team will use a variety of measures of learning, such as time to learn, number of errors made, number of hints requested, and engagement, as well as qualitative measures such as student surveys that gauge self-efficacy and motivation. Student performance data will be analyzed using correlation, analysis of variance, regression and significance testing.
