CAREER: Educational Data Mining for Student Support in Interactive Learning Environments

This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).

Creating intelligent learning technologies from data has unique potential to transform the American educational system, by building a low cost way to adapt learning environments to individual students, while informing research on human learning. This project will create the technology for a new generation of data-driven intelligent tutors, enabling the rapid creation of individualized instruction to support learning in science, technology, engineering, and mathematics (STEM) fields. This has the potential to make individualized learning support accessible for a broad audience, from children to adults, including students that are traditionally underrepresented in STEM fields.

This project will (1) develop computational methods to derive cognitive models from data that can be used to support individual learners through guidance, feedback, and help; (2) develop approaches to providing student support that leverage data to provide hints and guidance based on information such as frequency of student responses, probability of future errors, and solution efficiency; (3) develop interactive visualization tools for teachers to learn from student data in real time, to allow teachers and instructional designers to tailor instruction to address actual, rather than perceived, student problem areas; and (4) conduct formal empirical evaluations of pedagogical effectiveness.

The new software will construct adaptive support for teaching and learning in logic, discrete mathematics, and other STEM domains using a data-driven approach. From the extensive but tractable student performance data in computer-aided learning environments, student cognitive models will be automatically constructed. These cognitive models will build on the investigator's prior work using Markov Decision Processes and dimensionality reduction methods that leverage past data to assess student performance, direct a student?s learning path, and provide contextualized hints. Machine learning techniques will be used to expand problem-specific models into more general cognitive models to bootstrap the construction of new tutors and learn about student learning. For teachers and learning researchers, web-based visualization and analysis tool will be developed to graphically and interactively model student solutions annotated with performance data that reflect frequency, tendency to commit future errors, and closeness to a final solution. Through these new tutors and tools, experiments will be conducted to investigate student learning in a variety of contexts and domains, including logic, algebra, and chemistry. A team of diverse students and colleagues will be engaged to bring interdisciplinary expertise to this research and share findings broadly.