Abstract:

Diversity has long been recognized as an important component of science and engineering workforce. Professionals with different social and cultural backgrounds bring varied perspectives and thought process, making a group of effective problem solvers. Inclusion of people with diverse ethnicity helps form a welcoming environment, encouraging recruitment of talented individuals from different groups.

Promoting diversity means that the education system has to move away from a one-size-fits-all approach. We have to respect the uniqueness of each individual, and design education keeping in mind that each one has their own pace of learning. We need a personalized approach to learning, to ensure all students in a science or engineering class attain a same degree of learning in concepts and skills.

Until recently, making a personalized learning system was a daunting task. It involved hundreds of hours of intense work from skilled computer programmers and a strong understanding of pedagogy to make personalized tutoring systems. However, today technology has grown far more accessible and empowered teachers to create their own learning systems. The recent introduction of freely available tutor authoring software called Cognitive Tutor Authoring Tools (CTAT) has enabled every teacher - who may not be a computer programmer - author their own personalized tutor for their students.

In this work, we describe the process of constructing a tutor for personalized learning in Computer Science, and demonstrate its efficacy through learning analytics data. This tutor was constructed to help students learn an important concepts in Data Structures and Algorithms - a gateway course in undergraduate computer science. By splitting a complex problem solving exercise into a reasonable number of easy to master steps, we helped pace student learning in bite-sized mini drills.

Fine grained analysis of student interaction data with our tutor showed that solving the exercise consisted of several steps. By fitting student data to an exponential decay type learning curve, we show that students got better as they worked with our tutor, indicating learning. We split aspects of student learning by analyzing three important features: prior knowledge (knowledge students have before working with tutor), learning rate (quantity indicating how fast students master a skill), and residual error (learning gaps that students are left with, even after completing tutoring exercise).

While some skill were easy for students to learn, others were relatively hard, as indicated by differences in learning rate. We quantitatively identify areas with relatively high residual error, and discuss ways to eliminate them in future exercises. As the main strength of our insights come from student data, we would like to encourage student participation from international collaborators.

Keywords:

Science education, engineering education, computer science education, mathematics education, personalized learning, intelligent tutoring systems.