Significant efforts are underway to transition learning from today’s “one-size-fits-all” approach to one that personalizes the learning experience according to the needs and skills of particular students. Several systems have achieved some success by writing out a massive set of rules dictating how to teach a student a topic given that student’s prior learning performance. These rule-based systems are painstakingly constructed by a team of domain experts, computer scientists, and cognitive scientists for one specific course. This approach has two major disadvantages. The resources required to build the system for one course makes scaling up to many courses prohibitively expensive. And even assuming scalability, rule-based approaches are fragile: what happens when a student deviates from the course’s predetermined rules and how can these global rules be truly personalized for different kinds of learners?
OpenStax Tutor, a collaboration between engineers and cognitive scientists at Rice University and Duke University, addresses these challenges by fusing cognitive science learning strategies and modern machine learning algorithms; the result is an automated, personalized, and optimized learning experience for today’s courses and students.
OpenStax Tutor can marshal many different open educational resource (OER) sources in its quest to improve student learning, but two repositories stand out: Connexions (cnx.org) for rich e-texts and Quadbase (quadbase.org) for assessments. Connexions is one of the world’s first and largest OER projects. Connexions’ repository of free, open-source educational content is accessible to students, instructors, and authors worldwide. Quadbase is an open access question bank, focused on serving instructors and educational platforms with support for multiple question types and embedding options. Both platforms thrive on community-submitted and -curated content, and access remains free to all under a Creative Commons attribution license.
OpenStax Tutor runs on a unique pairing of two suites of advanced machine learning algorithms. The first takes a large corpus of OER materials, including texts, questions, videos, and simulations, automatically determines how these materials interrelate, and stores them in Linkify (linkify.org), a new system for understanding semantic links between web resources. The second takes the relationships from Linkify along with student learning records, and chooses each student’s next learning step to optimize their progress. These decisions are based on each student’s individual learning record, in addition to knowledge of other student’s prior successes and failures. OpenStax Tutor also builds on proven learning strategies from cognitive science. Questions for a topic are asked when the topic is introduced and at several later times to lock the knowledge in. Students also receive immediate feedback when working through assessments. Outcomes from the machine learning algorithms also provide fine-grained learning diagnostics on instructor and student dashboards.
Building off our initial success at Rice University, OpenStax Tutor is currently in its second round of pilot testing at several universities.
In this 45 minute session, OpenStax Tutor architect, JP Slavinsky, and algorithms expert, Richard Baraniuk, will overview the primary components of the system. Project Manager, Daniel Williamson, will overview the community approach to developing content in both Connexions and Quadbase.
