Principal Investigator: Neil Heffernan
CoPrincipal Investigator(s): Korinn Ostrow, Neil Heffernan, Jacob Whitehill
Organization: Worcester Polytechnic Institute
Abstract:
This is a project that will use machine learning to personalize messages about student homework. The project will apply technologies used by Google’s Smart Reply, a functionality that uses machine learning to generate and suggest human-like email responses, to provide teachers a quick and effective way to respond to student online homework. An important part of many State standards is the need for math students to communicate their ideas through writing. With the number of schools digitizing their classrooms on the rise, teachers are inundated with student data. Teachers are often unable to review and provide feedback in an effective and timely manner. This project will help teachers be more efficient and at the same time cause more effective student learning. The Dialogue Reinforcement Infrastructure for Volitional Exploratory Research – Soliciting Effective Actions from Teachers (DRIVER-SEAT) will be designed to help teachers more efficiently and effectively communicate with students in a way that feels personalized, while supported by advances in computer science. By applying a feature similar to Google’s Smart Reply in an educational setting, DRIVER-SEAT offer teachers suggestions of automated messages that can be used for more personalized feedback, thereby revolutionizing digital learning by re-incorporating teachers in an efficient and productive way.
The project will enlist teachers to create DRIVER-SEAT. These teachers will use a prototype equivalent to Google’s Smart Reply, to establish a library of trusted messages that teachers choose to provide their students. The methodology behind Google’s Smart Reply utilizes standard sequence-to-sequence machine learning techniques to automatically generate responses, grouping them into 100 clusters (with each cluster representing a specific semantic intent), and selecting messages from these clusters to suggest to users. In a similar fashion, sequence-to-sequence deep learning techniques are used to generate and suggest messages. However, instead of communicating via email, teachers will be using these messages to provide feedback for their students’ math homework. Based on student performance and system-detected affect and behavior, three appropriate feedback responses are selected to initiate interaction with each student. Cooperating teachers will help craft the library by piloting the prototype system and selecting feedback to send their students. Library development will enable machine learning to discover how to help teachers efficiently reply to their students. By implementing this technology, there is great potential to narrow the achievement gap in mathematics classrooms across the nation. This effect could then extend to science, technology, and engineering classrooms in a similar fashion. The transformative aspects of the proposed work will lead to adjustments in the way teachers and students interact in online learning environments.
This award reflects NSF’s statutory mission and has been deemed worthy of support through evaluation using the Foundation’s intellectual merit and broader impacts review criteria.
