← Back to Research
Supporting Novice Data Science Learners through Personalized Feedback
This project explored novice data science learners' common errors and misconceptions, and built a feedback-generation system that automatically identifies what went wrong to deliver personalized hints, at scale.
Methods
Analysis of programming logs + Qualitative analysis of incorrect submissions
Instructor interviews
Automatic feedback generation tool developed at Microsoft
Course Deployment at University of Michigan
Collected student feedback and shared with Microsoft
My Contributions
- Bridged collaboration between University of Michigan and Microsoft, led data sharing and helped both teams understand each other's domains and methodologies.
- Worked with cross-functional team of engineers, product managers, and researchers at Microsoft to advance their understanding of online learning and data science pedagogy.
- Developed taxonomy of data science programming mistakes to enabled scalable, personalized support.
- Shared findings with executive stakeholders at Microsoft to help improve data science developer tools.
Impact
- Deployed across 2 semesters of the Master of Applied Data Science program at the University of Michigan.
- Enabled real-time support on programming assignments, which was previously unavailable to asynchronous online learners in the program.