Clustering Entity Relationship Diagrams: Enhancing Feedback Quality and Grading Consistency in Large Database Courses

This innovative practice full paper introduces a tool for clustering Entity Relationship Diagrams (ERDs) and explores its application in large classes. ERDs are fundamental for database design in courses related to databases, data science, and software engineering. However, processing ERD homework submissions in large classes poses significant challenges due to the variety of design decisions made by students, leading to numerous diagram variations. This paper presents an ERD clustering tool designed to group similar ERD submissions, aiding instructors and teaching assistants in identifying popular solutions and common mistakes. The tool employs advanced object detection, OCR, and clustering technologies. We evaluated the tool using four datasets from two large public U.S. universities, with submissions ranging from 130 to 430 diagrams. Various clustering methodologies were assessed, highlighting the importance of incorporating ERD structure into the clustering process. Our findings indicate that the tool successfully generated adequate clusters, and that aiming for 10 clusters is appropriate regardless of the dataset size. The generated clusters included common approaches and mistakes, proving helpful for providing feedback and simplifying the grading process.