Researcher(s)
- Francis Murphy, Computer Science, University of Delaware
Faculty Mentor(s)
- Austin Bart, Computer and Information Sciences, University of Delaware
Abstract
Instructors in high-enrollment CS1 courses often lack visibility into why students fail exam questions, beyond a pass/fail grade. We present an instructor-facing dashboard that integrates with the Pedal library to give instructors fine-grained insight into student exam submissions. The tool ingests student code and unit test results and classifies each submission by its most pedagogically relevant outcome (failing unit tests, syntax errors, blank submissions, unused variables), allowing instructors to sort, filter, and browse submissions by outcome, grade, and individual student.
Building on this foundation, we introduce a knowledge component (KC) model in which instructors define the skills or concepts each exam question is meant to assess. Instructors author parameterized tests, executed via Pedal, that detect whether a given KC was correctly demonstrated in a student’s code submission. KC-level performance metrics for each student enable analysis of which concepts students struggle with and how their demonstrated knowledge evolves over the duration of the course. We present the dashboard’s design, our approach to KC authoring and automated detection, and early results from applying our pipeline to CS1 exam data at the University of Delaware. We also discuss our ongoing work extending this KC-level data into a feedback system to give students personalized, written guidance and direct them to targeted practice resources.