McMining: Automated Discovery of Misconceptions in Student Code.
| Authors | Erfan Al-Hossami & Razvan C. Bunescu |
| Year | 2026 |
| Venue | EACL 2026 |
| Paper | View on ACL Anthology |
| PDF | Download |
Abstract
Abstract not yet available in this stub. Read the full paper →
Engineering Breakdown
Plain English
This paper presents McMining, an automated system for detecting misconceptions in student code by analyzing common error patterns. The system uses NLP techniques to identify systematic bugs and misunderstandings in programming assignments, enabling instructors to provide targeted feedback at scale without manual review of every submission.
Key Engineering Insight
The core innovation is treating misconception detection as a pattern mining problem rather than a classification problem—the system discovers what misconceptions exist in a student population first, then identifies students exhibiting those patterns, rather than trying to predict pre-defined error types.
Why It Matters for Engineers
For engineers building educational platforms and automated grading systems, this approach solves the real problem of providing personalized, actionable feedback to thousands of students. Rather than generic error messages, the system can surface the specific conceptual gaps that groups of students share, enabling both better UX and more effective learning outcomes.
Research Context
Previous work on automated code feedback focused either on syntax checking or matching against solution rubrics. McMining advances this by discovering emergent misconception patterns from student work itself—filling the gap between generic feedback and instructor-written rubrics. This enables scalable, adaptive feedback systems that improve as more student data accumulates.
:::tip Subscribe Get weekly breakdowns of papers like this in AI Letters - the newsletter for engineers building production AI systems. :::