EduRecom AI An Educational Recommendation System Based on Students' Programming Errors

Abstract

Learning programming is often challenging for students due to the difficulty of understanding technical error messages, creating a gap between error detection and meaningful feedback. This thesis presents EduRecom AI, a hybrid intelligent system designed to support beginner Python students by providing pedagogical explanations and personalized recommendations. Unlike conventional debugging tools, the proposed system employs a four-stage diagnostic pipeline that combines deterministic techniques (compilation, execution, and static analysis) with Machine Learning methods. To ensure reliable error classification, a Multinomial Naive Bayes model was trained on a refined subset of 150,000 authentic student errors extracted from the IBM Project CodeNet dataset. In addition, a guarded generative AI layer based on GPT-4o-mini converts technical diagnostics into human-readable explanations and code corrections adapted to students’ needs. Implemented using FastAPI and React, the platform provides personalized feedback and progress monitoring, offering a scalable and effective solution for enhancing programming education.