Data-Driven Team Assignment in Software Engineering Education: A GitHub-Informed, Difficulty-Aware Approach

Résumé fourni par la source

This is a research-to-practice full paper. Team-based projects play a pivotal role in software engineering education, shaping students' technical expertise and collaborative skills. A key determinant of success in such projects is the formation of well-balanced and compatible teams, where members' skills and experiences align with the complexity and requirements of the assigned projects. In this study, we propose a structured, data-driven framework that leverages GitHub-derived numerical metrics (such as repository counts, commits, and language proficiency) and textual artifacts (such as README files) to match students to projects in a way that promotes both individual contribution and collective performance. Our framework integrates three complementary components: (1) skill-level clustering employing a Constrained K-Means algorithm to ensure balanced team composition, (2) semantic similarity matching using Sentence-BERT embeddings to align student experience with project requirements, and (3) difficulty-aware project ranking based on linguistic coherence and code complexity. Empirical results reveal that substantial performance gains are achieved only when semantic alignment and difficulty calibration are jointly applied alongside skill balancing. Isolating any individual component leads to diminished outcomes, underscoring their interdependence. Teams formed through the full pipeline significantly outperformed both randomly assigned groups and partial configurations, highlighting the pedagogical value of holistic, data-informed team formation in project-based learning environments.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.