Template-based question generation from code using static code analysis

Abstract This study presents a multilingual system for automatic question generation from source code in Python, C++, Java, and C. Using static code analysis, and template-based methods; it extracts code structure and generates questions aligned with Bloom's Taxonomy. A dataset of 456 questions from 19 algorithms and three code complexity levels was used. Current systems are monolingual; this approach handles four programming languages with a unified framework. The system was evaluated using several metrics, including the overall quality score. The automatic evaluation shows that C achieved a slightly higher overall quality score of 0.59, while the other languages scored 0.57. Human evaluation complements the automated evaluation, providing educators' perspective on question quality.