ReDef: Do Code Language Models Truly Understand Code Changes for Just-in-Time Software Defect Prediction?

1. 1Curate a high-confidence dataset of function-level modifications from large-scale projects
2. 2Evaluate existing code language models on the ReDef dataset to assess their understanding of code changes
3. 3Analyze the results to identify areas where models excel or struggle in predicting defects
4. 4Use the insights to fine-tune models and improve their performance in just-in-time software defect prediction

Software engineers and AI researchers on a team can benefit from ReDef to improve the accuracy of defect prediction models and prioritize risky code changes during code review

💡 Code language models' ability to understand code changes is crucial for accurate defect prediction, and ReDef provides a high-confidence dataset to evaluate and improve these models

🚨 ReDef: a new benchmark for just-in-time software defect prediction 🚨