Just-in-time software defect prediction via bi-modal change representation learning

Yuze Jiang, Beijun Shen, Xiaodong Gu

Abstract: Highlights•A novel code change pre-training model that extracts bi-modal semantic information from code changes.•A novel pre-training objective allowing the model to explicitly learn the semantic association between commit messages and code changes.•An approach for just-in-time defect prediction (JIT-DP) based on semantic representations extracted by the pre-trained BiCC-BERT.•Significantly outperform the state-of-the-art approaches.