Code Change Representation

Code change representation focuses on encoding modifications to source code into numerical vectors, enabling automated analysis of software evolution. Current research emphasizes incorporating hierarchical structural information into sequence-based models, such as transformers and graph neural networks, to capture both global and local code relationships, often leveraging pre-trained models for improved performance. These improved representations are crucial for various software engineering tasks, including automated commit message generation, defect prediction, and code summarization, ultimately improving software development efficiency and reliability.