READSUM: Retrieval-Augmented Adaptive Transformer for Source Code Summarization

Code summarization is the process of automatically generating brief and informative summaries source code to aid in software comprehension maintenance. In this paper, we propose a novel model called READSUM, REtrieval-augmented ADaptive transformer for SUMmarization, that combines both abstractive extractive approaches. Our proposed generates an manner, taking into account structural sequential information input code, while also utilizing approach leverages retrieved summary similar increase frequency important keywords. To effectively blend original at embedding layer stage, obtain augmented representation through multi-head self-attention. addition, develop self-attention network adaptively learns representations encoder stage. Furthermore, design fusion capture relation between decoder The guides generation based on summary. Finally, READSUM extracts keywords using high-quality considers code. We demonstrate superiority various experiments ablation study. Additionally, perform human evaluation assess quality generated