Code Structure–Guided Transformer for Source Code Summarization

Code summaries help developers comprehend programs and reduce their time to infer the program functionalities during software maintenance. Recent efforts resort deep learning techniques such as sequence-to-sequence models for generating accurate code summaries, among which Transformer-based approaches have achieved promising performance. However, effectively integrating structure information into Transformer is under-explored in this task domain. In article, we propose a novel approach named SG-Trans incorporate structural properties Transformer. Specifically, inject local symbolic (e.g., tokens statements) global syntactic dataflow graph) self-attention module of inductive bias. To further capture hierarchical characteristics code, are designed distribute attention heads lower layers high Extensive evaluation shows superior performance over state-of-the-art approaches. Compared with best-performing baseline, still improves 1.4% 2.0% on two benchmark datasets, respectively, terms METEOR score, metric widely used measuring generation quality.