EnCoSum: enhanced semantic features for multi-scale multi-modal source code summarization

Code summarization aims to generate concise natural language descriptions for a piece of code, which can help developers comprehend the source code. Analysis of current work shows that the extraction of syntactic and semantic features of source code is crucial for generating high-quality summaries. To provide a more comprehensive feature representation of source code from different perspectives, we propose an approach named EnCoSum, which enhances semantic features for the multi-scale multi-modal code summarization method. This method complements our previously proposed M2TS approach (multi-scale multi-modal approach based on Transformer for source code summarization), which uses the multi-scale method to capture Abstract Syntax Trees (ASTs) structural information more completely and accurately at multiple local and global levels. In addition, we devise a new cross-modal fusion method to fuse source code and AST features, which can highlight key features in each modality that help generate summaries. To obtain richer semantic information, we improve M2TS. First, we add data flow and control flow to ASTs, and added-edge ASTs, called Enhanced-ASTs (E-ASTs). In addition, we introduce method name sequences extracted in the source code, which exist more knowledge about critical tokens in the corresponding summaries and can help the model generate higher-quality summaries. We conduct extensive experiments on processed Java and Python datasets and evaluate our approach via the four most commonly used machine translation metrics. The experimental results demonstrate that EnCoSum is effective and outperforms current state-of-the-art methods. Further, we perform ablation experiments on each of the model’s key components, and the results show that they all contribute to the performance of EnCoSum.