Structure and Sequence Aligned Code Summarization with Prefix and Suffix Balanced Strategy

Area of Science:

• Software Engineering
• Natural Language Processing
• Artificial Intelligence

Background:

• Source code summarization is crucial for code comprehension and maintenance.
• Existing methods often neglect structural code information, leading to incomplete or inconsistent descriptions.
• Human factors contribute to missing or inaccurate code documentation.

Purpose of the Study:

• To propose a unified transformer-based model (SSCS) for source code summarization.
• To effectively capture both structural and sequential information from source code.
• To improve the accuracy and consistency of automatically generated code descriptions.

Main Methods:

• Developed SSCS, a structure-induced transformer encoder-decoder architecture.
• Incorporated multi-scale structural information capture with an adapted fusion strategy.
• Utilized a hierarchical encoding strategy for textual information and a bidirectional decoder for balanced summary generation.

Main Results:

• SSCS demonstrated superior performance on public Java and Python datasets.
• The model effectively integrates multi-scale structural and hierarchical textual information.
• The bidirectional decoder improved summary generation consistency.

Conclusions:

• SSCS offers a significant advancement in source code summarization.
• The proposed architecture effectively addresses limitations of previous methods.
• SSCS enhances code comprehension and facilitates software maintenance.