Title:Enhancing Source Code Representations for Deep Learning with Static Analysis

Abstract:Deep learning techniques applied to program analysis tasks such as code classification, summarization, and bug detection have seen widespread interest. Traditional approaches, however, treat programming source code as natural language text, which may neglect significant structural or semantic details. Additionally, most current methods of representing source code focus solely on the code, without considering beneficial additional context. This paper explores the integration of static analysis and additional context such as bug reports and design patterns into source code representations for deep learning models. We use the Abstract Syntax Tree-based Neural Network (ASTNN) method and augment it with additional context information obtained from bug reports and design patterns, creating an enriched source code representation that significantly enhances the performance of common software engineering tasks such as code classification and code clone detection. Utilizing existing open-source code data, our approach improves the representation and processing of source code, thereby improving task performance.