Binary Code Modularization Method Based on Graph Embedding

In the field of cybersecurity, software reverse engineering serves as a crucial foundational technology, particularly in the analysis of software vulnerabilities and malicious code. A key challenge within this domain is the partitioning of binary code into functional modules, which aids analysts in swiftly and accurately comprehending software structure and functionality, thereby enhancing analysis efficiency. This paper presents a novel approach for binary code module partitioning based on graph embedding, aiming to address limitations inherent in traditional methodologies employed in software reverse engineering. By abstracting software systems into attribute graphs and utilizing graph embedding-based clustering techniques to embed and cluster function nodes, this method adequately considers both node attributes and similarities, thus improving the accuracy and robustness of module partitioning. Notably, the approach employs a graph embedding clustering method based on multi-head attention mechanisms, effectively facilitating module partitioning of binary files. Experimental results demonstrate the significant effectiveness and performance advantages of the proposed method on binary files.