Cross-Architecture Binary Code Similarity-Detection Method Based on Contextual Information

With the rapid growth of software scale, binary code similarity detection is of great significance in security analysis tasks, such as malicious code detection and vulnerability mining. However, due to differences in instruction sets and inconsistent intermediate languages used by different compilers, with existing methods it is difficult to effectively implement cross-architecture detection. To address the problem of insufficient cross-architecture feature extraction in existing methods, we propose a cross-architecture binary code similarity-detection method based on contextual information. We design an assembly instruction-classification method that maps instructions implementing the same function under different architectures to the same semantic space, and makes the model learn the common features of semantically similar instructions under different architectures more efficiently through comparative learning to better capture semantic context information. In order to better capture the contextual structural information between basic blocks, we introduce graph attention neural networks to reduce the interference of noisy nodes that contain fewer instructions. The combination of semantic contextual information as well as structural contextual information ultimately improves the detection accuracy. Experimental results show that compared with existing methods, the proposed method has better performance in accuracy, precision, recall and F1-score.