Learning Semantic Representation for Binary Code Similarity Detection

To avoid the human bias introduced by numerical statistical features and overcome difficulty in cross platform and optimization detection methods, this paper proposes a binary code similarity detection method based on semantic representation generated from deep learning models. This method uses the intermediate representation to eliminate the differences in syntax caused by compilation options. The feature vector of basic block with context semantic information is generated through natural language processing model, and the attention mechanism is introduced into the graph embedding neural network to realize the accurate construction of function features. The results show that this method has better results in accuracy, precision, recall, and F1 value compared to advanced technologies in this field.