Similarity Measure for Binary Function Based on Graph Mover’s Distance

Graph matching has a wide range of applications. However, graph matching faces two challenges: high computational complexity and loss information of graph embedded. In order to solve these two problems and correctly identify the similarity between two function control flow graphs, we propose a binary function similarity comparison method based on Graph Mover's Distance. The model takes two graphs as input, and first learns the spatial structure of the nodes in the graph based on the graph attention neural network to form the node embedding. Then we use the sum of the distance between the two sets of node embeddings to represent the distance between the graphs as the graph similarity index. We conduct experiments on the control flow graph of the program and prove the effectiveness of the method.