The paper introduces *struc2vec*, a novel framework for learning latent representations that capture the structural identity of nodes in a network. Structural identity refers to the symmetry in network structures where nodes are identified based on their relationships and the network structure itself, rather than their attributes or labels. *struc2vec* uses a hierarchy to measure node similarity at different scales and constructs a multilayer graph to encode structural similarities and generate structural context for nodes. The method is evaluated through numerical experiments on toy examples and real networks, demonstrating superior performance compared to state-of-the-art techniques such as *DeepWalk*, *node2vec*, and *RoX*. *struc2vec* is shown to better capture structural identity, even in the presence of noise, and outperforms other methods in classification tasks where node labels depend more on structural identity. The paper also discusses the scalability and optimizations of *struc2vec*, highlighting its potential for large-scale network analysis.The paper introduces *struc2vec*, a novel framework for learning latent representations that capture the structural identity of nodes in a network. Structural identity refers to the symmetry in network structures where nodes are identified based on their relationships and the network structure itself, rather than their attributes or labels. *struc2vec* uses a hierarchy to measure node similarity at different scales and constructs a multilayer graph to encode structural similarities and generate structural context for nodes. The method is evaluated through numerical experiments on toy examples and real networks, demonstrating superior performance compared to state-of-the-art techniques such as *DeepWalk*, *node2vec*, and *RoX*. *struc2vec* is shown to better capture structural identity, even in the presence of noise, and outperforms other methods in classification tasks where node labels depend more on structural identity. The paper also discusses the scalability and optimizations of *struc2vec*, highlighting its potential for large-scale network analysis.