Year of Graduation
Felipe javier Vaca ramirez
Urban Networks Clustering Using Graph Embeddings
Applied Statistics with Network Analysis
The most recent approaches for learning representations of graphs have been motivated for achieving higher efficiency on graph algorithms and overcoming feature engineering problems. One of such approaches consists on graph embedding methods, which are representations in a vector space that try to preserve the graph properties. Despite these methods have been already used in several areas (e.g., Biology, Chemistry, and Linguistics) for machine learning tasks (e.g., link prediction, graph classification), other application scenarios are yet to be explored. Thus, this work proposes an application of graph embeddings in urban morphology. Specifically, we use “graph2vec” and “Anonymous Walk Embeddings” algorithms for analysing a set of urban networks. Our results suggest that the obtained vector representations allow to distinguish tree-like networks from grid-like networks. The differences in dead-ends culs-de-sac predominance and other network patterns support these results. Furthermore, this kind of partition is in the line with one of the taxonomy tendencies existing in the literature, which are partially rooted in network evolution processes.