Feature Extraction and Representation of Urban Road Networks Based on Travel Routes

摘要: Extraction of traffic features constitutes a key research direction in safety planning. In previous tasks, road network are extracted manually. contrast, Network Representation Learning aims to automatically learn low-dimensional node representations. Enlightened by feature learning Natural Language Processing, representation urban nodes is studied as supervised task this paper. Following line thinking, deep framework, called StreetNode2VEC, proposed for representations the based on travel routes, and then model parameter calibration performed. We explain effectiveness from visualization, similarity analysis, link prediction. naturally present clustered pattern, different clusters correspond regions network. Meanwhile, still retain their spatial information analysis. The method StreetNode2VEC obtains AUC score 0.813 prediction, which greater than that obtained Graph Convolutional (GCN) Node2vec. This suggests can be used effectively credibly predict whether should established between two nodes. Overall, our work provides new way representing network, have potential planning field.