A multi-criteria algorithm for automatic detection of city communities

Characterizing urban communities is essential for understanding citizens' needs and neighborhood-wise dynamics. Discriminating factors are population mobility patterns, neighborhood structural characteristics, and distance to other areas of the city. Available approaches focus on one aspect and, often, suffer from isolated nodes and excessive geographical fragmentation of solutions. For these reasons, we formulate the problem of urban community clustering considering all three aspects and provide an algorithm that combines hierarchical aggregation with node adjustment and relocation. We evaluate our approach on a real-world data set and the obtained results show its efficacy. Finally, we also show the importance of using map embedding for characterizing neighborhood from the structural standpoint.