Criterions for locally dense subgraphs

Community detection is one of the most investigated problems in the field of complex networks. Although several methods were proposed, there is still no precise definition of communities. As a step towards a definition, I highlight two necessary properties of communities, separation and internal cohesion, the latter being a new concept. I propose a local method of community detection based on two-dimensional local optimization, which I tested on common benchmarks and on the word association database.