A declarative approach to constrained community detection

Community detection in the presence of prior information or preferences on solution properties is called semi-supervised or constrained community detection. The task of embedding such existing kinds of knowledge effectively within a community discovery algorithm is challenging. Indeed existing approaches are not flexible enough to incorporate a variety of background information types. This paper provides a framework for semi-supervised community detection based on constraint programming modelling technology for simultaneously modelling different objective functions such as modularity and a comprehensive range of constraint types including community level, instance level, definition based and complex logic constraints. An advantage of the proposed framework is that, using appropriate solvers, optimality can be established for the solutions found. Experiments on real and benchmark data sets show strong performance and flexibility for our proposed framework.