Background: Complex systems involved in biochemistry, neuroscience, physics, engineering and social science are primarily studied and modeled through network structures. The connectivity patterns within these interaction networks are discovered through clustering-like techniques. Community discovery is a related problem to find patterns in networks. Objectives: Existing algorithms either try to find few large communities in networks; or try to partition network into small strongly connected communities; that too is time consuming and parameter-dependant. Method/Results: This paper proposes a chromatic correlation clustering method to discover small strong communities in an interaction network in heuristic manner to have low time complexity and a parameter free method. Comparison with other methods over synthetic data is done. Conclusion: Interaction networks are very large, sparse containing few small dense communities that can be discovered only through method specifically designed for the purpose.

No abstract

The social network can be viewed as a chromatic graph of relations and entities. Thus, the community discovery in social network is essentially a problem of chromatic correlation clustering. This paper aims to develop metrics to measure the performance of community discovery algorithms in view of nonoverlapping strong communities. Three performance metrics, namely, chromatic density (CD), chromatic cut ratio (CCR) and chromatic conductance (CC), were proposed for thorough analysis on the output quality of chromatic clustering algorithms. In addition, synthetic graph generator was developed to generate sparse networks with few dense and strong communities. Five algorithms for chromatic correlation clustering, i.e. CB, ICB, LCB, OCB and RECB, were evaluated by the proposed metrics. The evaluation shows that the RECB is the most suitable algorithm for the discovery of strong communities in social network. The research results shed important new light on the detection of communities in social network.