Abstract
As the massive size of contemporary social networks poses a serious challenge to the scalability of traditional graph clustering algorithms and the evaluation of discovered communities, we develop, in this manuscript, an approach used to discover hierarchical community structure in large networks. The introduced hybrid technique combines the strengths of bottom-up hierarchical clustering method with that of top-down hierarchical clustering. In fact, the first approach is efficient in identifying small clusters, while the second one is good at determining large ones. Our mixed hierarchical clustering technique, based on the assumption that there exists an initial solution composed of k classes and the combination of the two previously mentioned methods, does not the change of the number of partitions, modifies the repartition of the initial classes. At the end of the introduced clustering process, a fixed point, representing a local optimum of the cost function which measures the degree of importance between two partitions, is obtained. Consequently, the introduced combined model leads to the emergence of local community structure. To avoid this local optimum and detect community structure converged to the global optimum of the cost function, the detection of community structures, in this study, is not considered only as a clustering problem, but as an optimization issue. Besides, a novel mixed hierarchical clustering algorithm based on swarms intelligence is suggested for identifying community structures in social networks. In order to validate the proposed method, performances of the introduced approach are evaluated using both real and artificial networks as well as internal and external clustering evaluation criteria.