A generalized game theoretic framework for mining communities in complex networks

Since the community structure is able to reveal the potential law behind complex networks, mining hiding communities has gained particular attention from various applications. A variety of objective functions, such as Modularity, weighted clustering coefficient (WCC), etc., have been developed to characterize the cohesiveness of a community, and thus many community detection approaches are proposed by optimizing a predefined objective function. This paper offers a urgent study on how to integrate different objective functions into a generic framework, which aims to enhance the flexibility of expert systems that are designed to identify communities from complex networks. Specifically, we formulate the process of community detection as a strategic game and give a general form of utility function for each agent from the perspective of game theory. Furthermore, we prove that if the parameters in the generalized utility function can be specified carefully, the strategic game could well match a potential game and be able to converge to a pure Nash equilibrium. In addition, we choose some commonly used objective functions to match the generalized utility function and design a synchronous learning model to test the performance of different global models. Compared with existing approaches, experimental results on synthetic and real-world data sets demonstrate that the proposed model achieve higher accuracy and efficiency.