A Novel Label Propagation Method for Community Detection Based on Game Theory

Community is a mesoscopic feature of the multi-scale phenomenon complex networks, which bridge to revealing formation and evolution networks. Due high computational efficiency, label propagation becomes topic considerable interest within community detection, but its randomness yet produces serious fluctuations. Facing inherent flaws propagation, this paper proposes series solutions. Firstly, presents heuristic algorithm named Label Propagation Algorithm use Cliques Weight (LPA-CW). In algorithm, labels are expanded from seeds propagated based on node linkage index. Seeds produced complete subgraph, index related neighboring nodes. This method can produce competitive modularity Q not Normalized Mutual Information (NMI), compensate with existing methods, such as Stepping Detection Similarity (LPA-S). Secondly, in order combine advantages different algorithms, introduces game theory framework, design profit function participant algorithms attain Nash equilibrium, build an integration model for detection (IA-GT). Thirdly, above model, IA-GT (LPA-CW-S), integrates LPA-CW LPA-S solves incompatibility between NMI. Fully tested both computer-generated real-world gives better results indicators NMI than effectively resolving contradiction theoretical real community. Moreover, significantly reduces runs faster.