Abstract: Community detection algorithms are critical research tools in the field of complex networks. However, traditional community detection genetic algorithms have the problems with poor initial population quality and low computational efficiency under large-scale networks. To address this, an improved community detection genetic algorithm based on matrix computation acceleration is proposed. To tackle the problem of subpar initial population quality, a novel initialization operator is proposed to construct high-quality initial communities using the closure coefficient with biases node selection. To address the issue of low computational efficiency, the traditional community detection genetic algorithm operators are restructured based on matrix operations, enabling the use of GPU acceleration to enhance computational efficiency. Experimental results indicate that the proposed algorithm maintains good partitioning accuracy and demonstrates higher computational efficiency than other algorithms of the same type under different scales of real networks and LFR synthetic networks.

Key words: complex network, community detection, genetic algorithm, matrix operation, modularity