Title:Enhancing Graph Topology and Clustering Quality: A Modularity-Guided Approach

Abstract:Current modularity-based community detection algorithms attempt to find cluster memberships that maximize modularity within a fixed graph topology. Diverging from this conventional approach, our work introduces a novel strategy that employs modularity to guide the enhancement of both graph topology and clustering quality through a maximization process. Specifically, we present a modularity-guided approach for learning sparse graphs with high modularity by iteratively pruning edges between distant clusters, informed by algorithmically generated clustering results. To validate the theoretical underpinnings of modularity, we designed experiments that establish a quantitative relationship between modularity and clustering quality. Extensive experiments conducted on various real-world datasets demonstrate that our method significantly outperforms state-of-the-art graph construction methods in terms of clustering accuracy. Moreover, when compared to these leading methods, our approach achieves up to a hundredfold increase in graph construction efficiency on large-scale datasets, illustrating its potential for broad application in complex network analysis.