A degree-corrected stochastic block model for community discovery in signed networks with heterogeneous degree distributions

• DCS2BM for signed networks with heterogeneous degree distributions. • Explicitly captures the underlying generative process of community structures. • Learning algorithm with simultaneous model selection and parameter estimation. • Enhancing scalability for analyzing large-scale networks with complex degree patterns. Community discovery in signed networks has garnered considerable attention as a key topic in network science. However, existing methods often struggle with heterogeneous degree distributions and face scalability issues due to high computational complexity-particularly when the number of communities is unknown. To overcome these challenges, we propose the degree-corrected signed stochastic block model (DCS2BM), a probabilistic framework designed to accurately detect community structures in signed networks, with varying degree distributions. We integrate the minimum message length criterion with a component-wise expectation-maximization algorithm to develop a scalable approach that performs parameter estimation and model selection. Experimental results on both synthetic and real-world datasets demonstrate that DCS2BM outperforms state-of-the-art methods in both accuracy and efficiency. On synthetic networks with heterogeneous degrees, our model achieves an average NMI (normalized mutual information) of 0.99-exceeding the best baseline by 4.2 %. Furthermore, for large-scale networks with up to 40,000 nodes, DCS2BM maintains consistently high performance ( N M I > 0.99 ) while reducing computation time by 87.4 % compared to traditional approaches.