Network inference with infection frequency matrix by improved Bayesian method

Most network inference methods based on epidemic spreading models rely on binary-state time series to reconstruct the underlying network structure. However, because binary-state time series only qualitatively describe node states and lack quantitative information on infection histories, accurate network reconstruction typically requires extensive iterative computation and suffers from low efficiency. To overcome this limitation, this work proposes a Bayesian network inference approach that converts binary-state time series into an infection frequency matrix encoding pairwise infection events and uses the resulting likelihood to jointly infer the contact network and transmission-related parameters. This infection frequency representation reduces computational complexity, improves inference accuracy, and enables fast, high-fidelity reconstruction of contact networks, providing a principled basis for optimizing intervention strategies in biological, social, and cyber-physical systems.