A network inference problem focuses on discovering the structure of a diffusion network from observed cascades.This problem is significantly more challenging in several settings in which this type of an inference is desirable or necessary because of heterogeneity in the diffusion process.The heterogeneity of the diffusion process in different life stages results in the inaccuracy of a common assumption of constant influence strength.In this study,a Life Stage Heuristic (LSH) method is proposed to model life stage heterogeneity by decoupling the popularity level of an item under propagation from a true strength of social ties to improve inference accuracy.The proposed LSH is incorporated into almost all existing state-of-the-art network inference algorithms to improve estimation accuracy with only minimal changes in the implementation and maintaining the same running time.Additionally,NetRate,NetInf,and ConNIe are used as three examples to demonstrate the power of the proposed method.Furthermore,clustering of cascades prior to the LSH is proposed to eliminate noise,and the optimized method is termed as Clustered Life Stage Heuristic (CLSH).Extensive experiments on synthetic and real world datasets indicate that both LSH and CLSH methods significantly improve the accuracy of network inference.