Inferring Information Diffusion Networks without Timestamps

The topology of diffusion networks plays an essential role in understanding information propagation dynamics and conducting social network analysis. However, diffusion networks are often unobservable in practical applications, leading to wide research on network inference from information cascades over the past decade. At present, novel cascades-based methods have been further developed to recover temporal dynamics and network topology by exploring the utilization of node temporal information, resulting in notable advancements. However, it requires high costs to acquire extensive temporal information, and the performance of network inference may decrease due to potential observational errors. Therefore, this paper specifically focuses on the time-independent scenario to address these limitations. Firstly, this paper models the node statuses of each diffusion process by leveraging the assumption of propagation trees based on the well-known independent cascade model. Subsequently, a gradient-based approach is developed to estimate the influences between nodes, facilitating the inference of network structure. Furthermore, this paper proposes a Monte Carlo EM-based approach to enhance the efficiency of network inference while maintaining comparable accuracy. Extensive experiments are conducted to verify the efficiency and effectiveness of our approaches on both synthetic and real-world networks.