EPOC: A Survival Perspective Early Pattern Detection Model for Outbreak Cascades

The past few decades have witnessed the booming of social networks, which leads to a lot of researches exploring information dissemination. However, owing to the insufficient information exposed before the outbreak of the cascade, many previous works fail to fully catch its characteristics, and thus usually model the burst process in a rough manner. In this paper, we employ survival theory and design a novel survival perspective Early Pattern detection model for Outbreak Cascades (in abbreviation, EPOC), which utilizes information both from the static nature and its later diffusion process. To classify the cascades, we employ two Gaussian distributions to get the optimal boundary and also provide rigorous proof to testify its rationality. Then by utilizing both the survival boundary and hazard ceiling, we can precisely detect early pattern of outbreak cascades at very early stage. Experiment results demonstrate that under three practical and special metrics, our model outperforms the state-of-the-art baselines in this early-stage task.