Information Cascade Popularity Prediction Using Structural and Temporal Features
TL;DR
The basic ideas of online diffusion are introduced by formalising cascade and global graphs, and the extraction of structural and temporal features is described, and the shift toward modelling stochastic cascade dynamics is highlighted.
Abstract
Prediction of the spread of information cascades can help explain how information spreads in society and can support multiple applications. Early models often used aggregated user data and linear approximations; more recent approaches use graph representations to capture complex spatio-temporal variations. This paper studies the development of popularity prediction models and introduces some types, such as feature-engineered statistical methods, advanced representation learning and generative frameworks. The basic ideas of online diffusion are introduced by formalising cascade and global graphs, and the extraction of structural and temporal features is described. An introduction to the new frameworks is provided here, including random walk models (DeepWalk, node2vec), graph convolutional networks (GCNs), and the newer CasDO (Cascade Diffusion and Neural ODEs), which introduces continuous-time dynamics to address irregular sampling and diffusion uncertainty. The above comparisons highlight the shift toward modelling stochastic cascade dynamics. Finally, the current limitations, such as cross-domain applicability, are mentioned, and the potential integration of large language models (LLMs) and context-aware generation in future research is also introduced.
