Cascade Prediction

Cascade prediction is the problem of forecasting whether an information cascade (reshare, retweet, comment chain, or other spreading content) will grow further and to what extent. Given early observations of a cascade, the goal is to predict its eventual size, structure, shape, or dynamics using features from network topology, user behavior, temporal properties, and content characteristics.

Key questions

• How predictable is cascade growth from early observations?
• Which feature classes (content, temporal, structural, user-centric) are most informative?
• Do different content types or platforms exhibit different predictability?
• How does the observation window size affect prediction accuracy?
• Can cascade structure (depth, branching, virality) be predicted?

Key papers

• A Survey of Information Cascade Analysis: Models, Predictions, and Recent Advances — Comprehensive survey covering 250+ papers on information cascade analysis; provides taxonomy of prediction methods across feature-based, generative, and deep learning approaches with systematic evaluation of methodologies.
• Can Cascades be Predicted? — Foundational work showing cascades are predictable (~80% accuracy) and temporal features dominate.
• DeepCas: an End-to-end Predictor of Information Cascades — End-to-end deep learning approach using random walk graph representations and attention mechanisms; outperforms feature-based and embedding baselines on Twitter and citation cascades.
• Topological Recurrent Neural Network for Diffusion Prediction — Extends cascade prediction via dynamic DAG modeling; introduces diffusion topologies and Topo-LSTM architecture; achieves 20–56% relative improvement in MAP over DeepCas.
• Hierarchical Propagation Networks for Fake News Detection: Investigation and Exploitation
• Beyond News Contents: The Role of Social Context for Fake News Detection

Related topics

• Information diffusion in social networks (broader concept)
• Viral Spread
• Temporal Prediction