Information diffusion in social networks

Information diffusion studies how ideas, news, and claims spread through social networks via sharing, retweeting, forwarding, and other propagation mechanisms. A core insight is that what gets shared and how fast it spreads depends on properties of both the message and the network structure.

Key dimensions of diffusion research:

Message properties — The linguistic, emotional, and factual characteristics of content affect spread. Moral-emotional language (combining moral and emotional content) increases diffusion by ~20% per word. Emotional arousal, sentiment polarity, novelty, and controversy all influence virality.

Network structure — Information spreads differently depending on network topology, homophily (in-group clustering), bridge density between communities, and user influence (follower counts, centrality). Messages spread faster within ideologically coherent groups than across group boundaries, contributing to polarized "echo chambers."

Behavioral dynamics — Retweet patterns, sharing thresholds, and cascade dynamics determine how far information penetrates. First movers and high-influence users (accounts with many followers) disproportionately determine reach.

Key papers

• Rumor Cascades — Large-scale empirical study of 16.7K rumor cascades on Facebook
• A Survey of Information Cascade Analysis: Models, Predictions, and Recent Advances — Comprehensive survey of information cascade modeling and prediction; covers 250+ papers systematizing feature-based, generative, and deep learning approaches; analyzes temporal, structural, user/item, and content features; compares prediction strategies and methodologies.
• Shu, Bernard & Liu (2018) — Studying Fake News via Network Analysis — detailed treatment of temporal dimension of news dissemination; temporal diffusion networks model user engagement sequences; proposes RNN-based methods for learning temporal representations of how individual users engage with news over time
• Detecting and Tracking the Spread of Astroturf Memes in Microblog Streams — Analyzes diffusion patterns of political memes on Twitter; shows network topology is highly predictive of deceptive vs. organic spread; foundational work on using diffusion structure for misinformation detection
• Can Cascades be Predicted? — Predicting cascade growth from early observations; shows temporal features most predictive (~80% accuracy)
• DeepCas: an End-to-end Predictor of Information Cascades — Deep learning approach to cascade prediction; learns cascade graph representations via random walks and attention mechanisms without hand-crafted features
• Topological Recurrent Neural Network for Diffusion Prediction — Topological RNN for diffusion prediction; models cascades as dynamic DAGs and learns topology-aware node embeddings; 20–56% improvement over prior deep learning methods
• Cinelli et al. (2020) — The COVID-19 Social Media Infodemic — Applies epidemic models (EXP and SIR) to characterize platform-specific information reproduction numbers; finds all major platforms exhibit infodemic conditions (R₀ > 1)
• Cinelli et al. (2021) — The echo chamber effect on social media — Analyzes information diffusion bias across platforms using SIR epidemic models; shows that messages on Facebook and Twitter propagate preferentially within ideological in-groups while Reddit exhibits unbiased diffusion toward all political leanings
• Brady et al. (2017) — Emotion shapes the diffusion of moralized content: demonstrates moral-emotional language increases retweet rates by ~20% per word across gun control, same-sex marriage, and climate change; shows stronger diffusion within political in-groups than out-groups.

Connections

• Social networks and online communities — the infrastructure through which diffusion occurs
• Emotional language in online discourse — message properties that amplify spread
• Political polarization and ideological echo chambers — diffusion asymmetries within vs. between ideological groups contribute to polarization
• Misinformation spread and diffusion — how false claims specifically propagate