Skip to content

Network analysis of misinformation

Network analysis studies the topology and dynamics of information diffusion networks, treating social media users and their interactions (retweets, shares, mentions) as a graph. This approach reveals structural patterns in how misinformation spreads: the role of influencers, clustering of communities, and core-periphery organization.

Key network concepts

  • k-core decomposition: Recursive removal of low-degree nodes to identify nested "shells" of increasing connectivity. Applied to misinformation networks to reveal dense cores dominated by particular user types.
  • Network core: The densest, most tightly-connected subset of nodes. Often populated by high-volume spreaders and automated accounts.
  • Polarization: Community structure where nodes within clusters are highly connected but between-cluster connections are sparse. Misinformation and fact-checking often form distinct communities with minimal interaction.
  • Centrality measures: Node importance measured by in-degree, out-strength (weighted out-degree), betweenness (bridging role), PageRank (eigenvector-based importance).
  • Information cascades: Temporal sequences of retweets/shares tracing how a claim or article spreads through the network.

Key papers and datasets in this wiki

Foundational: - Fortunato, S. (2009) — Community detection in graphs — Comprehensive survey of algorithms and theory for partitioning networks into communities; covers modularity optimization, spectral methods, and applications to biological, social, and technological networks; provides theoretical grounding for network-based misinformation analysis.

Survey and framework: - Shu, Bernard & Liu (2018) — Studying Fake News via Network Analysis: Detection and Mitigation — comprehensive chapter surveying network properties (echo chambers, filter bubbles, malicious accounts), three homogeneous and three heterogeneous network types, feature learning via network embeddings (NMF, RNNs), detection methods (interaction embedding, temporal diffusion, credibility propagation, knowledge network matching), and mitigation strategies (provenance identification, leader selection, influence minimization)

Dataset paper: - Shu et al. (2018) — FakeNewsNet — Multi-dimensional repository with rich network-level features: follower/following distributions, network cascades, user interaction graphs, and social network topology for both PolitiFact and GossipCop data.

Analysis and methods: - Detecting and Tracking the Spread of Astroturf Memes in Microblog Streams — Pioneering network analysis approach to detect astroturfing; computes diffusion network statistics (degree, strength, clustering, injection points, connected component size) and shows network topology is more discriminative than sentiment for identifying deceptive campaigns - Shao et al. (2018) — Anatomy of an online misinformation network — Network analysis of fact-checking vs. misinformation spread during 2016 US election; k-core decomposition reveals strong segregation between communities and that fact-checking disappears in the denser core; proposes robustness analysis and node-removal strategies - Vosoughi et al. (2017) — The spread of true and false news online — Large-scale temporal analysis of 126,000 cascades over 10 years; false news spreads faster and wider than true news; studies network properties and retweeter behavior - Shao et al. (2017) — The spread of low-credibility content by social bots — Network analysis of bot-driven amplification; bots occupy central positions and drive cascade shape

Connections