Network Based Misinformation Detection

Network-based misinformation detection frames the problem as inference of information reliability through the structure of social networks and information sharing patterns, rather than sole reliance on content analysis. The core insight is that users' collective sharing behavior—which sources co-occur, which domains cluster together—encodes implicit signals about source credibility that can be exploited for automated assessment.

Key approaches

Domain co-occurrence networks: Aggregate sharing patterns at the domain level to construct networks where nodes are news sources and edges reflect how often they co-occur in shared content. Domain reliability often exhibits assortative mixing—unreliable sources cluster together, high-quality sources cluster separately—a pattern that can be learned by graph neural networks even without content analysis.

Propagation graph structure: Analyze how misinformation spreads through social networks. Fake news often exhibits different propagation patterns than true news (faster initial spread, deeper trees, different user engagement), and these structural features can be used for classification via GNNs.

User-source networks: Model bipartite graphs of users and sources, capturing user behavior patterns (which sources each user shares, user credibility inferred from their sharing history) and source characteristics jointly.

Key papers

• Network Information Enhances Unreliable News Domain Detection — Shows that Telegram domain co-sharing networks exhibit assortative mixing by reliability; GraphSAGE on domain networks achieves 0.63 accuracy and provides complementary signals to content analysis.
• Network-based Fake News Detection: A Pattern-driven Approach — Pattern-driven fake news detection using network representation learning.
• Hierarchical Propagation Networks for Fake News Detection: Investigation and — Hierarchical attentive heterogeneous graph learning for fake news detection.
• FakeNewsNet: A Data Repository with News Content, Social Context and Spatiotemporal — FakeNewsNet: A data repository for studying news content and social context; includes propagation and network analysis tools.
• Rumor Detection on Social Media with Bi-Directional Graph Convolutional Networks — Bidirectional GCN for rumor detection exploiting both top-down and bottom-up propagation structure.

Connections

• Graph Neural Networks (methodological foundation)
• Misinformation spread and diffusion (phenomenon being analyzed)
• Credibility assessment for fake news detection (complementary perspective)