Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

When GenAI Meets Fake News: Understanding Image Cascade Dynamics on Reddit

Saumya Chauhan, Mila Hong, Maria Vazhaeparambil

TL;DR

This study investigates how misinformation and GenAI-generated images propagate on Reddit by building a multimodal diffusion framework that combines textual sentiment, visual markers, and diffusion dynamics. It reconstructs repost cascades across five ideologically diverse communities using DSU clustering and canonical repost graphs, and evaluates virality prediction at post- and cascade-level with SHAP-based interpretability. The results show post-level accuracy of AUC 0.83 and cascade-level accuracy of AUC 0.998 when combining content and diffusion features, with mixed-flag content driving the most expansive cascades. These findings provide actionable guidance for moderating synthetic and misleading visuals by linking content signals to diffusion patterns and cross-community spread.

Abstract

AI-generated content and misinformation are increasingly prevalent on social networks. While prior research primarily examined textual misinformation, fewer studies have focused on visual content's role in virality. In this work, we present the first large-scale analysis of how misinformation and AI-generated images propagate through repost cascades across five ideologically diverse Reddit communities. By integrating textual sentiment, visual attributes, and diffusion metrics (e.g., time-to-first repost, community reach), our framework accurately predicts both immediate post-level virality (AUC=0.83) and long-term cascade-level spread (AUC=0.998). These findings offer essential insights for moderating synthetic and misleading visual content online.

When GenAI Meets Fake News: Understanding Image Cascade Dynamics on Reddit

TL;DR

This study investigates how misinformation and GenAI-generated images propagate on Reddit by building a multimodal diffusion framework that combines textual sentiment, visual markers, and diffusion dynamics. It reconstructs repost cascades across five ideologically diverse communities using DSU clustering and canonical repost graphs, and evaluates virality prediction at post- and cascade-level with SHAP-based interpretability. The results show post-level accuracy of AUC 0.83 and cascade-level accuracy of AUC 0.998 when combining content and diffusion features, with mixed-flag content driving the most expansive cascades. These findings provide actionable guidance for moderating synthetic and misleading visuals by linking content signals to diffusion patterns and cross-community spread.

Abstract

AI-generated content and misinformation are increasingly prevalent on social networks. While prior research primarily examined textual misinformation, fewer studies have focused on visual content's role in virality. In this work, we present the first large-scale analysis of how misinformation and AI-generated images propagate through repost cascades across five ideologically diverse Reddit communities. By integrating textual sentiment, visual attributes, and diffusion metrics (e.g., time-to-first repost, community reach), our framework accurately predicts both immediate post-level virality (AUC=0.83) and long-term cascade-level spread (AUC=0.998). These findings offer essential insights for moderating synthetic and misleading visual content online.

Paper Structure

This paper contains 25 sections, 1 equation, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Our Contribution
  3. Methods
  4. Data Collection & Labeling
  5. Safe Search
  6. Misinformation
  7. GenAI
  8. Feature Extraction & Virality Metric
  9. Textual Features (Post-Level)
  10. Visual Features (Post-Level)
  11. Metadata & User-Interaction Features
  12. Virality Attention Index (VAI)
  13. Cascade Construction (DSU Clustering)
  14. Canonical Repost Graph (For Each Cascade)
  15. Structural Virality (Wiener Index)
  16. ...and 10 more sections

Figures (3)

  • Figure 1: Overall Architecture Pipeline
  • Figure 2: Experimental Setup Overview
  • Figure 3: Top 5 Mean SHAP Contribution for Virality Prediction at the (a) Post-level (b) Cascade-level (image, text) (c) Cascade-level (image, text, cascade-dynamics). GenAI and misinformation counts per cascade are by far the strongest virality predictors, outpacing any other image or text feature. Cascades rich in pseudo-labeled content spread faster, persist longer, and cross community boundaries, posing a moderation challenge within Reddit communities.