Structure-aware Propagation Generation with Large Language Models for Fake News Detection

TL;DR

This work tackles fake news detection under incomplete propagation by introducing StruSP, a structure-aware synthetic propagation framework that combines bidirectional evolutionary propagation (BEP) learning with structure-aware propagation enhancement (SPE). BEP jointly learns propagation structure via structure-aware hybrid sampling and masked propagation modeling, while SPE guides LLMs to extend propagation in a way that preserves real-world structural patterns. Empirical results on three public datasets show StruSP improves detection accuracy, enhances early detection, and strengthens cross-platform generalization, with ablations confirming the importance of each BEP component. The approach demonstrates that aligning semantic plausibility with realistic propagation structures yields more informative synthetic data and superior detection performance in practical settings.

Abstract

The spread of fake news on social media poses a serious threat to public trust and societal stability. While propagation-based methods improve fake news detection by modeling how information spreads, they often suffer from incomplete propagation data. Recent work leverages large language models (LLMs) to generate synthetic propagation, but typically overlooks the structural patterns of real-world discussions. In this paper, we propose a novel structure-aware synthetic propagation enhanced detection (StruSP) framework to fully capture structural dynamics from real propagation. It enables LLMs to generate realistic and structurally consistent propagation for better detection. StruSP explicitly aligns synthetic propagation with real-world propagation in both semantic and structural dimensions. Besides, we also design a new bidirectional evolutionary propagation (BEP) learning strategy to better align LLMs with structural patterns of propagation in the real world via structure-aware hybrid sampling and masked propagation modeling objective. Experiments on three public datasets demonstrate that StruSP significantly improves fake news detection performance in various practical detection scenarios. Further analysis indicates that BEP enables the LLM to generate more realistic and diverse propagation semantically and structurally.

Figures (4)

• Figure 1: A comparison of LLM-generated propagation (e.g., DELL wan2024dell) and original propagation of the news "Donald Trump has been disqualified from running for president. "
• Figure 2: Overall framework of StruSP. We first perform bidirectional evolutionary propagation learning to capture the structural evolution of real propagation. We sample a set of propagation paths for both breadth-wise and depth-wise evolution based on the propagation tree, and then train the LLM to reconstruct masked nodes in forward and backward directions along the sampled path. Then, we generate synthetic propagation to enrich existing propagation for the downstream detection.
• Figure 3: Results of StruSP and comparison methods on early detection. RAGCL is the backbone model. Full refers to using all available propagation data for prediction, i.e., general detection. None refers to only using limited real-world propagation for early detection. DELL and StruSP use both limited real-world and LLM-generated propagation for early detection.
• Figure 4: Results of micro-level propagation analysis on the combined Twitter and PHEME5 datasets. The distributions of Structural Entropy, Instance-level Depth/Breadth, and Semantic Homogeneity across propagation generated by different methods