Enhancing Fake News Detection in Social Media via Label Propagation on Cross-modal Tweet Graph

TL;DR

This work tackles fake-news detection on social media by densifying the interaction graph with cross-modal connections derived from CLIP, addressing sparsity in traditional social-context graphs. It introduces FCN-LP, combining a Feature Contextualization Network with a signed-label Propagation Network to leverage positive and negative correlations among tweets. A domain-generalization loss based on Maximum Mean Discrepancy encourages feature consistency between seen and unseen events, improving generalization to new contexts. Evaluations on Twitter, PHEME, and Weibo show consistent improvements over state-of-the-art multimodal detectors, and ablations confirm the benefits of contextualization, sign-aware propagation, and domain-generalization.

Abstract

Fake news detection in social media has become increasingly important due to the rapid proliferation of personal media channels and the consequential dissemination of misleading information. Existing methods, which primarily rely on multimodal features and graph-based techniques, have shown promising performance in detecting fake news. However, they still face a limitation, i.e., sparsity in graph connections, which hinders capturing possible interactions among tweets. This challenge has motivated us to explore a novel method that densifies the graph's connectivity to capture denser interaction better. Our method constructs a cross-modal tweet graph using CLIP, which encodes images and text into a unified space, allowing us to extract potential connections based on similarities in text and images. We then design a Feature Contextualization Network with Label Propagation (FCN-LP) to model the interaction among tweets as well as positive or negative correlations between predicted labels of connected tweets. The propagated labels from the graph are weighted and aggregated for the final detection. To enhance the model's generalization ability to unseen events, we introduce a domain generalization loss that ensures consistent features between tweets on seen and unseen events. We use three publicly available fake news datasets, Twitter, PHEME, and Weibo, for evaluation. Our method consistently improves the performance over the state-of-the-art methods on all benchmark datasets and effectively demonstrates its aptitude for generalizing fake news detection in social media.

Figures (7)

• Figure 1: By utilizing cross-modal search, we can construct a more comprehensive connection to relevant tweets that may contain potential positive and negative correlations. These connections can be exploited to enable more credible detection of fake news.
• Figure 2: An overview of FCN-LP for fake news detection.
• Figure 3: Parameter sensitivity analysis for similarity threshold $\tau$: (a) accuracy curves with different similarity thresholds; (b) the average number of connected (relevant) tweets per node (Avg. CT / node).
• Figure 4: Accuracy values obtained for various loss balancing hyper-parameters $\lambda$ and $\mu$ on the Twitter dataset.
• Figure 5: 2D t-SNE plot of the cross-modal tweet graph constructed on the PHEME dataset. The various markers signify distinct events within the dataset. A red border indicates the tweet is real, while a black border denotes a fake tweet.