Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Triple Path Enhanced Neural Architecture Search for Multimodal Fake News Detection

Bo Xu, Qiujie Xie, Jiahui Zhou, Linlin Zong

TL;DR

The paper tackles multimodal fake-news detection under partial modality scenarios by introducing MUSE, a triple-path neural architecture search framework. It combines two dynamic NAS-driven fusion/transformation paths with a static auxiliary path to robustly fuse text and image features, using continuous relaxation over operator sets and discrete selection to adapt its architecture to data quality. The final prediction fuses outputs from all three paths with learnable weights and optimizes via binary cross-entropy, backed by extensive ablations showing the importance of each path and operator choice. Empirical results on WEIBO and PHEME demonstrate strong performance—especially under missing-modality conditions—and validate the method's ability to generalize across datasets with varying data quality and modality availability.

Abstract

Multimodal fake news detection has become one of the most crucial issues on social media platforms. Although existing methods have achieved advanced performance, two main challenges persist: (1) Under-performed multimodal news information fusion due to model architecture solidification, and (2) weak generalization ability on partial-modality contained fake news. To meet these challenges, we propose a novel and flexible triple path enhanced neural architecture search model MUSE. MUSE includes two dynamic paths for detecting partial-modality contained fake news and a static path for exploiting potential multimodal correlations. Experimental results show that MUSE achieves stable performance improvement over the baselines.

Triple Path Enhanced Neural Architecture Search for Multimodal Fake News Detection

TL;DR

The paper tackles multimodal fake-news detection under partial modality scenarios by introducing MUSE, a triple-path neural architecture search framework. It combines two dynamic NAS-driven fusion/transformation paths with a static auxiliary path to robustly fuse text and image features, using continuous relaxation over operator sets and discrete selection to adapt its architecture to data quality. The final prediction fuses outputs from all three paths with learnable weights and optimizes via binary cross-entropy, backed by extensive ablations showing the importance of each path and operator choice. Empirical results on WEIBO and PHEME demonstrate strong performance—especially under missing-modality conditions—and validate the method's ability to generalize across datasets with varying data quality and modality availability.

Abstract

Multimodal fake news detection has become one of the most crucial issues on social media platforms. Although existing methods have achieved advanced performance, two main challenges persist: (1) Under-performed multimodal news information fusion due to model architecture solidification, and (2) weak generalization ability on partial-modality contained fake news. To meet these challenges, we propose a novel and flexible triple path enhanced neural architecture search model MUSE. MUSE includes two dynamic paths for detecting partial-modality contained fake news and a static path for exploiting potential multimodal correlations. Experimental results show that MUSE achieves stable performance improvement over the baselines.
Paper Structure (14 sections, 12 equations, 2 figures, 5 tables)

This paper contains 14 sections, 12 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Method
  3. General Framework
  4. Multimodal Feature Extraction
  5. Multimodal Feature Fusion
  6. Dynamic Path in MUSE
  7. Static Path in MUSE
  8. Fake News Prediction
  9. Experiments And Results
  10. Experimental Setup
  11. Performance Comparison
  12. Ablation Experiments for Operators
  13. Ablation Experiments for Paths
  14. Conclusion

Figures (2)

  • Figure 1: Example for various types of fake news.
  • Figure 2: Overall architecture of the MUSE model.