FMNV: A Dataset of Media-Published News Videos for Fake News Detection

2025-04-10

AI Summary

Existing fake news detection datasets predominantly rely on low-quality user-generated videos, failing to capture the societal harm posed by professionally fabricated videos—such as those disseminated by media organizations with political or viral intent. Method: To address this gap, we introduce FMNV, the first benchmark dataset specifically designed for detecting fake news in professionally produced media videos, covering four representative professional forgery patterns and proposing the first systematic taxonomy for media-level fake news video classification. We further design an LLM-driven authentic-video tampering generation strategy to enhance data diversity and propose FMNVD, a dual-stream collaborative attention model that fuses CLIP and Faster R-CNN features via cross-modal co-attention to strengthen inconsistency modeling. Results: Experiments demonstrate that FMNV significantly improves the generalization of mainstream models, while FMNVD outperforms baselines across multiple metrics, validating its high detection accuracy and robustness against high-harm professional forgeries.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Application Domains: Misinformation & Fake News
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Web Mining and Content Analysis: Web data provenance, reliability, and authenticity
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

News media, particularly video-based platforms, have become deeply embedded in daily life, concurrently amplifying risks of misinformation dissemination. Consequently, multimodal fake news detection has garnered significant research attention. However, existing datasets predominantly comprise user-generated videos characterized by crude editing and limited public engagement, whereas professionally crafted fake news videos disseminated by media outlets often politically or virally motivated pose substantially greater societal harm. To address this gap, we construct FMNV, a novel dataset exclusively composed of news videos published by media organizations. Through empirical analysis of existing datasets and our curated collection, we categorize fake news videos into four distinct types. Building upon this taxonomy, we employ Large Language Models (LLMs) to automatically generate deceptive content by manipulating authentic media-published news videos. Furthermore, we propose FMNVD, a baseline model featuring a dual-stream architecture integrating CLIP and Faster R-CNN for video feature extraction, enhanced by co-attention mechanisms for feature refinement and multimodal aggregation. Comparative experiments demonstrate both the generalization capability of FMNV across multiple baselines and the superior detection efficacy of FMNVD. This work establishes critical benchmarks for detecting high-impact fake news in media ecosystems while advancing methodologies for cross-modal inconsistency analysis.

Problem

• Detect fake news in professionally crafted media videos
• Address lack of datasets for media-published fake videos
• Improve multimodal fake news detection with LLMs

Innovation

• FMNV dataset with media-published news videos
• LLMs generate deceptive content automatically
• Dual-stream model with CLIP and Faster R-CNN