Research map / Vision-language and multimodal / Cross-modal learning
External Reliable Information-enhanced Multimodal Contrastive Learning for Fake News Detection
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes ERIC-FND, a multimodal fake news detector that adds Wikipedia entity-description enhancement, contrastive image-text alignment with cross-modal attention, and adaptive feature fusion, reporting gains on Weibo and X.
Problem
Multimodal fake news detection struggles to fully exploit text-image information and relies on external information that is either unreliable or static.
Approach
Enriches news text with first-sentence Wikipedia descriptions of extracted entities via attention, aligns text and image features with InfoNCE contrastive loss, applies cross-modal attention interaction, and adaptively weights text, interaction and image features before classification.
Key results
- On Weibo, ERIC-FND reaches 0.946 accuracy versus 0.922 for KMAGCN(bert), the strongest baseline reported.
- On X, ERIC-FND reaches 0.945 accuracy versus 0.809 for MCAN, the best baseline reported.
- Ablation on Weibo: removing the external information module drops accuracy from 0.946 to 0.926; removing the multimodal interaction module drops it to 0.925.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the adaptive fusion-based classification module (ablation) | ERIC-FND · Weibo | accuracy: 0.938 -> 0.946 | Helped |
| add the multimodal information interactive enhancement module (ablation) | ERIC-FND · Weibo | accuracy: 0.925 -> 0.946 | Helped |
| add the external information enhancement module (ablation) | ERIC-FND · Weibo | accuracy: 0.926 -> 0.946 | Helped |
| add the adaptive fusion-based classification module (ablation) | ERIC-FND · Weibo and X | accuracy: 0.8% | Helped |
| add the multimodal information interactive enhancement module (ablation) | ERIC-FND · Weibo | accuracy: 2.1% | Helped |
| add the multimodal information interactive enhancement module (ablation) | ERIC-FND · X | accuracy: 1.6% | Helped |
| add the external information enhancement module (ablation) | ERIC-FND · Weibo | accuracy: 2% | Helped |
| add the external information enhancement module (ablation) | ERIC-FND · X | accuracy: 3.1% | Helped |
Surprising findings
- The reported X accuracy gain (0.945 vs 0.809) is far larger than typical gains on this benchmark, and the paper gives no explanation for it.
Limitations
- The paper does not discuss limitations.
- Reproducibility details are deferred to an appendix that is not included in the provided text.
Experimental setup
Datasets: Weibo X (MediaEval)
Benchmarks: Weibo X
Metrics: accuracy precision recall F1-score
Baselines: att-RNN EANN MVAE CAFE MCAN MRML KMAGCN
Builds on: BERT ResNet-50 InfoNCE KMAGCN
Closest papers by meaning
- Reinforced Adaptive Knowledge Learning for Multimodal Fake News Detection (AAAI 2024 · similarity 0.91)
- DAMMFND: Domain-Aware Multimodal Multi-view Fake News Detection (AAAI 2025 · similarity 0.90)
- Learning Complex Heterogeneous Multimodal Fake News via Social Latent Network Inference (AAAI 2025 · similarity 0.89)
- Each Fake News Is Fake in Its Own Way: An Attribution Multi-Granularity Benchmark for Multimodal Fake News Detection (AAAI 2025 · similarity 0.88)
- RaCMC: Residual-Aware Compensation Network with Multi-Granularity Constraints for Fake News Detection (AAAI 2025 · similarity 0.86)
- Bad Actor, Good Advisor: Exploring the Role of Large Language Models in Fake News Detection (AAAI 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.