Research map / Label-efficient and robust learning / Recognition and long-tail
DAMMFND: Domain-Aware Multimodal Multi-view 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 DAMMFND, a domain-aware multimodal multi-view fake news detector that disentangles domain and semantic features, gates domain information per view, and weights views with a domain-enhanced transformer.
Problem
Multi-domain multimodal fake news detection suffers from inaccurate domain identification from predefined labels, imbalanced data across domains causing negative transfer, and domain-specific differences in how much each modality contributes to veracity decisions.
Approach
Disentangles domain-invariant and domain-specific features per modality via soft masks with domain classification and uniformity losses, filters domain features through per-view gates (inspired by ERASE), and uses a domain-aware transformer to predict per-view decision weights for text, image, and multimodal views.
Key results
- DAMMFND reaches 94.4% accuracy on Weibo and 94.7% on Weibo21, outperforming the best multi-domain baseline MMDFND (93.4% on Weibo) and multimodal baseline MRHFR (92.2% / 93.2%).
- Ablation shows removing the Domain-Enhanced Multi-view Decision Layer drops accuracy by 2.5 points on Weibo and 2.8 on Weibo21.
- Removing domain disentanglement lowers accuracy from 94.4% to 92.7% on Weibo.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use DAMMFND instead of MMDFND | multimodal multi-domain fake news detection · Weibo | F1: 0.934 -> 0.943 | Helped |
| use DAMMFND instead of MMDFND | multimodal multi-domain fake news detection · Weibo-21 | F1: 0.939 -> 0.947 | Helped |
| add Multi-channel Features Aggregation (ablation) | DAMMFND · Weibo | accuracy: 0.934 -> 0.944 | Helped |
| add Multi-channel Features Aggregation (ablation) | DAMMFND · Weibo-21 | accuracy: 0.938 -> 0.947 | Helped |
| add Domain Disentanglement (ablation) | DAMMFND · Weibo | accuracy: 0.927 -> 0.944 | Helped |
| add Domain Disentanglement (ablation) | DAMMFND · Weibo-21 | accuracy: 0.931 -> 0.947 | Helped |
| add domain information filtering in DAMVD (ablation) | DAMMFND · Weibo | F1 score (Real News): 0.933 -> 0.945 | Helped |
| add domain information filtering in DAMVD (ablation) | DAMMFND · Weibo-21 | F1 score (Real News): 0.937 -> 0.948 | Helped |
| retain the three modality views in DAMVD (ablation) | DAMMFND · Weibo | F1 score (Real News): 0.932 -> 0.945 | Helped |
| retain the three modality views in DAMVD (ablation) | DAMMFND · Weibo-21 | F1 score (Real News): 0.934 -> 0.948 | Helped |
| add Domain-Enhanced Multi-view Decision Layer (ablation) | DAMMFND · Weibo | accuracy: 0.924 -> 0.944 | Helped |
| add Domain-Enhanced Multi-view Decision Layer (ablation) | DAMMFND · Weibo-21 | accuracy: 0.922 -> 0.947 | Helped |
Limitations
- Relies on predefined domain labels for supervision of domain classification, which the paper itself argues are imperfect.
- Evaluated only on two Chinese social media datasets (Weibo, Weibo21) with text and image modalities.
- Weibo-based results have limited reported variance or significance testing.
Experimental setup
Datasets: Weibo Weibo21
Benchmarks: Weibo Weibo21
Metrics: accuracy precision F1 AUC per-domain F1
Baselines: MMoE MoSE MDFEND M3DFEND KATMF MMDFND SpotFake CAFE CMC BMR MRHFR
Builds on: MDFEND M3DFEND MMDFND KATMF ERASE
Closest papers by meaning
- External Reliable Information-enhanced Multimodal Contrastive Learning for Fake News Detection (AAAI 2025 · similarity 0.90)
- Each Fake News Is Fake in Its Own Way: An Attribution Multi-Granularity Benchmark for Multimodal Fake News Detection (AAAI 2025 · similarity 0.88)
- Bad Actor, Good Advisor: Exploring the Role of Large Language Models in Fake News Detection (AAAI 2024 · similarity 0.88)
- Reinforced Adaptive Knowledge Learning for Multimodal Fake News Detection (AAAI 2024 · similarity 0.87)
- Text-Guided Fine-grained Counterfactual Inference for Short Video Fake News Detection (AAAI 2025 · similarity 0.87)
- RaCMC: Residual-Aware Compensation Network with Multi-Granularity Constraints for Fake News Detection (AAAI 2025 · similarity 0.87)
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.