Organizations: Xinjiang University, Urumqi, China · Joint International Research Laboratory of Silk Road Multilingual Cognitive Computing, Xinjiang University, Urumqi, China · Sichuan University, Chengdu, China
Multimodal fake news detection is crucial for mitigating societal disinformation. Existing approaches attempt to address this by fusing multimodal features or leveraging Large Language Models (LLMs) for advanced reasoning. However, these methods suffer from serious limitations, including a lack of comprehensive multi-view judgment and fusion, and prohibitive reasoning inefficiency due to the high computational costs of LLMs. To address these issues, we propose LLM-Guided Multi-View Reasoning Distillation for Fake News Detection ( LLM-MRD), a novel teacher-student framework. The Student Multi-view Reasoning module first constructs a comprehensive foundation from textual, visual, and cross-modal perspectives. Then, the Teacher Multi-view Reasoning module generates deep reasoning chains as rich supervision signals. Our core Calibration Distillation mechanism efficiently distills this complex reasoning-derived knowledge into the efficient student model. Experiments show LLM-MRD significantly outperforms state-of-the-art baselines. Notably, it demonstrates a comprehensive average improvement of 5.19% in ACC and 6.33% in F1-Fake when evaluated across all competing methods and datasets. Our code is available at https://github.com/Nasuro55/LLM-MRD
Figures & tables
Figure 1: Comparison of our proposed ”teacher-student” framework with previous methods. (a) Traditional fusion methods (e.g., Attention, Add) lack deep contextual understanding and are easily misled by manipulations like image-text mismatching, leading to misjudgments. (b) Our approach employs an LLM teacher for deep multi-view reasoning and distills this knowledge to a efficient student model, enabling it to learn robust features for accurate classification.
Figure 2: LLM-MRD architecture overview. The student uses BERT [ 35 ] , MAE [ 34 ] , and CLIP [ 36 ] to encode textual, visual, and cross-modal view features. The LLM teacher generates multi-view reasoning chains. Calibration Distillation aligns student features with the teacher’s semantic space using projection layers before applying distillation losses. Calibrated features are integrated by Multi-View Fusion via cross-attention for prediction, while a Deep Supervision layer provides auxiliary guidance.
Weibo
Weibo-21
GossipCop
Category
Method
Acc
F1-Fake
F1-Real
AUC
Acc
F1-Fake
F1-Real
AUC
Acc
F1-Fake
F1-Real
AUC
Multimodal multi-domain methods
EANN [ 28 ]
0.827
0.829
0.825
0.873
0.870
0.862
0.875
0.894
0.864
0.594
0.920
0.852
FND-CLIP [ 29 ]
0.907
0.908
0.907
0.953
0.943
0.940
0.946
0.962
0.880
0.638
0.928
0.871
MIMoE-FND [ 30 ]
0.928
0.928
0.928
0.972
0.956
0.955
0.957
0.977
0.895
0.698
0.938
0.879
KEN [ 31 ]
0.935
0.935
0.934
0.967
0.935
0.937
0.932
0.971
0.881
0.646
0.928
0.873
Multimodal single-domain methods
RaCMC [ 24 ]
0.915
0.917
0.914
0.921
0.942
0.938
0.943
0.962
0.879
0.641
0.927
0.838
Table 1: Comparison of LLM-MRD with the latest multi-domain fake news detection methods from Weibo, Weibo-21, and GossipCop.
Model
Weibo
Weibo-21
GossipCop
Acc
F1-Fake
F1-Real
Acc
F1-Fake
F1-Real
Acc
F1-Fake
F1-Real
LLM-MRD
0.953
0.954
0.952
0.959
0.957
0.968
0.902
0.719
0.946
Iterative Calibration distillation
- w/o Ltext
0.929 ↓ 2.5%
0.924 ↓ 3.1%
0.920 ↓ 3.4%
0.937 ↓ 2.3%
0.933 ↓ 2.5%
0.938 ↓ 3.1%
0.881 ↓ 2.3%
0.672 ↓ 6.5%
0.915 ↓ 3.3%
- w/o Limage
0.925 ↓ 2.9%
0.918 ↓ 3.8%
0.922 ↓ 3.2%
0.923 ↓ 3.8%
0.935 ↓ 2.3%
0.937 ↓ 3.2%
0.874 ↓ 3.1%
0.674 ↓ 6.3%
0.913 ↓ 3.5%
- w/o Lcross
0.917 ↓ 3.8%
0.907 ↓ 4.9%
0.912 ↓ 4.2%
0.945 ↓ 1.5%
0.921 ↓ 3.8%
0.927 ↓ 4.2%
0.889 ↓ 1.4%
0.665 ↓ 7.5%
0.901 ↓ 4.8%
Table 2: Ablation study results showing performance drops for different model variants.
Figure 3: Analysis of hyperparameter sensitivity. This figure shows the impact of four different hyperparameters on the model’s F1-real score across three datasets.
Table 3: LLM-MRD explainability case study on two challenging examples from the Weibo21 test set.
Figure 4: T-SNE visualization of test set features. Same color dots indicate the same label.
Method
Params
Reason latency
Time Complexity
MIMoE-FND
397M
20.5ms
O(k⋅ne⋅(n2d))
KEN
382M
19.7ms
O(L⋅(n2d))
RaCMC
424M
18.8ms
O(L⋅(n2d))
GLPN-LLM
7.07B
158ms
O(Lllm⋅nllm2d)
LLM-MRD
358M
17.2ms
O(L⋅(nd))
Table 4: Comparative Reasoning efficiency analysis of LLM-MRD and baseline models
In recent years, multimodal multidomain fake news detection has garnered increasing attention. Nevertheless, this direction presents two significant challenges: (1) Failure to Capture Cross-Instance Narrative Consistency: existing models usually evaluate each news in isolation, fail to capture cross-instance narrative consistency, and thus struggle to address the spread of cluster based fake news driven by social media; (2) Lack of Domain Specific Knowledge for Reasoning: conventional models, which rely solely on knowledge encoded in their parameters during training, struggle to generalize to new or data-scarce domains (e.g., emerging events or niche topics). To tackle these challenges, we introduce Retrieval-Augmented Multimodal Model for Fake News Detection (RAMM). First, RAMM employs a Multimodal Large Language Model (MLLM) as its backbone to capture cross-modal semantic information from news samples. Second, RAMM incorporates an Abstract Narrative Alignment Module. This component adaptively extracts abstract narrative consistency from diverse instances across distinct domains, aggregates relevant knowledge, and thereby enables the modeling of high-level narrative information. Finally, RAMM introduces a Semantic Representation Alignment Module, which aligns the model's decision-making paradigm with that of humans - specifically, it shifts the model's reasoning process from direct inference on multimodal features to an instance-based analogical reasoning process. Extensive experimental results on three public datasets validate the efficacy of our proposed approach. Our code is available at the following link: https://github.com/li-yiheng/RAMM
Yiheng Li, Weihai Lu, Hanyi Yu +1
University of International Business and Economics · Beijing, China · Peking University +4
Multimodal fake news detection (MFND) aims to verify news credibility by jointly exploiting textual and visual evidence. However, real-world news dissemination frequently suffers from missing modality due to deleted images, corrupted screenshots, and similar issues. Thus, robust detection in this scenario requires preserving strong verification ability for each modality, which is challenging in MFND due to insufficient learning of the low-contribution modality and scarce unimodal annotations. To address this issue, we propose Head-wise Modality Specialization within Multimodal Large Language Models (MLLMs) for robust MFND under missing modality. Specifically, we first systematically study attention heads in MLLMs and their relationship with performance under missing modality, showing that modality-critical heads serve as key carriers of unimodal verification ability through their modality specialization. Based on this observation, to better preserve verification ability for the low-contribution modality, we introduce a head-wise specialization mechanism that explicitly allocates these heads to different modalities and preserves their specialization through lower-bound attention constraints. Furthermore, to better exploit scarce unimodal annotations, we propose a Unimodal Knowledge Retention strategy that prevents these heads from drifting away from the unimodal knowledge learned from limited supervision. Experiments show that our method improves robustness under missing modality while preserving performance with full multimodal input.
Kai Qian, Weijie Shi, Jiaqi Wang +7
Soochow University, China · Hong Kong University of Science and Technology, China · Computer Science and Engineering, The Chinese University of Hong Kong, China +5
The rapid dissemination of multimodal content has intensified the spread of fabricated news, presenting a substantial threat to social integrity. A formidable challenge for current detection systems is identifying misinformation related to novel events in zero-shot scenarios. Prevailing zero-shot methods typically assess news items in isolation via semantic matching, a strategy that fails to recognize the recycled disinformation tactics from past campaigns and lacks the sophisticated reasoning needed to identify subtle, cross-modal discrepancies. To surmount these deficiencies, we introduce MRAFnd, a novel Multimodal Retrieval-Augmented Framework for Zero-Shot Fake News Detection. MRAFnd emulates a collaborative team of analysts to verify news veracity. The framework initiates with Multimodal Similarity-based News Retrieval to assemble a corpus of contextually analogous articles from an unlabeled reference database. Subsequently, during the Bifurcated Evidential Reasoning stage, agents perform a dual-directional analysis to extract critical patterns from the retrieved evidence. Finally, a Multi-Agent Collaborative Debate, involving Analyst and Arbiter agents, engages in a structured discourse to arrive at a definitive and robust conclusion. Comprehensive experiments on three benchmark datasets reveal that MRAFnd markedly surpasses state-of-the-art baselines, achieving an accuracy gain of up to 2.35% on the demanding Weibo-21 dataset.
Lehan Zhang, Yinlei Cheng, Shiqi Hu Yiheng Zhou +2
Beijing Institute of Fashion Technology, Beijing, China · Shenyang University of Technology, Shenyang, China