Fake News Detection on FakeSV
90.93AccuracyMAGIC3+VLM (ours)
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| MAGIC3+VLM (ours)Evaluation Protocol=Supervised2026.03 | 90.93 | 90.41 | 90.85 | 90.95 | 91.1 | 88.9 | 509.71 | 24.68 | |
| FakeSV-VLMEvaluation Protocol=Supervised2026.03 | 89.6 | 89.4 | 89.2 | 90.1 | 90.5 | 88 | 18.67 | 24.68 | |
| MAGIC3 (ours)Evaluation Protocol=Supervised2026.03 | 86.71 | 86.84 | 86.52 | 83.61 | 87.5 | 89.8 | 680.43 | 1.67 | |
| KDSGAT-FNVDEvaluation Protocol=Supervised2026.03 | 84.73 | 84.68 | 82.23 | 83.61 | 87 | 85.86 | — | — | |
| FakingRecipeEvaluation Protocol=Supervised2026.03 | 82.87 | 83.52 | 91.35 | 71 | 80.67 | 94.74 | — | — | |
| SVFENDEvaluation Protocol=Supervised2026.03 | 80.88 | 80.17 | 74.53 | 83.61 | 85.82 | 77.63 | — | — | |
| FANVMEvaluation Protocol=Supervised2026.03 | 79.52 | 78.81 | 80.98 | 69.75 | 78.64 | 87.17 | — | — | |
| HCFC-MedinaEvaluation Protocol=Supervised2026.03 | 76.38 | 75.83 | 74.77 | 69.75 | 77.5 | 81.58 | — | — | |
| HCFC-HouEvaluation Protocol=Supervised2026.03 | 74.91 | 73.61 | 77.72 | 60.08 | 73.46 | 86.51 | — | — | |
| TikTecEvaluation Protocol=Supervised2026.03 | 73.43 | 73.26 | 68.08 | 74.37 | 78.37 | 72.7 | — | — | |
| GPT-4.1-miniEvaluation Protocol=Zero-shot2026.03 | 70.3 | 70.25 | 68.12 | 73.53 | 73.1 | 68.21 | — | — | |
| GPT-4o-miniEvaluation Protocol=Zero-shot2026.03 | 68.08 | 68.05 | 65.35 | 73.97 | 74.41 | 64.93 | — | — | |
| InternVL2.5-MPOEvaluation Protocol=Zero-shot2026.03 | 65.13 | 61.07 | 63.75 | 59.16 | 69.17 | 65.08 | — | — | |
| InternVL2.5Evaluation Protocol=Zero-shot2026.03 | 64.39 | 57.89 | 63.8 | 57.5 | 73.24 | 63.5 | — | — | |
| Qwen2.5-VLEvaluation Protocol=Zero-shot2026.03 | 64.21 | 60.79 | 62.5 | 60.54 | 66.6 | 62.5 | — | — |