Fake News Detection on FakeTT
89.52AccuracyMAGIC3+VLM (ours)
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| MAGIC3+VLM (ours)Evaluation Protocol=Supervised2026.03 | 89.52 | 88.14 | 89.7 | 88.5 | 87.2 | 90.1 | 530.73 | 22.76 | |
| FakeSV-VLMEvaluation Protocol=Supervised2026.03 | 88.9 | 87.6 | 88.2 | 86.5 | 86.6 | 89.1 | 30.25 | 22.76 | |
| MAGIC3 (ours)Evaluation Protocol=Supervised2026.03 | 84.95 | 82.28 | 94.12 | 80 | 68.99 | 89.9 | 693.76 | 1.67 | |
| KDSGAT-FNVDEvaluation Protocol=Supervised2026.03 | 81.94 | 78.94 | 92.68 | 76 | 64.44 | 87.88 | — | — | |
| FakingRecipeEvaluation Protocol=Supervised2026.03 | 79.26 | 77.87 | 89.66 | 78 | 64.8 | 81.82 | — | — | |
| SVFENDEvaluation Protocol=Supervised2026.03 | 77.14 | 75.63 | 87.91 | 76.33 | 62.33 | 78.79 | — | — | |
| HCFC-HouEvaluation Protocol=Supervised2026.03 | 73.24 | 72 | 87.04 | 70.5 | 56.93 | 78.79 | — | — | |
| FANVMEvaluation Protocol=Supervised2026.03 | 71.57 | 70.21 | 85.28 | 69.5 | 55.15 | 75.76 | — | — | |
| TikTecEvaluation Protocol=Supervised2026.03 | 66.22 | 65.08 | 82.35 | 63 | 49.32 | 72.73 | — | — | |
| HCFC-MedinaEvaluation Protocol=Supervised2026.03 | 62.54 | 62.23 | 84.92 | 53.5 | 46.24 | 80.81 | — | — | |
| GPT-4o-miniEvaluation Protocol=Zero-shot2026.03 | 61.54 | 61.2 | 81.62 | 52 | 47.28 | 79.8 | — | — | |
| GPT-4.1-miniEvaluation Protocol=Zero-shot2026.03 | 49.16 | 48.54 | 77.42 | 48 | 37.5 | 71.4 | — | — | |
| InternVL2.5Evaluation Protocol=Zero-shot2026.03 | 46.82 | 45.29 | 77.6 | 49 | 52.24 | 69.46 | — | — | |
| Qwen2.5-VLEvaluation Protocol=Zero-shot2026.03 | 45.82 | 45.31 | 72.1 | 45 | 41.28 | 65.84 | — | — | |
| InternVL2.5-MPOEvaluation Protocol=Zero-shot2026.03 | 43.14 | 40.84 | 74 | 47 | 49.8 | 65.46 | — | — |