CEVA: Complementary Evidence Verification via Frame-Level Reorganization for Short-Video Fake News Detection
Organizations: School of Cyber Science and Engineering, Sichuan University · School of Cyber Engineering, Xidian University · School of International Studies, Sichuan University
Abstract
Short-video platforms are now a common way for the public to follow unfolding events. Their rapid, visually persuasive format also makes fake news easier to spread. Unlike text-image posts, short-video news interleaves multiple modalities and changing scenes, making it difficult to identify which frames verify a claim. Existing detectors either pool sampled frames homogeneously or select them by text-frame similarity. Both overlook that aligned frames may repeat a misleading claim, whereas less dominant context may challenge it. We present CEVA (Complementary Evidence Verification Approach), to our knowledge the first short-video fake news detection framework to explicitly organize frames by their evidential roles before multimodal fusion. Its Dual-Branch Complementary Reorganization (DBCR) module separates explicit-alignment frames from candidate complementary-context frames, and cross-modal fusion with bidirectional contrastive regularization operates on the reorganized evidence. On FakeSV and FakeTT, CEVA achieves the best overall results among the compared methods. Under fixed frame budgets and backbones, DBCR outperforms uniform, random, and similarity-based selection and improves two structurally different detectors. Together, these results establish pre-fusion evidence organization as an effective and transferable component of short-video fake news detection.
Figures & tables
| FakeSV | FakeTT | |||||||
| Method | Acc | M-F1 | M-Prec | M-Rec | Acc | M-F1 | M-Prec | M-Rec |
| BERT | 76.89 | 76.83 | 77.21 | 76.90 | 71.24 | 69.31 | 68.98 | 70.85 |
| ViT | 71.22 | 71.04 | 71.04 | 71.33 | 65.55 | 64.39 | 65.17 | 67.11 |
| TikTec | 73.43 | 73.26 | 73.23 | 73.54 | 66.22 | 65.08 | 65.84 | 67.87 |
| FANVM | 79.52 | 78.81 | 79.81 | 78.46 | 71.57 | 70.21 | 70.22 | 72.63 |
| SV-FEND | 80.29 | 80.03 | 80.70 | 80.34 | 77.62 | 77.54 | 77.56 | 77.53 |
| Configuration | Acc | M-F1 | M-Rec |
|---|---|---|---|
| CEVA (Full) | 87.98/86.67 | 87.96/85.58 | 87.97/84.86 |
| w/o | 83.71/82.40 | 83.44/81.35 | 83.39/81.27 |
| w/o DBCR | 84.93/83.73 | 84.89/82.64 | 84.86/82.35 |
| w/o + | 84.48/83.47 | 84.40/82.38 | 84.33/82.13 |
| w/o | 85.69/84.27 | 85.66/83.38 | 85.63/82.82 |
| w/o Cross-Attn. | 84.17/85.33 | 84.15/84.83 | 84.13/84.43 |
| (a) Frame-selection strategy | ||||||||
|---|---|---|---|---|---|---|---|---|
| Fake | Real | |||||||
| Strategy | Acc | M-F1 | Prec | Rec | F1 | Prec | Rec | F1 |
| Uniform | 87.37 | 87.32 | 89.04 | 84.28 | 86.59 | 85.96 | 90.27 | 88.06 |
| Random | 84.02 | 83.99 | 85.11 | 81.68 | 83.36 | 83.05 | 86.27 | 84.63 |
| Title-Top | 86.45 | 86.35 | 90.46 | 80.50 | 85.19 | 83.42 | 92.04 | 87.52 |
| Title-Bottom | 82.95 | 82.90 | 83.88 | 80.19 | 81.99 | 82.15 | 85.55 | 83.82 |
| Configuration | Acc | M-F1 | M-Rec |
|---|---|---|---|
| SV-FEND on FakeSV | |||
| Baseline | 80.29 | 80.03 | 80.34 |
| + DBCR | 84.08 | 83.98 | 84.02 |
| FakingRecipe on FakeTT | |||
| Baseline | 79.53 | 77.42 | 78.29 |
| + DBCR | 81.61 | 79.86 | 80.89 |
Fake News Detection
Fake news detection identifies false or misleading claims by analyzing content, context, evidence, and how information spreads.
See all 104 papers