Content Moderation on Fine-grained Moderation Dataset
89.2Average F1CHAIRO (Ours)
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| CHAIRO (Ours)Category=Proposed Methods2026.04 | 89.2 | 89.3 | 71.5 | 97.8 | 82 | 96.1 | 98.6 | |
| E2E Reliability (RQ3)Category=Proposed Methods2026.04 | 86.9 | 88.3 | 93.2 | 96 | 98 | 82.2 | 63.7 | |
| Analogy-Rule Quality (RQ2)Category=Proposed Methods2026.04 | 84.7 | 84.4 | 68.4 | 92.8 | 75.4 | 90.3 | 96.9 | |
| Rule Impact (RQ1)Category=Proposed Methods2026.04 | 83.9 | 83.8 | 67.5 | 92.2 | 73.7 | 90.6 | 95.7 | |
| DeepSeek V3Category=General LLMs2026.04 | 80.3 | 79 | 90.3 | 89.8 | 95 | 70.5 | 62.5 | |
| DeepSeek R1Category=General LLMs2026.04 | 77.1 | 72.7 | 91.4 | 86.1 | 94.3 | 64.6 | 59.7 | |
| Qwen2.5-32B-InstructCategory=General LLMs2026.04 | 74.3 | 59.1 | 91.1 | 84.4 | 95.4 | 67.9 | 54.2 | |
| GPT-4Category=General LLMs2026.04 | 72.3 | 58.6 | 88.7 | 79.8 | 92.7 | 64.3 | 56.8 | |
| QwQ-32BCategory=General LLMs2026.04 | 69.1 | 75.4 | 69.6 | 72 | 84.9 | 60.7 | 54.6 | |
| LLaMA3-8BCategory=General LLMs2026.04 | 67.5 | 58.5 | 55.9 | 81 | 90.6 | 65.2 | 44.2 | |
| LLaMA-Guard-3-8BCategory=Specific LLMs2026.04 | 39.7 | 12 | 74.1 | 41.8 | 29.4 | 45.7 | 35.6 |