Charges Prediction on JurisMM-Text (test)
88.2AccuracyJurisMMA
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| JurisMMABackbone=GPT-4o, Knowledge Base (KB)=true, Multi-agent collaboration (MA)=true2026.01 | 88.2 | 60 | 64.2 | 60.3 | |
| JurisMMABackbone=GPT-4o, Knowledge Base (KB)=false, Multi-agent collaboration (MA)=true2026.01 | 80.1 | 20.3 | 19 | 18.7 | |
| JurisMMABackbone=GPT-4o, Knowledge Base (KB)=true, Multi-agent collaboration (MA)=false2026.01 | 77.3 | 58.1 | 59.9 | 56.6 | |
| TOPJUDGE2026.01 | 75.1 | 51.6 | 51.9 | 49.1 | |
| MPBFN2026.01 | 70.7 | 42.7 | 41.6 | 38.7 | |
| TextCNN2026.01 | 69.6 | 49.2 | 45.7 | 44.1 | |
| JurisMMABackbone=Qwen2.5-VL-7B, Knowledge Base (KB)=true, Multi-agent collaboration (MA)=true2026.01 | 61.8 | 38.4 | 42.4 | 37 | |
| JurisMMABackbone=Qwen2.5-VL-7B, Knowledge Base (KB)=true, Multi-agent collaboration (MA)=false2026.01 | 60.4 | 34.1 | 38.2 | 33.6 | |
| Qwen2.5-7B-Instruct2026.01 | 59 | 16.1 | 14.4 | 13.4 | |
| Qwen2.5-VL-7BSFT=true2026.01 | 51.5 | 29.5 | 27.7 | 25.2 | |
| GLM-4V-9B2026.01 | 47.6 | 16.9 | 12 | 11.9 | |
| GPT-4o-202503262026.01 | 43.6 | 25.3 | 18.6 | 18.2 | |
| JurisMMABackbone=Qwen2.5-VL-7B, Knowledge Base (KB)=false, Multi-agent collaboration (MA)=true2026.01 | 41.4 | 19.1 | 11.6 | 12.9 | |
| mPLUG-7B2026.01 | 31.9 | 9.3 | 6 | 5.8 | |
| Qwen2.5-VL-3B-Instruct2026.01 | 25.8 | 10.9 | 6.8 | 6.8 | |
| Qwen2.5-VL-7B-Instruct2026.01 | 13.4 | 6.1 | 2.1 | 2.6 |