Multi-class classification on Social VR harassment dataset
70.57AccuracyTransformer-based
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Transformer-basedModel Type=Baseline2026.04 | 70.57 | 67.58 | 65.37 | 65.71 | |
| HarassGuardBackbone=GPT-4o, Fine-tuning=true, Prompting Strategy=+ Few-shot2026.04 | 68.85 | 60.35 | 57.16 | 56.78 | |
| HarassGuardBackbone=GPT-4o, Fine-tuning=true, Prompting Strategy=Baseline Prompt2026.04 | 65.74 | 56.13 | 54.34 | 53.61 | |
| HarassGuardBackbone=GPT-4o, Fine-tuning=true, Prompting Strategy=+ Context2026.04 | 65.4 | 59.23 | 55.78 | 53.8 | |
| HarassGuardBackbone=GPT-4o, Fine-tuning=true, Prompting Strategy=+ CoT2026.04 | 64.22 | 62.05 | 56.04 | 54.64 | |
| HarassGuardBackbone=GPT-4o, Fine-tuning=false, Prompting Strategy=+ Few-shot2026.04 | 54.67 | 48.89 | 49.76 | 44.33 | |
| HarassGuardBackbone=GPT-4o, Fine-tuning=false, Prompting Strategy=Baseline Prompt2026.04 | 52.26 | 50.58 | 48.17 | 42.07 | |
| HarassGuardBackbone=GPT-4o, Fine-tuning=false, Prompting Strategy=+ CoT2026.04 | 50.15 | 54.23 | 46.71 | 40.75 | |
| HarassGuardBackbone=GPT-4o, Fine-tuning=false, Prompting Strategy=+ Context2026.04 | 48.27 | 64.31 | 46.33 | 38 | |
| LSTM/CNN-basedModel Type=Baseline2026.04 | 43.04 | 23.27 | 25.04 | 15.24 |