Psychological Defense Classification on PsyDefConv 2026 (official leaderboard)
64.83AccuracyMinistral-8B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ministral-8Bmode=fine-tuned (SOTA)2026.05 | 64.83 | 31.48 | |
| Qwen3-8B LoRAconfiguration=SeedA Ensemble + v2decode2026.05 | 64.19 | 39.17 | |
| InternLM3-8Bmode=fine-tuned2026.05 | 63.98 | 30.53 | |
| Llama 3.1-8Bmode=fine-tuned2026.05 | 62.92 | 30.51 | |
| DeBERTa-v3-basevalidation=5-fold2026.05 | 59.11 | 23.58 | |
| Qwen3-8B LoRAconfiguration=Grouped CV + Bias Tuning2026.05 | 58.43 | 35.48 | |
| Gemini 2.5 Promode=zero-shot2026.05 | 56.36 | 25.99 | |
| DeepSeek-V3.2mode=zero-shot, technique=CoT2026.05 | 55.72 | 26.17 | |
| Qwen3-8B LoRAconfiguration=Baseline Finetuned2026.05 | 54.45 | 24.91 | |
| GPT-5mode=zero-shot, source=task paper2026.05 | 52.75 | 19.53 | |
| RoBERTa-base2026.05 | 51.27 | 26.97 |