Personal fact classification on Personal Facts Annotation held-out (test)
79.4Main Category Macro F1 (MC)Gemma-300M (Multi-Head)
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Gemma-300M (Multi-Head)Architecture=Multi-Head encoder, Evaluation Protocol=mean±std over 5 seeds2026.05 | 79.4 | 85.7 | 84.7 | 80.4 | 84.2 | 59 | 78.8 | 81.6 | |
| GPT-5.4-miniMethod Family=Few-Shot LLMs, Evaluation Protocol=Single evaluation2026.05 | 76.5 | 80.75 | 77.75 | 76.13 | 78.96 | 53.81 | 72.13 | 72.92 | |
| RoBERTa-large (Multi-Head)Architecture=Multi-Head encoder, Evaluation Protocol=mean±std over 5 seeds2026.05 | 74.8 | 86.2 | 83.2 | 81.4 | 83.5 | 48.3 | 80.2 | 76.5 | |
| BGE-large (Multi-Head)Architecture=Multi-Head encoder, Evaluation Protocol=mean±std over 5 seeds2026.05 | 74.4 | 86 | 85.5 | 80.4 | 84.2 | 53.5 | 78 | 76.5 | |
| GPT-5-miniMethod Family=Few-Shot LLMs, Evaluation Protocol=Single evaluation2026.05 | 74.38 | 82.09 | 79.29 | 78.46 | 80.44 | 49.37 | 79 | 72.2 | |
| Llama-3.3-70B-InstructMethod Family=Few-Shot LLMs, Evaluation Protocol=Single evaluation2026.05 | 71.96 | 82.11 | 82.48 | 76.2 | 54.24 | 49.46 | 71.44 | 70.41 | |
| DeepSeek-V3.2Method Family=Few-Shot LLMs, Evaluation Protocol=Single evaluation2026.05 | 70.36 | 79.59 | 73.47 | 74.46 | 74.12 | 44.5 | 70.67 | 68.69 | |
| Multi-E5-base (Multi-Head)Architecture=Multi-Head encoder, Evaluation Protocol=mean±std over 5 seeds2026.05 | 67.1 | 85.1 | 82 | 78.9 | 83.9 | 36.8 | 77.1 | 71.7 | |
| Qwen3-235B-A22B-2507Method Family=Few-Shot LLMs, Evaluation Protocol=Single evaluation2026.05 | 65.59 | 82.33 | 79.43 | 71.51 | 53.28 | 47.68 | 74.34 | 70.04 | |
| SVMMethod Family=Traditional ML baselines, Evaluation Protocol=mean±std over 5 seeds2026.05 | 61.1 | 72.9 | 71 | 63.3 | 70.3 | 40 | 69.7 | 61.1 | |
| Logistic Reg.Method Family=Traditional ML baselines, Evaluation Protocol=mean±std over 5 seeds2026.05 | 60.1 | 71.1 | 69 | 64.3 | 71.2 | 40.3 | 68.6 | 60.2 | |
| Gradient Boost.Method Family=Traditional ML baselines, Evaluation Protocol=mean±std over 5 seeds2026.05 | 57.2 | 67.7 | 64.8 | 60 | 69.8 | 38.4 | 64.7 | 57.9 |