Character Embedding Evaluation on Character Embedding Benchmark (CEB)
98.6Gender AccuracyCN
Evaluation Results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CNEvaluation Protocol=Linear Probing2026.05 | 98.6 | 36.1 | 75 | 96.7 | 76.6 | 49.5 | — | — | 34.4 | — | 85.6 | 95.5 | 80.2 | 87.1 | |
| Roberta+CN+PosEvaluation Protocol=Linear Probing2026.05 | 98.1 | 43.2 | 92.4 | 97.8 | 82.9 | 48.5 | — | — | 42.7 | — | 83.9 | 95.6 | 81.2 | 86.9 | |
| Static CNEvaluation Protocol=Linear Probing, Graph Type=Static Character Network2026.05 | 96.1 | 41.8 | 82.4 | 99 | 79.8 | 54.9 | 71.1 | 78.7 | 54.4 | 64.8 | 74.1 | 96.4 | 80.7 | 83.7 | |
| No CNEvaluation Protocol=Linear Probing, Graph Type=None2026.05 | 96 | 44.8 | 81.7 | 99.5 | 80.5 | 52.6 | 68.8 | 75.6 | 51.6 | 62.2 | 78.3 | 95.6 | 81 | 85 | |
| DHCNEvaluation Protocol=Linear Probing, Graph Type=Dynamic Heterogeneous Character Network2026.05 | 95.7 | 48.7 | 82.8 | 99.3 | 81.6 | 52.9 | 67 | 75.5 | 52.1 | 61.9 | 76.3 | 94.6 | 79.9 | 83.6 | |
| E5-LargeEvaluation Protocol=Linear Probing2026.05 | 95.3 | 42.6 | 84.2 | 98 | 80.1 | 54 | 68.8 | 74.4 | 51.7 | 62.2 | 78.4 | 95 | 78.9 | 84.1 | |
| DCNEvaluation Protocol=Linear Probing, Graph Type=Dynamic Character Network2026.05 | 93.9 | 41.6 | 75.8 | 97.6 | 77.2 | 51.4 | 63 | 68 | 44.4 | 56.7 | 72.5 | 95.3 | 73.1 | 80.3 | |
| Qwen3-Emb-4bEvaluation Protocol=Linear Probing2026.05 | 87.1 | 34.2 | 62.1 | 98.3 | 70.4 | 56.7 | 54.8 | 72.9 | 48.7 | 58.3 | 72.8 | 87.3 | 79.4 | 79.8 |