Semantic Entity Recognition on XFUND
90.47Accuracy (ZH)LiLT[InfoXLM] BASE
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| LiLT[InfoXLM] BASEMultitask fine-tuning=true, Pre-training Docs Language=English only, Pre-training Docs Size=11M2022.02 | 90.47 | 80.88 | 83.4 | 85.77 | 87.92 | 87.69 | 84.93 | 85.85 | |
| LayoutXLM_BASEMultitask fine-tuning=true, Pre-training Docs Language=Multilingual, Pre-training Docs Size=30M2022.02 | 89.73 | 79.64 | 77.98 | 81.73 | 82.1 | 83.22 | 82.41 | 82.01 | |
| LiLT[InfoXLM]-BASEPre-training Data Language=English only, Pre-training Docs Size=11M, Evaluation Protocol=fine-tuning on X, testing on X2022.02 | 89.38 | 79.64 | 79.11 | 79.53 | 83.76 | 82.31 | 82.2 | 82.51 | |
| LayoutXLM-BASEPre-training Data Language=Multilingual, Pre-training Docs Size=30M, Evaluation Protocol=fine-tuning on X, testing on X2022.02 | 89.24 | 79.21 | 75.5 | 79.02 | 80.82 | 82.22 | 79.03 | 80.56 | |
| InfoXLM-BASEPre-training Data Language=-, Pre-training Docs Size=-, Evaluation Protocol=fine-tuning on X, testing on X2022.02 | 88.68 | 78.65 | 62.3 | 70.15 | 67.51 | 70.63 | 70.08 | 72.07 | |
| XLM-R_BASEMultitask fine-tuning=true2022.02 | 88.3 | 77.86 | 62.23 | 70.35 | 68.14 | 71.46 | 67.26 | 71.49 | |
| XLM-RoBERTa-BASEPre-training Data Language=-, Pre-training Docs Size=-, Evaluation Protocol=fine-tuning on X, testing on X2022.02 | 87.74 | 77.61 | 61.05 | 67.43 | 66.87 | 68.14 | 68.18 | 70.47 | |
| InfoXLM_BASEMultitask fine-tuning=true2022.02 | 87.41 | 78.55 | 59.79 | 70.57 | 68.26 | 70.55 | 67.96 | 71.06 | |
| LiLT [InfoXLM] BASEPre-training Docs Language=English only, Pre-training Docs Size=11M, Evaluation Protocol=Cross-lingual zero-shot transfer2022.02 | 61.52 | 51.84 | 51.01 | 59.23 | 53.71 | 60.13 | 63.25 | 60.61 | |
| LayoutXLM_BASEPre-training Docs Language=Multilingual, Pre-training Docs Size=30M, Evaluation Protocol=Cross-lingual zero-shot transfer2022.02 | 60.19 | 47.15 | 45.65 | 57.57 | 48.46 | 52.52 | 53.9 | 55.61 | |
| InfoXLM_BASEEvaluation Protocol=Cross-lingual zero-shot transfer2022.02 | 44.08 | 36.03 | 31.02 | 40.21 | 28.8 | 35.87 | 45.02 | 41.19 | |
| XLM-R_BASEEvaluation Protocol=Cross-lingual zero-shot transfer2022.02 | 41.44 | 30.23 | 30.55 | 37.1 | 27.67 | 32.86 | 39.36 | 38.24 |