您尚未登录账户请先登录您的atyun账户，方可使用该功能
您的企业账号申请正在审核中审核通过后即可使用此功能，请耐心等待~
模型:
yeshpanovrustem/xlm-roberta-large-ner-kazakh
英文为哈萨克语设计的命名实体识别模型
- 该模型的灵感来自于该论文 LREC 2022 .
- 该模型在 ner_kazakh 上进行了3个周期的训练.
- 该论文的原始仓库可在 https://github.com/IS2AI/KazNERD 找到.
如何使用
您可以使用Transformers流水线进行NER。
from transformers import AutoTokenizer, AutoModelForTokenClassification
from transformers import pipeline
tokenizer = AutoTokenizer.from_pretrained("yeshpanovrustem/xlm-roberta-large-ner-kazakh")
model = AutoModelForTokenClassification.from_pretrained("yeshpanovrustem/xlm-roberta-large-ner-kazakh")
nlp = pipeline("ner", model = model, tokenizer = tokenizer)
example = "Қазақстан Республикасы — Шығыс Еуропа мен Орталық Азияда орналасқан мемлекет."
ner_results = nlp(example)
print(ner_results)
在验证集和测试集上的评估结果
| Validation set | Test set | ||||
|---|---|---|---|---|---|
| Precision | Recall | F 1 -score | Precision | Recall | F 1 -score |
| 96.58% | 96.66% | 96.62% | 96.49% | 96.86% | 96.67% |
模型对验证集中的NE类的性能
| NE Class | Precision | Recall | F 1 -score | Support |
|---|---|---|---|---|
| ADAGE | 90.00% | 47.37% | 62.07% | 19 |
| ART | 91.36% | 95.48% | 93.38% | 155 |
| CARDINAL | 98.44% | 98.37% | 98.40% | 2,878 |
| CONTACT | 100.00% | 83.33% | 90.91% | 18 |
| DATE | 97.38% | 97.27% | 97.33% | 2,603 |
| DISEASE | 96.72% | 97.52% | 97.12% | 121 |
| EVENT | 83.24% | 93.51% | 88.07% | 154 |
| FACILITY | 68.95% | 84.83% | 76.07% | 178 |
| GPE | 98.46% | 96.50% | 97.47% | 1,656 |
| LANGUAGE | 95.45% | 89.36% | 92.31% | 47 |
| LAW | 87.50% | 87.50% | 87.50% | 56 |
| LOCATION | 92.49% | 93.81% | 93.14% | 210 |
| MISCELLANEOUS | 100.00% | 76.92% | 86.96% | 26 |
| MONEY | 99.56% | 100.00% | 99.78% | 455 |
| NON_HUMAN | 0.00% | 0.00% | 0.00% | 1 |
| NORP | 95.71% | 95.45% | 95.58% | 374 |
| ORDINAL | 98.14% | 95.84% | 96.98% | 385 |
| ORGANISATION | 92.19% | 90.97% | 91.58% | 753 |
| PERCENTAGE | 99.08% | 99.08% | 99.08% | 437 |
| PERSON | 98.47% | 98.72% | 98.60% | 1,175 |
| POSITION | 96.15% | 97.79% | 96.96% | 587 |
| PRODUCT | 89.06% | 78.08% | 83.21% | 73 |
| PROJECT | 92.13% | 95.22% | 93.65% | 209 |
| QUANTITY | 97.58% | 98.30% | 97.94% | 411 |
| TIME | 94.81% | 96.63% | 95.71% | 208 |
| micro avg | 96.58% | 96.66% | 96.62% | 13,189 |
| macro avg | 90.12% | 87.51% | 88.39% | 13,189 |
| weighted avg | 96.67% | 96.66% | 96.63% | 13,189 |
模型对测试集中的NE类的性能
| NE Class | Precision | Recall | F 1 -score | Support |
|---|---|---|---|---|
| ADAGE | 71.43% | 29.41% | 41.67% | 17 |
| ART | 95.71% | 96.89% | 96.30% | 161 |
| CARDINAL | 98.43% | 98.60% | 98.51% | 2,789 |
| CONTACT | 94.44% | 85.00% | 89.47% | 20 |
| DATE | 96.59% | 97.60% | 97.09% | 2,584 |
| DISEASE | 87.69% | 95.80% | 91.57% | 119 |
| EVENT | 86.67% | 92.86% | 89.66% | 154 |
| FACILITY | 74.88% | 81.73% | 78.16% | 197 |
| GPE | 98.57% | 97.81% | 98.19% | 1,691 |
| LANGUAGE | 90.70% | 95.12% | 92.86% | 41 |
| LAW | 93.33% | 76.36% | 84.00% | 55 |
| LOCATION | 92.08% | 89.42% | 90.73% | 208 |
| MISCELLANEOUS | 86.21% | 96.15% | 90.91% | 26 |
| MONEY | 100.00% | 100.00% | 100.00% | 427 |
| NON_HUMAN | 0.00% | 0.00% | 0.00% | 1 |
| NORP | 99.46% | 99.18% | 99.32% | 368 |
| ORDINAL | 96.63% | 97.64% | 97.14% | 382 |
| ORGANISATION | 90.97% | 91.23% | 91.10% | 718 |
| PERCENTAGE | 98.05% | 98.05% | 98.05% | 462 |
| PERSON | 98.70% | 99.13% | 98.92% | 1,151 |
| POSITION | 96.36% | 97.65% | 97.00% | 597 |
| PRODUCT | 89.23% | 77.33% | 82.86% | 75 |
| PROJECT | 93.69% | 93.69% | 93.69% | 206 |
| QUANTITY | 97.26% | 97.02% | 97.14% | 403 |
| TIME | 94.95% | 94.09% | 94.52% | 220 |
| micro avg | 96.54% | 96.85% | 96.69% | 13,072 |
| macro avg | 88.88% | 87.11% | 87.55% | 13,072 |
| weighted avg | 96.55% | 96.85% | 96.67% | 13,072 |