metadata
language: en
tags:
- ner
- bert
- mountain-ner
- named-entity-recognition
license: mit
datasets:
- NERetrieve
- Few-NERD
- mountain-ner-dataset
metrics:
- accuracy
- f1
- precision
- recall
pipeline_tag: token-classification
model-index:
- name: mountain-ner-bert-base
results:
- task:
type: token-classification
name: Named Entity Recognition
dataset:
name: mountain-ner-dataset
type: Gepe55o/mountain-ner-dataset
metrics:
- type: accuracy
value: 0.9919
- type: f1
value: 0.9048
- type: precision
value: 0.8899
- type: recall
value: 0.9202
Model Description
mountain-ner-bert-base is a fine-tuned model based on the BERT base architecture for mountain names Entity Recognition tasks. The model is trained on the merging of two datasets: NERetrieve, Few-NERD, Mountain-ner-dataset. The model is trained to recognize two types of entities: LABEL_0 (other), LABEL_1 (mountain names).
- Model Architecture: BERT base
- Task: mountain names entity recognition
- Training Data: mountain-ner-dataset
Performance
Metrics:
| Epoch | Training Loss | Validation Loss | Accuracy | Precision | Recall | F1 |
|---|---|---|---|---|---|---|
| 1 | 0.027400 | 0.030793 | 0.988144 | 0.815692 | 0.924621 | 0.866748 |
| 2 | 0.020600 | 0.024568 | 0.991119 | 0.872988 | 0.921036 | 0.896369 |
| 3 | 0.012900 | 0.024072 | 0.991923 | 0.889878 | 0.920171 | 0.904771 |
Best model performance achieved at epoch 3 with:
- F1 Score: 0.9048
- Accuracy: 0.9919
- Precision: 0.8899
- Recall: 0.9202
How to use
from transformers import AutoModel, AutoTokenizer, pipeline
model = AutoModel.from_pretrained("Gepe55o/mountain-ner-bert-base")
tokenizer = AutoTokenizer.from_pretrained("Gepe55o/mountain-ner-bert-base")
text = "Mount Everest is the highest mountain in the world."
nlp = pipeline("ner", model=model, tokenizer=tokenizer)
result = nlp(text)