metadata
language: fr
tags:
- camembert
- sentiment-analysis
- french
license: mit
pipeline_tag: text-classification
datasets:
- tblard/allocine
demo: https://huggingface.co/spaces/visalkao/sentiment-analysis-french
French Sentiment Analysis with CamemBERT
This model predicts sentiment (0: NEGATIVE, 1: POSITIVE) for French product reviews.
How to Test the model
You can go to the space attach to this model - https://huggingface.co/spaces/visalkao/sentiment-analysis-french
Note:
This model is finetuned on a subset of tblard/allocine dataset (7000 random samples 50-50).
Example Output
{
"label": "POSITIVE",
"score": 0.98
}
Training procedure
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 8
- eval_batch_size: 8
- weight_decay=0.01
- num_epochs: 10
Training results
Here is the result of training:
| Epoch | Training Loss | Validation Loss | Accuracy on Validation | Precision on Validation |
|---|---|---|---|---|
| 1 | 0.224300 | 0.166868 | 0.954000 | 0.955621 |
| 2 | 0.224000 | 0.174095 | 0.959500 | 0.959500 |
| 3 | 0.145400 | 0.194145 | 0.961500 | 0.961873 |