|
--- |
|
language: |
|
- uz |
|
tags: |
|
- transformers |
|
- uzroberta |
|
- uzbek |
|
- latin |
|
license: apache-2.0 |
|
widget: |
|
- text: Menga yoqdi, juda yaxshi ekan. |
|
metrics: |
|
- precision |
|
- recall |
|
- f1 |
|
- accuracy |
|
model-index: |
|
- name: uzroberta-sentiment-analysis |
|
results: [] |
|
--- |
|
|
|
# uzroberta-sentiment-analysis |
|
|
|
This is a roBERTa-base model trained on ~23K reviews (more than 323K words) and finetuned for sentiment analysis of customer reviews. This model is built as part of author's project at the Uz-NLP 2022 Hackathon and it is suitable for Uzbek language. |
|
|
|
<b>Labels</b>: LABEL_0 -> Negative; LABEL_1 -> Positive |
|
|
|
## Model description |
|
|
|
This model is a fine-tuned version of [rifkat/uztext-3Gb-BPE-Roberta](https://huggingface.co/rifkat/uztext-3Gb-BPE-Roberta) on the [Uzbek App reviews for Sentiment Classification](https://github.com/SanatbekMatlatipov/uzbek-sentiment-analysis) dataset. It achieves the following results on the evaluation set: |
|
- Loss: 0.5718 |
|
- Precision: 0.9113 |
|
- Recall: 0.8869 |
|
- F1 Score: 0.8989 |
|
- Accuracy: 0.896 |
|
|
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 2e-05 |
|
- train_batch_size: 16 |
|
- eval_batch_size: 32 |
|
- seed: 42 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: cosine |
|
- lr_scheduler_warmup_ratio: 0.1 |
|
- num_epochs: 4 |
|
- mixed_precision_training: Native AMP |
|
|
|
### Training results |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | Precision | Recall | F1 | Accuracy | |
|
|:-------------:|:-----:|:----:|:---------------:|:---------:|:------:|:------:|:--------:| |
|
| 0.1595 | 1.0 | 1125 | 0.4438 | 0.8971 | 0.8523 | 0.8741 | 0.872 | |
|
| 0.1891 | 2.0 | 2250 | 0.4157 | 0.8961 | 0.9012 | 0.8987 | 0.894 | |
|
| 0.1201 | 3.0 | 3375 | 0.5024 | 0.9074 | 0.8830 | 0.8950 | 0.892 | |
|
| 0.0772 | 4.0 | 4500 | 0.5718 | 0.9113 | 0.8869 | 0.8989 | 0.896 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.20.1 |
|
- Pytorch 1.12.0+cu116 |
|
- Datasets 2.3.2 |
|
- Tokenizers 0.12.1 |
|
|
