distilbert-base-multilingual-cased-finetuned
This model is a fine-tuned version of distilbert-base-multilingual-cased on the Emotone Arabic dataset, which includes tweets labeled for various emotions: none, anger, joy, sadness, love, sympathy, surprise, and fear. It achieves the following results on the evaluation set:
- Loss: 1.3099
- Accuracy: 0.6632
- F1: 0.6647
Model description
This model is designed for emotion recognition in Arabic text. It can classify tweets into one of the eight emotional categories.
Intended uses & limitations
This model is intended for applications in sentiment analysis and emotion detection in Arabic tweets. It may not perform well on texts outside the domain of social media or on languages other than Arabic.
Training and evaluation data
The model was fine-tuned on the Emotone Arabic dataset, which consists of tweets labeled with the following emotions:
- none
- anger
- joy
- sadness
- love
- sympathy
- surprise
- fear
Label Mapping
Label Name | Numeric Label |
---|---|
none | 0 |
anger | 1 |
joy | 2 |
sadness | 3 |
love | 4 |
sympathy | 5 |
surprise | 6 |
fear | 7 |
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 10
Training results
Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 |
---|---|---|---|---|---|
1.0026 | 1.0 | 252 | 1.0417 | 0.6408 | 0.6321 |
0.8422 | 2.0 | 504 | 1.0355 | 0.6508 | 0.6425 |
0.7114 | 3.0 | 756 | 1.0611 | 0.6364 | 0.6342 |
0.5709 | 4.0 | 1008 | 1.0672 | 0.6692 | 0.6665 |
0.459 | 5.0 | 1260 | 1.1167 | 0.6731 | 0.6693 |
0.3694 | 6.0 | 1512 | 1.1709 | 0.6637 | 0.6672 |
0.2975 | 7.0 | 1764 | 1.2094 | 0.6716 | 0.6699 |
0.2402 | 8.0 | 2016 | 1.2777 | 0.6642 | 0.6633 |
0.209 | 9.0 | 2268 | 1.2997 | 0.6692 | 0.6685 |
0.1792 | 10.0 | 2520 | 1.3099 | 0.6632 | 0.6647 |
Example Outputs
Here are some example inputs and their corresponding model predictions:
Input Tweet | Predicted Emotion | Numeric Label |
---|---|---|
"أنا سعيد جدًا اليوم!" | joy | 2 |
"هذا أمر محبط حقًا." | sadness | 3 |
"لا أستطيع تحمل هذا بعد الآن." | anger | 1 |
"أحب كل من يدعمني." | love | 4 |
Framework versions
- Transformers 4.44.2
- Pytorch 2.4.1+cu121
- Datasets 3.0.0
- Tokenizers 0.19.1
- Downloads last month
- 35