---
license: apache-2.0
datasets:
- AiLab-IMCS-UL/go_emotions-lv
- AiLab-IMCS-UL/twitter_emotions-lv
language:
- lv
base_model:
- AiLab-IMCS-UL/lvbert
---
# Latvian Basic Emotion Classifier

A fine-tuned version of [LVBERT](https://huggingface.co/AiLab-IMCS-UL/lvbert) for multi-label text classification of six basic emotions (+neutral) in Latvian, as defined by Ekman’s theory.

The model is trained on a combined dataset of [go_emotions-lv](https://huggingface.co/datasets/AiLab-IMCS-UL/go_emotions-lv) and [twitter_emotions-lv](https://huggingface.co/datasets/AiLab-IMCS-UL/twitter_emotions-lv).

Predicted labels:
```yaml
0: anger
1: disgust
2: fear
3: joy
4: sadness
5: surprise
6: neutral
```

The random seed used for initialization was 42:
```python
def set_seed(seed=42):
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    if torch.cuda.is_available():
        torch.cuda.manual_seed_all(seed)
```

Training parameters:
```yaml
max_length: null
batch_size: 32
shuffle: True
num_workers: 4
pin_memory: False
drop_last: False
optimizer: adam
lr: 0.000005
weight_decay: 0
problem_type: multi_label_classification
num_epochs: 3
```

## Evaluation

Evaluation results on the test split of [go_emotions-lv](https://huggingface.co/datasets/AiLab-IMCS-UL/go_emotions-lv/viewer/simplified_ekman/test):
|              |Precision|Recall|F1-score|Support|
|--------------|---------|------|--------|-------|
|anger         |     0.57|  0.36|    0.44|    726|
|disgust       |     0.42|  0.29|    0.35|    123|
|fear          |     0.59|  0.43|    0.50|     98|
|joy           |     0.78|  0.80|    0.79|   2104|
|sadness       |     0.65|  0.42|    0.51|    379|
|surprise      |     0.62|  0.38|    0.47|    677|
|neutral       |     0.66|  0.58|    0.62|   1787|
|micro avg     |     0.70|  0.59|    0.64|   5894|
|macro avg     |     0.61|  0.46|    0.52|   5894|
|weighted avg  |     0.68|  0.59|    0.63|   5894|
|samples avg   |     0.62|  0.61|    0.61|   5894|

Evaluation results on the test split of [twitter_emotions-lv](https://huggingface.co/datasets/AiLab-IMCS-UL/twitter_emotions-lv/viewer/simplified_ekman/test):
|              |Precision|Recall|F1-score|Support|
|--------------|---------|------|--------|-------|
|anger         |     0.94|  0.87|    0.90|  12013|
|disgust       |     0.92|  0.92|    0.92|  14117|
|fear          |     0.74|  0.80|    0.77|   3342|
|joy           |     0.87|  0.88|    0.87|   5913|
|sadness       |     0.81|  0.80|    0.81|   4786|
|surprise      |     0.93|  0.57|    0.71|   1510|
|micro avg     |     0.89|  0.87|    0.88|  41681|
|macro avg     |     0.74|  0.69|    0.71|  41681|
|weighted avg  |     0.89|  0.87|    0.88|  41681|
|samples avg   |     0.86|  0.87|    0.86|  41681|

## See also

https://huggingface.co/AiLab-IMCS-UL/mbert-lv-emotions-ekman

## Acknowledgements

This work was supported by the EU Recovery and Resilience Facility project [Language Technology Initiative](https://www.vti.lu.lv) (2.3.1.1.i.0/1/22/I/CFLA/002).