Edit model card

Short Description

The rubert-tiny2-russian-emotion-detection is a fine-tuned rubert-tiny2 model for multi-label emotion classification task, specifically on Russian texts. Trained on custom ru-izard-emotions dataset, so this model can recognize a spectrum of 9 emotions, including joy, sadness, anger, enthusiasm, surprise, disgust, fear, guilt, shame + neutral (no emotion). Project was inspired by the Izard's model of human emotions.

For more information about model, please check Github repository

Training Parameters:

Optimizer: AdamW
Schedule: LambdaLR
Learning Rate: 1e-4
Batch Size: 64
Number Of Epochs: 10

Emotion Categories:

0. Neutral (Нейтрально)
1. Joy (Радость)
2. Sadness (Грусть)
3. Anger (Гнев)
4. Enthusiasm (Интерес)
5. Surprise (Удивление)
6. Disgust (Отвращение)
7. Fear (Страх)
8. Guilt (Вина)
9. Shame (Стыд)

Test results:

Neutral Joy Sadness Anger Enthusiasm Surprise Disgust Fear Guilt Shame Mean
AUC 0.7319 0.8234 0.8069 0.7884 0.8493 0.8047 0.8147 0.9034 0.8528 0.7145 0.8090
F1 micro 0.7192 0.7951 0.8204 0.7642 0.8630 0.9032 0.9156 0.9482 0.9526 0.9606 0.8642
F1 macro 0.6021 0.7237 0.6548 0.6274 0.7291 0.5712 0.4780 0.8158 0.4879 0.4900 0.6180

Citations

@misc{Djacon,
  author={Djacon},
  year={2023},
  publisher={Hugging Face},
  journal={Hugging Face Hub},
}
Downloads last month
520
Safetensors
Model size
29.2M params
Tensor type
I64
·
F32
·
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Dataset used to train Djacon/rubert-tiny2-russian-emotion-detection

Space using Djacon/rubert-tiny2-russian-emotion-detection 1