rubert-tiny2-cedr-emotion-detection
| Property | Value |
|---|---|
| Parameter Count | 29.2M |
| Language | Russian |
| Task | Emotion Detection |
| Training Dataset | CEDR |
| Paper | Data-Driven Model for Emotion Detection in Russian Texts |
What is rubert-tiny2-cedr-emotion-detection?
This is a specialized Russian language model fine-tuned for emotion detection in text. Based on the rubert-tiny2 architecture, it's designed to perform multilabel classification, capable of detecting multiple emotions in a single sentence. The model was trained on the CEDR dataset and can identify six distinct emotional states: no emotion, joy, sadness, surprise, fear, and anger.
Implementation Details
The model was trained using the Adam optimizer over 40 epochs with a learning rate of 1e-5 and batch size of 64. Performance metrics show impressive results, with AUC scores ranging from 0.75 to 0.95 across different emotions, and particularly strong F1 scores averaging 0.93 for micro-averaging.
- Architecture: BERT-based transformer model
- Training Configuration: Adam optimizer, 40 epochs, 1e-5 learning rate
- Performance: Mean AUC of 0.8956 across all categories
- Size: 29.2M parameters
Core Capabilities
- Multilabel emotion classification in Russian text
- Detection of 6 emotional states
- High accuracy with F1 micro score of 0.9280
- Efficient processing with relatively small parameter count
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its specialized focus on Russian language emotion detection, combining the efficiency of the tiny2 architecture with strong performance metrics. Its multilabel classification capability makes it particularly valuable for complex emotional analysis tasks.
Q: What are the recommended use cases?
The model is ideal for sentiment analysis in Russian text, social media monitoring, customer feedback analysis, and any application requiring nuanced emotional understanding of Russian language content. It's particularly useful when multiple emotions need to be detected in a single piece of text.
