rubert-tiny-sentiment-balanced
| Property | Value |
|---|---|
| Parameter Count | 11.8M parameters |
| Model Type | Sentiment Analysis |
| Language | Russian |
| Downloads | 88,370 |
| Framework | PyTorch |
What is rubert-tiny-sentiment-balanced?
This is a lightweight Russian language model specifically designed for sentiment analysis, based on the rubert-tiny architecture. It's been fine-tuned to perform three-class sentiment classification (positive, neutral, negative) on Russian texts, offering an efficient balance between model size and performance.
Implementation Details
The model utilizes the Transformers architecture and has been trained on multiple Russian sentiment datasets collected by Smetanin. It employs a balanced training approach, using both up-sampling and down-sampling techniques to ensure equal representation across different sources and sentiment classes.
- Compact architecture with only 11.8M parameters
- Implements PyTorch backend with Transformers framework
- Uses Safetensors format for model storage
- Supports both CPU and CUDA execution
Core Capabilities
- Three-class sentiment classification (negative, neutral, positive)
- Achieves impressive macro F1 scores: up to 0.98 on mokoron dataset
- Handles various text domains (banks, telecom, news, reviews)
- Provides flexible output formats: labels, scores, or probabilities
Frequently Asked Questions
Q: What makes this model unique?
The model stands out for its balanced approach to Russian sentiment analysis, combining a compact architecture (11.8M parameters) with robust performance across different domains. Its balanced training methodology ensures consistent performance across various text sources.
Q: What are the recommended use cases?
The model is ideal for analyzing Russian text sentiment in various contexts, including social media monitoring, customer feedback analysis, and general text sentiment classification. It's particularly effective for production environments where efficiency and accuracy are both crucial.
