rubert-tiny2-russian-sentiment

Maintained By
seara

rubert-tiny2-russian-sentiment

PropertyValue
Parameter Count29.2M
LicenseMIT
LanguageRussian
TaskSentiment Analysis
FrameworkPyTorch/Transformers

What is rubert-tiny2-russian-sentiment?

This is a specialized sentiment analysis model based on RuBERT-tiny2 architecture, fine-tuned specifically for Russian text classification. It performs three-class sentiment classification (positive, negative, neutral) with impressive accuracy metrics, achieving a weighted F1-score of 0.75 and ROC-AUC of 0.90.

Implementation Details

The model was trained on a comprehensive collection of Russian sentiment datasets, including Kaggle Russian News, Linis Crowd 2015/2016, RuReviews, and RuSentiment. Training was performed with a maximum sequence length of 512 tokens, using Adam optimizer with a learning rate of 1e-5 over 5 epochs.

  • Batch size: 64
  • Train/validation/test split: 80%/10%/10%
  • Optimized for efficiency with only 29.2M parameters
  • Supports inference via Hugging Face Transformers pipeline

Core Capabilities

  • Three-class sentiment classification (neutral, positive, negative)
  • High precision for positive sentiments (0.84)
  • Robust performance across different text types
  • Efficient inference with minimal computational requirements

Frequently Asked Questions

Q: What makes this model unique?

The model combines the efficiency of the tiny2 architecture with comprehensive training on multiple Russian sentiment datasets, making it particularly suitable for production environments where computational resources are limited but accuracy is crucial.

Q: What are the recommended use cases?

This model is ideal for analyzing Russian social media content, customer reviews, news articles, and any short-to-medium length Russian text where sentiment analysis is required. It's particularly effective for real-time applications due to its lightweight architecture.

The first platform built for prompt engineering