angryelizar/ruBert-base-sentiment-classifier-final

Russian Review Sentiment Classifier

A model for classifying the sentiment of Russian-language user reviews (positive / negative / neutral), fine-tuned from ai-forever/ruBert-base.

Model Description

• Base model: ai-forever/ruBert-base
• Task: sentiment analysis / text classification
• Language: Russian
• Dataset: angryelizar/sentiment_dataset_splitted
• Training epochs: 1

Metrics

Metric	Value
Accuracy	77%

How to Use

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_name = "angryelizar/ruBert-base-sentiment-classifier-final"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

text = "Классный сервис - мне нравится!"
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)

with torch.no_grad():
    logits = model(**inputs).logits

predicted_class_id = torch.argmax(logits, dim=-1).item()
print(model.config.id2label[predicted_class_id])

An interactive UI for trying out the model is also available — see the Links section below.

Training Data

The model was trained on the angryelizar/sentiment_dataset_splitted dataset.

Links

• 📊 Dataset: https://huggingface.co/datasets/angryelizar/sentiment_dataset_splitted
• 🧠 Base model: https://huggingface.co/ai-forever/ruBert-base
• 🔧 Fine-tuning code: https://github.com/angryelizar/review-classifier-finetune
• ✅ Testing code: https://github.com/angryelizar/review-classifier-tester
• 🖥️ Interactive UI: https://github.com/angryelizar/ru-review-classifier-app

Project Context

This model is part of Capstone #1: Text Classifier on PyTorch & BERT (Attractor School, 2026).

Limitations

• The model was trained on a mixed-domain review dataset (marketplaces, movies, anime, geolocation reviews, etc.), so accuracy may differ on narrower, more specialized domains.
• The model was fine-tuned for only 1 epoch — accuracy could potentially be higher with additional training epochs.