Yelp Review Classification Model

This model takes the text from a Yelp review and predicts a label corresponding to the number of stars for the review. The labels are:

• "Very Negative" = 1 star
• "Negative" = 2 stars
• "Neutral" = 3 stars
• "Positive" = 4 stars
• "Very Positive" = 5 stars

Base Model

Tabularis.ai Multilingual Sentiment Classification Model

Dataset

Yelp Review Full

• train samples: 520000
• validation samples: 130000

Training Hyperparameters

• initial learning rate: 1e-5
• learning rate decay: linear
• batch size: 64
• epochs: 5

Evaluation

• test samples: 50000
• accuracy: 0.6772

Usage

The easiest way to inference the model is through its Hugging Face checkpoint.

Alternatively, clone this repo and run the code with uv:

uv run src/inference.py