ordek899/roberta_1to5rating_pred_for_restaur_trained_on_hotels

Hotel and Restaurant Review Rating Model

This model predicts the rating of a hotel or restaurant review on a scale from 1 to 5. It was trained on a hotel review dataset from Hugging Face and further tested with a small restaurant dataset. The model is based on the RoBERTa architecture and uses PyTorch for training and inference.

Model Details

• Model Name: ordek899/roberta_1to5rating_pred_for_restaur_trained_on_hotels
• Training Data: Trained on the Hugging Face hotel dataset and tested on a small restaurant dataset.
• Output: A rating prediction from 1 to 5 based on the input review.

Key Features

• Detailed performance metrics: Including classification report with precision, recall, and F1-score for each rating.
• Instance-level predictions: A CSV file containing the true label, predicted label, and correctness for each instance.
• Confusion matrix visualization: A confusion matrix to evaluate the model’s performance across different ratings (1 to 5).
• Comprehensive model information: Includes detailed insights into model training, evaluation, and predictions.

How It Works

The model uses a transformer-based architecture (RoBERTa) for classifying reviews into one of five ratings (1 to 5). It processes hotel and restaurant reviews and predicts a numerical rating based on the sentiment and content of the review.

Dependencies

To use this model, you will need to install the following Python packages:

pip install transformers datasets torch

Example Usage

Making Predictions with the Model To use the model for making predictions on a review, you can run the following Python code:

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# Load the model and tokenizer
model_name = "ordek899/roberta_1to5rating_pred_for_restaur_trained_on_hotels"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

# Example review - Replace this text with your own review
review = """Insert your review here."""

# Tokenize the input review
inputs = tokenizer(review, return_tensors="pt", truncation=True, padding=True, max_length=512)

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

# Convert logits to predicted rating (from 1 to 5)
predicted_rating = torch.argmax(logits, dim=-1) + 1 # Adding 1 because labels are typically 0-indexed
print(f"Predicted rating: {predicted_rating.item()}")

Model Evaluation and Training

The model was trained on the Hugging Face hotel dataset and evaluated with a small restaurant dataset. It predicts the rating of a review on a scale from 1 to 5.

The following files are generated during the evaluation process: Instance-Level Results: A CSV file containing the true label, predicted label, and correctness for each instance. Classification Report: A CSV file with detailed metrics like precision, recall, and F1-score for each rating. Confusion Matrix: A heatmap that visually shows how well the model performed across all five rating categories (1 to 5).

Example Output

When you run the prediction code with an input review, the output will show something like:

Predicted rating: 3

This indicates that the model has predicted a rating of 3 for the provided review.