This repository contains a fine-tuned DeBERTa model for sentiment analysis using the McDonald's review dataset from Kaggle. The model has been fine-tuned to classify customer sentiments as positive, negative, or neutral.

Model Details

Base Model: microsoft/deberta-v3-base
Task: Sentiment Analysis
Dataset: McDonald's review dataset (Kaggle)
Framework: Hugging Face Transformers, PyTorch
Fine-Tuning Approach:
- Tokenized using DeBERTa tokenizer
- Trained with cross-entropy loss
- Optimized using AdamW
- Learning rate scheduler applied

Dataset

The dataset consists of customer reviews from McDonald's, labeled with sentiment categories:

0: Negative
1: Neutral
2: Positive

Training Hyperparameters

Batch Size: 16
Epochs: 3
Learning Rate: 2e-5
Optimizer: AdamW
Scheduler: Cosine learning rate decay with warmup (10% warmup ratio)
Weight Decay: 0.01
Evaluation Strategy: Epoch-based
Save Strategy: Best model saved at the end
Mixed Precision Training: Enabled (fp16)
Gradient Accumulation Steps: 2
Logging Steps: 50

How to Use the Model

To load and use the model for inference:

python

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_name = "iSathyam03/McD_Reviews_Sentiment_Analysis"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

def predict_sentiment(text):
    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
    with torch.no_grad():
        outputs = model(**inputs)
    logits = outputs.logits
    prediction = torch.argmax(logits, dim=1).item()
    sentiment_labels = {0: "Negative", 1: "Neutral", 2: "Positive"}
    return sentiment_labels[prediction]

# Example
text = "The fries were amazing but the burger was stale."
print(predict_sentiment(text))

Evaluation

The model was evaluated using accuracy, F1-score, and confusion matrix.
Performance metrics suggest that the fine-tuned DeBERTa model achieves high accuracy on customer sentiment classification.

Deployment

You can deploy this model using Hugging Face's Inference API or via Streamlit/Gradio for an interactive UI.

Citation

If you use this model, please cite this repository and the original DeBERTa paper:

@article{he2020deberta,
  title={DeBERTa: Decoding-enhanced BERT with Disentangled Attention},
  author={He, Pengcheng and Liu, Xiaodong and Gao, Jianfeng and Chen, Weizhu},
  journal={arXiv preprint arXiv:2006.03654},
  year={2020}
}