Sentiment Analysis with DistilBERT

Overview

This project implements sentiment analysis using the DistilBERT model, leveraging PyTorch and the Hugging Face Transformers library. The model is fine-tuned on a dataset comprising 10,000 IMDB movie reviews.

Project Components

1. Data Preprocessing:

   • The dataset is preprocessed using the DistilBERT tokenizer to convert raw text into a format suitable for model training.

2. Model Architecture:

   • The DistilBERT model is employed for sentiment analysis. The architecture, including hyperparameters, is defined to suit the task.

3. Training Process:

   • The model is trained using a TrainingArguments object, incorporating gradient accumulation and a learning rate schedule for effective training.

4. Evaluation Metrics:

   • The model's performance is assessed on a separate test dataset, with metrics such as accuracy and F1-score computed to gauge its effectiveness.

5. Model Sharing:

   • The trained model is shared and made accessible via the Hugging Face Hub, facilitating easy integration into other projects.

6. Inference with Pipeline API:

   • The Pipeline API is utilized for sentiment analysis on new data using the uploaded model. The model demonstrates accurate classification of text polarity as positive or negative.

Evaluation Metrics

• Accuracy: Measures the overall correctness of the model's predictions.
• F1-Score: A metric that balances precision and recall, providing a comprehensive evaluation of model performance.

Usage

1. Training:

   • Fine-tune the DistilBERT model on your dataset by following the training process outlined in the code.

2. Evaluation:

   • Assess the model's performance on a test dataset using the provided evaluation metrics.

3. Model Sharing:

   • Share your trained model on the Hugging Face Hub for easy accessibility and integration.

4. Inference:

   • Use the Pipeline API to perform sentiment analysis on new text data, benefiting from the accurate predictions of the trained model.

Dependencies

• PyTorch
• Hugging Face Transformers
• Other dependencies as specified in the project code.

Acknowledgments

• The project leverages the power of the DistilBERT model and the Hugging Face Transformers library.
• Inspiration and guidance from the natural language processing, machine learning, and deep learning communities.

Streamlit Link -

https://huggingface.co/Satyajithchary/Sentiment_Analysis_using_BERT_With_10000_Samples/tree/main/