Modojojo / imdb-sentiment-analysis

Sentiment analysis on IMDB dataset using DistilBERT. Fine-tuned a pre-trained DistilBERT model on the IMDB dataset.

IMDB Sentiment Analysis

Classify a movie review into Positive or Negative review using DistilBERT Transformer.
A pre-trained DistilBERT transformer model was used which was fine-tuned on the IMDB reviews dataset.

Webapp

Home page

Positive review

Prediction for Positive review

Negative Review

Prediction for negative review

Repository Structure

main
|── models
|    |── model                   # Holds model files
|    └── tokenizer               # Holds tokenizer files 
|
|── screenshots                  # Contains webapp screenshots
|
|── src
|    |── __init__.py     
|    └── models_utils.py         # Model loading and prediction utility
|
|── templates
|    |── index.html 
|    └── error.html
| 
|── training
|    └── imdb_sentiment_analysis_training.ipynb    # Training notebook
|
|── .gitignore
|── README.md
|── app.py
|── params.yaml
|── requirements.txt
└── setup.py

Setup

1. Create a new conda environment

conda create -n imdb-sentiment-analysis python==3.7

2. run the setup file
   Note - If any error occurs for any package, comment the corresponding package in requirements.txt file and install the dependency separately.

pip install .

3. Use the notebook provided in Training folder to train a DistilBERT Transformer model on the imdb reviews dataset.
4. Download/Save the trained Model and Tokenizer files and paste them in the corresponding folder in the models directory.
5. Run the app

python app.py