Sentimental Analysis IMDB

Description

Classification of sentences depending on polarity- Positive or Negative achieved using training on the IMDB Dataset

Concepts

1. RNN
2. LSTM
3. Word Embeddings
4. Multilayer and Bi-Directional RNNs

RESULTS

Trained for 5 epochs on Kaggle.

Test Accuracy of ~90%

Dataset

IMDB

Code Structure

Simple-RNN : Used simple one layer RNN with no pretrained embeddings.

Improved-RNN : Used LSTM along with glove-6b embeddings, with a 2 layered bi-directional RNN.

This project is in PyTorch

Requirements:

• PyTorch
• Torchtext (Used torchtext.legacy)
• Spacy

Usage

• You can pretrain the model
• You can use the predict sentiment to try your own sentences

Acknowledgements

Thanks to the author of this for this amazing repo and its explanation and help.