Introduction

This project uses different pre-trained models to predict the sentiments of people based on reviews in Hindi.

Dataset Description

The dataset is a collection of Hindi reviews (2006 training examples) with 4 sentiments:

1. Happy
2. Sad
3. Angry
4. Neutral

Models used

1. LSTM
2. BERT
3. mBERT
4. DistillBERT
5. IndicBERT

Steps

Transliterate

1. In the Transliterate Directory 1st git clone below repository:

   git clone https://github.com/yourusername/torch-transformer-hinglish2hindi-translator.git
   cd torch-transformer-hinglish2hindi-translator

(go to the repository to check how to use it)

2. Change the model_path in transliterate.ipynb to any of the models and run that file.

Model building

1. Run Hindi_Sentiment_analysis.ipynb which uses IndicBert model.