Text Style Transfer: Informal to Formal

Overview

This project aims to perform text style transfer, specifically converting informal text to formal text. It leverages the NUS Social Media Text Normalization and Translation Corpus. The repository explores three primary approaches:

1. A simple encoder-decoder model.
2. An encoder-decoder model with an attention mechanism.
3. Pre trained model (facebook bart-large)

Getting Started

Prerequisites

Ensure you have the following Python libraries installed:

• NumPy
• pandas
• Matplotlib
• seaborn
• TensorFlow
• NLTK (for nltk.translate.bleu_score)
• joblib
• scikit-learn
• NLPAug

Data Preprocessing

Execute the Data_Preprocessing.ipynb notebook to prepare the data for training:

jupyter notebook Data_Preprocessing.ipynb

Simple Encoder-Decoder Model

To run the simple encoder-decoder model:

jupyter notebook Simple_Encoder_Decoder.ipynb

Encoder-Decoder Model with Attention Mechanism

For the attention mechanism-enhanced model:

jupyter notebook Final_Attention_Model.ipynb

Using Pretrained Model Facebook Bart-large

For the pretrained model:

jupyter notebook Pre_Trained_Model.ipynb

Streamlit Interface

This project includes a user-friendly interface built using Streamlit, a powerful framework for creating interactive web applications with Python. The interface allows users to interact with the model seamlessly and obtain results in real-time.

Usage

To run the Streamlit interface, execute the following command in your terminal:

streamlit run app.py