Skip to content

Latest commit

 

History

History
36 lines (29 loc) · 1.85 KB

README.md

File metadata and controls

36 lines (29 loc) · 1.85 KB
Raw

Bangla-Punctuation-Corrector

This project focuses on restoring punctuation from Bangla text. We have prepared this as our "Final Year Design Project" course work. The project contains two major part:

  1. Prepare and train a model
  2. Deploy and create an application interface for testing.
It was quite challenging to decide the 1st part as there were no existing work related to this task for Bangla. After investigating some deep learning model(mainly based on NLP) we choose Bi-Directional Recurrent Neural Network (GRU) with Attention mechanism as our base model. Our accurecy was not so high, but it gives us a threshold for any future work related to bangla punctuation task. We have also performed sevaral postprocessing technique to increase our machine output accurecy.

Training

Please visit here to read about training stage (Will be pulished soon)

Testing Interface

We have used Flask API as a lightweight python framework to create a testing interface. Screenshots along with test case are given below. We intentionally kept the restored punctuation with their name to identify model output.

To Test the Model

  1. Clone or download this repo
  2. Download the pretrainned model from here
  3. Put the model .pcl file into project directory
  4. Create a vertual env in yout machine and install dependencies given in requirements.txt
  5. run python api.py
  6. It will open the app into your localhost
  7. Play :)

Tools for Interface