NLP Disaster Tweets Classifier

Leveraging the power of BERT finetuned on a curated corpus of tweets for enriched Natural Language Understanding, our project achieved to reach the 27th spot among 871 contenders in Kaggle's NLP challenge. This achievement was realized as part of the final project for the AI course ECSE526, at McGill University.

Setup

To install the depencies needed for this project, you'll need to install conda, then run :

conda env create -f environment.yml

Important files

Most of our work can be found in the notebook, here is a brief summary of what you can find in them :

• data_exploration.ipynb explores the data set and the pre-processing pipeline
• BERT.ipynb is about the fine-tuning of BERT, its optimization and how to create a simple ensemble prediction
• LSTM.ipynb is about the training of a LSTM network