This project is a work on detecting tweets that require an action from first responders or the government during an emerging crisis, such as a natural disaster or an attack. We use a domain-adaptation technique between different events, by adversarially training a classifier removing the importance on the event per se, allowing the trained model to focus on the criticality of the tweet.

• BERT Embeddings of dataset (json)
• BERT Embeddings of dataset (npy)

pip install easydict
pip install pytorch-revgrad
pip install bert-embedding
pip install pyyaml
pip install pytorch-transformers
pip install umap-learn