Twitter is one of the most popular social media these days, and the best place to get breaking news instantly. Monitoring Twitter is interesting for a lot of companies, e.g. disaster relief organizations. In this regard, we are going to build a model for the disaster tweets classification task of whether the tweets are pointing to a real disaster or not. To overcome this problem, we proposed a model using contextualized BERT embedding layer. Our proposed model shows performance improvement in comparison to the base model with LSTM structure in terms of F1 score.
You can find out more details in our paper.
The only consideration for running notebook is that you have to run it (main.ipynb) with GPU.