Disaster Tweets Classification

Twitter is a popular social media platform where users share posts short texts to express their feelings to others. Such Tweets often contain useful information about critical situations such as disasters. If the information is extracted and processed efficiently, it could be used to quickly respond to the situation and make informed decisions. However, it is a challenging task to separate the disaster-related Tweets from others due to the unstructured nature of text data.

In this project, I have developed a transfer learning-based approach to detect whether a Tweet is about a disaster or not. I have fine-tuned a pretrained DistilBERT model that on the Kaggle “Natural language processing with disaster Tweets” dataset. The model achieves more than 83% accuracy on the test set.

Click here try-out the model yourself, or you can access the source code here.