French COVID-19 Tweets Classification Using FlauBERT Layers
- Lecture notes on data engineering and communications technologies , 114 : 805-816
Résumé
Late in 2019, Wuhan a city in China recorded its first case of coronavirus. Over time, the virus has spread to all continents, resulting in numerous victims. Many techniques have been developed to contain the spread of the virus, ranging from preventive to curative approaches. However, these solutions are still a luxury for developing countries. In this work, we have proposed a framework based on the Twitter datasets that can allow them to follow the propagation of the virus in real time. This low-cost solution could allow them to have information on the tendencies of contamination of the virus and consequently to take measures to contain it. In this framework, we proceeded to a tweet cleaning using the pre-trained Glove and FastText models. The text coverage rate was, respectively, 73.95% for the Glove model and 73.3, 58% for the FastText in the training dataset. We then trained a deep neural network using the BERT layers with a hyperparameter lot size of 32 and a hidden layer of 12. This allowed us to obtain an accuracy of 0.98%.
Mots-clés
COVID-19 pandemic, tweet classification, Gustave Flaubert, BERT, gloves, FastText, text cleaning, French language, Twitter, low-cost information systems