Early alert of an outbreak based on event detection and tracking on social networks: The case of Covid-19 and meningitis

In this paper, we propose an early warning approach for the occurrence of meningitis or Covid-19 epidemics based on the analysis of Twitter messages (tweets). To achieve this, we initially employed automatic classification techniques using Natural Language Processing and domain ontologies to filter the tweets. Subsequently, we utilized the metadata of the tweets to geolocate them and extract their publication dates. The real-time collection of tweets is facilitated through Kafka’s ecosystem.For the real-time identification of tweets, we trained SVM and CNN models. Our best results were obtained with the CNN model, which yielded an accuracy of 0.99. Tweets pertaining to infections are then geolocated by considering various key elements within the tweet that can provide information about the location.