A Geolocation Approach for Tweets Not Explicitly Georeferenced Based on Machine Learning
- International Symposium on Distributed Computing and Artificial Intelligence , 740 : 223-231
Résumé
In this paper, we describe an inference model for deducing the location of a tweet whose geolocation information is not available in the metadata. The approach we propose is based on machine learning techniques and uses the information contained in the tweets such as the places mentioned in the tweets and the profile of the authors of the tweets. The objective of the study is to contribute to setting up an early warning system for epidemics based on the monitoring of events on social networks like twitter. For this we need to geolocate the messages even if the smartphone’s GPS is not active. We trained on three models and obtained the best result with K-nearest neighbors model with an accuracy of 0.90.
Mots-clés
Geolocation, Machine Learning, infectious diseases, Twitter , real-time