Abstract

Online social networks like Twitter facilitate instant communication during natural disasters. A key problem is to distinguish in real-time the most assertive and contingent tweets related to the current disaster from the whole streaming. To address this problem, machine learning allows to classify tweets according to their relevance or credibility. In this article, it is proposed to use centrality measures to improve the training data sample of active learning classifiers. As a case study, tweets collected during the massive floods in Santiago of Chile at 2016 are considered. This approach improves the consistency and pertinence of the labeling process, as well as the classifiers' performance.