Mineração de redes sociais para detecção e previsão de eventos reais

Abstract

Online social networks are part of the everyday life of millions of people worldwide. More and more people use these networks to interact, provide feedback and share content about several topics such as entertainment, weather, work, family, traffic and even their health. In short, social networks have become a social place with its own meanings and evolving dynamicity. Many events are perceived and released later bythe traditional media, but can occur in social networks in real time, being capable of being detected and subsidize the construction of predictive models. The objective of this dissertation is to use the data available on social networks to detect the occurrence and provide real-life events. To accomplish these tasks, we propose a methodology that extends from the collection of messages on social networks to predict the occurrence of events, through analysis of correlation between the nature of the message content and the occurrence of events in terms of volume, time and space. The proposed methodology was applied to two types of actual events: floods and dengue epidemics. In the case of the dengue epidemic, we found a high correlation (0.74) between messages expressing personal experience and the incidence of the disease, which allowed the building of an warning system of the epidemic by location with an accuracy greater than 90% for cities with high incidence . We also got comparable results for the second type of event, being able to detect the occurrence of flooding points and predict its intensity every day, demonstrating the applicability of our proposal to complement traditionalsurveillance mechanisms, often allowing anticipated actions and minimizing the impact on the affected population.