Towards Exploiting Social Networks for Detecting Epidemic Outbreaks

Social networks are becoming a valuable source of information for applications in many domains. In particular, many studies have highlighted the potential of social networks for early detection of epidemic outbreaks, due to their capability to transmit information faster than traditional channels, thus leading to quicker reactions of public health officials. Anyhow, the most of these studies have investigated only one or two diseases, and consequently to date there is no study in the literature trying to investigate if and how different kinds of outbreaks may lead to different temporal dynamics of the messages exchanged over social networks. Furthermore, in case of a wide variability, it is not clear if it would be possible to define a single generic solution able to detect multiple epidemic outbreaks, or if specifically tailored approaches should be implemented for each disease. To get an insight into these open points, we collected a massive dataset, containing more than one hundred million Twitter messages from different countries, looking for those relevant for an early outbreak detection of multiple disease. The collected results highlight that there is a significant variability in the temporal patterns of Twitter messages among different diseases. In this paper, we report on the main findings of this analysis, and we propose a set of steps to exploit social networks for early epidemic outbreaks, including a proper document model for the outbreaks, a Graphical User Interface for the public health officials, and the identification of suitable sources of information useful as ground truth for the assessment of outbreak detection algorithms.