Committed to collecting information mentioning the new coronavirus or the covid-19 disease, researchers from USP’s Institute of Mathematical and Computer Sciences (ICMC) are dedicated to improving the artificial intelligence techniques of Websensors, a tool that uses the automatic extraction of events from from news. With it, it is possible to identify subjects, names of people and organizations, all with georeferenced locations. It is an additional support for specialists dedicated to fighting the disease that has taken over the news in recent months and which, it seems, will remain very present in our daily lives for some time. In this way, it is possible to outline strategies based on what was successful and also on what did not work in fighting the virus. Solange Rezende, project coordinator, explains the importance of something like this: “When we look at the future evolution of the disease contamination curve and only take into account data on contagions that happened in the past, we have a limited view of the problem. If it is possible to enrich this view, adding information extracted from reliable sources to the forecast, we believe that we can increase our view and, who knows, build predictive models closer to reality ”. It is a fact that there is a massive amount of information circulating on the internet about the pandemic and that this tends to increase more and more. Being a source of data from all over the world, with the right tool, it is possible to use the network as a kind of sensor that makes it possible to build models to predict what is coming. With Websensors, numerous links are captured through GDELT, an international platform that tracks news from every corner of the globe in more than 100 languages. Using filters that categorize the desired terms and the validity of the sources, a pre-processing is performed, transforming texts into a set of signals – something that a computer can work with, after all, human capacity is somewhat limited in terms of speed when compared to an artificial intelligence system. Then the magic begins to happen. This information is released in a neural network, responsible for the identification of varied patterns, from the most subtle to the most evident. For example, if a certain amount of news talks about the contamination curve of somewhere, all the data are taken into account and expand the model created virtually – in an analysis that can consider both the recovery rate of a certain age group and the actions taken that can explain the phenomenon, in addition to other details.
