Automatic Understanding of Human Sensors Testimonials – CATCH

After the fire at the Lubrizol factory on September 26, 2019, many inhabitants of Rouen and its surroundings were victims of inconveniences such as very strong odors or symptoms that could lead to the suspicion of harmful health effects. Between September 26, 2019 and September 30, 2020, more than 6,000 reports were received on the ODO web platform set up by Atmo Normandie. Many of these reports were supported by a comment expressing an emotion such as astonishment, fear or anger. These testimonials were also found on social networks (Twitter in particular), but it was difficult to really estimate their nature, their intensity and their evolution over time in order to effectively orient actions, adapt communication according to the situation and alert the organizations involved.

In light of this situation, the CATCH project proposes to use artificial intelligence tools, and in particular deep learning, to automatically exploit these testimonials relating to an industrial accident and its environmental and health consequences. More precisely, we have two goals: (i) to draw up a precise cartography of the nuisances allowing to follow the propagation and the evolution of the phenomena in time, and (ii) to analyze and characterize the sentiment of the population and its evolution throughout the crisis. To achieve these two objectives, we will exploit the data from the ODO platform, together with selected messages collected on the Twitter platform. We will be able to rely on the complementary expertise of the three actors of the consortium who are gathered around this project: Atmo Normandie, thanks to its perfect knowledge of the subject, can interpret the results and evaluate the relevance of the answers provided; the LITIS laboratory brings its scientific expertise in deep learning and automatic natural language processing; and Saagie brings its experience in data engineering (Big Data) and provides technical tools in data science and AI. In addition, it should be noted that Saagie and the LITIS laboratory have already collaborated on related scientific issues, notably within the scope of the RAPID-DGA SAPhIRS project (System for the Analysis of the Propagation of Information in Social Networks) and a CIFRE PhD thesis, for which solutions have been developed and maintained and which could benefit this project.

The solutions developed in this project will be made freely accessible, so that they can be capitalized and improved beyond the project. In addition, these solutions will be designed to be transferable to other areas in which communication is a key issue in crisis management. As such, we propose to make the applications and the dataset made up of tweets collected and annotated by CATCH available to the community to help in the digital processing of events and crises of all kinds.