In the event of a flood, we no longer call the fire brigade, we send a tweet! Social networks are, of course, the place of emotion, the most intense propaganda and the most malignant fake news, but also that of crucial instantaneous information. CNRS researchers in Toulouse and Paris have developed unique software in France which could see the light of day by 2026. Decision support for faster intervention in the event of a crisis. But, for lack of means and will, the risk is real of a possible recovery by banks and insurance companies…
Both angels and demons… If Twitter and social networks welcome all kinds, they are also places of information, especially in matters of emergency and natural disaster. It is from this observation that Alda Mari, linguist at the Jean-Nicod Institute, at the CNRS, in Paris (1) and her research partner based in Toulouse, Farah Benamara – lecturer in computer science at the university Toulouse III Paul Sabatier – joined forces in 2018 to manufacture a model, called Intact (intention-action), “capable of revolutionizing the apprehension of a major crisis endangering human lives that can save lives” did not yet exist in France. Until then, it was “do-it-yourself” based on volunteers…
We struggle for the project to remain at the CNRS. If the Ministry of the Interior does not wake up, it will go to banks and insurance…”
Every time there’s a crisis, “emergency services are still relying on an association, Visov, whose volunteers read the social networks live! It’s titanic and just impossible! We have developed software, currently in prototype stage, which sorts tweets automatically and sends them to the rescue. And this, since last May. For this software, financed for another year by CNRS innovation, to equip the emergency services, a company must finance the interface. “For this, there are SATTs, technology transfer units, which mediate between researchers and the business world. The problem is the change of personnel which takes place continuously at the Ministry of the Interior, we do not even know who will do the follow-up. since Macron’s new term of office, everything has changed and we have lost our time interlocutors…”
The other problem is that insurance and banks, in particular, are just waiting for this to financially exploit this software. And, incidentally, have the respective contributions paid accordingly. “We struggle, says Alda Mari, so that the project remains in the hands of the CNRS. If the Ministry of the Interior does not wake up”… It will go into highly capitalist hands. “But if a big company like EDF takes over this project, everything can go quickly, by 2026” for a project initially planned for 2024.
Decision support for public authorities
for faster emergency response
This model is a decision-making aid for the public authorities for a faster rescue intervention with a better knowledge of the seriousness of the situation. The prototype is just waiting to grow: “Our model now also needs to apply to droughts, a typical example of crises where we are not in the urgency of the moment but so-called silent crises but which, on the other hand, last a long time”, emphasizes Alda Mari again. Crises, other difficulty, “that many insurance companies refuse to reimburse. So, we try to test this model on all natural disasters that could cause victims. Frost, too, conversely. Can we help the population to organize itself better, to anticipate better than we do now in the face of major events that sometimes happen suddenly? This is our central question…”
Several areas of expertise requested
Farah Benamara (1) and Alda Mari have known each other since the end of their university studies twenty years ago before deciding, in 2018, to respond together to a call for projects from the Ministry of the Interior, aimed at analyzing text messages shared on social networks. The goal ? Decision support. In order to understand these messages, to extract useful information and to automate their analysis, several fields of research are then solicited, in particular linguistics and computational linguistics, two fields of expertise of the teams of Farah Benamara and Alda Mari.
In the pre-maturation phase so that the prototype acquires robustness and can be useful to a maximum of actors specialized in crisis management”
“The call for projects came more specifically from the Directorate General for Civil Security and Crisis Management. We then responded to a second call for projects, explains computer scientist Farah Benamara, we already have a proof of concept and are in the prematurity phase so that the prototype acquires robustness and can be useful to a maximum of actors specialized in crisis management. The objective was to automatically analyze and extract the most useful and relevant information in order to predict the actions to be taken in the event of a crisis and we succeeded in doing so.”
Some 13,000 tweets were analyzed!
The Alda Mari team took care of the strictly linguistic part of the project: “We manually annotated a corpus of 13,000 tweets posted over several crisiss. With the Directorate General for Civil Security and Crisis Management, we analyzed all these messages, tried to determine which are the most important, how they are formulated and what urgency of a situation they express. Some messages are of no use but others are important: information on human or material damage, a vehicle asking for help, etc.
On Twitter, it’s like writing to your “mom” during a crisis, to tell her that you’re stuck; ditto for the fire at Notre-Dame, in Paris: the first act of people was not to contact the 18 but to write messages on Twitter…”
Meaning, syntax, morphology… Alda Mari specializes in the more specific analysis of semantics, the meaning of words. “The purpose of this model is to have an analysis of social networks in the event of an ecological crisis: floods, storms, droughts…” So many insults that affect the South, the most affected part of the country. “We worked a lot on the floods, explains Alda Mari again ; well we notice that people no longer call the fire brigade or any help: they write what happens to them on social networks! As a result, the emergency services did not have the information immediately. Hence this request from the Ministry of the Interior in the face of the absence of appeals.”
The click happened during a day of heavy snow blocking all the way to the highways. “The Ministry found out from the…TV…” Clearly, on social networks, we write our amazement; “On Twitter, it’s like writing to your ‘mom’, during a crisis, to tell her that you’re stuck; ditto for the fire at Notre-Dame, in Paris: the first act of people was not to contact the 18 but to write messages on Twitter…”decodes Alda Mari.
We started building a text-based system to send the right tweets to the right people. It is a very difficult task…”
Faced with this observation, the 18 having perhaps become obsolete, the Ministry of the Interior had then wished to create an automatic monitoring of social networks in the event of a major ecological crisis. To learn as fast as “mother” and at least as fast as the media.
But what was there to look for as interesting information in social networks? “We mainly worked on Twitter, an open network, and for data accessibility stories, otherwise there can be a lot of constraints related to private data. The selected tweets must be relevant for paramedics, gendarmes, firefighters, etc. And we started building a text-based system to send the right tweets to the right people. It is a very difficult task. Our collaboration between linguists and computer scientists aimed to create a robust model. Because this system generates a lot of data in real time.”
It is also very difficult from the point of view of understanding. For instance : “Any system is still based on keywords. But for a crisis we recover anything. For example on the floods, people say: “I have a flood”, whereas it is only their bathtub which overflows…”
Artificial intelligence is enriched by humans
An unexpected flood of the river, the word “flood” can mean a lot of things: the feminine past participle of believing; it is also the opposite of ripe; it is also the rising of the river; or else I was not believed. “Well, our system starts from automatic categorization, which is done very well in artificial intelligence. We’re going to ask a human – our students – who we give 2,000 tweets.”
What do we ask him? “He is asked to answer – is it relevant? Urgent ? and if it is urgent, for whom is it urgent? in which category: human damage, material damage… – so that the artificial intelligence program learns from its responses. And the degree of interpretation of the urgency.” Since then,” the machine will try to create a criterion that will classify the tweets. And it works pretty well when you know that 80% of tweets are “noise”, messages to be put in the trash. VSIt is still the firefighters and/or the emergency services who have the freedom to decide whether or not to go to the scene. “And we are improving this software on language understanding (do tweets include exclamations, for example)…”
(1) Farah Benamara works at the Computer Science Research Institute of Toulouse1, in the Methods and engineering of languages, ontologies and discourse team, of which she is co-head. Alda Mari leads the Language, Thought and Behavior team at the Jean Nicod Institute, in the Cognitive Sciences department of the École Normale Supérieure in Paris.
We want to say thanks to the author of this article for this remarkable material
Ecological Crises: How Social Media Could Save Lives – Tell Them!
Find here our social media accounts as well as the other related pageshttps://www.ai-magazine.com/related-pages/