AI uses tweets to help researchers analyze the flood situation

Today's famous social networking sites like Facebook, Twitter or Instagram are getting more and more criticized about the negativity that it brings to society, running out of scams, misrepresenting, impersonating, and coming back to beat User data theft. However, it is unjust to deny all the merits of these social media platforms. Recently, the world's second largest social networking site, Twitter, has recently contributed greatly to the success of a research project that has a great impact on our lives. More specifically, there has recently been information about a scientific research project published on Arxiv.org with the title: 'Integrating Social Media into a Pan-European Flood Awareness System: A Multilingual Approach' (roughly translated) : Integrating social media into the European flood warning system: Multilingual approach), using a method called Social Media for Flood Risk (SMFR), receives Get a lot of attention from meteorologists as well as people all over the world.

1. Invite to listen to Rock music created by artificial intelligence (AI)

Accordingly, scientists at the Joint Research Center - a center of knowledge and science research under the European Commission, described in detail how real-time reporting information is posted by users. on social media platforms (especially Twitter) can be effective for the European flood warning system (EFAS).

In fact, this work was built primarily based on the inspiration from three other research projects that were successfully conducted earlier. The first is that studies were published by Harvard University and Google in August 2018, describing in detail a model of AI capable of predicting the location of aftershocks within a year after one Great earthquake appeared. The second is another study done by Facebook AI researchers in December, which successfully developed a method to more effectively analyze satellite images through intellectual models. artificially, thereby helping to quantify damage from widespread forest fires and other natural disasters more accurately. In addition, scientists at Google recently announced a retrospective of a machine learning system capable of accurately predicting the flood situation in rivers with accuracy. up to 75%.

1. Amazon AI system helps to cut 15% of speech recognition error on Alexa

In another related study, computer experts in the UK used machine learning algorithms, which used tweets to find places where violence could occur in violent incidents. disturbances, and at the same time allow them to predict relatively accurately the time when major protests might take place, as well as identify those who instinctively behind the protests.

'Over the past decade, social media have emerged as a source of information closely related to disasters, and this has attracted researchers from many fields. Different areas are more interested in how to take advantage of this useful information. Through analysis and practical evaluation, social media platforms have demonstrated great potential in being able to provide timely, valuable information about development related to space and time of a crisis, or any catastrophe, as well as assistance in identifying important disaster-related events, 'the researchers said.

Back to the new EU study. If you do not know, EFAS is part of the Copernicus Emergency Management Service (Copernicus EMS) and is operated directly by the European Commission's Emergency Coordination Center (ERCC). At the same time, the ERCC is also part of the European Commission, established to be responsible for humanitarian aid and protection activities, as well as to support response responses before, during and after disasters occur both inside and outside Europe. More specifically, the main task of the ERCC is to monitor potential hazards and risks, collect and analyze disaster data to prepare plans for timely implementation options. . In addition, the ERCC will also provide forecasts for EFAS - mainly forecasts of storms and floods, seasonal weather forecasts, as well as early impact assessments and warnings.

1. Disney's AI model creates animated sequences from scripts

In general, the researchers' warning system is responsible for determining when the risk of flooding in a given geographic area exceeds the permissible safety threshold. This has led EFAS's team to come up with the idea of ​​collecting relevant data from social networking sites, especially Twitter by adjusting and filtering up to 400 keywords at once.

However, extracting tweets with related keywords (ie words that may indicate information about an imminent or recent flood) is not an easy task for the Researcher at EFAS, because Europe is simply a large area with a population of over 741 million people and using 27 different languages. The solution here is to use a multi-lingual classification system. This classification system will use nonverbal mathematical representations, or embedded words, to infer the similarities between the keywords in the four main languages ​​in Europe, namely German, English, and Spanish. and french.

This system is actually a machine learning model, and in order to train it, scientists had to use a database of more than 7,000 annotated messages (from 1,200 to 2,300 messages for each language language). Meanwhile, they also use a separate model to deliver 'representative' messages (tweets with at least 90% of the likelihood of flooding) to areas at risk of flooding. prediction.

1. EU: AI systems should be developed and implemented in a transparent and responsible manner

To test the feasibility of this approach, scientists integrated SMFR into EFAS and deployed it in flooding affecting Calabria, Italy, in early October 2018. SMFR was collected. A total of 14,347 tweets were valid for 2 days, then proceeded with relevant analysis. The research team reported that these AI-filtered messages have a very close correlation with the actual flood situation, and at the same time it is a promising start to a system. can significantly shorten the response time in the early stages of a disaster:

'During the course of any disaster, the messages collected can bring tremendous value to international rescue coordinators, as they contribute to providing greater insight into counter-reactions. Local specific applications, and about situations where people affected by a disaster or disaster warning may face. For future research activities, we can imagine a similar system that is applied on a global scale, including dozens of different languages, and at the same time promoting a lot of use. more and more different social media platforms such as data sources can provide information for real-world predictions based on artificial intelligence. '