Automatic analysis of tweets for a rapid description of flash flood intensity and their effects

img projet

Name of the project : Automatic analysis of tweets for a rapid description of flash flood intensity and their effects

Type of project : Internships

Type of contract : Internship

Summary: It is often hard to get a quick grasp on the extent of the consequences of natural disaster, especially with information trickling in piecemeal from the field. And yet disaster management – starting with relief efforts and disaster assistance – must be organized based on this diagnosis. Over the past ten years, experience has shown that the occurrence of natural disasters often results in a rapid and massive dissemination of messages on social media, especially Twitter, whose short message format is particularly suited to instantaneous expression of live testimonials. Like other unpredictable rapid phenomena such as earthquakes, flash floods in particular lead to marked spikes activity on Twitter, with an explosion of tweets in the minutes following its occurrence.

This response could be useful for calibratingdetection models that are capable of automatically identifying flash flood events and estimating their impact. In France, the BRGM (Bureau de Recherches Géologiques et Minières) recently headed the development of a participatory platform for the semi-automatic analysis of natural disaster-related tweets to promote this rapid feedback of information by “citizen sensors”. The platform (, which the intern will use for his work, will offer a robust tool for collecting tweets related to flash floods, as well as automatic enrichment functions (on-the-fly classification of information and geolocation) as part of the internship

As such, the internship has five objectives:

  1. To establish the main characteristics of flash floods for which useful information should be targeted in messages exchanged on Twitter;
  2. To provide an overview of the latest developments in automatic tweet analysis methods relative to flash floods and other similar, fast-moving phenomena;
  3. Build a recent flash flood reference catalog for France and, based on this, an automated Tweet analysis strategy (a catalog gathering Tweet datasets as well as reference data regarding the intensity of the phenomenon and its effects). This catalog could be based on the case studies already identified within the framework of the ANR PICS projectand selected on the criteria of intensity of the hydrological reactions observed and intensity of impact;
  4. To propose an adaptation/hybridization of existing methodologies (“topic modeling” by supervised/semi-supervised/unsupervised classification, automatic language processing, etc.);
  5. To demonstrate/prove the feasibility of some of the tweet datasets gathered in the reference catalog. This notably will involve illustrating how the information gathered can be used to summarize the intensity of the crisis at different spatial scales (waterway segment, municipality, watershed, department, etc.). The possibility of identifying flooded areas from this information could also be assessed
  6. Name of supervisor(s) : Olivier PAYRASTRE (GERS-EE/UGE Nantes) / Cécile Gracianne (BRGM) et Faïza Boulahy (BRGM)

    Name of student(s): Axel Rambaud (Université de Nantes)

    Start and end of the project : 17 March 2020 - 9 July 2020

    Duration : 4 months