Group-15

Flood detection from social media text

Overview :: Description :: Directory Structure :: Technologies :: Getting started
Results :: Conclusion :: Future Scope :: Video :: Group Members

Overview

Social media has emerged as a source of quick communication and information. This can be used as an information source for natural disaster detection and assessment. However, using social media for disaster assessment is difficult due to the lack of trustworthiness brought on by anonymity and uncertainty.

Many methods, including the use of textual and visual features, have been tested to enhance the detection of natural disasters in social media posts. The results demonstrate that the features have a positive impact on distinguishing flood texts. From metadata, we considered only the textual metadata.

Description

Recently, a significant number of individuals use cellphones and write about their daily lives on social media. The analysis of this immense amount of social media data has the potential to significantly improve response times in the event of a natural disaster.

The project's objective is to identify floods from a given text which is associated social media metadata. We intend to put into practice a model for flood detection that makes use of the metadata.

In order to create an effective model as part of the fusion, we would like to investigate various 12 Natural Language Processing techniques for feature extraction from the social media information.

Directory Structure

.github/workflows/
   python-app.yml
   pdoc-app.yml
docs/
  src
  proj1rubricComments.pdf
  proj1rubric.md
src/
  README.md
  Application/
      static/
          base.jpeg
          water.jpeg
          main.js
          style.css
      templates/
          index.html
          login_socialmedia.html
          predict.html
      App.py
      app.yaml
      model_prediction.py
  Training/
      bert+svm_flood_detection.ipynb
      training_model.py
test/
  README.md
  Web Results/
      Home.png
      login.png
      output_prediction.png
  __init__.py
  test_index.py
  test_login.py
  test_modelprediction.py
  test_predict.py
  test_return.py
  test_runner.py
.gitignore
.travis.yml
CITATION.md 
CODE-OF-CONDUCT.md
CONTRIBUTING.md
INSTALL.md
LICENSE.md
README.md
requirements.txt
setup.py         

Technologies

Python
Java Script
CSS3
HTML 5
Jupyter Notebook

Gettingstarted

• Prerequisite:

  • Download Python3.x.

• Installation:

  Steps to setup virtual environment

  • Create a virtual environment:

    python3.8 -m venv app_env

  • Activate the virtual environment:

    source app_env/bin/activate

  • Build the dependencies in virtual environment:

    pip install -r requirements.txt

• Instructions to Run the application.

  To run/test the site:

  • Clone Flooddetection github repo.

  • Navigate to project directory.

  • Run python3 App.py

  • Site will be hosted at:(localhost) http://127.0.0.1:3000/

Results

Conclusion

• Our Current Application takes post tile and description as Input.
• We trained our model using BERT + SVM machinelearning model.
• Depending on the inputs our trained machine learning model predicts the outcome whether the flood exists or not.

Future scope

• Moreover, due to advancement of social media, users now can write in these social media using their native language. So, an extension to a social media app will be of good use.
• Our current model predicts the flood using current text analysis, including the images along with the text could improve the accuracy of the model
• We have a limited training data for the model in our application. It can be improved by training the model with more data.

Video

GD996R2A7J7NGIXU.mp4

Group Members

• Nelluru, Dedeepya (dnellur)
• Kanamarlapudi, Venkata Gnana Vardhani (vkanama)
• Vengali, Sai Kumar Goud (svengal)
• Immidisetti, Ratan (rimmidi)
• Chirumamilla, Raviteja (rchirum)