Save the Children (United Neighbors Mission)

Hacking for Human Rights 10/10/2020 - 10/24/2020, UVA School of Data Science

Problem Description

The goal of this hackathon is to explore and prototype different options for generating infrastructure damage values that can be applied to MDI’s predictive analytics model.

Our Contributions

• Explored extracting information from tweets scraped from Twitter
• Using a tweet scraper, we collected tweets matching a variety of keywords relating to conflicts and displacement
• We used a Named Entity Recognition tagger to extract tags from each tweet that captured the most important words from each tweet
• We explored two main primary sources taken from the IDMC and UNHCR datasets; performed data visualization and feature extraction

System Design

Case Study 1: Afghanistan

1. Collected tweets using the following queries:
   • Afghanistan, conflict
   • Afghanistan, refugee
   • Afghanistan, displacement
   • Afghanistan, crisis
   • Afghanistan, war
   • Afghanistan, conflict, children
   • Afghanistan, refugee, children
   • Afghanistan, displacement, children
   • Afghanistan, crisis, children
   • Afghanistan, war, children
2. Example Tweet Output: Example Tweet: this is heartbreaking!!!\nWe are the victims. No mother or father would want to see their children live their whole life with no legs. \nWar brings only misery. Let's bring an end to it. \n#Afghanistan #Farah #Kandahar #Taliban\n#Landmines #War #Children #ChildrenUnderAttack https://t.co/u4cAbCDVbQ
3. Extracted Tags: Afghanistan, Farah, Kandahar, Taliban
4. Queried for "Afghanistan" from primary sources
5. Parsed data into standardized format

Future Work

• Utilize more sophisticated text models to process tweets
• One way to enhance the information extracted from tweets would be to analyze images along with text from tweets
• This could make use of a captioning model to describe the image or using the image for geolocation using a model such as Google’s PlaNet
• Create map showing the prevalence of given topics in specific regions to help characterize how different areas have been affected
• Use map and location details to provide estimates of where people who have been displaced might travel
• Work to ensure data privacy, making sure the data extracted does not compromise the safety of those who have been displaced

How to Use this Repo

• data_exploration contains two colab notebooks, one for the UNHCR: Data on forcibly displaced populations and stateless persons: https://data.humdata.org/organization/unhcr?groups=afg&q=&ext_page_size=25, and one for the IDMC: Demographics and Locations of Forcibly Displaced Persons: https://data.humdata.org/dataset/unhcr-population-data-for-afg. These notebooks contain data visualizations and basic machine learning models for predicting number of displacements in following years. Note, your import path might need to change in order to access the datasets
• data_sources contains .csv files of the datasets used for this hackathon
• final_pipeline contains two colab notebooks, one for the extraction of primary data using the two sources and one for querying the Twitter API. It also includes an example of how to query for a specific country and returns standardized dataframe.

Detailed Overview of Research and Work

https://drive.google.com/drive/folders/1NOInwowDxQ53maoIytwMtRervMZt-yKQ?usp=sharing