Protecting Humanity with Technology
Human Rights First: Blue Witness
Blue Witness Program
United States of America has always been the country that represents freedom. As stated in the constitution, every citizen has the “…freedom of speech, freedom of assembly, freedom of religion, freedom from cruel and unusual punishment…” In order to protect the basic human rights and freedom of the citizens, Human Rights First came up with a program that focuses on the unjust police use of force… The Blue Witness Project
For more than three decades, Human Rights First has been a clarion voice in defense of human dignity and the rights and freedom of people everywhere.
— Susan E. Rice, Former U.S. National Security Advisor
Since technology has played a big role on fast sharing of information, everyone has a voice! Now, more than ever, we have the power to bring awareness about the things we care about. With that in mind, the Blue Witness program takes public information on tweets and reddit reports on police use of force, and consolidates them in an app for citizens to see the unbiased reports on police brutality happening in the United States by the Americans for Americans.
Data Science of Things
Bidirectional Encoder Representations from Transformers (BERT) Model
“BERT model is a transformer-based machine learning technique for natural language processing (NLP) pre-training developed by Google.” — wiki
This project aims to use the BERT model to identify the context and thought of texts instead of just looking at individual words.
By using this model, we can scrape data from twitter and properly identify them as police presence incident or not.
Data Collection
In order to come up with an accurate model, we need to scour thousands of tweets to train the BERT model! A chunk of our efforts were directed to coming up with a clean trainable dataset, big enough for the model to accurately identify police use of force.
Since it’s tedious and time-consuming to gather tweets one-by-one, we decided to use the current approved incidents, and searched for datasets with confirmed police brutality. To make our dataset balance, we also decided to scrape tweets that are related to police presence, but not police use of force such as reports on the black lives matter and other socio-economic movements with police presence that happened recently.
Scraping Tweets
Everyday, there are millions and millions of tweets in the United States alone. From the minute details of someone’s life to big global event happening in the communities, PEOPLE TWEET THEM! That’s the beauty of using the twitter platform! We can find ANYTHING! However, also because of this, we needed to put constraints to limit the tweets to police use of force.
Here are our main parameters:
- q: to search certain words to filter the tweets
- lang: limit the tweets in English
- geocode: set the coordinates in Kansas and set the radius big enough to cover continental United States
Results
Here is a working sample of the BERT model:
Stand-alone model:
- Input: tweet
- Output: Categorized police use of force
Vision
Human Rights First envisions the Blue Witness Project to incorporate a twitter auto-reply functionality in the future. Since twitter has turned off the geotag in tweets, the twitter bot functionality would have to interact with the user to get important information such as location, any other detail about the police presence.
The future holds a lot of opportunities for growth to move the protection of humanity forward. We are just starting the era where many of our human rights problems can be solved with technology. We can protect each others’ freedom of speech, freedom of assembly, freedom of religion, freedom from cruel and unusual punishment. Just imagine how much more we can use data to make the world better!
