-
-
A rough schematic of how our project works.
-
The Official Transparrot Twitter Bot Page
-
-
-
Inspiration
Over the past few years, the issue of misinformation has become increasingly apparent. While there are many fake news operations that contribute to this issue by spreading lies, many mainstream news outlets also worsen the issue with bias. The effects of this have been shown to be disastrous, with the anti-vax movement, QAnon, and even recent presidential elections.
Currently, there is no way to report on current events without some level of bias, especially in politics. Bias is a part of human nature, therefore any work created by people will be made with an ulterior motive in mind. That is why for this hackathon, we used artificial intelligence to pick up where humans have failed.
Enter Transparrot, an innovative, multiplatform webapp solution to solve the misinformation crisis by translating government press briefings and other sources into quick, easily-digestible government news.
What it does
The name "Transparrot" comes from translating parrot, as it translates and repeats information from reliable sources.
Transparrot is a progressive web application with a companion Twitter bot created to help combat the spread of fake news. It utilizes GPT-3, a state-of-the-art Natural Language Processing model, to summarize and spread official briefings directly from various governments around the world.
There are two distinct parts to this project: the Twitter bot and the website itself. The Twitter bot posts summaries every time a new press briefing announced. This is done through monitoring official government Twitter accounts for tweets that announce a new briefing. Once a tweet such as that is found, the Twitter bot web-scrapes the text of the press briefing using and summarizes it using GPT-3 with an OpenAI API call. The summarized text is then tweeted with a label including the date and origin.
The other part of this project is the website. The website contains a collection of all the summaries posted by the Twitter bot as well as allowing the user to create summaries of whatever news article they would like.
Through the blending Twitter and our site, we hope to inform as many people as possible with news people can trust. As it only parrots what governments officially announce, Transparrot will be virtually immune to bias as it has no emotion, cannot be paid off, and only summarizes factual information from governments.
How we built it
After hours of brainstorming and planning, we decided to tackle this challenge by dividing the project amongst ourselves. Adam and Ben worked on creating the Twitter bot and GPT-3 summarization while Sung-Jon and James created the website.
We coded the entire web app in the following languages/frameworks: We used HTML/CSS/JS, along with Jquery and Bootstrap to design the website hosted on Github Pages. The frontend calls the backend through an AJAX request. The backend is PHP and Node.js hosted on a Rasberry-Pi Server and an ASUS Laptop (it's complicated!). We created the Twitter Bot with Python, specifically the Tweepy library coupled with the Twitter API. We also used Requests and Beautiful Soup for the web-scraping and GPT-3 to summarize the scraped text. All data was scraped from official government sites.
Challenges we ran into
The primary challenge we ran into was creating the Twitter Bot. None of us had any experience making one and we were not able to get our developer accounts verified in time for the hackathon. However, we persevered and have a working bot ready to go live as soon as our developer accounts get verified.
In addition, we only got access to the closed GPT-3 beta extremely recently and had never before worked with a machine learning tool of that caliber. Implementing it into both a Twitter bot and website was a huge challenge. We are all proud of the hard work we put in to overcome these hurdles and create a finished product.
Accomplishments that we're proud of
We are incredibly proud of how our team creating a distinctive yet viable solution to fight the misinformation crisis. We are extremely proud of developing a solution that has never been previously considered or implemented with a transformer model. Most importantly, we were able to achieve our goals for the weekend by completely finishing our app, which we found unbelievable.
What we learned
Our team found it fulfilling to use cutting-edge machine learning tools to help combat one of the greatest issues of our time: misinformation. We believe that our project is ready to have a real impact on society after just 24 hours of development!
From a software perspective, we were proud to make our first website that heavily depended on API calls. No one on our team had ever really worked with these, so they were quite intimidating to implement on the scale we ended up using them. We are also proud of using a Rasberry Pi as a unique way to host our website's backend.
What's next for Transparrot
Currently, Transparrot only has support for the United States in summarizing press briefings automatically. We hope to add more and more countries over time and implement some of GPT-3's lesser-known features, such as semantic language translation. As it grows, we would likely have to change our Rasberry Pi hosting to something more scalable, such as Google Cloud. The most important next step however is to simply grow a larger following for Transparrot, so that more people have access to unbiased news.
Side Note Password will be LAHacks2021!!! for the website demo
Log in or sign up for Devpost to join the conversation.