Every day, we see blatant and dangerous lies about the coronavirus pandemic posted online by strangers and family alike. We wanted to help people sort out misinformation to help them keep safe in a data-driven way.
What it does
Covfefe-19 is an API packaged into a browser extension which helps people validate the information they see online about COVID-19. It takes in statements about the coronavirus from users' browsers and returns relevant sources -- including how similar each source's information is to the statement -- along with our own summary statistics (credibility, toxicity, etc.). In the browser extension, users can see all of this data to keep them informed in the face of so much misleading information.
How we built it
The frontend is built using React packaged into a browser extension. All backend infrastructure is hosted entirely in AWS, including serverless Lambda functions and a deep learning model (BERT) hosted in EC2. The backend is used to run searches on the data and generate scores for different aspects of the information.
Challenges we ran into
We ran into a few issues with how browser extensions access data from your window, but we managed to sort it out with some creative listener methods. We also had some problems with efficiently containing the backend services in serverless functions, but we ended up splitting the workload to reduce waiting times.
The main issue was definitely figuring out how to actually validate random statements about COVID-19; after some research and brainstorming, we came up with some interesting methodologies and statistics that leveraged Google's search algorithm, BERT (a deep learning language model) to measure text similarity, and compilations of trustworthy/untrustworthy sites. We didn't want to force subjective opinions onto our users, but rather provide them the tools to better inform themselves.
Accomplishments that we're proud of
The thing we're most proud of is that we tackled and solved a genuine problem our friends and family face. We've already had many of them test our tool, and they'll continue using it to stay informed.
What we learned
The main thing we learned as a team was how impactful online misinformation can be. A single tweet or WhatsApp post can put dozens or more in danger, but it's also not easy to conclusively judge a subjective issue.
On top of that, we also learned more about our technical areas. Sidharth and Arjun learned more about using EC2 to serve a deep learning model, and Meghna learned about how to use browser extensions to positively impact UX for online users.
What's next for Covfefe-19
The next step would be to get our tool in as many hands as possible and get feedback. We want feedback on performance, usefulness, user experiences, everything. We specifically think we can improve some of the statistics and information we provide, but we definitely want to get as much data as possible.