Even though we live in a world in which information has never been more accessible, that has the downside of misinformation because anyone can post anything, and anyone can access any post with any content. Regardless of generation, people are unfortunately pretty easy to trick, including me. I was on TikTok and saw people post "Tweets" from the Iran Military about recent events. Looking back at it now, it would make no sense for a foreign country to tweet about war. I thought that I would never fall for the same tricks that moms on Facebook easily fall for, but I was totally wrong.
FactChecker Pro is a Chrome extension that performs the work that you need to do to verify a certain claim. It looks through at the important claims on a page, and then essentially performs a Google search and webscrapes some of those searches (this is simplified, as webscraping Google searches is not allowed). Then, it looks at those websites and checks for those claims and verifies it with whatever you are seeing on your website.
We built the extension around two primary JavaScript files: content.js and fact-schecking-service.js. The content.js file is the content script that runs directly on the web pages the user visits. It scans the page to find text elements, identifies statements that are facts, sends the facts to our fact-checking-service, and finally receives the results and displays them in the proper format. The fact-checking-service.js is the core backend file that contains all the logic for evaluating a statement's accuracy. This file extracts claims using regex, analyzes it through a series of search functions that verifies the information against various reliable sources, and then assigns it a proper confidence score. We used a couple of Google APIs, such as the Google Natural Language API for the analysis of the text to find the most important words (noun, verb, settings, etc.)
There was an attempt to use the Google Fact Checker and Google Custom Search APIs, but those had a lot of restrictions because we don't have the special perms to essentially take the work that Google is doing. Furthermore, our CSS had conflicts with a lot of website's CSS (overriding issues), so even though the fact-checking algorithm worked, it looked like we used Inspect Element and ruined an entire website's design. Furthermore, we ran into some issues about the accuracy percentages. For example, clearly incorrect statements were underlined with green (meaning high accuracy) because there weren't enough articles that stated otherwise.
Luckily, we ended up finding solutions to most of the problems that we ran into, and if we didn't, we found ways to kind of make such problems less apparent.
We learned that Cursor and Anthropic are awesome and vibe-coding is a cost-efficient practice. Without AI, this project is something that would take up an entire summer.
What's next? Well, we want to make FactChecker Pro even more accurate. We also want to maybe provide confidence intervals and a word similarity percentage to make it more credible.
Log in or sign up for Devpost to join the conversation.