-
-
The app gives you a simple field to paste in the URL from any website you've read
-
It then generates a report based on expert.AI's analysis of emotional language and hate speech.
-
Some more details on the report.
-
You can even see a history of everything you've read, letting you compare all the different news outlets in your life.
-
GIF
The app in action
-
The app has a reactive layout so it works great on a tablet or computer, as well as your phone.
Inspiration
We know through court cases, etc. that some websites (e.g. Infowars which was recently fined $965m over hoax claims) can be disreputable and we know to take the information on those websites with a pinch of salt. This works for big news websites like Infowars, but does not scale well. For example, how are we to know whether a search result from an obscure website is trustworthy or not? Plus, doomscrolling, or seeing negative article after negative article, is also bad for our mental health, but we seem to consume so much without realising. What if there's a way to know if the news we're reading is bad for us?
What it does
It turns out that there are certain tell-tale attributes that tend to come with untrustworthy news sources: hate speech, and fanning negative emotions such as fear and outrage. We use Expert.AI's best-in-class hate speech detection API along with its emotional traits API to find the densities of these elements that indicate an untrustworthy article to compute a score, and compare those with websites known to be relatively impartial (e.g. the BBC) to give each article an easily interpretable grade.
How does it work?
News Report was built using the Expert.AI API, Python (Flask) for the backend, and Svelte for the frontend. This is then deployed to Microsoft Azure using docker.
Accomplishments that we're proud of
Making something that works in such a short amount of time, and is useful for the wider public.
What's next?
Prototyping and making this app was so much fun and it actually seems to be something that's pretty useful from our limited testing, so we'd like to develop this further, perhaps into a browser extension so that we can integrate it seamlessly with any web browsing we do.
Log in or sign up for Devpost to join the conversation.