Fakebuster’s Dream? Building The World’s Largest Public Fake News Database

Algorithms and crowdsourcing are the recipe for this brand new platform, which focuses very closely on the content consumed by the “gen Z”. Explore.

The Snipfeed team is a player: its objective is to interest people under 25 in the background news, assuming that they spend about seven hours on a smartphone, and have only six seconds of attention to content. That’s a hell of a constraint. But Rédouane Ramdani and his fellow travellers, two other French people from Silicon Valley, are convinced that Generation Z has a strong desire to learn and grow: “They are very committed to the challenges of society. The problem is that traditional sources of information are being abandoned because their formats are not suitable. In the United States, the first two media for gen Z are Instagram and Snapchat…”, says the entrepreneur. The young media company Snipfeed therefore took the codes of its latter, with vertical and horizontal videos, podcasts and articles cut into blocks of 400 characters maximum, and opted for a model of information distribution closer to a Spotify than a New York Times. All that was missing was a brick, which had become essential: cleaning the content base of fake news teeming on the web and social networks. That’s how Fakebuster was born.

“We offer them – publishers – a new way of monetizing and, thanks to Fakebuster, their articles will not be lost in the middle of fake news.”

Developed by a machine learning researcher at Stanford, Fakebuster artificial intelligence has learned to recognize a fake news by training on hundreds of thousands of articles. And his learning continues, since Internet users can now submit URLs of US sites to him. This algorithmic lie detector assigns a reliability rate to articles via a scoring system. The first of these is the writing style. “We know, for example, that slanderous sites use a more aggressive vocabulary, and a rather similar syntax, with many exclamation marks…”, explains Rédouane Ramdani. The AI also hunts down the “clic bate” by analyzing if there is a correlation between the headline and the body of the text. The third score is based on fact checking. It has two sub scores: reliability of the source and matching of the facts mentioned with the results found in the first nine pages of Google.

The ultimate goal is to build the world’s largest public fake news database. To achieve this, the startup also uses crowdsourcing: users are invited to give their opinion on the submitted articles: “Pretty sure this article is real” or “hmmm… I think it’s a fake”. The tool, integrated into the global Snipfeed content platform, would be of interest to major American publishers: “They like our model because we can offer their content in a personalized way to young people, we offer them a new way of monetizing and, thanks to Fakebuster, their articles will not be lost in the middle of fake news,” the team is pleased.

“Pretty sure this article is real” or “hmmm… I think it’s a fake”.

Accelerated at UC Berkeley, the startup is currently in its first fundraising round. A version adapted to the French-speaking community of its platform is currently being tested in Algeria. Snipfeed also plans to set up its AI research centre in Paris. Europe seems to appeal to investors… a destination where the salary of a good AI engineer is not as extravagant as in California, and where you can get interesting subsidy support.

Leave a Reply

Your email address will not be published. Required fields are marked *

Related Posts