FABULA AI announces breakthrough in algorithmic fake news detection
Fabula AI leverages patented Geometric Deep Learning to identify fake news with 93% accuracy within hours of dissemination, paving the way for the creation of the world’s truth-risk scoring house for news.
FABULA AI, the London-based technology venture, today announced breakthrough results in its efforts to leverage advanced machine learning for the automated and rapid detection of fake news on social media.
The extensive and systemic spread of “fake news” – stories published on social media containing intentionally false information – has been at the centre of global attention since the highly debated 2016 presidential elections in the United States and the Brexit vote in the United Kingdom. In both cases, fake news is alleged to have manipulated public opinion, and influenced the vote outcome. The fake news phenomenon is a global problem considered very seriously by governments as a threat to democracy and even a threat to human lives. It also presents potentially the biggest existential threat to the social media platforms Facebook, Twitter and YouTube, who face the prospect of government regulation and large-scale economic fines if they fail to bring fake news under control.
Solving fake news in a scalable and reliable way that preserves freedom of speech while containing the damage of misinformation, is possibly one of the biggest technological challenges faced by society today. Democracy, lives, and trillions of dollars of value are at stake.
However, automatically detecting fake news poses challenges that defy existing approaches based on linguistic content analysis. News are often highly nuanced and their interpretation requires the knowledge of political or social context or “common sense”, which current natural language processing algorithms lack. Furthermore, content-based approaches are language-dependent and in general can be defied by a sufficiently sophisticated adversary. That is why a lot of the efforts around fake news detection presently centre on manual, human fact-checking that is prohibitively expensive and impossible to scale.
FABULA AI takes a radically different approach to algorithmic fake news detection. Instead of considering the content of news, the technology looks at the manner in which stories spread on social media. Fake and real news spread differently, making it possible to find characteristic patterns telling them apart. The problem has been how to develop a technology that can use this insight to detect fake news rapidly, accurately, and at scale. The issue is exacerbated by the fact that conventional deep learning only works on grid-like data sets, but not on distributed networks like social media.
To solve this problem, FABULA have developed and patented a new class of machine learning algorithms called “Geometric Deep Learning” that are capable of learning patterns on complex, distributed data sets such as social networks. The underlying core algorithms are a generalization of convolutional neural networks to graphs that have been developed by the team over the past years.
Among the key advantages of FABULA’s technology compared to content-based methods is that it is agnostic to the content and language of news, and much harder to beat by adversarial techniques, as it relies on the collective behaviour of the social platform users. Geometric Deep Learning also allows naturally fusing heterogeneous data such as the social network connectivity, news spreading, user profile and behaviour.
Said Prof Michael Bronstein, co-founder and Chief Scientist: “FABULA’s engine is already able to detect fake news with extraordinary accuracy (>93% ROC AUC), within milliseconds of processing and after only a very short spread time (2-20 hours post publishing). By comparison, it takes a manual fact-checker on average 8 hours to clear one story. The performance of FABULA’s Geometric Deep Learning algorithm is unparalleled”.
Said Ernesto Schmitt, co-founder and President: “Today’s results pave the way for FABULA to become the world’s independent, decentralised truth-risk scoring platform for publishers, just like Clearscore and Equifax are the decentralised credit-risk scoring platforms for providers of consumer finance. From late 2019, FABULA aims to make available our APIs to third parties to obtain truth-risk scores in real time, for any hosted piece of content, against a configurable TPR/FPR confidence interval, allowing publishers and social media to flag content as suspect, remove it, downgrade it, or leave it as is – according to their own priorities and trade-offs.”
ABOUT FABULA AI
Fabula AI was founded in 2018 by Prof Michael Bronstein, Chair in Machine Learning and Pattern Recognition at London Imperial’s Department of Computer Sciences, Federico Monti of USI Lugano, Damon Mannion, and Ernesto Schmitt. The company employs 10 computer scientists and operators, and is headquartered in London, UK.