Deep-learning tool detects whoppers with 90 per cent accuracy
Image credit: Getty images
University of Waterloo engineers have developed a machine learning tool which could be deployed by social media companies and news companies.
The tool uses deep-learning algorithms: a type of machine learning algorithm which processes data through successive layers to extract increasingly meaningful and complex information.
This algorithm – which the researchers were motivated to create by the proliferation of politically motivated viral deception online – determines whether claims made in news stories or social media posts are supported by other content on the same subject.
“If they are: great, it’s probably a real story,” said Professor Alexander Wong, a systems design engineering expert at the University of Waterloo. “But if most of the other material isn’t supportive, it’s a strong indication you’re dealing with fake news.”
The algorithm was trained with tens of thousands of claims paired with stories that either supported or rejected them.
The researchers tested their system using a dataset created for the 2017 Fake News Challenge. It achieved 90 per cent accuracy in an area of research known as “stance detection”; given a single claim in a piece of content to compare with other stories on the same subject, it can correctly determine whether the claim is supported or not nine times out of ten.
This achievement is an important step towards developing an accurate and fully automated system for detecting fake news.
For now, the Waterloo engineers suggest that their system could be used as a tool by human fact-checkers: “It augments their capabilities and flags information that doesn’t look quite right for verification. It isn’t designed to replace people but to help them fact-check faster and more reliably,” said Wong.
Chris Dulhanty, a graduate student who led the project, commented: “We need to empower journalists to uncover truth and keep us informed. This represents one effort in a larger body of work to mitigate the spread of disinformation.”
Social media companies are under pressure to crack down on appropriate content, including politically-motivated fake news, on their platforms. Fake news on social media has been characterised by British politicians as a threat to democracy.
Platforms like Facebook use a combination of human moderators and automated systems for identifying and taking down this content. Their efforts are supported by fact-checking services like FullFact, researchers and panels of established news media organisations.
However, Facebook has received criticism for failing to tackle the problem effectively and outright refusing to remove lies posted by politicians.
Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.