GoodNews · Fake news detection in social networks using geometric deep learning

In the past decade, social media have become one of the main sources of news for people around the world. Yet, it comes with the danger of exposure to intentionally false information. The extensive spread of fake news has recently become a centerpiece of controversy following the highly debated elections of President Donald Trump and the Brexit vote. It is alleged that the outcome of these votes resulted from the public opinion manipulation by a massive injection of fake news, possibly sponsored by hostile foreign governments, constituting perhaps one of the most serious and unprecedented threats to the modern democracies. The ambition of GoodNews is to build the technological capability for algorithmic fake news detection in social media using a novel paradigm. Instead of following the traditional approach of analyzing the news content, we will analyze the news spreading patterns in social networks. The algorithmic core of GoodNews is based on a novel class of geometric deep learning algorithms developed in the ERC project LEMAN (Learning on Manifolds and Graphs). Our research group was among the first in the world to propose, implement, and patent generalizations of popular convolutional neural network architectures to graph-structured data such as social networks, allowing to do deep learning on graphs. The ability to learn fake news spread patterns on social networks will provide the needed breakthrough in the task of automated fake news detection. GoodNews will convert the geometric deep learning technology into a commercial application of fake news detection in social media. The focus of the project will be three-fold: developing a demo system for fake news detection with real data from social media; verifying and solidifying our IP portfolio and its licensing terms; analyzing the market and coming up with a financeable business plan. We target establishing a company at the end of the project and attracting investment to develop a commercial-grade product.