Detecting fake news on social media: a new approach

The spread of misinformation on social media poses a significant threat to our society. A scalable strategy to detect and ultimately prevent the spread of fake news is therefore urgently needed. In his Master’s thesis, carried out at the Chair of Management Information Systems, MSc MTEC graduate Francesco Paolo Ducci proposes a new algorithm to detect fake news on social media based on crowd behavior.

detecting-fake-news
"Information spreads on social media when posts or tweets are continually shared by users. This creates a cascade."

Whether Facebook, Twitter or Instagram, social media is one of the most important sources of information for many people today. However, a large number of users are exposed to false stories on topics such as electoral debates and medical discussions as well as conspiracy theories. Indeed, during the 2016 U.S. elections, social media users interacted more with fake content than with real news. Algorithms that quickly and effectively detect and respond to misinformation are therefore crucial to ensure the reliability of social media as a source of information.

Given the importance of the topic, much research is devoted to the investigation of misinformation spreading on online social media. However, most current research is limited to descriptive studies or statistical models based on aggregated data. Stefan Feuerriegel and Mathias Kraus of the Chair of Management Information Systems at D-MTEC worked with Francesco Paolo Ducci to develop a novel deep neural network that detects misinformation on Twitter based on its propagation dynamics: The Cascade-LSTM.

Information spreads on social media when posts or tweets are continually shared by users. This creates a cascade. Compared to models that reduce the cascade to a set of aggregate statistics (e.g. depth, number of users involved), the Cascade-LSTM is carefully designed to investigate the full structural properties of the cascade. This not only allows the model to perform better at detecting misinformation, but also to detect it only a few minutes after publication. This lays the foundations for early intervention in the spread of fake news.

Francesco Paolo Ducci’s paper, "external page Cascade-LSTM: A Tree-Structured Neural Classifier for Detecting Misinformation Cascades" has been accepted for presentation at the ACM conference on Knowledge Discovery and Data Mining (external page KDD 2020), the premiere venue for applied data science with an acceptance rate of only 5.8%.

JavaScript has been disabled in your browser