Fraunhofer uses machine learning in tool against fake news

The German Fraunhofer Institute has developed a tool that recognizes fake news. The tool uses machine learning to analyze metadata and language errors to give an indication that it is fake news.

Fraunhofer FKIE’s tool targets Twitter posts and the potentially fake news sites that tweets link to. As a first step, the institute’s research team developed databases containing serious reporting on the one hand and messages that users labeled as fake news on the other. Based on these datasets, the system trained with machine learning to distinguish features of fake news.

For example, the system pays attention to metadata. “Metadata provide critical leads”, according to prof.dr. Ulrich Schade, scientist at Fraunhofer FKIE. This concerns, for example, the transmission frequency and the time at which messages are sent to uncover bots. The Fraunhofer tool can also map and display networks of account followers in the form of heatmaps and other visualisations. The structure of networks and nodes shows where a fake news campaign started.

The software further classifies formulations and word combinations on a semantic level. Wording in a certain context may indicate intentional hate speech and language errors may indicate that the message originated in a different country than the language in which the message was written. According to Fraunhofer, the software can be adapted and trained for specific purposes. The tool could be used by both companies and governments and could also warn at an early stage that a fake news campaign has started.

Twitter and Facebook, among others, are fighting fake news. Last weekend, it emerged that the Snopes organization was ending its partnership with Facebook for fact-checking news. The work is very labour-intensive, the company says. The Associated Press is also unsure whether it will continue with the same work for Facebook.