There are some posts on social media that most users instantly dismiss as fake news. Fortunately, a large group of people are able to separate the wheat from the chaff. But the more reliable the source of the messages looks, the harder it becomes to distinguish fake news from the real thing. Scientists at the German universities of Göttingen and Frankfurt and the Jožef Stefan Institute in Ljubljana (Slovenia) now seem to have found an answer to this.
The events in the United States over the past few months have made it clear that even democracy may be at risk if people can no longer distinguish between genuine news and false reports. Nevertheless, there are other dangers at play as well. For example, criminals are increasingly spreading fake news about companies in order to manipulate share prices.
German and Slovenian researchers have developed a technique that can recognise these kinds of fake news. Even when the news content has been repeatedly changed. The results of the research were published in the Journal of the Association for Information Systems.
Criminals spread fictitious material on social media that portray certain companies in a positive light. The scientists used machine learning methods to detect these posts. They developed language models that can detect suspicious messages on the basis of their content and certain linguistic characteristics.
“By doing this, we look at other aspects of the text that make up the message, such as the comprehensibility of the language and the mood that the text conveys,” explains Professor Jan Muntermann from the University of Göttingen. In principle, this technique is already known through its use in e.g. spam filters. However, the main problem with the current methods is that fraudsters are constantly adapting the content and avoiding certain words that are used in the identification of fake news.
Reading between the lines
This is where the researchers’ new strategy comes in. The new models read between the lines, so to speak, and they are still capable of concluding that a certain message is fake news based on the composition of the text.
“Therefore, even if ‘suspicious’ words are deleted from the text, the fake news is still recognised on the basis of its linguistic characteristics. This poses a dilemma for scammers. They can only avoid detection if they modify the mood of the text. For example, by changing it from positive to negative,” explains researcher Michael Siering. “But then they miss the mark in their attempts to persuade investors to buy certain shares.”
This new method can be used, for example, to monitor stock exchange trading. For one thing, whenever false information is discovered, trading in certain shares can be halted temporarily. It also provides investors with invaluable information to help them avoid falling for such scams.
Also interesting: Detect fake news automatically, with software from Fraunhofer