As fake news has spread, so too have the number of intended solutions to the problem.  The latest of these comes from a team from the Fraunhofer Institute in Germany.  Their approach utilizes both the content of the story and its metadata to supposedly filter out fake news and disinformation with a high degree of accuracy.

The classification tool the team developed is capable of processing huge quantities of social media data, whilst also taking into account the metadata associated with content.

“Our software focuses on Twitter and other websites. Tweets are where you find the links pointing to the web pages that contain the actual fake news. In other words, social media acts as a trigger, if you like. Fake news items are often hosted on websites designed to mimic the web presence of news agencies and can be difficult to distinguish from the genuine sites. In many cases, they will be based on official news items, but in which the wording has been altered,” the team explain.

Spotting fake news

The researchers began by building up a library of serious news articles alongside a similar library of stories that had been identified as fake news.  These libraries were used to train the machine learning based system to look for specific markers in both the text of the article and its metadaa.

“When we supply the system with an array of markers, the tool will teach itself to select the markers that work. Another decisive factor is choosing the machine learning approach that will deliver the best results. It’s a very time-consuming process, because you have to run the various algorithms with different combinations of markers,” the team explain.

The metadata associated with each article proved to be a valuable marker to differentiate between an authentic story and a fake one.  For instance, the team believe that things such as the frequency of posts being made, their timing and so on can be a giveaway about the authenticity of the content.  For instance, a high send frequency can be a telltale sign that it is bot activity at work.

By analyzing this metadata, the team are able to develop a kind of heat map to chart send frequency, follower networks and the send data.  These networks then offer insight into the source of the fake news.

The tool is also able to detect hate speech posted online.  The team believe that hate speech and fake news are often tightly linked, with one often leading to the other.

“The important thing is to develop a marker capable of identifying clear cases of hate speech. Examples include expressions such as ‘political scum’ or ‘nigger’,” they explain.

The team hope to offer their tool to the market, either to public bodies or private organizations.  They’re one of a number of efforts to tackle fake news (more of which in a later post), and with so many projects emerging, it is hopefully a problem that we’re increasingly getting on top of.