The credibility of messages broadcast on social networks, often accused of allowing misinformation, could be assessed by a model based on the vocabulary used.
- The choice of words used in a Twitter message give an indication of the credibility of the information
- It is by analyzing Donald Trump’s tweets that researchers have developed a model to flush out “fake news”
- This model would be applicable to all online information sharing
Actors in political life have an interest in choosing the words they use when communicating on social networks. While France is about to enter the campaign for the presidential election for good, work carried out by the University of Rotterdam on the tweets broadcast by Donald Trump during his presidency has just shown how the vocabulary of political communication can reveal misleading or approximate statements. This study was published in the journal Psychological Science.
A model to predict whether information is true or false
Particularly verbose on the Twitter network – he reportedly posted more than 33 messages a day during his last year as President of the United States – Donald Trump offered researchers a reference of choice! They studied both the tweets of the American leader and the way in which they were analyzed by all those who tracked down the “fake news” that these messages could contain. As a result, they demonstrated that the choice of words used was different when the message was false or incorrect. And they thus determined a kind of universal model allowing to predict if a tweet contained true or false information.
“We have created language models that can indicate whether statements are misleading and these models can help detect lies in other contexts”, warn the researchers through the voice of Sophie van der Zee, first author of the article. And she specifies that in the case of Donald Trump, “his language was so consistent that in three quarters of the time our model based only on the use of certain words could predict whether his tweets were factual or not”.
Trump’s words true in three out of four tweets
It is by crossing three months of presidential tweets with the factual elements of messages broadcast by the Washington Post that the researchers created a model making it possible to affirm that the words of Donald Trump were true in three out of four tweets. To ensure that the data used did not constitute a bias that could distort the results of their work, they deleted all messages that did not reflect Donald Trump’s own vocabulary, i.e. quotes or retweeted remarks.
“Our paper serves as a warning to anyone sharing information online as we show that the words people use in this case may reveal sensitive information about the sender, including an indication of their trustworthiness”concludes Sophie van der Zee.
.
