This AI-based system can detect the users which spread misinformation on Twitter

The developers all around the world evaluated the vocabulary used by the users to spread the news.

An AI system has been designed by the Researchers at the University of Sheffield, the social media user who tried to spread misinformation will be identified by the AI system even before the post is actively shared.

Researches discover that users who usually post content from inaccurate sources, their tweets are mainly on religion or political issues. And others who rewrite credible sources talk most about their personal lives.

Co-author Dr. Nikos Aletras, a lecturer in Natural Language Processing at the University of Sheffield stated “We also observed that high level of political violence is related to the use of hateful language and the spread of misleading content.”

After reviewing more than 1 million tweets from about 6,200 Twitter users, the team announced their results.

The team starts their analysis by gathering tweets from a list of Twitter news media pages, these tweets would be rated as either reliable or inaccurate in four categories: satire, deception, hoax, and clickbait.

Then they used the Twitter public API to get access to the latest 3200 tweets for each category and extract all the retweets related to those posts so only original posts were left behind.

Next, satirical pages, such as The Onion, which have comedic rather than misleading content, were excluded from the list of 251 credible sources, such as the BBC and Reuters, and 159 inaccurate sources, including Infowars and Disclose.tv.

Then, the team put approximately 6,200 Twitter users into two distinct groups: in one group all those Twitter users who post inaccurate sources more than three times are placed and in another group, those users are there who only post content from reliable sources.

Finally, the researchers with the help of vocabulary skills in the tweets train a set of models to determine whether the individual is going to propagate misinformation.

The most powerful approach used was a neural network called T-BERT. The team claims it can determine with 79.7 percent accuracy when in the future the user will post inaccurate sources: “This reveals that neural models will instantly disclose (non-linear) correlations between users created textual information in the data and the frequency of that users to repost tweet in the future from credible or inaccurate news sources.”

The team also conducted an analysis on the language features to identify variation in the language of the two groups.

They observed that the most commonly used terms from the users with false sources are "liberal," "government and media" and sometimes related to Islam or Middle East politics. On the other hand, people who posted trustworthy sources often tweet about their social experiences and feelings, using words like "mood," "want to and "birthday."

Researchers expect that their studies will help social media platforms to fight against misinformation.


Photo: LIONEL BONAVENTURE via Getty Images

No comments:

Post a Comment