To Find Trolls From Tweets

A new algorithm developed by a researcher in Germany uses as few as 50 tweets to differentiate trolls from genuine tweets on Twitter.

An illustration of trolling on Twitter
Representative illustration (Source: Derreck Johnson/iStock)

A researcher from Germany has developed a new algorithm which can differentiate between troll tweets and genuine tweets using as few as 50 sample tweets. This can help Twitter differentiate between genuine users and trolls, a problem they have been tackling for a long time. Researcher Sergei Monakhov used the sociolinguistic characteristics of tweets to create two computer algorithms that can differentiate troll tweets from that of notable public figures.


Trolling in Internet terms refers to certain miscreants trying to incite hate, flame wars or intentionally upset other users on social networking websites. In a world where everyone has an opinion to share online, trolls have become a menace for genuine users of websites such as Twitter. The power of these troll accounts go so far as to being responsible for making or breaking a government, well illustrated by the 2017 United States Presidential election. Many attempts have been made by Twitter to curb such activity waging a war against trolls to promote ‘conversational health’ on the platform. It has previously used features such as reply limiting and hiding responses to keep internet trolls at bay.


Previous efforts to curb trolling have involved monitoring hashtags, tweet timings, mentions, etc. of known troll accounts. However, Sergei’s algorithms look at the contents of the tweets themselves. The researcher used a large sample of tweets from known Russian troll accounts and tweets from members of congress in the United States. He used the algorithm to show distinctive patterns which can easily identify a troll’s tweet from that of US Congressmen. The data obtained showed patterns of differences in the choice of words to the pairing of words in the tweets. On testing the algorithms, Manakhov found that only 50 tweets were needed to detect these patterns and find the trolls.


Such a method of curbing troll tweets and accounts could in theory help prevent hate speech, fake news, cyber bullying and the use of social media for political gain. The method can also in the future be implemented in different areas such as malicious messages and cyber fraud detection. However, it is going to be a while before a more refined version of this algorithm is deployed on a large scale. At the moment the algorithms are capable of differentiating only public figures from trolls ruling out regular Twitter users. Further research is on regarding the accuracy of the algorithms in pointing out troll tweets from tweets of genuine users who are not public figures.


The news has re-sparked the debate of Twitter’s approach of allegedly curbing free speech and suppressing tweets which carry strong or unpopular opinions. However, considering the amount of fake news circulating on social media platforms today, Manakhov’s research comes as a blessing to netizens providing a safer and better surfing experience. Looking at the above from the context of an Indian user, such an algorithm would curb the constant practice of hate speech on the platform while also keeping state backed troll accounts at bay.

0 comments