Identifying Twitter Troll Messages Using This New Strategy

In the internet world, a troll is someone who starts flame wars to intentionally upset people through posting off-topic, extraneous, and digressive messages in social media sites, blogs, forums, or chat rooms.

But in recent years, the word "troll" has slowly become synonymous with online harassment as most of them aim to cause grief to families. For instance, Twitter has lots of trolls, and so the company decided to create a new feature in its app to combat these trolls.

There is also a new strategy developed by Sergei Monakhov of Friedrich Schiller University in Jena, Germany, in identifying these troll messages to help inform efforts to combat hybrid warfare while still preserving the freedom of speech.

Cyber Security Concerns In The Global Wake of Hacking Threat

(Photo: Getty Images)
LONDON, ENGLAND - AUGUST 09: In this photo illustration, the logo for the Twitter social media network is projected onto a man on August 09, 2017, in London, England. With around 328 million users worldwide, Twitter has gone from a small start-up in for the public 2006 to a broadcast tool of politicians and corporations in 2017. (Photo by Leon Neal/Getty Images)

Identifying Twitter Troll Messages

On August 12, Monakhov presented his findings of Twitter troll messages in the open-access journal PLOS ONE. According to him, there are two algorithms that account for the distinctive use of repeated words or pair of words requiring as few as 50 tweets to distinguish Twitter troll messages.

In 2018, 13 Russian nationals were accused of using fake personas to interfere with the 2016 US presidential elections through social media posts.

Although previous research has looked into the common characteristics of troll messages, such as the timing, hashtags, and geographical location, only a few of them have investigated the linguistic features of the tweets.

That is where Monakhov's studies were focused on. He made the study assuming that trolls have a limited number of messages but needed to do so multiple times using diverse wording and topics to fool the readers.

Comparing the Russian troll tweets and genuine tweets from the US, Monakhov showed that these troll messages lead to distinctive patterns of repeated words and word pairs that are different from the ones seen in genuine tweets, or the non-troll tweets.

He then used an algorithm that uses these distinctive patterns to distinguish which are the genuine tweets. Monakhov found that the algorithm required 50 tweets to correctly identify troll tweets from tweets by famous people such as Donald Trump.

The Purpose of Identifying Troll Messages

According to Monakhov, this new strategy for quickly identifying troll messages could help those researchers who try to combat the massive troll messaging online while also preserving the freedom of speech.

He advises doing further studies to determine whether the algorithms used infamous people in determining trolls can be applicable to non-famous social media users.

He added that although troll writing is thought of as being permeated using recurrent messages, its most distinguishing characteristic is the anomalous distribution of repeated words and word pairs, creating messages that aim to cause harm to families.

Using the ratio of the proportions of these repeated words and word pairs as a quantitative measure, one can easily distinguish genuine tweets from troll messages.