We studied 1 million VKontakte profiles and almost 10 million posts
On March 27th, 2019, in the first round of the Ukrainian presidential elections, our company, in collaboration with Internews Ukraine, conducted research later published as ‘Chaos and Hate: What The Russian Social Network VKontakte Says About The Ukrainian Election’. We studied 1 million VKontakte profiles and almost 10 million posts to understand what people were saying about the nine most popular candidates and what was going on ahead of the Ukrainian elections.
The research for Internews Ukraine wasn’t concentrated on studying social bots; nevertheless, we decided to present these data (which were detected occasionally) in this case study.
Natural language processing and artificial intelligence approaches, and data research analysis.
How the social bots were detected
In the process of text clustering, we detected and studied 417 VK users that were writing about politics. Our script determined 152 clusters of topics in more than 9K texts. As a result of clustering, we identified 40 users who exhibited strange activity on VKontakte.
The distinctive feature of these profiles was frequent usage of the following words: ‘политика’ (politics), ‘канал’ (channel), ‘русский’ (Russian), ‘срочно’ (urgent), ‘Украина’ (Ukraine), ‘Путин’ (Putin). The pages of these VK users also contained posts about nature, art, religion, music, and reposts from neutral groups. So, given the massive amount of text, it would be hard to detect such accounts manually. The most suspicious aspects of these posts were the short time intervals between posts, as well as inauthentic photos.
We studied these accounts more closely and detected that the time intervals before the posts were 3, 5 or 10 minutes. It was clear that the posts were shared automatically, and that the accounts shared the characteristics of social bots.
The text content was critical of the current President of Ukraine, Poroshenko, and contained anti-Ukraine sentiments.
One of the detected social bot pages
We determined that these users belonged to the following groups: Новости RТ на русском (RT News in Russian), Русская весна (Russian spring), Украины больше нет (There is no more Ukraine), Новости ДНР (DPR News), Новости ЛНР (LPR News), МИР (Peace), Антибандеры (Anti-banders), Антифашист (Anti-fascist), Луганск первые новости (Lugansk First News), Эту страну не победить (This country can’t be defeated), Русские не сдаются (Russians don’t give up), and so on.
Social media is one of the instruments for manipulation of people’s decisions in political contests. VKontakte is becoming more and more pro-Russian. Specially created social bots communicate autonomously on social media to influence the political opinions of the readers.
Neural networks are one of the most effective instruments for the automatic detection of such accounts and for drawing conclusions about their authenticity.