Chef's Trap: filtering governmental spam on social networks.

Interview with the cyber-activists behind the browser extension that fights Russian troll army on online and offline propaganda landscapes. 


The name of 'Chef's Trap' comes from Evgeny Prigozhin, Putin's right hand, who directs the network of notorious digital propaganda companies. Prigozhin got the nickname of "Putin's chef" because his restaurants host dinners attended by Vladimir Putin. The pun of this username comes from the excessiveness of power that Prigozhin possesses in opposition to his declared status. While Chef's Trap is realised for Russian Vkontakte (VK), it is made in collaboration with a similar initiative operating on YouTube. Both projects - Observer (for YouTube) and Chef's Trap (for Vkontakte) simultaneously work as a database, accumulating accounts used by pro-governmental trolls, public page and a browser extension.


Chef's trap: public page; browser extension; setup instruction

Observer: public page; browser extension



        Anna Engelhardt: What are the current functions of your project, and how do they complement each other? Website of the Chef's Trap accumulates the database of governmental trolls, offering different means to interact with it. How would you like this database to be used?


        Chef's Trap: The primary function of the project is to create a database of accounts used by pro-governmental trolls. We are making it accessible through the website and actionable through the browser extension.

First of all, I would like this database to be used by the VK admins to ensure a quicker response and a ban of such accounts. At the moment, there has been no research that would engage the collected data (at least I was not informed). Still, third-party developers used our database to develop the script that highlights paid comments and make it into the browser extension. We accidentally learned about a similar project for YouTube, and, after contacting its creators, we got to know about a similar extension for video hosting. They already had a template for the script for Vkontakte and, after small cooperative changes and improvements, we made it available to everyone. This browser extension, colouring the trolls' comments, is an excellent solution for users to read the opinions of their fellow citizens, allowing them not to stumble upon a paid lie.

There is a function of an information card designed to notify other comment readers in a thread, who are not familiar with the project, about governmental accounts from which comments are written for money. We introduced it several months after the launch of the project, as a result of our attempts to briefly inform people in the comment thread about the activity of the specific account. The information card is designed simply: the algorithm puts the information, available in the database about the account in question, against a pre-prepared background, you can acquire it from a dedicated page on our website. The card contains the list of pages affected by this troll, the number of distributed fake comments, the registration date of the account and the number of accounts massively registered on that day if there is such information in the database. In general, such a method of propaganda [through pro-governmental trolls] doesn't seem successful, judging by the reaction of people who express themselves negatively against the activity of trolls after reading an information card. Still, one needs to spend the state budget!

One can also use the database to get acquainted with the current direction of propaganda, to see which agenda is promoted and what tricks are employed in attempts to influence public opinion. During its work, Chief's Trap has also become a source of news regarding the current positions promoted by trolls. We repeatedly noticed that the change in the direction of trolls' "activity" took place abruptly — let's take as an example the recent poisoning of Alexei Navalny. Up to then, the overwhelming majority of comments were about Belarus (simulating a nationwide declaration of love to its "acting" president). Still, on that very day [of Navalny's poisoning], they stopped writing about practically anything but Navalny. This abrupt change indicates the distribution of some mass instructions inside the troll factory which decide the "direction" of their activities for some period. It's only been a couple of hours since the news about the poisoning came, when Lakhta [NOTE] began to actively spread lies, trying to mislead the users with idiotic messages about the alleged poisoning with alcohol, antidepressants, etc.

The tricks used are relatively primitive and absurd. They involve: simulated "discussions" of news "in the right way" between two fake accounts as if they were two real people; labelling anyone who expresses disagreement or discontent as a Ukrainian or Navalnist (oppositional) bot (probably an attempt to play on the reactions of a rather close-minded public); a bizarre insistence that "s/he" is a real person with his/her own "opinion", and similar pathetic attempts in this vein.


        A.E.: As far as I understand from your public page, you created the Chef's Trap in 2019, so the project is more than a year old now. Were you inspired by any other initiatives when you started it? How did its aims and principles change over the year?


        Ch.Tr.: The project was conceived exclusively as a temporary means to lay bare the work of this utterly ridiculous troll factory. I didn't know any similar projects at that time. I expected that the trolls would be made inoperative after several months. Still, even despite mass bans and a predominantly negative attitude of society towards the trolls, they continue their attempts to this day. In the course of our work, we discovered many local bot farms that work only for groups and public pages of the particular region — Vologda, Primorsky, St. Petersburg and Yekaterinburg. They share the tasks, similar for everyone: delivering positive attitude in news discussions, simulating "genuine" responses to their own comments and artificially inflating likes. At the same time, the localisation of such bot farms takes place offline, since the comments are apparently written by people living in those regions. They have access to the niche information about the work of local governors, city halls and so on. Activity statistics also speak in favour of this theory, as accounts are active following the time zones in a particular area. Although it is possible that all of them work from one place, but then, as it seems to me, they would have problems with obtaining and operating information about the particular region and events there.


        A.E.: In studies of online propaganda, accounts are usually categorised into bots — fully automated accounts, trolls — people who spread pro-governmental messages for money or on their own initiative, and cyborgs — people who use scripts to spread pro-governmental messages. Do you follow a similar categorisation in the project? How do you view your interaction with these categories of accounts? Which account type do you prioritise to monitor?


        Ch.Tr.: It wouldn't be possible to automate comments on VK. Firstly, the social network itself blocks such activity and, secondly, there is no sense in using text templates, since every piece of news is unique, no matter what keywords for trolls it includes. Therefore, fully automated bots, just as "cyborgs", do not survive there for more than a few hours. As for unpaid pro-governmental users, I have seen only a couple of them for all the time (except subscribers and supporters of the "national liberation movement" and alike organisations, of course). Our database includes only accounts employed to write propaganda for money. That is, comments stored in our database with a particular marking are written by real people manually. These people are employees of the comment department who, during their shifts, have access to shared accounts, belonging to one of the media groups located in the Lakhta business centre (as can be judged from the open sources).


        A.E.: When processing such an amount of data, one begins to notice different connections and rhythms not initially expected. Which ideas came only after working with the data on bots and which preconceived notions did you confirm?


        Ch.Tr.: If we talk about some connections and rhythms invisible to an ordinary viewer or reader, the troll factory can also read that and, albeit very unlikely, make it more challenging to identify their accounts. In general, all their comments and activities are incredibly pathetic and clumsy, and, for me, personally, one glance at the text is enough to understand who is in front of me. Sometimes they attempt to deviate from their standard behaviour, trying to portray a real person to the best of ability, but it all looks ludicrously, of course.

Many ideas were confirmed or fundamentally disproved. For example, the project did not initially consider likes on trolls' comments. Still, after adding this function, the number of paid accounts identified through the likes equalled the number of trolls determined by the standard method of searching comments. We had an idea of working only with the text of comments (the standard method) and taking likes into account, but we didn't know how to count and spot fake likes, as this activity is drastically different. Tracking and processing likes now help detect about as many bots as processing comments.


        A.E.: Why did you choose Vkontakte as your platform? Do you plan to reach other social networks? Does the platform it monitors mean any fundamental difference to the operation of such a project?


        Ch.Tr.: The task and the main goal of the project is to reveal to the public the appalling things happening on social networks when the state is trying to propagandise on behalf of supposedly real people. Vkontakte was chosen because I believe it is the most popular information platform in the country, including all kinds of bots. We haven't yet planned to cover other social networks, since running the existing one on VKontakte is already quite complicated. Nevertheless, everyone can use our project, our webpage with statistics by day, to get acquainted with how bots react to certain events in the country and the world at specific points in time. It teaches people to define similar statements anywhere else easily. It seems to me that there can be no remarkable difference in the project's functioning depending on the social network it monitors since different departments of the "ministry of truth" share the same task.