Bots are a phenomenon which is impossible to avoid and dangerous to ignore. They can raise a hullabaloo about events of minor importance, direct or stifle a discussion, impact the opinion of other users. To avoid confusion, it is impossible to find out if there are any ways to evade the influence of bots and the agenda they impose.

Together with the company TheRespo, we have analyzed one of the country’s most popular and significant Facebook accounts – the page of Ukraine President Petro Poroshenko. How actively do the bots interfere in discussions of matters of national, public, or personal importance, pretending to present a first-person opinion?

“I congratulate you on the Day of Independence of Ukraine! I wish you peace and unity. Glory to Ukraine!” – these fairly ordinary lines appeared in the official account of the then recently elected Ukrainian President Petro Poroshenko on August 24, 2014. However, the response to such a regular congratulation was rather unusual: more than 500* comments, nearly 1,400 reposts, and 10,000 likes. Remarkably, this post broke the absolute record for the number of comments left by bots: 111, or 21%. Even more unexpected was the fact that the number of negative comments below that post was almost twice as high as the number of positive ones.

During Poroshenko’s first three years as President (June 2014 – May 2017), more than 79,000** unique Facebook users wrote at least one comment on his official page. The total number of entries they made was about 176,000. TheRespo analyzed every single user and comment: the study revealed that 98% of the commenters (77,500) were real people. On the other hand, the remaining 2% (1,500) were much more productive as they left about 26,000 comments or 15% of all the entries. It means that the bots made one out of every six comments.

How to catch a bot

A bot is a human or an algorithm which is deliberately leaving comments intended to influence their readers in a particular way. Two major types of bots can be defined:

1. first, an FB account is created, and then an algorithm designed to leave a preset comment under preset posts is connected to it (sometimes more complicated logic can be used);
2. similarly, a FB account is created; then comments are made on its behalf by a real person acting in essence as an algorithm does.

In spite of a rather simple algorithm of actions, bots often remain unidentified by ordinary users. Indeed, to “catch” a bot, one has to conduct a detailed analysis of his posts as well as of the frequency of his commenting. The starting point for our study was the number of comments left by Facebook users on Poroshenko’s page during the three-year period under review. Fig. 1 shows that most people left from one to five comments.

Figure 1. Commenters’ activity

Users who were leaving more than 30 entries were assigned to “the risk group” (other criteria were used as well). Following an analysis of these profiles, a number of specific features were defined:

• “suspected bots” usually have an account picture showing a rather unclear human image far in the background or even an abstract image. The photos they post do not show the same person (the account’s presumed owner) more than once; instead, photos of large groups of people are presented;
• “suspected bots” usually do not publish any posts on their behalf; instead, they regularly repost entries from 1-2 groups;
• their photos and posts either receive no likes/shares at all or get only a few responses from other users;
• the number of friends is not a revealing characteristic; bots are somewhat more frequently encountered among accounts befriended by less than 20 or more than 1,000 users.

Moreover, bots have an informal classification of their own. Tentatively, they can be divided into three categories:

• an economy option, without any friends or with just a few of them: the cheapest and most identifiable alternative;
• spambots, which distribute content via friends; usually, they have more than 1,000 friends;
• well-masked bots – the number of friends shows a distribution pattern similar to that of ordinary humans (most have 100-1,000 friends; less than a quarter have up to 100 friends; and less than 10% have more than 1,000 friends). These bots are the most expensive and most frequently used ones in politics.

Bot identification methodology: to determine whether an account is a bot, several indicators were used:

• primary: a publication of the same (as a rule, negative) comment several times, publication of a significant number of links to side resources;
• secondary: more than 30 comments below posts on Poroshenko’s page, absence of personal photos on the own page, absence or very few personal posts (most often, no individual posts at all; sometimes 1-3 posts), minimal subscriber response to reposts, regular distribution of content from a narrow range of sources (usually, 1-2 thematic groups), unrealistic personal data (workplace, position).

To identify bots, we constructed a model which analyzed each account for the presence of the above-mentioned individual factors and their combinations. In most cases, secondary indicators were used in the analysis, since not all of the bots have primary indicators. As a result, we have found a total of 1,500 bots, with an estimation error of 5%.

Poroshenko bots: good, bad, evil

The 1,500-strong bot army “fighting” below the President’s posts is devoid of unity as it had no consolidated position. Within the framework of this study, the bot’s “tonality” was identified and analyzed; as a result, we divided them into three groups: negative, positive, and neutral.

• Negative: bots for which the aggregate emotion of all "anger" (or "sorrow") comments is twice as intense as the total emotion of all "joy" comments. This group accounts for 29% of all the bots;
• Positive: bots for which the aggregate emotion of all "joy" comments is twice as intense as the aggregate emotion of all "anger" or "sorrow" comments. This group accounts for 27% of all the bots;
• Neutral: bots using negative and positive emotions in their comments on Poroshenko’s page with more or less the same frequency. This category accounted for 44% of all the bots.

The standoff between bots writing negative comments and those leaving positive comments began in the very first days of Poroshenko’s presidency. August and September of 2014 appeared to be the leading months in regards to the number of comments from fake accounts is topped by August and September of 2014: 3,203 and 2,712, respectively. The share of fake comments in the total number of comments was highest in November 2014, when bots wrote 26% of all the comments.

In absolute terms, bots’ activity was lowest in April and July 2016: 235 and 238 comments, respectively. In relative terms, their activity was lowest in August and October of that year, when only 11% of all the comments were written by bots. However, bots’ activity has been on the rise since the end of last year.

“Places of residence” of positive bots

After the beginning of Poroshenko’s presidency, 388 “positively minded” bots left their comments on his page, which amounts to 27% of the total number of bots. Judging by the number of comments, positively minded bots were most attracted by Poroshenko’s posts that stirred up a heated discussion involving many participants. For example, the birthday congratulation addressed to Poroshenko’s wife Maryna evoked 680 comments, where bots left one out of ten. Remarkably, nearly 60% of them were of a positive nature, and a mere 9% were negative. Bots also wrote many positive comments on a post in which Poroshenko expressed condolences and sorrow to Tetiana Chornovol because of her husband’s death in battle: one of every five comments came from bots, and 37% of them were positive. The third place is obtained by Poroshenko’s post on his conversation with Merkel about the RF’s aggression in the Donbas (29% of positive comments). Benevolent bots often comment on albums as well as on posts about diplomatic cooperation or about granting decorations to heroes. They most often use in their comments such expressions as “Hurray!”, “Thank God!”, “Cool!”, and “Welcome,” supporting these words with numerous smiles.

The evil world

Negative bots on Poroshenko’s Facebook page were a little more numerous than positive ones: 417, or 29%. Their vocabulary also includes some clear favorites, such as “Shame!” (“Hanba!”), “Disgraceful!”, “How sad all of this is,” and “Very sad.” Negative bots also often refer to mass media’s critical materials about Poroshenko, photo memes, and hashtags.

Negative bots were most active in commenting on the post about prospects for freeing Roman Sushchenko, the news about a tranche from the US, and the MH17 album.

Whose bot is this one?

It is quite easy to identify the general algorithm of bots’ behavior; it is much harder, however, to expose the “owner” (or “operator”) of a particular bot. We will try to resolve this problem in the future; however, the current study also allowed us to establish the new facts and to bust the old myths.

• The number of bots leaving comments on the President’s Facebook page is relatively small: 1,500 accounts, or about 2% of the total number of account owners that left their comments on Poroshenko’s official page during the three years under review;
• Still, the commenting productivity of these two percent is several times higher than that of “normal” users. Bots account for 15% of all the comments below the President’s posts;
• Poroshenko’s Facebook page is not an exclusive reserve for positively minded bots eulogizing the President. The share of bots with a clear prevalence of positive sentiments is only 27%;
• The share of negative bots is higher – 29%. The remaining 44% display no prevailing mood.
• Not all of the negative bots working on the President’s Facebook page are specifically targeted at Poroshenko. Some of these bots “dislike” Ukraine in general and so they leave anti-Ukrainian comments. There is also a group of spamming bots;
• Not all of the positive bots belong to or are influenced by Poroshenko’s communication team. This conclusion is supported by the fact that many bots call themselves “porokhobots” (see the picture) or the like, which looks inconsistent with the viewpoint of Poroshenko’s communication team. (87 bots are similarly named and/or display identical behavior, which may be indicative of the same owner).

• ● No significant patterns of dependence of “good” and “evil” bots’ involvement on specific topics raised by the President on his Facebook page were revealed.

* Here and in subsequent analysis, we used publicly available comments left on Poroshenko’s page. We disregarded deleted (hidden) comments as well as comments containing abusive language.

** The analyzed period is between June 7, 2014, and May 31, 2017.