Bot makes random comments to find popular and opposing opinions on Facebook, which is extremely popular as an online social media in the country! Bots are so stupid that they even post irrelevant comments under tech reports.
Before and after the 12th National Parliament elections, DismissLab has discovered a bot network that visited 1,369 Facebook profiles of various media and opposition parties on behalf of Awami League. A data-checking team at Digital Right, a research firm on the impact of technology, identified 21,221 political comments from 42 target pages from the accounts.
The FactCheck research report, titled ‘ Stupid Bots: Facebook’s Organized Network of Political Disinformation ‘, was published on its website on Thursday. The report was prepared by researchers Minhaj Aman and Abrar Ifaz. And the original engineering was discovered by engagement editor Partha Pratim Das.
Research has shown that more than 21 thousand comments have been made in 197 posts in favor of Awami League from Facebook profiles. They made the same comments in different posts. Same comment posted from different profiles.
The number of bot accounts and comments made by them may be higher. Because, this research has been done on the basis of information found in total 197 posts.
The head of the research team, Minhaj Aman, said that on June 21 of this year, on the website and Facebook page of BDNews24, “Is there an invisible code on every page printed on a color printer?” A news headline appeared. Usually, such news escapes the attention of devoted readers of politics, but a researcher at DismissLab caught the eye of a researcher under the Facebook post, with a comment: “This election will be transparent. People will be able to vote freely in the upcoming elections. And this time, BNP is afraid to come to the election because it cannot steal votes and is making so many stories.”
The following comments under that post are eye-rolling. Almost six months after the election, users have been commenting one after another on various political issues including criticism of the then opposition party BNP, hope for a successful and fair election of the 12th National Parliament, demands to hold the election under the ruling government. But why would someone make such a comment so long after the election, that too in a completely non-political news?
Then begins the identification and research of commenters. Every comment from that Facebook post of BDNews24 was collected for this study. A Google search on that list revealed 196 more posts with the same comment. A database of about 35,000 comments was created by scraping every comment from those posts. Bot accounts and more than 21,000 political comments made by them were isolated from that database. Only comments that have been posted more than 10 times are kept in the final database. The number of bot accounts and comments they make may be higher. Because, this research has been done on the basis of information found in total 197 posts.
According to research by DismissLab, the bot network became active just before the 12th parliamentary elections. As of June 2024, they have made 474 political comments in turn. The comments were made before the election, but they posted the same comments after the election. They targeted the country’s established media and Facebook pages associated with the then opposition political party Bangladesh Nationalist Party.
The bot network usually comments on political posts, and when they find certain political keywords or words in a Facebook post, they jump to that post. But sometimes they are fooled. As is the case with the keyword “EC” (Election Commission). In the case of BDNews24, the summary of the news about the color printer mentioned the “Machine Identification Code (MIC)”, there they saw hundreds of comments after seeing the keyword “EC”. Several other examples of bot stupidity have come up in research.
| ১৩৬৯ | ২১২২১ | ৪২ |
| Number of bot accounts | Political comments | Target page |
The final database also provides an idea of the list of possible political keywords and comments used by the bot network. Analyzing each account associated with the network separately found most of them: profiles locked or private, they were activated before the election, profile pictures were missing or stolen, they had few or no friends, and most accounts followed two specific pages. And these characteristics match the conventional definition of a political bot.
The activities of these profiles are also consistent with coordinated inauthentic behavior (CIB) or coordinated hostile campaigns. Meta defines such activities as, “We see CIB as a coordinated activity in which public debate is manipulated to achieve a strategic goal. At the center of these activities are fake accounts. In each case, the individuals involved in this work coordinate with each other and use fake accounts to mislead others as to who they really are and what they are doing.”
Apparently, the operation of this bot network is at least partially, if not completely, computational or automated. They automatically select posts based on specific keywords from one list and post comments selected from another list. In their process, there is no human monitoring of where and what comments are being posted.
University of Maryland Philip Merrill College of Journalism and Information Studies assistant professor Dr. Naemul Hasan said, “DismissalLab’s research results indicate that computational tools are used here.” However, it is also important to acknowledge the limitations of this analysis. Although the consistent repetition of similar comments suggests a computational mechanism, this finding did not provide direct evidence for it.
Regarding the limitations of the research, this research has been done only through the reverse search of the comments found below the Facebook post of BDNews24. As a result, only those Facebook posts that were associated with these comments were identified. There may also be more comments and related accounts – which could not be included in this study.
How does Bots look like?
DismissLab found a total of 1369 accounts in this study. Most of these profiles (77%) are in women’s names. There is also a strange similarity in that name. For example, 24% of female profiles end with the name “Akter” (Akter, Aktar). For example: Dia Aktar, Riya Aktar, Liza Aktar, Lima Aktar, Lisa Aktar, Misa Aktar etc.
A large proportion (16%) of male profiles used “Ahmed” as their last name. For example: Naeem Ahmed, Nadeem Ahmed, Kamil Ahmed, Mahin Ahmed, Sameer Ahmed, etc. The common feature found in both male and female accounts is that 90 percent of the names are two words. Somewhere, without much trouble, a name has been broken into two words. For example: Ri Pa, Mi Na, Li Ja, Yu Thi, Lam Ya, Mu Na, Ne Ha, Jos Na etc.
Fact-checking firm Snopes and cybersecurity firms CloudFlare and McAfee ( 1 , 2 , 3 ) suggest looking for common features to determine whether a Facebook profile is real or fake. The characteristics are: bot profiles are overly secretive, don’t provide enough information about themselves (in the About section), the number of posts on the profile is either too few or too high, the profile picture is collected from elsewhere on the Internet, and the number of friends is usually low.
Accordingly, DismissLab reviewed the privacy settings, number of friends, posting trends, contact information and profile pictures of 1,369 profiles. It showed that 247 profiles were locked. Of the remaining 1,122 accounts, 70 percent had profile pictures collected from elsewhere on the Internet. That is, the pictures belong to others. Somewhere the same picture is used in different profiles.
For example, a profile picture not found anywhere else on the internet was found to be used by Fahad Islam , Rajib Ahmed , Mamun Ahmed , Ahmed Tahsin , Riyad Khan. Tanjim Ahmed, Jahangir Alom and Ariyan Islam in a total of 8 accounts. All profiles uploaded the photo on the same day, 30 November 2023. Another 19 profile pictures were found to be not available on the internet, but being used in multiple profiles. A total of 111 accounts used these 20 profile pictures, which were not found anywhere else by searching the Internet.
Accounts opened in the name of women also used the same profile picture in different accounts. “Bots and fraudulent accounts use beauty as a shield to lure ‘friends’ and often” use pictures of attractive women or handsome men on their profiles or pages, McAfee says in describing techniques for identifying fake Facebook profiles. A similar trend was observed in this network. Several accounts used profile pictures of actresses and models such as Ashna Habib Bhavna from Bangladesh , Tama Mirza , Naila Naim , and Malavika Menon from South India.
Of the 1,124 profiles that were not locked, 85 percent had no contact information in the About section. 93 percent have no public posts other than profile picture, cover photo. Almost half (49 percent) of profiles do not display friend counts, and 45 percent of profiles have less than 200 friends. Only seven accounts appear to post regularly and they often share similar posts.
Analyzing the data using the time when a bot account first posted content as an indicator of activity, the profiles were active mainly between June and November of last year, before the election. But between September and November 2023, only half of the accounts became active. 344 accounts posted for the first time between November 23rd and 30th alone. Earlier, 240 profiles were activated between September 2 and 12.
Interconnected bot networks and coordinated behavior
The study found 21,221 political comments. But among these there were only 474 unique comments. That is, the bots returned to different posts and made these 474 comments. The interconnection between the bots is clearly visible (see the visualization below). For example, if you hover over a comment, it will show the total number of accounts that have made that comment. If you move the cursor or click on an account, it will show the total number of posts commented on by that account. Similarly, clicking on a post will show the total number of profiles on the network that have commented on it. Thus, the connection between posts, profiles and comments is clear.
For example, a profile named Riya Akhtar has made 138 comments on various posts in the last six months. On May 18, long after the elections, one of his comments was: “Bangladesh is an independent, sovereign country. An Independent Election Commission (EC) has been constituted here by law. I hope the upcoming National Assembly elections will be neutral.” A total of 96 posts with the same comment were made at different times from a total of 109 bot accounts including Dia Akhtar, Raisa, Rafia Akhtar, Nahid and Nipa.
Again, “those who have no confidence in the people are afraid. So, BNP is afraid to go to the polls” — this comment has received the most, 244 times. The comment was made on various posts from a total of 217 bot profiles. In a game report on Channel I’s Facebook page, a total of eight people including Aryan Munna, Nusrat Faria and Ratna Akhtar were seen making the same comment.
There are dozens of examples where multiple bot accounts have posted comments with the exact same misspelled words. In 2022, Meta also noted a similar linguistic error in a CIB network in China.
Another illustration of some kind of connection between profiles comes from the tendency to like two specific pages. 70 percent of the 1,124 accounts that do not have locked profiles are Bangla Khabar and Awami League Media Cell page is no longer available since the fall of the Awami League government on August 5); Follow either or both of these two pages.
Analyzing the commenting time on Facebook posts, 88 percent of the comments were made within two hours. That is, the time difference between the first bot-comment and the last bot-comment on each post is less than two hours. For example, on Prothom Alo’s Facebook page, 151 bot comments were made under a news link about BNP chairperson Khaleda Zia’s medical treatment from 11:45 am to 1:40 pm, in a span of 1 hour and 40 minutes.
Meta has earlier taken action against various pages and profiles in Bangladesh for coordinated malicious campaigns. They deleted 9 Facebook pages and 6 profiles before the National Assembly elections in 2018 . And in 2024 deleted 98 pages and 50 profiles. In both cases, pro-government commentary was shared from these pages and profiles, Meta noted.
Commentary and target of the bots
The bot network targeted 42 Facebook pages, mainly of various media outlets and political opponents of the Awami League. Out of this, 68 percent of the pages belong to the first-rate and well-known media and 31 percent of the pages belong to BNP or are related to BNP. Mass solidarity movement pages within political parties were also targeted.
According to the review, 86 percent of the bot comments criticize or attack BNP and party leaders. For example, “Fire-terrorist organization BNP. I demand their punishment” or “BNP is plotting to loot Bangladesh through money laundering” or “BNP is working as the main player in the international conspiracy against Bangladesh”. The remaining 14 percent of the comments were praising the government and the Election Commission and calling for peaceful or fair elections.
Out of the half of the posts which were commented on the pages of the media, 36 percent were related to BNP. The content of 17 percent was about Upazila election, national parliament election and election commission. BNP, Nationalist Party, Election Commission, EC, Khaleda Zia, Tariq Rahman, Awami League and Sheikh Hasina were the national keywords, the network commented in a coordinated way on those news or posts.
The stupidity of the bot
Bangladesh’s twelfth parliamentary election was held on January 7 this year and Awami League formed the government for the fourth time in a row. But it can be seen that the bot-comments posted in February, March, April, May or June are using the national term “upcoming 12th National Parliament Election” or “the time of 12th National Parliament Election is getting closer”. That is, the comments were made before the election but were arbitrarily posted after the election.
For example: “The forthcoming twelfth national election should be made free, fair and at the same time participatory and competitive so that other countries and people find this election credible. And the Election Commission is working towards that goal” — this comment was made in 49 posts, 123 times. All posts were published on Facebook after January 7. At least 15 such comments were found where the context of the comment was before the election, but posted after the election. That is, comments are posted without human oversight, there is no context check.
Another example of bot stupidity is political commentary in irrelevant news posts. At least 61 instances were found (about a third of the total posts) in which technology, health, sports, crime, international affairs, and even business news included irrelevant criticism of the opposition and praise of the government of the day. For example, looting money by breaking a bank safe in Bogura, the death of former Iranian president Ibrahim Raisi, a request to the ICC to change the venue of a cricket match in Pakistan or the news of an actress getting sick and being admitted to hospital.
It is difficult to find the rationale behind such seemingly political comments under these posts. But analyzing the captions or text accompanying these posts gives a general picture. Every single one of these posts has the word “EC” in the description in one form or another. Eg: ICU, ICC, REC, IFIC, OIC, ICT, Dialysis or Crisis. Every word has “EC”. The BDNews24 post about color printers that started this research also had the word “MIC” (Machine Identification Code).
EC stands for Election Commission. But the bot network consistently made political comments wherever they found the EC phrase on their target pages, whether or not it was relevant. This indicates that the network uses an automated method based on keywords to select which posts to comment on.
The Russian CIB network that Meta identified in 2022 contained bot-comment descriptions under many irrelevant posts. The social media firm also mentions a similar Israeli network in its first quarter 2024 Adversarial Threat Report .
Computational Propaganda and Democracy
While reviewing the results of the study individually, University of Maryland Assistant Professor Dr. Naemul Hasan said, “Using modern technology, these tasks are being done collaboratively with the help of human and computational tools.”
The whole point of such campaigns may not be the technology that is done. Samuel C. Uli and Philip N. As Howard, in their book Computational Propaganda: Political Parties, Politicians, and Political Manipulation on Social Media (2018), states, “For the most effective use of computational propaganda, both algorithmic distribution and human curation—are combined. Software bots and humans work together.” Their Taiwan study also shows that China’s propaganda on social media is not entirely automatic; Rather, they are rigorously coordinated by real people.
From the US presidential election in 2016 to Brazil, Russia, Ukraine, Poland, the Philippines, such computational propaganda and political bots have been used in various countries around the world. According to the Oxford Internet Institute’s 2020 Global Inventory of Organized Social Media Manipulation report, they documented the spread of computational propaganda and political disinformation on social media in 81 countries that year.
Research shows the proliferation of bot networks can have profound effects on democratic processes and accurate information systems. According to another study published by Woolley and Howard in the Journal of Democracy (2016), such networks can distort public discourse by promoting certain viewpoints while suppressing other viewpoints. This manipulation can create a false sense of consensus, polarize public opinion, and ultimately erode trust in democratic institutions.
Naemul Hasan said, “Future technologies will be more advanced using Artificial Intelligence (AI). For example, there will be no need to use the same profile picture in different fake accounts, AI will generate multiple profile pictures instantly making it difficult to detect coordinated disinformation attacks. So, it is necessary to create awareness among the public about these issues from now.”
Research Methodology
DismissLab’s research began with the irrelevant political comments found under a Facebook post by Bangladesh’s online news media, BDNews24, as mentioned earlier. In order to know that such comments have been made in any other post, 109 comments found under that post of bdnews24 have been searched separately on Google. From there, 196 other Facebook posts were found where similar comments were made.
Comments under 197 Facebook posts including the BDNews24 post have been scraped through an online tool called Apify. A total of 35930 comments are found there. Comments with more than 10 were selected to ensure that only bot comments were included in the analysis. Thus, the dataset of 21221 comments under this study was created.
This dataset yielded 474 unique comments, which were posted more than 10 times. These comments were made on 42 Facebook pages. These pages are divided into two main sections: press and opposition political parties.
The 1369 Facebook profiles from which these comments were made were looked at to analyze the characteristics of the number of friends, information in the About section, when the first post was made, whether the profile was locked, and whether the profile picture was available elsewhere on the Internet.
