A Pro-BJP and Pro-Modi network of Twitter accounts are attempting to influence the narrative of politically sensitive topics in India with newly created accounts posting the same text and amplifying specific hashtags.
The content amplified is supportive of the Bharatiya Janata Party (BJP), led by India Prime Minister Narendra Modi. The influence operation attempts to artificially distort the narrative on issues such as the farmers’ debate and discredit opposing parties.
This is not the first time Pro-BJP content has been identified as attempting to influence social media narratives on political issues. In the past the BJP, through its IT Cell, has been alleged to have exploited online communication methods to influence voters, as well as coordinate influence and automate Twitter activity for hashtag manipulation.
Overview of the Pro-BJP Network
The two main issues that appear to have surged in online activity over December and January targeted by the campaign are:
- The All India Trinamool Congress (TMC) – with the #TMCHataoBanglaBachao tag, and
- The subject of Indian farmers and the BJP – with the #KrishokSurokhaAbhijan tag
I captured activity on Twitter using those two hashtags over an intermittent period between January 2 – January 10.
I visualised the network through Gephi, as seen below. The visualisation is made up of accounts (nodes) and their retweets and mentions (edges – seen as lines). This visualisation was the basis for further analysis of the influence operation.
Accounts captured in the network published content under the #TMCHataoBanglaBachao and #KrishokSurokhaAbhijan hashtags. Much of the content used the exact same text, published through scores of accounts, many of which were newly created. This same text publishing is known as the ‘copypasta’ technique.
I first noticed the copypasta campaign on the #TMCHataoBanglaBachao tag on January 2. Subsequently many of the accounts I was monitoring moved on to amplify content under the #KrishokSurokhaAbhijan tag, indicating that the coordination of accounts has either the same, or mutual goal.
There are varying types of text under these hashtags all using the copypasta technique. This can be seen in the screenshots below.
This report is broken down into two sections to address each hashtag individually, and the accounts amplifying the copypasta text under those tags, before looking at a crossover between the two networks at the end of this report.
#TMCHataoBanglaBachao Network
The activity on Twitter under the #TMCHataoBanglaBachao tag focussed on the amplification of the following message:
“Bengal has been suffering from last 10 years. The TMC regime has built its foundation on ‘Cut Money’”.
Evidence can be seen below of a sample of the scale of the amplification effort.
Some of the posts also included infographics. An example can be seen below.
It should be noted that there were other pro-BJP copypasta campaigns using the same hashtag running different text packets. Some of those can be seen in the screenshots further above. This initial investigation commenced with the “Bengal has been suffering from last 10 years” text, so I have chosen to run an assessment on the network of that text use.
Assessing Accounts in the #TMCHataoBanglaBachao Network
Using Hoaxy, a visualisation can be created of the repeated text seen in the above copypasta campaign.
This visualisation allows for the display of links between the accounts in a network visualisation to identify trends and commonality between accounts.
Hoaxy uses a ‘bot score’ system to assess the likely level of automation of an account where 5 may indicate a large amount of automation, and 0 for little to none. More information on the details of the automation detection tool can be found here.
While the score is a good indicator, other factors should be taken into account, such as a deeper and more granular inspection of the accounts, in conjunction with identification of activity such as the copypasta evidence seen in the previous screenshots.
It may be the case that a network of accounts can be operated by human users in a ‘troll farm‘, and that the use of those accounts may exhibit automated tendencies. However these should not be classified as ‘bots’, but rather human-use accounts operating in a coordinated manner to target specific talking points and agendas.
In the below image, as indicated on the right, I have chosen to display all of the accounts which are identified on the range of 3-4 out of the 0-5 scaling used by the automation ranking system.
Selecting the ‘highly likely’ accounts, indicated in red, shows the spread of those accounts, similar to the orange ones, across the network visualisation.
In the identification of the red accounts for further analysis, many of them had moved on from the TMCHataoBanglaBachao amplification to amplify the KrishokSurokhaAbhijan content.
Analysis of Individual Accounts of the #TMCHataoBanglaBachao Campaign
Looking at the user-level data shows how some of these accounts operate on a granular level.
To identify this detailed information, I have used accountanalysis to provide that breakdown and chose three accounts from the above visualisation that were indicated in the red category of 4-5 as ‘highly likely’ to be automated.
I have blacked out the identification data of the specific users, for the purpose of protecting potential hijacked accounts or stolen profile images.
What we can see in the above data, is first, the high posting tendency at specific times. For example, in one of the accounts in a single hour on a Saturday there were 218 tweets. In the context of ‘tweets’, this also includes retweets, and in line with the high retweet to tweet ratio of the accounts it is likely that the majority of those 218 tweets were retweets.
In assessing that data, it may be the case that these accounts are not automated, but rather are shelf accounts for human use, and are specifically made to amplify certain content. After enough amplification has been made by the account, the human user may log out and access another account in the next hour to repeat the process.
As mentioned further above, in looking at the granular detail of the accounts we can also get an indication as to the other tweet/retweet content. In the three account samples (as well as many of the others I looked at) there was cross-content with the other analysed hashtag #KrishokSurokhaAbhijan, as well as the following other tags:
- #ModiWithFarmers
- #MaynaguriBJP
- #BengalWithBJP
- #AmitShahInBengal
#KrishokSurokhaAbhijan Network
The copypasta activity on Twitter under the #KrishokSurokhaAbhijan tag focussed on the amplification of the following message:
“নতুন কৃষি আইনের দ্বারা কৃষিজ পণ্য বিক্রি করা হবে আরও সহজ। বিজেপি সরকার আনছে কৃষক সুরক্ষা অভিযান।“
Using Google Translate, in English this says:
“The new agricultural law will make it easier to sell agricultural products. The BJP government is bringing farmers protection campaign.”
A very helpful source, that wishes to remain anonymous, has confirmed that translation as correct.
The text is used complementary with the tag #KrishokSurokhaAbhijan and evidence of the copypasta text packet and the tag can be seen below.
Similar to the text under the #TMCHataoBanglaBachao tag, some of the posts also included infographic images with text, as seen below.
Assessing Accounts in the #KrishokSurokhaAbhijan Network
Again using Hoaxy, I have created a visualisation of the repeated text seen in the above network, and then display the links between those accounts in a network visualisation fashion to identify links and trends.
The network of amplified content appears to be larger than the effort focussed on the #TMCHataoBanglaBachao tag, this may suggest specific interests and priorities of the influence operation.
As performed above in the previous section, I have illustrated the accounts that are labelled as ‘likely’ automated as ranked from 3-4 indicated as orange.
And below we can see the accounts that are ‘highly likely’ automated (ranked between 4-5) and indicated as red.
In looking at the creation dates of many of these accounts, as well as those in the #TMCHataoBanglaBachao copypasta campaign, there appears to be a batch of accounts made in December 2020 and January 2021 (note that this report was written on January 10). While other accounts have intermittent creation dates, the indication of these larger than normal creation dates is more indicative of the inauthentic use of these accounts. I have covered more on the creation dates further below.
Analysis of Individual Accounts of the #KrishokSurokhaAbhijan Campaign
When viewing the granular data of the accounts posting the copypasta content with the #KrishokSurokhaAbhijan tag, similar posting patterns can be identified as were seen in the granular assessment above.
I have selected a test sample of three of the accounts to illustrate a sample of how they operate. These are accounts from the visualisation above that were flagged red from 4-5 as ‘highly likely’ automated.
There are similar patterns as to the accounts seen that were using the copypasta tactic on the #TMCHataoBanglaBachao tag. Specifically the mass tweeting in a short amount of time.
One of the accounts, for example, on a Saturday evening between 6pm and 7pm made 303 tweets (including retweets). That same account was created in December 19, 2020 and in the dataset of 1000 tweets of that account analysed it had only ever retweeted – and never posted original content.
While both the #KrishokSurokhaAbhijan and #TMCHataoBanglaBachao tags were present, other tags featured in the sample accounts’ data, such as:
- #BengalwithBJP
- #AmitShahInBengal
- #BengalWelcomesNadda
- #UrjaAatmanirbharta
- #BJPBISHNUPUR
- #AmphanTakaChorTMC
- #SwagatamAmitShah
- #EbarBanglayBJP
Creation Dates of Accounts in Both #TMCHataoBanglaBachao and #KrishokSurokhaAbhijan Networks
In the analysis of the individual accounts in the above networks there appears to be an indication for the red-flagged accounts that were responsible for the pro-BJP copypasta information campaign on Twitter, to have been created in December 2020 and January 2021.
One approach to analysing the size of that network in the wider range of Twitter activity on those two hashtags is to analyse the entire network of activity of the two hashtags, and then visualise the accounts that were created in December 2020 and January 2021.
Using Twitter’s API, I captured activity over an intermittent period spanning eight days for activity under the #TMCHataoBanglaBachao and #KrishokSurokhaAbhijan tags. I captured 1669 accounts to analyse.
The visualisation below, made in Gephi, illustrates the entire network captured over that period.
These are the accounts, created in late December 2020, and early January 2021, illustrated in the network with a larger node size (at 40pts) and tagged in red.
To do this, I went through the spreadsheet individually flagging the accounts that were created between December 1, 2020 – January 10, 2021. There were a total of 408 accounts created between that window.
In comparison to the other creation dates, this is indicative of accounts being made for a purpose, such as use in a coordinated network. The ‘batching’ of these creation dates is visible in an analysis of the CSV data.
As seen below, I have highlighted the December/January accounts.
In looking at the number of accounts assessed, it is important to keep in mind I do not have access to all of the data available on the platform and the data I do have access to appears to be incomplete to make a whole figure assessment on just how many accounts exist in this influence operation, as well as to make a precise and evidential attribution of the network.
However, out of the 1669 accounts captured that tweeted, or retweeted, content using the hashtags #TMCHataoBanglaBachao and #KrishokSurokhaAbhijan, 24.4% of them were created either in December 2020, or the 10 days of January 2021.
UPDATE: Jan 12, 2021: Identification of Listed Tweet Content on Facebook Pages
There are a number of Facebook accounts, some of which identify as ‘BJP employees’, sharing a list of Tweet texts in Hindi, English and Bengali.
I have anonymised the specific user data so as to not reveal the identity of the person for either repercussion against their actions, or in case the account has been hijacked or using a stolen image.
Below are two posts where text packets featured in both networks analysed in this campaign are present.
Running a search for these texts on Twitter shows that each of these text packets have been echoed by large networks in the same ‘copypasta’ method as assessed in this report.
For example, the text below in Hindi.
And the network’s replication on Twitter.
The same can be seen for the Bengali text which appeared in the Facebook posts (seen below)
And again, echoed on Twitter by the network.
UPDATE: Jan 12, 2021: Google Document of Pro-Modi Copypasta Text Identifies Author Details
A Google Document with suggested tweets for the same copypasta network has emerged and is an open document for the public to view. The hashtag being promoted in that document is the #YuvaShaktiWithModi tag, with pro-Modi messaging.
The content in the document is a list of tweets in English, Hindi and Bengali, and were spammed on Twitter by the network that also featured above in this analysis. A screenshot of the document can be seen below.
The highlighted text can be seen on Twitter replicated by the copypasta network.
Again, the same content for tweets in non-English.
And the Hindi tweets replicating that message.
By looking at the specific data of the Google Document, an author can be identified. I have kept the privacy of the author as they may be required to conduct this role for their job.
The email address is listed as a contact address on the bio of a LinkedIn account. The owner of that LinkedIn account is a social media team leader at MyGov India.
Summary of the Pro-BJP #TMCHataoBanglaBachao and #KrishokSurokhaAbhijan Networks
There is clear evidence both through network visuals, screenshots of activity from the platform, and metrics, that there is coordinated activity on Twitter’s platform of pro-BJP accounts amplifying content in an attempt to influence the narrative on politically sensitive issues in India.
The tools and methods used in this assessment are freely available and the findings can be replicated by anyone pending Twitter API access.
Tools Used in Assessment
- Gephi – for API capturing and visualisation
- Hoaxy – for search string visualisation
- Botometer – automation assessment
- Accountanalysis – account assessment
- Social Bearing – hashtag assessment
- Tweetdeck – for account monitoring
- Twitter API – for data collection
Great Article
Well studied & compiled nicely! Thnx for your good effort.
Thank you very much for such a tricky analysis.