Hyderabad: That Twitter has a huge bot/fake account problem has been known for years together now. It has been documented and reported multiple times. Yet, the numbers still manage to astound.
According to research by an activist of tweets made over the last two weeks on Twitter — amid a politically charged climate owing to opposition to CAA and NRC — more than 85 per cent replies made to tweets of the most popular accounts were found to be by bots. And these bots tweeted on subjects across the political spectrum.
Twitter bots, short for ‘robot’, are software tools that, once programmed, are capable of tweeting, re-tweeting, liking, following, un-following or even direct messaging other accounts. They are often used to inflate the number of followers, re-tweets and so on.
Kumar (name changed) performed his study using a painstakingly designed procedure. He got the idea for this study after noticing replies to popular tweets.
For starters, he divided them into three categories: tweets by the user’s followers and verified accounts, tweets by popular troll and anonymous accounts and tweets by unknown accounts. “I ignored the first two categories since the probability of them being bots is low. The third one interested me and drew my attention,” he said.
To identify accounts of this kind, Kumar parsed through Twitter Trends’ political section. Over 15 days, his algorithm identified 1,000 of the most popular accounts.
The algorithm took these accounts and parsed through their followers and tweets. It identified accounts with common features: unverified, no profile picture or profile picture of random objects like flags, solid colours or repetitive photos (leaders), and were active for more than three months. This way the algorithm resulted in a list of 5.7 lakh accounts.
And that is precisely where the actual work began. “I worked on identifying the following features in these 5.7 lakh accounts. Some of them were the length of the user name, default picture, account age, number of unique profile descriptions, number of friends, number of followers, number of tweets (per hour and total), number of replies and so on,” explained Kumar.
The study resulted in astonishing findings. As many as 2.3 lakh of the 5.7 lakh accounts were found to be bots. “I found that 80-85 per cent of all replies made to tweets by popular accounts were made by bots,” he said.
Interestingly, a whopping 89 per cent of all these bots had started in June-July 2013. Kumar refused to speculate on an explanation for this finding.
“I tried to keep my study completely unbiased. Whatever bias I might have, I made sure it did not reflect in my algorithms,” he said.
However, comments on Kumar’s post about his findings on Reddit attribute this spike of bot accounts to the beginning of the 2014 general election season.
This was when, one commenter speculated, all major parties started their groundwork for the elections. These bot accounts exhibited quite similar characteristics when it came to time between tweets. Nearly 74 per cent have tweeted roughly around 5,000 times, with a time-interval of one tweet per 20 minutes. This indicates a predictable pattern as compared to humans, whose tweets are unpredictable.
Also, 18 per cent of the accounts tweeted more than 60,000 times with a time-interval of one tweet per eight minutes. Only one per cent of the tweets were original tweets; 18 per cent were replies and a vast majority, 81 per cent, were re-tweets. “As part of my work, I wrote algorithms to report these bots. For that I had to create my own fake accounts, report these bots, and delete my accounts. Once these bots get reported enough number of times, Twitter deletes them. I must clarify here that none of the accounts did anything else such as participate in Twitter polls,” said Kumar.
Confirming that these bots were bots is that less than one per cent challenged the deletion. “If they were human-run accounts, the deletion would have been challenged much more,” he said.
Kumar said that Twitter needed to do better by controlling the number of fake accounts on its platform. “Twitter’s claim that only 2-3 per cent of accounts are fakes is wrong by a huge margin. Until Twitter does something, the developer community needs to come together and go about the cleansing operation.
