It's not easy to figure out how many fake accounts there are on Twitter

It's not easy to figure out how many fake accounts there are on Twitter

The actual number of fake Twitter accounts ended up at the center of Elon Musk's complex acquisition of the social network, which in April offered $ 44 billion to gain control of it. Musk argues that fake and often automatic profiles (“bots”) are many more than those indicated by Twitter, and that consequently the value of the social network is lower than that declared, because it would have fewer real users; Twitter argues instead that the counts are accurate enough.

In the middle there are experts, analysts and investors, who begin to suspect that Musk has raised the issue of bots to try to obtain more advantageous purchase conditions or to withdraw from the acquisition, despite the commitments now formally made also with the supervisory authorities. of markets in the United States.

On Friday May 13, Musk had announced on Twitter that he had “temporarily suspended” his initiative, saying he was waiting for more solid data on the fact that fake and spam accounts are actually less than 5 percent, as claimed by the social network. He later said he was “still busy” in the acquisition, but without giving further details. The announcement had led Twitter's shares to lose several points on the stock exchange, raising new suspicions about the intentions of Musk, known for setting himself sometimes unattainable goals and for suddenly changing his mind about his projects.

Keeping the proliferation of fake accounts under control is one of the most difficult and costly activities for social networks. Facebook, the largest of all, eliminates several billion fake profiles every year, used for online scams, to spread fake news or to make the profiles and pages of other users (often public figures) or companies appear more followed and famous.

Many fake profiles work through bots – automatic systems or systems that can be controlled in bulk by a single person – to quickly put thousands of “likes” or comments, in order to bring out posts and conversations hoping that they will then be highlighted by the algorithms that organize the contents on social networks.

The bot problem has plagued Twitter for many years and the company has over time received various criticisms and accusations of not wanting to seriously address the problem, because this would involve deleting millions of accounts reducing the declared amount of subscribers to the social network. Twitter derives a large part of its revenues from advertising and sponsored tweets, the price of which for the promoter derives from what they are seen. Fake accounts can end up in the view count, with the risk of returning incorrect statistics and inflated prices for advertisements.

In the tax documents that it must periodically submit to regulatory authorities in the United States, Twitter publishes data on the amount of its users and also provides information on the estimated amount of bogus accounts and bots present on its platform. The company's most recent estimates say the amount of bots is around 5 percent, but the calculation does not convince all analysts and experts, for whom the problem is underestimated.

Musk also thinks so and for this reason last week he announced the suspension of the acquisition. However, many observers have reported that the Twitter bot problem had been known for years, including doubts about the 5 percent estimates, defining at least strange that Musk only noticed it after having submitted a purchase proposal worth tens of billions. dollars. Some have also hypothesized that the circumstance could become a pretext for abandoning the purchase operation.

Musk could try to show that the percentage of bots is much higher and that consequently the real Twitter users are less than stated, because this would have an impact on the actual value of the company. In this way, more favorable conditions could arise for the acquisition or withdraw its proposal, but in both cases Musk would face legal consequences considering that the purchase price had already been formalized to the control authorities and penalties could be applied.

A few days after Musk's tweets about bots, Twitter CEO Parag Agrawal wrote a long thread explaining how difficult it is to keep tabs on fake accounts on a platform where hundreds of thousands of accounts are activated and deleted every day. Agrawal explained that: “Spam is not 'binary' (human / non-human). The most advanced spam campaigns use a combination of coordinated humans and automated systems. They also hack real accounts and then employ them to run their campaigns. So: they are sophisticated and difficult to catch “.

According to data provided by Agrawal, Twitter suspends about half a million spam accounts every day and blocks millions of spam accounts every week if they fail certain tests of their authenticity, such as associating a phone number with account and verify it with a code: «The real difficulty is that many accounts that appear to be fake are actually authentic. And some of the spam accounts that are more dangerous – because they cause more harm to users – may appear completely authentic at first glance. “

Agrawal had then confirmed the estimate of less than 5 percent of bots, compared to the rest of the users active daily on Twitter and counted for the ad views: “Based on the margins of error on our estimates we are confident of our statements that we provide to every quarterly cash report “.

Concluding his explanation, Agrawal said he did not believe such an accurate evaluation of the bots by an external party is feasible «Given the need to use public and private information (which we cannot share). From the outside it is not even possible to know which accounts are counted for advertising views each day ». The reference was to the analysis that Musk had announced on May 14 that he wanted to do on a sample of 100 Twitter accounts. Musk had responded to Agrawal's explanation with a rather laconic and at the same time eloquent tweet:

💩

– Elon Musk (@elonmusk) May 16, 2022

In recent days, the 100 account test proposed by Musk has been carried out by various users and companies, including some who have already tried to independently estimate the percentage of bots on Twitter in the past. Among these is IV.ai, which has analyzed the characteristics of a hundred profiles following the account of Tesla, the electric car company of Musk. According to the results, 20 percent of accounts had a high probability of being bots; a subsequent manual analysis estimated an even higher percentage, around 50 per cent.

However, the calculation is complicated by the short duration of the fake accounts, probably resulting from the constant activity of Twitter to remove or suspend the suspicious ones. Most of those detected by IV.ai, for example, had disappeared shortly after the analysis. This would show that bot recognition systems work quite well and quickly remove a significant portion of the bogus accounts. However, the test was on a very limited group of accounts, considering Tesla's profile is followed by 15.5 million accounts.

There are algorithms that perform more refined analyzes and on a greater number of accounts, but their results are not always satisfactory. Botometer is considered one of the most reliable, but in a recent analysis it reported that Elon Musk's account itself is likely to be a bot, demonstrating how difficult it can be to distinguish fake accounts from authentic ones.

Is @elonmusk a bot? One of the most advanced algorithms for bot detection thinks so, which illustrates just how difficult it might be to clean up this bird place.

¯_ (ツ) _ / ¯

try the tool: https://t.co/sedxkIEQRC and technical references here: https://t.co/rwh7Yao1yV pic.twitter.com/mtfua9RA1L

– Chris Bail (@chris_bail) May 17, 2022

Botometer's algorithm employs huge amounts of data to make its algorithms automatically learn the best way to perform analyzes, improving their reliability over time. The system could be much more accurate if it had access not only to data that everyone can see on Twitter, but also to what only the social network can see. Twitter managers argue that this is not possible for privacy reasons, but according to various analysts the problem could be overcome by anonymizing sensitive data, in order to offer more relevant data sets for those who search through the social network.

It is not clear what the next evolutions of the comparison between Musk and Twitter on bots and the methodologies to calculate them could be. In recent years, Twitter has shown that it has some trouble making its own estimates, but is unlikely to change estimates that have been presented on various occasions to investors and regulators.

Meanwhile, Musk's statements and the suspension of the acquisition are having serious consequences for Twitter. In two weeks, the company has lost about a quarter of its stock market value and various executives have chosen to resign, leaving a situation that some sources within the company define as “chaotic”. The board of directors, which initially announced that it would block the acquisition and then reconsider after a few days, said it was unwilling to change the terms of the agreement.