Elon Musk has wrong approach to count fakes, spam on Twitter: experts
TeslaElon Musk, CEO sent TwitterShares tumbled on Friday after he stated that he would put his own money. $44 billion acquisition of the social network “on hold”While he studies the percentage of spam and fake accounts on the platform,
Musk said that he is still committed to the deal but continued to debate the problem of fake accounts. Musk tweeted that his team would conduct their own analysis, and raised doubts about the reliability of data reported by Twitter in its latest financial filings.
The following is a description of the product: first-quarter earnings report this yearTwitter admitted that there were “false and spam accounts” present on the platform. However, it also acknowledged legitimate daily active users (mDAU) with monetizable monetizable usage. According to Twitter, they reported that an “internal review” of some accounts showed less than 5% of their mDAU in the quarter.
TwitterIn addition, the company admitted that it had overestimated user numbers by between 1.4 million and 1.9 million over three years. Twitter revealed that the company had launched a feature in March 2019 which allowed users to connect multiple accounts to make it easier to switch between them. An error occurred at the time. All linked accounts were counted as MDAU because they had been accessed via the primary account.
Musk might be curious and understandable, but experts in statistical analysis, social media disinformation, and statistics say his suggestion for further analysis is seriously flawed.
This is what SpaceX has to say about it. TeslaThe CEO stated that he will do his best to find out how many duplicate, spam and fake accounts are on Twitter.
To find out the answer, I will be randomly sampling 100 @twitter followers. The process is open to others who want to replicate it and share their findings. In subsequent tweets, he clarified his method, saying: “Pick any account that has a lot followers.” Then, “Ignore the first 1000 followers and then choose every 10th.” “I’m open for better ideas.”
Musk claimed, but did not provide evidence, that he chose 100 for the study sample, as that is what Twitter uses to calculate their earnings numbers.
It is okay to use any sensible random sampling method. You can see if multiple people are getting similar results on fake/spam/duplicate account %. I picked 100 as the sample size number, because that is what Twitter uses to calculate <5% fake/spam/duplicate.”
Twitter did not respond to a question about whether his explanation of the company’s methodology was correct.
FacebookDustin Moskovitz (co-founder) addressed the subject via his Twitter account.
“Also, it feels like the “doesn’t believe the Twitter team to pull the sample” is its own red flag.
BotSentinelChristopher Bouzy, founder and CEO of CNBC, stated that his company’s analysis indicated that between 10% and 15% of Twitter accounts are “inauthentic.” This includes fakes and spammers as well as duplicates and nefarious bots. Also, “single purpose hate accounts,” which target individuals and spread disinformation to harass them.
BotSentinel receives funding primarily from crowdfunding. The company independently examines Twitter and determines if there is any inauthentic activity using machine learning software, as well as teams of human reviewers. Today, the company monitors over 2.5 million Twitter accounts. These are primarily English-language users.
Bouzy stated that Twitter does not have the ability to classify fake and spam accounts.
According to his warning, the amount of fake accounts on Twitter can vary depending on what topic is being covered. BotSentinel found that more fake accounts are tweeting about cryptocurrency, politics, climate change and other controversial topics than people who discuss non-controversial subjects like kittens or origami.
Musk’s silly sampling plan is just too bizarre for me to believe.
Carl T Bergstrom
Author, “Calling Bulls—“
Carl T. Bergstrom (University of Washington) was a co-author a book to help people understand dataCNBC was told by CNBC that sampling 100 followers from any one Twitter account is not sufficient to conduct “due diligence” in order to make a $44-billion acquisition.
According to Musk, a 100-member sample is orders of magnitude less than what social media researchers use for this kind of research. Selection bias is the greatest problem Musk could face when using this method.
Bergstrom sent a message about CNBC saying, “There is no reason to believe the followers of the official Twitter accounts are representative of all the accounts on this platform.” It is possible that bots have a lower likelihood of following this account, in an effort to avoid detection. Maybe they are more likely to do so to appear legit. It’s possible. Musk may be doing something other than trolling with us through this ridiculous sampling scheme, but I don’t think so.