Original link: https://www.williamlong.info/archives/6800.html
According to reports, Tesla CEO Elon Musk’s acquisition of Twitter (Twitter) has been a deal breaker. Musk himself sampled the proportion of fake Twitter accounts, but his method was judged by experts to be too “stupid”. On Friday, local time, he abruptly announced that he would put the $44 billion acquisition of Twitter on hold, citing a need to personally study the proportion of fake and spam accounts on the platform. The move sent Twitter shares tumbling on the day.
Questioning Twitter data
Later, Musk clarified that he was still committed to acquiring Twitter, but he still emphasized the troubles caused by the fake account problem. Musk revealed that his own team will conduct independent research and analysis. Musk expressed skepticism about the fake account data Twitter released in its most recent earnings report.
In announcing its first-quarter earnings, Twitter management said there were indeed some fake and spam accounts on the platform, as well as “legitimately profitable” daily active users.
Twitter said it conducted an internal analysis of a sample of Twitter user accounts and, in the first quarter of this year, estimated that fake and spam accounts accounted for less than 5 percent of daily active users.
Twitter also admitted that it had overestimated its active users by 1.4 million to 1.9 million over the past three years. According to reports, in March 2019, Twitter once launched a function that allows a user to link different Twitter accounts to facilitate switching, but this function made a mistake, that is, some interactive activities carried out by the parent account, As a result, all associated accounts are identified as daily active users.
Some of the actions Musk will take next may not be enough, experts say.
Musk’s statistical approach
In the tweet, Musk described how his team will identify fake accounts, spam accounts and duplicate accounts on Twitter.
Musk’s team will follow a random sample of 100 followers on “the official Twitter account,” and he invites other users to repeat the approach. Musk further explained that you can just find a Twitter account with a lot of followers, ignore the first 1,000 followers, and then pick one out of every 10 followers (as a statistical sample), Musk also expressed his willingness to listen to other better ones. research method.
Musk also said that the reason for taking 100 people as a sample of fans is because this is the unit that Twitter calculates the number of users in its earnings report.
Any reasonable random sampling process is fine, Musk said, and if many independent accounts come up with similar proportions of fake accounts, spam accounts, and duplicate accounts, that could reflect an important message.
Twitter declined to comment on whether Musk clearly described the statistical approach the team would use.
not really random
Dustin Moskovitz, co-founder of the Facebook platform, also tweeted to join the discussion. Musk’s statistical methods are not truly random, he said, using too few samples, which could lead to huge biases.
Moskovitz also said that if you don’t trust the samples drawn by the Twitter team, that’s a problem in itself.
Christopher Bouzy, founder and CEO of the American technology company “BotSentinel”, said that according to his company’s analysis, the proportion of accounts that are not authentic on Twitter is as high as 10% to 15%, including fake accounts, spam accounts, duplicate accounts, and malicious activities. bot accounts, as well as single-purpose hate accounts (primarily attacking and harassing others), plus a large number of accounts that deliberately spread false information.
BotSentinel’s research method is mainly based on crowdsourcing. The company uses a combination of computer software and human reviewers to study Twitter’s inauthentic account activity. Currently, the company monitors 2.5 million accounts on Twitter, mostly English-language users.
Butch also said that the number of inauthentic accounts on Twitter also fluctuates depending on the relevant content topic. Research by his own company found that on controversial topics such as politics, cryptocurrencies, climate change, and the new coronavirus, the number of inauthentic accounts is higher than that of other non-controversial topics, such as cat raising and origami art.
Musk’s method is too “stupid”?
Carl T. Bergstrom, a professor at the University of Washington in the United States, said that for large-scale mergers and acquisitions of up to 44 billion US dollars, it is not qualified due diligence to take a sample of 100 fans of a single user.
The 100-piece sample is orders of magnitude smaller than what social media researchers do, Bergstrom said. Musk’s own statistical methods could be prone to what’s known as “selection bias.”
The academic said there was no reason to support the idea that followers of an official Twitter account were a representative sample of users of the Twitter platform. Maybe bot accounts may not follow official accounts in order to avoid being discovered, and maybe these bot accounts are more keen to follow others than human accounts, but Musk’s sampling research method is a bit “stupid”.
Source: Sina.com
This article is reprinted from: https://www.williamlong.info/archives/6800.html
This site is for inclusion only, and the copyright belongs to the original author.