Clicky

Twitter top 100. Research

Twitter top 100. Research


The question of how much of the audience of the popular accounts is not a real, attracted attention since the first accounts gained a million followers.


Over the last 5 years there have been many publications on this topic. For 99% of the articles were used as sources, only two systems which provide information on the bots:


Twitter Audit and Status People. In the analysis of of separate accounts, journalists and researchers are also using the service Bot Or Not. Although it is not designed for a mass audience verification, we consider its methodology for the identification of bots, as it is the most sophisticated and detailed.


Each of the systems have their own advantages and disadvantages, they all provide services for the identification of bots.



Twitter audit


Twitter audit methodology is open and the number of examined accounts for each test is known. They estimate 5,000 followers (if that amount is actually there) to 5 parameters, and then extrapolates the findings in proportion to the current number of followers. Some of the results were obtained a year or more ago.


5 Twitter audit scores are similar to key metrics STAQ and allow using a small sample to determine the number of level 1 bots quite accurately. At the moment, this is the best online service for express analysis of the audience.


But Twitter Audit does not identify bots 2 and Level 3, or even implies they might exist. A high percentage of real users on the audit of this service does not guarantee the absence of more sophisticated bots.




Fakers


Fakers does not disclose its methodology and the number of accounts studied for each analysis. We can only examine the results of their work not knowing exactly how they were obtained.


All the followers the system divides into 3 categories: fake, inactive, good. Inactive accounts are the largest part of the audience in most cases. But who are these inactive somebody? Inactive people or inactive bots? Moreover, it is one of the key features for the bots 1 level.


The second and the most important thing. Results of the study of the same audience according to Status Рeople can be diametrically opposite. For example, two Top 100 accounts of Indian showbiz stars, Hrithik Roshan @iHrithik (15.1M) and Priyanka Chopra @priyankachopra (15.3M) have 9.5M (63%) common followers. But the same audience in various cases is assessed quite differently, from 70% to 0% of fakes.


iHrithik 70% fakes, 21% inactive, 9% good.


priyankachopra 0% fakes, 17% inactive, 83% good.


So we can’t recommend this analytical system as a trusted source.



BotOrNot


Bot or not is intended to identify the single advanced bots. It does not define the primitive bots, it does not check the accounts which do not have tweets. It does not aim to analyze a large audience, but considers this an important and urgent problem.


“Yet many research questions remain open. For example, nobody knows exactly how many social bots populate social media… link.


The system analyzes many parameters, but none of them is critical. As a result, the analytical conclusion is given in the form of a recommendation to consider this account as a bot with some probability, expressed as a percentage. The problem of this system, and all other existing systems is the lack of a complete classification of bots. We need the set of metrics with which we have 100% confidence it is a bot.


Various types of bots have different goals, so their stats and various behavioral models differ fundamentally. Checking different types of bots the same algorithm can not provide reliable results. Characteristic features for one type of bots are not typical for the other. For example, some are inactive most of the time, other tweets at a furious pace.


Our bot detection technology is based on the results of analysis of millions of accounts. This method has been confirmed by experiment with the purchase of bots. According to its results was detected 18 million botnet, as well as the majority of its customers. top-7 accounts consist mainly of these bots are still in the Top 1000 Twitter.


Also, we had found a stand up comedian, openly bought 1 million followers as a social experiment, and writing about it in the description of the account. Based on the study of his audience, it was part of the same botnet, which was discovered by us in the experiment.


In a nutshell, the bot 1 level can be described as an inactive mass follower. It has no purpose to look like a human, it just needs to increase the size of someone's audience. Any activity farther away are not required. It exists just as a static statistical index, and nothing more.


The object of our research is the audience, rather than a single account. There are no separate bots that try to imitate people. Even high-quality bots are part of a botnet.


In detecting bots 1 level, we cut off bots second and third levels. To determine advanced bots other researchers are needed. But we do not need to look for them among the level 1 bots further. This study allowed us to filter out almost 60% of the 523 million accounts.