Communications of the ACM
Bot Hunting Is All About the Vibes
Some tools use machine learning to detect who's human and who's not.
Though several bot detection systems are available, there is no standard definition of "bot" and the results of the detection tools can differ.
Bot Sentinel's tool is trained to identify "problematic accounts" that engage in toxic trolling, not just those deemed to be automated, says the company's Christopher Bouzy. Meanwhile, the Botometer tool looks for accounts that are at least partly automated. And Carnegie Mellon University's Kathleen Carley, who helped develop BotHunter and BotBuster, defines bot accounts as those "run using completely automated software."
There also is no standard criteria that can accurately predict whether an account is a bot, and humans decide on the data used to develop and train these tools. Bouzy says Bot Sentinel is trained with tweets from users labeled problematic by Twitter based on its own policies.
Indiana University Distinguished Professor Filippo Menczer, who contributed to the Botometer project, says bot detection tools cannot be used blindly or expected to perform beyond their capabilities.
From Wired
View Full Article
Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA
No entries found