Christopher Bouzy is try to stay ahead of the bots. As the person behind Bot Sentinel, a popular bot detection system, he and his team are constantly updating their machine learning models for fear of getting “old”. The task? Sort 3.2 million tweets from suspended accounts into two folders: “Bot” or “Not”.
To detect bots, Bot Sentinel’s models must first learn what problematic behaviors are due to data exposure. And by providing the model with tweets in two different categories — bot or no bot — Bouzy’s model can calibrate itself and allegedly find the gist of what he thinks makes a tweet problematic.
Training data is at the heart of any machine learning model. In the rapidly expanding field of bot detection, the way bot hunters define and label tweets determines the way their systems interpret and classify bot-like behavior. According to experts, this may be more of an art than a science. “Ultimately, it’s all about a vibe when you do the labeling,” says Bouzy. “It’s not just about the words in the tweet, the context matters.”
He’s a Bot, She’s a Bot, Everyone’s a Bot
Before anyone can hunt bots, they need to figure out what a bot is — and that answer changes depending on who you ask. The internet is full of people accusing each other of being bots because of minor political differences. Trolls are called bots. People with no profile picture and few tweets or followers are called bots. Even among professional bot hunters, the answers differ.
Bouzy defines bots as “problematic accounts” and trains Bot Sentinel to eradicate them. Filippo Menczer, a professor of computer science and computer science at Indiana University, says the tool he’s helping develop, Botometer, defines bots as accounts that are at least partially controlled by software. Kathleen Carley is a professor of computer science at Carnegie Mellon University’s Institute for Software Research and helped develop two bot detection tools: BotHunter and BotBuster. Carley defines a bot as “an account managed with fully automated software”, a definition that matches Twitter’s. “A bot is an automated account – nothing more, nothing less,” the company says wrote in a May 2020 blog post about platform manipulation.
Just as the definitions differ, the results these tools produce do not always match. For example, an account marked as bot by Botometer may come back as perfectly human on Bot Sentinel and vice versa.
Part of this is by design. Unlike Botometer, which aims to identify automated or partially automated accounts, Bot Sentinel hunts for accounts engaged in toxic trolling. According to Bouzy, you recognize these accounts when you see them. They may be automated or human-operated, and they engage in harassment or disinformation and violate Twitter’s terms of service. “Only the worst of the worst,” says Bouzy.
Botometer is maintained by Kaicheng Yang, a doctoral student in computer science at the Observatory on Social Media at Indiana University, who co-created the tool with Menczer. The tool also uses machine learning to classify bots, but when Yang trains his models, he isn’t necessarily looking for harassment or violations of terms of service. He’s just looking for bots. According to Yang, when he labels his training data, he asks himself one question: to believe does the tweet come from a person or an algorithm?”
How to train an algorithm
Not only is there no consensus on how to define a bot, but there is also no single criterion or signal that can pinpoint a researcher that accurately predicts whether an account is a bot. Bot hunters believe that exposing an algorithm to thousands or millions of bot accounts helps a computer detect bot-like behavior. But the objective efficiency of any bone detection system is clouded by the fact that people still have to decide what data to use to build it.
Take Botometer, for example. Yang says Botometer has been trained on tweets from about 20,000 accounts. While some of these accounts self-identify as bots, most are manually categorized by Yang and a team of researchers before being cracked by the algorithm. (Menczer says some of the accounts used to train Botometer come from datasets from other peer-reviewed research. “We try to use whatever data we can get as long as it comes from a reliable source,” he says. .)
