Artificial intelligence is used to find differences between human users and fake accounts on Twitter.
Emilio Ferrara at the University of Southern California in the US, and his colleagues have trained AI to detect bot on Twitter based on the different patterns of activity between real and fake accounts.
The team analyzed two separate data sets of Twitter users, who had been classified either manually or by pre-existing algorithms as bots or humans.
Manually verified dataset consists of 8.4 million tweets from 3500 human accounts, and 3.4 million tweets from 5,000 bots.
The researchers found that human users answered four to five times more often for other tweets than bots. Real users gradually become more interactive, with a small proportion of replies increasing during an hour-long Twitter usage session.
The length of tweets from human users also decreases as sessions increase. “The amount of information exchanged is reduced,” Ferrara said. He believes that the change may be the result of cognitive decline over time, where people are less likely to spend mental effort in compiling original content.
Bots, on the other hand, do not show changes in interactivity or the length of information they tweet from time to time.
The team also analyzed the amount of time between two consecutive tweets from one user. When this distribution is plotted, bots show spikes for specific time gaps, such as tweeting at 30 minute or 60 minute intervals.
The team then combined these steps to train an existing bot detection algorithm, called the Botometer, about differences in activity patterns. AI is significantly more likely to detect accurately to fake accounts than when it does not account for posting time.
Algorithms can be used to complement the others bot detection a tool that analyzes language in writing, Ferrara said.
One limitation of the research is that the Twitter data that the team analyzed came from three years ago. At that time, it is possible for bots to become more human-like in their activity patterns.
Journal reference: Border in Physics, DOI: 10.3389 / fphy.2020.00125
More about this topic:
to request modification Contact us at Here or [email protected]