Mon Apr 27 2020
A joint UK/US research team has trained an AI to be able to tell the difference between bots and humans on Twitter
Based on their activity patterns a team of AI researchers have claimed to have trained an AI (artificial Intelligence) to be able to spot the difference between humans and bots on the social media platform Twitter.
Apparently, the secret is that real people tend to respond a lot more frequently to tweets from other users when compared to bots.
The researchers conducted their study using a large Twitter database, analysing the changes in behaviour exhibited by both humans and bots over the course of an ‘activity session’.
Two different types of datasets were used in the study.
The first, which they designated French Elections (FE), contained over 16 million tweets, posted by more than 2 million different users between the 25th April 2017 and 7th May 2017.
The second dataset, which the researchers designated as Hand-Labelled (HL) contained “three groups of tweets produced by bot accounts active in as many viral spamming campaigns at different times, plus a group of human tweets."
During the analysis of both datasets the researchers examined numerous factors including:
The results of the study clearly showed that ‘real’ users responded almost four to five times more frequently to tweets from other users than the bots did.
As well as that though, the accounts operated by real users also showed a pattern of becoming more interactive over the course of an hour-long session, although with a declining amount of content produced in their tweets, whereas the bots remained consistent.
Emilio Ferrara, a professor at the University of Southern California's Information Sciences Institute, believes this behaviour could be a result of cognitive depletion in humans over time, in which they become less likely to spend mental efforts to create original content.
Bots, on the other hand, showed no major variations in interaction behaviour or the length of tweets posted over time on the platform.
The study also looked at the amount of time between consecutive messages between users and found that bits used clearly defined intervals to post (30 minutes, one hour etc).
All these results were then used to train Botometer, a pre-existing bot-detection algorithm which, after the data was fed in, showed much better performance in detecting bots than before.
Mon Apr 27 2020