PAT PILCHER discovers a cool free utility to help you detect Twitter bots and conspiracy crazies.
The sheer volume of false information swirling around on Twitter is boggling. While most of it’s obviously fake, some has been cunningly crafted to seem almost credible, even if it is total bullshit. The impact of false information can be catastrophic.
A potent example of this is the bogus conspiracy nuttiness that has seen people setting fire to cell towers in the mistaken belief that 5G is behind the coronavirus pandemic.
This and other online cray-cray is often disseminated by bots, apps that can control a Twitter account to tweet human-like responses. Estimates vary, but it’s suspected that 15 percent of all Twitter users are bot accounts.
None of this has escaped the attention of the folks from NortonLifeLock Research Group (NRG), who have developed a machine learning model that scans tweets to tell a human from a bot with a surprising degree of accuracy.
Best of all, NRG has released a beta version of this tech called BotSight. It is available as a free download and takes the form of a browser extension on most major web browsers.
Having downloaded it, I noticed that BotSight gave each Twitter user in my timeline a vivid green bot/human-probability score. Because these are dished out in real-time, telling if a wacky tweet is the work of a pile of code (or a human) is dead easy.
Any Twitter account with a score of over 90 percent is almost always going to be a human.
That said, it was a close call with my account which (much to my relief) garnered a score of 94 percent, which thankfully means I am real – who’d have thought!?
Bizarrely, Botsight found that Judith Collins tweets were not the work of a Terminator and was more human than I was. They had a 98 percent score indicating that she is, in fact, human (just what the remaining 2 percent is doesn’t bear thinking about).
Botsight uses five techniques to work its magic. These include IP-based correlation which looks at how accounts are linked geographically. Additionally, there’s a time-based correlation which examines Twitter activity over time. Add to this signs of automation such as usernames and other account metadata, as well as tweet content analysis and you’re well on your way to detecting (and blocking/muting) bots.
NRG found that 5 per cent of the overall volume of tweets examined were the work of bots. This is decreasing, even if the number of bots ramps up with trending topics like #5G or #covid19.
If you’re tired of conspiracy theory nuttiness and want a simple, low fuss way to sort the Twitter wheat from the chaff, BotSight might be just what the doctor ordered.