Reverse image search can be a useful tool for identifying potentially bogus Twitter accounts. We decided to compare a few different sets of accounts and see how they measure up in the "borrowed photo" department.
cc: @ZellaQuixote
cc: @ZellaQuixote
@ZellaQuixote For this experiment, we focused on Yandex's reverse image search (we also use Google and TinEye.) Since the site is hosted in Russia, we elected to use Tor to connect.
@ZellaQuixote Botnet operators sometimes like to put a human face (or faces) on their automated works. Yandex reverse image search flagged the profile photos of 93 of the surviving 147 members (63.3%) of the "invisible spaces" botnet we recently posted about as duplicates, for example.
@ZellaQuixote Fully automated accounts aren't the only ones that use purloined pictures, of course. 61 of 228 (26.8%) of the tweetdecker/lobster accounts (users that engage in coordinated retweeting and later undo the retweets) currently on our radar use apparently borrowed profile pics.
@ZellaQuixote Next, we looked at a sample of 500 #resistance and 500 #MAGA accounts. Although many of them use pictures seen elsewhere on the Internet, most aren't photos of humans - only 5.2% of the #resistance and 3.6% of the #MAGA accounts are using possibly stolen pics of other people.
@ZellaQuixote Here's an overall comparison of nine sets of accounts. The percentage of accounts with likely stolen pics of other human beings as profile photos (rightmost column) might be a reasonable indicator of degree of inauthenticity in different sets of related accounts.
