,
4 tweets,
4 min read
Read on Twitter
Can anything of diagnostic value be gleaned by studying patterns in how Twitter accounts reply? To start off, we'll plot the delay in reply (number of minutes between original tweet and reply) against the time of day of the original tweet for our own accounts.
cc: @ZellaQuixote
cc: @ZellaQuixote
Same technique applied to @ThreadReaderBot. The pattern makes it obvious that the bot generally replies either A) immediately or B) on roughly an hour delay, in contrast with our own plots which show a wide distribution in the delay time in cases where we don't reply immediately.
Next up: four accounts we've previously A) diagnosed as suspicious in general and B) have a large volume of replies. Original diagnoses linked for reference:
Comparison between a control group of six accounts and a set of accounts we've previous identified suspicious reply behavior from. The plots for the suspicious accounts look noticeably different from the control (interestingly, this is true for both bot and human troll accounts).
