The problem is this one:
https://rxivist.org/papers/8472
Which had 33,000+ downloads added by a bot. A sample size of 1 is a disaster for detecting these things going forward, but can we develop some kind of rule that will flag suspicious patterns? Could the pattern simply be "An unreasonable increase in the download count of a single month, compared to the months on either side"? Is there a tight enough correlation between tweets and downloads that we could use that?
The problem is this one: https://rxivist.org/papers/8472 Which had 33,000+ downloads added by a bot. A sample size of 1 is a disaster for detecting these things going forward, but can we develop some kind of rule that will flag suspicious patterns? Could the pattern simply be "An unreasonable increase in the download count of a single month, compared to the months on either side"? Is there a tight enough correlation between tweets and downloads that we could use that?