Bouzy is #TheMostUnseriousPerson. This 8 month-old account has all of 300 tweets, but scored as “Disruptive?” Afaict, their biggest offense is using hashtags supporting LawTubers @RekietaLaw and @natlawyerchic 🤔:
https://twitter.com/LauraC0526/status/1571608068173201408
2/ The stats on this account makes it pretty plausible that there’s some eigenvector centrality aspect to how network connections are accounted for. That or the model is really bad at classifying low-activity accounts 🤔:
https://twitter.com/angelaheinrich6/status/1572041315923148803
3/ I wonder how much, if any, weight the model gives to retweets (often done excessively by actual bots) or comments (no obvious indicator of a vaguely accurate sentiment classifier):
https://twitter.com/tsd_dragon/status/1572090035301908480
4/ The number of times I’ve heard people say they’ve never used BS before, yet they still have scores is interesting 🤔:
https://twitter.com/carmella_nicole/status/1572434458338529280
5/ Based on BS’ Privacy Policy, usernames entered into the “Analyze Account” function could be considered user-submitted data. So, theoretically, even if you never use the site, BS could include your account in their dataset if a third party analyzes your username?: 
6/ Kinda makes sense why CB glommed on to pro-Amber Heard Twitter and then started beef with LawTube. More publicity (good or bad) = more usernames as user-submitted data
7/ Alright, can confirm that your info will be included in BS’ dataset even if you’ve never used the site. Basically, anyone can reverse-search someone’s user ID by analyzing their account 😑
8/ Alright, read some articles on disinformation classifiers. TL;DR? Confirms everyone’s read of BS as either being really bad at what CB claims it’s doing or not doing what he says it’s doing
9/ In this study, they compared BS and Botometer to their own tool. Neither online tool performed that well when classifying troll and bot accounts lol: ncbi.nlm.nih.gov/pmc/articles/P…
10/ Per that article:
1. Both detectors have issues with false positives (“okay” accounts labeled as “bad”) and false negatives (“bad” accounts labeled as “okay”)
2. When the task was classifying troll/bot/disinfo accounts, BS’ performance is basically “idk, you tell me 🤷🏻♂️”
1. Both detectors have issues with false positives (“okay” accounts labeled as “bad”) and false negatives (“bad” accounts labeled as “okay”)
2. When the task was classifying troll/bot/disinfo accounts, BS’ performance is basically “idk, you tell me 🤷🏻♂️”
11/ In case you need further proof that CB has no idea wtf he’s talking about, I give you this quote from a 2020 interview: impakter.com/saving-twitter…
12/ He also claimed BS mastered classifying non-English Twitter accounts in a few weeks despite him not speaking Portuguese or knowing anything about Brazilian politics. He then said it’s “completely autonomous,” but also said it needs user help to learn: impakter.com/saving-twitter…
13/ Suffice to say, BS seems like a disingenuous way to use @TwitterDev API tools and get around @Twitter @TwitterSupport TOS RE: “targeted harassment.” Because CB isn’t actively tweeting targets for his followers to report en masse, it doesn’t qualify as “platform manipulation?”
14/ Also, BS should probably have a clearly indicated opt-out option for CA residents visiting the site. Afaict, there isn’t one 🤔:
https://twitter.com/carmella_nicole/status/1573740046829514753
15/ If @RonColeman or @NatetheLawyer need further examples of CB making contradictory statements RE: BS, look no further than this quote from 5/15/2020 and the BS FAQ from 5/28/2020:
@threadreaderapp unroll, please 🙏🏽
• • •
Missing some Tweet in this thread? You can try to
force a refresh
