Post

1/6 Out of 155 followers, 90+ were bots! Pretty annoying. They're gone now and I have a clean slate to start from. What really shocked me is how easy it was to identify them. Seriously, a few simple heuristics would capture almost 100% of cases. Is this something @XEng can help with? 🧵👇

Lost And Founding

@lostandfounding.com

6 Jan 2025

2/6 For example, by just looking at the usernames e.g. "bO9yMtn63I4Y8kA" and filtering all those with a high percentage of digits, I catch about 50% of cases straight away.

Lost And Founding

@lostandfounding.com

6 Jan 2025

3/6 If I then go a bit deeper, and look at accounts with a Following:Follower ratio of about 50:1 without a single Post or Reply on their account, I'd have caught another ~25%

Lost And Founding

@lostandfounding.com

6 Jan 2025

4/6 If they weren't already filtered by the above, I just looked at anyone with a female profile pic and a lack of Posts/Replies. That found a bunch more. There is definitely something in common with these PFPs. I can't articulate it but I'm sure a simple ML classifier could do it.

Lost And Founding

@lostandfounding.com

5/6 Lastly, I dealt with the impersonator type bots i.e. those who had taken a real account and varied the username by a few characters. These are harder to automate and, clearly, they do dupe a lot more people as they had a large number of followers in some cases.

6/1/25, 6:00 pm0 reposts0 likes

Lost And Founding

@lostandfounding.com

6 Jan 2025

6/6 Bottom line, I think something like a Bot Score would really help here. Similar to how @XEng has added the "Probably Spam" to replies, it would be good to have a "Probable Bot" or "Bot Rating: High" or something so that we could define our own thresholds for removing them