How to spot a bot

What's a bot? An Automated account, often post at high rates. Mostly RT's, likes, follows. Often doesn't write it own post

Transcript

1. How to spot a bot Andy Carvin Sr. Fellow, DFRLab acarvin@atlanticcouncil.org
2. What’s a bot? • Automated account • Often posts at a high rate • Mostly RTs, likes, follows • Often doesn’t write own posts
3. Caveats • Not all bots are bad • Not all bots are actually bots • Don’t write “this is a bot” • Better: “This behaves like a bot” or “this exhibits bot-like behavior”
4. How bots impact public opinion • They target algorithms more than people • They make topics/campaigns/people appear more popular or influential than they really are • Used to drown out conversations
5. Bot indicators 1. High levels of activity 2. Anonymity 3. Amplification 4. No profile picture 5. Stolen/shared profile 6. Suspicious handle 7. Posting in multiple languages 8. Automation software 9. Posting 24/7 10. Hashtag spamming Four or more = potentially a bot
6. High levels of activity @sunneversets100 11/16/2016-8/28/2017: 203,197 tweets 203,197 tweets % 288 days = 705 posts per day
7. What about this one? 76,322 tweets in one year 76,322 tweets / 365 days = 209 posts per day. Bot?
8. Nope. It’s me. I’m not a bot. What’s the lesson?
9. Nope. It’s me. I’m not a bot. What’s the lesson? Some people tweet a lot.
10. Anonymity • Lack of info about a real person • Images purposely don’t show anyone • Handle/username mismatch
11. Is this a bot or a person?
12. It’s Tsokhar Tsarnaev.
13. Amplification • Almost all RTs/shares, likes, etc • Copies a lot of text from other sources • Repeating the same text in many posts
14. @sunneversets100
15. @sunneversets100 www.twitonomy.com
16. No profile pic Some of @JessikkaAro’s Twitter followers
17. Stolen or shared photo
18. images.google.com
20. Suspicious handle Machine-generated user handles with random sequences or sequential numbering
21. Mismatched names Mismatched genders
22. Posting in multiple languages • It’s unusual to see people tweeting in more than two languages • But it’s not that unusual, so you have to scrutinize the context
24. Automation software • Online tools that automates posting to social media platforms • ift.tt and ow.ly among the most common examples • Important: Auto-posting has many legitimate uses, including news publishing and political campaigns
26. Tweeting 24/7
28. Hashtag spamming • Hashtags are extremely popular on Twitter and Instagram, among other platforms • Not unusual for influencers and business to include lots of hashtags • Their goal: attract attention via popular tags monitored by lots of people
30. #instagood: 976 million posts on instagram
33. Identifying botnets Four or more = likely coordinated 1. Speech patterns 2. Verbatim posts 3. Handle patterns 6. Date and time of creation 7. Identical activities 8. Location patterns
34. Case study: Libya
35. Speech patterns
36. Verbatim posts
37. Handle patterns
38. Shared creation date/time
39. Identical posting activity
40. Location patterns Similar or matching places in user profile 1. Greater London 2. England more broadly, but not the rest of the UK 3. Metropolitan Paris
41. Let’s go bot-spotting! • @amelia74698445 • @30de2e4d6c4f416 • @p79252228585 • @mariangelap17
42. facebook.com/DFRLab For the latest from the DFRLab, follow us on social media: @DFRLab @dfrlab