HP Labs Research on "Twitter Cyborgs" and spam


Published on

June 2010

Published in: Business
  • Be the first to comment

  • Be the first to like this

No Downloads
Total Views
On Slideshare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

HP Labs Research on "Twitter Cyborgs" and spam

  1. 1. TWITTER CYBORGS<br />Miranda Mowbray, HP Labs<br />miranda.mowbray @ hp.com<br />and Nazareno Andrade, TU Delft<br />
  2. 2. “A cyborg is essentially a man-machine system in which the control mechanisms of the human portion are modified externally by drugs or regulatory devices so that the being can live in an environment different from the normal one"<br />New York Times<br />22 May 1960<br />
  3. 3. TWIT(TER)MINOLOGY<br />Tweets<br />Public timeline<br />Followers<br />Direct messages<br />Followees<br />Follow ratio<br />Trend<br />
  4. 4. SOME RELATED WORK<br />Java et al: follower and followee counts are correlated and have approx power-law distributions. Some automated use detected. <br />A Java, X Song , T Finin and B Tseng, “Why We Twitter: Understanding Microblogging Usage and Communities”, August 2007<br />Hubspot: the median number of tweets, followers, and followees are all zero. <br />D Zarella, “State of the Twittersphere”, June 2009<br />Cheng and Evans, Sysomos.com: 24% of all tweets made by accounts posting >150 tweets/day. <br />A Cheng and M Evans, “Inside Twitter: An in-depth look inside the Twitter world”, June 2009, and “Inside Twitter: An In-Depth Look at the 5% of Most Active Users” August 2009<br />Mowbray: dramatic rise in Summer ‘09 of accounts with > 100 tweets/day.<br />M Mowbray, “The Twittering Machine”, April 2010 http://www.hpl.hp.com/techreports/2010/HPL-2010-54.pdf<br />
  5. 5. OUR DATA<br />13447 twitterers<br />Randomly sampled from public timeline between 22 April and 3 May 2010<br />http://api.twitter.com/1/statuses/public_timeline.xml<br />1257 spammers <br />Reported to Twitter over 7 days <br />
  7. 7. RISE OF THE TWITTER CYBORGS, IIPercentage of sampled tweets that are from users sending >100 tweets/day<br />
  8. 8. “It’s about humans connecting with each other, and often in ways in which they couldn’t otherwise”<br />EvWilliams<br />Co-founder of Twitter<br />BBC interview, 5 August 2009<br />
  9. 9. WHAT ARE THE CYBORGS DOING?<br />Of the 33 sampled cyborgs tweeting > 700 times/day:<br />marketing and e-commerce<br />Both legit businesses and porn/phishing sites <br />services for which frequent updates are useful<br />Clocks, weather data, software update announcements, news headlines<br />services for which they aren’t all that useful<br />Torrent site, website rating site, online pastebin<br />
  10. 10. FOLLOWEE COUNTS<br />
  11. 11. FOLLOWBACK DATA<br />Experiment by Catalin Cosoi<br />5/10 for fake users with many tweets<br />Follow ratios for sampled spammers<br />0.3/10 for spammers with < 5 tweets<br />4.2/10 for spammers with ≥ 5 tweets<br />So if you’re a spammer, tweet at least 5 times<br />
  12. 12. THE PROBLEM WITH MEGAPHONES<br />Image by Jonas Boni / daoro on Flickr, http://www.flickr.com/photos/daoro/3382798413<br />CC licence http://creativecommons.org/licenses/by/2.0/deed.en_GB<br />
  13. 13. FOUR CYBORG SPAMMING TECHNIQUESand how Twitter might defend against them<br />
  14. 14. CYBORG SPAMMING 1<br />Mention spam<br />Mention usernames in tweets<br />Tweets are automatically sent to users whose names are mentioned<br />May tempt the user to follow you<br />Possible defence by Twitter:<br />Limit number of different usernames one user can mention in a time period<br />Excluding names of their followers<br />
  15. 15. CYBORG SPAMMING 2<br />Trend spam<br />Automatically include trend words in your tweets<br />Notoriously, #iranelection<br />Tweet will be shown if a user clicks on the trend in the trends list<br />Or does a Twitter-search for the trend<br />Possible defences by Twitter:<br />Order search results based on reputation<br />Suppress multiple-trend tweets<br />
  16. 16. CYBORG SPAMMING 3<br />Fake-retweet spam<br />Abuse retweet convention<br />“RT @name” convention for quoting others’ tweets<br />Abuse this to make your ad or phishing URL appear to be from someone else<br />Possible defence by Twitter: <br />Track retweets, suppress “implicit” ones<br />Machinery to do this is already in place<br />
  17. 17. CYBORG SPAMMING 4<br />Follow spam<br />Follow lots of users, send tweets or DMs to those that follow back<br />Some target types of user to follow <br />Some churn followees to get round Twitter’s ~2000 limit <br />“Auto-Follow, Auto-Unfollow, Auto-Tweet & DM.”<br />Possible defence by Twitter:<br />Require a CAPTCHA solution to follow a user<br />Unless the user follows you first<br />
  18. 18. A SAD STORY FROM XKCD<br />Image from xkcd.com, http://xkcd.com/632 , <br />CC licence http://creativecommons.org/licenses/by-nc/2.5/<br />
  19. 19. VARIANTS<br />Twitter zombies<br />Used to be human, account compromised<br />Eg. Barack Obama’s account<br />“To increase your follower count, type your Twitter password here”<br />Twitter werewolves<br />Automated only at certain times, human otherwise<br />Eg. Facebook games advertising on players’ Twitter accounts<br />Eg. Human account owner agrees to include automated ads<br />
  20. 20. LESSONS FOR HUMAN+CYBORG ENVIRONMENTS<br />Be careful when you release your API<br />But do release it<br />Limit automated use of the system <br />Increase cost (in time, money or effort) of unwanted cyborg behaviours <br />Opt-in-only marketing can still have large negative externalities<br />Don’t let cyborg marketers do lots of attention-catching at small or zero cost<br />
  21. 21. Thanks<br />