Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
What	  Stops	  Social	  Epidemics?	              Greg	  Ver	  Steeg	     Rumi	  Ghosh	  &	  Kris:na	  Lerman	       USC	  ...
Informa:on,	  viruses,	  etc.	  spread	  from	  	       node	  to	  node	  on	  a	  network	                              ...
•  What	  is	  an	  epidemic?	  We	  observe	  many	     cascades	  that:	     –  	  Grow	  quickly	  ini:ally	     –  	  ...
What	  is	  an	  epidemic?	     On	  an	  infinite	  graph,	  an	  epidemic	  is	  any	  process	  that	  spreads	  to	  a	...
Social	  news:	  
Distribu:on	  of	  cascade	  size	  on	  -­‐-­‐-­‐-­‐-­‐-­‐	                                                    #nodes	  ∼...
Why	  are	  these	  cascades	  so	  small?	                                                                               ...
Maybe	  graph	  structure	  is	  responsible?	                                                                            ...
What	  about	  the	  spreading	           mechanism?	                                       Infected	                     ...
How	  important	  are	  repeat	           exposures?	                                          More	  than	  half	        ...
How	  do	  people	  respond	  to	    repeated	  exposure?	                                       Not	  much.	             ...
Big	  consequences	  for	  epidemic	  growth	  •  Most	  people	  are	  exposed	  to	  a	  story	  more	  than	     once	 ...
Weak	  response	  to	  repeated	  exposure	                                                     Take	  effect	  of	        ...
Also	  explains	  dynamics	                               Number	  of	  new	                               people	  expose...
Transmissibility:	  the	  percentage	  of	  new	  people	    exposed	  who	  end	  up	  infected/vo:ng	                   ...
Structure	  +	  Behavior	                                                                                                 ...
Summary	  •  Informa:on	  spread	  ≠	  Disease	  spread	  •  Big	  consequences	  for	  epidemics	  •  Repeat	  exposures	...
Decay	  of	  novelty	                               18	  
Weak	  response	  to	  repeated	  exposure	                                       Standard	  model	                       ...
What	  is	  an	  epidemic	  on	  a	  finite	  graph?	                                                           Infected	  ...
What	  is	  an	  epidemic	  on	  a	  finite	  graph?	                       Same	  number	  of	  red	  dots	               ...
Sub-­‐epidemic	  cascade	  size	                Reproduc:ve	  number,	  R0	  is	  the	                average	  number	  o...
1Sub-epidemic cascade size = 1 +                                1 − R0                              200                   ...
Satura:on	  on	  a	  real	  graph	                                              24	  
25	  
26	  
Upcoming SlideShare
Loading in …5
×

Icwsm-2011-What stops social epidemics

1,101 views

Published on

Look for my talk on videolectures.net

Published in: Technology
  • Be the first to comment

  • Be the first to like this

Icwsm-2011-What stops social epidemics

  1. 1. What  Stops  Social  Epidemics?   Greg  Ver  Steeg   Rumi  Ghosh  &  Kris:na  Lerman   USC  Informa:on  Sciences  Ins:tute    
  2. 2. Informa:on,  viruses,  etc.  spread  from     node  to  node  on  a  network   Infected   Not  Infected   Transmissibility,  λ   =  Probability  to   infect  your   neighbor   2  
  3. 3. •  What  is  an  epidemic?  We  observe  many   cascades  that:   –   Grow  quickly  ini:ally   –   But  remain  too  small  for  standard  (viral)  epidemic   models    •  Informa:on  cascades  differ:   –  Response  to  repeated  exposure  is  important  on   Digg  (and  TwiVer)   –  Dras:cally  alters  predic:ons  about  size  of   epidemics   3  
  4. 4. What  is  an  epidemic?   On  an  infinite  graph,  an  epidemic  is  any  process  that  spreads  to  a   frac:on  of  all  the  nodes   1  Frac:on  of  nodes  infected   Epidemic  threshold   predicted  for  many   0   cascade  models     Transmissibility,  λ   4  
  5. 5. Social  news:  
  6. 6. Distribu:on  of  cascade  size  on  -­‐-­‐-­‐-­‐-­‐-­‐   #nodes  ∼300k   Most  cascades   less  than  1%  of   total  network   size!   A  small  frac:on  is   s:ll  a  frac:on,   though,  right?   6  
  7. 7. Why  are  these  cascades  so  small?   Standard  model   of  epidemic   growth   Most  cascades  fall  in   (Heterogenous   this  range   mean  field  theory,   SIR  model,  same   degree  distribu:on   as  Digg)   λ,  Transmissibility  Transmissibility  of  almost  all  Digg  stories   7  fall  within  width  of  this  line?!  
  8. 8. Maybe  graph  structure  is  responsible?   ←  Mean  field  predic:on   (same  degree  dist.)    ←  Simulated  cascades   on  a  random  graph  with   same  degree  dist.   epidemic   threshold      Simulated  cascades  on   the  observed  Digg  graph   transmissibility λ    clustering  reduces  epidemic  threshold  and  cascade  size,              but  not  enough!   8  
  9. 9. What  about  the  spreading   mechanism?   Infected   Not  Infected  ? Independent   Cascade  Model   implicit  in  many   epidemic  models   9  
  10. 10. How  important  are  repeat   exposures?   More  than  half   exposed  to  a   story  more   than  once!   10  
  11. 11. How  do  people  respond  to   repeated  exposure?   Not  much.   We  have  similar   results  for   TwiVer  -­‐-­‐-­‐-­‐-­‐-­‐-­‐   Also  noted  by   Romero,  et  al,   WWW  2011   11  
  12. 12. Big  consequences  for  epidemic  growth  •  Most  people  are  exposed  to  a  story  more  than   once  •  Repeated  exposures  have  liVle  effect  •  Growth  of  epidemics  is  severely  curtailed   (especially  compared  to  Ind.  Cascade  Model)   12  
  13. 13. Weak  response  to  repeated  exposure   Take  effect  of   repeat  exposure   into  account:   Actual  Digg   Epidemic   cascades   threshold   unchanged   Result  of   simula:ons   λ*,  Tλ*   ransmissibility   13  
  14. 14. Also  explains  dynamics   Number  of  new   people  exposed  to  a   story  (who  don’t   vote  on  it)   Number  of  new   people  exposed  to  a   story  (who  do  vote)   14  
  15. 15. Transmissibility:  the  percentage  of  new  people   exposed  who  end  up  infected/vo:ng   Approximate  :me  of  story   15   promo:on  to  front  page  
  16. 16. Structure  +  Behavior   = Accurate   + model     of  behavior   MrBabyMan TalSiach kevinrose xdvx Independent   + = AmyVernon absolutelytrue oboymsaleem cascade   Bukowsky skored badwithcomputer Burento model   anderzole upick jaybollouiebaur IvanB vtbarrera 1KrazyKorean noupsell MrBabyMan TalSiach kevinrose xdvx + Accurate   = AmyVernon absolutelytrue oboy msaleem model     Bukowsky skored badwithcomputer Burento of  behavior   anderzole upick jaybol 16   louiebaur IvanB vtbarrera 1KrazyKorean noupsell
  17. 17. Summary  •  Informa:on  spread  ≠  Disease  spread  •  Big  consequences  for  epidemics  •  Repeat  exposures  are  important  on  Digg  and   TwiVer  •  On  Digg,  people  don’t  respond  to  repeat   exposure   –  Epidemic  threshold  unchanged   –  Dras:cally  reduces  size  of  epidemics   17  
  18. 18. Decay  of  novelty   18  
  19. 19. Weak  response  to  repeated  exposure   Standard  model   of  epidemic   growth   (Heterogenous   mean  field  theory,   SIR  model,  same   degree  distribu:on   as  Digg)   λ*   19  
  20. 20. What  is  an  epidemic  on  a  finite  graph?   Infected   Not  Infected   We  would  call  this  an   epidemic,  right?   And  now?   Epidemics  saturate  the   graph   20  
  21. 21. What  is  an  epidemic  on  a  finite  graph?   Same  number  of  red  dots   Epidemics  saturate  the   graph   21  
  22. 22. Sub-­‐epidemic  cascade  size   Reproduc:ve  number,  R0  is  the   average  number  of  new   spreaders  reached  by  each   spreader.     R0  <  1        No  Epidemic   R0 = kλ Average   number  of   Transmissibility   friends   22  
  23. 23. 1Sub-epidemic cascade size = 1 + 1 − R0 200 150 Cascade size 100 50 0 1average degree Λ, transmissibility Transmissibility  of  almost  all  Digg  stories   fall  within  width  of  this  line?!   23  
  24. 24. Satura:on  on  a  real  graph   24  
  25. 25. 25  
  26. 26. 26  

×