Networked user engagement

210 Aufrufe

Veröffentlicht am

Networked user engagement

Veröffentlicht in: Daten & Analysen
0 Kommentare
0 Gefällt mir
Statistik
Notizen
  • Als Erste(r) kommentieren

  • Gehören Sie zu den Ersten, denen das gefällt!

Keine Downloads
Aufrufe
Aufrufe insgesamt
210
Auf SlideShare
0
Aus Einbettungen
0
Anzahl an Einbettungen
2
Aktionen
Geteilt
0
Downloads
3
Kommentare
0
Gefällt mir
0
Einbettungen 0
Keine Einbettungen

Keine Notizen für die Folie

Networked user engagement

  1. 1. Networked user engagement user engagement optimization workshop - CIKM 2013 In  collabora*on  with:   Mounia  Lalmas,     Ricardo  Baeza-­‐Yates,     Elad  Yom-­‐Tov   Jane%e  Lehmann  
  2. 2. outline 1.  Mo%va%on   How  to  measure  user  engagement?       2.  Network  metrics   From  site  to  network  engagement.       3.  Measuring  networked  user  engagement   Case  study  based  on  the  network  of  Yahoo  sites.   Lights  on  by  JC*+A!  
  3. 3. 182-­‐365+1  by  meaganmakes   How  to  measure   User   Engagement?  
  4. 4. Measuring user engagement 4  JaneHe  Lehmann   MoJvaJon   User  engagement  on  Yahoo!   Web  traffic  reports:  Google  Analy%cs,  Alexa.com,  …   Popularity  (#Users)   Ac%vity    (DwellTime)   Loyalty    (AcJveDays)  
  5. 5. Measuring user engagement 5  JaneHe  Lehmann   MoJvaJon   However,  Yahoo!  has  many  sites:     maps,  flickr,  weather,  news,  shine,  etc.     How  to  measure  engagement  within     a  network  of  sites?  
  6. 6. NETWORK OF SITES 6  JaneHe  Lehmann   MoJvaJon   frontpage   tv   sports   shopping   autos   search   daJng   jobs   news   shine   groups   maps   local   health   answer   weather   games   mail   omg   homes   travel   flickr   finance   Large  online  providers     (AOL,  Google,  Yahoo!,  etc.)     offer  not  one  site,     but  a  network  of  sites  
  7. 7. NETWORK OF SITES 7  JaneHe  Lehmann   MoJvaJon   frontpage   tv   sports   shopping   autos   search   daJng   jobs   news   shine   groups   maps   local   health   answer   weather   games   mail   omg   homes   travel   flickr   finance   Each  site  is  usually     op%mized  individually,     with  some  effort  to  direct     users  between  them  
  8. 8. NETWORK OF SITES 8  JaneHe  Lehmann   MoJvaJon   frontpage   tv   sports   shopping   autos   search   daJng   jobs   news   shine   groups   maps   local   health   answer   weather   games   mail   omg   homes   travel   flickr   finance   Online  mul%tasking   Users  switch  between  sites   within  one  online  session     (e.g.  emailing,  reading  news,     social  networking,   search)  
  9. 9. NETWORK OF SITES 9  JaneHe  Lehmann   MoJvaJon   frontpage   tv   sports   shopping   autos   search   daJng   jobs   news   shine   groups   maps   local   health   answer   weather   games   mail   omg   homes   travel   flickr   finance   Measuring  user     engagement  should     account  for     user  traffic  between     sites...  
  10. 10. NETWORK OF SITES 10  JaneHe  Lehmann   MoJvaJon   frontpage   tv   sports   shopping   autos   search   daJng   jobs   news   shine   groups   maps   local   health   answer   weather   games   mail   omg   homes   travel   flickr   finance   …  to  detect   sites  that  are     used  together  
  11. 11. NETWORK OF SITES 11  JaneHe  Lehmann   MoJvaJon   frontpage   tv   sports   shopping   autos   search   daJng   jobs   news   shine   groups   maps   local   health   answer   weather   games   mail   omg   homes   travel   flickr   finance   …  to  detect   isolated   sites  
  12. 12. NETWORK OF SITES 12  JaneHe  Lehmann   MoJvaJon   frontpage   tv   sports   shopping   autos   search   daJng   jobs   news   shine   groups   maps   local   health   answer   weather   games   mail   omg   homes   travel   flickr   finance   …  to  detect   important   sites  
  13. 13. Danboard  by  sⓘndy°   From  site  to  network   engagement     Looking  beyond  one           own  nose!  
  14. 14. ENGAGEMENT NETWORKS 14  Networked  User  Engagement  JaneHe  Lehmann   State  of  the  art   Our  approach   Non- provider service Provider service Provider network Traffic Provider service under study G=(V,  E,  λ)   V:   n<i>  -­‐  Sites,  services,  funcJonaliJes,  etc.   n<e>  -­‐  External  sites   E:   e<i>  -­‐  Traffic  between  internal  nodes   λ(e):   Traffic  volume  (#Users)  
  15. 15. Network  metrics            TrafficVolume          Def.:  Sum  of  edge  weights          A  high  value  is  a  sign  of  a  high  networked  user  engagement;            users  navigate  oHen  between  the  sites  of  the  network.            Density          Def.:  ProporJon  of  edges  to  maximum  possible          A  high  value  is  a  sign  of  a  high  networked  user  engagement;            users  navigate  between  many  different  sites.            Reciprocity          Def.:  RaJo  of  number  of  edges  in  both  direcJons          A  high  value  is  a  sign  of  a  high  networked  user  engagement;          users  do  not  only  navigate  from  one  site  to  another,  they  also  tend  to  return            to  previously  visited  sites.   Network-LEVEL METRICS 15  Networked  User  Engagement  JaneHe  Lehmann   30 78 22 15 45 9 90 ConnecJvity  Size  
  16. 16. Network  metrics            Modularity  [subNWmod]          Def.:  Captures  the  existence  of  modules  based  on  a  random  walk  approach          A  high  modularity  indicates  that  users  visit  many  sites  of  one  subnetwork,            but  hardly  navigate  to  other  subnetworks.            Global  Clustering  Coefficient  [GCC]            Def.:  Captures  the  existence  of  Jghtly  connected  groups  of  nodes          A  high  value  is  a  sign  of  a  high  networked  user  engagement;          users  access  sites  directly,  instead  of  using  front  pages.               Network-LEVEL METRICS 16  Networked  User  Engagement  JaneHe  Lehmann   Low GCC High GCC Modularity  
  17. 17. ENGAGEMENT NETWORKS Modularity  [subNWmod]   Yahoo!  Network  with  3  countries   17  Networked  User  Engagement  JaneHe  Lehmann   low  modularity  è  low  engagement   high  modularity  è  high  engagement  
  18. 18. ENGAGEMENT NETWORKS Modularity  [subNWmod]   Yahoo!  Network  with  3  countries   18  Networked  User  Engagement  JaneHe  Lehmann   low  modularity  è  low  engagement   high  modularity  è  high  engagement   low  modularity     è  high  engagement   Country  
  19. 19. lem  by    [  embr  ]     Case  Study     Measuring     Networked  user   engagement  
  20. 20. ENGAGEMENT NETWORKS Interac%on  data   •  July  2011   •  2M  user,  25M  sessions   •  728  Yahoo!  sites  (news,  mails,  search,  etc.)     Networks   •  12  weighted  directed  networks,  filtered  by   »  Edge  types    (internal,  internal  +  returning)   »  Time    (Wednesday,  Sunday)   »  User  loyalty    (causal,  acJve,  VIP)   »  Country    (5  EU  countries)     •  Applying  metrics  on  networks,  results  are   normalized  using  the  z-­‐score   20  Networked  User  Engagement  JaneHe  Lehmann   external   returning  edge   z-­‐score   trafficVolume Country1 Country3 Country4 Country2 Country5 average above average below average
  21. 21. Network-LEVEL METRICS Edge-­‐based  networks                 Leaving  the  network  does  not  necessarily  entail  less  engagement   •  Traffic  increases  significant  when  accounJng  for  returning  traffic  [18.26%]   •  ConnecJvity  increases  à  User  use  new  navigaJon  paths   •  Modularity  increases  à  Users  stay  in  the  same  part  of  the  network     21  Networked  User  Engagement  JaneHe  Lehmann   trafficVolume density reciprocity subNWmod GCC Size Connectivity Modularity Edgeint Edgeint+ext Timewed Timesun Usercasual Usernormal UserVIP
  22. 22. Network-LEVEL METRICS Time-­‐based  networks                 Naviga%on  is  more  goal-­‐oriented  during  the  week   •  TrafficVolume  and  density  are  lower  on  Sunday  à  Less  acJvity   •  Reciprocity,  subNW,  and  GCC  are  higher  on  Sunday  à  User  return  more  omen  to  already   visited  sites  and  visit  more  sites  than  usual  (browsing  with  less  specific  goals)     22  Networked  User  Engagement  JaneHe  Lehmann   trafficVolume density reciprocity subNWmod GCC Size Connectivity Modularity Edgeint Edgeint+ext Timewed Timesun Usercasual Usernormal UserVIP
  23. 23. Network-LEVEL METRICS User-­‐based  networks                 Loyal  users  have  also  a  higher  networked  user  engagement   •  TrafficVolume  is  higher  for  VIP  user  à  More  acJvity   •  ConnecJvity  and  modularity  is  higher  for  VIP  user  à  Users  are  interested  in  more  sites   •  GCC  is  lower  for  causal  users  à  Users  access  sites  using  the  front  pages     23  Networked  User  Engagement  JaneHe  Lehmann   trafficVolume density reciprocity subNWmod GCC Size Connectivity Modularity Edgeint Edgeint+ext Timewed Timesun Usercasual Usernormal UserVIP
  24. 24. trafficVolume density reciprocity subNWmod GCC Size Connectivity Modularity Country1 Country3 Country4 Country2 Country5 Network-LEVEL METRICS Country-­‐based  networks                 Networked  user  engagement  differs  between  countries   •  Country1:  Highest  trafficVolume  +  GCC,  high  density  +  reciprocity,  lowest  subNWmod  à  high   engagement   •  Country5:  Low  trafficVolume,  but  high  density    +  GCC,  low  subNWmod  à  less  popular,  but  high   engagement   24  Networked  User  Engagement  JaneHe  Lehmann  
  25. 25. CONCLUSION AND FUTURE WORK •  Network  analysis  enhances  the  understanding  of  user  engagement.   It  allows:   •  To  analyze  users  browsing  behavior  on  a  global  scale   •  To  compare  networks  of  sites  (e.g.  of  different  countries)   •  To  observe  differences  over  Jme   •  Leaving  the  network  does  not  entail  less  engagement   •  Some  network  metrics  are  more  useful  than  others   Future  work:   •  DefiniJon  of  metrics  that  combine  network  traffic  and  engagement   •  Comparing  traffic-­‐  and  hyperlink-­‐networks   •  Studying  the  effect  of  changes  in  the  network   25  Networked  User  Engagement  JaneHe  Lehmann  
  26. 26. Janebe  Lehmann   Universitat  Pompeu  Fabra,  Spain   lehmannj@acm.org     Mounia  Lalmas   Yahoo  Labs  London   mounia@acm.org     Ricardo  Baeza-­‐Yates   Yahoo  Labs  Barcelona   rbaeza@acm.org     Elad  Yom-­‐Tov   Microsof  Research,  Herzliya,  Israel   elad@ieee.org    

×