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.

Social Media for Safety Characterising Online Interactions between Citizens and Police

695 views

Published on

Social Media for Safety Characterising Online Interactions between Citizens and Police

Niharika Sachdeva, Ponnurangam Kumaraguru, Munmun De Choudhury

Published in: Data & Analytics
  • Be the first to comment

Social Media for Safety Characterising Online Interactions between Citizens and Police

  1. 1. Social Media for Safety Characterizing Online Interactions between Citizens and Police Niharika  Sachdeva   Ponnurangam  Kumaraguru,  Munmun  De  Choudhury   HCI  2016   CERC,  IIITD  and  Georgia  Tech  University,  USA   niharikas@iiitd.ac.in  
  2. 2. cerc.iiitd.ac.in   Expressing Safety Concerns 2  
  3. 3. cerc.iiitd.ac.in   Traditional Methods – Police  rely  on  one  –  way  communicaKon  to   reach  ciKzens   3   69%   45%   hRp://www.accenture.com/us-­‐en/Pages/insight-­‐police-­‐ciKzen-­‐survey-­‐policing-­‐social-­‐media.aspx  
  4. 4. cerc.iiitd.ac.in   Better Interactive Medium for Citizen Participation 4   50%   Website  and  Portals   23%   Smartphone   81%   Preferred   Plaorm   hRp://www.accenture.com/us-­‐en/Pages/insight-­‐police-­‐ciKzen-­‐survey-­‐policing-­‐social-­‐media.aspx   Social  Media  may   be  best  channel   for  ciKzen-­‐police   engagement  
  5. 5. cerc.iiitd.ac.in   Police Use Social Media 5  
  6. 6. cerc.iiitd.ac.in   Top Officials also on Twitter 6   InteresKng  Hashtags  Strike  a  conversaKon  
  7. 7. cerc.iiitd.ac.in   #myNYPD 7  
  8. 8. cerc.iiitd.ac.in   Measuring Human Behavior   - Exploring  the  feasibility  of  social  media  in   quanKfying  aRributes  of  communicaKon   - IdenKfying  behavioral  aRributes  like  affecKve   expression,  engagement  and  social  and  cogniKve   response  processes   8   CiKzen  to  CiKzen   CiKzen  to  Police   Police  to  CiKzen     Police  to  Police  
  9. 9. cerc.iiitd.ac.in   Research Questions – RQ  1:  Topical  Characteris3cs   - Nature  of  content  and  topics  that  characterize   social  media  discussion  threads     – RQ  2:  Engagement  Characteris3cs   - How  do  ciKzens  and  police  engage  in  social  media   discussion  threads?     – RQ  3:  Emo3onal  Exchanges     - Nature  of  emoKons  and  affecKve  expression  that   manifest  on  social  media     – RQ  4:  Cogni3ve  and  Social  Orienta3on   - What  are  the  linguisKc  aRributes  that  characterize   cogniKve  and  social  response  processes?     9  
  10. 10. cerc.iiitd.ac.in   Methodology 10   85  Public  and  official  Police  Department   Average  age  3  years  (from  2010  –  April   2015)   47,474  wall  posts  and  85,408  status   updates  
  11. 11. cerc.iiitd.ac.in   Data Categorization 11    DT  w/  ≥  1   Comment   P&C   C  Total  DT   85,408   47,474   46,845     24,984   5,519   17,196   41,326   7,788   PP&C   CP&C   PC   CC  
  12. 12. cerc.iiitd.ac.in   Measures of Behavior 12   Topics   •  N  Gram  Analysis   •  K-­‐means  Clusters   Engagement   •  No.  of  police  and  ciKzen  who  comment  in  DTs   •  DisKnct  ciKzens  who  comment  in  DTs   •  Shannon’s  Wiener  Diversity  index   •  Average  no.  of  likes  and  comments   Emo6onal     •  Valence   •  Arousal   Social  and   cogni6ve   •  Interpersonal  Focus   •  Social  OrientaKon     •  CogniKon     LIWC  and  Anew  DicKonary   LIWC  DicKonary  
  13. 13. cerc.iiitd.ac.in   Topic Characteristics –          13   Unigram   Freq.   Unigram   Freq.   rules   0.015   safety   0.012   safety   0.014   following   0.011   violaKons   0.014   noKce   0.010   challans   0.011   prosecuted   0.009   please   0.011   movement   0.008   ciKzens   0.01   complaint   0.008   Focus  on  advisories,  the  status  of  different  cases  being   invesKgated   (U  =  700,  p  <  .05,  z  =  −3.57)  
  14. 14. cerc.iiitd.ac.in   Topic Characteristics 14   Most  posts  tend  to  request  police  to  take  acKon  on   their  complaints   Unigram   Freq.   Unigram   Freq.   please   0.026   people   0.022   take   0.021   please   0.02   acKon   0.019   one   0.019   people   0.019   take   0.016   one   0.019   acKon   0.015   Kme   0.017   Kme   0.015   near   0.017   number   0.013   Higher  Reference  to  “people”  
  15. 15. cerc.iiitd.ac.in   Clusters of Topics – Police  iniKated  discussions  are  more  focused   than  ciKzen  iniKated.   15   Awareness  drive  /  safety  campaigns     Road  sense  is  the  offspring  of  courtesy  and  the  parent  of  safety   Prosecuted  /  ac6on  taken  reports   Ac3on  taken  by  [Withheld],  Reg  your  tweet  pe33on,   @[withheld];  33  parking  tag  &  6  no  parking,  1  foot  path   parking.  Cases  booked  on  hospital  road   Advisories  on  situa6ons   Good  -­‐-­‐  Morning  to  all  the  Commuters  of  Shillong  City,   there  is  heavy  movement  over  NH  -­‐  40  –  44  and   Madanr3ng  down  side,  Lumdiengjri  area  stretch.   Please  do  not  overtake  
  16. 16. cerc.iiitd.ac.in   Clusters of Topics 16   –  Police  iniKated  discussions  are  more  focused  than   ciKzen  iniKated.   Apprecia6on   Hear3est  congratula3ons  to  [withheld]  police  for  nabbing   [withheld]  agent  within  24hrs.  wow!!!  Kudos  and  respect   Newspaper  ar6cles    Please  ACT:  h]p://3mesofindia.india3mes.com/videos/news/…   Ci6zen  6ps  and  complaints   4th  Nov  2014  [withheld]:  Driving  in  wrong  side  at  Teghoria  U  Turn   Neighbourhood  problems   “Learn  from  the  Delhi  incident  and  ensure  that  no  buses  in  Kolkata  have  3nted   glasses.  One  such  bus  was  spo]ed  on  Gariahat  road  Regn.  #.  [Withheld].   Kindly  take  appropriate  ac3on.  Thank  you   Missing  people   “Sir  plz  help  find  my  nephew,  he  is  missing  since  today   morning,  he  is  from  kodagu,  contact  [withheld]  
  17. 17. cerc.iiitd.ac.in   Research Questions – RQ  1:  Topical  Characteris3cs   - Nature  of  content  and  topics  that  characterize   social  media  discussion  threads     – RQ  2:  Engagement  Characteris3cs   - How  do  ciKzens  and  police  engage  in  social  media   discussion  threads?     – RQ  3:  Emo3onal  Exchanges     - Nature  of  emoKons  and  affecKve  expression  that   manifest  on  social  media     – RQ  4:  Cogni3ve  and  Social  Orienta3on   - What  are  the  linguisKc  aRributes  that  characterize   cogniKve  and  social  response  processes?     17  
  18. 18. cerc.iiitd.ac.in   Engagement Characteristics – Content  GeneraKon         18   Police  +  CiKzens   55,028   1,79,176   17,124   12,630   CiKzens  Only   54,982   1,79,176   17,081   12,630   Entropy   4.39   4.96   3.23   3.6   Police   CiKzen   26%  lower   CP  &C  discussion  threads  might  be  contribuKng  more  towards  social  capital   building  6  (due  to  engaging  both  police  and  ciKzens)  than  those  involving   ciKzen  commentary  only    
  19. 19. cerc.iiitd.ac.in   Engagement Characteristics – Content  GeneraKon       19   Police  +  CiKzens   55,028   1,79,176   17,124   12,630   CiKzens  Only   54,982   1,79,176   17,081   12,630   Entropy   4.39   4.96   3.23   3.6   Police   CiKzen   10.28%   lower   Lower  entropy:  large  number  of  comments  are   posted  by  a  small  number  of  ciKzens  and  police   CP&C  Contribute  less  towards  police  endeavours   to  obtain  mass  public  parKcipaKon  in  community   policing  
  20. 20. cerc.iiitd.ac.in   Engagement Characteristics 20   Comments*   Likes**   Avg.   Std.  dev   Avg.   Std.  dev   Cp&c   3.34   19.19   9.4   253.85   Cc   3.69   13.79   13.38   201.57   – Content  InteracKon     Ci#zen  post:  “My  family  and  I  are  gegng  the  unwanted  calls  from   the  given  number  [withheld].  Especially  he  is  misbehaving  with  a   female  member.  My  Number  is  -­‐  [withheld]”   Police  reply:  “Dear  [withheld],  Please  visit  at  your  nearest  Police   Sta3on  and  lodge  a  complaint  with  details  and  they  will  assist  you   in  this  regard...  Thankyou”   9.49%  lower  29.75%  lower   Police  suggests  an  appropriate  acKon  and  the  discussion   tends  to  close  early,  resulKng  in  lower  interacKon  
  21. 21. cerc.iiitd.ac.in   Engagement Characteristics – Content  InteracKon     21   Comments   Likes   Avg.   Std.  dev   Avg.   Std.  dev   PP&C   19.68   86.17   114.71   805.55   PC   9.88   74.92   88.05   1025.53   30.28%  higher  99.19%  higher   PP&C  discussions  are  mainly  advisories  where  police   requests  ciKzens  for  some  acKon  or  shares  informaKon   with  them,  these  may  generate  more  engagement  
  22. 22. cerc.iiitd.ac.in   Research Questions – RQ  1:  Topical  Characteris3cs   - Nature  of  content  and  topics  that  characterize   social  media  discussion  threads     – RQ  2:  Engagement  Characteris3cs   - How  do  ciKzens  and  police  engage  in  social  media   discussion  threads?     – RQ  3:  Emo3onal  Exchanges     - Nature  of  emoKons  and  affecKve  expression  that   manifest  on  social  media     – RQ  4:  Cogni3ve  and  Social  Orienta3on   - What  are  the  linguisKc  aRributes  that  characterize   cogniKve  and  social  response  processes?     22  
  23. 23. cerc.iiitd.ac.in   Emotional Expressions – NegaKve  senKment  higher  in  ciKzen  iniKated   threads   23   CP&C   CC   Avg   Std.  dev   Avg   Std.  dev   NA   0.021   0.03   0.018   0.04   Anx   0.001   0.01   0.003   0.02   Anger   0.006   0.02   0.005   0.02   Arousal   4.4   1.74   3.9   2.16   16.67%   higher  in  CP&C    
  24. 24. cerc.iiitd.ac.in   Emotional Expressions – NegaKve  senKment  higher  in  ciKzen  iniKated   threads   24   Cp&c   Cc   Avg   Std.  dev   Avg   Std.  dev   NA**   0.021   0.03   0.018   0.04   Anx**   0.001   0.01   0.003   0.02   Anger**   0.006   0.02   0.005   0.02   Arousal**   4.4   1.74   3.9   2.16   200%  higher   in  Cc   I  am  just  worried  if  Hyderabad  Traffic  Police  [HTP]   makes  things  worse  like  always  and  create  more   chaos.  Frankly  speaking...  it's  the  lower  income   group  or  the  people  who  are  not  aware  using  high   beams.  Try  to  educate  people  on  road.  
  25. 25. cerc.iiitd.ac.in   Emotional Expressions – NegaKve  senKment  higher  in  ciKzen  iniKated   threads   25   CP&C   CC   Avg   Std.  dev   Avg   Std.  dev   NA**   0.021   0.03   0.018   0.04   Anx**   0.001   0.01   0.003   0.02   Anger**   0.006   0.02   0.005   0.02   Arousal**   4.4   1.74   3.9   2.16   12.82%   higher  in  Cpc   Higher  arousal  and  nega#ve  affect  to  be  markers  of  sensi#sa#on   because  of  crime!  
  26. 26. cerc.iiitd.ac.in   Research Questions – RQ  1:  Topical  Characteris3cs   - Nature  of  content  and  topics  that  characterize   social  media  discussion  threads     – RQ  2:  Engagement  Characteris3cs   - How  do  ciKzens  and  police  engage  in  social  media   discussion  threads?     – RQ  3:  Emo3onal  Exchanges     - Nature  of  emoKons  and  affecKve  expression  that   manifest  on  social  media     – RQ  4:  Cogni3ve  and  Social  Orienta3on   - What  are  the  linguisKc  aRributes  that  characterize   cogniKve  and  social  response  processes?     26  
  27. 27. cerc.iiitd.ac.in   Social and Cognitive Orient. – Discussion  threads  involving  just  the   ciKzens  are  highly  self-­‐aRenKon  focused   27   Likely  ciKzens  mostly  express  their  own  concerns  that  they   face  with  others   CP&C   CC   ppron   0.062   0.059   0.045   0.056   i   0.008   0.017   0.014   0.033   shehe   0.002   0.01   0.003   0.003   they   0.005   0.013   0.008   0.008   75%  More   I  have  lived  in  the  UK  and  all  the  3me  I  have  never  heard   anyone  honking.  Honking  is  not  required  if  you  know  how   to  drive  [...]  Can  anyone  advise  me  where  to  complain  if  I   see  anyone  who  don't  comply  ?  
  28. 28. cerc.iiitd.ac.in   Why it matters? – Understanding  that  helps  police  improve  policing   and  community  sensing   - Facebook  can  be  used  to  record  and  sense  behavioural   aRributes  such  as  engagement,  emoKons,  and  social   support     – Enable  police  and  ciKzen  community  to  enhance   emoKonal  support  to  residents  experiencing   safety  issues   - Discussion  threads  with  police  and  ciKzen  commentary   showed  reduced  levels  of  anxiety,  showing  police   interacKons  can  be  calming  to  ciKzens.   28  
  29. 29. cerc.iiitd.ac.in   Technological Implications – Helping  communiKes  to  make  consensus  based   decisions  regarding  support  and  acKons  they  seek   from  police   – Help  gauge  changing  emoKons  and  behaviour  among   ciKzens   - Timely  and  early  predicKve  analyKcal  systems   – Sense  and  record  the  reacKons  of  ciKzens  and  share   these  records  with  decision  makers   - Take  Kmely  measures  and  gain  beRer  insights   29  
  30. 30. cerc.iiitd.ac.in   Acknowledgement – TCS  research  for  funding  the  project   – Members  of  Cybersecurity  EducaKon  and   Research  Centre  (CERC)  and  Precog  who   have  given  us  conKnued  support  throughout   the  project   – Special  thanks  to  Siddhartha  Asthana   30  
  31. 31. Thank  you!       niharikas@iiitd.ac.in   cerc.iiitd.ac.in  

×