Brown   U niversity  

Cheap  Guanxi  Identification     In  the  Chinese  Web   CSCI  2950U  Course  Project        

Weiyi  Liu   CS  Department

 

September,  2011      

Weiyi  Liu  

2    

Background   Researchers  have  noticed  that  the  Chinese  web  culture  might  differ  from  that  in  the  West.  Hyperlinks  between   Chinese  web  sites  might  reflect  the  underlying  manifestation  of  a  complex  dyadic  social  construct  in  China,  known  as   “guanxi”[1].  Though  the  direct  translation  of  “guanxi”  is  “relationship”,  the  concept  as  it  is  used  and  applied  in  Chinese   culture  is  much  richer.  Here  we  give  a  definition  of  Guanxi:  An  informal,  particularistic  personal  connection  between   two  individuals  who  are  bounded  by  an  implicit  psychological  contract  to  follow  the  social  norm  of  guanxi  such  as   maintaining  a  long  term  relationship,  mutual  commitment,  loyalty,  and  obligation.  [2]   A  quality  guanxi  is  characterized  by  the  mutual  trust  and  feeling  developed  between  the  two  parties  through   numerous  interactions  following  the  self-­‐disclosure,  dynamic  reciprocity,  and  long-­‐term  equity  principles.  A  cheap   guanxi  is  established  for  the  sole  purpose  of  sharing  resource  or  profit.  In  other  words,  cheap  guanxi  is  profit-­‐ oriented.  Once  a  side  in  a  cheap  guanxi  could  not  afford  the  profit  reciprocity,  the  guanxi  will  collapse.  

Problem  Statement   As  applied  to  web,  mutual  links  which  are  established  solely  for  promoting  one’s  own  website,  either  in  the  aspect  of   increasing  traffic  or  PageRank,  might  be  a  form  of  cheap  guanxi.  In  other  words,  there  are  no  requirements  on  the   website  content  or  purpose  when  two  websites  establish  a  cheap  guanxi.1   The  large-­‐scale  existence  of  cheap  guanxi  makes  link  analysis  algorithms  such  as  PageRank  less  effective,  which  might   overstate  the  importance  of  some  websites  who  trade  links  as  resources  with  other  websites.  Our  goal  is  to  identify   cheap  guanxi  websites  in  the  Chinese  web  through  studying  the  correlation  between  link  pattern  and  PageRank.   It  is  worthy  to  mention  that,  such  phenomenon  exists  not  only  in  China.  There  are  mutual  link  trading  websites  found   in  other  languages,  such  as  Mutual  Link  Market  (http://www.mutuallinkmarket.com).  We  select  Chinese  websites  as   our  research  resource,  because  the  existence  of  mutual  links  based  on  cheap  guanxi  is  more  obvious  in  China,   according  to  the  research  results  presented  in  [2].  Further  more,  it  is  typical  that  a  website  has  an  exclusive  part  at  the   bottom  of  its  homepage,  called  “Friendly  Links”  or  “Collaboration  Links”,  which  could  be  used  to  examine  the   rationality  of  our  results.  And  we  believe  that,  by  studying  a  culture  in  which  the  dyadic  relationship  plays  a  critical   role,  we  could  get  a  better  understanding  of  social  networks.  

                                                                                                                                        1  E.g.  from  a  travel  agency  web  page,  we  find  such  a  note:  “Our  website  have  PageRank  score  of  4  ...  we  welcomes   friendly  links.  There  are  no  requirements  on  content  or  purposes  to  link  to  our  page;  however,  the  web  site  has  to  be   legal,  with  no  viruses  ...  Each  web  site’s  PageRank  score  has  to  be  ≥  4.”  

Weiyi  Liu  

3    

Assumptions   Our  work  will  base  on  the  assumption  that  for  websites  establishing  cheap  guanxi,  the  objective  is  to  gain  PageRank  or   traffic.  Based  on  Theorem  1  of  [3],  given  website  i  and  website  j  with  PageRank  P(i)  and  P(j),  P(j)-­‐P(i)  is  correlated  to   the  amount  of  PageRank  website  i  will  gain  or  lose,  if  i  establishes  a  mutual  link  with  node  j.  So  that  websites  generally   prefer  to  establish  cheap  guanxi  with  websites  of  higher  PageRank.  However,  since  neither  web  site  is  willing  to  lose   PageRank  from  the  establishment  of  a  mutual  link,  we  speculate  cheap  guanxi  websites  tend  to  have  similar  PageRank   scores  with  their  partners.   Based  on  these  assumptions,  some  studies  on  Chinese  websites  have  given  an  empirical  threshold,  which  identify   websites  have  at  least  one  of  the  following  features  to  be  cheap  guanxi  websites:   1.

Websites  that  have  >30%  of  their  outgoing  links  reciprocated[2].  

2.

Websites  that  have  >50%  of  the  neighboring  websites  connected  by  mutual  links  with  PageRank  difference  of   one  or  zero  [2].

Proposal   There  will  be  two  critical  parts  in  our  project,  calculating  PageRank  and  counting  mutual  links.  Outgoing  links  without   reciprocations  will  also  be  counted,  while  it  is  trivial  comparing  to  the  two  critical  parts.  As  a  website  may  send  a  link   through  one  page  and  receive  a  link  back  to  another  page,  we  consider  mutual  links  between  websites  rather  than   merely  mutual  links  between  web  pages.  To  make  the  whole  process  scalable,  our  work  will  be  implemented  on   Hadoop.  Since  PageRank  has  already  been  well  developed  with  MapReduce,  we  will  be  focusing  on  counting  mutual   links  with  MapReduce  efficiently.  As  for  now,  we  plan  to  accomplish  these  two  jobs  separately.  Hopefully,  we  could   find  a  way  to  combine  them  together  into  one  job,  after  getting  a  better  understand  of  MapReduce  PageRank   Algorithm.  After  these  two  parts,  we  will  calculate  several  statistical  data,  such  as  the  percentage  of  website’s  outgoing   links  with  reciprocation,  the  PageRank  difference  between  two  mutual  linked  websites,  etc.  Then  we  will  use  the   empirical  threshold  mentioned  in  Assumption  part  to  identify  websites.  

Hypotheses  &  Experiments   Here  we  expect  the  result  of  identification  to  be  rational.  If  it  is  not,  one  plausible  reason  is  we  are  using  an   inappropriate  empirical  threshold.  By  tuning  it,  we  might  get  a  better  result.  There  might  be  some  other  reasons,  such   as  if  the  dataset  is  qualified  or  generalized  enough,  or  if  the  given  assumptions  are  correct.  By  analyzing  the  result,  we   could  find  out  a  rational  explanation.   To  confirm  the  hypotheses,  we  will  examine  it  in  the  following  ways:  

Weiyi  Liu  

4    

1.

As  we  mentioned  above  in  Problem  Statement,  typical  Chinese  cheap  guanxi  website  usually  has  a  key  part  on   one  of  its  webpages,  which  is  called  “Friendly  links”  or  “Collaboration  Partner  Links”.  By  matching  these   textual  hints  with  the  result  we  have,  we  could  figure  out  if  the  result  is  rational.  

2.

Since  a  website  who  establishes  a  cheap  guanxi  is  highly  possible  to  appear  in  other  similar  cheap  guanxi,  we   expect  those  websites  identified  to  be  cheap  guanxi  websites  will  compose  into  different  small  clusters   according  to  their  contents,  user  groups,  locations,  etc.    

3.

If  such  clusters  exist,  we  expect  the  PageRank  of  websites  in  a  cluster  to  be  similar  or  ranging  smoothly.    

Since  it  might  be  impossible  to  prove  (either  mathematically  or  logically)  the  identification  result  made  by  a  certain   threshold  is  correct,  artificial  judgment  will  play  a  critical  role  when  examine  the  rationality  of  the  result.    

Reference   [1]   [2]   [3]  

V.  King,  L.  L.  Yu,  and  Y.  Zhuang,  “Guanxi  in  the  chinese  web-­‐a  study  of  mutual  linking,”  in  Proceeding  of  the  17th   international  conference  on  World  Wide  Web,  2008,  no.  October,  pp.  1161–1162.   V.  King,  L.  Yu,  and  Y.  Zhuang,  “Guanxi  in  the  Chinese  Web,”  2009  International  Conference  on  Computational   Science  and  Engineering,  pp.  9-­‐17,  2009.   K.  Avrachenkov  and  N.  Litvak,  “The  Effect  of  New  Links  on  Google  Pagerank,”  Stochastic  Models,  vol.  22,  no.  2,   pp.  319-­‐331,  Jul.  2006.