Augmented	
  Visualization	
  of	
  Association	
  Rules	
  for	
  
                        Data	
  Mining	
  

               Wilson	
  Castillo-­‐Rojas1,	
  Alexis	
  Peralta1,	
  and	
  Claudio	
  Meneses2	
  
      1Faculty	
  of	
  Engineering	
  and	
  Architecture,	
  Arturo	
  Prat	
  University,	
  Iquique	
  -­‐	
  Chile	
  



      wilson.castillo@unap.cl,alexisperalta@unap.cl
          2Faculty	
  of	
  Engineering	
  and	
  Geological	
  Sciences,	
  North	
  Catholic	
  University,	
  	
  

                                                        Antofagasta	
  -­‐	
  Chile	
  

                                                 cmeneses@ucn.cl


         Abstract.	
   This	
   paper	
   describes	
   a	
   proposal	
   for	
   enhanced	
   visualization	
   of	
   a	
  
         data-­‐mining	
   model	
   generated	
   with	
   Association	
   Rule	
   (AR)	
   techniques	
   by	
  
         applying	
   Self-­‐Organizing	
   Maps	
   (SOM).	
   A	
   representation	
   of	
   visual	
   percep-­‐
         tion	
  model	
  of	
  AR	
  based	
  on	
  a	
  method	
  called	
  AVM-­‐DM	
  (Augmented	
  Visualiza-­‐
         tion	
   Models	
   for	
   Data	
   Mining)	
   is	
   established,	
   together	
   with	
   data	
   and	
   pat-­‐
         terns,	
  which	
  support	
  the	
  visual	
  exploration	
  stage,	
  thus	
  fitting	
  in	
  the	
  context	
  
         of	
  the	
  KDD	
  (Knowledge	
  Discovery	
  in	
  Databases)	
  process.	
  This	
  methodology	
  
         seeks	
  to	
  answer	
  generic	
  user	
  questions	
  regarding	
  the	
  inner	
  workings	
  of	
  the	
  
         model,	
  and	
  to	
  support	
  understanding	
  the	
  generated	
  model.	
  The	
  use	
  of	
  the	
  
         SOM	
  technique	
  as	
  a	
  visual	
  enhancer	
  applied	
  to	
  an	
  AR	
  model,	
  serves	
  a	
  dual	
  
         purpose:	
  to	
  obtain	
  the	
  spatial	
  distribution	
  of	
  the	
  subset	
  of	
  data	
  associated	
  
         with	
  a	
  rule,	
  and	
  to	
  display	
  this	
  subset	
  using	
  a	
  map.	
  	
  The	
  visualization	
  of	
  the	
  
         RA	
  model,	
  proposed	
  in	
  this	
  work,	
  is	
  implemented	
  through	
  a	
  software	
  tool	
  
         giving	
   users	
   different	
   interaction	
   mechanisms.	
   Results	
   of	
   user	
   experiments	
  
         demonstrate	
  the	
  usefulness	
  of	
  the	
  proposed	
  SOM	
  technique	
  in	
  visually	
  en-­‐
         hancing	
  and	
  helping	
  to	
  understand	
  	
  the	
  AR	
  model.	
  

         Keywords:	
   Data	
   mining,	
   visual	
   data	
   mining,	
   visualization	
   of	
   data	
   mining	
  
         models,	
  visualization	
  of	
  association	
  rules.	
  


1        Introduction

In	
  a	
  KDD	
  process,	
  the	
  utility	
  of	
  a	
  Data	
  Mining	
  (DM)	
  model	
  depends	
  mainly	
  
on	
   two	
   factors:	
   the	
   ability	
   of	
   the	
   model	
   to	
   discover	
   interesting	
   patterns	
  
and	
  the	
  ease	
  with	
  which	
  the	
  model	
  structure	
  can	
  be	
  understood	
  and	
  ad-­‐
justed	
  by	
  users.	
  Thus,	
  along	
  with	
  the	
  predictive	
  and	
  descriptive	
  power	
  of	
  
a	
  DM	
  model,	
  its	
  structure	
  should	
  be	
  well	
  understood	
  and	
  interpreted	
  by	
  
the	
  users,	
  because	
  the	
  classification	
  or	
  description	
  of	
  the	
  data	
  without	
  an	
  
explanation	
   model	
   induced	
   from	
   data,	
   can	
   reduce	
   the	
   credibility	
   of	
   the	
  
results	
  of	
  the	
  KDD	
  process	
  [1].	
  In	
  this	
  regard,	
  appropriate	
  visualizations	
  
of	
   DM	
   models	
   can	
   transform	
   them	
   into	
   understandable	
   tools	
   that	
   convert	
  
data	
   into	
   knowledge.	
   In	
   addition,	
   appropriate	
   visualizations	
   of	
   patterns	
  
can	
  facilitate	
  the	
  task	
  of	
  discovering	
  knowledge	
  to	
  interpret	
  and	
  evaluate	
  
these	
  patterns	
  visually	
  [2,	
  3].	
  	
  	
  
This	
   work	
   proposes	
   to	
   visually	
   enhance	
   the	
   DM	
   model	
   generated	
   by	
   an	
  
Association	
  Rule	
  mining	
  (AR)	
  technique,	
  by	
  combining	
  the	
  SOM	
  technique	
  
and	
   creating	
   complementary	
   views	
   of	
   the	
   different	
   rules	
   or	
   model	
   com-­‐
ponents.	
   This	
   method	
   seeks	
   to	
   answer	
   generic	
   user	
   questions	
   regarding	
  
the	
  inner	
  workings	
  of	
  the	
  model.	
  This	
  approach	
  is	
  based	
  on	
  the	
  Augment-­‐
ed	
  Visualization	
  Models	
  for	
  Data	
  Mining	
  (AVM-­‐DM)	
  [4]	
  method	
  that	
  pro-­‐
poses	
  a	
  model	
  of	
  visual	
  perception	
  and	
  user	
  interaction,	
  focusing	
  on	
  the	
  
stage	
  of	
  adjustment	
  or	
  refinement	
  of	
  the	
  DM	
  model	
  generated	
  within	
  the	
  
wider	
  context	
  of	
  the	
  entire	
  KDD	
  process.	
  	
  
The	
  proposed	
  work	
  includes	
  the	
  implementation	
  of	
  part	
  of	
  the	
  AVM-­‐DM	
  
method	
  in	
  a	
  prototype	
  tool	
  that	
  accepts	
  a	
  set	
  of	
  appropriate	
  data	
  and	
  an	
  
AR	
   model.	
   Finally,	
   a	
   subjective	
   evaluation	
   of	
   the	
   prototype	
   is	
   presented	
  
through	
   the	
   user	
   evaluation	
   experiment,	
   consisting	
   of	
   a	
   survey	
   .	
   Partici-­‐
pants	
   provided	
   information	
   about	
   the	
   performance,	
   usability,	
   manage-­‐
ment	
  views	
  and	
  support	
  provided	
  by	
  the	
  developed	
  tool	
  in	
  understanding	
  
a	
  previously	
  generated	
  DM	
  model.	
  


2         Visualization of Association Rules

AR	
   represent	
   the	
   relationships	
   between	
   several	
   variables,	
   i.e.,	
   consider	
  
that	
  AR	
  is	
  an	
  implication	
  of	
  the	
  form	
  X	
   →	
  Y,	
  where	
  X	
  is	
  a	
  set	
  of	
  items	
  ca-­‐
lled	
  antecedents,	
  and	
  Y	
  is	
  the	
  set	
  of	
  consequent	
  items.	
  At	
  least	
  five	
  para-­‐
meters	
  should	
  be	
  considered	
  in	
  the	
  visualization	
  of	
  an	
  AR:	
  the	
  set	
  of	
  ante-­‐
cedent	
   items,	
   consequent	
   items,	
   associations	
   between	
   antecedents	
   and	
  
consequents,	
  the	
  rule’s	
  support,	
  and	
  its	
  confidence	
  [5].	
  	
  
Research	
   on	
   visualization	
   of	
   AR	
   can	
   be	
   categorized	
   into	
   three	
   main	
  
groups,	
   depending	
   on	
   whether	
   they	
   are	
   based	
   on	
   tables,	
   matrices,	
   or	
  
graphs.	
  Tabl-­‐based	
  techniques	
  are	
  the	
  most	
  common	
  and	
  traditional	
  ap-­‐
proach	
   to	
   represent	
   AR.	
   The	
   columns	
   of	
   a	
   table	
   generally	
   represent	
   the	
  
items	
  of	
  the	
  AR	
  model	
  while	
  each	
  row	
  represents	
  a	
  rule.	
  Examples	
  of	
  te-­‐
chniques	
   based	
   on	
   tables	
   can	
   be	
   found	
   in	
   several	
   commercial	
   systems,	
  
including	
   SAS	
   Enterprise	
   Miner	
   and	
   DB	
   Miner	
   [6].	
   Matrix-­‐based	
   tech-­‐
niques	
   such	
   as	
   those	
   implemented	
   in	
   MineSet	
   [7]	
   and	
   InfoVis	
   [8]	
   use	
   a	
  
coordinate	
   axes	
   grid	
   that	
   represents	
   the	
   antecedents	
   and	
   consequents.	
  
The	
  last	
  group	
  consists	
  of	
  the	
  techniques	
  that	
  are	
  based	
  on	
  graphs	
  using	
  
nodes	
   to	
   represent	
   the	
   items	
   and	
   edges	
   to	
   represent	
   the	
   associations	
   of	
  
items	
  in	
  the	
  rules.	
   Some	
  of	
  these	
  techniques	
  have	
  proposed	
  several	
  types	
  
of	
  representations	
  known	
  to	
  study	
  a	
  large	
  set	
  of	
  data,	
  such	
  as	
  hyperbolic	
  
trees	
  [9].	
  	
  
In	
   summary,	
   although	
   these	
   efforts	
   to	
   improve	
   the	
   visualization	
   of	
   ARs	
  
were	
   able	
   to	
   supplement	
   the	
   rule	
   mining	
   with	
   graphics	
   that	
   allow	
   us	
   to	
  
observe	
   each	
   rule	
   in	
   detail,	
   we	
   failed	
   to	
   find	
   visualization	
   tools	
   that	
   allow	
  
an	
  interaction	
  with	
  each	
  rule,	
  while	
  also	
  visualizing	
  how	
  the	
  data	
  in	
  each	
  
rule	
  are	
  spatially	
  distributed.	
  A	
  comparative	
  review	
  of	
  visualization	
  tools	
  
for	
   DM	
   (including	
   AR)	
   techniques,	
   by	
   Castillo	
   [5],	
   concluded	
   that:	
   a)	
   most	
  
research	
   recommends	
   using	
   a	
   combination	
   of	
   DM	
   techniques	
   with	
   ap-­‐
propriate	
   views,	
   b)	
   it	
   is	
   essential	
   to	
   consider	
   in	
   the	
   design	
   of	
   views,	
   the	
  
mechanisms	
   for	
   user	
   interaction,	
   and	
   c)	
   the	
   role	
   of	
   visualization	
   in	
   the	
  
KDD	
  process	
  must	
  be	
  extended	
  in	
  all	
  its	
  stages.	
  


3         The AVM-DM Scheme

The	
   AVM-­‐DM	
   scheme	
   proposed	
   by	
   Castillo	
   in	
   [4]	
   considers	
   the	
   character-­‐
istics	
   of	
   the	
   analyzed	
   models	
   of	
   perception,	
   and	
   includes	
   the	
   most	
   rele-­‐
vant	
   aspects	
   of	
   each,	
   particularly	
   with	
   regard	
   to	
   the	
   integration	
   of	
   the	
  
display	
   in	
   step	
   adjustment	
   or	
   refinement	
   and	
   evaluation	
   of	
   DM	
   models.	
  
AVM-­‐DM	
  brings	
  the	
  concept	
  of	
  “Augmented	
  Visualization”	
  for	
  DM	
  models,	
  
and	
  suggests	
  that,	
  given	
  a	
  DM	
  technique	
  to	
  be	
  visualized,	
  called	
  Primary	
  
DM	
  Technique	
  (PT-­‐DM),	
  should	
  allow	
  the	
  user	
  to	
  incorporate	
  in	
  this	
  dis-­‐
play,	
  different	
  visuals	
  regarding	
  the	
  type	
  of	
  model	
  and	
  data	
  domain,	
  and	
  
in	
  turn	
  need	
  to	
  apply	
  another	
  DM	
  technique,	
  called	
  Secondary	
  DM	
  Tech-­‐
nical	
  (ST-­‐DM),	
  as	
  a	
  visual	
  enhancer	
  that	
  allows	
  exploring	
  the	
  PT-­‐DM.	
  The	
  
selected	
   ST-­‐DM	
   technique	
   must	
   meet	
   the	
   requirements	
   of	
   being	
   a	
   de-­‐
scriptive	
  DM	
  technique	
  that	
  is	
  appropriate	
  to	
  the	
  domain	
  data	
  being	
  ana-­‐
lyzed	
  within	
  the	
  PT-­‐DM.	
  	
  


4         Augmented Visualization of AR Model using SOM

In	
  the	
  case	
  of	
  the	
  AR	
  techniques,	
  several	
  visualization	
  methods	
  analyzed	
  
in	
  [5]	
  propose	
  a	
  static	
  display,	
  without	
  any	
  possibility	
  for	
  the	
  user	
  to	
  in-­‐
teract	
  with	
  each	
  rule.	
  Most	
  DM	
  visualization	
  tools	
  delivered	
  an	
  overview	
  
of	
  the	
  ARs	
  but	
  cannot	
  combine	
  DM	
  techniques	
  to	
  provide	
  information	
  on	
  
model	
   rules	
   and	
   instances	
   supporting	
   each	
  rule,	
  and	
   only	
  a	
   few	
   tools	
   pro-­‐
vide	
   interaction	
   mechanisms	
   for	
   the	
   user.	
   The	
   proposed	
   use	
   of	
   the	
   SOM	
  
technique	
  as	
  an	
  AR	
  visual	
  augmenter,	
  serves	
  a	
  dual	
  purpose:	
  to	
  obtain	
  the	
  
spatial	
   distribution	
   of	
   the	
   subset	
   of	
   data	
   associated	
   with	
   each	
   rule,	
   and	
   to	
  
display	
  this	
  partition	
  using	
  a	
  map.	
  
                                                                                                                                 	
  
                             Fig.	
  1.	
  Main	
  interface	
  of	
  the	
  prototype	
  software.	
  

The	
   prototype	
   implements	
   the	
   AVM-­‐MD	
   scheme	
   for	
   hierarchical	
   struc-­‐
ture	
  techniques	
  in	
  DM	
  (decision	
  tree	
  &	
  AR),	
  and	
  in	
  this	
  paper,	
  we	
  concen-­‐
trate	
  on	
  the	
  AR	
  mining	
  technique.	
  It	
  incorporates	
  a	
  set	
  of	
  visual	
  elements;	
  
data	
   table,	
   pie	
   chart	
   (by	
   rule	
   and	
   general),	
   dot	
   plot,	
   and	
   parallel	
   coordi-­‐
nates	
  plot.	
  Also,	
  available	
  interaction	
  mechanisms	
  include	
  zoom,	
  selection	
  
rules,	
   and	
   setting	
   of	
   the	
   parameters.	
   Figure	
   1	
   shows	
   the	
   main	
   interface	
   of	
  
the	
  experimental	
  prototype,	
  where	
  ARs	
  are	
  displayed	
  in	
  the	
  central	
  part,	
  
together	
   with	
   complementary	
   views	
   and	
   visual	
   elements	
   on	
   the	
   right	
  
side.	
   In	
   this	
   tool	
   all	
   architecture	
   components	
   of	
   the	
   proposed	
   AVM-­‐DM	
  
scheme	
  are	
  implemented.	
  The	
  user	
  can	
  maximize	
  the	
  image	
  located	
  in	
  the	
  
c)	
  section	
  of	
  the	
  interface	
  by	
  clicking	
  with	
  the	
  mouse,	
  opening	
  a	
  window	
  
that	
   presents	
   a	
   detailed	
   view	
   of	
   this	
   technique.	
   They	
   can	
   re-­‐configure	
  
their	
  initial	
  parameters	
  on	
  a	
  selected	
  rule	
  and	
  apply	
  the	
  SOM	
  technique.	
  
Also,	
  the	
  user	
  can	
  see	
  the	
  shape	
  of	
  the	
  distribution	
  of	
  the	
  instances	
  cov-­‐
ered	
  by	
  this	
  rule.	
  
5           Controlled Experiment: Evaluations & Analysis

The	
   following	
   controlled	
   experiment	
   provides	
   a	
   comparison	
   and	
   subjec-­‐
tive	
  evaluation	
  of	
  the	
  visualization	
  of	
  ARs	
  obtained	
  through	
  a	
  DM	
  task	
  to	
  
be	
   performed	
   by	
   a	
   set	
   of	
   users,	
   whose	
   aim	
   is	
   to	
   check	
   if	
   the	
   SOM-­‐based	
  
visualization	
   enhanced	
   AR	
   mining	
   along	
   with	
   the	
   set	
   of	
   visual	
   elements	
  
provided	
   by	
   the	
   prototype	
   software,	
   can	
   improve	
   the	
   understanding	
   of	
  
the	
   model,	
   such	
   as	
   looking	
   at	
   the	
   distribution	
   of	
   data	
   in	
   each	
   rule,	
   com-­‐
pared	
  with	
  the	
  visualization	
  provided	
  by	
  another	
  DM	
  tool,	
  that	
  does	
  not	
  
have	
  this	
  focus	
  or	
  visualization	
  scheme.	
  This	
  experiment	
  was	
  conducted	
  
with	
  17	
  users	
  of	
  varying	
  levels	
  of	
  expertise	
  in	
  DM	
  processes,	
  and	
  the	
  use	
  
of	
  DM	
  tools.	
  We	
  asked	
  participants	
  to	
  perform	
  a	
  generic	
  task	
  description	
  
and	
  could	
  answer	
  questions	
  about	
  the	
  model	
  and	
  its	
  components,	
  and	
  to	
  
relate	
  the	
  model	
  to	
  the	
  characteristics	
  of	
  the	
  data	
  from	
  which	
  the	
  model	
  
was	
  generated.	
  	
  




                                                                                                                                                            	
  
Fig.	
   2.	
   a)	
   Level	
   of	
   acceptance	
   of	
   views	
   available	
   to	
   describe	
   the	
   AR	
   model.	
   b)	
   Ability	
   to	
  
describe	
  the	
  data	
  on	
  the	
  AR	
  model	
  using	
  the	
  SOM	
  technique.	
  

Subsequently,	
   once	
   the	
   DM	
   task	
   was	
   prepared	
   for	
   this	
   experiment,	
   the	
  
users	
  had	
  to	
  answer	
  a	
  survey	
  designed	
  to	
  gather	
  the	
  subjective	
  opinion	
  of	
  
the	
   group,	
   regarding	
   the	
   performance	
   of	
   both	
   tools,	
   the	
   visualization	
   of	
  
the	
  generated	
  AR	
  model,	
  usability,	
  utility	
  of	
  visual	
  elements,	
  the	
  desirabil-­‐
ity	
   of	
   combining	
   the	
   SOM	
   technique	
   to	
   achieve	
   a	
   visually	
   augmented	
  
model,	
   and	
   the	
   efficiency	
   in	
   understanding	
   of	
   the	
   model.	
   Users	
   mostly	
  
stated	
  that	
  both	
  the	
  combination	
  of	
  the	
  SOM	
  technique	
  applied	
  to	
  the	
  AR	
  
model,	
  and	
  the	
  use	
  of	
  graphic	
  elements	
  on	
  the	
  data	
  rules,	
  allowed	
  them	
  to	
  
improve	
   their	
   understanding	
   of	
   the	
   generated	
   AR	
   model,	
   achieving	
   a	
  
score	
   distribution	
   of	
   54,	
   9%	
   good	
   and	
   33.33%	
   very	
   good,	
   which	
   can	
   be	
  
seen	
   from	
   the	
   graph	
   in	
   Figure	
   2	
   a).	
   Also,	
   as	
   shown	
   in	
   Figure	
   2	
   b),	
   users	
  
expressed	
   mostly	
   positive	
   ability	
   (76.5%	
   high	
   and	
   11.8%	
   very	
   high)	
   to	
  
obtain	
  an	
  augmented	
  visualization	
  of	
  the	
  AR	
  model.	
  


6         Conclusions and Future Work

The	
   preliminary	
   results	
   of	
   the	
   presented	
   study	
   allow	
   us	
   to	
   confirm	
   the	
  
suitability	
  and	
  utility	
  of	
  combining	
  the	
  AR	
  mining	
  technique	
  with	
  the	
  SOM	
  
technique	
   for	
   achieving	
   augmented	
   visualization	
   for	
   the	
   AR	
   model,	
   and	
  
for	
   visualizing	
   the	
   spatial	
   distribution	
   of	
   the	
   data	
   covered	
   by	
   each	
   rule,	
  
thus	
   helping	
   improve	
   the	
   understanding	
   of	
   their	
   inner	
   workings.	
   Also	
   the	
  
visual	
  tools	
  provided	
  in	
  the	
  prototype	
  software	
  support	
  the	
  analysis	
  and	
  
examination	
   of	
   the	
   AR	
   model.	
   As	
   future	
   work,	
   we	
   are	
   evaluating	
   other	
  
descriptive	
   DM	
   techniques	
   that	
   can	
   provide	
   alternative	
   views	
   for	
   visually	
  
enhanced	
  AR	
  models.	
  


7         Acknowledgements

We	
  thank	
  the	
  anonymous	
  reviewers	
  for	
  their	
  helpful	
  suggestions.	
  In	
  par-­‐
ticular,	
   we	
   thank	
   all	
   the	
   effort	
   in	
   the	
   editing	
   phase	
   which	
   substantially	
  
improved	
  the	
  readability	
  of	
  the	
  paper.	
  


8         References
 1. Keim,	
   D.A.,	
   (1997).	
   Visual	
   Techniques	
   for	
   Exploring	
   Databases.	
   Third	
   International	
  
    Conference	
  on	
  KDD	
  &	
  Data	
  Mining.	
  Newport	
  Beach,	
  CA,	
  August.	
  
 2. Meneses,	
  C.	
  J.	
  &	
  Grinstein,	
  G.	
  G.,	
  (2001).	
  Visualization	
  for	
  Enhancing	
  the	
  Data	
  Mining	
  
    Process.	
   In	
   Proceedings	
   of	
   the	
   Data	
   Mining	
   &	
   KDD:	
   Theory,	
   Tools,	
   and	
   Technology.	
   III	
  
    Conference.	
  Orlando-­‐FL,	
  April.	
  	
  
 3. Thearling,	
  K.,	
  Becker,	
  B.,	
  Mawby,	
  B.,	
  Pilote,	
  M.,	
  Sommerfield,	
  D.	
  (1998).	
  Visualizing	
  Da-­‐
    ta	
  Mining	
  Models.	
  In	
  Proceedings	
  of	
  the	
  Integration	
  of	
  Data	
  Mining	
  and	
  Data	
  Visuali-­‐
    zation	
  Workshop,	
  Springer-­‐Verlag.	
  
 4. Castillo-­‐Rojas,	
  W.,	
  Meneses,	
  C.,	
  &	
  Medina,	
  F.	
  (2013).	
  Augmented	
  Decision	
  Tree	
  Models	
  
    Using	
   SOM.	
   6th	
   Latin	
   American	
   Conference	
   on	
   Human	
   Computer	
   Interaction,	
   Costa	
  
    Rica.	
  Proceedings	
  pp.	
  148-­‐155.	
  Springer	
  LNCS	
  8278,	
  ISBN	
  978-­‐3-­‐319-­‐03067-­‐8.	
  
 5. Castillo-­‐Rojas,	
  W.,	
  &	
  Meneses,	
  C.	
  (2012)	
  Comparative	
  Review	
  of	
  Schemes	
  of	
  Multidi-­‐
    mensional	
  Visualization	
  for	
  Data	
  Mining	
  Techniques.	
  III	
  International	
  Congress	
  of	
  In-­‐
    formatics.	
  August,	
  Arica–Chile.	
  
 6. Han,	
  J.,	
  Kamber,	
  M.	
  (2001).	
  Data	
  Mining	
  Concepts	
  and	
  Techniques.	
  Morgan	
  Kaufmann.	
  
 7. Brunk,	
  C.,	
  Kelly,	
  J.	
  and	
  Kohavi,	
  R.	
  (1997).	
  MineSet:	
  An	
  Integrated	
  System	
  for	
  Data	
  Min-­‐
    ing.	
  Proc.	
  of	
  Third	
  Intel:	
  Knowledge	
  Discovery	
  and	
  Data	
  Mining,	
  pages	
  135-­‐138.	
  
 8. Wong,	
  P.C.,	
  Whitney,	
  P.,	
  (1999).	
  Visualizing	
  association	
  rules	
  for	
  text	
  mining.	
  INFOVIS.	
  
    Pages	
  120–123.	
  
 9. Lamping,	
   J.,	
   Rao,	
   R.,	
   and	
   P.	
   Pirolli.	
   (1995).	
   A	
   focus+context	
   technique	
   based	
   on	
   hyper-­‐
    bolic	
   geometry	
   for	
   visualizing	
   large	
   hierarchies.	
   In	
   Proceedings	
   of	
   the	
   ACM	
   confer-­‐
    ence	
  on	
  Human	
  Factors	
  in	
  Computing	
  Systems,	
  ACM	
  Press.	
  USA,	
  pages	
  401–408.