SlideShare a Scribd company logo
1 of 31
Download to read offline
Beyond	
  Sen)ment	
  Hype:	
  
Conversa)on	
  Context	
  for	
  Accurate	
  
          Discovery	
  


              Hadley	
  Reynolds	
  
                NextEra	
  Research	
  
Agenda	
  
•  Where	
  we	
  are	
  now	
  –	
  market	
  drivers	
  &	
  
   technology	
  dynamics	
  
•  The	
  Sen)ment	
  Bubble	
  considered	
  
•  Differen)a)ng	
  levels	
  of	
  analysis	
  
•  Prac)cal	
  dimensions	
  of	
  analysis	
  and	
  examples	
  
•  Discussion	
  
Market	
  Drivers	
  for	
  Sen)ment	
  Analysis	
  
Market	
  Drivers	
  for	
  Sen)ment	
  Analysis	
  
Market	
  Drivers	
  for	
  Sen)ment	
  Analysis	
  
Market	
  Drivers	
  for	
  Sen)ment	
  Analysis	
  
Market	
  Drivers	
  for	
  Sen)ment	
  Analysis	
  


          Addi$onal	
  Web	
  2.0	
  Content:	
  
                             Blogs	
  
                  Discussion	
  Forums	
  
      Amazon	
  (Yelp,	
  Trip	
  Advisor	
  etc.)	
  Reviews	
  
          User	
  Generated	
  RaAngs	
  Data	
  
                    “Like”	
  Google+	
  
             And	
  more,	
  much	
  more…	
  
Sen)ment	
  Technology	
  Providers	
  
45	
  
40	
  
35	
  
30	
  
25	
  
20	
  
15	
  
          Corpora Software
10	
  
  5	
  
  0	
  
   2003	
      2004	
     2005	
     2006	
     2007	
     2008	
     2009	
     2010	
     2011	
  
Where	
  Does	
  Sen)ment	
  Belong?	
  
Early	
  Social	
  Monitoring	
  
Naïve	
  Sen)ment	
  
Credibility:	
  comes	
  from	
  accuracy	
  &	
  insight	
  

Ambiguity:	
  is	
  the	
  enemy	
  of	
  accuracy	
  
Challenges	
  for	
  Sen)ment	
  Analysis	
  
•  Level	
  of	
  analysis	
  
•  Timeframes	
  for	
  analysis	
  
•  Rela)ve	
  sophis)ca)on	
  of	
  analysis	
  
Level	
  of	
  Analysis	
  
•  Corpus	
  (Do	
  the	
  bloggers	
  like	
  us?)	
  
•  Document	
  (Does	
  this	
  author	
  like	
  us?)	
  
Document	
  Sen)ment	
  Math	
  
                          Posi)ve	
  document	
  =	
  4	
  points	
  or	
  above	
  
                          Nega)ve	
  document	
  =	
  -­‐2	
  points	
  or	
  below	
  
                          Neutral	
  document	
  =	
  -­‐2	
  through	
  +3	
  
   good	
  
                                                             Value	
         Score	
  
   great	
  
                           Term	
  
                           good	
                              2	
              2	
  
   o.k.	
                  great	
                             3	
              3	
  
                           o.k.	
                              1	
              1	
  
                           disappointed	
                     -­‐4	
           -­‐4	
  
                                              Total:	
                        +2	
  

disappointed	
  




Neutral	
  Document	
  
Document	
  Sen)ment	
  Math	
  
                                      Posi)ve	
  document	
  =	
  4	
  points	
  or	
  above	
  
                                      Nega)ve	
  document	
  =	
  -­‐2	
  points	
  or	
  below	
  
Product	
  A	
     Product	
  B	
     Neutral	
  document	
  =	
  -­‐2	
  through	
  +3	
  
   good	
             ok	
  
                     good	
                                             Value	
         Score	
  
    great	
  
                                      Term	
  
                       ok	
  
                                      Product	
  A	
  good	
              2	
             2	
  
    o.k.	
                            Product	
  A	
  great	
             3	
             3	
  
                                      Product	
  A	
  o.k.	
              1	
             1	
  
                                      Product	
  A	
                     -­‐4	
          -­‐4	
  
                                      disappointed	
  
                                      Product	
  B	
  good	
              1	
             1	
  
disappointed	
  
                                      Product	
  B	
  ok	
                1	
             2	
  
                   disappointed	
  
                                      Product	
  B	
                     -­‐4	
          -­‐8	
  
                   disappointed	
  
                                      disappointed	
  
Nega)ve	
  Document	
                 	
  
                                                           Total:	
                      -­‐3	
  
Level	
  of	
  Analysis	
  
•    Corpus	
  (Do	
  the	
  bloggers	
  like	
  us?)	
  
•    Document	
  (Does	
  this	
  author	
  like	
  us?)	
  
•    Sentence	
  (What	
  is	
  this	
  person’s	
  comment?)	
  
•    En)ty/A`ribute	
  (What	
  is	
  it	
  about	
  us	
  that	
  she	
  
     likes	
  or	
  doesn’t	
  like?)	
  
En)ty-­‐level	
  Analysis	
  
                                                                                    Sources	
  


                      Person	
      Opinion	
   Target	
  En)ty	
  
                                                (Feature)	
  
(Profile)	
   Person	
   (Emo)on)	
  Opinion	
   (Feature)	
   Target	
  En)ty	
  
                                                (Feature)	
  

      (Social	
  Network)	
  
Timeframes	
  of	
  Analysis	
  
•  Retrospec)ve	
  analy)cs/business	
  intelligence	
  
•  Predic)ve	
  analy)cs	
  –	
  quality	
  issues,	
  future	
  
   performance	
  
•  Trend	
  emergence	
  
•  Real-­‐)me	
  –	
  customer	
  interac)ons,	
  social	
  
   interac)ons/engagements	
  	
  
Sophis)ca)on	
  of	
  Analysis	
  
•  Keyword-­‐based	
  sen)ment	
  techniques	
  
   –  Sen)ment	
  terms:	
  elusive,	
  ambiguous,	
  in	
  flux	
  
   –  Sen)ment	
  lexicons:	
  incomplete,	
  non-­‐specific,	
  
      inflexible	
  
   –  Unable	
  to	
  understand	
  context	
  surrounding	
  an	
  
      expression	
  or	
  the	
  people	
  contribu)ng	
  
   –  Unable	
  to	
  understand	
  connec)ons	
  among	
  related	
  
      en))es	
  and	
  a`ributes	
  and	
  people	
  
   –  Unable	
  to	
  gauge	
  quality	
  of	
  source	
  materials	
  
Sophis)ca)on	
  of	
  Analysis	
  
•  Seman)c-­‐based	
  sen)ment	
  techniques	
  
   –  Sen)ment	
  terms	
  >>	
  incorporate	
  related	
  expressions,	
  
      fuzzy	
  logic	
  -­‐	
  NLP	
  
   –  Sen)ment	
  lexicons	
  >>	
  domain	
  ontologies	
  (available	
  or	
  
      buildable)	
  provide	
  analy)cal	
  context	
  
   –  Able	
  to	
  understand	
  context	
  surrounding	
  an	
  expression	
  
      or	
  the	
  people	
  contribu)ng	
  -­‐	
  machine	
  learning	
  &	
  other	
  
      techniques	
  
   –  Able	
  to	
  understand	
  connec)ons	
  among	
  related	
  
      en))es	
  and	
  a`ributes	
  and	
  people	
  -­‐	
  triples,	
  event	
  
      extrac)on	
  
Dimensions	
  of	
  Analysis	
  
•  Ontologies	
  around	
  opinion	
  objects	
  
•  Iden)fica)on	
  and	
  qualifica)on	
  of	
  en))es	
  &	
  
    a`ributes	
  &	
  rela)onships	
  
•  Emo)onal	
  content	
  of	
  expression(s)	
  
•  Quality	
  gauge	
  of	
  sources	
  
•  Profiles	
  of	
  individual	
  commenters	
  
•  Roles/interac)ons/sociology	
  of	
  commenters	
  and	
  
    their	
  affilia)ons	
  
•  Timeframe	
  for	
  expressions	
  and	
  responses	
  
Beyond	
  +/-­‐:	
  
Ontology-­‐based	
  analy)cs	
  

Same	
  Ontology	
  breakdown	
  
Same	
  Scale:	
  Expressed	
  Opinions	
  




Higher	
  values	
  for	
  cardiovascular	
  
diseases	
  with	
  Avas)n	
  



                                   Source:	
  BuzzStory	
  
Opinion::Emo)on	
  	
  




                          Source:	
  BuzzStory	
  
Quality	
  of	
  Content	
  Sources	
  
topix.com	
                                     cancergrace.org	
  
•  Quality:	
  4.48	
                           •  Quality:	
  16.78	
  
"I	
  know	
  of	
  one	
  method	
  that	
     "As	
  shown	
  above,	
  a	
  total	
  of	
  
would	
  be	
  really	
  scary	
  and	
         362	
  pa)ents	
  who	
  hadn't	
  
graphic	
  that	
  would	
  work	
              progressed	
  aser	
  first	
  line	
  
towards	
  gepng	
  people	
  to	
              chemo/Avas)n	
  were	
  
stop	
  pollu)ng	
  my	
  sea	
  breeze	
       randomized	
  to	
  either	
  of	
  the	
  
environment.	
                                  two	
  maintenance	
  therapy	
  
What	
  I	
  wonna	
  know	
  is	
  they	
      arms,	
  and	
  the	
  combina)on	
  
keep	
  pupng	
  down	
  smokers	
              arm	
  showed	
  a	
  significantly	
  
and	
  blaming	
  us	
  for	
                   longer	
  progression-­‐free	
  
evrything.”	
                                   survival	
  (PFS)	
  coun)ng	
  from	
  
                                                the	
  beginning	
  of	
  all	
  
                                                treatment,	
  at	
  10.”	
  
Affilia)on	
  Network	
  –	
  Map	
  of	
  
                     Affilia)ons	
  of	
  People	
  &	
  Topics	
  
                                                                            Supplements	
  



Tobacco	
  Addic)on	
  
                                                                                                             Prostate	
  Cancer	
  

                                                                                                       Breast	
  Cancer	
  




                                                                                              Co-­‐Morbidi)es	
  




                                                                                                          Thyroid	
  Disease	
  

                              Biomarkers	
             Lung	
  Cancer	
  
                          Targeted	
  Therapies	
     Chemotherapy	
  




                                                       H&N	
  Cancer	
  
                                                                                                             Source:	
  BuzzStory	
  
Sociology	
  of	
  Affilia)ons	
  &	
  Topic	
  
                          Groupings	
  
                                               Co-­‐Morbidi)es	
  
Tobacco	
  Addic)on	
  



                                                               Other	
  	
  Types	
  of	
  
                                                                  Cancer	
  



    Supplements	
  




 Misc.	
  Side-­‐Effects	
  




                                              Misc.	
  Side-­‐Effects	
  
        Biomarkers	
  



                                                                                              Source:	
  BuzzStory	
  
Where	
  Does	
  Sen)ment	
  Belong?	
  


 Contextual	
  
 Analy)cs	
  




                  Keyword	
  
                  technology	
  
Challenges	
  Remain	
  

“The	
  service	
  at	
  Reynards	
  is,	
  in	
  general,	
  friendly	
  and	
  
loose.	
  Though	
  they	
  couldn’t	
  find	
  a	
  reserva)on	
  for	
  
four	
  one	
  Friday	
  night,	
  they	
  compensated	
  with	
  so	
  
much	
  warmth	
  and	
  comped	
  wine	
  that	
  all	
  was	
  
forgiven.	
  In	
  some	
  ways,	
  Reynards	
  offers	
  what	
  one	
  
wishes	
  a	
  dining	
  experience	
  in	
  Manha`an	
  would	
  be:	
  
kindness	
  instead	
  of	
  aptude,	
  inoffensive	
  prices,	
  
glorious	
  food,	
  and	
  aesthe)c	
  variety—the	
  clientele	
  is	
  
split	
  roughly	
  in	
  half	
  between	
  the	
  stylish	
  and	
  the	
  
schlumpy.”	
  
	
  
The	
  New	
  Yorker,	
  September	
  24,	
  2012	
  
Resources	
  
•  Bing	
  Liu,	
  Sen$ment	
  Analysis	
  and	
  Opinion	
  
   Mining,	
  Morgan	
  &	
  Claypool,	
  2012	
  
•  Bo	
  Pang	
  and	
  Lillian	
  Lee,	
  Opinion	
  Mining	
  and	
  
   Sen$ment	
  Analysis,	
  (Founda$ons	
  and	
  Trends	
  
   in	
  Informa$on	
  Retrieval),	
  Now	
  Publishers,	
  
   2008	
  	
  
•  Sen)ment	
  Analysis	
  Symposium,	
  San	
  Francisco,	
  
   CA,	
  October	
  30,	
  2012	
  
Ques)ons? 	
  	
  


hadleyr@nexteraresearch.com	
  

More Related Content

Similar to Beyond Sentiment Hype: Conversation Context for Accurate Discovery

Search as Communication: Lessons from a Personal Journey
Search as Communication: Lessons from a Personal JourneySearch as Communication: Lessons from a Personal Journey
Search as Communication: Lessons from a Personal Journey
Daniel Tunkelang
 
Social text sentiment and tone analysis [aai 201] - (4160)
Social text sentiment and tone analysis [aai 201] - (4160)Social text sentiment and tone analysis [aai 201] - (4160)
Social text sentiment and tone analysis [aai 201] - (4160)
Ruben Pertusa Lopez
 
Jillian ms defense-4-14-14-ja
Jillian ms defense-4-14-14-jaJillian ms defense-4-14-14-ja
Jillian ms defense-4-14-14-ja
Jillian Aurisano
 
The future of scientific information & communication
The future of scientific information & communicationThe future of scientific information & communication

Similar to Beyond Sentiment Hype: Conversation Context for Accurate Discovery (15)

Frontiers of Computational Journalism week 1 - Introduction and High Dimensio...
Frontiers of Computational Journalism week 1 - Introduction and High Dimensio...Frontiers of Computational Journalism week 1 - Introduction and High Dimensio...
Frontiers of Computational Journalism week 1 - Introduction and High Dimensio...
 
Private Distributed Collaborative Filtering
Private Distributed Collaborative FilteringPrivate Distributed Collaborative Filtering
Private Distributed Collaborative Filtering
 
Improving search with neural ranking methods
Improving search with neural ranking methodsImproving search with neural ranking methods
Improving search with neural ranking methods
 
Search as Communication: Lessons from a Personal Journey
Search as Communication: Lessons from a Personal JourneySearch as Communication: Lessons from a Personal Journey
Search as Communication: Lessons from a Personal Journey
 
Reshaping Scientific Knowledge Dissemination and Evaluation in the Age of the...
Reshaping Scientific Knowledge Dissemination and Evaluation in the Age of the...Reshaping Scientific Knowledge Dissemination and Evaluation in the Age of the...
Reshaping Scientific Knowledge Dissemination and Evaluation in the Age of the...
 
Collective Opinion Spam Detection Bridging Review Networks and Metadata
Collective Opinion Spam Detection Bridging Review Networks and MetadataCollective Opinion Spam Detection Bridging Review Networks and Metadata
Collective Opinion Spam Detection Bridging Review Networks and Metadata
 
Validation and mechanism: exploring the limits of evaluation
Validation and mechanism: exploring the limits of evaluationValidation and mechanism: exploring the limits of evaluation
Validation and mechanism: exploring the limits of evaluation
 
Social text sentiment and tone analysis [aai 201] - (4160)
Social text sentiment and tone analysis [aai 201] - (4160)Social text sentiment and tone analysis [aai 201] - (4160)
Social text sentiment and tone analysis [aai 201] - (4160)
 
Jillian ms defense-4-14-14-ja
Jillian ms defense-4-14-14-jaJillian ms defense-4-14-14-ja
Jillian ms defense-4-14-14-ja
 
SociologyExchange.co.uk Shared Resource
SociologyExchange.co.uk Shared ResourceSociologyExchange.co.uk Shared Resource
SociologyExchange.co.uk Shared Resource
 
Millburn - Flybase community curation
Millburn - Flybase community curationMillburn - Flybase community curation
Millburn - Flybase community curation
 
Methodology and IRB/URR
Methodology and IRB/URRMethodology and IRB/URR
Methodology and IRB/URR
 
Analytical Design in Applied Marketing Research
Analytical Design in Applied Marketing ResearchAnalytical Design in Applied Marketing Research
Analytical Design in Applied Marketing Research
 
The future of scientific information & communication
The future of scientific information & communicationThe future of scientific information & communication
The future of scientific information & communication
 
Improving Your Literature Reviews with NVivo 10 for Windows
Improving Your Literature Reviews with NVivo 10 for WindowsImproving Your Literature Reviews with NVivo 10 for Windows
Improving Your Literature Reviews with NVivo 10 for Windows
 

Beyond Sentiment Hype: Conversation Context for Accurate Discovery

  • 1. Beyond  Sen)ment  Hype:   Conversa)on  Context  for  Accurate   Discovery   Hadley  Reynolds   NextEra  Research  
  • 2. Agenda   •  Where  we  are  now  –  market  drivers  &   technology  dynamics   •  The  Sen)ment  Bubble  considered   •  Differen)a)ng  levels  of  analysis   •  Prac)cal  dimensions  of  analysis  and  examples   •  Discussion  
  • 3. Market  Drivers  for  Sen)ment  Analysis  
  • 4. Market  Drivers  for  Sen)ment  Analysis  
  • 5. Market  Drivers  for  Sen)ment  Analysis  
  • 6. Market  Drivers  for  Sen)ment  Analysis  
  • 7. Market  Drivers  for  Sen)ment  Analysis   Addi$onal  Web  2.0  Content:   Blogs   Discussion  Forums   Amazon  (Yelp,  Trip  Advisor  etc.)  Reviews   User  Generated  RaAngs  Data   “Like”  Google+   And  more,  much  more…  
  • 8. Sen)ment  Technology  Providers   45   40   35   30   25   20   15   Corpora Software 10   5   0   2003   2004   2005   2006   2007   2008   2009   2010   2011  
  • 9. Where  Does  Sen)ment  Belong?  
  • 12. Credibility:  comes  from  accuracy  &  insight   Ambiguity:  is  the  enemy  of  accuracy  
  • 13. Challenges  for  Sen)ment  Analysis   •  Level  of  analysis   •  Timeframes  for  analysis   •  Rela)ve  sophis)ca)on  of  analysis  
  • 14. Level  of  Analysis   •  Corpus  (Do  the  bloggers  like  us?)   •  Document  (Does  this  author  like  us?)  
  • 15. Document  Sen)ment  Math   Posi)ve  document  =  4  points  or  above   Nega)ve  document  =  -­‐2  points  or  below   Neutral  document  =  -­‐2  through  +3   good   Value   Score   great   Term   good   2   2   o.k.   great   3   3   o.k.   1   1   disappointed   -­‐4   -­‐4   Total:   +2   disappointed   Neutral  Document  
  • 16. Document  Sen)ment  Math   Posi)ve  document  =  4  points  or  above   Nega)ve  document  =  -­‐2  points  or  below   Product  A   Product  B   Neutral  document  =  -­‐2  through  +3   good   ok   good   Value   Score   great   Term   ok   Product  A  good   2   2   o.k.   Product  A  great   3   3   Product  A  o.k.   1   1   Product  A   -­‐4   -­‐4   disappointed   Product  B  good   1   1   disappointed   Product  B  ok   1   2   disappointed   Product  B   -­‐4   -­‐8   disappointed   disappointed   Nega)ve  Document     Total:   -­‐3  
  • 17. Level  of  Analysis   •  Corpus  (Do  the  bloggers  like  us?)   •  Document  (Does  this  author  like  us?)   •  Sentence  (What  is  this  person’s  comment?)   •  En)ty/A`ribute  (What  is  it  about  us  that  she   likes  or  doesn’t  like?)  
  • 18. En)ty-­‐level  Analysis   Sources   Person   Opinion   Target  En)ty   (Feature)   (Profile)   Person   (Emo)on)  Opinion   (Feature)   Target  En)ty   (Feature)   (Social  Network)  
  • 19. Timeframes  of  Analysis   •  Retrospec)ve  analy)cs/business  intelligence   •  Predic)ve  analy)cs  –  quality  issues,  future   performance   •  Trend  emergence   •  Real-­‐)me  –  customer  interac)ons,  social   interac)ons/engagements    
  • 20. Sophis)ca)on  of  Analysis   •  Keyword-­‐based  sen)ment  techniques   –  Sen)ment  terms:  elusive,  ambiguous,  in  flux   –  Sen)ment  lexicons:  incomplete,  non-­‐specific,   inflexible   –  Unable  to  understand  context  surrounding  an   expression  or  the  people  contribu)ng   –  Unable  to  understand  connec)ons  among  related   en))es  and  a`ributes  and  people   –  Unable  to  gauge  quality  of  source  materials  
  • 21. Sophis)ca)on  of  Analysis   •  Seman)c-­‐based  sen)ment  techniques   –  Sen)ment  terms  >>  incorporate  related  expressions,   fuzzy  logic  -­‐  NLP   –  Sen)ment  lexicons  >>  domain  ontologies  (available  or   buildable)  provide  analy)cal  context   –  Able  to  understand  context  surrounding  an  expression   or  the  people  contribu)ng  -­‐  machine  learning  &  other   techniques   –  Able  to  understand  connec)ons  among  related   en))es  and  a`ributes  and  people  -­‐  triples,  event   extrac)on  
  • 22. Dimensions  of  Analysis   •  Ontologies  around  opinion  objects   •  Iden)fica)on  and  qualifica)on  of  en))es  &   a`ributes  &  rela)onships   •  Emo)onal  content  of  expression(s)   •  Quality  gauge  of  sources   •  Profiles  of  individual  commenters   •  Roles/interac)ons/sociology  of  commenters  and   their  affilia)ons   •  Timeframe  for  expressions  and  responses  
  • 23. Beyond  +/-­‐:   Ontology-­‐based  analy)cs   Same  Ontology  breakdown   Same  Scale:  Expressed  Opinions   Higher  values  for  cardiovascular   diseases  with  Avas)n   Source:  BuzzStory  
  • 24. Opinion::Emo)on     Source:  BuzzStory  
  • 25. Quality  of  Content  Sources   topix.com   cancergrace.org   •  Quality:  4.48   •  Quality:  16.78   "I  know  of  one  method  that   "As  shown  above,  a  total  of   would  be  really  scary  and   362  pa)ents  who  hadn't   graphic  that  would  work   progressed  aser  first  line   towards  gepng  people  to   chemo/Avas)n  were   stop  pollu)ng  my  sea  breeze   randomized  to  either  of  the   environment.   two  maintenance  therapy   What  I  wonna  know  is  they   arms,  and  the  combina)on   keep  pupng  down  smokers   arm  showed  a  significantly   and  blaming  us  for   longer  progression-­‐free   evrything.”   survival  (PFS)  coun)ng  from   the  beginning  of  all   treatment,  at  10.”  
  • 26. Affilia)on  Network  –  Map  of   Affilia)ons  of  People  &  Topics   Supplements   Tobacco  Addic)on   Prostate  Cancer   Breast  Cancer   Co-­‐Morbidi)es   Thyroid  Disease   Biomarkers   Lung  Cancer   Targeted  Therapies   Chemotherapy   H&N  Cancer   Source:  BuzzStory  
  • 27. Sociology  of  Affilia)ons  &  Topic   Groupings   Co-­‐Morbidi)es   Tobacco  Addic)on   Other    Types  of   Cancer   Supplements   Misc.  Side-­‐Effects   Misc.  Side-­‐Effects   Biomarkers   Source:  BuzzStory  
  • 28. Where  Does  Sen)ment  Belong?   Contextual   Analy)cs   Keyword   technology  
  • 29. Challenges  Remain   “The  service  at  Reynards  is,  in  general,  friendly  and   loose.  Though  they  couldn’t  find  a  reserva)on  for   four  one  Friday  night,  they  compensated  with  so   much  warmth  and  comped  wine  that  all  was   forgiven.  In  some  ways,  Reynards  offers  what  one   wishes  a  dining  experience  in  Manha`an  would  be:   kindness  instead  of  aptude,  inoffensive  prices,   glorious  food,  and  aesthe)c  variety—the  clientele  is   split  roughly  in  half  between  the  stylish  and  the   schlumpy.”     The  New  Yorker,  September  24,  2012  
  • 30. Resources   •  Bing  Liu,  Sen$ment  Analysis  and  Opinion   Mining,  Morgan  &  Claypool,  2012   •  Bo  Pang  and  Lillian  Lee,  Opinion  Mining  and   Sen$ment  Analysis,  (Founda$ons  and  Trends   in  Informa$on  Retrieval),  Now  Publishers,   2008     •  Sen)ment  Analysis  Symposium,  San  Francisco,   CA,  October  30,  2012  
  • 31. Ques)ons?     hadleyr@nexteraresearch.com