SlideShare a Scribd company logo

2012 11 7 TAR Webinar Part 3 Sigler

Sonya Sigler, profile picture
Sonya Sigler

The document discusses technology-assisted review (TAR) in detail, focusing on various underlying technologies such as rules-based systems, Bayesian classification, and latent semantic indexing. It covers different search technologies, including dtsearch and lucene, and their functionalities like keyword searching, fuzzy searching, and boolean operators. The document emphasizes the importance of defensibility in the TAR process, highlighting the need for transparency and quality control.

1 of 31
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Demys&fying  Technology   Assisted  Review   Part  3:  Deconstruc&ng  the   Technology   Sonya  L.  Sigler  
Agenda     Review/Overview     Underlying  Search  Technology     dtSearch     Lucene  (open  source)     Others  –  My  SQL,  etc.     Underlying  StaCsCcal  Based  Technology     Rules  Based  Technology  (LinguisCc  or  StaCsCcal)     Bayesian  ProbabilisCc  Technologies     Latent  SemanCc  Indexing     Q  &  A                        Demys&fying  Technology  Assisted  Review  
Review/Overview  -­‐  Search  &  Review  Spectrum   Linear  Review   Culling   IteraCve  search   Review   Accelerated  Review     Email  Threading   Near  Duplicate  DetecCon   Automated  Review     Per     CA  -­‐  Clustering   Relevance  Ranking   Document   CategorizaCon  (Supervised)   Machine  Learning   Cost   Latent  SemanCc  Indexing   (staCsCcal  probability)   PaRern  Analysis   Sampling  Data  for  High   Precision  and  Recall  Rates   Organiza3on  Commitment                        Demys&fying  Technology  Assisted  Review  
Underlying  Technologies   Rules  Based  Systems   dtSearch   Key  word  Search   Ontologies   Lucene   Other  Search  Engines   LinguisCc  –  word  based   StaCsCcal  -­‐  #s  based   Bayesian  ClassificaCon   Support  Vector  Models   Latent  SemanCc  Indexing                        Demys&fying  Technology  Assisted  Review  
Database  NormalizaCon   From:  Nuala  Coogan  Nuala@SFLData.com   Subject:  EDI  Summit  –  Florida   Date:  October  3,  2012  10:11:21  AM  PDT   To:  Sigler  L.  Sonya  Sonya@sigler.name     From:  Nuala  Coogan   Subject:  EDI  Summit  –  Florida   Date:  10/03/12   To:  Sonya  Sigler                        Demys&fying  Technology  Assisted  Review  
TokenizaCon     Words,  Phrases,  Symbols     Mostly  at  the  word  level     Numbers     PunctuaCon     Meaningful  Elements  or  Pieces  –>  Tokens     Parsing  and  Text  Mining     Treatment  of  ContracCons,  Hyphenated  words,   EmoCcons  and  Larger  Constructs  (like  urls)     Look-­‐up  tables                        Demys&fying  Technology  Assisted  Review  
LinguisCc  Based  Technologies   Keyword  Sample   Ontology  Sample    Simple:      q  ((+(std:%CapacityReports_%  std:%DINCapacity_ %)    (std:%ACMEEPPlant_%  std:%ProductName_%))    "legal  systems"  OR  legalsystems   (+(std:%ACMEPNPlant_%  std:%ProductName_%)  +  "Mike  Custodian”     (std:%ProducCveCapability_%  std: %CapacityReports_%))  (+(std:%CapacityCreep_%   std:%OperaConsImprovement_%  std:  Medium:   %CapacityExpansion_%  std:%CapacityRestoraCon_  mail(custodian@domain.com)  AND  "legal  systems”   %)  +(std:%ACMEPNPlant_%  std:%ProductName_ %))  (+(std:%EquipmentReplacement_%  std:  (Custodian  w/3  (Mike  OR  Michael  OR  M))   %FinishingColumn_%)  +(std:%ACMEPNPlant_%  std: %ProductName_%))  (std:%Audit_%  actor:%Audit_  Complex:   %)  (+(std:%SeRlementNegoCaCons_%  std: %ContractNegoCaCons_%  )  +(actor:  (privilege  OR  privileged  OR  legally  OR  "work   %ACMEOutsideCounsel_%  std: product")  NOT  w/35  (((original  OR  intended  OR   %ACMEOutsideCounsel_%  actor:%ACME   designated  OR  named)  w/3  (recipient  OR   UBOutsideCounsel_%  std: recipients  OR  addressee  OR  addressees  OR   %AcmeSubOutsideCounsel_%  actor:%AcmeSub_%   solely))  OR  ("message  in  error")  OR  ("received  in   std:%AcmeSub_%))  (std:%FTC_%  actor:%FTC_%)   ((+subject:%ProductName_%  +(std:swap   error")  OR  ("named  above")  OR  ((electronic  or   std:"supply  agreement"  std:"exchange  agreement"   email  or  e-­‐mail)  w/3  (message  or  transmission))   std:"agree  to  exchange"))  std:"name   OR  ("confidenCality  noCce"))                        Demys&fying  Technology  Assisted  Review  
Search  Engines     dtSearch     dtSearch  Corp.,  founded  1991     Incorporated  into  Symantec’s  Norton  Navigator     SDKs  available,  most  license  off  the  shelf     hRp://support.dtsearch.com/faq/search.html     Lucene     Open  source  -­‐  hRp://lucene.apache.org/core/     Doug  Cukng,  1999,  Part  of  Apache  projects  in  2001     APIs,  Customizable     Other  –  My  SQL,  SQL,  (DBMS,  RDBMS)                        Demys&fying  Technology  Assisted  Review  
dtSearch     RelaCvity,  Concordance,  Viewpoint,  others     Single  User  desktop  license  $199     LiRle  CustomizaCon  –  more  similariCes  across  apps     Includes  Boolean  operators     Includes  Proximity  searching     Includes  Fuzzy  Searching     Alphabet  -­‐>  Alphaqet,  alpphabet,  alpkaqet                        Demys&fying  Technology  Assisted  Review  
Lucene     Clearwell,  Intella,  Cataphora,  SHIFT,  others     Open  Source  Tool  –  meant  to  be  customized     LiRle  SimilariCes  Across  Apps     Know  your  defaults!     Includes  Boolean  Operators     Includes  Proximity  Searching                        Demys&fying  Technology  Assisted  Review  
dtSearch  –  Fuzzy  Searching     Degrees  of  Fuzziness     1-­‐10;  dtSearch  uses  1-­‐3     Marked  by  use  of  %  symbol     InserCon:  co%t  →  coat     DeleCon:  coat  →  co%t     SubsCtuCon:  coat  →  cost     TransposiCon  cots  →  cost     Fuzziness  Degrees     Alphabet  –  Alphaqet,  Alpkaqet                        Demys&fying  Technology  Assisted  Review  
Boolean  Operators  –  AND,  OR  ,  NOT   dtSearch   Lucene     Search  for     Depends  on     MulCple  words   customizaCon   treated  as  a  phrase     OR     ANY  –  treats  word     AND   list  as  separated  by   OR     Know  your  defaults     ALL  –  treats  word  list     Spell  out    variaCons   as  separated  by  AND                        Demys&fying  Technology  Assisted  Review  
Proximity   dtSearch   Lucene     Pre/post     w/  order  doesn’t     w/  order  doesn’t   maRer   maRer    House  white     No  pre  usage    White  house     Pre/  finds  first  word   prior  to  second   word    White  house                        Demys&fying  Technology  Assisted  Review  
Punctua&on   dtSearch   Lucene     LeRers     All  punctuaCon     Space   treated  as  a  word     Ignored     break     Hyphens     %  -­‐  fuzzy  searching     _  -­‐  ignored                        Demys&fying  Technology  Assisted  Review  
dtSearch  Hyphen  Example                        Demys&fying  Technology  Assisted  Review  
Noise  Words  –  Unindexed,  Ignored   dtSearch   Lucene     Unindexed,  Can     Ignores  *  in  quotes   create  Custom  Index     (Quality  Control*)  =     Many,  but  a  few   Quality  Control  but   examples:  Do,  not,   nothing  else   for,  your,  only,  under,   made,  way     Know  defualts                        Demys&fying  Technology  Assisted  Review  
Stemming  v.  Wild  Cards   Stemming   Wild  Cards     SyntacCc  VariaCons     Strings  of  characters     Replacements  for  beginning,     Regular  Verbs   parts,  or  endings     Irregular  Verbs     Lucene  -­‐  *     dtSearch  performs     dtSearch  -­‐  ?  For  single   character,  *  for  any  #  of   poorly  with  irregular   characters   verbs     Time  consuming     Spelling  out  recommended     Wild  cards  in  quotes                        Demys&fying  Technology  Assisted  Review  
Stemming  v.  Wild  Cards  Example   Stemming   Wild  Cards   Catch  –  Lucene   Catch*   Catch~  -­‐  dtSearch     Catch     Catch     Catches     Catches     Catching     Catching     Catcher     Catcher     Catch1234  –  not  in     Caught  -­‐  not  in  dtSearch     stemming                        Demys&fying  Technology  Assisted  Review  
StaCsCcal  Technologies     Rules  Based     Bayesian  ClassificaCon     Vector  Space  Modeling     Latent  SemanCc  Indexing                        Demys&fying  Technology  Assisted  Review  
StaCsCcal  Based  Technologies     Concept  -­‐  Clustering     Machine     Unsupervised     Quickly  understand      data     Uncontrolled  Clusters                        Demys&fying  Technology  Assisted  Review  
StaCsCcal  Based  Technologies     Concept  -­‐  Categoriza&on       User  Created     Supervised     Control  Topics     Time  Consuming                        Demys&fying  Technology  Assisted  Review  
StaCsCcal  Based  Technologies   Rules  Based  Systems     If..  Then…     If  email  =  person  1  to  person  2  then  return  it     If  email  =  person  1  or  person  2  then  return  it     ArCficial  Intelligence  Systems     EnCty  extracCon  (&  dicConaries)     Time  consuming     Mirror  human  thinking     Case,  subject  maRer     Transparent  System                        Demys&fying  Technology  Assisted  Review  
StaCsCcal  Based  Technologies     Bayesian  ClassificaCon       ProbabilisCc     Co-­‐occurrence     Frequency     Spam  Filters     Viagra     Concepts     Words,  phrases                        Demys&fying  Technology  Assisted  Review  
Bayesian       Bayesian  illustraCon     Baseball,  glove,  diamond,  bats,  hit,  home  run     Diamond,  pendant,  jewelry     Co-­‐occurrence     Local  –  within  a  document     Global  –  across  document  populaCon     Frequency  –  how  ozen  does  it  appear     WeighCng  –  uniqueness  counts                        Demys&fying  Technology  Assisted  Review  
StaCsCcal  Based  Technologies     Vector  Space  Modeling     Latent  Seman&c  Indexing/Analysis     Words     Phrases,  Concepts     Tables     Algebraic  equaCons  represenCng  docs     WeighCng  Algorithms                        Demys&fying  Technology  Assisted  Review  
Latent  SemanCc  Indexing  Example     Exclude  Noise  Words     The,  and,  or,  etc.     Vector  Space  Modeling     Build  Document  Profile     Diamond     Base,  ball     Necklace,  pendant     Diamond  Saw                        Demys&fying  Technology  Assisted  Review  
MathemaCcal  FoundaCon     Tables  built  with  0s,  1s     Yes  it  has  that  word  or  phrase     No  it  doesn’t                        Demys&fying  Technology  Assisted  Review  
Simple  Matrix  with  WeighCng                        Demys&fying  Technology  Assisted  Review  
Weighted  by  Document  (not  just  type)                        Demys&fying  Technology  Assisted  Review  
Defensibility  Report       Document,  Document,  Document     Transparency     Workflow     What  Was  Considered,  By  Whom?     QC  Process     Metrics                        Demys&fying  Technology  Assisted  Review  
Q&A - Thank you!   Post  your  ques&ons  to  the   presenter  in  the  chat  secCon   Sonya  L.  Sigler   Vice  President,  Product  Strategy  &  Consul&ng   SFL  Data   415-­‐321-­‐8385   sonya@sfldata.com     www.sfldata.com                          Demys&fying  Technology  Assisted  Review  

More Related Content

PDF
spChains: A Declarative Framework for Data Stream Processing in Pervasive App...
Fulvio Corno
 
PDF
2012 6 27 TAR Webinar Part 1 Sigler
Sonya Sigler
 
PDF
2013 3 27 TAR Webinar Part 4 Getting Started Sigler
Sonya Sigler
 
PDF
Who's Afraid of eDiscovery?
CallPM
 
PPTX
Georgetown lecture 2012 6 2 full
Sonya Sigler
 
PDF
2013 7 24 TAR Webinar 5 Tips & Myths Sigler
Sonya Sigler
 
PDF
Julia Brickell - Your "Big Buckets" Are Full Of "Big Data" - The Information ...
ARMA International
 
PPT
Sejarah perkembangan akuntansi syariah
Titiek Widyastuti
 
spChains: A Declarative Framework for Data Stream Processing in Pervasive App...
Fulvio Corno
 
2012 6 27 TAR Webinar Part 1 Sigler
Sonya Sigler
 
2013 3 27 TAR Webinar Part 4 Getting Started Sigler
Sonya Sigler
 
Who's Afraid of eDiscovery?
CallPM
 
Georgetown lecture 2012 6 2 full
Sonya Sigler
 
2013 7 24 TAR Webinar 5 Tips & Myths Sigler
Sonya Sigler
 
Julia Brickell - Your "Big Buckets" Are Full Of "Big Data" - The Information ...
ARMA International
 
Sejarah perkembangan akuntansi syariah
Titiek Widyastuti
 

Similar to 2012 11 7 TAR Webinar Part 3 Sigler (20)

PDF
2012 8 29 TAR Webinar Part 2 Sigler
Sonya Sigler
 
PDF
Concept Searching Portal Solutions Search Engine Face Off
martingarland
 
PDF
DDC2011 - Association
Buhwan Jeong
 
PPTX
Taxonomy Assessments - Part Two
Access Innovations, Inc.
 
PDF
Semantic Technology: State of the arts and Trends
Won Kwang University
 
PDF
Concept Searching Overview Google Vs Fast
martingarland
 
PDF
Large-Scale Search Discovery Analytics with Hadoop, Mahout, Solr
DataWorks Summit
 
KEY
A Data Scientist And A Log File Walk Into A Bar...
Paco Nathan
 
PDF
Semantic Search Trend
Won Kwang University
 
PPTX
랭킹 최적화를 넘어 인간적인 검색으로 - 서울대 융합기술원 발표
Jin Young Kim
 
PPTX
Чираг Шах «Коллективный поиск, взаимодействие пользователей: подходы к изучен...
Yandex
 
PDF
How Search 2.0 Has Been Redefined by Enterprise 2.0
Enterprise 2.0 Conference
 
PDF
Tutorial kcc-2011
Won Kwang University
 
PPTX
Machine Learned Relevance at A Large Scale Search Engine
Salford Systems
 
PDF
Scale, Structure, and Semantics
Daniel Tunkelang
 
PDF
FAST Search for SharePoint
C/D/H Technology Consultants
 
KEY
Evolution: It's a process
Christine Connors
 
PDF
Introduction to Information Retrieval & Models
Mounia Lalmas-Roelleke
 
PPTX
SF Women in eDiscovery Sept 2011
Sonya Sigler
 
PPTX
Large Scale Search, Discovery and Analytics with Hadoop, Mahout and Solr
Grant Ingersoll
 
2012 8 29 TAR Webinar Part 2 Sigler
Sonya Sigler
 
Concept Searching Portal Solutions Search Engine Face Off
martingarland
 
DDC2011 - Association
Buhwan Jeong
 
Taxonomy Assessments - Part Two
Access Innovations, Inc.
 
Semantic Technology: State of the arts and Trends
Won Kwang University
 
Concept Searching Overview Google Vs Fast
martingarland
 
Large-Scale Search Discovery Analytics with Hadoop, Mahout, Solr
DataWorks Summit
 
A Data Scientist And A Log File Walk Into A Bar...
Paco Nathan
 
Semantic Search Trend
Won Kwang University
 
랭킹 최적화를 넘어 인간적인 검색으로 - 서울대 융합기술원 발표
Jin Young Kim
 
Чираг Шах «Коллективный поиск, взаимодействие пользователей: подходы к изучен...
Yandex
 
How Search 2.0 Has Been Redefined by Enterprise 2.0
Enterprise 2.0 Conference
 
Tutorial kcc-2011
Won Kwang University
 
Machine Learned Relevance at A Large Scale Search Engine
Salford Systems
 
Scale, Structure, and Semantics
Daniel Tunkelang
 
FAST Search for SharePoint
C/D/H Technology Consultants
 
Evolution: It's a process
Christine Connors
 
Introduction to Information Retrieval & Models
Mounia Lalmas-Roelleke
 
SF Women in eDiscovery Sept 2011
Sonya Sigler
 
Large Scale Search, Discovery and Analytics with Hadoop, Mahout and Solr
Grant Ingersoll
 
Ad

2012 11 7 TAR Webinar Part 3 Sigler

  • 1. Demys&fying  Technology   Assisted  Review   Part  3:  Deconstruc&ng  the   Technology   Sonya  L.  Sigler  
  • 2. Agenda     Review/Overview     Underlying  Search  Technology     dtSearch     Lucene  (open  source)     Others  –  My  SQL,  etc.     Underlying  StaCsCcal  Based  Technology     Rules  Based  Technology  (LinguisCc  or  StaCsCcal)     Bayesian  ProbabilisCc  Technologies     Latent  SemanCc  Indexing     Q  &  A                        Demys&fying  Technology  Assisted  Review  
  • 3. Review/Overview  -­‐  Search  &  Review  Spectrum   Linear  Review   Culling   IteraCve  search   Review   Accelerated  Review     Email  Threading   Near  Duplicate  DetecCon   Automated  Review     Per     CA  -­‐  Clustering   Relevance  Ranking   Document   CategorizaCon  (Supervised)   Machine  Learning   Cost   Latent  SemanCc  Indexing   (staCsCcal  probability)   PaRern  Analysis   Sampling  Data  for  High   Precision  and  Recall  Rates   Organiza3on  Commitment                        Demys&fying  Technology  Assisted  Review  
  • 4. Underlying  Technologies   Rules  Based  Systems   dtSearch   Key  word  Search   Ontologies   Lucene   Other  Search  Engines   LinguisCc  –  word  based   StaCsCcal  -­‐  #s  based   Bayesian  ClassificaCon   Support  Vector  Models   Latent  SemanCc  Indexing                        Demys&fying  Technology  Assisted  Review  
  • 5. Database  NormalizaCon   From:  Nuala  Coogan  Nuala@SFLData.com   Subject:  EDI  Summit  –  Florida   Date:  October  3,  2012  10:11:21  AM  PDT   To:  Sigler  L.  Sonya  Sonya@sigler.name     From:  Nuala  Coogan   Subject:  EDI  Summit  –  Florida   Date:  10/03/12   To:  Sonya  Sigler                        Demys&fying  Technology  Assisted  Review  
  • 6. TokenizaCon     Words,  Phrases,  Symbols     Mostly  at  the  word  level     Numbers     PunctuaCon     Meaningful  Elements  or  Pieces  –>  Tokens     Parsing  and  Text  Mining     Treatment  of  ContracCons,  Hyphenated  words,   EmoCcons  and  Larger  Constructs  (like  urls)     Look-­‐up  tables                        Demys&fying  Technology  Assisted  Review  
  • 7. LinguisCc  Based  Technologies   Keyword  Sample   Ontology  Sample    Simple:      q  ((+(std:%CapacityReports_%  std:%DINCapacity_ %)    (std:%ACMEEPPlant_%  std:%ProductName_%))    "legal  systems"  OR  legalsystems   (+(std:%ACMEPNPlant_%  std:%ProductName_%)  +  "Mike  Custodian”     (std:%ProducCveCapability_%  std: %CapacityReports_%))  (+(std:%CapacityCreep_%   std:%OperaConsImprovement_%  std:  Medium:   %CapacityExpansion_%  std:%CapacityRestoraCon_  mail(custodian@domain.com)  AND  "legal  systems”   %)  +(std:%ACMEPNPlant_%  std:%ProductName_ %))  (+(std:%EquipmentReplacement_%  std:  (Custodian  w/3  (Mike  OR  Michael  OR  M))   %FinishingColumn_%)  +(std:%ACMEPNPlant_%  std: %ProductName_%))  (std:%Audit_%  actor:%Audit_  Complex:   %)  (+(std:%SeRlementNegoCaCons_%  std: %ContractNegoCaCons_%  )  +(actor:  (privilege  OR  privileged  OR  legally  OR  "work   %ACMEOutsideCounsel_%  std: product")  NOT  w/35  (((original  OR  intended  OR   %ACMEOutsideCounsel_%  actor:%ACME   designated  OR  named)  w/3  (recipient  OR   UBOutsideCounsel_%  std: recipients  OR  addressee  OR  addressees  OR   %AcmeSubOutsideCounsel_%  actor:%AcmeSub_%   solely))  OR  ("message  in  error")  OR  ("received  in   std:%AcmeSub_%))  (std:%FTC_%  actor:%FTC_%)   ((+subject:%ProductName_%  +(std:swap   error")  OR  ("named  above")  OR  ((electronic  or   std:"supply  agreement"  std:"exchange  agreement"   email  or  e-­‐mail)  w/3  (message  or  transmission))   std:"agree  to  exchange"))  std:"name   OR  ("confidenCality  noCce"))                        Demys&fying  Technology  Assisted  Review  
  • 8. Search  Engines     dtSearch     dtSearch  Corp.,  founded  1991     Incorporated  into  Symantec’s  Norton  Navigator     SDKs  available,  most  license  off  the  shelf     hRp://support.dtsearch.com/faq/search.html     Lucene     Open  source  -­‐  hRp://lucene.apache.org/core/     Doug  Cukng,  1999,  Part  of  Apache  projects  in  2001     APIs,  Customizable     Other  –  My  SQL,  SQL,  (DBMS,  RDBMS)                        Demys&fying  Technology  Assisted  Review  
  • 9. dtSearch     RelaCvity,  Concordance,  Viewpoint,  others     Single  User  desktop  license  $199     LiRle  CustomizaCon  –  more  similariCes  across  apps     Includes  Boolean  operators     Includes  Proximity  searching     Includes  Fuzzy  Searching     Alphabet  -­‐>  Alphaqet,  alpphabet,  alpkaqet                        Demys&fying  Technology  Assisted  Review  
  • 10. Lucene     Clearwell,  Intella,  Cataphora,  SHIFT,  others     Open  Source  Tool  –  meant  to  be  customized     LiRle  SimilariCes  Across  Apps     Know  your  defaults!     Includes  Boolean  Operators     Includes  Proximity  Searching                        Demys&fying  Technology  Assisted  Review  
  • 11. dtSearch  –  Fuzzy  Searching     Degrees  of  Fuzziness     1-­‐10;  dtSearch  uses  1-­‐3     Marked  by  use  of  %  symbol     InserCon:  co%t  →  coat     DeleCon:  coat  →  co%t     SubsCtuCon:  coat  →  cost     TransposiCon  cots  →  cost     Fuzziness  Degrees     Alphabet  –  Alphaqet,  Alpkaqet                        Demys&fying  Technology  Assisted  Review  
  • 12. Boolean  Operators  –  AND,  OR  ,  NOT   dtSearch   Lucene     Search  for     Depends  on     MulCple  words   customizaCon   treated  as  a  phrase     OR     ANY  –  treats  word     AND   list  as  separated  by   OR     Know  your  defaults     ALL  –  treats  word  list     Spell  out    variaCons   as  separated  by  AND                        Demys&fying  Technology  Assisted  Review  
  • 13. Proximity   dtSearch   Lucene     Pre/post     w/  order  doesn’t     w/  order  doesn’t   maRer   maRer    House  white     No  pre  usage    White  house     Pre/  finds  first  word   prior  to  second   word    White  house                        Demys&fying  Technology  Assisted  Review  
  • 14. Punctua&on   dtSearch   Lucene     LeRers     All  punctuaCon     Space   treated  as  a  word     Ignored     break     Hyphens     %  -­‐  fuzzy  searching     _  -­‐  ignored                        Demys&fying  Technology  Assisted  Review  
  • 15. dtSearch  Hyphen  Example                        Demys&fying  Technology  Assisted  Review  
  • 16. Noise  Words  –  Unindexed,  Ignored   dtSearch   Lucene     Unindexed,  Can     Ignores  *  in  quotes   create  Custom  Index     (Quality  Control*)  =     Many,  but  a  few   Quality  Control  but   examples:  Do,  not,   nothing  else   for,  your,  only,  under,   made,  way     Know  defualts                        Demys&fying  Technology  Assisted  Review  
  • 17. Stemming  v.  Wild  Cards   Stemming   Wild  Cards     SyntacCc  VariaCons     Strings  of  characters     Replacements  for  beginning,     Regular  Verbs   parts,  or  endings     Irregular  Verbs     Lucene  -­‐  *     dtSearch  performs     dtSearch  -­‐  ?  For  single   character,  *  for  any  #  of   poorly  with  irregular   characters   verbs     Time  consuming     Spelling  out  recommended     Wild  cards  in  quotes                        Demys&fying  Technology  Assisted  Review  
  • 18. Stemming  v.  Wild  Cards  Example   Stemming   Wild  Cards   Catch  –  Lucene   Catch*   Catch~  -­‐  dtSearch     Catch     Catch     Catches     Catches     Catching     Catching     Catcher     Catcher     Catch1234  –  not  in     Caught  -­‐  not  in  dtSearch     stemming                        Demys&fying  Technology  Assisted  Review  
  • 19. StaCsCcal  Technologies     Rules  Based     Bayesian  ClassificaCon     Vector  Space  Modeling     Latent  SemanCc  Indexing                        Demys&fying  Technology  Assisted  Review  
  • 20. StaCsCcal  Based  Technologies     Concept  -­‐  Clustering     Machine     Unsupervised     Quickly  understand      data     Uncontrolled  Clusters                        Demys&fying  Technology  Assisted  Review  
  • 21. StaCsCcal  Based  Technologies     Concept  -­‐  Categoriza&on       User  Created     Supervised     Control  Topics     Time  Consuming                        Demys&fying  Technology  Assisted  Review  
  • 22. StaCsCcal  Based  Technologies   Rules  Based  Systems     If..  Then…     If  email  =  person  1  to  person  2  then  return  it     If  email  =  person  1  or  person  2  then  return  it     ArCficial  Intelligence  Systems     EnCty  extracCon  (&  dicConaries)     Time  consuming     Mirror  human  thinking     Case,  subject  maRer     Transparent  System                        Demys&fying  Technology  Assisted  Review  
  • 23. StaCsCcal  Based  Technologies     Bayesian  ClassificaCon       ProbabilisCc     Co-­‐occurrence     Frequency     Spam  Filters     Viagra     Concepts     Words,  phrases                        Demys&fying  Technology  Assisted  Review  
  • 24. Bayesian       Bayesian  illustraCon     Baseball,  glove,  diamond,  bats,  hit,  home  run     Diamond,  pendant,  jewelry     Co-­‐occurrence     Local  –  within  a  document     Global  –  across  document  populaCon     Frequency  –  how  ozen  does  it  appear     WeighCng  –  uniqueness  counts                        Demys&fying  Technology  Assisted  Review  
  • 25. StaCsCcal  Based  Technologies     Vector  Space  Modeling     Latent  Seman&c  Indexing/Analysis     Words     Phrases,  Concepts     Tables     Algebraic  equaCons  represenCng  docs     WeighCng  Algorithms                        Demys&fying  Technology  Assisted  Review  
  • 26. Latent  SemanCc  Indexing  Example     Exclude  Noise  Words     The,  and,  or,  etc.     Vector  Space  Modeling     Build  Document  Profile     Diamond     Base,  ball     Necklace,  pendant     Diamond  Saw                        Demys&fying  Technology  Assisted  Review  
  • 27. MathemaCcal  FoundaCon     Tables  built  with  0s,  1s     Yes  it  has  that  word  or  phrase     No  it  doesn’t                        Demys&fying  Technology  Assisted  Review  
  • 28. Simple  Matrix  with  WeighCng                        Demys&fying  Technology  Assisted  Review  
  • 29. Weighted  by  Document  (not  just  type)                        Demys&fying  Technology  Assisted  Review  
  • 30. Defensibility  Report       Document,  Document,  Document     Transparency     Workflow     What  Was  Considered,  By  Whom?     QC  Process     Metrics                        Demys&fying  Technology  Assisted  Review  
  • 31. Q&A - Thank you!   Post  your  ques&ons  to  the   presenter  in  the  chat  secCon   Sonya  L.  Sigler   Vice  President,  Product  Strategy  &  Consul&ng   SFL  Data   415-­‐321-­‐8385   sonya@sfldata.com     www.sfldata.com                          Demys&fying  Technology  Assisted  Review