a semantic enrichment approach based on the vector ...

4 downloads 129 Views 9MB Size Report
connotaNons on their own domain. ▫ The presented work uses KM, IR tools and domain ontologies as a. semanNc referenNal to represent, enrich and retrieve ...
A  SEMANTIC  ENRICHMENT  APPROACH   BASED  ON  THE  VECTOR  SPACE  MODEL   SUPPORTING  COLLABORATION  IN  THE   MANUFACTURING  DOMAIN Paulo  Figueiras,  Raquel  Melo,  Ruben  Costa,  Carlos  AgosDnho,  Celson  Lima,  Ricardo  Gonçalves

PresentaDon  Outline • IntroducDon • Main  Concepts • Knowledge  Enrichment  Process • C2NET  E-­‐Procurement  Approach • PracDcal  Use-­‐Case • Conclusions  &  Future  Work

IntroducDon The  problem  is  …

§  European   SMEs   need   to   have   access   to   advanced   management   systems   and     to   collaboraDve   tools   due   to   their   restricted   resources.

§  § 

§ 

SMEs   manufacturing   value   chains   are   distributed   and   dependent   on   complex  informa9on  and  material  flows  requiring  new  approaches  to   reduce  the  complexity  of  manufacturing  management  systems. They   need   ubiquitous   tools   suppor9ng   collabora9on   among   value   chain   partners   and   providing   advanced   algorithms   to   achieve   op9miza9on  of  manufacturing  assets  and  to  respond  faster  and  more   efficiently  to  unforeseen  changes. Manufacturing   Industry   is   in   need   of   tools   to   have   real-­‐9me   communica9on  with  the  shop  floor  to  overcome  the  lengthy  process   of  faults  detec9on,  monitoring  and  produc9on  planning.

IntroducDon §  Knowledge  is  of  crucial  importance  to  companies  in  the  manufacturing   domain  and  it  is  difficult  to  store  and  share. §  Knowledge   Management   (KM)   and   InformaDon   Retrieval   (IR)   are   mechanisms   for   achieving   a   beZer   capitalizaDon   over   the   available   business  informaDon. §  One   scenario   in   which   efficient   knowledge   management   and   sharing   would   help   companies   collaborate   for   achieving   beZer   results   is   collaboraDve   e-­‐procurement.   Companies   share   a   common   understanding   of   the   knowledge   in   order   to   use   the   same   semanDc   connotaDons  on  their  own  domain. §  The   presented   work   uses   KM,   IR   tools   and   domain   ontologies   as   a   semanDc  referenDal  to  represent,  enrich  and  retrieve  crucial  knowledge   that  can  be  used  during  the  e-­‐procurement  process.     § 

It   also   aims   at   automa9ze   ontology   enrichment   through   na   ontology   evolu9on   process,  in  order  to  close  the  loop  of  knowledge  crea9on  and  capitaliza9on  in  e-­‐ procurement  tasks.

Main  Concepts • Knowledge  Management • InformaDon  Retrieval • Indexing  and  Searching • SemanDc  ReferenDals  &  Ontologies

• IndexaDon  through  SemanDc  Vectors  &  the  Vector  Space   Model

Knowledge  Enrichment  Process 1.  Pre-­‐Processing  Stage

Load  Domain   Search  and  label  relevant   Store  relevant   Ontology  and   Knowledge  Sources  with   Knowledge  Sources   RelaDons’   support  of  Domain   on  Knowledge   weights Experts repository

•  The  pre-­‐processing  stage  holds  the  preparaDon  of  both  operaDonal  environment  and  input   sources  (e.g.  domain  ontologies,  documents) •  Experts  play  a  key  role  to  help  inspecDng  and  pre-­‐labelling  those  relevant  knowledge   sources  (KS) •  All  relevant  KSs  are  selected  and  stored  in  a  Knowledge  Base  repository,  to  help  to  deal   with  the  management  of  all  sources  to  be  indexed. •  Experts  may  also  need  to  validate  weights  of  ontological  relaDons  (next  step).

Knowledge  Enrichment  Process 2.  Ontology  EvoluDon

Enriching  ontology  with  new  concepts  and   relaDons

•  The  ontology  evoluDon  happens  when  new  KSs  are  included  in  the  knowledge  repository •  New  concepts  must  be  added,  new  relaDons  may  be  idenDfied  and  equivalent  terms  can   be  extended. •  Allows  the  assessment  of  the  relevance  of  the  domain  ontology  used  regarding  the  current   knowledge  base  repository. •  AssociaDon  rules  are  automaDcally  discovered  based  on  the  semanDc  liaisons  connecDng   ontological  concepts.

Knowledge  Enrichment  Process 3.  SemanDc  Vector  CreaDon

• 

• 

• 

CreaDon  of  SemanDc  Vectors

the  system  calculates  the   G-­‐idf  scores  for  all  terms,   to  reduce  the  size  of  the   staDsDcal  vector   according  to  a  certain   relevance   another  procedure   normalizes  the  staDsDcal   vector  aaer  pruning  the   terms.   Next,  the  semanDc   enrichment  is  responsible   for  the  generaDon  of  the   keyword,  taxonomy  and   ontology-­‐based  vectors,   respecDvely.

Knowledge  Enrichment  Process 4.  ClassificaDon

5.  EvaluaDon

Classify  input  Knowledge  Sources   into  predefined  categories

Analyse  evidence  /  ExplanaDon   for  ClassificaDon

•  Relies  on  the  applicaDon  of  unsupervised   classificaDon  algorithm  (K-­‐Means   clustering),  in  order  to  group  KSs  into   various  categories,  called  clusters

•  Assesses  the  overall  approach  using   classical  precision  and  recall  metrics  to   measure  performance

C2NET  Cloud  Plaeorm  Architecture

C2NET  E-­‐Procurement  Approach KS  semanDc  vectors  to  find  crucial   knowledge  within  an  e-­‐Procurement   collaboraDve  task  within  C2NET  (best   supplier  or  logisDcs  operator,  in  terms   of  price,  speed  of  delivery,  distance,   cargo  space  and  quanDDes,  etc) Will  collect  business,  producDon  and   operaDonal  informaDon  from  different   sources  (PDM/PLM,  producDon   systems,  ERP,  smart  sensors) Will  use  a  semanDc  search  module   based  on  the    Vector  Space  Model,   and  calculates  the  proximity  between   a  query  inpuZed  by  a  user  and  the  KSs   in  the  C2NET  KB C2NET  users  upload  their  KSs,  such  as   catalogues,  inventory  lists,  supplier   and  logisDcs  plans,  among  others,   these  will  be  pre-­‐processed  and  the   corresponding  semanDc  vectors  will  be   created  and  stored.  Ontology  evoluDon   is  also  executed.

C2NET  Metalworking  Networked  SME’s

The  companies  were  not  able   to  easily  share  informaDon   between  to  help  them  achieve   common  goals  (even  being   geographically  close)

Conclusions By  using  the  semanDc  search  tool,  companies  within  C2NET  will  be  able  to   easily  find  documents  and  other  knowledge  sources  which,  otherwise,   would  be  difficult  to  capitalize  on,  and  to  share  with  other  companies.

This  assumpDon  will  be  evaluated  during  the  Dme  of  execuDon  of  the   C2NET  project  by  both  the  evaluaDon  measures  made  to  the  knowledge   enrichment  and  ontology  evoluDon  processes  and  the  project’s  pilot   companies  themselves,  during  the  pilot  execuDon.  

Nevertheless,  the  aim  is  to  help  these  companies  to  organize  their   knowledge,  and  to  solve  their  knowledge  capitalizaDon  issues,  both   internally  and  when  collaboraDng  with  third  parDes.

Thank  you!  Any  quesDons?

Suggest Documents