AeroDAML%20Applying%20Information%20Extraction%20to%20Generate%20DAML%20Annotations - PowerPoint PPT Presentation

Title: AeroDAML%20Applying%20Information%20Extraction%20to%20Generate%20DAML%20Annotations


1
AeroDAMLApplying Information Extractionto
Generate DAML Annotations

• Dr. Paul Kogut
• Lockheed Martin
• Management Data Systems

2
What is Information Extraction?
Information Extraction
Text or web pages
Entities
Relationships
Co-references
Events
Linguistic Knowledge
3
Extraction and Semantic Annotation

• Consumer-side extraction - 3rd party text -gt
  database
• Advantages
• Applicable to raw documents (most of the web)
• Disadvantages
• Must deal with full complexity of natural
  language
• Semantic annotation proposed to overcome
  difficulty of consumer-side extraction - but
  annotation is labor intensive
• Producer-side extraction - authored text -gt
  annotation
• Advantages
• Partial-automation - reduces manual effort
• Human assisted disambiguation
• Domain customization for intranets and B2B
  e-commerce
• Disadvantages
• Requires manual effort to correct and add rich
  set of relationships
• Domain customization requires up-front effort
  from the author/webmaster
• Both types of extraction will coexist.

4
AeroDAML Architecture
UBOT
Annotation Editor
refined annotation
basic annotation
basic annotation
Extraction to DAML Translation
DAML annotated text or web pages
DAML Ontologies
Text or web pages
Text Extraction
5
Client-Server AeroDAML

• Users
• personnel who routinely produce documents (e.g.,
  intelligence analysts)
• personnel who have a large collection of legacy
  documents

6
Web-based AeroDAML

• Users
• novice/infrequent DAML annotators
• people who want to do quick/simple annotation of
  a web page

7
AeroDAML Output Entities
ltaacABSOLUTEDATE rdfabout"December19,1997"gt
ltdamllabelgtlt!CDATADecember 19,
1997gtlt/damllabelgt lt/aacABSOLUTEDATEgt
ltaacAIRCRAFT rdfabout"Dash8Series400"gt
ltdamllabelgtlt!CDATADash 8 Series
400gtlt/damllabelgt lt/aacAIRCRAFTgt ltaacMEASURE
rdfabout"61-foot"gt ltdamllabelgtlt!CDATA61-
footgtlt/damllabelgt lt/aacMEASUREgt
8
AeroDAML Output Relationships
ltaacNATION rdfabout"Austria"gt
ltdamllabelgtlt!CDATAAustriagtlt/damllabelgt lt/aac
NATIONgt ltaacORGANIZATION rdfabout"TyroleanAir
ways"gt ltaacOrgToLoc rdfresource"Austria"/gt
ltdamllabelgtlt!CDATATyrolean
Airwaysgtlt/damllabelgt lt/aacORGANIZATIONgt
9
AeroDAML Output Co-reference
ltaacPERSON rdfabout"PierreLortie"gt
ltaacPersToOrg rdfresource"BombardierRegionalAir
craft"/gt ltdamlequivalentTo rdfresource"Lortie"/
gt ltdamllabelgtlt!CDATAPierre
Lortiegtlt/damllabelgt lt/aacPERSONgt
ltaacPERSON rdfabout"Lortie"gt
ltdamllabelgtlt!CDATALortiegtlt/damllabelgt
lt/aacPERSONgt
10
AeroDAML Plans

• Integrate with annotation editor
• Improve Web-based AeroDAML
• Allow user to select other ontologies besides the
  current AeroDAML default ontology for annotation
  generation
• OpenCyc or Cyc Upper Ontology
• CIA World Fact Book
• IEEE Standard Upper Ontology
• Dublin Core
• UNSPSC...
• Try AeroDAML!
• http//ubot.lockheedmartin.com/ubot/hotdaml/aeroda
  ml.html