CORPORUM-OntoExtract

1
CORPORUM-OntoExtract Ontology Extraction Tool
Author Robert Engels Company CognIT a.s
2
Overview

1. On-To-Knowledge project
2. CORPORUM
3. CORPORUM-OntoExtract
4. Discussion
5. Conclusion

3
What is Knowledge Management?
Knowledge Management is the collection of
processes that govern the creation,
dissemination, and utilization of knowledge.
--- Brian Newman, 1991
4
What is On-To-Knowledge (OTK) project?
Goals develop tools and methods for supporting
knowledge management relying on sharable and
reusable knowledge ontologies. The technical
backbone of On-To-Knowledge is the use of
ontologies for the various tasks of information
integration and mediation.
5
What is On-To-Knowledge (OTK) project?

• European project in EU Information Society
  Technologies (IST) Program EU-IST-10132
• Duration 2.5 years, January 2000 - June 2002
• Total effort cost 26 personyears, 2.5 M EUR
• Partners

1. CognIT a.s
2. AIdministrator
3. AIFB (University of Karlsruhe)
4. BT Research
5. Enersearch
6. Swiss Life Information Systems Research Group

6
CognIT a.s

• Established in Halden, Norway in 1996.
• 20 employees - 3 with PhD
• CORPORUMTM
• Develops Technology for

1. intelligent search by means of agents
2. text analysis and extraction
3. structuring and fusing data to build knowledge
4. knowledge bases and feedback of experience
5. data mining and text mining

7
On-to-Knowledge workbench

• CORPORUM-OntoExtract extract ontologies from
  unstructured documents and represent them in
  XML/RDF/OWL
• CORPORUM-OntoWrapper extract ontologies from
  structured documents and represent them in
  XML/RDF/OWL
• RDF-DB (Sesame)
• RDF-Ferret interface between users and RQL
• OntoEdit (Ontology Editor)
• RQL engine query RDF-DB
• DAML-OIL representation language

8
The OnToKnowledge system architecture
9
Introduction of CORPORUM
CORPORUM is a tool for information retrieval and
extraction developed by CognIT a.s.

• crawl the internet and intranet
• analyzing relevance and content
• maintain knowledge base (RDF-DB)

Features

• focus on the content
• searches, cataloguing, summaries and extractions
  can be performed according to user interests
• founded on CognlTs Mimir technology

10
The overall CORPORUM architecture
11
Introduction of CORPORUM
Core technology -- MIMIR includes

• Linguistic analysis through all levels and
  generate user interested ontology in RDF.
• Similar analysis obtain documents which are most
  pertinent to a specific analyzed text.
  (information retrieval and extraction)

12
Classical Natural Language processing
decomposed.
13
Mimir architecture
14
Introduction of CORPORUM
Informaton distribution
Histogram showing where the desired content in
the document can be found and to what degree it
is pertinent.
15
CORPORUM-OntoExtract

• The web-based version of a CORPORUM version
• Use same architecture as the CORPORUM
• Extract ontologies from unstructured web pages
• Represent extracted ontologies in XML/RDF/OIL

16
CORPORUM-OntoExtract

• CMOntoBuild taken care of overall control of the
  system and co-ordinating all information flows
• CMWebHandler responisble for collecting all
  (text-) documents from a specific site
• CMCogLib analysis texts, extracts information,
  exports a variety of formats
• CMLexEn language dependent support module for
  CMCoglib
• CMWebInteract communication component that takes
  care of all interaction of CORPORUM-OntoExtract
  with the RDF database. Responsible for querying
  the RDF-DB, as well as submitting final analysis
  results.
• DOMhandler integrated in CMWebInteract, the
  OpenXML DOM handler takes care of the
  interpretation of the results which are returned
  from the RDF server

17
CORPORUM-OntoExtract performs the following tasks

• CMOntoBuild is invoked by the user
• CMWebHandler is invoked by CMOntoBuild
• CMWebHandler retrieves the domain that is
  specified from the intra/internet and returns it
  to CMOntoBuild
• CMOntoBuild passes texts to the CMCoglib that
  analyses, interprets and extracts information
  from these texts, and returns a basic RDF
  representation to CMOntoBuild
• CMOntoBuild now analyses the generated RDF and
  queries the RDF Ontology repository to try to
  find knowledge that can augment the previously
  generated RDF
• When all querying that could be performed is
  done, and the RDF is augmented, the final RDF
  ontology for a specific document is sent to the
  RDF server together with a
• reference to the original text.

18
Client/Server based System Architecture of
CORPORUM-OntoExtract
19
The overall CORPORUM architecture
20
CORPORUM-OntoExtract output

• Namespace definitions
• Dublin Core based metadata
• Property definitions
• Ontology
• Facts/instances
• Cross-taxonomic relations

21
Discussion on use of CORPORUM technology in
OntoExtract
Content in natural language vs. content in
structure

• CORPORUM-OntoExtracte can capture content without
  considering the layout and structure of the
  texts.
• In some cases, the structure of texts has to be
  considered. Contracts, licenses.
• CORPORUM-OntoWrapper

22
Discussion on use of CORPORUM technology in
OntoExtract
Diversity of web pages (unknown intention)

• Diversity of documents on the web
• It is difficult to analyze a text according to
  the intention of the writers
• Combination of CORPORUM-OntoExtract with
  CORPORUM-OntoWrapper might some of these issues

23
Discussion on use of CORPORUM technology in
OntoExtract
Representational issues (A-box vs. T-box
reasoning)

• TBox Tbox consists of (class) concept inclusion
  axioms (and/or equivalence) -- e.g., "C subsumes
  D.
• ABox Abox consists of individual/tuple
  membership axioms - e.g., "x is an instance of C"
  or "ltx,ygt is an instance of R".
• Most of the CORPORUM-OntoExtract generated
  knowledge is TBox knowledge.

24
Discussion on use of CORPORUM technology in
OntoExtract
Domain specificity of extracted knowledge

• Since the ontologies are extracted from specified
  domains, the extracted information is expected to
  be restricted in these domains.
• Positive while many of the searches will also be
  rather domain specific, and knowledge about
  cross-taxonomic relations might come in very
  handy.
• Negative one may like to build up domain
  independent knowledge bases.

25
Conclusion

• CORPORUM helps web become more semantic.
• Semantic-based technology.
• Enhance usability of formal knowledge
  representations for end-users
• Decrease initial efforts when defining an
  ontology in new domains

26
Conclusion

• Dynamicity of the analysis, i.e. ease of use in
  dynamic environments
• Offer new ways of navigating knowledge bases and
  documents sets by visualization of contents and
  by means of semantic-based, graphic structures
• Extract of content-based meta-data from
  documents, such as important concepts, semantic
  structures, etc.
• Ability to offer domain-specific information as
  related-keywords

27
Comments

• Description is too general. No examples and
  details.
• Weak sentences. Complicate sentence structures.

28
Questions