Mining Patterns From The Web With Ontology

1
Mining Patterns From The Web With Ontology

• Eugene Khokhlov
• Eugene_at_ApplicaSoft.com

2
Ontology Generation from Tables WISE 2003

• Yuri Tijerino, Brigham Young University
• David W. Embley, Brigham Young
• Deryle W. Longsdale, Brigham Young
• George Nagy, Rensselaer Polytechnic

3
Ontology Generation from Tables

• Motivation
• We need to access information in multiple tables
  in different web pages.
• To understanding tables we rely on
• Acquired expertise
• Background information
• Practice

4
Ontology Generation from Tables

• Paper addresses the table processing issue by
  developing a new framework to table
  understanding.
• The ultimate goal is to develop an
  information-gathering engine to assimilate and
  organize knowledge.
• TANGO-(Table Analysis for Generating Ontologies)

5
Ontology Generation from Tables

• Four Steps
• Understand a tables structure and conceptual
  content
• Discover the constraints that hold between
  concepts extracted from the table
• Match the recognized concepts with ones from a
  more general specification of related concepts
• Merge the resulting structure with other similar
  knowledge representations for use in future
  situations.

6
Ontology Generation from Tables

• The idea is that information-gathering engine
  behind TANGO expands from a small set of rules
  rather than growing from scratch
• Examples in the paper are based on geopolitical
  facts. Basically for simplicity sake all
  examples have some relevance to geopolitical
  information.

7
TANGO Kernel concept

• Build a kernel application ontology.
• Having only few concepts (geopolitical in this
  paper)
• Central (the most important concepts)
• Example-rich (containing typical sample data)

8
Ontology

• For any given table, the system creates a
  mini-ontology based on its understanding of the
  table.
• The system attempts to integrate each new
  mini-ontology with the ontology it is building.

9
Ontology Integration Issues

• There may be alternative ways it can integrate
  the mini-ontology into the evolving global
  ontology
• Constraints may be inconsistent
• Adjustments to evolving ontology may be necessary
• Might need human intervention.

10
Kernel Example
11
Understand Table (associate the attributes with
values) Example 1
12
Discover Constraints

• Agglomeration gt Population
• Population gt Agglomeration, Continent, Country
• Country /gt Population, Country, Agglomeration
• Country /gt Population, Agglomeration.

13
Mini-Ontology
14
Match Merge Ontology
15
Example 2 Understand Table one table made up of
many small tables
16
Mini-Ontology
17
Discover Constraints Match Merge
18
The Future of TANGO

• Examples are sketches of TANGO functionality.
• Examples looked at the geopolitical domain, but
  given a reasonable kernel ontology TANGO could be
  used in any other domain.

19
The Future of TANGO

• TANGO should run in fully automatic mode as well
  as building ontology as directed by its users.
• Allowance for the human input whenever issues are
  encountered with multiple choices.

20
References

• GoogleSets, http//labs.google.com/sets
• Y.A. Tijerino, D.W. Embley, D.W. Lonsdale, and G.
  Nagy. Ontology Generation from Tables. In WISE
  2003, pages 242-252
• Y. Wang and J. Hu. A Machine Learning Based
  Approach for Table Detection on the Web. In WWW
  2002, pages 242-250, 2002