Ontology Based Extraction of RDF Data from the World Wide Web

1
Ontology Based Extraction of RDF Data from the
World Wide Web

• Tim Chartrand
• Masters Thesis
• Research Supported By NSF

2
Introduction

• World Wide Web
• Has a huge amount of existing information
• Designed primarily for human consumption
• Semantic Web
• Is an extension of WWW
• Gives information a well-defined meaning
• Allows automation of tasks
• DEG contribution Extract data from the WWW
• Solution
• Extract Semantic Web data from the WWW
• Superimpose extracted data

3
Research Overview
4
RDF What is it?

• Resource Description Framework
• Language of the Semantic Web
• Set of ltsubjectgtltpredicategtltobjectgt triples
• ltmailtotim_at_cs.byu.edugtltgenealogyagegt25
• ltmailtotim_at_cs.byu.edugtltgenealogyfatherOfgtltmail
  totyler_at_thechartrands.comgt

5
DAML

• Core Concepts
• damlclass defines a class
• damlproperty defines a binary relation, has a
  value
• rdfsdomain specifies class to which a property
  applies
• rdfsrange specifies possible values of a
  property
• damluniqueProperty, damlunambiguousProperty
  specify cardinality constraints for a property

6
Example Ontology

• . . .
• ltdamlClass rdfID"Program"gt
• ltrdfslabelgtProgramlt/rdfslabelgt
• lt/damlClassgt
• ltdamlClass rdfID"OperatingSystem"gt
• ltrdfslabelgtOperatingSystemlt/rdfslabelgt
• lt/damlClassgt
• . . .
• ltdamlDatatypeProperty rdfID"Name"gt
• ltrdftype rdfresource"damlUniqueProperty"/
  gt
• ltrdftype rdfresource"damlUnambiguousPrope
  rty"/gt
• ltrdfsdomain rdfresource"Program"/gt
• ltrdfsrange rdfresource"rdfsLiteral"/gt
• lt/damlDatatypePropertygt
• ltdamlProperty rdfID"supportsOperatingSystem"gt
• ltrdfsdomain rdfresource"Program"/gt
• ltrdfsrange rdfresource"OperatingSystem"/gt
• lt/damlPropertygt
• . . .

7
DAML ? OSM

• Class ? Non-lexical object set
• Property ? Binary relationship set between object
  sets
• Literal property ? Lexical object set and binary
  relationship set between non-lexical and lexical
  object sets
• Cardinality restriction ? Participation constraint

8
DAML ? OSM

• ltdamlClass rdfID"Program"gt
• ltrdfslabelgtProgramlt/rdfslabelgt
• lt/damlClassgt
• ltdamlClass rdfID"OperatingSystem"gt
• ltrdfslabelgtOperatingSystemlt/rdfslabelgt
• lt/damlClassgt
• . . .
• ltdamlDatatypeProperty rdfID"Name"gt
• ltrdftype rdfresource"damlUniqueProperty"/gt
  
• ltrdftyperdfresource"damlUnambiguousPropert
  y"/gt
• ltrdfsdomain rdfresource"Program"/gt
• ltrdfsrange rdfresource"rdfsLiteral"/gt
• lt/damlDatatypePropertygt
• ltdamlProperty rdfID"supportsOperatingSystem"gt
• ltrdfsdomain rdfresource"Program"/gt
• ltrdfsrange rdfresource"OperatingSystem"/gt
• lt/damlPropertygt

9
Data Frames

• Lexical object sets need data frames.
• Use data-frame library
• Match lexical object sets with data frames
• Compare stemmed names and aliases
• Levenshtein edit distance
• Soundex
• Longest common subsequence
• Weighted average
• Specialization heuristic
• Choose most similar data frame (above a threshold)

10
User Modification

• Provide graphical ontology editor
• Automate graph layout
• Allow the user to edit participation constraints
• Allow user to edit data-frame mapping
• Provide data frame editor

11
Extracting the Data
12
Pointing to the Data
lthtmlgt . . . ltbodygt lttablegt
lttrgt lttdgt
lta href"..."gtltbgtStick
Death 1.0lt/bgtlt/agtltbr /gt
Advance in levels, grab weapons, and unlock new
levels and
characters.ltbr /gt
ltbgtOSlt/bgt Windows 3.x/95/98/Me/NT/2000/XPltbr /gt
ltbgtFile
Sizelt/bgt2.66MBltbr /gt
ltbgtLicenselt/bgtFreeltbr /gt
lt/tdgt lttdgt05/14/2002ltbr /gt
ltigtltbgtnewlt/bgtlt/igt
lt/tdgt
lttdgtlt/tdgt lttdgt2,235lt/tdgt
lttdgtlta href"..."gtDownload
nowlt/agtltbr /gtltbr /gtlt/tdgt lt/trgt
. . .
xpointer(string-range(/html1/body1/table1/tr
1, , 10, 3))
13
Convert to RDF
14
Superimposed Data
15
Results

• RDF Data Extraction and Viewing
• Built 4 data-extraction ontologies
• 3 from DAML ontologies for data extraction
• 1 from an existing DAML ontology
• Most existing DAML ontologies not good for data
  extraction
• Data Frame Matcher
• 8 training ontologies, 16 test ontologies
• 128 lexical object sets, 40 correct matches, 12
  incorrect matches
• Precision 77
• Recall 89
• Experiment (apartment rentals) 6 students 3 data
  frames
• Phone 2.8 min
• RentalRate 16.5 min
• Bedrooms 17.5 min

16
Contributions

• Advancement of Semantic Web
• Application of Information Extraction to building
  Semantic Web content
• Semantic Web data as superimposed information
• Algorithm for ontology conversion

17
Future Work

• Data extraction
• Enhance name matcher with data values
• Support n-ary relationship sets
• RDF data generation
• Generate only one URI for an object
• Associate concepts from DAML ontologies to
  well-known DAML ontologies