Extracting RDF Data from Unstructured Sources Based on an RDF Target Schema

1
Extracting RDF Data from Unstructured Sources
Based on an RDF Target Schema

• Tim Chartrand
• Research Supported By NSF

2
Motivation

• Semantic Web Global machine understandable
  knowledge base
• WWW lots of information/data designed for human
  consumption
• DEG contribution Extract data from the human
  readable web
• Proposed solution Extract WWW data and
  structure it in the Semantic Web format (RDF)

3
Overview of Proposed Research
User
HTML Page
Extraction Ontology
RDF Schema
Extraction Engine
Relational Data
RDF Data
4
RDF What is it?

• Resource Description Framework
• Language of the Semantic Web
• Set of subject-predicate-object triples
• tim.html, creator, tim, tim.html, type,
  thesis
• ltRDFgt
• ltDescription abouttim.htmlgt
• ltCreatorgtTimlt/Creatorgt
• ltTypegtThesislt/Typegt
• lt/Descriptiongt
• lt/RDFgt

Tim
creator
tim.html
type
Thesis
5
RDF Schema Basics

• Core Concepts
• rdfsclass The usual concept of a class.
• Ex. Class Person
• rdfssubClassOf Specifies the generalization of
  a class
• Ex. Class Teacher is subClassOf Person
• rdfsproperty Can apply to a class. Has a value
  which.
• Ex. Class Person has property Name
• rdfsdomain Classes to which a property can
  apply.
• Ex. Property Name has domain Person
• rdfsrange Possible values of a property.
• Ex. Property Name has range Literal
• rdfssubPropertyOf Specifies the generalization
  of a property
• Ex. Property Nickname is subPropertyOf Name

6
Example RDF Schema

• Full Schema
• ltrdfsClass rdfIDPerson gt
• lt/rdfsClassgt
• ltrdfsClass rdfIDFuneral gt
• lt/rdfsClassgt
• ltrdfProperty rdfIDPFuneral" gt
• ltrdfsdomain rdfresource"Person"/gt
• ltrdfsrange rdfresource"Funeral"/gt
• lt/rdfPropertygt
• ltrdfProperty rdfID"Name" gt
• ltrdfsdomain rdfresource"Person"/gt
• ltrdfsrange rdfresource"rdfsLiteral"/gt
• lt/rdfPropertygt

7
RDF Schema Graph
8
Extraction Ontology

• Ontology Structure
• Classes map to object sets
• Properties map to binary relationship sets
  between classes
• Literal properties map to relationship sets
  between classes and lexical data frames
• Primary Object Constraints best guess based
  on heuristics\
• Data Frames
• Need a data frame library
• Match properties with data frame library
• Specialize the property data frames

9
User Modification

• Cardinality Constraints
• Allow the user to edit any of the generated
  constraints
• Keep track of changes affects database schema
• Data Frames
• Provide a data frame editor
• Allow user to modify the specialized data frames
• Usually only add key words

10
Input Web Page
11
Relational Data
12
Extracted RDF Data

• Full RDF
• ltobitPerson rdfID"1001"
• obitName"Lemar K. Adamson"
• gt
• ltobitFuneral rdfresource"5001" /gt
• lt/obitPersongt
• ltobitFuneral rdfID"5001"
• obitFuneralAddress"1540 E. Linden"
• obitFuneralDate""
• obitFuneralTime"1000 a.m."gt
• lt/obitFuneralgt

13
RDF Data Graph
14
Conclusions

• Converting RDF Schemas to Data Extraction
  Ontologies can be done with some user
  interaction.
• The nature and amount of user interaction
  necessary for good data extraction is a good
  topic for research
• Converting relational data to RDF data can be
  done automatically