RIDLUR: Relationship Identification Leveraging Userdefined Rules

1
RIDLURRelationship Identification Leveraging
User-defined Rules

• Meta-data Extraction Group
• CSCI6350
• Zixin Wu
• Vincent Howard
• Michael Moore

2
Background Reading

• How to Create Topic Maps
• Identifies all associations between two items in
  an index (topic map)
• Uses GATE/JAPE
• Classifying Semantic Relations in Bioscience
  Texts
• Identifies relations between two fixed entities
• S-CREAM - Semi-automatic Creation of Metadata
• Annotation and Creation of metadata-enriched
  documents
• Uses machine learning techniques to create rules
  for extraction
• Semantic Enhancement Engine (SEE)
• Does not use natural language processing
  techniques
• Automatically determines target ontology
• Keyword Extraction from the Web for FOAF Metadata
• Uses context to identify relationships (two
  entities occur in multiple web pages)
• Instance-Based Learning and Information
  Extraction for the Generation of Metadata
• Uses simple (Perl) based regular expressions to
  identify entities and relationships
• Doesnt work well for dirty data or a large
  variety of knowledge types

3
Research Problems

• Information Extraction of structured,
  semi-structured and unstructured data
• Entity Identification
• Entity Disambiguation
• Relationship Identification
• Mapping to ontology
• Classification
• Identifying importance to target domain (one and
  many)

4
Problems to be addressed

• Relationship Identification
• Identify relationships in unstructured text
• Applicability to target domain
• Only concerned with relationships defined in the
  target ontology
• Allows relationship extraction rules to be
  defined in the context of the relationship

5
System Architecture
6
System Architecture

• GATE (Natural language processing tool)
• English Tokeniser
• Separate words and punctuations as "Tokens".
• Sentence Splitter
• Separate every sentence.
• POS Tagger
• Annotate the grammar function to every token.
• Morphological analyzer
• Annotate the root of every verb.
• Gazetteer
• Annotate the type of every token, by looking up
  lists.
• OrthoMatcher
• For example, "IBM" and "Big Blue" is the same
  thing.
• Verb Group Chunker
• For example, "is going to investigate" is a verb
  chunk.
• NE Transducer
• Find the named entities. For example, "Mr. Smith"
  is a person.

7
System Architecture

• User perspective
• Define regular expressions based on relationships
  in target ontology
• Translated into JAPE Rules (complex regular
  expressions)
• Generate automata on-the-fly.
• Matching text returned to user
• Simple text
• RDF

8
Abilities and Limitations

• Abilities
• Present user with text that matched the regular
  expressions defined
• Identifies instances of relationships defined in
  the target ontology
• Associate each regular expression with one
  relationship in the target ontology
• Limitations
• Does no machine learning
• Does not handle coreference and subordinate
  clause
• Does not create instance data
• Simple user interface
• Only allows specifying one schema/knowledgebase

9
Example
Mr. Smith lived in Athens.
lived in Athens
subj
in
Verb2Subject subj Verb2Object in
Mr. Smith (Person)
Athens (Location)
Mr. Smith and Bob lived in Athens.
lived in Athens
subj
in
Verb2Subject subjset_member Verb2Object in
set
Athens (Location)
set_member
Mr. Smith (Person)
Bob (Person)
Verb2Subject subj(set_member)
10
References

• How to Create Topic Maps, Kerk, R. (University
  Halle-Wittenberg), Groschupf, S. (Media Style
  Labs).
• Classifying Semantic Relations in Bioscience
  Texts, Rosario, B, Hearst, M. UC Berkeley.
  Proceedings of ACL. 2004.
• S-CREAM - Semi-automatic Creation of Metadata,
  Handschuh, S., Stabb, S., Ciravenga, F. Institute
  AIFB, Department of CS, University of Sheffield.
• Semantic Enhancement Engine A Modular Document
  Enhancement Platform for Semantic Applications
  over Heterogeneous Content, Hammond, B.,
  Sheth, A., Kochut, K. Semagix and Department of
  Computer Science, University of Georgia.Real
  World Semantic Web Applications, 2002.
• Keyword Extraction from the Web for FOAF
  Metadata, Mori, J. et. all. University of Tokyo.
• Instance-Based Learning and Information
  Extraction for the Generation of Metadata,
  Lattner, A., Herzog, O. Center for Computing
  Technologies - TZI, University of Bremen,
  Germany. 3rd International Conference on
  Knowledge Management, 2003.

11
References

• Extracting Personal Names from Emails Applying
  Named Entity Recognition To Informal Text,
  Minkov, E., Cohen, W., Wang, R. Carnegie Mellon
  University. Computational Linguistics, 2004.
• Computational Linguistics Meets Metadata, or the
  Automatic Extraction of Keywords From Full Text
  Content, Deegan, M, et. all. 2004.
• Automatic Document Metadata Extraction using
  Support Vector Machines, Han, H. , et. all.
  JCDL, 2003.
• Exploiting Dictionaries in Named Entity
  Extraction Combining Semi-Markov Extraction
  Processes and Data Integration Methods, Cohen,
  W. (Carnegie Mellon University), Sarawagi, S.
  (IIT Bombay). KD, 2004.
• A Comparison of String Metrics for Matching
  Names and Records, Cohen, W., Ravikumar, P.,
  Flenberg, S. Carnegie Mellon University. 2003.
• From Unstructured Data to Actionable
  Intelligence, Rao, R. IEEE IT Pro, 2003.
• SemTag and Seeker Bootstrapping the Semantic
  Web via Automated Semantic Annotation, Dill, S.,
  et. all. IBM Almaden Research Center. 2003.
• Web Directories as Training Data for Automated
  Metadata Extraction, Kavalec, M., Svatek, V.,
  Strossa, P. University of Economics, Prague,
  Czech Republic. 2004.
• Adaptive Name Matching in Information
  Integration, Bilenkio, M., Mooney, R.
  (University of Texas at Austin), Cohen, W.,
  Ravikumar, P., Flenberg, S. (Carnegie Mellon
  University). IEEE Intelligent Systems, 2003.
• A Generalized Framework for an Ontology-Based
  Data-Extraction System, Wessman, A., Liddle, S.,
  Embley, D. ISTA. 2005.
• Automatic Discovery of WordNet Relations,
  Hearst, M. WordNet An Electronic Lexical
  Database. MIT Press. 1998.