Ontology Maintenance with an Algebraic Methodology: a Case Study

1
Ontology Maintenance with an Algebraic
Methodology a Case Study

• Jan Jannink, Gio Wiederhold
• Presented by
  Lei Lei

2
Challenges

• Obstacle Autonomy of diverse knowledge sources
• Data volatility and amount increases cost
• Major challeges Establish and maintain
  application specific portion of knowledge sources

3
An Algebraic Approach

• Construct virtual knowledge bases geared to a
  specific application
• Use composable operators to transform contexts
  into contexts
• Operators express relevant parts of a source and
  the conditions using rules
• Rules define a valid context transformation

4
On-line DictionaryWebster

• Autonomously maintained to develop a novel
  thesaurus application
• 120,000 entries, two million words
• Semi-annual updates
• Errors and inconsistencies help robustness

5
Target Application

• Construct a graph of the definitions to determine
  related terms, and automatically generate
  thesaurus entries

6
Related Work

• Ontology composition (Wiederhold 1994)
• Rule-based approach to semantic integration
  (Bergamaschi et al. 1999)
• Semantic reconciliation (Siegel 1991)
• Uschold et al. 1998
• Specification morphisms, (Smith 1993)
• WordNet system (Miller al. 1990)
• WHIRL (Cohen 1998)
• PageRank (PageBrin 1998)
• Latent semantic indexing (Deerwester 1990)
• Hypertext authority (Kleinberg 1998)

7
Outline

• Algebra Usage Scenario
• Background
• Context Creation
• Ontology Maintainance
• Future Work
• Conclusion
• My Evaluation

8
Typical Algebra Usage Scenario
A minimal sufficient set of Linkage between
items in different resources
9
Background
Canonical unary to establish and refine a context
within which the source knowledge meets the
application requirement

• Algebraic Operators

10
Background(Cont.)

• Semantic Context
• No global notion of consistency
• Defined as objects that encapsulate other
  objects
• Congruity relevance of source info. to
  application
• Similarity equivalent and mergeable
  objects
• between different
  sources

11
Rule Language(Cont.)

• Allow uninterpreted components of an object to
  become attributes of the object
• Constructors
• create new objects
• Constructors
• generate proxy objects
• Editors convertors
• modify the objects

12
Object Model(Cont.)

• Subsume existing models
• Only objects have an identity to which others can
  refer
• Correspond to XML supplemented with obj. identity
• Rich to model complex relationship

13
Context Creation

• Summarize Operator (S operator)
• Transforms source data based on a predicate
• Create object Encapsulates populates
• Data classificationGroups source into
  equivalent classes
• Syntax (given
  contexts c1,c2, a matching rule e)

14
Context Creation(Cont.)

• 1.Predicate e partitions the objects of c1 into n
  equivalent parts
• 2. c2 consists of n1 values e,s1,s2,,sn
• 3.One is an exception class, not match e

15
Example with Websters Dictionary

• Automatic Thesaurus Extraction from Dictionary

16
Example(Cont.)

• Construct a directed
• graph from definition
• 1.Each head word and
• definition grouping is a node
• 2.Each word in a definition node
• is an arc to the node having
• that head word
• Definition
  from the dictionary data for Egoism

17
Context Creation(Cont.)

• Syllable and accent markers in head words
• Misspelled head words
• Mis-tagged fields
• Stemming and irregular verbs(Hopelessness)
• Common abbreviations in definitions(etc.)
• Undefined words with common prefixes(un-)
• Multi-word head words(Water Buffalo)
• Underfined hyphenated and compound words
• Set 99 accuracy in the conversion from data to
  graph stru.

18
Constructing the Congruity Expression

• An object that represents the entire source
• Subdivided into chunks
• One head word
• One definition
• Express congruity relationship between the
  dictionary and thesaurus application

19
Ontology Maintenance

• Context Refinement
• Return the ten longest head words of the
  dictionary

20
Maintaining the Ontology

• Changes in source help correct and extend dict.
• Maintain statistics with the S operator when
  extracting the relevant parts of the dictionary
• Find no longer needed rules
• Note which rules no longer needed
• A comparison of the terms reveals new errors

21
Future Work

• A web based interface to display ArcRank
  algorithm based on PageRank
• (http//www-db.stanford.edu/SKC)

22
Conclusion

• An on-line dictionary is good test-bed
• An algebraic approach improving maintainability
• Congruity simplified identification and handling
  of changes
• Use Summarize to define and refine a context that
  prepare the dictionary data for thesaurus service
  use

23
My Assessment

• Strength
• Decouple the selection of congruent parts of
  the source data
• Congruity and similarity measure use algebra
  rather than single language
• Mirror classes using operators of the algebra
  instead of low level abstract primitives that are
  difficult to compose
• Weakness
• Details of ciS(ci) are needed
• Difficult to grasp the capability of S
  operator
• Accuracy and error accumulation problem
• Ambiguous Rules Generation

24
Questions?