S calable K nowledge C omposition Ontology Interoperation January 19, 1999 Jan Jannink, Prasenjit...

Preview:

Citation preview

Scalable

Knowledge

Composition

Ontology Interoperation

January 19, 1999

Jan Jannink, Prasenjit Mitra, Srinivasan Pichai,

Danladi Verheijen, Gio Wiederhold

Database Group (Infolab), Stanford University

KC

S

Prasenjit Mitra SKC 2

KC

S

Road Map

• SKC Project Overview

– The Problem» The Approach

» Issues

» Example: NATO Web

– The Algebra & Its Application

–Conclusion & Future Directions

Prasenjit Mitra SKC 3

KC

S

The Approach

• Integration of Knowledge from Multiple Sources

–Preserve the autonomy of sources

–Compose ontologies using the algebra» Spreads the maintenance cost

» Scales smoothly to more complex inferences

–Reuse existing sources and knowledge for new applications

Prasenjit Mitra SKC 4

KC

S

Issues

–Semantic Mismatch» mismatch in terms

» automatic discovery and resolution expensive

» difficulty in processing and matching terms

– Incomplete Specifications» full semantics not specified

– Inconsistent Data» data from multiple sources inconsistent

Prasenjit Mitra SKC 5

KC

S

Example: NATO Web

• URLshttp://www.nato.int/family/countries.htm

http://www.nato.int/php/partners.htm

• Partial Contentslegislature (parliament, house, senate)governmentstate head prime minister

Prasenjit Mitra SKC 6

KC

S

Austria

Prasenjit Mitra SKC 7

KC

S

England

Prasenjit Mitra SKC 8

KC

S

Finland

Prasenjit Mitra SKC 9

KC

S

SKC methodology

• Construct an embedding for a frame-like object in terms of semistructured data as in the OEM data model

• A rule language for explicitly resolving semantic mismatches and for restructured views

• Contexts over semistructured data using the rules to circumscribe areas of interest (similar to views over relations)

• Unary and Binary operations on these contexts

Prasenjit Mitra SKC 10

KC

S

Road Map

• SKC Project Overview

– The Problem

– The Algebra & Its Application» Unary Operators

» Binary Operators

» Rule Primitives

» Application: Intersection

–Conclusion & Future Directions

Prasenjit Mitra SKC 11

KC

S

Unary operators

• Flatten : Build a glossary of terms from an ontology

• Circumscribe : Induce a restricted ontology which is of interest for a specific application. The articulation rules work only on the circumscribed ontology.

• Filter : Select the instance objects satisfying a specific condition

Prasenjit Mitra SKC 12

KC

S

Binary Operators

• A knowledge based algebra for contexts.

• Binary operations

– Intersection : Find the common schema and instances between contexts

– Union : Compose contexts to enrich information

–Difference : Determine the transform between contexts

Prasenjit Mitra SKC 13

KC

S

Rule Primitives

• Provide articulation primitives for matching concepts between ontologies and restructuring objects.

–Match nodes, Add a Child, Merge nodes, Block nodes etc.

• Extraction rules allow us to create contexts from information sources

–Create Nodes, Sequence a list

–Create explicit nodes to accommodate implicit assumptions

–Conversion between instances and schema items permitted

Prasenjit Mitra SKC 14

KC

S

Application: Intersection• Restructuring of two NATO graphs

– 1: Extract the two labeled graphs from the NATO web sources

– 2: Match the two graphs to identify corresponding nodes

– 3: Filter out only matching nodes and restructure one graph to match the structure of the other

Prasenjit Mitra SKC 15

KC

S

Application: Intersection

• Matching of Nodes

–Content Based Matching» Construct list of labels describing each node

» Preprocess labels (if required, to root words)

» Rule-based matching

» Type checking

» Generate heuristic estimates of extent of match

» Accept or reject match based on threshold

–Structure Based Matching

Prasenjit Mitra SKC 16

KC

S

Road Map

• SKC Project Overview

– The Problem

– The Algebra & Its Application

–Conclusion & Future Directions» Future Work

» Summary

Prasenjit Mitra SKC 17

KC

S

Future Work

• Estimate maintenance costs to validate our claims

– n sources of size s ; m articulation agents

– Is n * maint[s] + m * maint[agent] < maint[n * s]

• Enable inference within the source of contexts

• Proofs on properties of the operators and rewriting expressions.

Prasenjit Mitra SKC 18

KC

S

Summary• Algebra enables interoperation by– dealing explicitly with differences using

rulesets– keeping source domains autonomous

• Assumes domain has a common ontology– composing domain ontologies requires the

algebra to manage the linkages where articulation occurs

• Articulation knowledge is distributed– allows specialists to work independently– supports multiple intersections and views

• Maintenance is structured and partitioned

Recommended