Learning Ontologies in a Multiagent System for the World Wide Web

Much research has been done on intelligent agent knowledge sharing through the use of common, pre-defined ontologies. However, several domains, including the World Wide Web, often precipitate intelligent agents selfishly inventing ontologies based on their utility for the task at hand. If such agents are able and willing to teach each other semantic concepts based on their individual ontologies, then the entire group of agents can accomplish their group and individual tasks more efficiently and effectively. We present our prototype DOGGIE (Distributed Ontology Gathering Group Integration Environment) which demonstrates how a group of agents with diverse Web ontologies can learn the location and translation of semantic concepts.

Click here to start

Table of Contents

Learning Ontologies in a Multiagent System for the World Wide Web

PPT Slide

Intelligent Agent

Multiagent Learning

Ontology Definition

Research Problem

A Childís Ontology

ìCommonî Ontology Problems

Research Issues

Motivation

Related Work

Research Goal

Proposed Solution

Solution

Ontologies and the World Wide Web Domain

Web Ontologies

PPT Slide

Ontology Learning in a MAS

Semantic Concept Learning

Semantic Concept Interpretation

Interpretation Thresholding

K Region

Single M Region

Recursive Semantic Context Rule Learning (RSCRL)

Recursive Semantic Context Rule Learning (cont.)

Multiple M Regions

Concept Cluster Integration (CCI)

Concept Cluster Integration

D Region

PPT Slide

DOGGIE Distributed Ontology Gathering Group Integration Environment

Multiagent Architecture

PPT Slide

User Interface

Experiments

Performance Measurements

Distributed Concept Locations

Agent Model Learning Fully Connected Network

PPT Slide

PPT Slide

PPT Slide

PPT Slide

PPT Slide

PPT Slide

Disparate References to Concept

Concept Translation Learning Fully Connected Network

Different Vocabularies

Recursive Semantic Context Rule Learning Fully Connected Network

Different Vocabularies Query Forward / Concept Translation

PPT Slide

Query Forward / Concept Translation Summary

Discovering Concept Relations

Concept Cluster Integration Results

Concept Precision Summary

Concept Recall Summary

Discussion

Future Work

Conclusion

Author: Andrew B. Williams
Electrical and Computer Engineering
University of Iowa

Andrew B. Williams received the B.S. in Electrical Engineering in 1988 from the University of Kansas, the M.S. in Electrical & Computer Engineering in 1995 from Marquette University, and the Ph.D. in Electrical Engineering from the University of Kansas in 1999. He is currently an assistant professor in the Electrical & Computer Engineering Department at the University of Iowa. He previously worked at GE medical Systems and Allied Signal Aerospace Company. He also has internship experience at GE Corporate Research & Development and Lucent Technologies. Dr. Williams' research activities include developing multiagent learning algorithms for AI-assisted browsing and search for the World Wide Web. He has also worked on intelligent image classification and applied AI applications. Dr. Williams was a Patricia Roberts Harris Fellow in 1991 and a GEM Ph.D. Fellow sponsored by General Electric Company.