|
|
Quenten Thomas
thomaq01@csse.uwa.edu.au
Entry year: 2004
Enrolment status: confirmed
Degree: BSc/BE
Degree status: complete semester 2, 2004
Project: The Adaptive Concept Network: Freeing Semantics from Syntax
Supervisor(s): Wei Liu
Project status: complete semester 2, 2004
|
|
Knowledge representation is a key field of Artificial Intelligence.
Typically, knowledge is represented via some symbolic language.
Symbols can be interpreted in the context of a given syntax to convey semantics.
However, we believe that concepts can be modeled without the use of a symbolic language.
By eliminating the requirement of language to label knowledge, we generalize the knowledge that can be expressed.
Our approach borrows ideas from both classical and connectionist paradigms to represent the fundamental semantics directly, rather than as an imperfect representation.
We define a computational model of a concept as simply a set of relationships with associated concepts and extend the idea of Associations from Topic Maps to model these inter-concept relationships.
This model leads to a network-like structure; with concepts being represented as nodes, and relationships between concepts represented as Associations, connecting nodes.
While representation was our focus, we explain how Spreading Activation techniques can be used as the network's principle processing algorithm.
Through its use, we improve the standard keyword-based query, turning it into a semantically meaningful concept search.
Spreading activation can also be used to adapt the structure of the network to implement knowledge abstraction and generalization.
Hence, we dubbed our network of concepts and Associations, an `Adaptive Concept Network'.
|
 Proposal
Dissertation
|
Last update: Mon Mar 7 21:25:31 2005
|
|
