Information Technology Reference
In-Depth Information
1 Introduction
The term
has been used previously (Jarke et al. 1995 ; Zhao et al.
2007 , for example) and has been adopted in Greer ( 2011 ) to describe a database of
heterogeneous sources, representing information that has been received from the
environment and stored in the database for processing. The key point is that the
information received from one source does not have to be wholly consistent with
information received from another source. The uncertain environment in which it
operates, means that information can be much more fragmented, heterogeneous, or
simply unrelated to other sources. This could be particularly true in a sensorised
environment, when sensors provide a relatively small and speci
'
concept base
'
c piece of infor-
mation. As the sensor-based information would be determined by the random and/
or dynamic environment in which it operates, there can be much less cohesion
between all of the different input sources. For example, event 1 triggers sensor A
with value X and shortly afterwards, event 2 triggers sensor B with value Y. Later,
event 1 again triggers sensor A with value X, but instead, event 3 occurs and
triggers sensor B with value Z. While the nature of the data is much more random,
statistical processes can still be used to try to link related pieces of information. This
linked data can then represent something about the real world. The term
'
can be used to describe a single value or a complex entity equally and so the
concept base can consistently store information from any kind of data source.
Intelligent linking mechanisms can be used to try to turn the smaller, more sim-
plistic and separate concepts into larger, more complex and meaningful ones. This
is probably also more realistic in terms of what humans have to deal with, in our
interaction with the real world.
While information might be input and stored in an ad hoc manner, it is probably
the case that some level of structure must
'
concept
firstly be added to the information, before
it can be processed, data mined, or reasoned over. When looking for patterns or
meaningful relations; then if the data always appears to be random, it is more
difficult to find the consistent relations and so a first stage that does this would
always be required. This paper looks at a very generic and simplistic way of adding
structure to the data, focusing particularly on using whatever existing structure there
is, as a guide. Other statistical processes can then use the structure to try to generate
some knowledge. Thinking of the sensors or data streams, for example—if
if it can be
determined that concepts A and B usually occur together, while concepts C and D
also occur together; knowledge might be able to tell us that when A-B occurs, C-D
is likely to occur soon afterwards, or maybe should occur as something else. The
current context is to extract this structure from textual information sources, but this
is only an example of how the method would work. If consistent patterns can be
found, they can be used to grow
. A concept tree is essentially and
AND/OR graph of related concepts that grows naturally from the ordering that
already exists in the data sources. This paper is concerned with describing the
structure of these concept trees and how the process might work. Note that this is at
the structure-creation level and not the knowledge-creation level just mentioned.
'
concept trees
'
Search WWH ::




Custom Search