Concept Trees: Building Dynamic Concepts from Semi-structured Data Using Nature-Inspired Methods - Complex System Modelling and Control Through Intelligent Soft Computations - page 232

Fig. 2 Example of ' pruning ' and optimising the structure when new trees are added

default. Separating and linking is also an optimisation or normalisation feature. If
the process concludes that

is an important concept in its own right, it
should not be duplicated in different places, but rather it should be stored and
updated in one place and referenced to by other trees. The understanding of
'

' drank milk '

drinking milk ' is the same for all 3 animals. Then of course, also the triangular

count rule.

4.2 Combining with Unrelated Branches

Another situation would be if the concept base then receives, for example, 3
instances of: The thirsty elephant drank milk and ate grass. This would automat-
ically add ' ate grass ' to the ' drank milk ' tree and the count would still be OK. The

sequence however, only relates to the original elephant branch in this

case. There is no indication that the boy or cat ate grass. The

' ate grass '

final solution to this is
another new tree and also a new indexing key, to link the elephant with both the
'

trees. This might happen however after a stage of monitoring
(see Sect. 4.3 ). The other tree branches keep the original index, where Fig. 3 shows
what the new set of structures might look like. Note the way that existing links only

milk ' and the ' grass '

Fig. 3 New tree and key value changes the indexes