Concept Trees: Building Dynamic Concepts from Semi-structured Data Using Nature-Inspired Methods - Complex System Modelling and Control Through Intelligent Soft Computations - page 240

in the graph and each node position in the cluster space, is represented by its
position in the graph.
In Fig. 4 , the nodes have a perfect descending order. Using the process of linking
to your closest neighbour, this could lead to the whole set of nodes creating a single
cluster in one go. In Fig. 5 , the perfect ordering is broken, where node 3 will link
with node 2 only. This forces two clusters to be formed, or forces a break in the
sequence. We also have the idea of a minimal energy, or entropy (Shannon 1948 ).
This has already been used to cluster or sort text documents, for example Decision
Trees (Quinlan 1986 ) and the principle of entropy can also be applied to a concept
tree. If one considers the simplistic sorting mechanism in Fig. 4 again, it can be seen
that the most ef cient sort, causing the least amount of energy to move from one

place to the next, is in fact the uniform decreasing of the entity lengths, from largest
to smallest. If each energy change is 1 unit, then a total of 7 units are required. Any
change in this order, for example Fig. 5 , would require a larger amount of energy to
traverse all of the entities

9 units in this case. As natural systems like lower energy
states, a self-organising system might favour the lower energy state. This therefore
supports the idea of not adding larger counts to smaller ones, because the required
energy amount for the same entity set increases, as in Fig. 5 . It could increase and
then decrease uniformly, but in general, it would support the rule. Entropy also

—

Fig. 4 Energy of 7 required
to traverse all elements

Fig. 5 A single change
increases the energy amount
to 9