Mining Matrix Pattern from Mobile Users - Distributed Artificial Intelligence, Agent Technology, and Collaborative Applications - page 242

Figure 8. Algorithm for generation of single layer matrix pattern
Void FindPattern (Single Layer Matrix M) {
For (I = 0, M.Size(), M++) {
If M.I.Significance = True {
Display M.I;
Display M.I.Confidence();
}
}
}
less important than the original confidence threshold because at this stage, only highly significant static
nodes are taken into consideration. Those low significant static nodes have been taken out of consider-
ation and adjusting a higher relationship confidence threshold than the confidence threshold will filter
out the minor inaccuracies.
Step : Generate Single Layer Matrix Pattern
Figure 8 shows that the final stage of this mobile user data mining process is the generation of single
layer matrix pattern. The generation of this piece of knowledge contains two parts. The first part consists
of the generation of knowledge based on the confidence alone, that is, a list of static nodes in a matrix
structure that contains high confidence threshold. This can be generated by means of a single layer
matrix with the static nodes with high confidence identified with some marker. This piece of knowledge
shows the static nodes that are highly significant and the relative positions of them.
The second part of the creation of single layer matrix pattern is the generation of the relationship.
Each relationship consists of two static nodes and the strength of the relationship in percentage. The
stronger the relationship confidence between two static nodes, the stronger their significance. It is im-
portant to group the order of patterns according to relationship confidence. The highest relationship
confidence should be displayed first then goes down to the lowest relationship confidence. Finally, the
matrix is also displayed and lines are drawn across two static nodes that have highly significant rela-
tionship confidence.
Multi-Layer Matrix
A multi-layer matrix , as the name suggests, contains multiple layers of matrices. Each layer is asso-
ciated with a particular domain, that is, a logical theme. A domain or logical theme is the context of
the problem which the data miner wishes to solve. For example, two very common logical themes are
entertainment and shopping, when the data miner wishes to find out the relationship of mobile user
behaviours between an entertainment question and a shopping question. In a multi-layer matrix, each
layer will be representing different single logical domain. The advantage of a multi-layer matrix is that
it provides a good structure for mobile user data mining by having the ability to analyse the behaviour
of mobile users across several logical domains.
In a multi-layer matrix, each layer must have been assigned with a confidence threshold. Each layer
can share the same value for the confidence threshold, or they can share different levels of confidence