42 /342
為頻繁單變量不確定樣式產生摘要
42
modified
k
-medoids algorithm is similar to the procedure of the
k
-means clustering algorithm
(MacQueen, 1967) except that the former uses medoids rather than means. The set of clusters
generated by the modified
k
-medoids algorithm may contain some
unsatisfied clusters
. An
unsatisfied cluster contains one or more FU2Ps that are more similar to the FU2Ps in the
other clusters than the FU2Ps in the unsatisfied cluster; this requires the unsatisfied cluster
should be further decomposed into several clusters to make sure each FU2P in a cluster is
most similar to the other FU2Ps in the same cluster. We apply the modified
k
-medoids
algorithm to each of these unsatisfied clusters in the third and fourth steps. Then, we
examine all the clusters to determine if there exist unsatisfied clusters. We continue to apply
the modified
k
-medoids algorithm to the unsatisfied clusters until there exist no unsatisfied
clusters. In the fifth step, we merge similar clusters. Finally, the summary of the FU2Ps is
generated in the sixth step.
Determine the initial set of
representative FU2Ps
Merge similar
clusters
Generate
summary
End
Apply the modified
k
-medoids algorithm to all
FU2Ps
Determine the initial set of
representative FU2Ps for
each unsatisfied cluster
Apply the modified
k
-medoids algorithm to
each unsatisfied cluster
Unsatisfied
clusters exist?
No
Yes
Figure 1 The Flow Chart of the SFC Algorithm