Table of Contents Table of Contents
Previous Page  41 /342 Next Page
Information
Show Menu
Previous Page 41 /342 Next Page
Page Background

臺大管理論叢

27

卷第

2S

41

The distance between two FU2Ps is a weighted sum of

D

ExSup

and

D

App

:

(4)

The range of

w

is between 0 and 1. If w is larger than 0.5, the U2 patterns in a cluster

are more likely to have similar expected supports. In contrast, if

w

is smaller than 0.5, the

FU2Ps in a cluster are more likely to have similar appearances.

Table 6 contains a set of FU2Ps, where each FU2P shows its serial number, appearance,

and expected support. In the figure, some FU2Ps have similar appearances but present

significant differences in their expected supports, e.g.,

F

4

and

F

6

. If we set

w

as 0.1 or 0.9,

the distance between

F

4

and

F

6

is 0.0333 or 0.2997. (

D

ExSup

= 0.333 and

D

App

= 0).

Table 6 A Set of FU2Ps

Serial

number

FU2P

Expected

support

Serial

number

FU2P

Expected

support

F

1

[

A

1

:[427, 433]]

1707.000

F

11

[

A

4

:[368, 374]]

3613.998

F

2

[

A

1

:[427, 435]]

1876.000

F

12

[

A

5

:[190, 194]]

3692.028

F

3

[

A

2

:[335, 337]]

2074.049

F

13

[

A

5

:[195, 197]]

1576.016

F

4

[

A

3

:[35, 37]]

1775.109

F

14

[

A

5

:[188, 193]]

3365.006

F

5

[

A

3

:[30, 33]]

1736.733

F

15

[

A

5

:[192, 197]]

2077.285

F

6

[

A

3

:[34, 37]]

2662.734

F

16

[

A

1

:[427, 435],

A

3

:[34, 37]]

1590.300

F

7

[

A

3

:[30, 34]]

2315.562

F

17

[

A

2

:[335, 337],

A

4

:[368, 374]]

1682.998

F

8

[

A

3

:[33, 35]]

3018.757

F

18

[

A

3

:[33, 35],

A

4

:[362, 365],

A

5

:[190, 194]]

1614.752

F

9

[

A

4

:[362, 365]]

1901.214

F

19

[

A

3

:[30, 34],

A

4

:[361, 367],

A

5

:[195, 197]]

1575.002

F

10

[

A

4

:[361, 367]]

2182.797

F

20

[

A

3

:[33, 35],

A

4

:[368, 374],

A

5

:[190, 194]]

3100.002

3.3 The SFC Algorithm

In this section, we present our proposed SFC algorithm. The SFC algorithm adopts a

hierarchical clustering technique to derive clusters of FU2Ps, where the medoid of each

cluster serves as the representative FU2P. Guided by three user-specified parameters, the

SFC algorithm automatically determines the number of clusters. Figure 1 shows the flow

chart of the SFC algorithm.

In the first step, we choose an initial set of representative FU2Ps. In the second step, a

modified

k

-medoids algorithm is applied to the full set of FU2Ps. The procedure of the