Page 244 -

226 8 Mining Additional Perspectives

Fig. 8.8 Social network based on similarity of profiles. Resources that execute similar collections
of activities are related. Sara is the only resource executing e and f . Therefore, she is not connected
to other resources. Self-loops are suppressed as they contain no information (self-similarity)

The weight of role Expert is only 1.15, because the two experts (Sue and Sean) only
execute activity b which, on average, is executed 1.15 times per case. The weights
of the arcs are directly taken from Table 8.6. Clearly, handovers among the roles
Assistant and Manager are most frequent.
Counting handovers of work is just one of many ways of constructing a social
network from an event log. In [104] various types of social networks are presented.
For example, one can simply count how many times two resources have worked on
the same case, i.e., two nodes have a strong relationship when they frequently work
together on common cases. One can also use Table 8.4 to quantify the similarity of
two resources. Every row in the resource-activity matrix can be seen as the profile
of a resource. Such a vector describes the relevant features of a resource. For ex-
ample, Pete has profile P Pete = (0.30,0.0,0.345,0.69,0.0,0.0,0.135,0.165),Mike
has profile P Mike = (0.5,0.0,0.575,1.15,0.0,0.0,0.225,0.275), and Sara has pro-
file P Sara = (0.0,0.0,0.0,0.0,2.3,1.3,0.0,0.0). Clearly, P Pete and P Mike are very
similar whereas P Pete and P Sara are not. The distance between two profiles can be
quantified using well-known distance measures such as the Minkowski distance,
Hamming distance, and Pearson’s correlation coefficient. Moreover, clustering tech-
niques such as k-means clustering and agglomerative hierarchical clustering can be
used to group similar resources together based on their profile (see Sect. 3.3). Two
resources in the same cluster (or in close proximity according to the distance metric)
are strongly related whereas resources in different clusters (or far away from each
other) have no significant relationship in the social network.
For the resource-activity matrix shown in Table 8.4, it does not matter which dis-
tance metric or clustering technique is used. All will come to the conclusion that
Pete, Mike, and Ellen are very similar and thus have a strong relationship in the so-
cial network based on similarity. Similarly, Sue and Sean have a strong relationship
in the social network based on similarity. Sara is clearly different from the resources
in the two other groups. Figure 8.8 shows the social network based on similarity.
Here one can clearly see the roles Assistant, Expert, and Manager mentioned be-
fore. However, now the roles are discovered based on the profiles of the resources.