Page 245 -

8.3 Organizational Mining 227

Fig. 8.9 Organizational model discovered based on the event log

8.3.2 Discovering Organizational Structures

The behavior of a resource can be characterized by a profile, i.e., a vector indicating
how frequent each activity has been executed by the resource. By using such pro-
files, various clustering techniques can be used to discover similar resources. Fig-
ure 8.8 showed an example in which three roles are discovered based on similarities
of the profiles of the six resources. In Sect. 3.3, we introduced k-means clustering
and agglomerative hierarchical clustering. For k-means clustering the number of
clusters is decided upfront. Agglomerative hierarchical clustering produces a den-
drogram allowing for a variable number of clusters depending on the desired gran-
ularity. Additional relevant features of resources (authorizations, salary, age, etc.)
can be added to the profile before clustering. This all depends on the information
available. After clustering the resources into groups, these groups can be related to
activities in the process. Figure 8.9 shows the end result using the roles discovered
earlier.
The three roles Assistant, Expert, and Manager in Fig. 8.9 have the property that
they partition the set of resources. In general this will not be the case, e.g., a resource
can have multiple roles (e.g., a consultant that is also team leader). Moreover, each
activity corresponds to precisely one role. Also this does not always need to be the
case. Figure 8.10 sketches a more general situation.
The hypothetical organizational model in Fig. 8.10 connects the process model
and the resources seen in the event log. There are eighth organizational entities:
oe1,...,oe8. The model is hierarchical, e.g., oe4 contains resource r5 and all re-
sources of oe6, oe7, and oe8. Hence five resources belong to organizational entity