Process Maps

In order to discover structured process models from real life event logs, process dis-covery techniques should be able to abstract from undesired details and provide a high level view on processes. A similar challenge is faced in the eld of cartography, i.e., the simplication of highly complex and unstructured topologies such as towns, roads, etc. Analogous with thematic cartography, that is concerned with creating maps for a specic purpose and for a select part of the overall topology, process discovery techniques aim at providing a hierarchical view of processes with the fa-cility to zoom in or zoom out of the process map [11], [16]. A generic approach for the generation of understandable and context-dependent business process maps is presented in Discovering Process Maps [16]. The mining techniques that we will use for the discovery of process maps is the Fuzzy miner [12], [11]. The plugin has been enhanced to facilitate the discovery of hierarchical process models that sup-port multiple levels of abstractions, with the facility to zoom in/out in the hierarchy.

The approach to construct hierarchical process maps is a two phase approach that aims to mine interactive and context-dependent business process maps based on common execution patterns [16].

1. The rst phase of the approach consists of the pre-processing of logs to a de-sired level of granularity.

A mapping M ⊆ 2^Σ× A is dened between the original alphabet of the event log Σ, and an abstract alphabet A. Let us consider the example event log de-picted as in Figure 3.1. The mapping for this event log is M = {({c, u, d, n}, Y), ({c, u}, Y), ({d, n}, Y), ({s, a, m}, X), ({f}, Z), ({h, i}, Z), ({g, i}, Z), ({i, k}, Z)}, ({f, h, l}, Z), ({g, h, i, k, l}, Z), ({f, g, h, i, k, l}, Z).

D = ∪_(A,a^{) ∈M}A denotes the set of activities in Σ for which a mapping is

de-ned. The original event log L, is transformed into an abstract log L⁰. Each trace t ∈ L is transformed into a corresponding trace t⁰ ∈ L⁰. In each trace t, the manifestation of each pattern captured by A is replaced with its abstract activity, a, in the transformed trace. When replacing the manifestation of each pattern, sub-logs are saved for each abstract activity. For our running example all traces in the event log L (e.g., sambcudnje, etc.) will be transformed into

3.2. Process Maps 43

Figure 3.1: Traditional approach vs. two-phase approach

the abstract log L⁰(e.g., XbZje, etc.). The activities in Σ\D being not involved in the denition of mapping indicate activities that are insignicant from the context of analysis and are ltered out from t during this transformation.

The denition of mapping M is facilitated by discovering common execution patterns in the log, selecting the desired context dependent patterns and den-ing abstractions out of them. Further on, abstractions can also be dened and properly controlled by the analyst using domain knowledge. The goal is to cre-ate an abstract activity that captures all sub-activities and activities belonging to a common functionality and create a sub-process that can be reached by zoom in/out feature. The selection of patterns and the creation of abstractions used to map events in logs is an interactive approach done using the Pattern Abstractions plugin [6].

2. The second phase of the approach consists of the discovery of hierarchical pro-cess maps that provide the functionality of zoom in/out in the sub-propro-cesses of previously dened abstractions. The sub-processes are created upon zooming, based on the sub-logs of each abstract activity.

Figure 3.1 presents the dierence between the traditional approach to pro-cess discovery and the two-phase approach. It is important to notice that the fuzzy model obtained using the traditional approach is spaghetti-like. The process map mined using the two-phase approach is more simple due to the fact that activities are abstracted based on the context, and sub-processes are created for these abstractions. There is the possibility to zoom into such a sub-process.

Figure 3.2: The process map of FSE for Converter Velara 8E case study

The approach to construct hierarchical process models presented in this section has been applied on the event logs of the Philips case study Converter Velara 8E

presented in Chapter 2. The resulting process map presented in Figure 3.2 represents the workow of the FSE for the case study. This process map has been created based on the event logs at command level and as we can see it is easy to comprehend in comparison to the model presented in Figure 2.16. The process map conforms with what domain experts wanted to see, activities being at the highest level with the possibility to drill down in the hierarchy and see also the sub-processes.

Moreover, in Figure 3.2 we can distinguish three type of nodes in the fuzzy map:

• Abstract nodes: are zoomable nodes that have a sub-process underneath. Ab-stract nodes have blue color and are rectangular in shape.

• Primitive nodes: are nodes that do not have sub-processes underneath. Prim-itive nodes have light brown color and are rectangular in shape.

• Cluster nodes: are abstract nodes automatically created by the Fuzzy miner.

While constructing a process map, the mining algorithm aggregates coherent groups of less signicant behaviour which are highly correlated into cluster nodes. In comparison with normal abstract nodes, cluster nodes cannot be subsumed in any other node (abstract or cluster), they are at the highest hier-archical level in the model. Cluster nodes have green color and are hexagonal in shape.