Once loaded in databases, the question appears what can the system do with the event data. First, the system benefits from the multidimensional structure of the OLAP cube. In that sense, inspecting different dimensions of the cube is possible. Moreover, the system supports a set of
(a) Dice filtering. Five elements are se- lected on the EVENT conceptEXT name dimension.
(b) Dice filtering result. While the event log corresponding to P C has 33 events, the event log corresponding to P Cdiced has
only 14 events.
Figure 5.3: Dice operation.
basic OLAP operations, e.g., slice, dice, drill-down, roll-up and pivoting. Filters can be created that would slice or dice the cube in various way. Default filters exist for drill-down and roll-up operations that can be applied at request on specific chosen dimensions. Each filter is stored for further use, unless not explicitly deleted. Not only can the event data in the cube filtered, it can also be visualized from different perspectives. This functionality is offered by employing the pivoting operation.
5.4.1
Dice & Slice
A dice operation is realized when multiple members are selected for one or more dimensions. Given a process cube P C, the result of a dice is a subcube P Cdicedfor which only a subset of members are selected on particular dimensions, and for the rest is the same with the initial cube.
Figure 5.3a shows a dice filter applied on the EVENT conceptEXT name dimension. With dice, multiple elements of a dimension can be selected. In Figure 5.3a there are five task names selected and the rest of the elements are just discarded for the EVENT conceptEXT name dimension. The result of the dice operation is shown in Figure 5.3b. From 33 events present in the event log corresponding to the process cube P C, only 14 are considered for the P Cdiced. The number of cases remains the same.
A dice operation can influence more than one dimension. For example, together with the filter on EVENT conceptEXT name dimension, a subset of timestamps can be selected on the EVENT TIME timeEXT timestamp 1. A dice operation allows the selection of any element of the time hierarchy. For example, one can select year 2012 and 2013 out of a set of years containing 2010, 2011, 2012 and 2013. The month dimension can also be considered for dice. For instance, selecting the 2012F eb month in 2012 is also a dice, since it contains the following set of elements: 2012F ebM on, the 2012F ebW ed and the 2012F ebT hu.
For dimensions with numerical members, a dice filter can be created, by selecting a certain
1In the dimension name, the TIME tag is used to recognize a dimension corresponding to a time at-
tribute. Other examples of such dimensions are: EVENT TIME dueDate, EVENT TIME plannedDate, EVENT TIME createdDate
(a) Slice filtering. Only a single event name, 01 HOOFD 060 is selected on the EVENT conceptEXT name dimension.
(b) Slice filtering result. While the event log corresponding to P C has 33 events, the event log corresponding to P Cslicedhas
only 2 events.
Figure 5.4: Slice operation.
range. For example, for the SUMLeges dimension, all the events with SUMLeges between 100.5 and 500.2 can be selected.
The slice operation is a particular type of dice. That is, a slice is performed when only a single member of one dimension is selected and the other members corresponding to the dimension are filtered out. Given a process cube P C, the result of a slice is a subcube P Csliced with the same dimensions as the cube P C, except for one, which has just a single member selected of the initial set of the dimension members.
Figure 5.4a shows a slice filter applied on the EVENT conceptEXT name dimension. From all the elements of this dimension, only 01 HOOFD 060 is selected. After creation, the slice filter is saved and, at request, is applied on the event data of the process cube. That is, only events with the event name 01 HOOFD 060 are considered for the new P Cslicedcube. Figure 5.4b depicts the slice result on the process cube. In the top window, a Log Dialog shows information on the initial event log. Note that the entire event log contains 4 cases and 33 events. The bottom window illustrates a Log Dialog containing information on the event log created after slicing. The new event log contains only 2 cases and 2 events. Consequently, there are only 2 events with the name 01 HOOFD 060 and they belong to 2 different cases.
For a dimension with time attributes, the slice can be performed while selecting a leaf member, situated at the day of week hierarchical level. For example, for a timestamp dimension containing 2012 at the year hierarchical level and 2012F eb at the month level and 2012F ebT ue at the day of week level, a slice can be executed by selecting the 2012F ebT ue element. Note that such a slice filters out all the events except for the ones that occurred on Tuesday in the February month of 2012, and not on all Tuesdays of the 2012 year or on all Tuesdays, in general.
5.4.2
Pivoting
The subcubes obtained after slice and dice operations can be visualized. In this project, the traditional 2D visualization is considered for the process cube visualization. As such, only two dimensions of the process cube can be visualized simultaneously. This is possible through the table of visualization. The rows of a table of visualization contain two dimensions of the process
cube and also the corresponding filters created by the user. Even though based on the elements of two process cube dimensions, the dimensions of visualization are usually not identical with the former. The main difference is that their elements can be both results of filtering and elements of different hierarchical levels. In that sense, two visualization neighbor-cells can contain overlapping data, while this is never the case for two neighbor-cells of the process cube.
The restriction of visualizing only two dimensions at a time has no influence on which two dimensions to select. That is, any combination is possible and any of the two dimensions can be substituted with a new PC dimension, at any time. By swapping from one dimension to another, the visualization perspective of the P C cube changes. This operation is known as pivoting or the rotation operation.
Figure 5.5: The result of the pivoting operation. Rotation is obtained by replacing the concept names dimension with the timestamp dimension and the SUMLeges is replaced by the concept names dimension.
Figure 5.5 shows the effect of the pivoting operation on the visualization table. In the visualiza- tion table from the top of the image, the SUMLeges and the event names are the two dimensions of visualization. In the second table of visualization, the same process cube is visualized through the event names and the timestamp dimensions. Also, while the event names was initially on the x axis, in the second table, it is changed on the y axis.
5.4.3
Drill-down & Roll-up
The drill-down operation is realized by unfolding a member situated on a hierarchically superior position in a set of members corresponding to a hierarchical level lower.
Figure 5.6 shows a table of visualization with one dimension corresponding to the timestamp and another dimension corresponding to the event name. Elements of the timestamp dimension can be selected from a hierarchy. For example, the 2012 member is selected and a drill-down operation is performed on it. As in the time hierarchy, months follow years, all the months corresponding to year 2012 are shown. Based on the definition of drill-down from Section 2.3.1, the children of 2012 are added to the timestamp dimension of the table of visualization and the 2012 element is removed. In our project, we keep also the 2012 element, because it is useful to compare process mining results corresponding to elements on different hierarchical levels, e.g. the process of 2012 with the process of 2012M ar.
Figure 5.6: Drill-down operation on the timestamp dimension. Year 2012 is drilled-down to its months.
The roll-up operation is realized by folding certain members of a dimension into one member, which is hierarchically superior.
Figure 5.7: Roll-up operation on the timestamp dimension. The months corresponding to year 2012 are folded back.
Figure 5.7 shows a table of visualization corresponding to the same timestamp and event name dimensions. Based on the definition of roll-up from Section 2.3.1, the children of 2012 are removed from the timestamp dimension of the table of visualization and the 2012 element is added. In our project, there is no need to add the 2012 element, as it is already present from the drill-down operation.