Visual Analysis of Business Process Execution

This master project focuses on devising and developing a tool to allow for analytical reasoning of the past execution of business processes. The process of reasoning will be supported by interactive graphical interfaces that provide several visual representations of the process execution and data. In particular, the tool is aimed to exploit the map metaphor. Each visual representation is represented by a map, a simple flat image, which may be a geographic map(e.g., a map of a factory or an office), a Gantt chart, an organizational diagram, social networks and several more. Dots represent the performed activities and are placed in meaningful positions: e.g., on a geographic map activities are placed where they were executed or on an organization diagram the role of the activity executor. The tool will rely on the concept of “activities’ replay”. It consists in analysing the execution log and, thus rebuilding what occurred in the past. In this way, activities can be positioned as dots on the maps. Since during the activities’ replay, activities are started, carried on and out, dots are continuously appearing and disappears. That results in building a movie where each frame represents a visual representation in a certain moment in time. These frames are then merged together in order to obtain a movie. By playing the movie, process analysts becomes aware of the past executions and able to find issues and bottlenecks. Since the amount of data visualized can be by far huge, the tool should be able with an end-user support to aggregate some pieces of information and filter out what has little relevancy.

Process-aware information systems, ranging from generic workflow systems to dedicated enterprise information systems, use work lists to offer so-called work items to users. Process-Aware Information Systems (PAISs) are frequently applied in a variety intra- and inter-organizational settings, e.g., as supporting systems to manage bank loans, insurances as well as in hospital for health care or in emergency management to deal with the aftermath of calamities.

A basic function of Process-aware Information Systems (PAISs) is to offer work to resources. The elementary pieces of work are called work items, e.g. “Approve travel request XYZ1234”. Work items are offered to users via so-called work-list handlers, which take care of work distribution and authorization issues. Typically, PAISs use a so-called “pull mechanism”, i.e. work is offered to all qualifying resources, which are relatively free to pick the next to work on. Typically work items are chosen, thus balancing the resources’ personal needs and the overall benefit of the organizations for which they work.

The aim of this master project is to devise a tool that allows process analysts to evaluate the past execution of process instances. The ultimate goal of the tool is to enable for finding issues, bottlenecks, and other structural delays and inefficiencies. Large-size companies are always confronted with several instances of many different processes that are running at the same time. For instance, let us consider a bank. Usually a bank offers several services and serves thousands of customers (e.g., different types of loans and investments). Every service is managed through a diverse process schema and each customer is associated to a different process instance.

Therefore, the amount of logging data produced is enormously vast. While purely automatic or purely visual analysis methods were developed in the last decades, the complex nature of the analysis of such process log data makes it indispensable to include humans in these evaluations. The tool will allow decision makers to combine their flexibility, creativity, and background knowledge with the enormous amount of logging data to gain insight into the complex problem to evaluate a posteriori the processes’ execution.

The tool is meant to exploit the so-called “map metaphor” that has been already used in a previous work, and that is proven to be valuable (see [1]). In particular, the analysis will be supported by interactive graphical interfaces that provide several visual representations of the process execution and data.

Each visual representation is represented by a map, which is a simple image, such as a JPG or PNG image. Maps can be of several different natures, such as geographic map, a Gantt chart, an organizational diagram, social networks and several more. These visual representations are meant to show the different perspectives behind a business process. Process tasks are visualized on these maps as dots, which are placed in meaningful positions. For instance, if the map shows the chart of the different offices, dots are placed on top of the office where the corresponding tasks need to be carried on. If the map shows the UML Activity Diagram of the process, work items are placed on top of the activities they correspond to.

Figure 1 shows a concrete example of the map: dots are placed on a city map and represent where the corresponding activities need to be executed. Whenever dots overlap in the map, they are merged in an aggregate dot, whose size increases logarithmically with the number of dots are merged. The figure shows that certain dots are also colored; the color gives further information about the activity status, such as the level of urgency, whether it is started or not, or the supposed level of ability of the process participant to execute it. Figure 2 shows a second example of a “timeline map”. While the position wrt. x-axis gives information about how urgent a certain activity is. The scale gives time information when the activity is going to expire (where expiration can be considered as a kind of failure). Hence, the closer the activity is to the left border, the earlier the activity is going to expire.

The tool to be devised for this project aims at enabling the analysis of log. If a given map is considered and a certain point in time is taken into account, the tool should extract the activities active at that time and, starting from that, compute the configuration of dots onto the map (i.e. the dots’ position taken on the map). If the temporal horizon of past process execution is sampled at fixed (and configurable) rate, different points in time are extracted and, for each of them, a different map configuration is obtained. The different computed map configurations can be considered as a “frame”, which, when merged together, lead to obtain a video. Please note that a video will be edited for each map (i.e. for each available perspective).

Through such videos, process analists can evaluate the performance of past process instances from different perspectives. The videos should also be prevented from showing irrelevant information, whereas the most significant should be somehow highlighted. Videos can be played at different speeds and users are allowed to switch from one to another video at any time. In general, a process instance execution can last days, even months. The sampling cannot be made too frequently since, otherwise, the video would be too long. On the other hand, a too low sampling rate is not significant, either, since too many events would be ignored. A possible solution is to choose a certain sampling rate and, later, to merge subsequent frame into a single “composite” one. The composite needs to summarize the information of all the frames that are merged. As far as this point, there are issues to take into account:

• In general, dots can change their positions in subsequent frames. When we merge them, dots should be placed, thus somehow considering all the different positions taken in the frames that are merged.
• Subsequent frames can have different sets of dots: when merging, a dot might have been presented in X% of the frame that are merged. The open issue is about how to visualize such a percentage, thus highlighting more the dots that are “more present”.

[1] Massimiliano de Leoni, Wil M. P. van der Aalst, Arthur H. M. ter Hofstede: Visual Support for Work Assignment in Process-Aware Information Systems. Proceedings of the BPM’08 Conference, pages 67-83.