Graph Layout Stability in Process Mining
This web-page contains videos showing the results of my master's thesis 'Graph layout stability in process mining', which can be found here.
ProcessGold filed a patent on the algorithm I developed and a paper was written based on my work.
Abstract
Process mining allows organizations to gain insight into their internal/external process(es) and thereby make data-driven decisions. Visualization of this internal/external process, which is done using a graph layout, plays an important role in process mining. Ideally, the process visualization satisfies two properties. Firstly, it should be of high quality, i.e., the graph layout is readable and understandable, and thereby properly shows the underlying process. Secondly, since process analysis is often an iterative task of interactively filtering the process data, the mental map of the user should be preserved. The current industry standard used for process visualization, is dot, a Sugiyama-based hierarchical graph layout algorithm. Dot, however, often fails to satisfy these two properties. In this work, we present two data structures which are computed based on the process data, and are used during the graph layout computation. Additionally, we present a novel crossing minimization algorithm. Finally, we use phased animation to further improve mental map preservation. A quantitative and qualitative evaluation show that our approach is better at preserving the mental map of the user and computes layouts of high quality (compared to dot). Additionally, our approach is significantly faster than dot, especially for graphs with more than about 250 edges.
Abstract
Process mining allows organizations to gain insight into their internal/external process(es) and thereby make data-driven decisions. Visualization of this internal/external process, which is done using a graph layout, plays an important role in process mining. Ideally, the process visualization satisfies two properties. Firstly, it should be of high quality, i.e., the graph layout is readable and understandable, and thereby properly shows the underlying process. Secondly, since process analysis is often an iterative task of interactively filtering the process data, the mental map of the user should be preserved. The current industry standard used for process visualization, is dot, a Sugiyama-based hierarchical graph layout algorithm. Dot, however, often fails to satisfy these two properties. In this work, we present two data structures which are computed based on the process data, and are used during the graph layout computation. Additionally, we present a novel crossing minimization algorithm. Finally, we use phased animation to further improve mental map preservation. A quantitative and qualitative evaluation show that our approach is better at preserving the mental map of the user and computes layouts of high quality (compared to dot). Additionally, our approach is significantly faster than dot, especially for graphs with more than about 250 edges.