in Multidimensional Process Mining C. Cordes, T. Vogelgesang, H.-J. - - PowerPoint PPT Presentation

A Generic Approach for Calculating and Visualizing Differences between Process Models in Multidimensional Process Mining

• C. Cordes, T. Vogelgesang, H.-J. Appelrath

TAProViz 2014

Process Mining: An introduction

16.09.2014 Thomas Vogelgesang 2

• Automatic generation

and analysis of process models

• Based on event logs

– Recorded during execution of process – Sequences of events – Process instances – Arbitrary attributes

According to: v. d. Aalst et al.: Process Mining Manifesto (2011)

Multidimensional process mining: Motivation

16.09.2014 Thomas Vogelgesang 3

• Process behavior often influenced by many factors
• E.g. healthcare processes

– Individuality of patients

• Age, sex, allergies, …

– Individuality of institution

• Equipment of hospital, experience of staff, …
• Process mining typically considers entire event log

– One model for all patients – Influence of properties gets blurred

Multidimensional process mining: The idea

16.09.2014 Thomas Vogelgesang 4

• Define groups of patients with similar properties
• Separate process mining for each group

Process model differences in MPM: Concept

16.09.2014 Thomas Vogelgesang 5

Process mining uses different modeling languages Diff algorithm independent of specific modeling language No restriction to specific diff algorithm Visualization in same modeling language as input Various ways to indicate differences

Intermediate model: TGraphs

16.09.2014 Thomas Vogelgesang 6

• Intermediate model based on TGraphs [1]

– Consists of linked nodes – Nodes and edges are mapped to nodes of TGraph – Annotated with key-value-pairs for additional information

[1] J. Ebert and A. Franzke. A Declarative Approach to Graph Based Modeling. In Graphtheoretic Concepts in Computer Science. Springer, 1995.

Difference calculation on intermediate models

16.09.2014 Thomas Vogelgesang 7

• Arbitrary difference calculation algorithms

– Interchangeable through interface

• Input: TGraph models
• Output: Tgraph-based difference model

– Merged from input models – Nodes marked as added, deleted, changed, unchanged

Difference visualization: View model

16.09.2014 Thomas Vogelgesang 8

• Same structure as TGraph
• Consists of different representation nodes

– Language-specific – E.g. transitions, places and arcs for Petri nets

• Representation type defined in language mapping
• Visualization logic contained in representation nodes

– Interchangeable and customizable – Allows visualization of additional information

Differencing multiple models

16.09.2014 Thomas Vogelgesang 9

• Approach (theoretically) allows to compare an

arbitrary number of models

– One model needs to be selected as reference model – All differences are related to the reference model

• In practice only applicable to very few models

– Exponential growth of possible diff states of a single node – Experience: not useful for more than three models

Prototype

16.09.2014 Thomas Vogelgesang 10

• Integrated into tool for

MPM (PMCube Explorer)

– Based on .NET

• Different modeling

languages

– Petri nets, causal nets, Process tree (BPMN-like visualization)

• Different diff algorithms

– (Extended) Snapshot diff – Block-structured diff [2,3]

[2] Li, Reichert, Wombacher. Representing Block-structured Process Models as Order Matrices: Basic Concepts, Formal Properties,

• Algorithms. Report. (2009)

[3] Li, Reichert, Wombacher. On Measuring Process Model Similarity Based on High-Level Change Operations. In ER'08. (2008).

Future work

16.09.2014 Thomas Vogelgesang 11

• Conceptually

– User-friendly way to visualize differences for more than two models – Handling of artificial nodes (e.g. silent transitions) in difference calculation

• Implementation

– Improving layout algorithm – Improving maintainability of visualization by using advanced .NET concepts

Conclusions

16.09.2014 Thomas Vogelgesang 12

• Generic concept for visualizing differences in MPM
• TGraphs as intermediate model

– Supporting arbitrary modeling languages

• Language-independent difference calculation

– Multiple algorithms possible

• View model with representation nodes

– Each node type defines its own visualization – Allows for visualization of additional information

• Highly customizable

– Diff algorithms, layout algorithms, view models

Questions?

16.09.2014 Thomas Vogelgesang 13

Thank ank you for your att ttention tion!