Interactive Workflow Mining

1
Interactive Workflow Mining

• Markus Hammori, Joachim Herbst, Niko Kleiner

2
1 Introduction
2 Requirements Analysis
3 Selected Concepts
3.1 MaximumRecognition Layout
3.2 Measure for completeness
4 Prototypical Implementation
5 Conclusion
3
1 Introduction
2 Requirements Analysis
3 Selected Concepts
3.1 MaximumRecognition Layout
3.2 Measure for completeness
4 Prototypical Implementation
5 Conclusion
4
Basic Workflow Mining Approach
Model
5
Basic Workflow Mining Approach
Model
6
Basic Workflow Mining Approach
Model
7
Basic Workflow Mining Approach
Model
8
Why use Workflow Mining?

• Acquisition Problem
• Reverse Engineering of Workflow Models
• Adaptation Problem

9
Why is Workflow-Mining interactive?

• Basic conditions
• Non-unique activity names ? Exponential search
  space
• Best model is a compromise between accuracy and
  model size
• Human users use soft measures as basis for a
  decision
• The Standard Workflow Mining Process

10
1 Introduction
2 Requirements Analysis
3 Selected Concepts
3.1 MaximumRecognition Layout
3.2 Measure for completeness
4 Prototypical Implementation
5 Conclusion
11
Approach for requirements analysis

• Initial situation
• InWoLvE ? non-Interactive WF Mining prototype
• ADONIS ? commercial BPMS, used for visualization
• Approaches for requirements analysis
• Systematic experiments using the combination of
  InWoLvE and ADONIS
• Evaluation of other Workflow-Mining Tools with
  respect to support for interactive usage
• Search for applicable techniques from Data-Mining

12
Interactive Requirements in the Standard
Workflow Mining Process

• Choosing the initial Parameters for InWoLvE
• Extract all helpful log information
• Visualizing and evaluating the results in ADONIS
• Close integration of user interface and mining
  tool
• Possibility to influence an ongoing calculation
• Layout algorithm featuring change resistance and
  well structured layout
• Modifying parameters for a new iteration
• Visualize all useful data associated to a model
• Choosing a result model
• Provide a measure for the completeness of a model

13
1 Introduction
2 Requirements Analysis
3 Selected Concepts
3.1 MaximumRecognition Layout
3.2 Measure for completeness
4 Prototypical Implementation
5 Conclusion
14
Layout Algorithm - Example for bad change
resistance
15
Layout Algorithm Example for bad structuring
16
Layout Algorithm

• Task Produce Layout that is resistant to small
  changes
• Problem Existing Approaches are based on
  incremental changes
• Solution MaximumRecognition Algorithm
• Static algorithm
• Solves dynamic requirements by applying a strict
  set of rules
• Exploit restrictions of graph model
• Block-structured split / join blocks
• Branches of splits are not connected
• Restricted number of predecessors / successors
  for some vertex types

17
Layout Algorithm - Explanation

• Sort the Vertices into Levels
• Modified breadth-first search
• Vertices must be positioned according to longest
  path

18
Layout Algorithm - Explanation

• Sort the Vertices into Levels
• Sort the Vertices into Columns
• Decompose the graph
• Split / join blocks
• Decision paths
• Compute the width of subgraphs
• Set vertex positions
• Sort paths according to well defined order

19
Layout Algorithm - Explanation

• Sort the Vertices into Levels
• Sort the Vertices into Columns
• Split the graph
• Compute the width of subgraphs
• Set vertex positions
• Refinements
• Draw backward edges around intermediate vertices
• Position stop vertex

20
1 Introduction
2 Requirements Analysis
3 Selected Concepts
3.1 MaximumRecognition Layout
3.2 Measure for completeness
4 Prototypical Implementation
5 Conclusion
21
Measure for the completeness of models

• Problem When can we reliably believe that we
  have seen enough examples?
• Solution Useful approach from Data Mining
• Divide examples in training-/test-set
• Estimate pred. accuracy using holdout measure
• Result gives indication of probability that
  important data was missed
• Missing Algorithm to validate a trace against a
  model

22
Validating a trace against a model - Basic Idea

• Sequential
• activity must match the node
• activities must be sequential
• Decision
• at least one path must be validated
• Split
• all required paths must be validated
• at least one path must be validated
• arbitrary order between different paths
• Entire model
• the stop vertex must be reached
• all activities in the trace must be used

A1, A2, A5, A4, A6
BPA1
BPA2
BPA3
BPA5
BPA4
BPA6
23
Validating a trace against a model - Basic Idea

• Sequential
• activity must match the node ?
• activities must be sequential ?
• Decision
• at least one path must be validated
• Split
• all required paths must be validated
• at least one path must be validated
• arbitrary order between different paths
• Entire model
• the stop vertex must be reached
• all activities in the trace must be used

A1, A2, A5, A4, A6
BPA1
BPA2
BPA3
BPA5
BPA4
BPA6
24
Validating a trace against a model - Basic Idea

• Sequential
• activity must match the node ?
• activities must be sequential ?
• Decision
• at least one path must be validated
• Split
• all required paths must be validated
• at least one path must be validated
• arbitrary order between different paths
• Entire model
• the stop vertex must be reached
• all activities in the trace must be used

A1, A2, A5, A4, A6
BPA1
BPA2
BPA3
BPA5
BPA4
BPA6
25
Validating a trace against a model - Basic Idea

• Sequential
• activity must match the node ?
• activities must be sequential
• Decision
• at least one path must be validated ?
• Split
• all required paths must be validated
• at least one path must be validated
• arbitrary order between different paths
• Entire model
• the stop vertex must be reached
• all activities in the trace must be used

A1, A2, A5, A4, A6
BPA1
BPA2
BPA3
BPA5
BPA4
BPA6
26
Validating a trace against a model - Basic Idea

• Sequential
• activity must match the node ?
• activities must be sequential ?
• Decision
• at least one path must be validated ?
• Split
• all required paths must be validated ?
• at least one path must be validated ?
• arbitrary order between different paths
• Entire model
• the stop vertex must be reached
• all activities in the trace must be used

A1, A2, A5, A4, A6
BPA1
BPA2
BPA3
BPA5
BPA4
BPA6
27
Validating a trace against a model - Basic Idea

• Sequential
• activity must match the node ?
• activities must be sequential
• Decision
• at least one path must be validated ?
• Split
• all required paths must be validated ?
• at least one path must be validated ?
• arbitrary order between different paths ?
• Entire model
• the stop vertex must be reached
• all activities in the trace must be used

A1, A2, A5, A4, A6
BPA1
BPA2
BPA3
BPA5
BPA4
BPA6
28
Validating a trace against a model - Basic Idea

• Sequential
• activity must match the node ?
• activities must be sequential ?
• Decision
• at least one path must be validated ?
• Split
• all required paths must be validated ?
• at least one path must be validated ?
• arbitrary order between different paths
• Entire model
• the stop vertex must be reached
• all activities in the trace must be used

A1, A2, A5, A4, A6
BPA1
BPA2
BPA3
BPA5
BPA4
BPA6
29
Validating a trace against a model - Basic Idea

• Sequential
• activity must match the node ?
• activities must be sequential ?
• Decision
• at least one path must be validated ?
• Split
• all required paths must be validated ?
• at least one path must be validated ?
• arbitrary order between different paths
• Entire model
• the stop vertex must be reached
• all activities in the trace must be used

A1, A2, A5, A4, A6
BPA1
BPA2
BPA3
BPA5
BPA4
BPA6
30
Validating a trace against a model - Basic Idea

• Sequential
• activity must match the node
• activities must be sequential
• Decision
• at least one path must be validated
• Split
• all required paths must be validated
• at least one path must be validated
• arbitrary order between different paths
• Entire model
• the stop vertex must be reached ?
• all activities in the trace must be used ?
• ? Validation successful

A1, A2, A5, A4, A6
BPA1
BPA2
BPA3
BPA5
BPA4
BPA6
31
Validating a trace against a model The Reality

• Simplified example
• Generally
• Non unique vertex names
• Nested concurrencies / loops
• Solutions
• Constructive approach that discards impossible
  mappings early
• Backtracking algorithm for solving concurrency
• Propagate each successful step to all branches
  (full backtracking)
• State based memory
• Visited vertices
• Depth of nesting

32
1 Introduction
2 Requirements Analysis
3 Selected Concepts
3.1 MaximumRecognition Layout
3.2 Measure for completeness
4 Prototypical Implementation
5 Conclusion
33
Prototypical Implementation

• MaximumRecognition Layout
• Measure for completeness
• Project Management Management of results and
  configuration parameters
• Post-pruning Hide paths with low probability
  after calculation stopped
• Temperature-Color-Map Colored highlighting of
  graph sections according to absolute visiting
  frequency

34
1 Introduction
2 Requirements Analysis
3 Selected Concepts
3.1 MaximumRecognition Layout
3.2 Measure for completeness
4 Prototypical Implementation
5 Conclusion
35
Conclusion

• Accomplished
• Located aspects of Workflow Mining which require
  user interaction
• Systematically extracted requirements for
  interactive Workflow Mining system
• Developed concepts to meet various requirements
• Prototypical implementation
• Future Work
• Improvement of the workflow mining algorithms
• Application in a real scenario
• Layout solution for very large models
• Diff function for locating differences between
  models
• Reliability measure based on cross-validation

36
Thank you for your attention!
37
Backup Slides
38
MaximumRecognition Layout
39
Layout Algorithms Related Work

• Simple Approach Use static Layout with Animation
• Incremental Graph Layout
• Modify Graph step by step
• Will only accept certain operation ( e.g. add
  vertex)
• ? Extract differences between two graphs, split
  into valid operations
• Foresighted Layout
• Produces perfect results
• Will use any static layout algorithm
• Needs Information about all graphs
• ? Can not be used for online layout

40
InWoLvE Explanantion of the SplitPar algorithm
41
The lifecycle Workflow-Modells
42
The Search Space
43
ProTo System setup