The document describes techniques for simplifying process models mined from event logs to make them more readable and understandable for users. It involves unfolding the mined model based on the event log to represent concurrency explicitly, then refolding and merging equivalent nodes. Implied places that do not restrict behavior can be removed. The resulting model has less complexity but the same behavior as the original model. Experimental results on benchmark logs show the techniques can significantly reduce model complexity while maintaining precision.

1. Dirk Fahland
Wil M.P. van der Aalst
Simplifying
Mined Process Models

2. Process Mining, Currently
event process mining process
log algorithm model
PAGE 1

3. Process Mining, Currently
readable
event process mining process process
log algorithm model model
PAGE 2

4. Post-Process the Model
readable
event process mining process
simplify process
log algorithm model model
can replay the entire log introduce:
post-processing can replay
operations on the the entire log
mined model
PAGE 3

5. …Based on Original Event Log
process mining process
algorithm model readable
event simplify process
log model
can replay the entire log introduce:
post-processing can replay
operations on the the entire log
mined model
PAGE 4

6. Analysis
process mining process
algorithm model readable
event simplify process
log model
discover ordering relations
 infer behavior
behavior
observed executions generalized behavior
incomplete knowledge
PAGE 5

7. Idea: Re-Adjust Generalization
process mining process
algorithm model readable
event simplify process
log model
unfold model wrt. log
model
complexity
fold, simplify,
generalize
behavior
log
PAGE 6

8. Unfold a Spaghetti-Model
PAGE 7

9. Unfold Model wrt. a Log
A
ABDA
ABCBDA
ABCBC
log C B
D
mined
process model
PAGE 8

10. Unfold Model wrt. a Log
unfold
A
A
ABDA
ABCBDA
B ABCBC
log C B
D
A D
mined
process model
PAGE 9

11. Unfold Model wrt. a Log
unfold
A
A
ABDA
ABCBDA
B ABCBC
log C B
C D
B A D
mined
D
process model
A
PAGE 10

12. Unfold Model wrt. a Log
unfold
A
A
ABDA
ABCBDA
B ABCBC
log C B
C D
B A D
mined
D
process model
C
A
PAGE 11

13. Unfold Model wrt. a Log
unfold
A
A
ABDA
ABCBDA
B ABCBC
log C B
C D
B A D
mined
D B
process model
C
B A
unfolding
wrt. the log PAGE 12

14. Represents Concurrency
unfold
A
A
AEBDA
ABECBDA
B E ABCBC
log C B E
C D
B A D
mined
D
process model
C
A
unfolding
wrt. the log PAGE 13

15. Represents Concurrency
A
AEBDA
ABECBDA
B E ABCBC
log
C D
• is a process model
B A • contains only behavior in the log
• is acyclic
C D
• represents concurrency explicitly
• labeled
(several tasks with same label)
A
unfolding
wrt. the log PAGE 14

16. Represents Concurrency
A
AEBDA
ABECBDA
B E ABCBC
log
C D
unfold
B A
fold,
simplify,
C D
generalize
A
unfolding
wrt. the log PAGE 15

17. Fold an unfolded model
A
merge equivalent nodes
B E necessary condition on
equivalent transitions
C D • same label
B A
C D
A
PAGE 16

18. Fold an unfolded model
A
merge equivalent nodes
B E necessary condition on
equivalent transitions
C D • same label
• equivalent pre-/post-places
B A
C D
A
PAGE 17

19. Fold an unfolded model
A
merge equivalent nodes
B E necessary condition on
equivalent transitions
C D • same label
• equivalent pre-/post-places
B A
various equivalences possible
(see paper for some)
C D
A
PAGE 18

20. Fold an unfolded model
A
merge equivalent nodes
B E
A
C D
C B E
B A
D
C D
A A
PAGE 19

21. Unfolding and Refolding
unfold
A
fold
A
C B E
C B E
D
D
refolded vs. original model
• less behavior
(replays the log and more)
A
• simpler structure
PAGE 20

22. Next: Simplifying and Generalizing
readable
process
simplify
simplify process
model
model
unfold
complexity
fold
simplify,
generalize
behavior
log
PAGE 21

23. Implied Places
A
implied place
• does not restrict transitions
B fold
A
remove from folded model
C D • simpler model
C B • same behavior
B A
D
various techniques to find
C D implied places
A A
PAGE 22

24. Special: Implied Places and Folding
A A
p p
A
B C D C p
fold
D B C
unfolding wrt. log
 folding may merge implied and non-implied places
 remove p: simpler model,
more behavior (generalization)
 let user decide
PAGE 23

25. Configurable Simplification
readable
process
simplify
simplify process
model
model
unfold
complexity
fold
configurable
simplify,
generalize
behavior
log
PAGE 24

26. ProM6 / Uma > www.processmining.org
PAGE 25

27. ProM6 / Uma > www.processmining.org
PAGE 26

28. ProM6 / Uma > www.processmining.org
PAGE 27

29. Experimental Results
 15 benchmark logs, 6 industrial logs
[www.promtools.org/prom5/]
PAGE 28

30. Experimental Results
 15 benchmark logs, 6 industrial logs
[www.promtools.org/prom5/]
model complexity = #arcs / #nodes
9.0
8.0
7.0
6.0
5.0
4.0
3.0
2.0
1.0
0.0
PAGE 29

31. Experimental Results
 precision: traces allowed by model and not in log
 1.0 = only log behavior allowed
 rises/falls within limits (can be controlled)
PAGE 30

32. from to
spaghetti lasagna?

33. from to less complex
spaghetti spaghetti

34. Lessons Learned
 techniques to navigate the model/behavior space
 use model and log together
 use model unfoldings
 break a rule and see what happens
unfold
model
complexity
fold
simplify,
generalize
behavior
log
PAGE 33

35. And next?
process mining process
algorithm model readable
event simplify process
log model
 process views
most simple model covering 80% of the log
 improve mining algorithms?
we showed: there is room for improvement
PAGE 34

36. Dirk Fahland
about.me/dirk.fahland
Simplifying
Mined Process Models