wil m.p. van der aalst simplifying mined process modelsdfahland/presentations/... · simplifying...
TRANSCRIPT
Simplifying
Mined Process Models
Dirk Fahland
Wil M.P. van der Aalst
PAGE 1
Process Mining, Currently
event
log
process mining
algorithmprocess
model
PAGE 2
Process Mining, Currently
event
log
process mining
algorithmprocess
model
readable
process
model
PAGE 3
Post-Process the Model
event
log
process mining
algorithmprocess
model
readable
process
modelsimplify
introduce:
post-processing
operations on the
mined model
can replay the entire log
can replay
the entire log
PAGE 4
…Based on Original Event Log
event
log
process mining
algorithmprocess
model readable
process
modelsimplify
introduce:
post-processing
operations on the
mined model
can replay the entire log
can replay
the entire log
PAGE 5
Analysis
event
log
process mining
algorithmprocess
model readable
process
modelsimplify
behavior
observed executions
incomplete knowledge
generalized behavior
discover ordering relations
infer behavior
simplify
PAGE 6
Idea: Re-Adjust Generalization
event
log
process mining
algorithmprocess
model readable
process
model
behavior
model
complexity
log
simplify
unfold model wrt. log
fold, simplify,
generalize
PAGE 7
Unfold a Spaghetti-Model
PAGE 8
Unfold Model wrt. a Log
A
BC
D
log
mined
process model
ABDA
ABCBDAABCBC
PAGE 9
Unfold Model wrt. a Log
A
BC
D
log
unfold
mined
process model
ABDA
ABCBDA
A
B
D
A
ABCBC
PAGE 10
Unfold Model wrt. a Log
A
BC
D
log
unfold
mined
process model
ABDA
ABCBDA
A
B
D
A
C
B
D
A
ABCBC
PAGE 11
Unfold Model wrt. a Log
A
BC
D
log
unfold
mined
process model
ABDA
ABCBDA
A
B
D
A
C
B
D
A
C
ABCBC
B
B
B
unfolding
wrt. the logPAGE 12
Unfold Model wrt. a Log
A
BC
D
log
unfold
mined
process model
ABDA
ABCBDA
A
B
D
A
C
B
D
A
C
ABCBC
unfolding
wrt. the logPAGE 13
Represents Concurrency
A
BC
D
log
unfold
mined
process model
AEBDA
ABECBDA
A
B
D
A
C
B
D
A
E
EABCBC
C
unfolding
wrt. the logPAGE 14
Represents Concurrency
log
AEBDA
ABECBDA
A
B
D
A
C
B
D
A
E
• is a process model
• contains only behavior in the log
• is acyclic
• represents concurrency explicitly
• labeled(several tasks with same label)
ABCBC
C
unfolding
wrt. the logPAGE 15
Represents Concurrency
log
AEBDA
ABECBDA
A
B
D
A
C
B
D
A
E
unfold
fold,
simplify,
generalize
ABCBC
C
PAGE 16
Fold an unfolded model
A
B
D
A
C
B
D
A
E
merge equivalent nodes
C
necessary condition on
equivalent transitions
• same label
PAGE 17
Fold an unfolded model
A
B
D
A
C
B
D
A
E necessary condition on
equivalent transitions
• same label
• equivalent pre-/post-places
merge equivalent nodes
C
PAGE 18
Fold an unfolded model
A
B
D
A
C
B
D
A
E
C
various equivalences possible
(see paper for some)
merge equivalent nodes
necessary condition on
equivalent transitions
• same label
• equivalent pre-/post-places
PAGE 19
Fold an unfolded model
A
B
D
A
C
B
D
A
E
C
A
BC
D
E
A
merge equivalent nodes
PAGE 20
Unfolding and Refolding
A
BC
D
E
A
A
BC
D
E
unfold
fold
refolded vs. original model
• less behavior
(replays the log and more)
• simpler structure
simplify
PAGE 21
Next: Simplifying and Generalizing
process
model
readable
process
model
behavior
complexity
log
simplify
unfold
simplify,
generalize
fold
PAGE 22
Implied Places
A
B
D
A
C
B
D
A
C
A
BC
D
A
fold
implied place
• does not restrict transitions
remove from folded model
• simpler model
• same behavior
various techniques to find
implied places
PAGE 23
Special: Implied Places and Folding
folding may merge implied and non-implied places
A
B
A
D
unfolding wrt. log
C C
remove p: simpler model,
more behavior (generalization)
let user decide
p
fold
A
BD C
p
p
PAGE 24
Configurable Simplification
process
model
readable
process
model
behavior
complexity
log
unfold
simplify,
generalize
fold
configurable
simplifysimplify
PAGE 25
ProM6 / Uma > www.processmining.org
PAGE 26
ProM6 / Uma > www.processmining.org
PAGE 27
ProM6 / Uma > www.processmining.org
15 benchmark logs, 6 industrial logs[www.promtools.org/prom5/]
PAGE 28
Experimental Results
15 benchmark logs, 6 industrial logs[www.promtools.org/prom5/]
model complexity = #arcs / #nodes
PAGE 29
Experimental Results
9.0
8.0
7.0
6.0
5.0
4.0
3.0
2.0
1.0
0.0
precision: traces allowed by model and not in log
1.0 = only log behavior allowed
PAGE 30
Experimental Results
rises/falls within limits (can be controlled)
from
spaghetti
to
lasagna?
from
spaghetti
to less complex
spaghetti
techniques to navigate the model/behavior space
use model and log together
use model unfoldings
break a rule and see what happens
PAGE 33
Lessons Learned
behavior
model
complexity
log
unfold
simplify,
generalize
fold
PAGE 34
And next?
simplifyevent
log
process mining
algorithmprocess
model readable
process
modelsimplify
process views
most simple model covering 80% of the log
improve mining algorithms?
we showed: there is room for improvement
Simplifying
Mined Process Models
Dirk Fahland
about.me/dirk.fahland