Differences Between Business Process Models Abel Armas Cervantes 1 Similar or not 2 3 Diagnosis statements In model 1 C occurs at most once whereas it occurs 01 or more ID: 476870
Download Presentation The PPT/PDF document "Diagnosing Behavioral" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Diagnosing BehavioralDifferences Between Business Process Models
Abel Armas Cervantes
1Slide2
Similar … or not?
2Slide3
3Diagnosis statements
In model
1, C
occurs
at most
once
;
whereas it
occurs 0,1 or more
times
in
model
2
In model 1, B and
C
are performed in
parallel
;
whereas in model 2,
task B
precedes
task C
Sequence
Exclusiveness
ParallelismSlide4
Comprehensibility
Accuracy
Desiderata
4Slide5
I
A
B
C
D
O
I
+
→
→
→
→
→
A
←
+
→
→
→
→
B
←
←
+
||
→
→
C
←
←
||
+
→
→
D
←
←
←
←
+
→
O
←
←
←
←
←
+
Proposition 4.11:
BPs can be highly inaccurate behavioral representations
Cleaveland
et al.
Existing approaches and their limitations
5
Behavioral Profiles
No standard notion of equivalenceSlide6
Overview of the proposed solution
6
DiagnosisSlide7
Behavioral abstractionPrime Event Structures
7Slide8
Branching process of a Petri net
8Slide9
Events and two behavioral relations: causality and conflict
Proposition 5.12
: The restriction of a PES to its observable behavior is equivalent to the original (
w.r.t
. visible
pomset
equivalece
)
Prime Event Structures
9Slide10
Cyclic Petri netsBranching process is infiniteComplete unfolding prefix
10Slide11
Causally-complete unfolding prefix
Proposition 5.3
: For any pair of causally-dependent events in the unfolding of a net, there is a representative of such events in the prefix that are causally-
dependent.
11
RepetitionSlide12
Recap12
Diagnosis
Isomorphism
⇓
Equivalence
Behavioral relations
⇒
Workflow patterns
Causality
Sequence
Conflict
Exclusiveness
Concurrency
ParallelismSlide13
ComparisonPartial Synchronized Product (PSP) – Visible pomset equivalence
13Slide14
Partial Synchronized Product
14Slide15
Partial Synchronized Product (2)
15Slide16
VerbalizationDiagnosis as natural language statements
16Slide17
Verbalization of differencesMismatching behavioral relationsRepetition
Not matched events
17
“In M1, there is a state after
<context>
where
<verbalization
of relation
1>
, whereas in the matching state in M2,
<verbalization
for relation
2>
”
“In
M1,
<activity> <verbalization
of multiplicity in
M1>
, whereas in M2, it
<verbalization
of activity multiplicity in M2
>
”
“In M1, there is a state after
<context>
where
<activity>
always occurs, whereas it cannot occur in the matching state in M2”Slide18
Reduction of diagnosis
18Slide19
ReductionFolding of event structures – History preserving bisimulation
19Slide20
Theorem 6.10 and 6.18: The folding operation preserves the behavior with respect to history preserving bisimulation
Folding of event structures
20Slide21
Canonicity in AES and FES21Slide22
2
1
Deterministic folding
Lexicographic order
Number of events to fold
Canonical numbering
1
2
3
4
5
Proposition 6.20
: The deterministic folding of a pair of isomorphic event structures produces isomorphic event structures.
22
1
2
2
1
1
2
3Slide23
Recap
23
Diagnosis
Folding of AES or FESSlide24
BPdiff
24Slide25
Summary of the proposed technique
Behavioral abstraction
Comparison
Verbalization
25
Natural language statements
Visible
pomset
equivalence
Hp-bisimulation
Causally-complete
Reduction
Comprehensibility
AccuracySlide26
Future workApply the proposed techniques in the context of process mining
Business process discovery Conformance checking
Improve the efficiency of the current technique
26Slide27
27Slide28
VerbalizationDiagnosis as natural language statements
28Slide29
Mismatching behavioral relations
In model 1,
there is a state after A where B has to occur before activity
C, whereas in the matching state in model 2, B and C are parallel
Verbalization of differences (1)
29Slide30
RepetitionVerbalization of differences (2)30
In
model 1,
C occurs at most once, whereas in
model 2,
it occurs
any number of times, but at least onceSlide31
Not matched eventsVerbalization of differences (3)31
In
model 2,
there is a state after C
where D
always occurs, whereas it cannot occur in the matching state in
model 1Slide32
Experiments – BIT IBM3 libraries: 152, 184, 16 process models
Library
# models
# elements (
avg
)
# events
(
avg
)
Computation time (sec)
PES
AES
PES
Canonical labeling
Minimal AES
folding
A
152
12.3
18.27
16.77
0.06
0.02
0.07
B3
184
9.1
9.07
8.53
0.02
0.01
0
C
15
17.3
57.93
36.27
52.61
0.36
36.76
32Slide33
Experiments – South and West Australia3 pairs of model
Model
#
elements# events
Computation time (sec)
PES
AES
PES
Canonical labeling
Minimal AES
folding
SA 1
37
13
13
0.25
0.05
0.01
SA 2
47
80
80
1.34
0.14
0.44
SA 3
36
5230
1.19
0.120.22WA 1
2814140.090.03
0WA 2
5080
800.950.081.33WA 3
3146
230.320.05
0.5433Slide34
Experiments – South and West Australia (2)
Model 1
Model 2Computation time (sec)
# Differences
PESAES
PES
AES
SA 1
WA 1
0.16
0.29
23
23
SA 2
WA 2
2.79
5.17
6
6SA 3
WA 3
98.56
145.52
104
80
34Slide35
Hp-bisimilar reduction
AES
FES
35