The abundance of event data in today’s information systems makes it possible to “confront” process models with the actual observed behavior. Process mining techniques use event logs to discover process models that describe the observed behavior, and to check conformance of process models by diagnosing deviations between models and reality. In many situations, it is desirable to mediate between a preexisting model and observed behavior. Hence, we would like to repair the model while improving the correspondence between model and log as much as possible. The approach presented in this article assigns predefined costs to repair actions (allowing inserting or skipping of activities). Given a maximum degree of change, we search for models that are optimal in terms of fitness—that is, the fraction of behavior in the log not possible according to the model is minimized. To compute fitness, we need to align the model and log, which can be time consuming. Hence, finding an optimal repair may be intractable. We propose different alternative approaches to speed up repair. The number of alignment computations can be reduced dramatically while still returning near-optimal repairs. The different approaches have been implemented using the process mining framework ProM and evaluated using real-life logs.
Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...
Impact-Driven Process Model Repair
1. 39th International Conference on Software Engineering
May 26, 2017 | Buenos Aires, Argentina
Impact-Driven Process Model Repair
(Journal-First ACM TOSEM Paper, DOI: 10.1145/2980764)
1 Queensland University of Technology, Brisbane, Australia
2 Eindhoven University of Technology, Eindhoven, The Netherlands
* Presenting/corresponding author
Artem Polyvyanyy1,*, Wil M. P. van der Aalst 2,1, Arthur H. M. ter Hofstede1,2, and Moe T. Wynn1
3. Conformance Checking
Artem Polyvyanyy | May 26, 2017 | Impact-Driven Process Model Repair | ICSE 2017 3
a
p1
t1 p2
b c
d e
p3
p4
p5
t2 t3
t4 t5
a b c
a b c
t1 t2 t3
γ1 =
a d e
a d e
t1 t4 t5
γ2 =
perfect alignments
a b x
a b
t1 t2
γ3 =
real-world traces
designed traces
e ≫
c
t3
≫ ≫
a b x
a b
t1 t2
γ4 =
e≫
c
t3
≫ ≫
a b
a
t1 t4
γ5 = ≫d
≫ x e
e
t5
≫
0 0 1 1 2 0 0 1 2 1 0 3 1 1 0+ + + + + + + + == =4 4 5+ + + +
move on trace move on modelsynchronous move
cost(γ3)=cost(γ4)=4 and cost(γ5)=5; γ3 and γ4 are optimal alignments!
4. Process Model Repair
Artem Polyvyanyy | May 26, 2017 | Impact-Driven Process Model Repair | ICSE 2017 4
a
p1
t1 p2
b c
d e
p3
p4
p5
t2 t3
t4 t5
a b x
a b
t1 t2
γ4 =
e≫
c
t3
≫ ≫
0 0 1 2 1
γ6 =
a e
a e
t1 t5
≫
d
t4
0 3 0
= 7R=({x},{d})
insert skip
x t6
t7
γ7 =
a
a
t1
0
b
b
t2
0
x
x
t6
0
c
t3
2
≫
1
e
≫ γ9 =
a
a
t1
0
≫
t7
0
b
≫
1
≫
1
x
0
e
e
t5
γ8 =
a
a
t1
0
≫
t7
0 0
e
e
t5
+ + + + = 3 + + + + = 2 + + = 0
= 2
Note the problem with x!
ττ
γ8 and γ9 are optimal alignments!
5. Towards Optimal Repair
Artem Polyvyanyy | May 26, 2017 | Impact-Driven Process Model Repair | ICSE 2017 5
R=({x},{d})
a
p1
t1 p2
b c
d e
p3
p4
p5
t2 t3
t4 t5
γ10 =
a
a
t1
0
d
t4
0
≫
1
b
≫
0
x
≫
e
e
t5
0
= 1
=
a e
a e
t1 t5
≫
d
t4
0 0 0
γ11
simulating repair via “free” moves
a
p1
t1 p2
b c
d e
p3
p4
p5
t2 t3
t4 t5
t7
x t6
= 1
γ12 =
a
a
t1
0
≫
t7
0
b
≫
1 0
e
e
t5
+ + + + = 1
x
x
t6
0
γ8 =
a
a
t1
0
≫
t7
0 0
e
e
t5
+ + = 0
ττ
γ8 and γ12 are optimal alignments!
Note no problem with x!
6. Optimal Repair (Theorem 5.3)
Artem Polyvyanyy | May 26, 2017 | Impact-Driven Process Model Repair | ICSE 2017 6
There exist optimal alignments between
the traces and the repaired model which:
(i) Demonstrate that the traces fit the
repaired model “at least as good” as
they fit the original model;
(ii) “Fulfill” the repair recommendation,
e.g., γ8 and γ12 contain no “bad”
moves (x,≫) and (≫,d).
a
p1
t1 p2
b c
d e
p3
p4
p5
t2 t3
t4 t5
a
p1
t1 p2
b c
d e
p3
p4
p5
t2 t3
t4 t5
t7
x t6
= 1
γ12 =
a
a
t1
0
≫
t7
0
b
≫
1 0
e
e
t5
+ + + + = 1
x
x
t6
0
γ8 =
a
a
t1
0
≫
t7
0 0
e
e
t5
+ + = 0
ττ
R=({x},{d})
γ8 and γ12 are optimal alignments!
Note no problem with x!
7. One More Example
Artem Polyvyanyy | May 26, 2017 | Impact-Driven Process Model Repair | ICSE 2017 7
a
p1
c
d
b f
g
h
e
t1
t2
t3
t4
t5
t6
t7
t8
t9
t10
t11
p2
p3
p4 p5
p6
p7
p8
p9
p10 p11
10 x
9 x
6 x
2 x
9 x
7 x
2 x
= 120
All deviations
cost one Peso!
8. A Possible Repair
Artem Polyvyanyy | May 26, 2017 | Impact-Driven Process Model Repair | ICSE 2017 8
R = ({f,e,x},{c,f,g})
= 47
insert skip
a
p1
c
d
b
f
g
he
t1
t2
t3
t4
t5
t6
t7
t8
t9
t10
t11
p2 p3
p4 p5 p6
p7
p8
p9
p10 p11
t12
t13
t14
x
f
e
t15
t16 t17
9. Another Possible Repair
Artem Polyvyanyy | May 26, 2017 | Impact-Driven Process Model Repair | ICSE 2017 9
R = ({f,e},{g,d,e,c})
= 33
insert skip
a
p1
c
d
b
f
g
h
e
t1
t2
t3
t4
t5
t6
t7
t8
t9
t10
t11
p2
p3
p4
p5
p6
p7
p8
p9
p10
p11
t13
t15
t16
t17
t18
e f
t12 t13
et14
10. Towards Optimal Repair Recommendation
Artem Polyvyanyy | May 26, 2017 | Impact-Driven Process Model Repair | ICSE 2017 10
R = ({f,x},{d,e,c,h})
= 25
insert skip
a
p1
c
d
b
f
g
he
t1
t2
t3
t4
t5
t6
t7
t8
t9
t10
t11
p2
p3
p4
p5
p6
p7
p8
p9
p10
p11
t13
t12
f
x
t14
f
t15
t16
t17
t18
costs of repair
recommendation
implemented
heuristics
numbers of discovered
recommendations
numbers of alignment
computations deviation costs
11. Conclusion and Future Work
Artem Polyvyanyy | May 26, 2017 | Impact-Driven Process Model Repair | ICSE 2017 11
In this work, we …
Defined, and proposed a solution to, the optimal process model repair problem
Defined, and proposed several (approximate) solutions to, the optimal repair
recommendation problem
Developed a publicly available tool that implements all the invented techniques,
https://svn.win.tue.nl/repos/prom/Packages/SelectivePRepair/Trunk/release/
Reported on lessons learned from experiments with real-world data, e.g.,
for small repairs proposed heuristics perform just as well as brute-force
Possible avenues for future work include:
Case studies to better understand requirements of process model repair
Aesthetic process model repair
Repair heuristics with guarantees
Repairs based on other conformance data structures
Repairs that go beyond trace fitness
Let’s work together:
Engage into research collaborations with us
Write your master and/or PhD thesis with us (scholarships available)
Editor's Notes
All synchronous moves have zero costs.
All non-synchronous moves not shown on this slide have the cost of one.
The repaired model fits traces “at least as good” as the original model
The repaired model “fulfills” the repair recommendation; recommendation ({x},{d}) in the running example
There exist optimal alignments that do not contain “bad” moves; moves (x,≫) and (≫,d) in the running example