user guided discovery of declarative process models
DESCRIPTION
Discovering business rules from event logsTRANSCRIPT
User-Guided Discovery of Declarative Process Models
Fabrizio Maria Maggi, Arjan Mooij,
Wil van der Aalst
Environment with a lot of variability
Department of Mathematics and Computer Science PAGE 223-09-14
Environment with a lot of variability
Department of Mathematics and Computer Science PAGE 323-09-14
Environment with a lot of variability
Department of Mathematics and Computer Science PAGE 423-09-14
Environment with a lot of variability
Department of Mathematics and Computer Science PAGE 523-09-14
Environment with a lot of variability
Department of Mathematics and Computer Science PAGE 623-09-14
Discovery of Spaghetti-like models
Department of Mathematics and Computer Science PAGE 723-09-14
Discovery of Spaghetti-like models
Department of Mathematics and Computer Science PAGE 823-09-14
Declarative approaches
Department of Mathematics and Computer Science PAGE 923-09-14
Declarative process discovery
• Avoid the discovery of spaghetti-like models• Traditional discovery techniques explicitly specify all
possible behaviours (closed models)
• Declarative process discovery: process behaviour described as a compact set of rules (open models)
Department of Mathematics and Computer Science PAGE 1023-09-14
Declarative process discovery
• Possibility to guide the discovery process towards specific properties of interest
Department of Mathematics and Computer Science PAGE 1123-09-14
Declarative process discovery
Department of Mathematics and Computer Science PAGE 1223-09-14
Declarative process discovery
Department of Mathematics and Computer Science PAGE 1323-09-14
A is always eventually followed by
B
Declarative process discovery
Department of Mathematics and Computer Science PAGE 1423-09-14
A is always eventually followed by
B
A or B always
occur but never
together
Declarative process discovery
Department of Mathematics and Computer Science PAGE 1523-09-14
A is always eventually followed by
B
A or B always
occur but never
together
A and B never
occur in sequence
Declare
• A is always eventually followed by B• RESPONSE• User-friendly graphical representation
• Semantics specified through LTL (for finite traces)
Department of Mathematics and Computer Science PAGE 1623-09-14
Core algorithm
PAGE 1723-09-14
LOG
User-guided discovery
Department of Mathematics and Computer Science PAGE 1823-09-14
Core algorithm
Department of Mathematics and Computer Science PAGE 1923-09-14
• W = {(A C B C), (C B A C), (A C A C A C B)}
Core algorithm
Department of Mathematics and Computer Science PAGE 2023-09-14
• W = {(A C B C), (C B A C), (A C A C A C B)}
Core algorithm
Department of Mathematics and Computer Science PAGE 2123-09-14
• W = {(A C B C), (C B A C), (A C A C A C B)}
Core algorithm
Department of Mathematics and Computer Science PAGE 2223-09-14
• W = {(A C B C), (C B A C), (A C A C A C B)}
Core algorithm
Department of Mathematics and Computer Science PAGE 2323-09-14
• W = {(A C B C), (C B A C), (A C A C A C B)}
Tuning the discovery process: PoE
• Percentage of Events (PoE) avoids the discovery of less-relevant constraints referring to event classes which rarely occur in the log• This parameter has also a positive effect on the execution
time of the algorithm
Department of Mathematics and Computer Science PAGE 2423-09-14
PoE parameter
Department of Mathematics and Computer Science PAGE 2523-09-14
• W = {(A C B C), (C B A C), (A C A C A C B)}
PoE parameter
Department of Mathematics and Computer Science PAGE 2623-09-14
• W = {(A C B C), (C B A C), (A C A C A C B)}− f(A) = 5/15
− f(B) = 3/15
− f(C) = 7/15
PoE parameter
Department of Mathematics and Computer Science PAGE 2723-09-14
• W = {(A C B C), (C B A C), (A C A C A C B)}− f(A) = 5/15
− f(B) = 3/15
− f(C) = 7/15
PoE parameter
Department of Mathematics and Computer Science PAGE 2823-09-14
• W = {(A C B C), (C B A C), (A C A C A C B)}− f(A) = 5/15
− f(B) = 3/15
− f(C) = 7/15
PoE parameter
Department of Mathematics and Computer Science PAGE 2923-09-14
• W = {(A C B C), (C B A C), (A C A C A C B)}− f(A) = 5/15
− f(B) = 3/15
− f(C) = 7/15
Tuning the discovery process: PoI
• Percentage of Instances (PoI) specifies that a constraint can still be discovered even if it does not hold for all process instances of the log • This parameter is useful in case of noisy logs, where rules
are violated in exceptional cases, but hold for most cases
Department of Mathematics and Computer Science PAGE 3023-09-14
PoI parameter
Department of Mathematics and Computer Science PAGE 3123-09-14
• W = {(A C B C), (C B A C), (A C A C A C B)}
PoI parameter
Department of Mathematics and Computer Science PAGE 3223-09-14
• W = {(A C B C), (C B A C), (A C A C A C B)}
3/32/3
2/31/3
1/3 1/3
PoI parameter
Department of Mathematics and Computer Science PAGE 3323-09-14
• W = {(A C B C), (C B A C), (A C A C A C B)}
3/32/3
2/31/3
1/3 1/3
Truncated semantics for DECLARE constraints
Department of Mathematics and Computer Science PAGE 3423-09-14
• W = {(A C D B C D A E F B A), (C A D B C A D C B F D A D B C A D E F), (A C B D A E B F A E F)}
Truncated semantics for DECLARE constraints
Department of Mathematics and Computer Science PAGE 3523-09-14
• W = {(A C D B C D A E F B A), (C A D B C A D C B F D A D B C A D E F), (A C B D A E B F A E F)}
Truncated semantics for DECLARE constraints
• Relevant when logs are not complete
• Literature on Truncated Semantics• C. Eisner, D. Fisman, J. Havlicek, A. Mcisaac, Y. Lustig, and D. V.
Campenhout, “Reasoning with Temporal Logic on Truncated Paths,” in In CAV Proceedings, LNCS 2725, pp. 27–40, 2003
Department of Mathematics and Computer Science PAGE 3623-09-14
Truncated semantics for DECLARE constraints
Department of Mathematics and Computer Science PAGE 3723-09-14
• After every prefix, four evaluations of a constraint− Satisfied : independent of future
− Temporarily satisfied : satisfied if this is the end of the log− Temporarily violated : violated if this is the end of the log− Violated : independent of future
• Three kinds of semantics:• Weak: temporarily xxx satisfied
• Neutral: temporarily xxx xxx
• Strong: temporarily xxx violated
Truncated semantics for DECLARE constraints
Department of Mathematics and Computer Science PAGE 3823-09-14
• W = {(A C D B C D A E F B A), (C A D B C A D C B F D A D B C A D E F), (A C B D A E B F A E F)}
Truncated semantics for DECLARE constraints
Department of Mathematics and Computer Science PAGE 3923-09-14
• W = {(A C D B C D A E F B A), (C A D B C A D C B F D A D B C A D E F), (A C B D A E B F A E F)}
(Weak semantics)
Vacuity detection in DECLARE discovery
Department of Mathematics and Computer Science PAGE 4023-09-14
• W = {(C B C B E F ), (C B C B C F B C B E F), (C B E F E F)}
Vacuity detection in DECLARE discovery
Department of Mathematics and Computer Science PAGE 4123-09-14
• W = {(C B C B E F ), (C B C B C F B C B E F), (C B E F E F)}
Vacuity detection in DECLARE discovery
Department of Mathematics and Computer Science PAGE 4223-09-14
• Literature on Vacuity Detection• O. Kupferman and M. Y. Vardi, “Vacuity Detection in Temporal
Model Checking,” International Journal STTT, vol. 4, pp. 224–233, 2003
Vacuity detection in DECLARE discovery
Department of Mathematics and Computer Science PAGE 4323-09-14
• A constraint is vacuously satisfied if the constraint is not really “activated”
• Instead of checking the validity of a constraint c we check the validity of witness(c) to be sure that the constraint is non-vacuously satisfied
Vacuity detection in DECLARE discovery
Department of Mathematics and Computer Science PAGE 4423-09-14
c
Vacuity detection in DECLARE discovery
Department of Mathematics and Computer Science PAGE 4523-09-14
witness(c)
Vacuity detection in DECLARE discovery
Department of Mathematics and Computer Science PAGE 4623-09-14
• W = {(C B C B E F ), (C B C B C F B C B E F), (C B E F E F)}
Vacuity detection in DECLARE discovery
Department of Mathematics and Computer Science PAGE 4723-09-14
• W = {(C B C B E F ), (C B C B C F B C B E F), (C B E F E F)}
Conclusion
Department of Mathematics and Computer Science PAGE 4823-09-14
• Novel approach to discover declarative models from logs that allows users to guide the discovery process towards specific properties
• Results on truncated semantics can be used to obtain significant results in the case that only partial logs are available
• Vacuity detection to identify the percentage of process instances where a constraint is really activated
Present and future work
Department of Mathematics and Computer Science PAGE 4923-09-14
• Better performance of the discovery algorithm• equivalent combinations
• combination of event classes occurring in the same trace
Present and future work
Department of Mathematics and Computer Science PAGE 5023-09-14
• Application of the approach to several case studies
• Given a constraint and a process instance where it is non-vacuously satisfied how many times it has been “activated” in the process instance
• Given a constraint and a process instance where it is violated level of “healthiness” of the process instance based on the number of violations
Visit the website
http://www.win.tue.nl/declare/declare-miner/
Department of Mathematics and Computer Science PAGE 5123-09-14