SLIDE 1
PAGE 1
PAGE 1 How did PM tooling develop Three key over time? When did - - PowerPoint PPT Presentation
PAGE 1 How did PM tooling develop Three key over time? When did - - PowerPoint PPT Presentation
PAGE 1 How did PM tooling develop Three key over time? When did observations process mining What are the start? main research challenges? Conclusion What are the main Why is process PM developments discovery so How about data in
SLIDE 2
SLIDE 3
PAGE 2
Conclusion How about data mining and business process management? When did process mining start? What are the main PM developments in this century? How did PM tooling develop
- ver time?
Three key
- bservations
Why is process discovery so difficult? What are the main research challenges?
SLIDE 4
PAGE 3
Conclusion How about data mining and business process management? When did process mining start? What are the main PM developments in this century? How did PM tooling develop
- ver time?
Three key
- bservations
Why is process discovery so difficult? What are the main research challenges?
SLIDE 5
Positioning Process Mining
4
process mining
Data Mining (DM)
(clustering, classification, rule discovery, etc.)
Business Process Management (BPM)
(process analysis/modeling, enactment, verification, etc.)
performance-oriented questions, problems and solutions compliance-oriented questions, problems and solutions
SLIDE 6
History and Origins of BPM
PAGE 5
database system user interface database system user interface database system application BPM system
1960 1975 1985 2000
application application application
BPM WFM
- ffice
automation data modeling
- perations
management scientific management business intelligence software engineering formal methods business process reengineering
Skip Ellis, Office Talk, 1979 Michael Zisman, SCOOP, 1977 Anatol Holt, Information Systems Theory Project, 1968 Carl Adam Petri, Petri nets, 1962
SLIDE 7
History and Origins of Data Mining
PAGE 6
Classical statistics (since 500 BC): descriptive statistics (e.g., sample mean) statistical inference (e.g., confidence interval, regression, hypothesis testing). Artificial intelligence (since 1950): making intelligent machines by applying human-thought- like processing to statistical problems. Machine learning (since 1950): construction and study of systems that can learn from data.
SLIDE 8
Data Mining: Supervised Learning
- Labeled data, i.e., there is a response variable that
labels each instance.
- Goal: explain response variable (dependent variable)
in terms of predictor variables (independent variables).
- Classification techniques (e.g., decision tree
learning) assume a categorical response variable and the goal is to classify instances based on the predictor variables.
- Regression techniques assume a numerical
response variable. The goal is to find a function that fits the data with the least error.
PAGE 7
SLIDE 9
Example: Decision tree learning
PAGE 8
logic failed (79/10)
- ≥ 8
passed (31/7) failed (101/8) linear algebra program ming
- perat.
research cum laude (20/2) <8 <6 <6 passed (82/7) ≥ 6 ≥ 6 passed (87/11) ≥ 7 <7 linear algebra ≥ 6 <6 failed (20/4)
SLIDE 10
Unsupervised Learning
- Unsupervised learning assumes unlabeled data, i.e.,
the variables are not split into response and predictor variables.
- Examples: clustering (e.g., k-means clustering and
agglomerative hierarchical clustering) and pattern discovery (association rules)
PAGE 9
SLIDE 11
Example: Association rules
PAGE 10
SLIDE 12
Example: Episode Mining
PAGE 11
a b c d E1 b c E2 a b c d E3
10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
a c b d e c b b c f a e e b c d c b E1 E2 (16x) E1 E3
SLIDE 13
PAGE 12
Conclusion How about data mining and business process management? When did process mining start? What are the main PM developments in this century? How did PM tooling develop
- ver time?
Three key
- bservations
Why is process discovery so difficult? What are the main research challenges?
SLIDE 14
Language identification in the limit (Mark Gold 1967)
- Mother uses sentences from some
language {aab, ab, ab, abc, …}.
- "Perfect child" listens to mother and
hypothesizes what the full language is like (given all sentences so far).
- Eventually the perfect child’s
hypothesis is correct and never changes again (without knowing), i.e., only finitely many wrong hypotheses are generated.
- A language is learnable in the limit if
such a perfect child exists.
PAGE 13
Language identification in the limit by E Mark Gold, Information and Control, 10(5):447–474, 1967.
SLIDE 15
Language identification in the limit (E. Mark Gold 1967)
- Gold showed that most languages cannot be
learned in the limit (including the most simple
- nes like regular languages (ab*(c|d)).
- He noted that it matters whether the child gets
positive and negative examples (corrections), whether the mother is evil, etc.
- Frequencies matter!
- Representational bias matters!
PAGE 14
sentence ≅ trace in event log language ≅ process model
SLIDE 16
Myhill-Nerode Theorem (1958) and the Biermann/Feldman Algorithm (1972)
- There is a unique minimal deterministic finite
automaton recognizing a regular language L ( shown by John Myhill and Anil Nerode in 1958).
- The equivalence classes defined by ≅ determine the
states of the automaton: x ≅ y if there is no z such that xz∉L and yz∈L.
- Cannot be applied to example traces: overfitting and
no generalization.
- Alan W. Biermann and Jerome A. Feldman propose
in 1972 techniques to learn finite state machines from examples (e.g., considering k-tails).
PAGE 15
- Nerode. Linear automaton transformations. Proc. Amer. Math. Soc. 9 1958 541-544.
Biermann and Feldman. On the synthesis of finite-state machines from samples of their behaviour. IEEE Transactions on Computers, 21:592–597, 1972.
SLIDE 17
SLIDE 18
Where/when did process mining start?
- Myhill/Nerode(1958)?
- Gold (1967)?
- Baum/Welch (1970)?
- Biermann/Feldman (1972)?
- Rakesh Agrawal (1994)?
− Apriori algorithm for frequent patterns, later extended to sequences, episodes, …
- Jonathan Cook and Alexander Wolf (1998)?
− "Discovering Models of Software Processes from Event-Based Data" − using techniques similar to Biermann/Feldman (k-tails) and Baum/Welch (Markov models)
- Rakesh Agrawal, Dimitrios Gunopulos, Frank Leymann?
− "Mining Process Models from Workflow Logs" (1998) − Flowmark process models without discovering type of splits and joins, no loops, etc.
- Anindya Datta (1998)?
− Automating the Discovery of AS-IS Business Process Models − Biermann/Feldman style work, embedded in BPM
PAGE 17
SLIDE 19
SLIDE 20
Initial team
PAGE 19
SLIDE 21
PAGE 20
Conclusion How about data mining and business process management? When did process mining start? What are the main PM developments in this century? How did PM tooling develop
- ver time?
Three key
- bservations
Why is process discovery so difficult? What are the main research challenges?
SLIDE 22
Workflow Mining
PAGE 21
diagnosis/ requirements configuration/ implementation enactment/ monitoring adjustment (re)design models data insight discussion verification performance analysis animation specification documentation configuration
SLIDE 23
Models, data, and systems coexist
PAGE 22
( r e ) d e s i g n implement/configure r u n & a d j u s t model-based analysis d a t a
- b
a s e d a n a l y s i s
SLIDE 24
SLIDE 25
Team in November 2007
PAGE 24
Some people are missing, e.g., Peter van den Brand.
SLIDE 26
Current process mining spectrum
(including alignments, operational support, and multiple perspectives)
PAGE 25
information system(s)
current data
“world”
people machines
- rganizations
business processes documents historic data resources/
- rganization
data/rules control-flow de jure models resources/
- rganization
data/rules control-flow de facto models provenance explore predict recommend detect check compare promote discover enhance diagnose
cartography navigation auditing event logs Models
“pre mortem” “post mortem”
SLIDE 27
PAGE 26
Conclusion How about data mining and business process management? When did process mining start? What are the main PM developments in this century? How did PM tooling develop
- ver time?
Three key
- bservations
Why is process discovery so difficult? What are the main research challenges?
SLIDE 28
Pre-ProM
(figure from March 2002!)
PAGE 27
Staffware InConcert MQ Series
workflow management systemen
FLOWer Vectus Siebel
case handling / CRM systemen
SAP R/3 BaaN Peoplesoft
ERP systems
gemeenschappelijk XML formaat voor het opslaan van workflow logs
EMiT Little Thumb
mining tools
InWoLvE Process Miner Exper- DiTo alpha algorithm including time analysis (BvD) predecessor
- f MXML
format predecessor of ProM's heuristic miner (TW) mining with duplicate tasks (Joachim Herbst) mining block structured models (Guido Schimm) evaluation tool (Laura Maruster) The first tool to support the alpha algorithm for process mining was the MiMo (Mining Module) tool based on ExSpect. Later it was implemented in EMiT and ProM.
Tobias Blickle (ARIS PPM)
SLIDE 29
PAGE 28
EMiT MiMo Little Thumb Process Miner
SLIDE 30
SLIDE 31
SLIDE 32
SLIDE 33
SLIDE 34
SLIDE 35
SLIDE 36
SLIDE 37
SLIDE 38
PAGE 37
Conclusion How about data mining and business process management? When did process mining start? What are the main PM developments in this century? How did PM tooling develop
- ver time?
Three key
- bservations
Why is process discovery so difficult? What are the main research challenges?
SLIDE 39
How good is my model: Four forces
PAGE 38
fitness simplicity generalization precision
Process Mining
ability ¡to ¡explain ¡
- bserved ¡behavior
avoiding ¡ underfitting Occam’s ¡Razor avoiding ¡
- verfitting
lift gravity thrust drag
Leaving out one of these dimensions during discovery will lead to degenerate cases!
SLIDE 40
SLIDE 41
PAGE 40
formal (not just a picture) fast (should not take years) sound (result should at least be free
- f deadlocks,
etc.) ability to balance all conformance dimensions (fitness, precision, generalization, and simplicity) incl. noise provide guarantees (not just a best effort) 1 2 3 4 5
SLIDE 42
PAGE 41
Conclusion How about data mining and business process management? When did process mining start? What are the main PM developments in this century? How did PM tooling develop
- ver time?
Three key
- bservations
Why is process discovery so difficult? What are the main research challenges?
SLIDE 43
PAGE 42
- conformance checking to diagnose deviations
- squeezing reality into the model to do model-based
analysis
#1 Alignments are essential!
SLIDE 44
PAGE 43
#2 Models are like the glasses required to see and understand event data!
SLIDE 45
SLIDE 46
PAGE 45
Conclusion How about data mining and business process management? When did process mining start? What are the main PM developments in this century? How did PM tooling develop
- ver time?
Three key
- bservations
Why is process discovery so difficult? What are the main research challenges?
SLIDE 47
PAGE 46
Finding sheep with five legs
we are getting close…
SLIDE 48
PAGE 47
Distributing process mining problems to cope with big data
SLIDE 49
PAGE 48
On-the-fly process mining Operational support
SLIDE 50
Concept drift
PAGE 49
Concept drift
SLIDE 51
Cross-organizational mining
PAGE 50
cross-organizational / comparative process mining
SLIDE 52
PAGE 51
context aware process mining
SLIDE 53
PAGE 52
Suppor upporting ing the he pr proces
- cess
- f
- f pr
proces
- cess mining
mining
SLIDE 54
PAGE 53
Conclusion How about data mining and business process management? When did process mining start? What are the main PM developments in this century? How did PM tooling develop
- ver time?
Three key
- bservations
Why is process discovery so difficult? What are the main research challenges?
SLIDE 55
SLIDE 56