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 this century? difficult? mining and business process management? PAGE 2

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 this century? difficult? mining and business process management? PAGE 3

Positioning Process Mining Business Process Management (BPM) (process analysis/modeling, enactment, verification, etc.) performance-oriented questions, compliance-oriented questions, problems and solutions problems and solutions process mining Data Mining (DM) (clustering, classification, rule discovery, etc.) 4

History and Origins of BPM user user interface interface application BPM system application application application Michael Zisman, Carl Adam Petri, SCOOP, 1977 Petri nets, 1962 Anatol Holt, Information Systems Theory database database database Skip Ellis, Project, 1968 Office Talk, system system system 1979 BPM 1960 1975 1985 2000 business business process intelligence reengineering WFM operations management data modeling formal methods office software scientific automation engineering management PAGE 5

History and Origins of Data Mining 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. PAGE 6

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

Example: Decision tree learning logic ≥ 8 - <8 failed program ≥ 7 (79/10) ming linear <7 cum laude algebra ≥ 6 (20/2) <6 linear algebra ≥ 6 passed (87/11) operat. <6 research ≥ 6 <6 passed (31/7) failed (20/4) failed passed (101/8) (82/7) PAGE 8

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

Example: Association rules PAGE 10

Example: Episode Mining b b b a d a d c c c E1 E2 E3 E2 (16x) a c b e d f c b b c a e b e c d c b 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 E1 E1 E3 PAGE 11

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 this century? difficult? mining and business process management? PAGE 12

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.

Language identification in the limit (E. Mark Gold 1967) • Gold showed that most languages cannot be learned in the limit (including the most simple ones 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! sentence ≅ trace in event log language ≅ process model PAGE 14

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). 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. PAGE 15 IEEE Transactions on Computers, 21:592–597, 1972.

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

Initial team PAGE 19

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 this century? difficult? mining and business process management? PAGE 20

Workflow Mining diagnosis/ requirements adjustment insight performance discussion animation analysis enactment/ (re)design monitoring data models verification documentation specification configuration/ implementation configuration PAGE 21

Models, data, and systems coexist model-based d a t a a - analysis b n a a s l y e s d i s n r u g n i s e & d a ) e d r j u ( s t implement/configure PAGE 22

Team in November 2007 Some people are missing, e.g., Peter van den Brand. PAGE 24

Current process mining spectrum (including alignments, operational support, and multiple perspectives) “world” people organizations business machines processes documents information system(s) provenance event logs “pre “post current historic mortem” mortem” data data cartography navigation auditing recommend compare diagnose promote discover enhance explore predict detect check Models de jure models de facto models control-flow control-flow data/rules data/rules resources/ resources/ organization organization PAGE 25

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 this century? difficult? mining and business process management? PAGE 26

Pre-ProM (figure from March 2002!) The first tool to support workflow management systemen case handling / CRM systemen ERP systems the alpha algorithm for Staffware FLOWer SAP R/3 process mining was the MiMo (Mining Module) tool based on ExSpect. InConcert Vectus BaaN Later it was implemented in EMiT predecessor MQ Series Siebel Peoplesoft and ProM. of MXML format gemeenschappelijk XML formaat voor het opslaan van workflow logs alpha algorithm including time analysis (BvD) Little Exper- Process EMiT InWoLvE Thumb DiTo Miner mining block structured models mining tools (Guido Schimm) predecessor of ProM's Tobias Blickle heuristic miner (TW) mining with (ARIS PPM) evaluation tool duplicate tasks (Laura Maruster) (Joachim Herbst) PAGE 27

MiMo Little Thumb EMiT Process Miner PAGE 28

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 this century? difficult? mining and business process management? PAGE 37