Digital and Knowledge Economy UG4/MSc Building an Ontology Slide - PDF document

Digital and Knowledge Economy UG4/MSc Building an Ontology Slide Set 8 Dr. Jessica Chen-Burger Computer Science, MACS Heriot-Watt University Y.J.ChenBurger@hw.ac.uk 1 What is Process Mining?  A young field arise more than a decade ago  A combined origin – Computational intelligence and data mining – Process modelling and analysis  Activities involved in PM – To discover actual processes – To help analyse processes (e.g. by comparing process models and actual processes) – To help improve actual processes 2 1

Why Process Mining? (1)  To understand processes that actually happened  To compare actual processes with processes that one thought to have been carried out  To compare actual processes with standard processes  To understand who carried out the processes  To understand what resources are used/needed by processes  To understand where processes are been carried out (location, organisation) 3 Why Process Mining? (2)  To streamline and improve actual processes, e.g. – To reduce cost – To speed up processes – To reduce redundancy – To discover/remove irregularities  To create improved standard process model  To reuse process models 4 2

PM Three Functions (re-cap) Three types of PM 5 Event Log 6 3

7 Three Main PM Functions 8 4

Where to apply Process Mining? 9 10 5

Sample Control-Flow Diagram 1 11 Sample Control-Flow Diagram 2 12 6

Sample Control-Flow Diagram 3 [Source: Celonis slides] 13 The Event Log (re-cap) [Source: Celonis slides] 14 7

Event Logs  May have to pull information from multiple places, e.g. – Database tables – Message logs – Mail archives – Transaction logs  Making sense of pulled information and create an integrated event log, inc. – Activities that is temporal based – May include information flow – May include process flow – May include cost information – May include actor/role information – May include mechanism used in an activity 15 Problems in Event Logs  Incorrectness – Errors in event logs – Assumed/recalled data, not always real data recorded  Missing information – Missing events – Sparse entries in variables (of events)  Missing definition (semantics) – Process and data level  Need to preserve privacy and security – E.g. healthcare or financial data – Do you need consent from the actor that information are being recorded and the way information is being used? 16 8

Object Management Group Business Process Model and Notation (BPMN not examinable) And – parallel gateway Loop XOR – exclusive gateway 17 Basic Process Flow Control  And gateway (parallel)(and-split/join) – All activities are executed – They can be executed in parallel, if running on a parallel machine/network  Xor gateway (choice; exclusive gateway)(Xor- split/join) – Exactly one activity is executed  Or gateway (inclusive gateway)(Or-split/join) – At least one activity is executed; – If more than one activity is executed, they can be executed in parallel, if running on a parallel machine/network 18 9

Process Mining Guidelines  Event data should aim for high quality  Log extraction should be driven by questions  Basic process flow control should be supported in the process model, i.e. – and/or/xor gateway  Events (stored in event logs) should be related to elements in the control-flow and process model  Models are abstractions of the reality that are created to support certain goals  Process mining is a continuous process 19 Challenges - 1  Find, merge and cleaning event data  Dealing complex event logs having diverse characteristics, e.g. – Different sizes in cases – Large amount of different types of events to handle – Incomplete real-world samples – Too low level of abstraction of events  Create representative benchmarks  Dealing with concept drift – the process is changing while being analysed 20 10

Challenges - 2  Avoid representational bias by selecting a suitable process modelling language – Can it represent all of the basic/important processes?  Balancing between quality criteria, i.e. – Fitness (show all links between PM and logs), – Simplicity (simple model) – Precision (no noise) – Generalisation (describe the domain, not just sample events) – Problems to capture real cases with low frequency 21 Challenges - 3  Cross-organisational mining, e.g. – End-to-end process in a Supply Chain – Cloud computing – different organisations executing the same/similar processes while sharing experiences, knowledge or a common infrastructure, e.g. salesforce.com – organisations may learn from one another  Provide operational support (real-time) – Detect (and alert) – Predict - use historical data as guide – Recommend 22 11

Challenges - 4  Combine process mining with other types of analysis, e.g. – Operations management – Data mining (e.g. pattern discovery) – Predicting future (e.g. use simulation) – Visual analytics – automated analysis with interactive visualisation  Improve usability for non-experts – Support living process model, not a static one – Use user-friendly UI, hide complexity  Improve understand-ability for non-experts – Indication of data accuracy; – Support for results generated 23 Celonis Tool Demo  http://www.macs.hw.ac.uk/~yjc32/proje ct/Teaching/DKE-2016-17/coursework-2/ – Demo 1 – Demo 2 24 12

Exercise - 1  What is process mining?  Why is process mining useful?  What are the main functions/types of process mining?  How process mining interact with software systems?  What sorts of results may be derived from process mining?  What are the challenges?  Given a real-world scenario, can you recommend a process mining solution, e.g. based on the 3 functions? 25 Exercise - 2  The Lothian Bus wishes to have a better understanding of their customers’ usage of their buses, in the interest of providing better services based on lower cost.  They are interested in, e.g. – Where most of the customers get on the bus and where they take off – When they take the bus – When is the peak time – Where and when the traffic is congested – Whether buses are on-time  Recommend relevant process mining techniques and explain why. 26 13

References  IEEE CIS Task Force on Process Mining, Process Mining Manifesto, 2011: http://www.win.tue.nl/ieeetfpm/doku.p hp?id=shared:process_mining_manifesto 27 Reference (not examinable) • BPMN model: http://www.bpmn.org/ • Jon Espen Ingvaldsen and Jon Atle Gulla, Model-Based Business Process Mining: https://scholar.google.com/citations?use r=lq5InSEAAAAJ&hl=en 28 14