Диссертация (1137084), страница 37
Текст из файла (страница 37)
They illustrate the applicability of presented model repairtechniques, their best use-cases, advantages and disadvantages.To summarize, the main contributions of this thesis are as follows:I. A new modular approach (scheme) for process model repair based on event logs is presented.It is based on process model decomposition. Process models are represented by workflow nets.The scheme may include various algorithms for model decomposition and sub-nets repair.Sufficient conditions of the approach effectiveness are formulated.II.
New algorithms are presented, performing local and non-local process model repair based onevent logs. These algorithms implement the general modular scheme, and employ the divideand conquer principle.164III. New algorithms for event log generation via process model simulation are presented.IV. Prototype implementations of the process model repair algorithms have been experimentallyevaluated using event logs generated by the simulation algorithms.Open Issues and Future Research DirectionsIn this thesis we proposed the algorithms to repair workflow nets according to the event logs.As usual, there are still several open issues which lay the foundations for possible future researchprojects.
They are discussed in this section.I. In this thesis, it is shown how to repair a process model such that the repaired model willperfectly fit a given event log. Besides, the proposed techniques preserve model structure. Thus,the repaired model is approximately as precise as the initial one. However, sometimes a user do notwant a perfectly fitting model, and it is enough to improve the fitness of the model. Possibly, thiscan be done using the less number of change operations.
The big challenge is to develop modularrepair techniques which will be able to balance the repaired model fitness and the severity of itschange.II. We have addressed the problem of fitness repair, since fitness is the key model characteristics.No one needs models which do not reflect the reality observed in event logs. However, fitness isnot the only model quality criterion. In particular, when evaluating repair results we also calculatehow precise the model is. Our current modular repair technique has no guarantees concerning theprecision of repaired models. This is mainly because the theoretical basis of decomposed precisioncalculation is not yet well established. To present a repair technique with guarantees concerningboth fitness and precision of a repaired model significant theoretical and practical studies areneeded.III.
The modular repair technique is appropriate in the cases when there are minorinconsistencies between a model and an event log, i.e. model fitness is not very low. A completere-discovery of a process model using the event log is a better choice when the fitness is low.Significant inconsistencies are usual for the non-local repairs. In these cases, the greedy repairtechnique may modify a model significantly. In the worst case, this technique re-discovers thewhole model. One of the future tasks is to develop new repair techniques for the cases with minorbut non-local inconsistencies.IV.
Yet another observation is that the combined strategies can be applied to repair modelswhich contain both local and non-local inconsistencies. The repair techniques presented inthis thesis can be combined with each other, and with the repair methods proposed by otherresearch groups. Of most interest are the methods of D. Fahland and W. van der Aalst [25] and165A. Polyvyanyy et al.
[108] These methods are appropriate when completely new activities areadded to a considered process, and they are observed in the event log. Thus, new transitionsshould be added to the process model in the appropriate positions. Perhaps, combined chains ofrepair methods will be able to re-arrange the transitions corresponding to the stable activity set,and then add new transitions, while preserving model structure.V.
Note also, that the modular repair scheme is general, and can be further refined. Variousdecomposition, discovery, and conformance checking algorithms can be used as its buildingblocks. Section 2.6 discusses many decomposition methods proposed in the literature. Some ofthese decompositions are valid, and thus may be used as building blocks of a modular repairtechnique. Examples of such a decomposition are SESE -based decomposition [137], and Passagebased decomposition [127]. Besides, many process discovery algorithms are available, as it wasstated in Section 1.2.1. Most of them do not guarantee that the synthesised model perfectly fitthe event log. Although an arbitrary discovery algorithm can not be directly used as a buildingblock of the modular scheme, further research can be done to determine the applicable discoveryalgorithms.VI.
More experiments are needed to comprehensively evaluate the modular repair technique.Especially useful will be experiments using real-life data. In this thesis, we have evaluated themodular repair technique using synthetic data which represents typical modelling patterns, becauseit is hard to select real-life process models with particular characteristics needed when the set ofactivities is stable. Thus, one of the future work directions is to evaluate the proposed repairtechniques in real-life cases.VII.
One more direction for future research is developing methods to repair process models invarious notations, such as: EPC s [27], well-known YAWL [186] and BPMN process models [164],UML Activity diagrams [29], and Heuristics nets [65]. Besides, there exist different types of highlevel Petri nets which are not considered yet: Coloured Petri nets [187], Petri nets with inhibitorarcs [188], Nested Petri nets [189], Petri nets with structured data [190], etc. Each notationdetermines various modelling elements.
Models in some of the notations are easy to decompose,while for the others principles of model decomposition should be developed. We believe, that ourmodular repair technique can be extended to the broad spectre of process modelling notations.166AcknowledgementsThe author of this thesis would like to acknowledge the following very important persons.First of all, I thank the supervisor of this thesis, the head of PAIS Lab14 prof. Irina Lomazova,and the honourable scientific director of PAIS Lab prof. Wil van der Aalst. Obviously, their adviceand supervision have been of key importance.Secondly, I thank my colleagues at PAIS Lab: Sergey Shershakov, Anna Kalenkova, and RomanNesterov.
We spent many hours in interesting conversations about process mining, softwareengineering, Petri nets, mathematics, teaching, and other not-so-formal subjects during last 5years. I also thank Roman for very valuable comments on this text.I also thank current and former student assistants at PAIS Lab: Ivan Shugurov, NataliaNikitina, Semen Tikhonov, Vladislav Lipyanin, Dmitry Yakovlev, and others. I feel happy to worktogether with all these people on the projects of PAIS Lab. Many of them have already graduated,and — I hope — also consider our collaboration joyful, and useful for their current and futureprofessional life.This thesis template is based on the Russian-Phd-LaTeX-Dissertation-Template 15 by AndreyAkinshin and other persons.
I thank all the contributors of this project.Last but not least, I thank my parents Olga and Alexander Mitsyuk, and my brother Anton.In some sense, they motivated me more than others mentioned above.1415Laboratory of Process-Aware Information Systems, NRU HSE: https://pais.hse.ru/en/Repository: https://github.com/AndreyAkinshin/Russian-Phd-LaTeX-Dissertation-Template167References1. Dumas M., van der Aalst W. M.
P., ter Hofstede A. H. M. Process-Aware InformationSystems: Bridging People and Software through Process Technology. — Wiley & Sons, 2005.2. van der Aalst W. M. P. Process-Aware Information Systems: Design, Enactment and Analysis // Wiley Encyclopedia of Computer Science and Engineering / Ed. by B.W. Wah. —Wiley & Sons, 2009.
— Pp. 2221–2233.3. Stahl Thomas, Voelter Markus, Czarnecki Krzysztof. Model-Driven Software Development:Technology, Engineering, Management. — USA: John Wiley and Sons, Inc., 2006.4. Brambilla Marco, Cabot Jordi, Wimmer Manuel. Model-Driven Software Engineering inPractice. — 1st edition. — Morgan and Claypool Publishers, 2012.5. Wil M. P. van der Aalst. Process Mining — Data Science in Action, Second Edition. —Springer, 2016.6. Fahland D., van der Aalst W. M.
P. Repairing Process Models to Reflect Reality // International Conference on Business Process Management (BPM 2012) / Ed. by A. Barros, A. Gal,E. Kindler. — Vol. 7481 of Lecture Notes in Computer Science. — Springer-Verlag, Berlin,2012. — Pp. 229–245.7. Generating event logs for high-level process models / Alexey A. Mitsyuk, Ivan S. Shugurov,Anna A. Kalenkova, Wil M. P. van der Aalst // Simulation Modelling Practice and Theory.— 2017.
— Vol. 74. — Pp. 1–16.8. Mitsyuk A. A., Lomazova I. A., van der Aalst W. M. P. Using Event Logs for Local Correction of Process Models // Automatic Control and Computer Sciences. — 2017. — Vol. 51,no. 7. — Pp. 709–723.9. Мицюк А. А., Ломазова И. А., ван дер Аалст В. М. П. Использование журналов событий для локальной корректировки моделей процессов // Моделирование и анализ информационных систем. — 2017. — Vol. 24, no. 4.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
