=Paper=
{{Paper
|id=Vol-1789/bpm-demo-2016-paper8
|storemode=property
|title=BPMN Miner 2.0: Discovering Hierarchical and Block-Structured BPMN Process Models
|pdfUrl=https://ceur-ws.org/Vol-1789/bpm-demo-2016-paper8.pdf
|volume=Vol-1789
|authors=Raffaele Conforti,Adriano Augusto,Marcello La Rosa,Marlon Dumas,Luciano García-Bañuelos
|dblpUrl=https://dblp.org/rec/conf/bpm/ConfortiARDG16
}}
==BPMN Miner 2.0: Discovering Hierarchical and Block-Structured BPMN Process Models==
https://ceur-ws.org/Vol-1789/bpm-demo-2016-paper8.pdf
BPMN Miner 2.0: Discovering Hierarchical and
Block-Structured BPMN Process Models
Raffaele Conforti1 , Adriano Augusto1 , Marcello La Rosa1 , Marlon Dumas2 , and
Luciano Garcı́a-Bañuelos2
1
Queensland University of Technology, Australia
{a.augusto, raffaele.conforti, m.larosa}@qut.edu.au
2
University of Tartu, Estonia
{marlon.dumas, luciano.garcia}@ut.ee
Abstract. We present BPMN Miner 2.0: a tool that extracts hierarchical and
block-structured BPMN process models from event logs. Given an event log in
XES format, the tool partitions it into sub-logs (one per subprocess) and discovers
a BPMN process model from each sub-log using existing techniques for discover-
ing BPMN process models via heuristics nets or Petri nets. A drawback of these
techniques is that they often produce spaghetti-like models and in some cases
unsound models. Accordingly, BPMN Miner 2.0 applies post-processing steps
to remove unsound constructions as well as a technique to block-structrure the
resulting process models in a behavior-preserving manner. The tool is available
as a standalone Java tool as well as a ProM and an Apromore plugin. The tar-
get audience of this demonstration includes process mining researchers as well as
practitioners interested in exploring the potential of process mining using BPMN.
Keywords: Process Discovery, BPMN, Structured Process Models, Hierarchical
Process Models, ProM, Apromore
Process mining is a discipline within Business Process Management that aims to
extract actionable insights from process execution logs [9]. Such execution logs, called
event logs for short, can be extracted from common or special-purpose IT systems avail-
able in today’s organizations, such as an ERP system or a Claims Management System.
One particular category of process mining methods is that of automated process
model discovery. These methods extract a process model from an event log. Real-life
logs, however, are typically affected by noise (e.g. in the form of infrequent behavior)
and are incomplete (i.e. they do not contain all possible behavior of the business process
under analysis). Consequently, process model discovery is not a trivial procedure and
leads to process models of varying quality, which often approximate the actual business
process, depending on the discovery algorithm that is employed.
A large number of automated process discovery algorithms have been proposed.
These algorithms strike various tradeoffs between accuracy (measured in terms of fit-
ness, precision and generalization [9]) and complexity (measured in terms of model size
and other complexity metrics [7]). Some algorithms tend to discover process models
with high accuracy at the cost of high model complexity. The importance of model
Copyright c 2016 for this paper by its authors. Copying permitted for private and academic
purposes.
40 Conforti et al.
complexity should not be underestimated as this underpins the understandability of the
extracted process model, and so ultimately its value to users. In this respect, empirical
studies [5] have shown that besides model size, an important proxy for process model
understandability is the structuredness of the model. This latter observation has led
to the design of discovery algorithms such as the Inductive Miner [6] and the Evolu-
tionary Tree Miner [2], which discover structured process models by design. Models
discovered using such algorithms may be further away from reality depending on the
degree of unstructuredness of the actual business process the log refers to. For exam-
ple, the Inductive Miner tends to over-generalize the behavior in the log, leading to low
precision while maximizing fitness [1]. On the other hand, discovery algorithms such
as the Heuristics Miner [11] and the Fodina Miner [10] tend to strike better results
in terms of accuracy but produce more complex and sometimes syntactically incorrect
and unsound process models [1]. Furthermore, the majority of discovery algorithms
produce models that are represented in languages that are either not widely accepted
among practitioners (e.g. Petri nets), too technical (e.g. Heuristics nets) or too abstract
and over-generalizing (e.g. fuzzy nets or process maps).
The BPMN Miner algorithm [3, 4] is designed to address the above limitations. This
algorithm discovers models in the BPMN 2.0 language [8], a de jure standard widely
supported by vendors and practitioners. Moreover, by exploiting implicit functional and
foreign-key dependencies between attributes in the event log, the algorithm can generate
hierarchical BPMN models, i.e. models organized over two or more abstraction levels
via the use of subprocess models. In order to further reduce the complexity of the dis-
covered models, the algorithm exploits an extensive set of notational elements provided
by the BPMN language, such as boundary events (to model interrupting exceptions),
activity markers (loops and multi-instance) and event subprocesses.
BPMN Miner relies on existing (baseline) automated process discovery algorithms
to generate an initial model of each subprocess. The baseline algorithms supported are
Inductive Miner, Heuristics Miner, Fodina, ILP Miner and Alpha Miner. Version 2.0 of
BPMN Miner additionally embeds an algorithm that discovers sound and maximally-
structured BPMN models, namely the Structured Miner [1]. This latter algorithm re-
moves unsound and unstructured constructs in the discovered model, thus further in-
creasing its potential usability.
The minimum input required by the tool is the log from which to discover a BPMN
model. By default, the Heuristics Miner is chosen as the baseline discovery algorithm.
It is however possible to customize a number of input parameters in order to tune the
results (see Figure 1).
Besides the baseline discovery algorithm, one can choose whether the partitioning
of the log into sublogs is achieved via a noise tolerant dependency discovery algorithm,
or not, in which case the log is assumed to be noise-free. Additionally, it is possible to
sort the input log based on the timestamp of its events, and to switch on the structuring
of the discovered process model. The latter function is only effective when the baseline
discovery algorithm does not already produce a structured model by design.
Additional parameters can be set to fine-tune the discovery of BPMN-specific no-
tational elements, on the basis of a number of heuristics. For example, one can set
tolerance levels for the identification of boundary timer and message events, and of
multi-instance activity markers.
BPMN Miner 2.0: Discovering BPMN Process Models 41
Once these parameters are set, the algo-
rithm retrieves event attributes which may
be used as primary keys for the partition-
ing of the log into sublogs. In this context,
a primary key is an attribute which is recur-
rent in all events that are related to the ac-
tivities of a particular subprocess. For exam-
ple, an attribute “invoice” would be recurrent
across all events related to the handling of
the invoice, as part of an overarching order-
to-cash process. All such events that are re-
lated to the handling of the invoice would be
isolated in a separate sublog, from which the
corresponding subprocess for handling in-
voices will then be discovered. The user has
the possibility of steering the use of partic-
ular event attributes by selecting/deselecting
them from a list that is automatically popu-
lated by the tool.
Next, the algorithm assigns a specific
primary key to each sublog. If more than
one primary key can be assigned to the same
sublog, the user is asked to choose from a
droplist, where the first key is the most fitting
one. At this point the event log is partitioned
into sublogs using the chosen primary keys,
and each sublog is passed as input to the
baseline discovery algorithm for model dis-
covery. When the discovery has completed,
each subprocess model is structured sepa-
rately, if this option is enabled, and assem-
bled together as part of a single hierarchical
BPMN model.
Figure 2 shows an example of hierar-
chical process model discovered by BPMN Fig. 1. Parameters.
Miner, which was fully structured after the
discovery. This model exhibits two expanded subprocesses (one with loop marker, the
other with multi-instance marker) and one event subprocess. Figure 3 shows an example
of hierarchical process model discovered with BPMN Miner in ProM. This latter model
is maximally structured and exhibits two expanded subprocesses (again, with loop and
multi-instance markers), two event subprocesses (one nested within a subprocess), and
boundary timer and message events with exception flows. Both models were discovered
from synthetic log of an order-to-cash process.
The accuracy and scalability of BPMN Miner have been extensively evaluated us-
ing over 600 event logs, including both artificial and real-life event logs. The majority
of the artificial logs were generated from the SAP R/3 and IBM BIT process model
collections, which group models from a variety of domains, including finance, sales,
42 Conforti et al.
Fig. 2. Example of fully-structured process model discovered by BPMN Miner in Apromore.
Fig. 3. Example of maximally-structured process model discovered by BPMN Miner in ProM.
accounting, logistics, communication and human resources. The results of these evalu-
ations are reported in [3, 4, 1].
The results of the experiments show that the tool scales well to large and noisy real-
life logs, performing within reasonable time bounds in the order of minutes. Moreover,
the results indicate a statistically significant improvement of discovery accuracy and
model complexity over all the baseline discovery algorithms supported by the tool.
BPMN Miner 2.0 is available as an OSGi plugin of the Apromore process model
repository, as a plugin of the ProM Framework, as well as a standalone command-
lineJava tool. Apromore is an online open-source ecosystem of advanced capabilities
for managing large process model collections, including process modeling, simulation,
filtering, querying, similarity search, behavioral comparison and model merging. ProM
is the largest on-source process mining framework, offering over 300 plugins (in its
latest incarnation) offering process model discovery, conformance checking, variants
and deviance mining, and log analysis capabilities.
BPMN Miner 2.0: Discovering BPMN Process Models 43
A screencast is available at https://youtu.be/eb0k2RO2PQ8. This video
illustrates different examples and provides a brief explanation of the tool settings along
with the possible outputs that can be obtained by varying the input parameters. BPMN
Miner 2.0 is embedded as an OSGi plugin in the online platform Apromore, which has
been used for the screencast (http://apromore.qut.edu.au). The artificial log
used in the screencast is available at https://goo.gl/AdvnEd.
BPMN Miner is also available as a ProM plugin (http://promtools.org)
and as a standalone Java tool (http://apromore.org/platform/tools).
References
1. A. Augusto, R. Conforti, M. Dumas, M. La Rosa, and G. Bruno. Automated discovery of
structured process models: Discover structured vs. discover and structure. In Proc. of ER.
Springer, 2016. Preprint available at http://eprints.qut.edu.au/95189/.
2. J. Buijs, B.F. van Dongen, and W.M.P. van der Aalst. On the role of fitness, precision,
generalization and simplicity in process discovery. In Proc. of CoopIS, volume 7565 of
LNCS, pages 305–322. Springer, 2012.
3. R. Conforti, M. Dumas, L. Garcı́a-Bañuelos, and M. La Rosa. Beyond tasks and gateways:
Discovering BPMN models with subprocesses, boundary events and activity markers. In
Proc. of BPM, LNCS, 2014.
4. Raffaele Conforti, Marlon Dumas, Luciano Garcı́a-Bañuelos, and Marcello La Rosa. Bpmn
miner: Automated discovery of bpmn process models with hierarchical structure. Informa-
tion Systems, 56:284–303, 2016.
5. M. Dumas, M. La Rosa, J. Mendling, R. Mäesalu, H.A. Reijers, and N. Semenenko. Un-
derstanding business process models: the costs and benefits of structuredness. In Proc. of
CAiSE, volume 7328 of LNCS, pages 31–46. Springer, 2012.
6. S.J.J. Leemans, D. Fahland, and W.M.P. van der Aalst. Discovering block-structured process
models from event logs - a constructive approach. In Proc. of PETRI NETS, volume 7927 of
LNCS. Springer, 2013.
7. J. Mendling. Metrics for Process Models: Empirical Foundations of Verification, Error Pre-
diction, and Guidelines for Correctness. Springer, 2008.
8. Object Management Group (OMG). Business Process Model and Notation (BPMN) ver. 2.0.
Object Management Group (OMG), January 2011.
9. W.M.P. van der Aalst. Process Mining - Discovery, Conformance and Enhancement of Busi-
ness Processes. Springer, 2011.
10. S.K.L.M. vanden Broucke, J. De Weerdt, J. Vanthienen, and B. Baesens. Fodina: a robust
and flexible heuristic process discovery technique. http://www.processmining.
be/fodina/. Last accessed: 03/27/2014.
11. A.J.M.M. Weijters and J.T.S. Ribeiro. Flexible Heuristics Miner (FHM). In Proc. of CIDM.
IEEE, 2011.