=Paper=
{{Paper
|id=Vol-3648/paper_4842
|storemode=property
|title=Noise and Outlier Detection in Task Mining
|pdfUrl=https://ceur-ws.org/Vol-3648/paper_4842.pdf
|volume=Vol-3648
|authors=Tom Hohenadl
|dblpUrl=https://dblp.org/rec/conf/icpm/Hohenadl23
}}
==Noise and Outlier Detection in Task Mining==
https://ceur-ws.org/Vol-3648/paper_4842.pdf
Noise Detection in Task Mining for RPA
Implementation
Tom Hohenadl1,∗
1
Business Administration and Business Informatics, Catholic University of Eichstätt-Ingolstadt, Germany
Abstract
Robotic Process Automation (RPA) has gained significant attention in recent years as a technology
enabling organizations to automate repetitive tasks and improve operational efficiency. Task mining, a
technique used to capture users interactions with software systems, plays a crucial role in understanding
user behavior and identifying automation opportunities in RPA implementations. However, task mining
data often contains noise, which is erroneous or irrelevant action data, that can affect the accuracy and
reliability of analysis results and synthesized bots. This extended abstract presents a research proposal
focused on noise filtering techniques specifically tailored for user interaction data recorded for RPA
implementation.
Keywords
Robotic Process Automation (RPA), Task mining, Noise Filtering
1. Introduction
Task mining is a growing research area and technology domain. It is a part of process mining
and Business Process Management (BPM). Both academia and industry have shown growing
interest in understanding user actions in business processes [1]. Currently, process mining
and BPM techniques are mainly focused on information systems [2]. Most techniques in the
process mining domain focus on static information provided from system logs [3]. Yet, capturing
ad-hoc user actions and user processes is a BPM problem to be solved in future research [1].
Task mining delivers process insights on a user interaction level through recording of clicks or
keystrokes and adds significant benefits for process enhancement [4]. In addition, the analysis
of task mining logs facilitates the implementation of Robotic Process Automation (RPA) scripts,
so called bots. Where currently a lot of manual initiative is necessary for identifying RPA bot
opportunities and their implementation [5], better bots can be created utilizing task mining
methods [6]. Leno et al. [7] combined the interconnection of task mining and RPA by creating
the Robotic Process Mining (RPM) framework.
While task mining focuses on capturing user actions in user interaction logs, discovering user
processes, checking behavior conformance, and enhancing manual processes [8], RPM uses task
mining techniques to generate bots without the need for manual analysis or implementation
ICPM Doctoral Consortium and Demo Track 2023
∗
Corresponding author.
Envelope-Open tom.hohenadl@stud.ku.de (T. Hohenadl)
Orcid 0000-0002-7501-97803 (T. Hohenadl)
© 2023 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
Workshop
Proceedings
http://ceur-ws.org
ISSN 1613-0073
CEUR Workshop Proceedings (CEUR-WS.org)
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
effort, thereby automating user behavior across different information systems [9]. Although
bots created using the available methods can work, they are still susceptible to erroneous or
irrelevant behavior recorded in the logs [10]. This behavior is defined as noise, e.g. attribute
noise [11], and is not relevant for analysis of manual processes or implementation [10]. When
implementing this behavior into bots automatically, the benefits of RPA, such as efficiency and
quality [12], are marginal due to erroneous RPA bot executions, which lead to failures and
increased maintenance effort [13]. These factors diminish the advantages of RPA compared to
manual work.
The consequences of faulty bot implementation, driven by noisy user interaction logs, high-
light two core areas of focus for this research proposal: the identification of noise-indicating
properties within user actions recorded in interaction logs, and the development of methods
for cleaning such noise from these logs. The primary objective of the research is to create an
artifact capable of identifying such noisy actions so that subsequently automated RPA bots are
not affected by erroneous or irrelevant user behavior. In addition, the purpose of this research
project is to evaluate the transformation of actual user actions into automated RPA routines
during bot implementation and to determine the actions that are important for automation.
2. Research Proposal and Research Questions
As outlined by Mayr et al. [8] there are three particular challenges to overcome in task mining
research at the moment: segmentation of logs, privacy preservation of user related informa-
tion, and noise filtering. This proposal focuses on the noise filtering aspects of the described
challenges.
So far, the preprocessing of user interaction logs is performed within the developed RPM
recorders, e.g. actionLogger [14] or smartRPA [9]. These developed Design Science Research
(DSR) artifacts use hard-coded static rules to remove unwanted or incorrect user behavior. On
the one hand, Leno et al. mention exemplary rules, such as repeated navigation or copy actions
without pasting [14]. On the other hand, Agostinelli et al. identify the quality of the recorded
fine-grained data as a weakness of the RPM artifact [9], so that not all contingencies may be
mapped to static rules. Thus, this proposal entails research targeting the detection and removal
of noise in user interaction logs and is grounded on the following research questions:
• What user actions are relevant for RPA?
• How can irrelevant actions be identified and removed after recording users and before
RPA bot creation?
• Do existing algorithms for noise detection from the process mining domain exceed the
capabilities of hard-coded static rule based filters from existing RPM DSR artifacts?
• Is it possible to create a better filtering method based on the results of the previous
research questions?
3. Research Methodology
Initially, a structured literature review based on guidelines by Kitchenham et al. [15] was
conducted. This literature review was used to identify user action properties described in the
existing literature. Based on the acquired data a taxonomy of user actions in manual processes
was created following the directive of Nickerson et al. [16].
Following the literature review and the identification of action categories, a semi-structured
interview study is used to identify translation patterns. A translation is the mapping of user
actions to bot functionality. The interview study is based on the methods described in Saldana
[17]. The guiding questions for the interview is ”Which user actions, intrinsic to users daily
routines, are being incorporated into the operational framework of an RPA bot?”. By answering
this question and encoding the interviewees’ answers, an overview of actions that are valuable
in RPA bots, as well as actions that are nonfunctional, can be created.
Based on the theoretical findings of the first two research questions, a comparative analysis is
conducted to examine existing process mining noise detection methods and their applicability
for task mining. Initially, students at the researchers institute are instructed to perform a range
of manual processes on their digital devices and record the corresponding actions to create a
reference data set. This data set serves as a baseline for the subsequent comparison of algorithms.
By applying Leno’s et al. [18] and Agostinelli’s et al. [19] hard-coded static noise filtering rules
on the data set, an initial base line to remove noise or irrelevant events from the log can be
established. The disparity between the original logs and the filtered logs will subsequently be
employed to identify appropriate noise detection algorithms. A set of outlier and noise detection
algorithms is reviewed by Koschmieder et al. [11] . The set of publicly available algorithms,
i.e. source code or DSR artifacts, will be used as an initial set of comparable algorithms to the
baseline data. Furthermore, a literature review and forward backward search will be used to
identify further relevant noise filtering algorithms.
The final objective of this study is to develop a DSR artifact aimed at eliminating the reliance
on hard-coded filtering. The findings from the comparative study of available algorithms as well
as the interview study will be used to elaborate noise filtering techniques for user interaction
logs. Based on these results, a DSR noise filtering artifact will be developed. The reference data
set will be labeled in collaboration with the students and will be used as a basis for evaluation
using scores such as precision, recall or the F-score. The artifact aims to establish a robust data
preprocessing foundation for the automated creation of RPA bots by leveraging filtered user
interaction logs.
4. Current Research Status
The initial literature review to identify user actions currently processed in task mining and
RPM has been conducted. The result is a taxonomy containing six categories of value adding
user actions and a category for not relevant actions. These value adding actions categories are
opening, navigating, transforming, transferring, concluding and closing actions. Furthermore,
the category of empty actions, i.e. doing nothing at all, was identified. However, a validation of
these categories through practical use cases or empirical evaluation is still necessary.
Following the literature review, the initial two stages of the qualitative interview study have
been accomplished. The first stage involved the formulation of interview questions and the
identification of the target group of participants, specifically focusing on RPA developers and
researchers in the fields of RPM or task mining. In the second stage, interviews were conducted
with a total of five RPA developers, four task mining researchers and four RPA business experts.
Following the interviews, an initial round of process coding is conducted. This will be followed
by the categorization of the codes and a second cycle coding approach.
5. Conclusion
In the evolving research and industry landscape, the automation of user processes is achieved
using RPA, which has opened up new avenues for improving process efficiency and effectiveness.
RPA implementation, in turn, is further enhanced by leveraging task mining and the RPM
framework. While research on task mining is increasing and industry interest shifts towards
automated automation, the task mining, RPA and RPM domain still has challenges. The described
research proposal aims at improving the automatic creation of RPA bots by removing noise from
user interaction logs. Consequently, this work addresses relevant aspects in the convergence of
task mining, RPA and RPM, bridging gaps in the pursuit of efficient and effective automation
solutions.
References
[1] I. Beerepoot et al, The biggest business process management problems to solve before we
die, Computers in Industry 146 (2023) 103837. doi:10.1016/j.compind.2022.103837 .
[2] M. Dumas, M. La Rosa, J. Mendling, H. A. Reijers, Fundamentals of business process
management, softcover re-print of the hardcover 2nd edition 2018 ed., Springer, Berlin,
2018. URL: https://link.springer.com/book/10.1007/978-3-662-56509-4.
[3] W. van der Aalst et al, Process mining manifesto, in: F. Daniel, K. Barkaoui, S. Dustdar
(Eds.), Business Process Management Workshops, Lecture Notes in Business Information
Processing, Springer, Berlin, 2012, pp. 169–194.
[4] W. van der Aalst, Process mining - Data science in action, second edition ed., Springer,
Berlin and Heidelberg, 2016. doi:10.1007/978- 3- 662- 49851- 4 .
[5] W. van der Aalst, On the pareto principle in process mining, task mining, and robotic
process automation, DATA 2020 - Proceedings of the 9th International Conference on
Data Science, Technology and Applications (2020). doi:10.5220/0009979200050012 .
[6] S. Agostinelli, A. Marrella, M. Mecella, Research challenges for intelligent robotic process
automation, Business Process Management Workshops, BPM 2019 (2019) 12–18. URL:
doi.org/10.1007/978-3-030-37453-2_2.
[7] M. Dumas, M. La Rosa, V. Leno, A. Polyvyanyy, F. M. Maggi, Robotic process mining, in:
W. M. P. van der Aalst, J. Carmona (Eds.), Process Mining Handbook, Springer International
Publishing, Cham, 2022, pp. 468–491. URL: doi.org/10.1007/978-3-031-08848-3_16.
[8] A. Mayr, L.-V. Herm, J. Wanner, Christian Janiesch, Applications and challenges of
task mining: A literature review, ECIS 2022 Research-in-Progress Papers (2022). URL:
https://aisel.aisnet.org/ecis2022_rip/55.
[9] S. Agostinelli, M. Lupia, A. Marrella, M. Mecella, Automated generation of executable rpa
scripts from user interface logs, in: A. Asatiani (Ed.), BPM 2020 Blockchain and RPA Forum
Proceedings, volume 393 of Lecture Notes in Business Information Processing, Springer Inter-
national Publishing, Cham, 2020, pp. 116–131. URL: doi.org/10.1007/978-3-030-58779-6_8.
[10] S. Agostinelli, M. Lupia, A. Marrella, M. Mecella, Smartrpa: A tool to reactively synthesize
software robots from user interface logs, in: S. Nurcan, A. Korthaus (Eds.), Intelligent
Information Systems, volume 424 of Lecture Notes in Business Information Processing,
Springer International Publishing and Imprint Springer, Cham, 2021, pp. 137–145. URL:
doi.org/10.1007/978-3-030-79108-7_16.
[11] A. Koschmider, K. Kaczmarek, M. Krause, S. J. van Zelst, Demystifying noise and out-
liers in event logs: Review and future directions, in: A. Marrella, B. Weber (Eds.),
Business Process Management Workshops, volume 436 of Springer eBook Collection,
Springer International Publishing and Imprint Springer, Cham, 2022, pp. 123–135. URL:
doi.org/10.1007/978-3-030-94343-1_10.
[12] A. Meironke, S. Kühnel, How to measure rpa’s benefits? a review on metrics, indicators, and
evaluation methods of rpa benefit assessment, in: Wirtschaftsinformatik 2022 Proceedings,
2022, pp. 1–19. URL: https://aisel.aisnet.org/wi2022/bpm/bpm/5.
[13] P. Noppen, I. Beerepoot, I. van de Weerd, M. Jonker, H. A. Reijers, How to keep rpa
maintainable?, in: D. Fahland, C. Ghidini, J. Becker, M. Dumas (Eds.), Business Process
Management, volume 12168 of Springer eBook Collection, Springer International Publishing
and Imprint Springer, Cham, 2020, pp. 453–470. URL: doi.org/10.1007/978-3-030-58666-9_
26.
[14] V. Leno, A. Polyvyanny, M. Dumas, F. Maggi, Action logger: Enabling pro-
cess mining for robotic process automation, in: Proceedings of the Dis-
sertation Award, Doctoral Consortium, and Demonstration Track at BPM 2019,
2019, pp. 124–128. URL: https://bia.unibz.it/esploro/outputs/conferenceProceeding/
Action-logger-Enabling-process-mining-for/991006186495001241.
[15] B. Kitchenham, Procedures for performing systematic reviews, Keele, UK, Keele Univ 33
(2004). URL: http://www.inf.ufsc.br/~aldo.vw/kitchenham.pdf.
[16] R. C. Nickerson, U. Varshney, J. Muntermann, A method for taxonomy development and
its application in information systems, European Journal of Information Systems 22 (2013)
336–359. doi:10.1057/ejis.2012.26 .
[17] J. Saldaña, The coding manual for qualitative researchers, 4e ed., SAGE, Los Angeles and
London and New Delhi and Singapore and Washington DC and Melbourne, 2021.
[18] V. Leno, A. Polyvyanyy, M. La Rosa, M. Dumas, F. M. Maggi, Action logger: Enabling
process mining for robotic process automation, in: Proceedings of the Dissertation Award,
Doctoral Consortium, and Demonstration Track at BPM 2019, volume 2420 of CEUR
Workshop Proceedings, CEUR-WS, 2019, p. 5. URL: https://hdl.handle.net/10863/19700.
[19] S. Agostinelli, A. Marrella, M. Mecella, Exploring the challenge of automated segmentation
in robotic process automation, in: Research Challenges in Information Science 2021, 2021,
pp. 38–54. URL: doi.org/10.1007/978-3-030-75018-3_3.