Recommender Systems in Business Process Management: A Systematic

Literature Review

Sebastian Petter and Stefan Jablonski

University of Bayreuth, Bayreuth, Germany

Keywords:

Business Process Management, Recommender System, Systematic Literature Review, Goal-Based Process

Improvement.

Abstract:

Recommender systems have the potential to enhance decision-making and to improve business process exe-

cution in the domain of Business Process Management (BPM). By analyzing data and providing personalized

recommendations, these systems can assist users in making profound decisions and so foster the achievement

of their business goals. In our study that is based on the PRISMA (Preferred Reporting Items for Systematic

reviews and Meta-Analyses) methodology, we examine the usage of recommender systems in BPM, focusing

on the objectives, methods, and input data utilized. We searched eight databases and included papers that focus

on process execution and recommendation methods while excluding those that are not digitally available, not

in English, patents, miscellany, or proceedings, or focused solely on business process modeling. This results

in 33 papers, addressing the research questions, that are analysed in detail. The discussion highlights research

gaps related to user preferences and input data, suggesting that further investigation is needed to enhance the

effectiveness of recommender systems in business process management.

1 INTRODUCTION

The core elements of Business Process Management

(BPM) are processes that reﬂect sequences of work-

ing steps having to be executed in a speciﬁc order

(Dumas et al., 2018; Weske, 2007). For process mod-

elling process models in a business process modelling

language are deﬁned. Each working step is repre-

sented by an activity within a process model. Fur-

thermore, control ﬂow (i.e., the order of the activi-

ties), data used, and resources eligible to execute the

activities are speciﬁed in the process model. Process-

aware Information Systems (PAISs) interpret process

models and execute them (Reichert and Weber, 2012).

A PAIS determines which activities to be performed

next can be executed next by what process partici-

pant. Tasks ready for execution are provided to the

process participants in so-called worklists. Once a

task is completed, information about its execution is

stored in a process event log (van der Aalst, W, 2016).

A process event comprises information about the task

itself, the resource, i.e. process participant, executed

the task, and further details like date, duration, and

optional customer data. Events belonging to the ex-

ecution of a speciﬁc instance of a business process

constitute a so-called (event) trace.

Since all kinds of enterprises rely on BPM pro-

cess improvement is a major success factor for them.

Various approaches for process improvement have

been developed. Typically, they aim to optimize

a business process execution in respect of certain

goals. Respective goals are reduction of process cy-

cle time, improvement of process outcomes, or in-

crease of user satisfaction (Jablonski and Bussler,

1996; Koulopolous, 1995; Lawrence, 1997; Petter

et al., 2022). It is common practice to deploy recom-

mendations in PAISs that support process participants

to better achieve those goals.

Recommender systems are mainly known from

e-commerce. A recommender system is a software

system that provides personalized recommendations

to users (Aggarwal, 2016). They have become in-

creasingly popular in e-commerce due to their poten-

tial to increase customer engagement and sales and

to improve customer experience. Three main meth-

ods have been developed to generate recommenda-

tions: collaborative ﬁltering, content-based ﬁltering,

and knowledge-based ﬁltering. Collaborative ﬁlter-

ing relies on users’ past explicit (e.g., issuing ”stars”)

and implicit (e.g. chosen actions) ratings and based

on that constitutes classes of users with similar pref-

erences. Personalized recommendations are then de-

Petter, S. and Jablonski, S.

Recommender Systems in Business Process Management: A Systematic Literature Review.

DOI: 10.5220/0012039500003467

In Proceedings of the 25th International Conference on Enterprise Information Systems (ICEIS 2023) - Volume 2, pages 431-442

ISBN: 978-989-758-648-4; ISSN: 2184-4992

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

431

rived from the behavior of an associated user class. In

content-based ﬁltering attributes of things (e.g. activ-

ities) a user likes are referred to as ”content”. A rec-

ommender system recommends similar things to user,

i.e. things that coincide in all/most of the content at-

tributes. Knowledge-based ﬁltering uses a user proﬁle

and evaluates whether a particular item is likely to be

of interest to a user, again based on the contextual at-

tributes of items. Furthermore, these three methods

can mutually be combined, forming what is known as

hybrid ﬁltering.

Over time, different methods of recommender sys-

tems have been integrated into BPM. These methods

vary concerning - among others - the objectives they

are aiming at, the input data they are exploiting, and

the algorithms they install. In some cases, different

methods aim at achieving the same objective but in

different ways.

Currently, there are two systematic literature re-

views regarding recommendations in BPM (Kubrak

et al., 2021; Yari Eili and Rezaeenour, 2022). Kubrak

et al. consider the recommendation of interventions

at runtime to prevent negative outcomes or poorly

performing cases. That means, this review focuses

on predictive process monitoring. They provide an

overview of recent research papers regarding per-

formance objectives with a focus on control ﬂow,

resource allocation, and other perspectives (Kubrak

et al., 2021). In a second literature review, Yari Eili

and Jalal Rezaeenour focus on the usage of event logs

as source for recommendations. These approaches

rely on process mining techniques to exploit event

log information. The authors discuss different types

of recommendations and evaluation metrics applied

in the considered research papers (Yari Eili and Reza-

eenour, 2022).

We broaden the perspective of a systematic litera-

ture review compared with the above introduced pre-

decessor reviews. We neither restrict ourselves to pre-

dictive process monitoring as objective, nor do we ex-

clusively focus on event logs as information resource.

To the best of our knowledge, we present a ﬁrst sys-

tematic literature review on the application of recom-

mender systems in the realm of BPM without any re-

strictions. To address this we focus on the following

research questions:

. What is the objective of applying recommen-

dation methods, respectively recommender sys-

tems, in the context of BPM?

. What data provided by a PAIS is used to gener-

ate recommendations?

. Which methods known from RS are applied?

focuses on the objectives the approaches fos-

ter applying RS during process execution. Many

research papers focus on optimizing process perfor-

mance (e.g., shortening processing time, increasing

quality), whereas few papers consider other goals like

improving user satisfaction (Petter et al., 2022). RQ

and RQ

consider the methods used to generate rec-

ommendations and the input input data they require.

For example, the trace of a currently executed process

instance can be used, as well as the complete event

log provided so far by the PAIS.

To answer the above research questions, we con-

duct a systematic literature review following the Pre-

ferred Reporting Items for Systematic Reviews and

Meta-Analysis (PRISMA) method (Page et al., 2021)

as far as applicable. After establishing the research

questions, the search strategy to identify research pa-

pers is deﬁned, including all accessed databases and

applied search queries. Furthermore, inclusion and

exclusion criteria are deﬁned to qualify the results of

such queries. Extracting data from the found records

leads to a detailed analysis of the remaining records.

The last step is the interpretation of the papers accord-

ing to the research questions.

The remainder of the paper is structured as fol-

lows: Section 2 gives a detailed overview of the meth-

ods used to identify relevant research papers. The

step-wise execution of the literature review and the

obtained results are presented in Section 3. Section 4

provides a discussion of the results, and ﬁnally, Sec-

tion 5 concludes the systematic literature review and

highlights essential directions for future research.

2 METHODS

This paper describes a systematic literature review

conducted according to PRISMA guidelines (Page

et al., 2021) to identify and analyse existing research

integrating recommender systems into BPM systems.

More speciﬁcally, we investigate why recommender

systems are used in the context of BPM (RQ

), what

data is used (RQ

), and which methods known from

genuine recommender systems come into use (RQ

The PRISMA method includes guidance on plan-

ning, conducting, reporting, and disseminating sys-

tematic reviews and meta-analyses transparently. It

involves formulating an appropriate research ques-

tion, performing comprehensive searches, selecting

studies that meet predeﬁned inclusion criteria, ex-

tracting relevant data from each study, synthesizing

results, and interpreting results critically with respect

to limitations of included studies and the quality of

evidence provided by them. This process allows re-

searchers to draw reliable conclusions about existing

ICEIS 2023 - 25th International Conference on Enterprise Information Systems

432

Table 1: The search queries applied.

ID Search Query

”Recommender System” AND

”Business Process Management”

”Recommender Systems” AND

”Business Process Management”

”Recommendation System” AND

”Business Process Management”

”Recommendation Systems” AND

”Business Process Management”

evidence on any given topic while minimizing bias in

the review process.

According to the PRISMA checklist, ﬁrst, we de-

ﬁne the databases to access. A total of eight different

data sources are searched to ﬁnd as many relevant re-

search as possible: ACM Digital Library, Emerald,

Google Scholar, IEEE Explore, ScienceDirect, Sco-

pus, SpringerLink, and Web Of Science. We apply

the software Publish or Perish (Harzing, 2007) for

the search with Google Scholar and Scopus. Since

Google Scholar limits the number of results for each

query to at most 1000 results, we split the query

into multiple queries according to different periods

and uniﬁed them afterward. The queries posted are

shown in table 1. Notice that we post four sepa-

rated queries for every database instead of connect-

ing the keywords by the logical operator OR since

some databases cannot handle this operator properly.

Furthermore, a distinction is made between Recom-

mender System and Recommender Systems, respec-

tively Recommendation System and Recommendation

Systems, as many databases provide different results

due to the plural form of the terms.

Second, eligibility criteria (e.g., inclusion and ex-

clusion criteria) must be deﬁned. In our literature re-

view, we address the objectives of the considered lit-

erature, the data used, and the methods applied. From

this basis, two inclusion criteria are derived:

. Process execution

. Recommendation methods (content-based ﬁlter-

ing, collaborative ﬁltering, knowledge-based ﬁl-

tering, hybrid ﬁltering)

Inclusion criterion IC

requires that papers con-

sidering process execution must be included. Also,

all papers that explicitly apply certain methods from

the domain of recommender systems to the domain of

BPM must be taken into account (IC

Furthermore, the following exclusion criteria are

deﬁned to exclude papers from the evaluation:

. Not digitally available

. Not written in English

. Patents, miscellany, and proceedings

. Only one domain is considered: BPM or RS

. Business process modeling optimization

Exclusion criteria EC

and EC

ensure that papers

are generally available and are written in English, so

they can be accessed by the majority of researchers.

says that patents, miscellany, and proceedings

”as a whole” are excluded. Nevertheless this does

not exclude the contents of these sources. For exam-

ple, papers of a proceeding are included as long as

they fulﬁll the inclusion and exclusion criteria. Exclu-

sion criterion IC

stipulates that both domains must

be considered, BPM and recommendation systems.

This criterion excludes papers that only cover one of

the two domains. Finally, EC

excludes those papers

that solely focus on optimization of process modeling

without looking at process execution.

In summary, the inclusion criteria IC

and IC

provide the basis for our systematic literature anal-

ysis, while the exclusion criteria EC

to EC

help to

narrow down the scope and focus of the study.

After having deﬁned search strategy and inclu-

sion and exclusion criteria (eligibility criteria), re-

spectively, we deﬁne our selection process. We

ﬁrst capture the metadata of a publication (title, au-

thor, publication venue, year, keywords, year, DOI).

Based on the metadata, duplicate entries are removed

semi-automatically (tool support from Microsoft Ex-

cel (Microsoft Corporation, 2022)). In order to con-

sider a found record in future steps, the record must

not violate the exclusion criteria. To ensure this, all

authors of this study check the title of each record in-

dependently. The resulting records are checked ac-

cording to their abstract afterwards, again indepen-

dently by each author of this paper. The records left

are entirely read by the authors. Depending on the

content, they are excluded because they violate one or

more exclusion criteria, or they are sorted into a table

and classiﬁed according to their objectives, data used,

and methods applied. To identify even more relevant

research, we conduct backward referencing (snow-

balling) to the papers found (Okoli and Schabram,

2010).

3 RESULTS

3.1 Applying the PRISMA Method

To get an overview of the existing research combin-

ing recommender systems and BPM, we applied the

PRISMA method described in Section 2. This method

comprises a ﬂow diagram (Page et al., 2021) that pro-

vides a visual representation of the search process and

Recommender Systems in Business Process Management: A Systematic Literature Review

433

the number of studies included or excluded at each

stage. It offers a rigorous and transparent view on

the literature ﬁnding and selecting process and clearly

illustrates the results of the systematic review. The

adapted PRISMA ﬂow diagram for our study is shown

in Figure 1.

Figure 1: PRISMA ﬂow diagram of our study.

Identiﬁcation. The search process is initiated by

searching eight different databases: ACM, Emerald,

Google Scholar, IEEE Xplore, Science Direct, Sco-

pus, Springer, and WebOfScience. For each database

we run the four queries introduced in Table 1. On Jan-

uary 10th, 2023 we conducted our search and iden-

tiﬁed a total of 8780 records. Table 2 depicts de-

tailed results of the query execution. The columns

represent the search queries, the rows identify the dif-

ferent databases considered. Each cell shows the re-

sults of query execution for the particular databases.

Notice that the number of entries found for Recom-

mender System and Recommender Systems, respec-

tively, and for Recommendation System and Recom-

mendation Systems, respectively, is the same for the

databases Emerald, IEEE Xplore, Science Direct, and

Scopus, whereas in the remaining databases differ-

ent numbers of entries are found for the singular and

plural forms of these terms. This highlights the im-

portance of applying a consistent and standardized

search strategy, as different databases may have dif-

ferent ways of indexing and categorizing literature.

According to the PRISMA method in the next step the

8780 papers have to be screened for duplicate entries.

4567 of these records are identiﬁed as duplicates and

are therefore excluded, leaving 4213 unique records.

Screening. Out of these, 65 are not available in digital

form and are also excluded from the study (EC

). The

remaining 4148 papers are subject to further screen-

ing. Another 300 papers are excluded since they are

not written in English (EC

). 892 additional entries

are excluded because they are either patents, miscella-

nies, or proceedings (EC

). 2872 papers are excluded

as they either do not consider both, BPM and recom-

mender systems, or have an inadequate focus on these

topics (EC

). Finally, 54 papers are excluded as they

focus on recommender systems in the context of pro-

cess modeling, which is also outside the scope of the

present study (EC

). This results in 30 papers that are

included in our study.

Included. A total of 3 papers are added through back-

ward referencing, resulting in a ﬁnal list of 33 papers

to be analyzed in this research.

3.2 Classiﬁcation of the Result

For a better overview of the 33 papers we have an-

alyzed and classiﬁed them as depicted in Table 3.

We identiﬁed three superior columns referring to the

research questions deﬁned in Section 1: Objectives

(RQ

), Input Data (RQ

), and Methodology (RQ

Each row of the table represents a study or a paper

to be analyzed.

Objectives. The Objectives column lists the most

frequently mentioned objectives that the reviewed pa-

pers are aiming at when applying recommender sys-

tems in the BPM context. Process performance goals,

such as shortening processing time and minimizing

costs, are the most frequently mentioned goals. Addi-

tionally, the papers consider ﬂexible goals that can be

customized based on a target function provided. For

example, a ﬂexible goal may aim at the optimization

of different KPIs, such as process completion time,

resource utilization, and customer satisfaction. Often,

process performance goals and ﬂexible goals over-

lap. Enhancing user experience and user satisfaction

is one of the original goals of recommender systems.

This goal is still relevant in the ﬁeld of Business Pro-

cess Management and is pursued by researchers. The

column Others summarizes goals that do not ﬁt into

the above categories. For example, a paper may aim

at the reduction of process errors.

Input Data. The Input Data column of the table lists

the different types of data exploited by the reviewed

papers as input for recommendations. The data listed

in this column can be divided into six sub-areas:

Event log, partial traces, worklist, resource informa-

tion, process model, and others. Many approaches

require the event log, which records the events that

occur in a business process, to generate recommen-

ICEIS 2023 - 25th International Conference on Enterprise Information Systems

434

Table 2: Published articles in journals and conferences.

Sum

ACM DL 16 37 18 23 94

Emerald 54 54 48 48 204

Google Scholar 961 1442 1293 792 4488

IEEE XPlore 199 199 576 576 1550

ScienceDirect 81 81 62 62 286

Scopus 24 24 15 15 78

SpringerLink 547 563 474 469 2053

Web Of Science 8 7 6 6 27

Sum 1890 2407 2492 1991 8780

dations. The event log can be extended to include

additional information such as process attributes or

resource preferences (Petter et al., 2022). Addition-

ally, partial traces (e.g., sequences of events) can be

used to generate recommendations based on the his-

tory of events in a process. A worklist contains in-

formation about the tasks that must be performed in a

business process. Recommendations can be generated

based on the information contained in such a worklist.

Furthermore, resource information, such as the avail-

ability of a resource, is often necessary to generate

adequate recommendations. A process model (con-

taining information about the activities that need to

be performed) and additional attributes, such as con-

textual information for activities, are used in some

reviewed papers to make recommendations. For in-

stance, recommendations are derived that are based

on the similarity of these task attributes. Any other

input data that does not ﬁt into the previously men-

tioned categories is grouped under Others. The use

of different types of input data can signiﬁcantly inﬂu-

ence the accuracy and efﬁciency of the recommenda-

tions. Furthermore, the methodology used to calcu-

late recommendations is depending on the input data

provided.

Methodology. The Methodology column of the ta-

ble categorizes the different methods applied in the

reviewed papers for recommendations in the ﬁeld of

BPM. Content-based ﬁltering uses the attributes of the

items being recommended (e.g., tasks in a business

process) and the execution history of the user consid-

ered to generate recommendations. A collaborative

ﬁltering method generates recommendations based on

users’ past behavior and preferences. It assumes that

users who have similar preferences in the past will

have similar preferences in the future. Knowledge-

based ﬁltering is based on knowledge representation

and reasoning to generate recommendations. It can

take into account various factors, such as user pro-

ﬁles. As the name suggests, hybrid ﬁltering combines

the advantages of multiple ﬁltering methods to gener-

ate recommendations. For example, a hybrid method

uses content-based ﬁltering to generate initial recom-

mendations and then reﬁne these recommendations

using collaborative ﬁltering. In addition to these stan-

dard methods, the column Others accommodates all

papers that do not ﬁt into the previously mentioned

categories.

3.3 Presentation of the Classiﬁed

Results

In the following section the results are discussed re-

garding their objectives. In this study, the authors an-

alyzed a total of 15 papers with the aim of optimizing

process performance goals. Out of the analyzed pa-

pers, 100% aim at shortening processing time, 40%

aim at minimizing costs, and 33% aim at improving

the quality of process outcome. Furthermore, 13 pa-

pers consider ﬂexible optimization goals, four papers

regard user preferences as optimization goal, and 14

papers are assigned to the column of ”other” objec-

tives.

Shorten Processing Time. The majority of the

papers focuses on the optimization of process per-

formance goals by shortening processing time (15

papers). This list comprises (Arias et al., 2016),

(Schobel and Reichert, 2017), (van der Aalst et al.,

2010), (van der Aalst et al., 2010), (Setiawan and

Sadiq, 2011), (Setiawan et al., 2011), (Agarwal et al.,

2022), (Branchi et al., 2022), (Huber et al., 2015),

(Weinzierl et al., 2020), (Bozorgi et al., 2021), (Scho-

nenberg et al., 2008), (Barba et al., 2012), (Barba

et al., 2013), (Aalst et al., 2009), and (Haisjackl and

Weber, 2011). These papers analyse an event log to

extract the potential process execution time. Some

papers, such as (Barba et al., 2012) and (Barba et al.,

2013), use the event log to derive information about

contextual information like execution times for the

single activities in a process model. Enacting the

model with such information during build time leads

to simple generation of recommendations during run-

time. During process execution, the trace with the

shortest remaining processing time can be recom-

Recommender Systems in Business Process Management: A Systematic Literature Review

435

Table 3: Overview of the analyzed research papers.

Research Paper Objectives Input Data Methodology

Shorten processing time

Minimize costs / Increase proﬁt

Quality

Flexible goal

User experience

Others

(Enhanced) Event log

Partial traces

Worklist

Resource Information

Process model

Others

Content-based ﬁltering

Collaborative ﬁltering

Knowledge-based ﬁltering

Hybrid ﬁltering

Others

(Yang et al., 2017) • • • •

(Arias et al., 2016) • • • • • • • •

(Pika and Wynn, 2021) • • •

(Arias et al., 2017) • • • •

(Schobel and Reichert, 2017) • • • • •

(Conforti et al., 2015) • • • •

(Zhao et al., 2016) • • • •

(van der Aalst et al., 2010) • • • •

(Petter et al., 2022) • • • • •

(Di Valentin et al., 2014) • • • •

(Khan et al., 2021) • • • •

(Leoni et al., 2020) • • • •

(Setiawan and Sadiq, 2011) • • • • • • • • •

(Setiawan et al., 2011) • • • • • • • • •

(Mertens et al., 2015) • • • •

(Agarwal et al., 2022) • • • • • • •

(Arias et al., 2018) • • • • •

(Branchi et al., 2022) • • • • • •

(Huber et al., 2015) • • • •

(Liu and Wu, 2018) • • • •

(Bidar et al., 2019) • • • •

(Gr

oger et al., 2014) • • • •

(Weinzierl et al., 2020) • • • • • •

(Bozorgi et al., 2021) • • • •

(Arias et al., 2016) • • • •

(Schonenberg et al., 2008) • • • • • • •

(Barba et al., 2012) • • • • • • •

(Trabelsi et al., 2021) • • • •

(Barba et al., 2013) • • • • • • •

(Pika and Wynn, 2020) • • •

(Aalst et al., 2009) • • • • • •

(Cabanillas et al., 2013) • • • • • •

(Haisjackl and Weber, 2011) • • • • • • • •

mended. The remaining 13 papers are calculating rec-

ommendations by deriving the potential process ex-

ecution time by comparing the current partial trace

with similar traces in the event log. In addition to

the event log and the partial trace, some papers con-

sider contextual information such as user expertise

(Arias et al., 2016; Setiawan and Sadiq, 2011; Seti-

awan et al., 2011), resource availability (Arias et al.,

2016; Barba et al., 2012; Barba et al., 2013), and ex-

plicit domain knowledge (Aalst et al., 2009) to opti-

mize the process execution time. None of the papers

regarding shortening processing time can be assigned

to the methodology of content-based ﬁltering or hy-

brid ﬁltering. The method of collaborative ﬁltering is

used in (Setiawan and Sadiq, 2011), (Setiawan et al.,

2011), and (Branchi et al., 2022), where the executed

events of the current user are compared to the exe-

cuted events of other users. Knowledge-based ﬁlter-

ICEIS 2023 - 25th International Conference on Enterprise Information Systems

436

ing can be found in (Barba et al., 2012) and (Barba

et al., 2013). Both papers enhance the process model

with contextual information providing recommenda-

tions during runtime. The remaining papers in the

section apply methods that cannot be assigned to the

standard methods known from recommender systems

since they are not user-speciﬁc.

Flexible Goals. Flexible goals that can be deﬁned

during build- or run-time have become increasingly

important for many organizations due to the chang-

ing nature of business requirements and goals. To

achieve these ﬂexible goals, different techniques and

approaches have been proposed and investigated in

the literature. In this study, we review 13 papers that

consider ﬂexible goals in their research and provide a

summary of the techniques and methods used: (Yang

et al., 2017), (Arias et al., 2016), (Khan et al., 2021),

(Leoni et al., 2020), (Setiawan and Sadiq, 2011), (Se-

tiawan et al., 2011), (Agarwal et al., 2022), (Branchi

et al., 2022), (Liu and Wu, 2018), (Weinzierl et al.,

2020), (Schonenberg et al., 2008), and (Haisjackl and

Weber, 2011). Most of the papers reviewed use Key

Performance Indicators (KPIs) to deﬁne their objec-

tives. These KPIs are metrics that are used to mea-

sure the performance of a process, system, or orga-

nization. For instance, minimizing costs (Agarwal

et al., 2022; Branchi et al., 2022), shortening pro-

cessing time (Agarwal et al., 2022; Branchi et al.,

2022; Weinzierl et al., 2020), and increasing customer

satisfaction (Setiawan and Sadiq, 2011) are some of

the common KPIs used to deﬁne objectives in the

reviewed papers. However, in (Schonenberg et al.,

2008), the objective is deﬁned via a target function.

To calculate recommendations for achieving the ob-

jectives, all papers use event log information. Some of

the papers use an enhanced event log that includes ad-

ditional information, such as customer satisfaction, to

generate more accurate recommendations. For exam-

ple, (Leoni et al., 2020), include customer satisfaction

information in their enhanced event log to generate

recommendations that can improve customer satisfac-

tion. Partial traces are used by 70% of the papers to

generate recommendations (Yang et al., 2017; Arias

et al., 2016; Leoni et al., 2020; Setiawan and Sadiq,

2011; Setiawan et al., 2011; Agarwal et al., 2022;

Branchi et al., 2022; Weinzierl et al., 2020; Schonen-

berg et al., 2008; Aalst et al., 2009; Haisjackl and We-

ber, 2011). By analyzing partial traces, recommenda-

tions can be generated that can help achieve the objec-

tives deﬁned in the KPIs. Worklists of current users

are also used in some papers to generate recommen-

dations (Khan et al., 2021; Setiawan and Sadiq, 2011;

Setiawan et al., 2011; Agarwal et al., 2022; Scho-

nenberg et al., 2008; Haisjackl and Weber, 2011).

Since worklists contain information about the tasks, a

user can potentially work on, this information can be

used to generate recommendations that can help the

user achieve their objectives. Less frequently consid-

ered input data for generating recommendations dur-

ing process execution are business data (Liu and Wu,

2018), domain knowledge (Aalst et al., 2009), and re-

source information (Setiawan and Sadiq, 2011; Seti-

awan et al., 2011). For example, (Aalst et al., 2009)

use domain knowledge to generate recommendations

that can improve the quality of a process. Collabora-

tive ﬁltering is a method used by some of the papers

to generate recommendations. Collaborative ﬁltering

is a technique that is commonly used in recommender

systems to provide recommendations to users based

on the behavior of similar users. In the context of

ﬂexible goals, collaborative ﬁltering can be used to

provide recommendations to users based on the be-

havior of other users who have similar objectives. For

instance, (Setiawan and Sadiq, 2011), (Setiawan et al.,

2011), and (Branchi et al., 2022) use collaborative ﬁl-

tering to generate recommendations. Finally, some of

the papers use methods that cannot be assigned to the

standard methods known from recommender systems.

Minimizing Costs. The objective of minimizing

costs during process execution is in the focus of six re-

search papers, namely (Arias et al., 2016), (Setiawan

and Sadiq, 2011), (Setiawan et al., 2011), (Branchi

et al., 2022), and (Aalst et al., 2009). The com-

mon approach adopted by these papers is to use event

logs and partial traces to generate recommendations.

While some of the papers consider the current work-

list of the user such as (Setiawan and Sadiq, 2011),

(Setiawan et al., 2011), and (Haisjackl and Weber,

2011), others like (Yang et al., 2017) incorporate in-

formation about resources to calculate recommenda-

tions. It is worth noting that the methods used by

these papers cannot be directly mapped to the existing

methods known from recommender systems. Only

three papers, namely (Setiawan and Sadiq, 2011),

(Setiawan et al., 2011), and (Branchi et al., 2022), can

be classiﬁed under the collaborative ﬁltering method.

Quality. The key research area of (Setiawan and

Sadiq, 2011), (Setiawan et al., 2011), (Agarwal et al.,

2022), (Schonenberg et al., 2008), and (Haisjackl and

Weber, 2011) is to improve the quality of the pro-

cess outcome. The authors of these research papers

analyse event logs and partial traces to generate rec-

ommendations, while also making use of the cur-

rent user’s worklist. (Setiawan and Sadiq, 2011) and

(Setiawan et al., 2011) deploy methods of collabora-

tive ﬁltering to improve process outcomes, applying

this technique to event logs and partial traces to de-

velop recommendations for process improvement. On

Recommender Systems in Business Process Management: A Systematic Literature Review

437

the other hand, the other papers utilize non-standard

recommendation methods to achieve their objectives.

These papers suggest that there are several methods

that can be used to generate recommendations for pro-

cess improvement, and that different methods may be

more efﬁcient in different contexts.

User Experience. The original goal of recommender

systems is to provide personalized recommendations

to users based on their preferences, behaviors, and

context information. In this study, we review four re-

cent studies that focus on improving user experience

in BPM through recommender systems: (Arias et al.,

2017), (Petter et al., 2022), (Di Valentin et al., 2014),

and (Bidar et al., 2019). These studies differ in their

focus, methodology, and application domain, but they

all share the common goal of enhancing user experi-

ence through personalized recommendations. The re-

search papers (Arias et al., 2017), (Petter et al., 2022),

and (Bidar et al., 2019) interpret event logs to gener-

ate recommendations for users. Event logs are used

to infer user preferences and behavior patterns. (Pet-

ter et al., 2022) enhances the event log with explicit

user preferences which are collected through surveys

or questionnaires. This approach allows for a more

accurate representation of user preferences and leads

to better recommendations. (Arias et al., 2017), (Pet-

ter et al., 2022), and (Bidar et al., 2019) consider ad-

ditional information beyond the event log to improve

recommendations. (Arias et al., 2017) incorporates

resource information, such as availability and cost, to

ﬁlter out irrelevant recommendations. (Petter et al.,

2022) and (Bidar et al., 2019) analyse the worklist of a

current user, which contains tasks that the user can ex-

ecute, to generate recommendations that are aligned

with the user’s goals and priorities. In addition to

these approaches (Petter et al., 2022) examines ad-

ditional attributes of activities to calculate similar ac-

tivities. This approach allows for more ﬁne-grained

recommendations that take into account the speciﬁc

characteristics of different activities, such as their du-

ration, location, or complexity. (Bidar et al., 2019)

also extracts implicit user preferences from the event

log, such as the frequency or actuality of certain ac-

tions, to recommend activities that are more likely to

be of interest to the user. The fourth study reviewed,

(Di Valentin et al., 2014), focuses on supporting em-

ployees in carrying out business processes under the

consideration of their personal proﬁle and context in-

formation. This study deploys a knowledge-based ﬁl-

tering approach, which involves the use of domain

knowledge to make recommendations. Speciﬁcally,

the study considers the employee’s role, skills, and

context information to generate personalized recom-

mendations.

Other. Objectives, summarized in the column Others

are considered in the following. One of the considered

objectives during process execution is the allocation

of work teams. (Arias et al., 2017) propose a content-

based ﬁltering approach to dynamically allocate work

teams based on historical process execution data and

expertise information. Another important objective is

to maximize ﬂexibility. (Mertens et al., 2015) pro-

pose a technique that allows the user to retain max-

imum ﬂexibility by considering partial traces and a

declarative process model. This technique enables

users to divert from recommendations later on and

to perform tasks in their own preferred way. (Arias

et al., 2018) and (Arias et al., 2016) propose tech-

niques to recommend the best ﬁtting resource for a

task. In (Arias et al., 2018), the authors use contex-

tual information, historical information, and weights

to calculate recommendations using content-based ﬁl-

tering methods. In (Arias et al., 2016), the authors

use resource information (e.g., expertise) and histori-

cal information about past process executions to make

recommendations. These approaches can help orga-

nizations to allocate resources more effectively and

reduce operational costs. Finally, another important

objective during process execution is to reduce the

risk of process failure. (Conforti et al., 2015) and

(Gr

oger et al., 2014) propose techniques to avoid pre-

dicted metric deviation, such as a process running out

of time. The goal is achieved by comparing running

process instances with historical process data and rec-

ommending next best actions according to the results.

These approaches can help organizations to avoid pro-

cess failure and improve their overall efﬁciency.

4 DISCUSSION

The integration of recommender systems in the con-

text of BPM has gained signiﬁcant interest in recent

years due to its potential to optimize process perfor-

mance goals. In the context of BPM, these systems

can be used to improve process performance by sug-

gesting the best possible next steps for the process.

This literature review provides an overview of ex-

isting approaches using recommender systems in the

context of BPM. The overview unveils several gaps

and associated implications for future research.

The main objective of the reviewed studies is

to optimize process performance goals using recom-

mender systems. The primary focus is on shortening

processing time; however, other performance objec-

tives are represented in only a few examples. The

use of various methods and techniques highlights the

diversity of research in this area and the potential

ICEIS 2023 - 25th International Conference on Enterprise Information Systems

438

for new and innovative approaches to be developed.

While there is still much work to be done, the re-

search conducted recently provides a strong founda-

tion for further exploration and development in this

exciting ﬁeld. The results from this study suggest

that using event logs and partial traces can effec-

tively optimize process execution regarding different

goals. While the methods used by most papers can-

not be directly mapped to existing methods known

from recommender systems, content-based ﬁltering is

the most widely used approach for the remaining pa-

pers. We observe that the majority of previous studies

use the event log to calculate recommendations. Ac-

cording to the domain of recommender systems, de-

pending on historical information is not good at all

since often there is no historical information available

(cold-start problem).

The review highlights several gaps in the current

literature. First, there needs to be more common ter-

minology for using recommender systems in BPM re-

search. Second, the majority of methods in the ﬁeld

aim to improve processes along the temporal perspec-

tive (e.g., cycle time, processing time, deadline viola-

tions); other performance dimensions are represented

in only a few examples. Thus, another research direc-

tion is to investigate other performance objectives that

could be enhanced via recommender systems. Third,

recommender systems are essential for businesses that

aim to improve the user experience. The primary goal

of these systems is to provide personalized recom-

mendations to users by analyzing their past behav-

ior, preferences, and interests. However, the use of

recommender systems in BPM is currently limited,

and there is signiﬁcant space for one of the critical

challenges yet to be considered is recommending the

next best actions according to the user preferences. To

achieve this, there is a need for explicit feedback from

users and the use of this feedback to generate rec-

ommendations. According to this feedback, content-

based and collaborative ﬁltering methods might be

used to generate recommendations to improve user

experience during process execution. Another pos-

sibility is using knowledge-based ﬁltering regarding

user-proﬁles and contextual information for the busi-

ness process.

In the following paragraph, we will address and

provide answers to the research questions. Answer-

ing RQ

, we found that most research papers con-

sider process performance goals as main objectives

followed by ﬂexible goals, improving process qual-

ity, and improving user experience. The input data

the most analyzed papers use to generate recommen-

dations (RQ

) are event logs next to partial traces.

Furthermore, the worklist of the current user, re-

source information, and the process model are con-

sidered. The methods known from recommender sys-

tems are hardly considered to generate recommenda-

tions in the context of business process management

(RQ

). Hybrid ﬁltering is used by only one paper

whereas content-based ﬁltering and collaborative ﬁl-

tering methods are used in seven, respectively, ﬁve

papers. Most papers calculate recommendations us-

ing other methods.

Overall, these studies demonstrate the impor-

tance of developing personalized recommender sys-

tems that can take into account a wide range of fac-

tors, such as user preferences, behavior patterns, con-

text information, and domain knowledge. While the

approaches and methodologies used in these studies

vary, they all share the goal of enhancing user support

and providing relevant, useful, and engaging recom-

mendations. As the use of recommender systems con-

tinues to grow in various applications, researchers and

practitioners must continue exploring new approaches

and methods for developing effective recommender

systems that can meet the evolving needs of different

users and applications.

Systematic literature reviews have several typical

pitfalls and threads to validity (Ampatzoglou et al.,

2019; Kitchenham and Charters, 2007). One po-

tential threat is missing relevant publications during

the search. This risk is mitigated by conducting a

two-phase search that includes a broad range of key

terms as well as backward referencing. Another po-

tential threat is to exclude relevant publications during

screening. This threat is mitigated by using explicitly

deﬁned inclusion and exclusion criteria. Additionally,

all unclear cases are examined and discussed by all

authors of this paper. A potential bias is subjectiv-

ity in applying the inclusion and exclusion criteria to

determine the subsumed studies and possible inaccu-

racies in data analysis. To mitigate these threats, each

article is independently assessed against the inclusion

and exclusion criteria by all authors. To provide va-

lidity, all articles are thoroughly reviewed by two au-

thors.

5 CONCLUSION

In this paper, we present a systematic literature re-

view of recommender systems deployed in the context

of BPM. The paper categorizes the research based on

the objectives, methodology, and input data. The ﬁrst

goal of the paper is to provide a comprehensive review

of the existing literature on the use of recommender

systems in the context of BPM. The paper identiﬁes

the research conducted in the past, categorizes the re-

Recommender Systems in Business Process Management: A Systematic Literature Review

439

search, and analyzes the ﬁndings. The second goal

is to identify the research gaps in the ﬁeld of recom-

mender systems in BPM.

The ﬁrst gap is the need for more detailed con-

sideration of user preferences in the process execu-

tion phase. Most of the studies focus on process im-

provement, ignoring the preferences and experiences

of the users. This gap presents an opportunity for fu-

ture research to explore the use of recommender sys-

tems to personalize the process execution for individ-

ual users. The second gap is the reliance on event logs

as input data. While event logs provide a rich source

of data for analyzing the process, they suffer from

the cold-start problem. This problem arises when a

new process or employee is introduced, and no data

is available to train the recommender system. Future

research can explore the use of other types of input

data, such as process models and user preferences, to

overcome this problem.

Finally, this paper is limited to process execution

and does not consider other research ﬁelds like us-

ing recommender systems in the context of business

process modeling. To the best of our knowledge, sev-

eral recommender systems have been proposed to as-

sist process designers in modeling business processes.

However, these approaches are not examined in this

study.

REFERENCES

W.M.P. van der Aalst, Pesic, M., and Schonenberg, H.

(2009). Declarative workﬂows: Balancing between

ﬂexibility and support. Computer Science - Research

and Development, 23:99–113.

Agarwal, P., Gupta, A., Sindhgatta, R., and Dechu, S.

(2022). Goal-oriented next best activity recom-

mendation using reinforcement learning. ArXiv,

abs/2205.03219.

Aggarwal, C. C. (2016). Recommender Systems - The Text-

book. Springer.

Ampatzoglou, A., Bibi, S., Avgeriou, P., Verbeek, M., and

Chatzigeorgiou, A. (2019). Identifying, categorizing

and mitigating threats to validity in software engineer-

ing secondary studies. Inf. Softw. Technol., 106:201–

230.

Arias, M., Munoz-Gama, J., and Sep

ulveda, M. (2017). A

multi-criteria approach for team recommendation. In

Dumas, M. and Fantinato, M., editors, Business Pro-

cess Management Workshops, pages 384–396, Cham.

Springer International Publishing.

Arias, M., Munoz-Gama, J., Sep

ulveda, M., and Miranda,

J. C. (2018). Human resource allocation or rec-

ommendation based on multi-factor criteria in on-

demand and batch scenarios. European Journal of In-

dustrial Engineering, 12:364–404.

Arias, M., Rojas, E., Lee, W. L. J., Munoz-Gama, J., and

Sep

ulveda, M. (2016). Resrec: A multi-criteria tool

for resource recommendation. In International Con-

ference on Business Process Management.

Arias, M., Rojas, E., Munoz-Gama, J., and Sep

ulveda,

M. W.M.P. van der Aalst (2016) A framework for

recommending resource allocation based on process

mining. In Reichert, M. and Reijers, H. A., edi-

tors, Business Process Management Workshops, pages

458–470, Cham. Springer International Publishing.

Barba, I., Weber, B., and Del Valle, C. (2012). Sup-

porting the optimized execution of business processes

through recommendations. In Daniel, F., Barkaoui,

K., and Dustdar, S., editors, Business Process Man-

agement Workshops, pages 135–140, Berlin, Heidel-

berg. Springer.

Barba, I., Weber, B., Del Valle, C., and Jim

enez-Ram

ırez,

A. (2013). User recommendations for the optimized

execution of business processes. Data & Knowledge

Engineering, 86:61–84.

Bidar, R., ter Hofstede, A., Sindhgatta, R., and Ouyang, C.

(2019). Preference-based resource and task allocation

in business process automation. In Panetto, H., De-

bruyne, C., Hepp, M., Lewis, D., Ardagna, C. A., and

Meersman, R., editors, On the Move to Meaningful In-

ternet Systems: OTM 2019 Conferences, pages 404–

421, Cham. Springer International Publishing.

Bozorgi, Z. D., Teinemaa, I., Dumas, M., Rosa, M. L., and

Polyvyanyy, A. (2021). Prescriptive process monitor-

ing for cost-aware cycle time reduction. In 2021 3rd

International Conference on Process Mining (ICPM),

pages 96–103. IEEE.

Branchi, S., Di Francescomarino, C., Ghidini, C., Mas-

simo, D., Ricci, F., and Ronzani, M. (2022). Learning

to act: A reinforcement learning approach to recom-

mend the best next activities. In Di Ciccio, C., Di-

jkman, R., del R

ıo Ortega, A., and Rinderle-Ma, S.,

editors, Business Process Management Forum, pages

137–154, Cham. Springer International Publishing.

Cabanillas, C., Garc

ıa, J. M., Resinas, M., Ruiz, D.,

Mendling, J., and Ruiz-Cort

es, A. (2013). Priority-

based human resource allocation in business pro-

cesses. In Basu, S., Pautasso, C., Zhang, L., and Fu,

X., editors, Service-Oriented Computing, pages 374–

388, Berlin, Heidelberg. Springer.

Conforti, R., de Leoni, M., La Rosa, M., van der Aalst,

W. M., and ter Hofstede, A. H. (2015). A recom-

mendation system for predicting risks across multiple

business process instances. Decision Support Systems,

69:1–19.

Di Valentin, C., Emrich, A., Werth, D., and Loos, P. (2014).

Context-sensitive and individualized support of em-

ployees in business processes: Conceptual design of a

semantic-based recommender system. In 2014 9th In-

ternational Workshop on Semantic and Social Media

Adaptation and Personalization, pages 77–82. IEEE.

Dumas, M., Rosa, M. L., Mendling, J., and Reijers, H. A.

(2018). Fundamentals of Business Process Manage-

ment. Springer, 2nd edition.

oger, C., Schwarz, H., and Mitschang, B. (2014). Pre-

scriptive analytics for recommendation-based busi-

ICEIS 2023 - 25th International Conference on Enterprise Information Systems

440

ness process optimization. In Abramowicz, W. and

Kokkinaki, A., editors, Business Information Systems,

pages 25–37, Cham. Springer International Publish-

ing.

Haisjackl, C. and Weber, B. (2011). User assistance during

process execution - an experimental evaluation of rec-

ommendation strategies. In zur Muehlen, M. and Su,

J., editors, Business Process Management Workshops,

pages 134–145, Berlin, Heidelberg. Springer.

Harzing, A. (2007). Publish or perish. Sofware. available

from https://harzing.com/resources/publish-or-perish.

Huber, S., Fietta, M., and Hof, S. (2015). Next step recom-

mendation and prediction based on process mining in

adaptive case management. In Proceedings of the 7th

International Conference on Subject-Oriented Busi-

ness Process Management, S-BPM ONE ’15, New

York, NY, USA. Association for Computing Machin-

ery.

Jablonski, S. and Bussler, C. (1996). Workﬂow Manage-

ment: Modeling Concepts, Architecture, and Imple-

mentation. Cengage Learning.

Khan, A., Le, H., Do, K., Tran, T., Ghose, A., Dam, H.,

and Sindhgatta, R. (2021). Deepprocess: Support-

ing business process execution using a mann-based

recommender system. In Hacid, H., Kao, O., Me-

cella, M., Moha, N., and Paik, H.-y., editors, Service-

Oriented Computing, pages 19–33, Cham. Springer

International Publishing.

Kitchenham, B. A. and Charters, S. (2007). Guidelines for

performing systematic literature reviews in software

engineering. Technical Report EBSE 2007-001, Keele

University and Durham University Joint Report.

Koulopolous, T. M. (1995). The Workﬂow Imperative:

Building Real World Business Solutions. John Wiley

& Sons, Inc.

Kubrak, K., Milani, F., Nolte, A., and Dumas, M. (2021).

Prescriptive process monitoring: Quo vadis? CoRR,

abs/2112.01769.

Lawrence, P. (1997). Workﬂow Handbook.

Leoni, M. d., Dees, M., and Reulink, L. (2020). Design and

evaluation of a process-aware recommender system

based on prescriptive analytics. In 2020 2nd Interna-

tional Conference on Process Mining (ICPM), pages

9–16. IEEE.

Liu, Q. and Wu, B. (2018). Prediction of business process

outcome based on historical log. In Proceedings of the

10th International Conference on Computer Model-

ing and Simulation, ICCMS ’18, page 118–122, New

York, NY, USA. Association for Computing Machin-

ery.

Mertens, S., Gailly, F., and Poels, G. (2015). Generating

business process recommendations with a population-

based meta-heuristic. In Fournier, F. and Mendling,

J., editors, Business Process Management Workshops,

pages 516–528, Cham. Springer International Pub-

lishing.

Microsoft Corporation (2022). Microsoft excel.

Okoli, C. and Schabram, K. (2010). A guide to conducting

a systematic literature review of information systems

research. SSRN Electronic Journal, 10.

Page, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron,

I., Hoffmann, T. C., Mulrow, C. D., Shamseer, L.,

Tetzlaff, J. M., Akl, E. A., Brennan, S. E., Chou,

R., Glanville, J., Grimshaw, J. M., Hr

objartsson, A.,

Lalu, M. M., Li, T., Loder, E. W., Mayo-Wilson, E.,

McDonald, S., McGuinness, L. A., Stewart, L. A.,

Thomas, J., Tricco, A. C., Welch, V. A., Whiting, P.,

and Moher, D. (2021). The prisma 2020 statement:

an updated guideline for reporting systematic reviews.

BMJ, 372.

Petter, S., Fichtner, M., Sch

onig, S., and Jablonski, S.

(2022). Content-based ﬁltering for worklist reorder-

ing to improve user satisfaction : A position paper.

In Proceedings of the 24th International Conference

on Enterprise Information Systems. Volume 2, pages

589–596. SciTePress, Portugal.

Pika, A. and Wynn, M. (2021). A machine learning based

approach for recommending unfamiliar process activ-

ities. IEEE Access, 9:104969–104979.

Pika, A. and Wynn, M. T. (2020). Workforce upskilling: A

history-based approach for recommending unfamiliar

process activities. In Dustdar, S., Yu, E., Salinesi, C.,

Rieu, D., and Pant, V., editors, Advanced Information

Systems Engineering, pages 334–349, Cham. Springer

International Publishing.

Reichert, M. and Weber, B. (2012). Enabling Flexibility

in Process-Aware Information Systems: Challenges,

Methods, Technologies. Springer.

Schobel, J. and Reichert, M. (2017). A Predictive Approach

Enabling Process Execution Recommendations, pages

155–170. Springer International Publishing, Cham.

Schonenberg, H., Weber, B., van Dongen, B., and van der

Aalst, W. (2008). Supporting ﬂexible processes

through recommendations based on history. In Du-

mas, M., Reichert, M., and Shan, M.-C., editors, Busi-

ness Process Management, pages 51–66, Berlin, Hei-

delberg. Springer.

Setiawan, M. A. and Sadiq, S. (2011). Experience driven

process improvement. In Halpin, T., Nurcan, S.,

Krogstie, J., Soffer, P., Proper, E., Schmidt, R., and

Bider, I., editors, Enterprise, Business-Process and

Information Systems Modeling, pages 75–87, Berlin,

Heidelberg. Springer.

Setiawan, M. A., Sadiq, S., and Kirkman, R. (2011). Facili-

tating business process improvement through person-

alized recommendation. In Abramowicz, W., editor,

Business Information Systems, pages 136–147, Berlin,

Heidelberg. Springer.

Trabelsi, F. Z., Khtira, A., and El Asri, B. (2021). Towards

an approach of recommendation in business processes

using decision trees. In 2021 International Symposium

on Computer Science and Intelligent Controls (ISC-

SIC), pages 341–347. IEEE.

van der Aalst, W. (2016). Process Mining: Data Science in

Action. Springer, 2 edition.

van der Aalst, W. M. P., Pesic, M., and Song, M. (2010). Be-

yond process mining: From the past to present and fu-

ture. In Pernici, B., editor, Advanced Information Sys-

tems Engineering, pages 38–52, Berlin, Heidelberg.

Springer.

Recommender Systems in Business Process Management: A Systematic Literature Review

441

Weinzierl, S., Dunzer, S., Zilker, S., and Matzner, M.

(2020). Prescriptive business process monitoring for

recommending next best actions. In Fahland, D., Ghi-

dini, C., Becker, J., and Dumas, M., editors, Business

Process Management Forum, pages 193–209, Cham.

Springer International Publishing.

Weske, M. (2007). Business Process Management - Con-

cepts, Languages, Architectures. Springer.

Yang, S., Dong, X., Sun, L., Zhou, Y., Farneth, R. A.,

Xiong, H., Burd, R. S., and Marsic, I. (2017). A data-

driven process recommender framework. In Proceed-

ings of the 23rd ACM SIGKDD International Con-

ference on Knowledge Discovery and Data Mining,

KDD ’17, page 2111–2120, New York, NY, USA. As-

sociation for Computing Machinery.

Yari Eili, M. and Rezaeenour, J. (2022). A survey on recom-

mendation in process mining. Concurrency and Com-

putation: Practice and Experience, 34.

Zhao, W., Liu, H., Dai, W., and Ma, J. (2016). An entropy-

based clustering ensemble method to support resource

allocation in business process management. Knowl-

edge and Information Systems, 48:305–330.

ICEIS 2023 - 25th International Conference on Enterprise Information Systems

442