Content-based Filtering for Worklist Reordering to improve User

Satisfaction: A Position Paper

Sebastian Petter

, Myriel Fichtner

, Stefan Sch

onig

and Stefan Jablonski

University of Bayreuth, Germany

University of Regensburg, Germany

Keywords:

Business Process Management, Recommender Systems, Content-based Filtering, User-centered Process

Improvement, User Satisfaction, Worklist Optimization, Process Activity Similarity.

Abstract:

Business Process Management (BPM) is an approach to optimize business processes regarding certain com-

pany goals, e.g., the duration or quality of process outcome. Human resources are essential to business pro-

cesses and are often neglected during process optimization. Considering domains focusing on users, it can be

observed that recommender systems are often used to support user decisions and increase user satisfaction.

This inspired us to use recommendation techniques in the context of BPM. Employee satisfaction signiﬁcantly

inﬂuences productivity, while employees are more satisﬁed when their preferences are taken into account dur-

ing process execution. In this work, we propose to adopt the concept of content-based ﬁltering to recommend

worklist items to process participants they probably prefer. Since this work is part of a research project, we

illustrate our approach on a simpliﬁed real-world business process from one of our application partners.

1 INTRODUCTION

Business processes can be understood as a series of

activities to be completed to achieve a company goal.

Optimizing such business processes is one of the pri-

mary purposes in research related to Business Process

Management (BPM). Usual goals are to shorten pro-

cessing time or to increase the quality of process out-

comes (Jablonski and Bussler, 1996), (Koulopolous,

1995), (Lawrence, 1997). To reach optimization,

companies need to gain insight into the structure of

their processes. An established method is to visual-

ize them using process modeling languages such as

the Business Process Model and Notation (BPMN)

(OMG, 2011). The resulting business process model

contains the sequence of required steps and process

entities involved, like process participants or data ob-

The project is ﬁnanced with funding provided

by the Federal Ministry of Education and Research

and the European Social Fund under the ”Future

of work” programme and managed by the Project

Management Agency Karlsruhe (PTKA). The author

is responsible for the content of this publication.

jects. Process-Aware Information Systems (PAIS) are

used to manage and execute operational processes

based on process models. Every executed activity

is recorded in a so-called process event log during

process execution. In addition, further details are in-

cluded, like execution durations or resources execut-

ing activities (van der Aalst, 2016). Process mining

techniques allow process optimization through the ex-

traction of knowledge from event logs (Marin-Castro

and Tello-Leal, 2021). Most approaches are activity-

centered and focus on process control ﬂow (Scho-

nenberg et al., 2008), (Haisjackl and Weber, 2010).

In contrast, many processes executed in small and

medium enterprises are human-driven. Since human

resources are an essential component of business pro-

cess executions in companies, they have to be consid-

ered as a target for optimization.

In times of a shortage of skilled employees, com-

panies have to distinctly focus on the needs and ex-

pectations of human workers. Consideration of such

aspects increases employee satisfaction and decreases

ﬂuctuation. Furthermore, happiness and satisfaction

with assigned work positively impact the employees’

productivity. It is proved that ”happy” workers are

13 % more productive (Bryson et al., 2015) (Bellet

et al., 2019). Furthermore, human resources that have

some degree of control over their work are more pro-

Petter, S., Fichtner, M., Schönig, S. and Jablonski, S.

Content-based Filtering for Worklist Reordering to improve User Satisfaction: A Position Paper.

DOI: 10.5220/0011092900003179

In Proceedings of the 24th International Conference on Enterprise Information Systems (ICEIS 2022) - Volume 2, pages 589-596

ISBN: 978-989-758-569-2; ISSN: 2184-4992

589

ductive and more motivated (Russell et al., 2016).

One way of increasing satisfaction and motivation

of employees is to regard their preferences. Discover-

ing preferences of human process participants can be

adopted from the domain of recommender systems.

Recommender systems are used to recommend items

to users. For example, they are used in online shops to

recommend products the user potentially prefers (Ag-

garwal, 2016). The use of such concepts has proven

to be proﬁtable for companies. These results inspired

us to adopt the methods of recommender systems and

using them in the context of BPM for user-centered

process improvement. To the best of our knowledge

existing research applies worklist reordering and op-

timization only for the purpose of improving key per-

formance indicators. User and employee satisfaction,

however, is neglected so far. In this work, we con-

sider the concept of content-based ﬁltering to recom-

mend activities of a running business process instance

to process participants. For this purpose, we intro-

duce two ideas to enhance process activities and pro-

cess event logs to adopt the concept of content-based

ﬁltering to BPM.

As a simple example, we consider a software

development process. Besides other activities, it

comprises the activities ”frontend development” and

”backend development”. Both must be executed by

software developers. Some of them prefer to use

JAVA to HTML. Due to missing knowledge about that

preference, assigning activities to developers occurs

randomly. The developers have to execute these ac-

tivities independent of their preferences and the pref-

erences of co-workers. This might lead to minor mo-

tivation to do the jobs.

Human resources have different preferences

which may affect their motivation (Novak, 1999).

By applying content-based ﬁltering, we create user-

centered process improvement by reordering the

worklist items based on preferences of human re-

sources. This creates a better working atmosphere,

strengthens employee loyalty, and raises productivity.

Furthermore, reordering the worklist has no negative

impact on process performance (Pﬂug and Rinderle-

Ma, 2015).

The main research question addresses the issue

whether the approach of content-based ﬁltering from

recommender systems can be applied in BPM to in-

crease user satisfaction. From this, two sub-tasks can

be derived:

1. Which information can be exploited by a recom-

mender system to infer user preferences?

2. How should these results be displayed to the

users?

To realize user-centered process improvement com-

bining BPM and recommender systems, an inter-

disciplinary consortium consisting of three research

partners and ﬁve SMEs - three application partners

and two implementation partners - collaborated in a

joint project called PRIME

(Process-based integra-

tion of human expectations in digitalized work envi-

ronments) funded by the Federal Ministry of Educa-

tion and Research and the European Social Fund. In

our work, we present an exemplary real-world pro-

cess of one of our partners to show the relevance of

this topic.

The remainder of the paper is structured as fol-

lows. In Section 2, we present the two main areas con-

sidered in this work: BPM and recommender systems

in general and the content-based ﬁltering method in

particular. Furthermore, we show existing approaches

combining these aspects. Section 3 describes the core

idea of recommending activities to the current user of

a PAIS. We conclude our work and give ideas for fu-

ture work in Section 4.

2 BACKGROUND

This work considers the two ﬁelds of BPM and rec-

ommender systems to accomplish user-centered pro-

cess improvement by means of worklist optimization.

First, we introduce the two research areas, BPM and

recommender systems. We then present related work

that addresses both topics.

Business Process Management. BPM is used to

increase efﬁciency and reduce costs of processes exe-

cuted in enterprises of all industries and sizes. BPM

considers modeling, executing, and analyzing busi-

ness processes (Dumas et al., 2018). In order to

execute a business process, ﬁrst, a business process

model is created. A Business Process Management

System executes the model and displays activities that

have to be executed. Activities that need to be ex-

ecuted are represented in a worklist. Since multi-

ple users with the same roles are usually involved

in the process execution, the worklist can be same

for them. Users can select their preferred activities

from the worklist as long as there are no further re-

strictions. Historical process executions of a business

process are stored in so-called (process) event logs.

Therefore each execution of an activity is recorded as

event. Such events encapsulate all relevant informa-

tion about the execution, e.g., execution timestamp or

the involved resources. Other properties can arbitrar-

ily extend this set of information. Events that refer to

https://prime-interaktionsarbeit.de/

ICEIS 2022 - 24th International Conference on Enterprise Information Systems

590

the same process instance are temporally ordered by

their timestamp and build so-called ”traces” (van der

Aalst, 2016).

Recommender Systems. Recommender systems

predict results for alternative options based on user

preferences. (Guo et al., 2019). Preference is one

of the fundamental attributes to support user (Gold-

smith and Junker, 2008). The value of recommender

systems has been demonstrated in various application

domains like E-commerce and other user-centric web

applications (Manouselis and Costopoulou, 2007),

(Nguyen and Haddawy, 1998). To calculate pre-

dictions, recommendation algorithms require infor-

mation about users and items that should be rec-

ommended. Four main recommendation methods

have been established, namely collaborative ﬁlter-

ing, content-based ﬁltering, knowledge-based ﬁlter-

ing, and hybrid recommendation, which differ mainly

in the input information used to calculate recommen-

dations. In our work, we focus on content-based

ﬁltering. Content-based ﬁltering aims to classify

items with speciﬁc keywords, learn about user prefer-

ences, look up those keywords, and recommend simi-

lar items. Content-based ﬁltering exploites two types

of input data: (i) user feedback (explicit and implicit)

and (ii) item attributes. User feedback is created while

using, for example, web applications. Consider a

webshop, where users can buy products and rate them

afterwards. Buying a product is considered implicit

user feedback (liking this product), whereas the rating

is explicit feedback. Any product of a webshop has

additional information about itself, so-called ”item at-

tributes”, e.g., a book has an author and a genre. User

preferences can be derived from feedback created by

users. The item attributes are used to compare activ-

ities. Combining item attributes and user feedback,

recommendation algorithms can calculate the proba-

bility of a user liking a speciﬁc product depending on

the similarity to other, already rated products. The

goal of content-based ﬁltering methods is to predict

the rating of each item considered using training data

where items are rated by the user. Content-based

ﬁltering follows three main steps: (i) Preprocessing

and Feature Extraction, (ii) Learning User Proﬁles,

and (iii) Filtering and Recommendation (Aggarwal,

2016). During step (i) features are extracted from the

considered item and are converted into a keyword-

based vector space. In (ii), a user-speciﬁc model is

created to predict user interest in items. The con-

struction of this model is based on the history of a

user buying or rating items. (iii) establishes recom-

mendations on items for the user to consider while

creating the model. A drawback of content-based

ﬁltering is the cold-start problem.It describes that it

is impossible to create recommendations if we have

no information about the user (e.g., in case of a new

user). Furthermore, only apparent items are often rec-

ommended because the algorithm recommends items

with similar attributes. In return, new items can be

recommended because they are comparable to exist-

ing items due to the item attributes, and the recom-

mendation is not based on preferences of other users.

Furthermore, item attributes can be derived from the

textual description of an item.

Related Work. Now, we provide an overview of

work that combines the two areas Business Process

Management and recommender systems. Most ap-

proaches consider process improvement through rec-

ommendations, whereas the focus is on optimizing

process results or shortening processing time. In

(Schonenberg et al., 2008) and (Haisjackl and We-

ber, 2010) the authors present an approach creating

recommendations on possible next steps through pre-

diction of a partial case. The recommendations are

based on the current execution trace and considera-

tion of similar process instances. The goal is to as-

sist the user in optimal decision-making. A similar

approach is presented in (van der Aalst et al., 2010).

Based on event logs, recommendations are made on

which activity should be executed next to receive a

certain goal like shortened processing time. To sup-

port users in ﬂexible processes, business process mod-

els are enhanced by estimations about runtime and

availability of resources in (Barba et al., 2012). Based

on this information, an optimized execution plan can

be generated. To reduce risks (e.g., exceeding dead-

lines during process execution), recommendations, on

which activity has to be executed next, are made to

the user in (Conforti et al., 2015). Such recommenda-

tions are also made in (Huber et al., 2015). Consid-

ering the successors, activities are recommended to

the user having the shortest runtime. In (Bidar et al.,

2019) the authors make a suggestion on how to de-

rive user-preferences from event logs. The optimal

allocation of resources at process level is computed

in (Cabanillas et al., 2013). In this work, a priority-

based resource allocation is proposed using the Se-

mantic Ontology of User Preferences (Garc

ıa et al.,

2010). Every time a new activity has to be executed,

it is allocated to the resource with the highest pref-

erence. Reordering worklist items to improve key

performance indicators is considered in (Pichler and

Edre, 2019). Furthermore, (Pﬂug and Rinderle-Ma,

2015) shows, that reordering the worklist of process

participants does not have negative impact on tempo-

ral parameters like throughput time. Summing up, ex-

Content-based Filtering for Worklist Reordering to improve User Satisfaction: A Position Paper

591

isting research mainly focuses on recommending next

activities and resource allocation for performance op-

timization.

3 CONCEPTUAL APPROACH

To support the comprehensibility of our approach, we

introduce a running example of a business process

modeled in the modeling language Business Process

Model and Notation (BPMN). The simpliﬁed busi-

ness process shown in Figure 1 is part of a real-world

example from one of our application partners. It con-

sists of ﬁve activities that must be executed by human

resources with the role Team leader. The ﬁrst step is

to accept an incoming order the Team leader receives

as an E-mail from a customer. Afterward, the Roster

for the technician must be created in Excel and dis-

patched to the technician by E-mail. During the step

”Preparation of order (ID4)”, the Team leader com-

municates with the customer (getting detailed order

information) and with the technician (send detailed

order information as Excel-ﬁle by E-mail). The last

step of this process is the treatment of special requests

the technicians send via E-mail.

As described in Section 2.2, user feedback and

item attributes are used as input information for

content-based ﬁltering. In BPM, we can consider ac-

tivities as items.

Thus, we need to get user feedback related to ac-

tivities and attributes describing activities. Implicit

user feedback is created during process execution and

is stored in an event log. This information is declared

as implicit feedback since it relates to execution pa-

rameters (e.g., execution time) and not to information

directly provided by executors. However, to identify

user preferences, explicit feedback in term of user rat-

ing is still missing. Therefore, we need to introduce

an additional step after executing user tasks: the rat-

ing of an executed activity. In our example, we simply

consider like and dislike as activity ratings. In general,

the ratings can be of any complexity.

Another preprocessing step for applying content-

based methods in BPM is tagging activities. In pro-

cess models, activities have labels interpretable as

textual instructions but do not contain information

about semantics. To get more information about an

activity and to be able to compare activities, we need

to add tags or labels to each activity. Our concept con-

sists of three basic elements: (i) Preprocessing and

Tagging, (ii) Content-based Filtering, and (iii) Work-

list Reordering.

Preprocessing and Tagging. The input data for

preprocessing are business process models, process

descriptions, an event log, and the current state of pro-

cess execution provided by a PAIS instance, e.g., the

current users and their worklists. As process model

the introduced example from above is taken. During

execution of this process model an event log is gen-

erated. Every event log entry must include a rating

of whether the executing process participant likes or

dislikes the activity.

To create recommendations based on content-

based ﬁltering, we also need activity tags containing

semantic information. These tags can be generated

from the process model automatically or manually

by a process expert during modeling. Furthermore,

a process expert can manually adapt these keywords

(adding new or removing irrelevant ones). An advan-

tage of this approach is the possibility of consider-

ing and therefore interpretively connecting activities

of multiple processes. Thus, executions of different

processes by the same user can be combined through

these activity tags. This enables to adopt preferences

for similar activities from other processes.

In our running example, we derive ﬁve tags from

the business process model: E-mail, Excel, commu-

nication with customer, communication with techni-

cian, and roster. None of the ﬁve activities has all

tags. For example, ”Accept order (ID1)” considers

the activity tags communication with customer and e-

mail. The ”Preparation of order (ID4)” activity con-

siders all tags but roster.

Content-based Filtering. To generate recommen-

dations, we consider training data which is deﬁned as

a set of rated activities A

. These activities are rated

by the process participant the items are recommended

to. Since the event log is enhanced by explicit user

feedback, it contains all information to calculate rec-

ommendations with content-based ﬁltering methods.

Therefore we consider it as training data. Every item

of the event log contains an activity ID, the informa-

tion which user executed what activity, and a set of

activity tags.

Besides the training set A

for content-based ﬁl-

tering, an unrated set of activities A

(worklist) is

considered. As described in Section 2, the goal of

content-based ﬁltering methods is to predict the rating

of each activity in A

using the training data. Both,

and A

, are user-speciﬁc sets of activities.

The issue we have to solve is similar to that of

classiﬁcation. An established technique to classify

data sets is the nearest neighbor classiﬁer (Chomboon

et al., 2015). We will use this classiﬁcation method in

our running example, while it is interchangeable and

ICEIS 2022 - 24th International Conference on Enterprise Information Systems

592

Figure 1: Example: Technician service management process.

can be replaced by any other classiﬁcation method.

We deﬁne a similarity function based on the near-

est neighbor classiﬁer. Similarity/distance functions

like the Euclidean distance or the Manhattan distance

are used for structured or multidimensional data. In

this work, we use the Euclidean distance to calculate

the distance between two vectors to compare two ac-

tivities. Afterward, we transform the results to a sim-

ilarity measure due to better comparability.

The Euclidean distance is calculated as follows:

d(p, q) =

∑

i=1

− p

)

(1)

Where p and q are a pair of vectors representing

activities with n attributes. Furthermore, a similarity

metric is needed to improve comparability of the dis-

tances:

sim(p, q) =

1 + d(p, q)

(2)

For each activity in A

, the similarity metric to

each activity from A

is calculated to identify the

nearest neighbor. Considering the rating of the near-

est neighbor activity, the rating of the related activity

in A

can be deduced.

In our running example, the training set A

based on the event log and the activity tags created

in the Preprocessing and Tagging step. In Table 1 the

training set A

is displayed. The activities with ID1

– ID4 have been executed several times and rated by

the current user. Activity ID5 is not displayed at the

table because the team leader has never executed it

until now. Furthermore, the tags for every activity are

considered. In our example the values 0 and 1 show

whether an attribute is present or not. The vector-

representation of activity ID4 is (1, 1, 1, 1, 0) since it is

related to the tags E-mail, Excel, communication with

customer and communication with technician. The set

of unrated activities A

is the worklist with tasks that

team leaders must execute. We assume that the work-

list of the team leaders contains currently ﬁve tasks in

the following order where every task can be executed

independent of others:

1. Create roster (ID2)

2. Dispatch roster (ID3)

3. Receive special request (ID5)

4. Accept order (ID1)

5. Preparation of order (ID4)

List 1: Worklist of team leader (FIFO).

To calculate the similarity between ”Receive spe-

cial request (ID5)” and ”Preparation of order (ID4)”

we have to deﬁne p as the activity ”Receive special

request (ID5)” in A

and q as the activity ”Prepa-

ration of order (ID4)” in A

. The two vectors v

(1, 1, 0, 1, 0) and v

= (1, 1, 1, 1, 0) are representations

of the activities. Then d(p, q) = 1 and sim(p, q) = 0.5.

Calculating all values for p of any activity in A

, we

notice that 0.5 is the highest value, whereas 1 means

the activities are similar and 0 means that the activ-

ities are completely different. Considering the user

rating of all ”Preparation of order (ID4)” activities,

we notice that the team leader always likes this ac-

tivity. Thus, the team leader possibly likes ”Receive

special request (ID5)” too. Applying this calculation

to all data considered, we achieve the following pre-

dictions: the ﬁrst item of the team leader’s worklist

(List 1),”Create roster (ID2)”, is rated as dislike. The

activity has been executed by the team leader four

times. Thus, the similarity metric sim(p, q) equals 1.

Content-based Filtering for Worklist Reordering to improve User Satisfaction: A Position Paper

593

Table 1: Training data A

of one team leader for running example.

ActivityId E-mail Excel Com. Customer Com. Technician Roster ... Like/Disklike

ID1 1 0 1 0 0 ... DISLIKE

ID2 0 1 0 0 1 ... DISLIKE

ID3 1 0 0 1 1 ... DISLIKE

ID4 1 1 1 1 0 ... LIKE

ID1 1 0 1 0 0 ... DISLIKE

ID2 0 1 0 0 1 ... DISLIKE

ID3 1 0 0 1 1 ... LIKE

ID2 0 1 0 0 1 ... LIKE

ID2 0 1 0 0 1 ... DISLIKE

ID3 1 0 0 1 1 ... LIKE

ID4 1 1 1 1 0 ... LIKE

Three of four times, the user did not like the activ-

ity though the prediction for a feature rating is dis-

like. In the same way, the rating for ”Create roster

(ID2)” is calculated, the rating for ”Dispatch roster

(ID3)”, ”Accept order (ID1)”, and ”Preparation of or-

der (ID4)” can be calculated because the user has al-

ready executed these activities. The only task the user

did not execute yet is ”Receive special request (ID5)”.

To predict this task, we need to calculate the nearest

neighbor by calculating the Euclidean distance and

this similarity metric shown in Formula (1) and (2).

The calculation results in a similarity metric of 0.5

regarding the activity ”Preparation of order (ID4)”.

This activity has been executed twice, and the user

liked it both times, thus with a high probability the

user likes the new activity.

Worklist Reordering. Most PAIS organize a work-

list according to the FIFO principle. Taking the re-

sults from step (ii) (Content-based Filtering), we can

rearrange the worklist and adapt it to the requirements

of a current user. The manipulation of the items only

affects the order of the worklist items. We do not omit

any tasks as well as we do not add any new tasks. The

active user can still decide which activity to execute

next.

While displaying the reordered, customized work-

list, it is required to present the reason for reordering

so the user can reproduce why speciﬁc activities are

recommended.

Taking the results of step (ii) (Content-based

Filtering) and adapting the worklist according to

these recommendations, the following reordered

worklist can be generated:

1. Dispatch roster (ID3)

2. Receive special request (ID5)

3. Preparation of order (ID4)

4. Create roster (ID2)

5. Accept order (ID1)

List 2: Preference-based reordered worklist of team leader.

In this case, the content-based ﬁltering method

recommends executing the tasks ”Dispatch roster

(ID3)”, ”Receive special request (ID5)”, and ”Prepa-

ration of order (ID4)” because these are the tasks the

team leader potentially likes, whereas the last two

tasks in List 2 are not among the user’s favorite tasks.

Perhaps another team leader with opposite prefer-

ences performs the latter tasks, thus increasing that

user’s satisfaction.

The example considers a worklist in ﬁrst-in-ﬁrst-

out order. However, some approaches exist to reorder-

ing the worklist due to better process performance or

similar (e.g., (Schonenberg et al., 2008)). These pre-

ordered worklists must be considered too. One pos-

sibility of including user preferences into preordered

worklists is to reorder only tasks with the same prior-

ity, not falsifying previous calculations.

Considering the real-world example, we notice

that the activities ”Receive special request (ID5)” and

”Preparation of order (ID4)” are similar, indeed. One

team leader available during process documentation

rated the activities ID1 - ID4 (Table 1). Activity ID5

has not been rated since we want to predict whether

the team leader likes it or not. Presenting the result

of the calculation, the team leader conﬁrmed that he

likes the activity ”Receive special request (ID5)”.

4 CONCLUSION AND FUTURE

WORK

In this work, we provide a novel approach to increase

user satisfaction during business process execution by

worklist optimization through recommender systems

and illustrate it with a running example. We apply

ICEIS 2022 - 24th International Conference on Enterprise Information Systems

594

the method of content-based ﬁltering to a given work-

list and the event log and present the result, a work-

list, ordered according to potential user preferences,

to the active user of the PAIS. An advantage of this

approach is that existing data and information like the

process model or the event log can be used to cal-

culate the user’s preferences. Furthermore, activities

of the process model can be enhanced by other tags

for more accurate predictions. We evaluated our work

prototypically with our project partners. The idea

received a lot of positive feedback, and the project

partners conﬁrmed that the presented idea leads to an

improvement in user satisfaction. Initial prototypical

implementations and evaluations showed the success

of the approach. Manual calculations of recommen-

dations provided promising results demonstrating the

need for implementation.

In future work, we will extend our approach by

alternate activity rating approaches as well as more

approaches to generating activity tags. Furthermore,

other methods known from recommender systems can

be considered to tackle drawbacks, like cold-start

problem for new users, or recommendation of appar-

ent items, of content-based ﬁltering method, e.g., col-

laborative ﬁltering or knowledge-based ﬁltering. Fi-

nally, we plan to implement a prototype and to include

a real-world evaluation of our approach.

REFERENCES

van der Aalst, W. M. P. (2016). Process Mining: Data Sci-

ence in Action. Springer, 2 edition.

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

Beyond process mining: From the past to present and

future. In Proc. of the 22nd International Confer-

ence on Advanced Information Systems Engineering

(CAiSE ’10), volume 6051.

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

book. Springer.

Barba, I., Weber, B., and Valle, C. D. (2012). Supporting

the optimized execution of business processes through

recommendations. In Proc. BPI, volume 99, pages

135–140.

Bellet, C. S., Neve, J.-E. D., and Ward, G. (2019). Does

employee happiness have an impact on productivity?

ıd Business School WP.

Bidar, R., ter Hofstede, A. H. M., Sindhgatta, R., and

Ouyang, C. (2019). Preference-based resource and

task allocation in business process automation. In

OTM 2019, volume 11877, pages 404–421.

Bryson, A., Forth, J., and Stokes, L. (2015). Does worker

wellbeing affect workplace performance? Technical

report, Department for Business, Innovation & Skills.

Cabanillas, C., Garc

ıa, J. M., Resinas, M., Ruiz, D.,

Mendling, J., and Cort

es, A. R. (2013). Priority-based

human resource allocation in business processes. In

ICSOC 2013, volume 8274, pages 374–388.

Chomboon, K., Chujai, P., Teerarassammee, P., Kerdpra-

sop, K., and Kerdprasop, N. (2015). An empirical

study of distance metrics for k-nearest neighbor algo-

rithm. In Proc. of the 3rd International Conference on

Industrial Application Engineering, pages 280–285.

Conforti, R., de Leoni, M., Rosa, M. L., van der Aalst, W.

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

mendation system for predicting risks across multi-

ple business process instances. Decis. Support Syst.,

69:1–19.

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

(2018). Fundamentals of Business Process Manage-

ment. Springer, 2nd edition.

Garc

ıa, J. M., Ruiz, D., and Ruiz-Cort

es, A. (2010). A

model of user preferences for semantic services dis-

covery and ranking. In The Semantic Web: Research

and Applications, volume 6089, pages 1–14.

Goldsmith, J. and Junker, U. (2008). Preference handling

for artiﬁcial intelligence. AI Magazine, 29:9–12.

Guo, Z., Zeng, W., Wang, H., and Shen, Y. (2019). An

enhanced group recommender system by exploiting

preference relation. IEEE Access, PP:1–1.

Haisjackl, C. and Weber, B. (2010). User assistance during

process execution - an experimental evaluation of rec-

ommendation strategies. In Proc. of BPI Workshop,

volume 66, pages 134–145.

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

mendation and prediction based on process mining in

adaptive case management. In 7th International Con-

ference on Subject-Oriented Business Process Man-

agement.

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

ment: Modeling Concepts, Architecture, and Imple-

mentation.

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

Building Real World Business Solutions.

Lawrence, P. (1997). Workﬂow Handbook.

Manouselis, N. and Costopoulou, C. (2007). Analysis and

classiﬁcation of multi-criteria recommender systems.

World Wide Web, 10(4):415–441.

Marin-Castro, H. M. and Tello-Leal, E. (2021). Event log

preprocessing for process mining: A review. Applied

Sciences, 11.

Nguyen, H. and Haddawy, P. (1998). Diva: Applying de-

cision theory to collaborative ﬁltering. In Proc. of the

AAAI Workshop on Recommender Systems.

Novak, J. M. (1999). Choice theory: A new psychology

of personal freedom. Brock Education: a Journal of

Educational Research and Practice, 9.

OMG (2011). Business Process Model and Notation

(BPMN), Version 2.0.

Pﬂug, J. and Rinderle-Ma, S. (2015). Analyzing the effects

of reordering work list items for selected control ﬂow

patterns. In IEEE 19th International Enterprise Dis-

tributed Object Computing Workshop, pages 14–23.

Pichler, H. and Edre, J. (2019). Towards look-ahead strate-

gies for work item selection. 2019 IEEE Jordan Inter-

national Joint Conference on Electrical Engineering

and Information Technology (JEEIT), pages 752–757.

Content-based Filtering for Worklist Reordering to improve User Satisfaction: A Position Paper

595

Russell, N., van der Aalst, W. M. P., and ter Hofstede,

A. H. M. (2016). Workﬂow Patterns: The Deﬁnitive

Guide. MIT Press.

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

Aalst, W. M. P. (2008). Supporting ﬂexible processes

through recommendations based on history. In In-

ternational Conference on Business Process Manage-

ment (BPM ’08), volume 5240, pages 51–66.

ICEIS 2022 - 24th International Conference on Enterprise Information Systems

596