Temporal, Semantic and Structural Aspects-based Transformation

Rules for Refactoring BPMN Model

Mariem Kchaou, Wiem Khlif and Faiez Gargouri

Mir@cl Laboratory, University of Sfax, Sfax, Tunisia

Keywords: Refactoring BPMN Models, Structural Measures, Performances Measures, Semantic and Structural Aspects,

Temporal Aspect, Transformation Rules.

Abstract: Refactoring a business process model (BPM) may improve its usability (understandability, modifiability)

performance, and/or ease its maintainability. So-far proposed refactoring approaches have used

either refactoring which focuses on structural aspects and/or semantic information. Nevertheless, these

aspects provide a partial view of the model. As we show in this paper, combining the semantic and structural

aspects with the temporal aspect decreases further the complexity of a business process modelled in BPMN

and enhances its performance. Our method uses a set of transformation rules. We illustrate their efficiency

through well-established performance measures (temporal and cost) and structural measures (complexity).

1 INTRODUCTION

A business process (BP) is a series of activities

occurring within a company that lead to the

production of a product or a service (ISO/IEC 19510,

2013). It allows organizations to keep or even

increase their competitiveness. For this reason,

companies manage business processes with the

adequate quality degree, i.e., without faults that

influence understandabiliy, modifiability or

performance of the BP among others features. Indeed,

understandability, modifiability and performance

have proved to be three of the most important features

to accomplish business processes with appropriate

quality degrees (Lanz et al., 2016).

For instance, to show the improvements of the

undestandabilty and modifiability, several researches

apply a set of well-known structural measures

expressing the complexity of the BP (Cardoso, 2006)

(Gruhn and Laue, 2006) such as the number of

sequence flows, tasks and connectors, connectivity,

density, average/maximum connector degree, control

flow complexity, etc. In addition, for organizations

seeking continuous improvements, the recent

literature on the BP performance measures (Lanz et

al., 2016) (Kis et al., 2017) has shown three trends of

approaches: those based on time, those centered on

cost and those combining the two aspects. For

instance, (Lanz et al., 2016) define the measure

“Activity Duration” to determine the necessary time

to perform an activity.

Furthermore, refactoring techniques are required

for a potential and promising solution in order to

improve understandability, modifiability and

performance of BP models. In this context, current

approaches focusing on refactoring are classified into

structure-based works, semantic-based works, and

works which combine the semantic and structural

aspects (Khlif et al., 2017).

The first type of refactoring approaches is based

on refactoring operations and social network

rediscovery-based methods (Oinas-Kukkonen et al.,

2010). The former defines a set of refactoring

operators and algorithms to change the internal

structure of BPM without altering its external

behavior (La Rosa et al., 2011). The latter considers

that social relationships (ie. structural relations) and

collaborative behaviors among people within an

enterprise affect the overall process performance

(Boulmakoul, and Besri, 2013).

The second type of these approaches relies on so-

cial network discovery-based methods which explore

the human perspectives (performers and their roles)

to discover knowledge expressing the relationships

among performers in a BPM (Kim et al., 2014).

The third type of approaches (Khlif et al., 2017)

shows how to apply the correlated structural (i.e.,

functional and organizational) and semantic aspects

to restructure BPM specified in the Business Process

Kchaou, M., Khlif, W. and Gargouri, F.

Temporal, Semantic and Structural Aspects-based Transformation Rules for Refactoring BPMN Model.

DOI: 10.5220/0007925401270138

In Proceedings of the 16th International Joint Conference on e-Business and Telecommunications (ICETE 2019), pages 127-138

ISBN: 978-989-758-378-0

127

Modelling Notation (BPMN) (ISO/IEC19510, 2013).

Our literature review reveals that the so-far

proposed refactoring techniques have been based on

the structural and or semantic aspects while

neglecting the temporal aspect that can also affect the

refactoring quality. More specifically, in this paper,

we propose to reuse the proposed strategy in (Khlif et

al., 2017) to define transformation rules related to all

perspectives. The authors browse the fragments based

on the defined top down approach (Khlif et al., 2017).

These transformation rules use the BPMN model

structure and the business context that describe

semantic information related to BPMN elements such

as the description of the activities, the information

about the performers associated to each activity, etc.

Toward this end, we tackle the limits presented in

the literature by enhancing the EVARES approach

(Khlif et al., 2017) with : 1) a way to annotate a

BPMN model by using the business context as a

means to encapsulate temporal information related to

the actor and to all BPMN elements. These

information are pertinent to the business logic and

organizational aspect; 2) a set of new temporal and

cost measures to improve the performance of a BP;

and 3) BPMN transformation rules that encapsulates

also the temporal constraints. To show the

improvements gained by applying the transformation

rules and their influence on the complexity and

performance dimensions of a model, we use a set of

well-known structural and performance measures.

The remainder of the paper is organized as follows:

Section 2 summarizes related work. Section 3,

introduces concepts related to the quality of a BPM. In

section 4, we propose temporal and cost measures to

evaluate the performance of a BPM. Section 5 presents

the definition of the business context in BPMN model.

In section 6, we illustrate a subset of our transformation

rules. Section 7 illustrate our transformation rules

through an example. Finally, section 8 summarizes the

presented work and outlines its extensions.

2 RELATED WORK

The works in this section are overviewed not only on

structural and performance measures but also on

refactoring methods which aim at optimizing the

model.

2.1 Structural and Performance

Measures

In the area of business process, a wide variety of

business quality measures has been developed. These

measures are classified into two categories: structural

measures and performance measures.

2.1.1 Structural Measures

Structural measures fall basically into three types:

complexity, coupling and cohesion (Cardoso, 2006).

In this paper, we focus on the complexity measures

such as the size measures (Rolón et al., 2006) that

calculates the number of each BPMN element (ie.

Number of Sequence flows (NSF), Number of Lanes

(NL), Total Number of Gateways (TNG), Total

Number of Events (TNE), Number of Activities

(NOA), Number Of Activities, Joins and Splits in the

process (NOAJS)). In addition, (Mendling, 2006)

identifies the Density (Den) measure.

Others complexity measures are defined in

(Cardoso, 2006) such as the Control Flow

Complexity (CFC); and Coefficient of Connectivity

(CNC). (Gruhn and Laue, 2006) present the Cognitive

Weight (CW) measure which reveals the effort

required for comprehending a given model.

2.1.2 Performance Measures

In this section, we overview works on the BP

performance based on measures. These works are

organized in three categories: time based-

performance measures, cost based-performance

measures and integrating time and cost to measure the

BP performance.

In the first category, (Lanz et al., 2016) identify

10 different time patterns which constitute solutions

for representing temporal constraints. They are

classified into 4 distinct groups: 1) Durations and

Time Lags. 2) Restricting Execution Times. 3)

Variability. 4) Recurrent Process Elements. Based on

time patterns, (Lanz et al., 2016) define for instance

the Activity Duration (AD) belongs to TP2.

(Del-Río-Ortega et al., 2013) propose the

PPINOT metamodel that has been created to allow

the modelling of Process Performance Indicators

(PPIs). PPINOT shows the main elements of a PPI

definition and the types of measures (Base, Derived

and Aggregated) that can be used to define a PPI.

PPINOT defines also two types of resource-aware

PPIs: Resource Measure and Group by Resources.

For instance, the authors present the measure Total

Number of Actors that perform Tasks in a period of

time (TNAT(period)).

In the second category, (Sampathkumaran and

Wirsing, 2013) propose a methodology for cost

calculation by dividing a business process into

patterns. For example, the cost of n tasks in a

ICE-B 2019 - 16th International Conference on e-Business

128

sequential order represents the total cost of all the

tasks in a sequential order.

In the third category, Kis et al., 2017) propose a

framework on how the four dimensions of the devil's

quadrangle (time, cost, quality and flexibility) can be

measured. In this context, (Kis et al., 2017) define the

measure Lead Time of the Activity which is

calculated as the sum of the duration of each activity

and the wait time.

In summary, the majority of the presented

measures in literature are related to the activity and

event elements, or resources measures. However, the

authors neglect the gateway, sequence flow and

lane/pool elements which have an impact on

evaluating BP performance. In addition, the authors

don’t use these measures to refactor the BPM and

improve their performance. Besides, there is no works

that combine the cost and temporal measures of all

BPMN elements with the actor’s measures.

2.2 Works on Business Process

Refactoring

The recent literature on business process refactoring

has shown three trends of approaches:

 those which are structure-based: these works

focus on refactoring operations and social

network rediscovery-based methods;

 those which are semantic-based: they represent

social network discovery-based methods;

 those which combine the semantic and

structural aspects (Khlif et al., 2017).

2.2.1 Structure-based Refactoring

Refactoring has been used in literature for improving

the quality of a BPM. For example, (Fernández-

Ropero et al., 2013) provide a mechanism to detect

the sub-set of refactoring operators which structurally

transform a model while preserving its behavior if the

pre-conditions are satisfying. (La Rosa et al., 2011)

identify twelve patterns which determine structural

fragments subject to reorganization. (Corradini et al.,

2018) optimize the understandability of the BPM

during refactoring.

In addition, several works are based on social

network, especially, on rediscovery-based methods

(Boulmakoul and Besri, 2013) (Kajan et al., 2014) in

order to recognize processes from execution event

logs and identify the structural relations among the

performers or organizational units. Adopting this type

of approach, (Kajan et al., 2014) propose a method

using a set of social relations that connect tasks,

persons and machines together to develop specialized

networks that capture the interactions during BP

execution. (Boulmakoul and Besri, 2013) propose

methods for assessment of organizational structure

based on structural analysis.

2.2.2 Semantic-based Refactoring

Besides the structure based, other approaches focus

on discovering social network knowledge. More

specifically, (Battsetseg et al., 2013) propose an

algorithm to discover an activity-performer affiliation

network model from an Information Control Net

(ICN) based workflow model. (Kim et al., 2014)

visualize the workflow performer-role affiliation

networking knowledge from an ICN based workflow

model.

2.2.3 Refactoring based on Combining

Semantic and Structural Aspects

In (Khlif et al., 2017), the authors propose a method

called EVARES Quality (EVAluation and Quality

Restructuring) to refactor and evaluate the quality of

BP models. The authors defined twenty-eight

transformation rules that consider the semantic and

structural information. These rules are related to the

behavioral and organizational perspectives. The

behavioral rules exploit only the structural aspect.

The organizational rules rely on the structural and

semantic (business context) information.

In summary, the majority of refactoring

approaches, presented in literature, use the semantic

and/or structural. However, there is no works that

take into account the temporal aspect. Evidently,

neglecting this aspect reduces the amount of

information that can be extracted and, therefore, may

reduce the scope for potential improvement solutions.

Our proposed method build on our preliminary work

in (Khlif et al., 2017) and combines structural,

semantic and temporal aspects in order to improve the

BP performance (ie. reduce the time, cost) and the

understandability and modifiability of the BP.

3 BACKGROUND

(ISO/IEC 25010, 2011) provides a guide for the use

of the international standard called Evaluation and

Requirements Software Quality Requirements and

Evaluation (SQuaRE). SQuaRE is composed of five

divisions. This paper focuses on the quality model

division which proposes eight quality characteristics

of the models (products): functional suitability,

reliability, performance efficiency, usability,

Temporal, Semantic and Structural Aspects-based Transformation Rules for Refactoring BPMN Model

129

operability, security, compatibility, maintainability

and transferability.

Each characteristic is composed of a set of sub

characteristics. For example, according to this

classification, understandability, modifiability and

reusability are sub-characteristics, respectively, of

usability and maintainability. In addition,

performance efficiency is shown by the sub

characteristics time behaviour and resource

utilization. Reliability is revealed by the sub

characteristics maturity and fault tolerance.

The understandability is expressed by the sub-

characteristics (appropriateness, recognisability). It is

defined as the clarity degree of the objectives of a

system for the evaluator (Azim et al., 2008).

Modifiability is defined by the easy modification

of a process model (Azim et al., 2008) while re-

usability allows to reuse BP fragments.

Time behaviour and Resource utilization are the

sub characteristics of the performance efficiency that

is defined by the capability of the BPMN element to

provide a performance relative to the resources and

the time used (Heinrich and Paech, 2010).

Time behaviour is defined as the appropriate transport

time between different BPMN elements and

processing times when executed; while Resource

utilization represents the capability of the BPMN

element to use appropriate amounts and the types of

resources when executed under stated conditions.

Maturity and fault tolerance are the sub

characteristics of the reliability which is determined

by the capability of the activity to maintain its

performance (Heinrich and Paech, 2010). Maturity is

the capability of the activity to avoid failure in the

activity. Fault tolerance is the capability of the

activity to maintain its performance in cases of faults.

To evaluate the presented characteristics, we use

measures that are shown in sections 2.1.1 and 2.1.2

and we propose also a set of performance measures

that cover all BPMN concepts and the social aspect

(actor) which affect the performance of a BP.

4 MEASURES FOR BUSINESS

PROCEESS PERFORMANCE

Before proposing a set of new measures, we define in

this section the availability and suitability

characteristics that are related to the actor.

Availability is the capability of the actor to be able to

perform the activity in the required unit of time.

Suitability focuses on actor skills that cover his

qualification, expertise, social competence, skills,

motivation and performance ability.

In addition, we propose the characteristic cost

which is expressed as a price or monetary value

associated to BPMN element and to actor in a period

of time.

Next, we propose measures that will be classified

according to the presented characteristics.

4.1 Measures to Assess the Suitability

and Availability of the Actor

In this sub-section, we propose the following

measures related to the Actor suitability:

 Shift Time of an Actor to perform an Activity

(ShT

Act

(A)): a period where an actor is

scheduled (planified) to perform an Activity.

 Actor’s BReaks when he performs an Activity

(BR

Act

(A)) : unproductive time where the actor

is scheduled not to work. A scheduled time

when workers stop working for a brief period.

 Ideal Cycle Time of an Actor to perform an

Activity (ICTAct(A)) : Theoretical minimum

time to perform an activity by an actor.

 Stop Time of an Actor when he performs an

Activity (ST

Act

(A)) : the time where the actor

was intended to work but was not due to

unplanned stops (breakdowns) or planned stops

(changeovers).

 Planned Production time of an Actor to

perform an Activity (PPT

Act

(A)): the total time

that an actor is expected to produce.

 Working Time spent by an Actor to perform an

Activity (WT

Act

(A)): It corresponds to the run

time which is simply calculated by the

difference between the Planned Production

Time and Stop Time.

 Percentage of an Actor Time Contribution in a

Lane per day (PTC

Act

(L)): it represents the

proportion of the working time spent per day

by an actor in a lane (L) and the total working

time of the same actor in all the process P.

 Performance of an Actor per Day (PerDay

Act

It expresses how fast the actor’s work ? In

addition, it represents all elements that causes

the process to operate at less than the maximum

possible speed, when running. It compares the

working Time spent by an actor per day to the

Ideal Cycle Time.

 Ratio of Defected Activities by an Actor per day

(RDA

Act

): is calculated by the Total Number of

Defected Activities performed by an actor

divided by the Total number of Activities

performed by the same actor.

ICE-B 2019 - 16th International Conference on e-Business

130

 Ratio of Good Activities performed by an Actor

(RGA

Act

): is calculated by the Total Number of

Good Activities realized by an actor in a day

divided by the Total number of Activities

performed by the same actor in one day.

 Availability of an Actor in a Day (AVDay

Act

represents the capability of the actor to be able

to perform the activity in the required unit of

time. It is calculated as the ratio of Working

Time spent by an actor on a day to Planned

Production Time.

4.2 Measures to Assess the Cost of the

Actor

In this sub-section, we propose cost measures related

to actor.

 Cost of an actor in a Lane per Day

(CosDay

act

(L)): is calculated by the product of

the total working time spent by an Actor in a

Lane per Day (TWTDay

Act

(L)) and its actual

Labour Costs per Hour (LCH

Act

4.3 Measures to Assess the Time of the

BPMN Elements

In this sub-section, we propose performance

measures related to BPMN elements (gateway,

sequence flow and lane/Pool).

 Lane Duration (LD): the sum of the needed

time to carry out all BPMN elements in a lane.

 Pool Duration (PD): It is calculated by the sum

of lanes duration in the process.

 Gateway Duration (GD (Gateway): represents

the duration of a gateway.

 Sequence Flow Duration (SeqFD): represents

the transfer time between BPMN elements

(activity, gateway and event).

We note that the sequence flow duration expresses

also the time lag (difference) between the start of an

element i+1 and the end of an element i.

4.4 Measures to Assess the Cost of the

BPMN Elements

In this sub-section, we propose cost measures related

to BPMN elements (gateway, activity and lane/Pool).

 Cost of an Activity realized by an actor (CA

Act

is calculated by the product of the actor’s actual

Labour Costs per Hour and the working time

spent by an Actor to perform an Activity.

 Cost of a Lane per Day (CosDay(L)):

determines the cost of all BPMN elements in a

Lane per Day. It includes the cost of transfer

time between them.

 Cost of a Sequence Flow (CosSeqF

Act

(BPMN

i+1

BPMN el

)): the product of the Sequence

Flow Duration (SeqFD) and the actor’s actual

Labour Costs per Hour (LCH

Act

5 BPM BUSINESS CONTEXT

The business context allows to classify the semantic

and temporal information according to five business

process perspectives: organizational, functional,

informational, behavioural and temporal.

The organizational perspective focuses on the

social context. It denotes "Where" and by "whom" the

activities are performed. "Pool" and "Lane" are the

main BPMN concepts in the organizational

perspective (Curtis et al., 1992). In particular, lane

and pool elements are described with the following

information: Unique identifier of the lane (ID

)/pool

(ID

, their labels, the list of actors affiliated with the

lane, their permissions and their role assignments, and

the relation between the actors that can be

hierarchical roles (Khlif et al., 2017). Recall that

hierarchical roles indicate partial ordering on roles.

Roles are partially ordered to reflect the

organizational hierarchy. Therefore, for two roles r

and r', r



r′ implies that permissions that exist within

r′ are subsumed into those in r (Khlif et al., 2017). We

extend the between actors by cooperation roles and

actors having the same position.We note that the

cooperation roles imply that the actors have the same

permissions and different roles; while the same

position indicates that the actors have the same roles

and permissions.

We extend the business context in the

organizational perspective by the semantic

information representing the social aspect related to

the actor and the corresponding temporal constraints:

the availability of an actor expressing his capability

to perform an activity in the required unit of time, his

suitability representing his capability to perform the

activity well, his performance expressing how fast his

work is and working time expressing how long an

actor performs an activity.

In the functional and temporal perspectives,

activity node represents the main concept which is

documented with the following context information:

the unique activity identifier (ID), its lane, the ID of

the actor responsible for performing it, the IDs of the

activities on which it directly depends (before and

after), the dependency type (authorization,

coordination or resource dependency), and its

Temporal, Semantic and Structural Aspects-based Transformation Rules for Refactoring BPMN Model

131

required objects which can be either shared or private

(Khlif et al., 2017). We extend this annotation by the

following temporal and semantic information:

Performance duration that denotes the starting and

finishing time of an activity, its cost, its state which

can be defected (IsDefected) or good (IsGood) and

the time lags between two activities expressing the

transfer time between them.

In addition, we define in the informational and

temporal perspectives, the business context

associated to the events. We suppose that the event in

a BPM is documented with the following context

information: Time date that specifies a fixed date

when trigger will be fired, its duration, its cost that

represents the cost of sending/receiving an event and

the time lags between event and other elements.

The behavioural and temporal perspectives focus

on the business context associated to the gateway and

sequence flow elements. We suppose that the

gateway and sequence flow nodes in a business

process model are documented with the following

context information: Unique identifier of the gateway

(ID

)/sequence flow (ID

SeqF)

, their labels, their

duration and costs. Note that the gateway can be also

expressed by the time lags between it and other

BPMN elements (gateway, activity and event)

expressing the transfer time between them.

6 TRANSFORMATION RULES

In this section, we use the proposed strategy in (khlif

et al., 2017) to define six transformation rules related

to all perspectives: functional, behavioural,

organizational, informational and temporal. In fact, to

apply a transformation rule, we browse the fragments

based on the defined top down approach (Khlif et al.,

2017). We note that the transformation rules preserve

the semantics of the transformed fragment. To prove

this property, we have compared the behavioral

profiles (Weidlich et al., 2011) of pattern fragments

before and after each transformation rule, and we

have verified that they effectively satisfy the behavior

preserving property; that is, both models have

equivalent trace sets.

In order to propose the organizational and

temporal rules, we studied the possible cases that may

be included in a model well-formed according to the

BPMN meta-model. Our method considers the

sequential tasks in the same lane (R1) and in different

lanes (R2) where the availability and the suitability of

the actors are an important factor for the

transformation rules. The same factors are considered

to delete a department (R3). In addition, R4 and R5

are represented by parallel fragments containing

duplicate tasks in different lanes where all the tasks

or the set of tasks are defected. In this case, R4

replaces a parallel fragment by a sequential one and

R5 removes department containing only defected

tasks. Finally, R6 presents conditions that allow

duplicating an activity.

R1: Merge directly connected activities

performed by two actors in the same lane and

associate the resulting activity with the actor who is

the most suitable and available to perform the original

activities.

Figure 1: Organizational and temporal annotations to

illustrate rule R1.

Figure 1 shows the semantic and temporal

information for annotating tasks and assigning actors

expressing the organizational and temporal aspect.

Since "Ali" is the leader of "Salah", the permission

attributed to "Salah" is also attributed to "Ali". The

leader has been available since Monday at 10:00 to

perform "Task2" and "Task3", and "Salah" can be

available on Monday at 14:00. In this case, we affect

"Task2" and "Task3" to "Ali" who has the higher

skills comparing to "Salah". In addition, when "Ali"

performs the tasks assigned to "Salah", he can reduce

the transfer time and even eliminate it. In fact, "Ali"

can perform these tasks with less time than "Salah".

We propose to merge the three tasks in one activity:

"SP1-2-3". Figure 2 illustrates the application

example for rule R1.

Figure 2: Application example for R1.

ICE-B 2019 - 16th International Conference on e-Business

132

Figure 2: Application example for R1 (cont.).

 Measure for its detection

− Structural measures: CW, NSF, NOAJS,

NOA, TNT.

− Temporal measures: LD , PD, AD(Task2),

AD(Task3), SeqFD(Task3

Salah

, Task2

Salah

TNAT(Monday), PerDay

Ali

, AVDay

Ali,

SeqFD(Task2

Salah

,Task1

Ali

),PTC

Ali

(Lane1).

− Cost measures: CA

Salah

(Task3),

CosSeqFD(Task2

Salah

,Task1

Ali

), CA

Salah

(Task2), CosDay

Salah

(Lane1), CosSeqFD

(Task3

Salah

,Task2

Salah

), CosDay (Lane1).

 Improvements after refactoring: increase the

understandability, modifiability and reusability

of the model. Decrease time behaviour, cost

and increase the availability and suitability of

the actor.

R2: Merge directly activities related and

performed by two actors with hierarchical roles in

two different lanes. Associate the resulting activity

with the actor who is available to carry out the

original activities and who has skills to save time.

Figure 3: Organizational and temporal annotations to

illustrate rule R2.

Figure 3 presents the information about the tasks

and the actors that perform each task. In this case,

since "Salah" is under the hierarchy of "Ali", the

permission attributed to "Salah" is also attributed to

"Ali". In addition, we note that "Ali" can be available

before Salah. In fact, we assign "Task3" to "Ali" who

can start performing this task on Tuesday since 10:00.

R2 affects "Task1", "Task2" and "Task3", to "Ali"

who has the skills to perform them in "Lane1". This

rule is illustrated in Figure 4.

Figure 4: Application example for R2.

 Measure for its detection

− Structural measures: CW, NSF, NOAJS,

NOA, TNT.

− Temporal measures: AD(Task3), LD, PD,

SeqFD(Task2

Ali

,Task1

Ali

), SeqFD

(Task3

Salah

, Task2

Ali

), SeqFD (Task4

Sami

Task3

Salah

), AVDay

Ali

, PTC

Ali

(Lane1),

PerDay

Ali

, TNAT(Tuesday).

− Cost measures: CosDay

Salah

(Lane2),

CosSeqFD (Task2

Ali

,Task1

Ali

), CosSeqFD

(Task3

Salah

,Task2

Ali

),CosSeqFD(Task4

Sami

ask3

Salah

),CA

Salah

(Task3), CosDay(Lane2).

 Improvements after refactoring: Increase the

understandability, modifiability and reusability

of the BP. Decrease time behaviour, cost and

increase the availability and suitability of the

actor.

R3: Delete a lane containing only activities that

are sequentially linked to other activities in another

lane and which are performed by actors related by a

hierarchical role. Then affect the resulting activity to

the actor who is available and has the most

appropriate skills (suitability) to perform the original

activities.

Figure 5: Organizational and temporal annotations to

illustrate rule R3.

Temporal, Semantic and Structural Aspects-based Transformation Rules for Refactoring BPMN Model

133

Figures 5 and 6 show an example where rule R3

can be applied. Since "Sami" is under the hierarchy of

"Ali", the permission attributed to "Sami" is also

granted to "Ali". In addition, the leader has been

available since Monday at 11:00 to perform "Task3"

and "Task4" within "Lane1". However, "Sami" can be

available on Monday at 12:00. In this case, we affect

"Task3" and "Task4" to "Ali" who has the higher

skills comparing to "Sami". We note that the more the

suitability value is minimal, the more the actor is

suitable to perform the tasks. R3 proposes to merge

"Task1", "Task2", "Task3" and "Task 4" into one

process: "SP1-2-3-4" and grant it to "Ali".

Figure 6: Application example for R3.

 Measure for its detection

− Structural measures: CW, NSF, NOAJS,

NOA, NL, TNT.

− Temporal measures: AD(Task3), LD, PD

AD(Task4), SeqFD(Task3

Sami

,Task2

Ali

SeqFD (Task4

Sami

, Task3

Sami

), PerDay

Ali

AVDay

Ali

, PTC

Ali

(Lane1), TNAT(Monday).

− Cost measures: CA

Sami

(Task3),

Sami

(Task4), CosDay(Lane2), CosSeqFD

(Task3

Sami

, Task2

Ali

), CosDay

Sami

(Lane2),

CosSeqFD (Task4

Sami

,Task3

Sami

 Improvements after refactoring: Increase the

understandability, modifiability and reusability

of the model. Improve performance by

decreasing time behaviour, cost and optimize

the availability and suitability of the actor.

R4: Replace a parallel fragment containing the same

or a subset of defected activities that belong to

different lanes by a sequence fragment, if the

activities are performed by actors affiliated to

different lanes and having cooperation relation.

Associate then the obtained sequential fragment to

skilled actor who can perform activities and terminate

them correctly.

Figure 7: Organizational and temporal annotations to

illustrate rule R4.

Figure 7 represents semantic and temporal

information used to annotate tasks and actors. Since

"Sami" and "Ali" have cooperation relation, the perm-

ission granted to "Sami" is also attributed to "Ali".

"Ali" can perform "Task1" correctly, while "Sami"

produces "Task1" in "Lane1" which represents failure

due to internal errors. In this case, we assign "Task1"

to "Ali" and delete the corresponding one associated

to "Sami" in "Lane1". However, "Sami" has more

skills than "Ali" to perform "Task2". R4 delete then

"Task2" in "Lane2" and give it to "Sami". This rule is

illustrated in Figure 8.

Figure 8: Application example for R4.

 Measure for its detection

− Structural measures: CW, NSF, NOAJS,

NOA, TNT, TNG, NL.

ICE-B 2019 - 16th International Conference on e-Business

134

− Temporal measures: LD, PD, TDADay

Sami

TADay

Sami

, RDA

Sami

, SeqFD(Task1

Ali

, G1),

SeqFD(G2, Task3

Sami

), SeqFD(G2, Task2

Ali

SeqFD(Task2

Ali

,Task1

Ali

), SeqFD (Task1

Sami

G1), SeqFD(Task2

Sami

, Task1

Sami

PerDay

Sami

, AVDay

Sami

, PTC

Sami

(Lane1),

GD(G1),GD(G2), TNAT (Monday).

− Cost measures: CA

Sami

(Task1),

Sami

(Task2), CosDay(Lane1), CosDay

(Lane2), CosSeqFD(Task1

Ali

, G1),

CosSeqFD (Task2

Ali

, Task1

Ali

), CosSeqFD

(Task2

Sami

,Task1

Sami

), CosDay

Ali

(Lane2),

CosSeqFD(G2, Task2

Ali

), CosGat(G1),

CosDay

Sami

(Lane1), CosSeqFD(Task1

Sami

G1),CosSeqFD(G2, Task3

Sami

), CosGat(G2).

 Improvements after refactoring: Increase the

understandability, modifiability and reusability

of the model. Increase performance since the

cost and the time are reduced, improve

reliability and fault tolerance and optimize the

availability and suitability of the actor.

R5: If a parallel fragment contains the same or a

subset of activities belonging to different lanes and if

one lane has only defected tasks that can be performed

by an actor who is related to other actors in another

lane by hierarchical roles or by having the relation

cooperation or the same position, then apply in order:

R4, R3.

Figure 9: Organizational and temporal annotations to

illustrate rule R5.

Figure 9 and 10 show an example where R5 can

be applied: "Lane2" contains only "Task1" and

"Task2" all of which are performed by "Ali". Since

"Ali" and "Sami" have the same position (Figure 9),

then rule R5 suggests to replace the parallel fragment

by a sequence fragment, to remove "Lane1" and to

associate "Task1" and "Task2" with "Ali" who is

allowed to perform them and can therefore do them

correctly.

Figure 10: Application example for R5.

 Measure for its detection

− Structural measures: CW, NSF, NOAJS,

NOA, TNT, TNG, NL.

− Temporal measures: LD, PD, TDADay

Sami

TADay

Sami

, RDA

Sami

, SeqFD(Task1

Ali

, G1),

SeqFD(Task2

Ali

,Task1

Ali

),PTC

Sami

(Lane 1),

GD(G1),GD(G2), SeqFD(Task1

Sami

, G1),

SeqFD(G2, Task2

Ali

), SeqFD(Task2

Sami

Task1

Sami

), SeqFD(G2, Task2

Sami

), PerDay

Ali

AVDay

Ali,

TNAT(Monday).

− Cost measures: CA

Sami

(Task1),

Sami

(Task2), CosGat(G2), CosGat (G1),

CosDay(Lane1), CosSeqFD (Task1

Ali

, G1),

CosSeqFD(Task1

Sami

,G1),CosSeqFD(Task2

Ali

, Task1

Ali

), CosSeqFD (G2,Task2

Ali

CosSeqFD(G2,Task2

Sami

),CosSeqFD(Task2

ami

,Task1

Sami

), CosDay

Sami

(Lane1).

 Improvements after refactoring: Increase the

understandability, modifiability and reusability

of the BP. Increase performance by decreasing

time behaviour, cost. Improve reliability and

fault tolerance and optimize the availability and

suitability of the actor.

R6: Duplicate a defected activity in a lane that

doesn’t belong to them, if it is followed and/or

preceded by a parallel fragment(s) which is

performed by an actor in another lane and who is

available and suitable to perform the original activity

correctly.

Figure 11 shows the semantic and temporal

information for annotating tasks and assigning actors

expressing the organizational and temporal aspects.

Temporal, Semantic and Structural Aspects-based Transformation Rules for Refactoring BPMN Model

135

Figure 11: Organizational and temporal annotations to

illustrate rule R6.

Figure 11 illustrates that "Sami" performs a

defected sub process "SP4-5" in "Lane2". In addition,

we note that "Sami" is under the hierarchy of "Ali"

and "Salah". So, the permission attributed to "Sami"

is also attributed to "Ali" and "Salah". Since "Salah"

is available before "Ali" and has more skills (skills

level =1) than him (skills level =2), he performs "SP4-

5". R6 assigns "SP4-5" to "Salah" who terminates

"SP4-5" correctly before "Ali". Figure 12 illustrates

this rule.

Figure 12: Application example for R6.

 Measure for its detection

− Structural measures: CW, CFC, NSF, TNG,

NOAJS, NOA, NL, TNT, CNC, Den.

− Temporal measures: GD(G2), TADay

Sami

TDADay

Sami

, TNAT(Monday), RDA

Sami

SeqFD(G2, Task7

Salah

), SeqFD(SP4-5

Sami

G2), AVDay

Salah

, PD, LD, PTC

Salah

(Lane3),

PerDay

Salah

− Cost measures: CosDay

Salah

(Lane2),

Sami

(SP4-5), CosSeqFD(G2, Task7

Salah

CosDay(Lane2),CosSeqFD(SP4-5

Sami

, G2),

CosGat(G2).

 Improvements after refactoring: Increase the

understandability, modifiability and reusability

of the model. Decrease time behaviour and cost

which allow to improve the performance of the

actor and therefore that of the BP. Improve

reliability and fault tolerance.

In summary, some of the proposed transformation

rules have negative aspects. For instance, R3 leads to

the deletion of "Lane2" that represents a department

of a company. This induce loss of some information

which might be necessary to better understand the

entire process. To preserve the semantic of the

fragments before and after the application of R3, the

designer should rename the remaining lane "Lane 1-

2" since it will contains also "Task 3" and "Task 4".

In addition, R5 propose to merge two identical

parallel process streams into one stream. This might

be feasible for a make-to-stock process where the

output is not that important. However, this type of

refactoring can be problematic for make-to-order

process types since the merge of two parallel process

streams into one stream with the reduction of the

workforce in this process results in a decreased

output. To verify that fragments in R5 satisfy the

behavior preserving property before and after

transformation, we eliminate only the actor who is not

suitable in "Lane1" and who produced defected

activities. In this case, we preserve the quantity of the

produced output since the actor in "Lane2" performs

the corresponding tasks very well.

7 APPLICATIVE EXAMPLE

In order to illustrate the transformation rules, we use

the "Sales management of items" example modelled

with BPMN in Figure 14. The model is annotated by

temporal constraints and semantic information (cost

and organizational aspects) that help designer to

refactor the BPMN model. Figure 13 shows the

semantic and temporal information for annotating

tasks and assigning actors expressing the

organizational and temporal aspect. The

transformation of the "Sales management of items"

example is illustrated in Figure 15.

For instance, we suppose that two actors ("Salah"

and "Hedi") have a hierarchical relation in the

"Payment" lane. Since "Salah" is the leader of "Hedi",

the permission attributed to "Hedi" is also attributed

to "Salah". The leader is available since Monday at

ICE-B 2019 - 16th International Conference on e-Business

136

08:40 to perform "T4: Deliver bill", and "Hedi" is

available on Monday at 09:00. In this case, we affect

"T4: Deliver bill" to "Salah" who has the higher skills

(skills level =1) comparing to "Hedi" (skills level =2).

In addition, when "Salah" performs the task assigned

to "Hedi", he can reduce the transfer time and even

eliminate it. In fact, "Salah" can perform this task

with less time than "Hedi". He terminates this task at

08:45. After that, R1 merges "T3: Establish payment"

and "T4: Deliver bill" in one activity: "SP2: Pay bill".

Figure 13: Organizational and temporal annotations to

illustrate the "Sales management of items" example.

Figure 14: "Sales management of items" example before the

transformation.

Furthermore, "Ali" performs a defected sub

process "SP1: Manage order" in the "Order

management" lane. We note that "Ali" is under the

hierarchy of "Sami" and "Salah". So, the permission

attributed to "Ali" is also attributed to "Sami" and

"Salah". Since "Salah" is available before "Sami" and

has more skills (skills level=1) than him (skills

level=2), he performs "SP1: Manage order".

Consequently, R6 assigns "SP1: Manage order" to

"Salah" who terminates it correctly before "Sami". To

preserve the semantic of the fragments before and

after the application of R6, the designer should

rename the remaining lane "Order management and

payment" since it will contains also "SP1: Manage

order".

Figure 15: "Sales management of items" example after the

transformation.

8 CONCLUSION

In this paper, we focused on improving the quality of

BPMN models in term of their complexity and

performance. To end this purpose, we first enriched

the existing measures by proposing a set of cost and

temporal ones related to BPMN elements and actors.

These measures are based on business context.

In addition, we proposed a set of transformation

rules that consider the semantic, structural and

temporal information to refactor a BPM. The

structural aspect describes the connections between

the BP elements. The semantic aspect is derived from

the context-enriched nodes in BPMN, and the

temporal aspect represents the temporal constraint

related to BPMN elements and to the actor.

Our future work will focus on defining an

algorithm to decide on the application order of the

transformations to produce the best performance of a

BPM. Then, we will study the impact of their

application to a well-structured model. In addition,

we are going to focus on analysing the correlations

among the transformation rules in order to ensure the

production of an optimal restructured BPMN.

Temporal, Semantic and Structural Aspects-based Transformation Rules for Refactoring BPMN Model

137

REFERENCES

Azim, A., Ghani, A., Koh Tieng, W.G., Muketha, M., Pei

Wen, W., 2008. Complexity metrics for measuring the

understandability and maintainability of business

process models using goal-question-metric. In Inte.

Jour. of Computer Science and Network Security, V. 8.

Battsetseg, B., Ahn, H., Lee, Y., Park, M., Kim, H., Yoon,

W., Pio Kim, K, 2013. Organizational closeness

centrality analysis on workflow-supported activity-

performer affiliation networks. In ICACT’15,

International Conference on Advanced Communication

Technology, PyeongChang, Korea,

Boulmakoul, A., and Besri, Z., 2013. Scoping enterprise

organizational structure through topology foundation

and social network analysis. In INTIS’3, International

Conference on Innovation and News Trends in

Information Systems, Tanger.

Curtis, B. Kellner, M. Over, J. (1992), "Process Modeling",

Communication of the ACM, 35(9).

Cardoso, J., 2006. Process control-flow complexity metric:

An empirical validation. In SCC’11, Inter. Conf. on

Services Computing, Chicago, USA. pp. 167–173.

Corradini, F., Ferrari, A., Fornari, F., Gnesi, S., Polini,

A., Re, B., Spagnolo, G.O., 2018. A Guidelines

framework for understandable BPMN models. In Inter.

Jour. of Data & Knowledge Engineering, pp. 129-154.

Del-Río-Ortega, A., Resinas, M., Cabanillas, C., Ruiz

Cortés, A., 2013. Defining and Analysing Resource-

Aware Process Performance Indicators. In CAiSE’25,

International Conference on Advanced Information

Systems Engineering, Ljubljana, Slovenia, pp. 57-64.

Fernández-Ropero, M., Pérez Castillo, R., Cruz-Lemus,

J.A. and Piattini, M., 2013. Assessing the best order for

business process model refactoring. In SAC’28, Annual

Conference of the Symposium on Applied Computing,

Coimbra, pp. 1400-1406.

Gruhn, V., Laue, R., 2006. Complexity metrics for business

process models. In BIS ’09, Inter. Conf. on Business

Information Systems, Klagenfurt, Austria. pp. 1–12.

Heinrich, R., Paech, B., 2010. Defining the Quality of

Business Processes. In Modellierung’10, 4th Inter.

Conf. of Modellierung, Klagenfurt, Austria, pp.133-

148.

ISO/IEC25010, 2011. Software and system engineering–

Software product Quality Requirements and Evaluation

(SQuaRE) Guide to SQuaRE. International

Organization for Standarization.

ISO/IEC19510, 2013. Object Management Group Business

Process Model and Notation BPMN.

Kajan, E., Faci, N., Maamar, Z., Loo, A., Pljaskovic,

A., Sheng, Q.Z., 2014. The Network-Based Business

Process. In International Journal of IEEE Internet

Computing, Volume 18, pp. 63-69.

Khlif, W., Ben-Abdallah, H., Ben Ayed, N.E., 2017. A

methodology for the semantic and structural

restructuring of BPMN models. In International

Journal of Business Process Management, V. 23, pp.

16-46.

Kim, H., Ahn, H., Kim, K.P., 2014. Modeling, discovering,

and visualizing workflow performer-role affiliation

networking knowledge. In Inter. Jour. of Transactions

on Internet and Information Systems, V. 8, pp. 689-706.

Kis, I., Bachhofner, S., Di Ciccio, C., Mendling, J., 2017.

Towards a Data-Driven Framework for Measuring

Process Performance. In BPMDS’17, 18

International

Conference on Enterprise, Business-Process and

Information Systems Modeling, Germany, pp. 3-18.

Lanz, A., Reichert, M., Weber, B., 2016. Process time

patterns: A formal foundation. In International Journal

of the Information Systems, Volume 57, pp. 38-68.

La Rosa, M., Wohed, P., Mendling, J., ter Hofstede,

A.H.M., Reijers, H.A., Van der Aalst, W.M.P, 2011.

Managing process model complexity via abstract

syntax modifications. In Inter. Jour of IEEE Tran-

sactions on Industrial Informatics, V. 7, pp. 614-629.

Oinas-Kukkonen,H., Lyytinen, K., Yoo, Y., 2010. Social

networks and information systems: ongoing and future

research streams. In International Journal of the

Association for Information Systems. Volume 11, pp. 3.

Rolón E., Ruiz F., García, F., Piattini M. (2006), "Applying

software process metrics inbusiness process model",

University Center Tampico-Madero, 89336 Tampico,

Tamps. September, México. RPM-AEMES, V.3, ISSN:

1698-2029.

Sampathkumaran, P.B., Wirsing, M., 2013. Financial

Evaluation and Optimization of Business. In Inter.

Journal of Information System Modeling and Design,

Volume 4, Number 2, pp. 91-120

Weidlich, M., Polyvyanyy, A., Desai, N., Mendling, J.,

Weske, M., 2011. Process Compliance Analysis based

on Behavioural Profiles. In International Journal of

Information Systems, Volume 36, pp. 1009-1025.

ICE-B 2019 - 16th International Conference on e-Business

138