A Context-Based Approach for Real-Time Adaptation Need Detection

Jamila Oukharijane

1 a

, Imen Ben Said

, Mohamed Amine Cha

abane

, Eric Andonoff

and Raﬁk Bouaziz

MIRACL Laboratory, University of Sfax, Tunisia

IRIT, University Toulouse 1-Capitole, 2 Rue du Doyen Gabriel Marty, 31042 Toulouse Cedex, France

Keywords:

Context, Real-Time Monitoring, Situation Reasoning, Situation Filtering, Adaptation Need Detection,

Context Analysis.

Abstract:

The increasingly dynamic and changing environment in which business processes are evolving requires com-

panies to adapt them frequently. Thus, we propose in this paper a new context-based approach recommending

to analyze well the environmental changes. This approach recommends to structure knowledge required for

process adaptation need in accordance with our BPMN4V-Context meta-model, which supports the modeling

of the operating environment, using the context along with the versions of the model of each BPMN process

and their use conditions. It also advocates (i) a ﬁltering activity to retain only signiﬁcant context changes in

the monitored data, as low-level context parameters, (ii) a reasoning activity to deduce high-level context pa-

rameters from ﬁltered low-level ones, enhancing the current situation of running processes and (iii) examining

the current situation before its analysis in order to resolve problems related to the used units and synonym

values. Finally, the feasibility and applicability of this approach is demonstrated by a case study from the

crisis domain and two performance tests.

1 INTRODUCTION

The dynamic environment in which companies oper-

ate forces them to frequently adapt their processes to

face changes occurring in this environment. Thus,

their capability to rapidly and efﬁciently adapt their

running processes to changes is an essential require-

ment for them. So, Business Process Management

Systems (BPMS) implementing processes often sup-

port manual adaptation; the process designer has to

identify which changes in the operating environment

require an adaptation and resolve them by deﬁning

the required adaptation operations and carrying them

out (Rinderle et al., 2004). However, the manual

process adaptation is a costly, time-consuming and

error-prone task (Masoumi et al., 2013). In fact, it

requires the presence of a business expert who must

possess knowledge of the environment in which su-

pervised processes operate to be able to identify each

adaptation need and resolve it. Consequently, self-

adaptation is seen as an effective solution to deal with

the complexity of process adaptation with minimum

human (De Lemos et al., 2013) (Oukharijane et al.,

2019).

https://orcid.org/0000-0001-6155-0215

As argued in (IBM, 2006), self-adaptation in the

general level encompasses various self-* properties

in the major level, including self-conﬁguring, self-

healing, self-optimizing and self-protecting. We con-

sider that the approach presented in this paper falls

under self-healing category. Self-healing is the ca-

pability of the adaptation engine of discovering, di-

agnosing and reacting to disruptions (Huebscher and

McCann, 2008). Self-healing can be classiﬁed into

self-diagnosing and self-repairing, where the former

concerns itself with identifying adaptation needs, and

the latter focuses on the resolution of the identiﬁed

adaptation needs, namely the deﬁnition and the ex-

ecution of the adaptation operations. As this paper

deals with the adaptation need detection issue, it pro-

vide a comprehensive approach for self-diagnosis.

On the other hand, self-adaptive systems must

have the capability of self-adjusting to the varia-

tions of their operating environment, which is of-

ten referred to as context. To address process self-

adaptation, we thus have to deal with the detection of

process adaptation needs. In the BPM area, the no-

tion of context is deﬁned in (Rosemann et al., 2008)

as “the minimum of variables containing all relevant

information that impacts the design and the execution

114

Oukharijane, J., Ben Said, I., ChaÃ

cbane, M., Andonoff, E. and Bouaziz, R.

A Context-Based Approach for Real-Time Adaptation Need Detection.

DOI: 10.5220/0012148500003552

In Proceedings of the 20th International Conference on Smart Business Technologies (ICSBT 2023), pages 114-125

ISBN: 978-989-758-667-5; ISSN: 2184-772X

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

of a process”. For his part, (Rosemann et al., 2008)

outlined that four relevant types of context must be

considered: (i) an immediate context, which covers

information on the behavioral, informational and or-

ganizational dimensions of processes, which are three

important dimensions to be considered when model-

ing processes, (ii) an internal context, which covers

information about the internal environment of an or-

ganization that impacts the processes, (iii) an exter-

nal context, which covers information about the ex-

ternal stakeholders of an organization, and ﬁnally, (iv)

an environmental context, which covers information

about external factors. If several literature contribu-

tions addressed the adaptation need detection issue,

their recommended solutions had drawbacks. Firstly,

they did not address the modeling of the operating

environment in a comprehensive way as they did not

consider all context types deﬁned in Rosemann’s tax-

onomy (Rosemann et al., 2008). Second, their models

that support the representation of the operating envi-

ronment are rather simplistic or even poor, as only

low-level context parameters are supported to model

the operating environment as a context. They do not

contain high-level context parameters inferred by a

reasoning component in order to deduce new infor-

mation that enhances the current situation knowledge.

To overcome these weaknesses, we recommend a

context-based approach, which addresses the adapta-

tion need detection issue and has the four following

features:

• First, it aims to ensure for a powerful monitor-

ing of the operating environment of running pro-

cesses, able to capture relevant context changes

and enhance context parameters. Therefore,

it allows better analysis of changes to identify

whether or not process adaptations are required.

• Second, it recommends the use of the push

mode (i.e., publish/subscribe communication), as

a technique for current situation acquisition. The

advantages of this mode are as follows: (1)

real-time monitoring and processing of context

changes and (2) loose coupling between the op-

erating environment and the adaptation engine,

which makes the integration of new sensors in the

operating environment easier as the adaptation en-

gine does not need to be modiﬁed.

• Third, it recommends the context notion to repre-

sent the operating environment of processes, i.e.,

the current situation in which running processes

operate, using context parameters belonging to

any type of context parameter types deﬁned in

Rosemann’s taxonomy.

• Finally, it recommends the reasoning on context

parameters, which contributes to the enhance-

ment of the current situation by high-level con-

text parameters and thus the improvement of the

decision-making for process adaptation.

Accordingly, the remainder of the paper is organized

as follows. Section 2 provides the state-of-the-art

on self-adaptation of processes, focusing mainly on

adaptation need detection. Section 3 introduces the

BPMN4V-Context we propose for the modeling of

knowledge required by the adaptation need detection.

Section 4 discusses in detail the context-based ap-

proach proposed for process adaptation need detec-

tion. Section 5 demonstrates the applicability of the

proposed approach on the case study, whereas, Sec-

tion 6 reports on the evaluation of the recommended

solution to demonstrate its feasibility. Finally, section

7 summarizes the paper contributions and gives some

directions for future research.

2 RELATED WORK

In the literature, there are several works that focus on

the automatic and autonomous context monitoring of

running business processes and eventually on adapt-

ing them to the occurring context changes (e.g., (Ay-

ora et al., 2012), (Ayoub and Elgammal, 2018), (Ferro

and Rubira, 2015), (Monteiro et al., 2008), (Oliveira

et al., 2013) and (Seiger et al., 2019)). Due to the

space limitation, this section considers only the recent

contributions that focus on autonomic detection of the

adaptation need.

For their part, Oliveira et al. (Oliveira et al., 2013)

introduced an approach that enables the modeling of

autonomic processes at the design-time and manag-

ing them at the run-time. At the design-time, this ap-

proach makes it possible to model an autonomic pro-

cess using models, which provide all the necessary

concepts that guide the self-adaptation of processes

at the run-time, i.e., which tasks must be monitored,

which context changes impact the execution of a pro-

cess and how to resolve them. At run-time, this ap-

proach goes through the following steps. ﬁrst, it peri-

odically requests to retrieve the value of any context

element in the operating environment. Then, it checks

all the variation points and evaluates the received val-

ues according to the modeled context in each variant

in order to detect the adaptation needs. Therefore, if

adaptations are required, it selects and executes an al-

ternative variant that satisﬁes the context of the oper-

ating environment. Moreover, this approach does not

address the separation of concerns between the adap-

tation logic and the operating environment. Further-

more, while the acquired context situations are mod-

A Context-Based Approach for Real-Time Adaptation Need Detection

115

eled according to a meta-model, the modeling of these

situations is not addressed in a comprehensive way

encompassing all the context types deﬁned in Rose-

mann’s taxonomy (Rosemann et al., 2008). In addi-

tion, this approach does not support the context sit-

uation reasoning in order to enhance the current sit-

uation with high-level context parameters; only the

low-level context situations are supported. ﬁnally, it

does not allow the detection and resolution of situa-

tion problems related to the used units and to the syn-

onymous values.

Moreover, another interesting contribution by

Ferro and Rubira (Ferro and Rubira, 2015) intro-

duced an agent-based adaptation engine for the self-

adaptation of processes at run-time. This adaptation

engine acts as follows: The monitor agent continu-

ously acquires and evaluates the current state of the

process instance and its context until it determines a

symptom that needs to be analyzed (e.g., business rule

violation). The adapter agent implements the deci-

sion making for adaptation, which is driven by goal

and business rule analysis and the operations required

for the resolution of the detected adaptation need. Fi-

nally, the executor agent sends adaptation operations

to the process engine to be executed. This contribu-

tion supports the pull mode for the acquisition of sit-

uations of internal context elements in the operating

environment, which implies that it embeds the current

situation capturing code into the adaptation engine

and therefore, leads to poor maintainability. Besides,

it does not consider all context types deﬁned in Rose-

mann’s taxonomy (Rosemann et al., 2008). Finally,

the context situation reasoning is not implemented,

which implies that the high-level context situations

are not supported.

On the other hand, Ayoub and Elgammel (Ay-

oub and Elgammal, 2018) introduced a framework

for monitoring and improving social business pro-

cess (i.e., process that integrates Web 2.0 technolo-

gies such as Facebook and Twitter). In addition to

this framework, these authors recommend a concrete

approach that enables the adaptation need detection

based on social data as it particularly uses data mining

and machine learning techniques to detect the need

for process adaptations. Thus it supports the predic-

tive analysis in detecting the adaptation needs. How-

ever, this approach does not separate the concerns

between the adaptation logic and the operating envi-

ronment. In addition, this approach is speciﬁc as it

gathers and models only social data. Thus it enables

the modeling of only external context type deﬁned in

Rosemann’s taxonomy (Rosemann et al., 2008). Fi-

nally, it does not support the interpretation of current

situation, i.e., enhancement of the current situation

by adding new knowledge and translation of situation

values to resolve problems related to their units and

synonyms.

Finally, Seiger et al. (Seiger et al., 2019) proposed

a framework that enables the self-adaptation of pro-

cesses in the cyber-physical systems. More precisely,

this framework monitors and analyzes the consistency

between the sensed physical world and the assumed

cyber world of task executions. In case an inconsis-

tency is detected, the resource involved in the con-

cerned task execution is replaced by another alterna-

tive resource, and then the task is executed. As in (Ay-

oub and Elgammal, 2018), this framework uses the

pull (i.e., request/response) mode for the retrieval of

internal context element values in the operating envi-

ronment without addressing the modeling of all con-

text types deﬁned in Rosemann’s taxonomy (Rose-

mann et al., 2008) and the separation of concerns be-

tween the adaptation logic and the operating environ-

ment. In addition, this framework does not support

the interpretation of current situation. Therefore, only

the low-level context situation is supported.

To overcome the weaknesses of the examined con-

tributions, we recommend a self-diagnosis approach

which takes into account the interesting features of

the examined adaptation contributions, namely, the

use of the context-based model to deﬁne the current

situation context of each running process. But our ap-

proach differs from them in the following respects.

First, it advocates taking into account the Rosemann’s

taxonomy for a comprehensive modeling of process

contexts, where the context parameters observed from

the immediate, internal, external and environmental

contexts of processes may be represented. Second, it

supports real-time context monitoring by implement-

ing the push technique for the current situation ac-

quisition from sensors. Finally, it recommends the

reasoning on context parameters in order to infer new

knowledges.

3 KNOWLEDGE FOR

ADAPTATION NEED

DETECTION

Our approach recommends to model knowledge re-

quired by the adaptation need detection using the

two levels of BPMN4V-Context meta-model recom-

mended in (Gamma et al., 1993). An excerpt of the

UML class diagram of BPMN4V-Context is given in

Figure 1. This ﬁgure adopts the following policies:

blue background for concepts related to private pro-

cess versions, grey background for concepts related to

ICSBT 2023 - 20th International Conference on Smart Business Technologies

116

the deﬁnition of the use conditions of versions, yellow

background for concepts corresponding to process ex-

ecutions, and ﬁnally green background for concepts

related to current situation modeling of running pro-

cess. In this ﬁgure, we particularly focus on two ab-

straction levels for the modeling of the context pa-

rameters: (i) the model level, in which we describe

the use conditions of versions of processes at design-

time, and (ii) the instance level, in which we describe

the current situation of running processes at run-time.

These levels declare the structures of this knowl-

edge as described below.

3.1 Model Level

The model level (left side of Figure 1) of BPMN4V-

Context meta-model concerns the concepts involved

in the deﬁnition of the process, sub-process and task

versions and their use conditions at design-time. In

fact, this meta-model introduces new classes to de-

ﬁne these conditions of use and the involved context

parameters. A use condition may be atomic or com-

posite. Therefore, the class of the Use condition is

specialized in two sub-classes: Atomic use condition

and Composite use condition. An atomic use condi-

tion refers to the context parameter involved in this

condition and deﬁnes the associated operator, value

and unit of measurement. A composite use condition

is an aggregation of several use conditions connected

together by logical operators (and, or). For each de-

ﬁned operator, we deﬁne the rank that indicates the

position of the use condition in the composite use

condition expression. It should be noted that the use

condition of a process (or sub-process) version is the

combination of the use conditions of activities (which

are versions of tasks or versions of sub-processes) that

make up this process (or sub-process) version.

3.2 Instance Level

This level (right side of Figure 1) of BPMN4V-

Context meta-model enables to model the operating

environment of running process versions. It deals

with the concepts involved in the deﬁnition of run-

ning process versions and their corresponding current

situations. Each occurrence of Running process is re-

lated to exactly one occurrence of Version of Process,

but an occurrence of Version of Process can be re-

lated to several occurrences of Running process. An

occurrence of Running process is composed of a set

of occurrences of Running activity. An occurrence

of Running activity can be an occurrence of Running

sub-process or of Running task. Moreover, current

situations of running tasks are modeled as instances

of the class Current situation, which may be either

atomic or composite:

• An atomic situation is deﬁned by a context pa-

rameter, a value, a timestamp and a state: con-

text parameter can be immediate, internal, exter-

nal or environmental context parameter according

to the Rosemann’ taxonomy; value speciﬁes the

acquired value by the Monitor component; unit

speciﬁes the used unit for measuring a context pa-

rameter value (e.g., degrees Celsius, pressure, me-

ter, etc.); timestamp indicates the time value when

the context parameter value was acquired.

• A composite situation is a set of atomic situations

connected together by the logical operator “and”.

Note that the current situation of a running process or

sub-process is the union of the current situations of

its running activities, i.e., running tasks and running

sub-processes.

Knowledge structures expressed at these two lev-

els are respectively implemented by the Models

repository and the Instances repository

4 THE CONTEXT-BASED

APPROACH FOR THE

PROCESS ADAPTATION NEED

DETECTION

Our approach addresses the adaptation need detection

aiming to analyze changes of the operating environ-

ment in order to identify whether or not process adap-

tations are required. This approach is based on (i)

sensors and the push mode to capture context param-

eter values in real-time, i.e., during the execution of

processes, and (ii) context notion to represent both

the operating environment of process model versions

(the current situation in which running process ver-

sions operate) and the use condition of each process

model version. Therefore, the notion of context is cru-

cial because it serves as a basis for identifying adap-

tation needs. Indeed, identifying adaptation needs is

like comparing these two contexts (i.e., current situa-

tion and use condition). The process driving this ap-

proach is shown in Figure 2 as a BPMN collaboration

diagram.

As shown in this diagram, sensors and process en-

gine listeners simultaneously observe the execution of

the running process versions and their operating envi-

ronment, and push to the adaptation engine the events

indicating changes occurring on the context param-

eters. But as indicated in (Da et al., 2014), the re-

ceived data from sensors, without any further inter-

pretation, can be meaningless, trivial, vulnerable, or

A Context-Based Approach for Real-Time Adaptation Need Detection

117

Figure 1: BPMN4V-Context meta-model as a UML class diagram.

Figure 2: Adaptation need detection process as a BPMN diagram.

ICSBT 2023 - 20th International Conference on Smart Business Technologies

118

uncertain bout small changes. To overcome this prob-

lem, our adaptation engine ﬁrst executes a situation

ﬁltering activity to ensure that the reasoning activity

will process only the signiﬁcant data changes. This

Situation ﬁltering activity ﬁlters the received data and

keeps only the signiﬁcant changes of context parame-

ters values. To do so, this activity compares the abso-

lute value of the difference between the received value

of each low-level context parameter and the last stored

value in the Context repository to a certain variation

threshold (var-threshold) stored also in the Context

repository. So, it veriﬁes if the received change of

each context parameter value is signiﬁcant (i.e., the

absolute value of the difference is higher than the vari-

ation threshold) or insigniﬁcant. In case all the re-

ceived context changes are insigniﬁcant, the Adapta-

tion engine does not trigger the situation reasoning,

nor the adaptation need identiﬁcation. Consequently,

both the reasoning and analysis times are reduced,

only by processing and analyzing the signiﬁcant con-

text changes. Otherwise (i.e., when there is one or

more signiﬁcant context changes are received), both

activities “Situation reasoning” and “Adaptation need

identiﬁcation” are triggered for enhancing this situa-

tion by high-level context parameters values and de-

tecting the need for process adaptation, respectively.

More precisely, the ﬁrst activity includes the follow-

ing both internal operations to deduce the high-level

context parameters from ﬁltered low-level ones:

• The ﬁrst operation, Situation aggregation, uses

aggregation functions (i.e., average, count, max,

sum, etc.) of (low-level/high-level) context pa-

rameters to deduce high-level context parameters.

For instance, the value of the high-level context

parameter precipitation amount is the aggregation

sum of the precipitation measurements acquired

by the “precipitation sensors” located at different

sources points of a river.

• The second operation, Situation deduction, exe-

cutes the rules that ensure deducing new high-

level context parameters using low-level/high-

level context parameters. Let us remember that

the deduction rules are deﬁned by domain experts

in the Context repository. The aim of this oper-

ation is to deduce a set of high-level context pa-

rameters enhancing the current situation of run-

ning processes, consequently helping the process

adaptation.

After executing these operations, the Adaptation

needs identiﬁcation activity is triggered to identify

the possible adaptation needs of the considered pro-

cess instance. This activity is modeled as sub-process

in the BPMN diagram, which includes the following

atomic activity: “Use condition identiﬁcation” and

“Similarity calculation”. The Use condition identi-

ﬁcation activity allows to obtain the use condition

corresponding to the considered process version V

;

it generates an appropriate query to access the Mod-

els repository implementing the Model part of the

BPMN4V-Context meta-model and retrieve the use

condition C of V

. Then, the second activity, Similar-

ity calculation, calculates the similarity between the

identiﬁed use condition C and the enhanced current

situation S of V

. The algorithm called Sim, shown be-

low, implements the similarity calculation. It receives

both use condition C and enhanced current situation

S of a process version V

as input, and returns a sim-

ilarity percentage “sim

” of V

. It calculates the sim-

ilarity value “sim” of V

as the number of its atomic

use conditions veriﬁed by the values of the context

parameters of the current situation. It compares the

value of each context parameter in the current situ-

ation “V PC

”, and the value of the context param-

eter speciﬁed in the atomic use condition “V PC

”.

Note that this comparison may require a conversion

of V PC

or its replacement with the appropriate syn-

onym when V PC

is expressed with a different unit

of measure or a synonym of V PC

S, respectively. If

an atomic use condition is veriﬁed, then sim is incre-

mented by 1. Finally, the algorithm returns the simi-

larity percentage sim

of the atomic usage conditions.

sim

is equal to the number of the veriﬁed atomic us-

age conditions of V

divided by its total number of

atomic usage conditions. This algorithm uses the fol-

lowing functions:

• convertUnit (vs, us, uc): it converts the context

parameter value vs if necessary from the unit of

measurement us deﬁned in the atomic situation to

the unit of measurement uc deﬁned in the atomic

use condition. For example, the current value of

the context parameter temperature is measured us-

ing the Fahrenheit unit (e.g., 77 °F), whereas the

unit of temperature is deﬁned in degrees Celsius

in the atomic use condition; then, the current value

of the context parameter temperature is converted

into degrees Celsius (e.g., 23 °C) by this function.

• synonyms (v): it resolves the possible synonymy

problem when the value v of a context parameter

is expressed by a synonym of that deﬁned in the

atomic use condition, using a dictionary of equiv-

alent representations. For instance, the value of

the context parameter resource availability can be

0 or 1 in the current situation, whereas the deﬁned

value of resource availability in the atomic use

condition is rather “available” or “unavailable”.

• checkCondition (v

,op

): It returns true if the

value vs of the context parameter involved in the

atomic situation is veriﬁed in the use condition,

A Context-Based Approach for Real-Time Adaptation Need Detection

119

otherwise, it returns false. More precisely, it com-

pares the current value vs of the context parameter

to the deﬁned value v

in the atomic use condition

using the operator op

• card (cs): It returns the number of atomic situa-

tions involved in the current situation cs.

Algorithm 1: Similarity calculation: Sim.

Require: S: Current situation,C: Use condition

Local

sim ← 0: real, ps: Atomic situation, c: Atomic use

condition

begin function

for each c in C do

if isCompositeCond (c) then

▷ Case where c is a composite use condition

SimilarityCalculation (S, c)

else

▷ Case of c is an an atomic use condition

for each s in S do

if (c.getContextPar () = s.getContextPar

()) then

if (s.getUnit() ̸= c.getU nit()) then

vs ⇐convertUnit (s.getValue (),

s.getUnit (), c.getUnit ())

end if

if checkCondition (vs,

c.getOperator (), Synonyms (c.getValue ()))

then

sim ⇐ sim + 1

end if

end for

end if

end for

Return sim/card(S)

end function

It should be noted that Algorithm Sim returns 1 as

similarity value when all atomic situations featuring

the current situation verify the use condition of the

considered process version. This means that the cur-

rent situation of the considered process version ex-

actly matches the use condition of this process ver-

sion. But when this algorithm returns a similarity

value ¡ 1, a need for process adaptation is identiﬁed.

In this case, a signal event (need process instance

adaptation event) is triggered for deﬁning the oper-

ations required to address this need using the hybrid

approach recommended in (Oukharijane et al., 2020;

Oukharijane et al., 2021).

5 CASE STUDY: THE FLOOD

MANAGEMENT PROCESS

This section presents the Flood Management Process

(FMP) case study and illustrates the applicability of

the approach. First, it shows how to model the Flood

Management Process and their context then, it uses

this case study to demonstrate the execution of a sam-

ple adaptation need detection using the recommended

approach.

5.1 Knowledge Modeling for FMP

This section presents a real case study related to

the process of ﬂood crisis management of the major

French river “the Loire”, on the city of Orleans. As

shown in Figure 3, this process is triggered in urban-

ized impacted area when the water level of the river

“Loire” in France rises above 3 m

. In response to this

event, the crisis cell decides whether or not to evacu-

ate people from the ﬂooded zones by assessing the

ﬂooding situation. In case an evacuation is needed,

the Prefect emits an evacuation order, then the COD,

which is the operational committee set up within the

crisis cell, and the town hall inform the media and the

population about the ﬂood. After that the gendarmes

proceed to the evacuation of people from the ﬂooded

zones. Finally, the crisis cell reports on the evacua-

tion, and the Prefect sends the report to the interior

ministry.

Like any other process, the FMP might be subject

to different operating environment variations, such as

water level, water ﬂow rate, amount of precipitation,

that could interrupt its functioning. In order to super-

vise this environment, we refer to the following con-

text parameters among others:

• Water level, which indicates the level of water

rising in the Loire. In the Orl

eans city, this param-

eter refers to the average water levels recorded at

the following two stations: the Pont Royal station

and the Quai du Roi station.

• Water velocity, which indicates the speed of the

water in the Loire in the different source points of

water to the Loire. This parameter refers to the

average water velocities recorded at Pont Royal

and Quai du Roi stations.

• Impacted area, which features the size of the

population potentially impacted,

• Road state, which can be not ﬂooded, ﬂooded and

derivable or ﬂooded and blocked. When the roads

are not ﬂooded, people evacuate themselves. On

the other hand, when the roads are ﬂooded but still

derivable, the evacuation method is carried out by

ICSBT 2023 - 20th International Conference on Smart Business Technologies

120

Figure 3: Flood management process model.

land, which means that the evacuation task is car-

ried out using speciﬁc vehicles with the help of

gendarmes. However, when the roads are blocked

because they are highly ﬂooded, people’s evacua-

tion must be carried out by ﬁreﬁghters with zodi-

acs or helicopters in the most extreme cases.

• Risk level, which can be low, moderate or high.

Indeed, the risk level of ﬂooded zones is de-

duced from the following context parameters: wa-

ter level and water velocity. Thus, when the water

level is below of 2 m

and the water velocity is

below of 2,500 m

/s, thus the risk level is ”low”,

whereas when the water level is more or equal to

2 m

and a water velocity is below of 2,500 m

/s,

so the risk level is ”moderate”. While, when the

water level is higher than 3 m

and water veloc-

ity is more than 2500 m

/s, thus the risk level is

“high”.

On the basis of the previous context description, the

parameters Impacted area, Water level of a station,

Water velocity of a station and Road state represent

the low-level context that must be measured by sen-

sors and process engine listeners. For each of these

parameters, of the numeric type, a threshold value

was deﬁned in Table 1 in order to verify if the re-

ceived change is signiﬁcant or not. We note that these

context parameters can be measured at different sta-

tion points; for example, for the Orleans city the wa-

ter level is measured at both the Pont Royal and the

Quai du Roi stations.

As for the high-level context parameters, such as

the Water level, Water velocity, Risk level, Person-

nel’s role and Type of equipment, they are deduced

from the low-level ones according to the aggregations

functions and deduction rules identiﬁed by domain

experts and which are given below:

On the other hand and according to cell crisis,

the ﬂood management process deﬁned in Figure 3 is

used in the following context (i.e., use condition): Im-

pacted area is urbanized, Water level is more than 3

, Water velocity is below of 2500 m

/s, road state

is ﬂooded and derivable, Type of equipment is ve-

hicle, personnel role of evacuation is gendarme and

Risk level is moderate.

5.2 Context-Based Approach

Simulation

In this section, we illustrate the simulation of our

context-based approach for the detection of the adap-

tation need. In order to demonstrate this, we suppose

that the ﬂood management process (cf., Figure 3) is

running for Orleans city affected by Loire’s ﬂoods.

For this illustration, we refer to the following changes

from the operating environment that need adaptation:

Water level of Pont Royal station = 3 m

and Water

level of Quai du Roi station = 4 m

and Water veloc-

ity of Pont Royal station = 3100 m

/s and Water ve-

locity of Quai du Roi station = 3300 m

and Impacted

area = “Urbanized” and Road state = “Flooded and

derivable”.

Once the adaptation engine receives these

changes, it starts its Situation ﬁltering activity to ﬁlter

and keep only the signiﬁcant changes. Then, it trig-

gers the ”Situation reasoning” activity to enhance the

current situation thanks to its both operations. The

ﬁrst operation enhances the situation thanks to aggre-

gation functions (average) and deduces the values of

the water level and velocity of the Loire from the mea-

sured values at both Pont Royal and Quai du Roi sta-

tions. These functions led respectively to the addition

of the following average values: ”3.5 m

” and ”3200

/s” for the two high-level context parameters ”Wa-

ter level” and ”Water velocity” describing the aver-

age water level and velocity of the Loire. Then the

second operation enhances the situation thanks to the

deduction rules R3 and R5 deﬁned in Table 2. Once

the deduction is made, the current situation will be

enhanced by new high-level context parameters ”risk

level”, Personnel role and ”Type of equipment” hav-

ing respectively “high”, “Gendarme” and ”vehicles”

values.

A Context-Based Approach for Real-Time Adaptation Need Detection

121

Table 1: Variation threshold of low-level parameters.

Context parameter Threshold value

Water level of a station 0.5 m

Water velocity of a station 150 m

Table 2: An expect of the deduction rules.

N Deduction rule

R1 If water level < 2 m

and Water velocity < 2500 m

/s then Risk level = ”low”

R2 If Water level ≥ 2 m

and Water velocity < 2500 m

/s then Risk level = ”moderate”

R3 If Water level > 3 m

and Water velocity ≥ 2500 m

/s then Risk level = ”high”

R4 If Road state = “Not ﬂooded” Then Type of equipment = “Vehicle”

R5 If Road state = “Flooded and derivable” Then Personnel role = “Gendarme” and Type of equipment = “Vehicle”

R6 If Road state = “Flooded and blocked” Then Personnel role = “Fireﬁghter” and Type of equipment = “Zodiac”

According to this enhanced situation and the use

condition deﬁned in the previous section, the Ana-

lyzer of the running process version calculates the

similarity using Algorithm 1. The latter gives a simi-

larity in the result equal to 0.67, which implies that the

conditions related to the context parameters of water

velocity, the risk level and the type of equipment are

not conﬁrmed, therefore, the running process needs

adaptation.

6 SIMULATION AND

EVALUATION RESULTS

We evaluate in this section our contributions, which

ensure the context monitoring and analysis function-

alities, by simulations. To this end, we perform two

performance tests. In the ﬁrst test, we evaluate the

reasoning execution times consumed by the adapta-

tion engine using or not using the Situation Filterer

to verify that the use of the Situation Filterer activity

reduces the reasoning time, and thus enhances the per-

formance of adaptation engine. Whereas, the second

test evaluates the impact of the reasoning on context

parameters to improve the outcome of the analyzer

(i.e., adaptation need, no adaptation need) using pre-

cision, recall, f-measure and error rate metrics.

6.1 The First Performance Test

Our aim in this sub-section is to report on an evalu-

ation that measures the advantages of using Situation

ﬁltering activity to enhance the performance of con-

text monitoring. To conduct this evaluation, we vary

the number of sensors/listeners from 50 to 200, and

we realize two experiments:

• First experiment (without ﬁltering the received

context changes from sensors/listeners): in this

experiment, all the received context changes are

Figure 4: Comparison of the reasoning execution times with

and without performing an upstream situation ﬁltering.

interpreted in order to enrich the current situation

with high-level context parameters.

• Second experiment (with ﬁltering the received

context changes from sensors/listeners): in this

experiment, the adaptation engine uses the vari-

ation thresholds of low-level context parameters

in order to ﬁlter the received changes by elimi-

nating all the insigniﬁcant ones. Therefore, only

notable changes are interpreted for inferring high-

level context parameters values.

Figure 4 illustrates the beneﬁts of an upstream situ-

ation ﬁltering on the reduction of the reasoning ex-

ecution times consumed by the Situation Reasoning

activity, respectively.

In the ﬁrst experiment (without using an upstream

situation ﬁltering), we let the adaptation engine inter-

prets all the received changes from sensors and lis-

teners without ﬁltering them. The results of this ex-

periment are displayed by the red charts which con-

ﬁrm that the reasoning execution times grow expo-

nentially with the number of sensors and listeners.

ICSBT 2023 - 20th International Conference on Smart Business Technologies

122

For the case of 200 sensors and listeners for exam-

ple, the time taken by the Situation reasoning activity

is about 1096.8 ms (cf. ﬁgure 4, red chart).

In the second experiment (with using an upstream

situation ﬁltering), we let the adaptation engine ﬁlters

all the received changes of context parameters val-

ues by using the variation threshold (Var-threshold)

of each context parameter and keeps only the sig-

niﬁcant ones. Figure 4 illustrates that the reason-

ing execution times are remarkably reduced, since the

context parameter value is updated only if the differ-

ence between the received value and the stored value

in the Context repository is higher than its variation

threshold. In the case of 200 sensors and listeners,

the reasoning execution time is decreased from more

than 1096.8 ms (cf. ﬁgure 4, red chart) to about

400.3 ms (cf. ﬁgure 4, blue chart) by performing a

situation ﬁltering upstream of the situation reason-

ing, for example. Indeed, although the number sen-

sors/listeners, sending changes to the adaptation en-

gine, is increased, the monitoring performance is not

seriously affected; this is the contribution of the Situ-

ation ﬁltering activity that eliminates all insigniﬁcant

context changes.

6.2 The Second Performance Test

As for the second test, we evaluate the effectiveness

of our approach for the detection of the adaptation

needs by reporting on an assessment that measures the

advantages of implementing the Situation reasoning

activity in the adaptation engine to enhance the cur-

rent situation, and thus increase the accuracy of the

adaptation needs. In this evaluation, we expose the

possible scenarios for the detection of the adaptation

needs. For each of these scenarios, we compared the

provided results with our approach in which the Sit-

uation reasoning activity is implemented against the

results provided without implementing the Situation

reasoning activity. These scenarios, ﬁve in number,

are as follows:

• SC1 (reasoning is not needed): detecting adapta-

tion needs considering (i) a use condition that con-

tains only low-level context parameters and (ii) a

sensed situation containing all the low-level con-

text parameters of the considered use condition.

• SC2 (unit translation is needed): detecting adapta-

tion needs considering (i) a use condition that con-

tains only low-level context parameters and (ii) a

sensed situation containing low-level context pa-

rameter values expressed with units different from

those speciﬁed in the considered use condition.

• SC3 (value translation is needed): detecting adap-

tation needs considering (i) a use condition that

contains only low-level context parameters and

(ii) a sensed situation containing low-level context

parameters values which are synonyms of those of

the considered use condition.

• SC4 (high-level context parameter deduction is

needed): detecting adaptation needs considering

(i) a use condition that contains both low and

high-level context parameters and (ii) a sensed sit-

uation that contains all low-level context parame-

ters of the deﬁned use condition of the supervised

task.

• SC5 (all the reasoning types are needed): detect-

ing adaptation needs considering (i) a use condi-

tion that contains both low and high-level context

parameters and (ii) a sensed situation containing

low-level context parameter values expressed with

units different from those speciﬁed in the consid-

ered use condition and others which are synonyms

of those of this use condition.

To evaluate the effectiveness of our context-based ap-

proach for the detection of the adaptation needs, we

have considered several changes related to the current

situation of the ﬂood management environment. For

each of these scenarios, we have classiﬁed the out-

come of adaptation engine using a confusion matrix.

In this matrix, each column represents the outcome of

the implemented adaptation engine (adaptation need,

no adaptation need), where each row represents the

real outcome. However, the cells of this matrix rep-

resent (i) True positives (TP), i.e., number of cases

with adaptation needs that are correctly identiﬁed, (ii)

False negatives (FN), i.e., number of cases with adap-

tation needs that are not identiﬁed, (iii) False positives

(FP), i.e., number of cases with adaptation needs that

are incorrectly identiﬁed and (iv) True negative (TN),

i.e., number of cases with no adaptation needs that

are correctly identiﬁed. Then, we have measured the

following metrics: (1) Precision deﬁnes how many

adaptation needs are correctly identiﬁed among all the

identiﬁed adaptation needs, (2) Recall deﬁnes how

many adaptation needs are correctly identiﬁed among

all the adaptation ones (correctly identiﬁed and not

identiﬁed),(3) F-measure refers to the balanced mean

between the precision and recall metrics, and (4) Er-

ror rate is calculated as the number of all incorrect

cases divided by the total number of test cases. The

best error rate is 0, whereas the worst is 1.

Table 3 summarizes the results obtained using

these metrics for the two adaptation need detection

approaches that respectively implement or not the sit-

uation reasoning.

Let us comment some of these results. For the

ﬁrst scenario SC1 and as indicated in Table 3, the

precision and recall metrics are both equal to 1 for

A Context-Based Approach for Real-Time Adaptation Need Detection

123

Table 3: Evaluation outcomes of the two context-based approaches for the detection of the adaptation needs.

Scenarios Results without considering situation reasoning Results with considering situation reasoning

Precision Recall F-measure Error Precision Recall F-measure Error

SC1 1 1 1 0 1 1 1 0

SC2 0.25 0.4 0.3 0.82 1 1 1 0

SC3 0.33 1 0.5 0.67 0.75 1 0.85 0.14

SC4 1 0.29 0.45 0.5 1 1 1 0

SC5 0 0 0 1 0.8 0.85 0.82 0.29

both approaches as the both versions of adaptation

engine (with and without implementing the Situation

reasoning activity) outcomes match exactly the real

outcomes for all scenario test cases. This is due to the

fact that each context parameter exists in both the use

condition and current situation with the same unit and

representation.

Regarding the result of SC2, detecting the adap-

tation needs with a translated situation thanks to the

Unit translator is more efﬁcient than detecting the

adaptation needs without translating the situation re-

ceived from the sensors. Indeed, the Unit translator

function takes care of translating the units of the re-

ceived situation from the sensors to the units deﬁned

in the use condition. Thus, the precision, recall and

f-measure metrics for adaptation engine implement-

ing this function are all equal to 1, and the error rate

metric is equal to 0. However, to detect the adapta-

tion needs considering the situation received from the

sensors without unit translation, the precision metric

is equal to 0.25 and the recall is equal to 0.4. This

means that (i) a set of incorrect adaptation needs are

identiﬁed for some scenario test cases and (ii) a set of

adaptation needs that must be detected are not iden-

tiﬁed for some other cases. As a consequence, the

values of the f-measure and error rate metrics are

equal to 0.3 and 0.82, respectively. As for the re-

sult of SC3, detecting the adaptation needs consid-

ering an enhanced situation thanks to the Value trans-

lator gives very acceptable results, since the values

of the precision and recall are on average equal to

0.75 and 1, respectively. On the other hand, detecting

the adaptation needs considering only the situation

received from sensors (without enhancement) sends

back the values of 0.33 for the precision and 1 for the

recall. As the recall metrics are equal to 1 for both

approaches of the detection of the adaptation needs,

it turns out that all adaptation needs that should have

been detected are identiﬁed in the different simulated

test cases from SC3. As a consequence, the values of

the f-measure and error rate metrics are equal to (i)

0.85 and 0.14, respectively, for our approach in which

the situation reasoning is considered and (ii) 0.5 and

0.67 for the approach in which the situation reason-

ing is not considered. For the result of SC4, detecting

the adaptation needs considering an enhanced situa-

tion with high-level context parameters thanks to the

Situation reasoning activity returns a value equal to

1 for the precision and recall metrics. On the other

side, detecting the adaptation needs considering a sit-

uation without enhancement returns values equal to

1 and 0.29, respectively, for the precision and recall

metrics. It should be noted that the precision met-

rics are always equal to 1, since all adaptation needs

are correctly identiﬁed in the different simulated test

cases from SC4, because they are detected on the ba-

sis of only the low-level context parameter values. As

a consequence, the values of the f-measure and er-

ror rate metrics considering the situation reasoning is

better than the values of these metrics without con-

sidering the situation reasoning. These results high-

light the importance of the Situation reasoning activ-

ity. Finally, regarding the result of SC5, detecting

the adaptation needs considering an enhanced situ-

ation is more efﬁcient than detecting the adaptation

needs without implementing the Situation reasoning

and translating. Indeed, the Situation reasoning activ-

ity (i) detects and resolves situation problems related

to the used units and to the synonymous values, and

(ii) enriches the current situation with high-level con-

text parameter values. Thus the error rate of the adap-

tation need detection is considerably reduced (from 1

to 0.29).

7 CONCLUSION

The complexity of processes, the dynamism of the en-

vironments in which they are operated, and the need

for process adaptation to context changes are growing

rapidly. So, process adaptations are often performed

manually by process designers or process owners.

However, a manual adaptation is a costly, time con-

suming and error prone task. For that reason, this pa-

per recommends an approach for real-time adaptation

need detection. Its contributions are as follows:

• It is context-based, which has enabled it to im-

ICSBT 2023 - 20th International Conference on Smart Business Technologies

124

prove the detection of the adaptation needs. The

context is represented by parameters from any

type of contexts deﬁned in Rosemann’s taxonomy.

• It recommends well modeling the knowledge re-

quired for the adaptation need detections, as al-

lowed by the two levels of BPMN4V-Context

meta-model.

• It recommends (i) the use of sensors and the push

mode to support a real-time monitoring of the op-

erating environment of processes, (ii) the context

changes ﬁltering in order to only analyze signif-

icant changes, and (iii) the reasoning on context

parameters to enhance the current situation.

• It allows comparing the operating environment of

process model versions and the use conditions of

these versions. Therefore, it serves as a basis for

identifying adaptation needs.

• It ensures the translation of the context parame-

ters values related to units and synonyms before

analyzing. This translation makes it possible to

avoid divergences of representation of these val-

ues in the current situation and in the use con-

ditions, and thus to improve decision-making for

process adaptations.

However, our approach can be improved. As future

work, we plan to incorporate ontologies to better cap-

ture the semantics of the current situation of running

processes and take advantage of semantic aspects. In

addition, we plan to study how to structure and exploit

historical data and how to conduct predictive analysis

of the current situation of the operating environment

in order to predict the future adaptation needs before

they arise. Moreover, we have to evaluate the usabil-

ity of our contributions.

REFERENCES

Ayora, C., Torres, V., Pelechano, V., and Alf

erez, G. H.

(2012). Applying CVL to business process variability

management. In VARiability for You Workshop: Vari-

ability Modeling Made Useful for Everyone, pages

26–31, Innsbruck, Austria. ACM.

Ayoub, A. and Elgammal, A. (2018). Utilizing Twitter data

for identifying and resolving runtime business process

disruptions. In OTM Confederated International Con-

ferences” On the Move to Meaningful Internet Sys-

tems”, pages 189–206. Springer.

Da, K., Montaury, P., Dalmau, M., Montaury, P., Roose, P.,

and Montaury, P. (2014). Kalimucho : Middleware for

mobile applications. In 29th Annual ACM Symposium

on Applied Computing, pages 413–419.

De Lemos, R., Giese, H., M

uller, H. A., Shaw, M., An-

dersson, J., Litoiu, M., Schmerl, B., Tamura, G., Vil-

legas, N. M., and Vogel, T. (2013). Software engi-

neering for self-adaptive systems: A second research

roadmap. In Software Engineering for Self-Adaptive

Systems II, pages 1–32. Springer.

Ferro, S. and Rubira, C. (2015). An architecture for dy-

namic self-adaptation in workﬂows. In International

Conference on Software Engineering Research and

Practice (SERP), pages 35–41.

Gamma, E., Helm, R., Johnson, R., and Vlissides, J. (1993).

Design patterns: Abstraction and reuse of object-

oriented design. Springer.

Huebscher, M. C. and McCann, J. A. (2008). A survey of

autonomic computing—degrees, models, and applica-

tions. ACM Computing Surveys (CSUR), 40(3):1–31.

IBM (2006). An architectural blueprint for autonomic com-

puting. IBM White Paper, 31.

Masoumi, A., Marrella, A., and Soutchanski, M. (2013).

Towards a Planning-Based Approach to the Auto-

mated Design of Chemical Processes. In International

Conference of the Italian Association for Artiﬁcial In-

telligence, volume 1101, pages 61–70, Trento, Italy.

Monteiro, P., Rodrigues, J., Oliveira, J., and Souza, J.

(2008). ABPM: Autonomic Business Process Man-

ager. In Latin American Autonomic Computing Sym-

posium, GRamado.

Oliveira, K., Castro, J., Espa

na, S., and Pastor, O. (2013).

Multi-level autonomic business process management.

In International Conference on Enterprise, Business-

Process and Information Systems Modeling, pages

184–198, Valencia, Spain. Springer.

Oukharijane, J., Ben Said, I., Cha

abane, M. A., Andonoff,

E., and Bouaziz, R. (2019). Towards a New Adapta-

tion Engine for Self-Adaptation of BPMN Processes

Instances. In 14th International Conference on Eval-

uation of Novel Approaches to Software Engineering

(ENASE), pages 218–225, Heraklion, Crete, Greece.

Oukharijane, J., Chaab

ane, M. A., Ben Said, I., Andonoff,

E., and Bouaziz, R. (2020). A Hybrid approach based

on Reuse Techniques for Autonomic Adaptation of

Business Processes. In International Conference on

System and Software Reuses, pages 1–17.

Oukharijane, J., Chaab

ane, M. A., Ben Said, I., Andonoff,

E., and Bouaziz, R. (2021). Self-adaptive business

processes: a hybrid approach for the resolution of

adaptation needs. Innovations in Systems and Soft-

ware Engineering, pages 1–23.

Rinderle, S., Reichert, M., and Dadam, P. (2004). Cor-

rectness criteria for dynamic changes in workﬂow

systems-a survey. Data & Knowledge Engineering,

50(1):9–34.

Rosemann, M., Recker, J., and Flender, C. (2008). Contex-

tualisation of business processes. International Jour-

nal of Business Process Integration and Management,

3(1):47–60.

Seiger, R., Huber, S., Heisig, P., and Assmann, U. (2019).

Enabling Self-adaptive Workﬂows for Cyber-physical

Systems. Software & Systems Modeling, 18(2):1117–

1134.

A Context-Based Approach for Real-Time Adaptation Need Detection

125