Ontology-Driven Conceptual Modeling for Early Warning Systems:

Redesigning the Situation Modeling Language

João L. R. Moreira

, Luís Ferreira Pires

, Marten van Sinderen

and Patricia Dockhorn Costa

Faculty of Electrical Engineering, Mathematics and Computer Science, University of Twente, Enschede, The Netherlands

Computer Science Department, Federal University of Espírito Santo (UFES), Espírito Santo, Brazil

Keywords: Early Warning System, Public Health Surveillance, Situation Modelling Language, Events Processing.

Abstract: An early warning system (EWS) is an integrated system that supports the detection, monitoring and alerting

of emergency situations. A possible application of an EWS is in epidemiological surveillance, to detect

infectious disease outbreaks in geographical areas. In this scenario, a challenge in the development and

integration of applications on top of EWS is to achieve common understanding between epidemiologists

and software developers, allowing the specification of rules resulted from epidemiological studies. To

address this challenge this paper describes an ontology-based model-driven engineering (MDE) framework

that relies on the Situation Modelling Language (SML), a knowledge specification technique for situation

identification. Some requirements are realized by revisiting SML, which resulted in a complete redesign of

its semantics, abstract and concrete syntaxes. The initial validation shows that our framework can accelerate

the generation of high quality situation-aware applications, being suitable for other application scenarios.

1 INTRODUCTION

In public health epidemiological surveillance an

early warning system (EWS) is a system of systems

that supports the detection, monitoring, decision

making, alerting and responding to outbreaks and

epidemics (Lai et al., 2015). These capabilities are

addressed by EWS components, e.g. sensors`

system for collecting data and a complex event

processing (CEP) for situation detection.

A challenge for software designers in the

development of applications that identify situations

in this domain is to clearly specify the rules over the

data, which are usually identified as results of

epidemiological studies. These rules reflect

temporal, causal and existential relationships.

Moreover, such components must realize a set of

non-functional requirements, e.g. changing rules at

runtime and adequate interoperability to integrate

with healthcare and emergency systems. Model-

driven engineering (MDE) is a common approach to

realize these needs (Boubeta-Puig et al., 2015, Bui

Thi Mai et al., 2015, Sobral et al., 2015, Martínez-

García et al., 2015, De Nicola et al., 2012, Lim Choi

Keung et al., 2015). In this paper we describe the

advances of our MDE-based development strategy

to create and integrate high quality applications on

top of EWS. We stress the redesign of the Situation

Modelling Language (SML), a graphical language

that supports the representation of real-world

situations, allowing developers to build systems that

react upon these situations. Here we describe how

SML can evolve to fulfil the requirements imposed

by the epidemiological context.

Ontology-driven modelling techniques are

extremely suitable for context modelling (Guizzardi

et al., 2015). SML redesign was grounded in

foundational ontologies, which provide

interpretation for theories of Barwisean situation

semantics and Endsley`s situation-awareness,

discussed in (Moreira et al., 2015). Although the

SML redesign was motivated by requirements from

epidemiology, the resulting language is not restricted

to this domain, but can still be used in other domains

whenever it is necessary to describe situations.

Section 2 summarizes the EWS framework we

developed for the improvement of semantics in

situational awareness. It includes some of the public

health surveillance requirements, the specification

strategy, technologies and the methodology used.

Section 3 presents our SML redesign in terms of its

semantics and syntax. Section 4 presents some

related work. Section 5 presents our final remarks.

L. R. Moreira J., Ferreira Pires L., van Sinderen M. and Dockhorn Costa P.

Ontology-Driven Conceptual Modeling for Early Warning Systems: Redesigning the Situation Modeling Language.

DOI: 10.5220/0006208904670477

In Proceedings of the 5th International Conference on Model-Driven Engineering and Software Development (MODELSWARD 2017), pages 467-477

ISBN: 978-989-758-210-3

467

2 EWS FRAMEWORK

The current United Nations agenda for disaster risk

reduction (Sendai) establishes the high-level

requirement of integrating global warning and

response systems (Bello-Orgaz et al., 2015).

2.1 Requirements

The data sources characterize if the epidemiological

surveillance system is event-based (reports, rumours

and unstructured data) or indicator-based (traditional

practices with official and structured reports). In

both cases these systems implement a set of rules

over the data to detect initial outbreaks, having

epidemiological studies as the foundation to identify

health outcomes such as infectious agents,

symptoms, risk factors and transmission behaviour.

In this context, measures of frequency and

association play an important role to describe how

likely a person can be infected in accordance to a

specific population and to identify cause-effect

factors. The measures of frequency are descriptive

multivariate functions (risk, odds, rate and

prevalence), which consider variables such as the

number of new and total cases, the number of

individuals at risk and all individuals in a

population, the number of individuals with and

without the health outcome and the total person-time

at risk. Measures of association aim at comparing

the association between a specific exposure and a

health outcome. An epidemiological study about the

risk of Zika outbreaks in Europe is described in

(Rocklöv et al., 2016), which uses information about

Aedes mosquito populations, climate (temperature

and precipitation), peak flow of air travellers and

areas for mosquito-borne transmission of Zika.

Once epidemiologists draw results from these

studies, they report their analyses through natural

language description of the complex rules over the

context elements required to detect the outbreak

situations. These rules must be implemented by

software developers responsible for the development

and integration of EWS for epidemiological

surveillance, which implies that the rules should be

understandable for them. The main functional

requirement here is to improve the common

understanding of such rules and the involved

context, i.e. to achieve successful communication

among epidemiologists and software modellers by

decreasing the semantic gap among them.

An EWS must provide temporal reasoning over

the events being monitored (Lai et al., 2015), which

relies on temporal relations to correlate events. For

example, the situation of a possible contagion of

Zika is characterized by the situation of high risk of

Zika infection overlapping the situation of a person

diagnosed with Zika within the same geographical

area. “Overlapping” is a temporal relation between

the two situations, usually referred as an operator

from Allen’s calculus for temporal reasoning.

In addition, an EWS must implement temporal

existential rules, i.e. rules that only match events

occurring in a specific time window (ranges of time

units). For example, the case of influenza-like illness

(ILI) is defined by WHO as “an acute respiratory

infection with: measured fever ≥ 38 C° and cough;

with onset within the last 10 days”. Therefore, ILI is

characterized by the existence of fever and coughing

within the past 10 days. Sliding time windows

allows the definition of existential rules for EWS

(Liu et al., 2015). An EWS for public health

surveillance also needs to allow the definition of

descriptive multivariate and aggregation functions,

for example to describe average epidemic levels and

sums in populations (Marsh et al., 2016).

An EWS requires a dynamic and adaptive

approach to be able to change the applications at

runtime (Al-Khudhairy et al., 2012). Moreover, an

EWS needs to have high interoperability to receive

upstream data from e-Health sensor platforms in

accordance to electronic health records standards,

e.g. HL7; and to send downstream information for

medical staff, e.g. EDXL (Wächter and Usländer,

2014). Finally, the specification of the EWS

components must be supported by a modelling tool

widely adopted in the healthcare community

(Martínez-García et al., 2015).

2.2 Architecture

To address the requirements listed above in the

development of an EWS for epidemiological

surveillance, we proposed an ontology-based MDE

framework in (Moreira et al., 2015). Figure 1 (top)

shows the specification and implementation phases

for the development of components of an integrated

EWS in our framework. Figure 1 (bottom) also

shows the integrated EWS at runtime. This

framework guides the implementation of

components (applications) with different roles in an

EWS, e.g. an application that monitors properties of

patients, as body temperature and other symptoms of

Zika, and an application that monitors posts about

Aedes mosquitos in social media. As usual in MDE,

the implementation is (partially) generated by MDE

transformations from the specification models. Here

the implementations make use of technologies such

MODELSWARD 2017 - 5th International Conference on Model-Driven Engineering and Software Development

468

as rule-based systems, complex event processing

(CEP) and business process management (BPM).

Figure 1: MDE framework for integrated EWS.

Our MDE-based framework has been designed to

support the needs of software designers to describe

situations, the response actions and to facilitate the

generation of CEP code and services compliant to

interoperability standards and ontologies. CEP is a

technology for detecting event patterns in real time,

usually applied for situation inference (Boubeta-Puig

et al., 2015). CEP is supported in our framework by

the SCENE platform, which extends Drools Fusion,

to implement situation-aware applications as rule-

based systems. It adds the notion of situation

management to the Drools Rule Language through

metadata annotations, applied for epidemiological

surveillance (Costa et al., 2016). The Situation

Notification Service platform (SiNoS) distributes the

processing of situation management of SCENE,

consisting of a service broker and a message-

oriented middleware (MOM) following the publish-

subscribe pattern. Business processes can be

triggered through an enterprise service bus (ESB)

whenever a situation instance is activated. For

example, emergency plans specify evacuation

actions that can be implemented as data flows within

an ESB and called as a subscriber of a service

exposed by SiNoS. These services follow

appropriate standards according to the type of

information. For example, an instance of a situation

identified in SCENE is transformed to EDXL data

structure, as described in (Moreira et al., 2016).

2.3 Specification Strategy

We consider both structural and behavioural models

as suggested in (Brambilla et al., 2012). We propose

the use of a high expressive modelling language that

includes clear real-world semantic distinctions to

model the observed contextual elements and the

situation patterns to be identified. We adopted the

ontology-driven conceptual modelling approach

(Guizzardi et al., 2015), where structural modelling

is grounded in a foundational ontology. In particular,

we use OntoUML, an extension of UML (a well-

founded ontological language), based on the Unified

Foundational Ontology (UFO). OntoUML can solve

classical problems of structural languages, e.g.

ambiguity and low expressivity of UML associations

(Guizzardi et al., 2015). OntoUML was chosen to

address the expressiveness requirement to specify

the context of EWS. Our core ontology, coined

OntoEmerge, was designed with OntoUML and

includes basic elements for the specification of

contextual elements, such as health units, patients,

risks, installations and plans. These elements can be

mapped from the context model specification onto

implementation code as Java classes.

SCENE addresses the requirements of

implementing temporal reasoning, existential rules

and descriptive multivariate and aggregation

functions, as well as the dynamic adaptive behaviour

to change rules at runtime, due to the rule-based

nature of Drools. The realization of epidemiology

requirements with SCENE (Costa et al., 2016)

showed that these capabilities could be useful to

support the definition of the SML. SML is a

graphical language for modelling complex rules,

having MDE transformations for SCENE (Costa et

al., 2012). In SML a Situation Type (ST) is the

representation of patterns of contextual elements in

time. A ST allows the characterization of rules

among these contextual elements and their changes

caused by events. SML relies on the theory of the

three levels of situation awareness: “(i) the

perception of the elements in the environment within

a volume of time and space, (ii) the comprehension

of their meaning and (iii) the projection of their

status in the near future” (Wickens, 2008).

Originally, the semantic formalization of SML

considered a mapping from the syntactic elements of

the language to a logic framework (frame-based

models). This framework was chosen because

frame-based models deal with the temporal aspects

discussed above, having frames and their properties

as a modelling primitive. Two essential (disjoint,

complete) categories of STs were defined: (i) Simple

ST is interpreted as an open sentence formula which

includes free variables (e.g. parameters), the

situation itself and the variables representing the

contextual elements (characterized by predicates);

and (ii) Complex ST has the same characteristics of a

simple situation but it considers new predicates for

Ontology-Driven Conceptual Modeling for Early Warning Systems: Redesigning the Situation Modeling Language

469

each situation that is part of the composition of the

complex (whole) situation, i.e. it allows the use of

situation participant.

There are two ways to compose a complex ST

with situation participants (references to other STs),

i.e. relate the participants within the whole ST. (1)

the use of equals relation between entity (or relator)

participants of the ST referred by the situation

participant, and (2) the use of temporal (Allen)

relations between situation participants.

One important aspect of SML is the management

of the situation lifecycle, i.e. the creation (activation)

and the deactivation of an instance of a ST. One or

more events in the context characterize the creation

of an instance of a ST. For example, the event of

increase over 37 degrees of a person`s body

temperature triggers the ST of fever, i.e.

characterizes the creation of an instance of fever ST.

This instance is said to be active (or current) while

the constraints hold; and is said to be deactivated

(past) when the constraints are not satisfied

anymore. A fever ST instance holds in time while

the body temperature of a person is greater than 37

degrees, but it is deactivated when the temperature

measured over the time stops to satisfy the

constraint. An instance is never reactivated, e.g. if a

fever instance is deactivated because the temperature

measured changed from 38 to 37 and right after the

temperature measured is 38, a new instance of fever

is created, rather than reactivating the deactivated

instance. The temporal relations between STs

depend on the status of each ST participant (current

or past). This is .reflected in the temporal relations

accepted by SCENE.

The framework also includes the design of the

reaction to a ST with BPMN, i.e. the processes to be

executed when a situation is detected (activated), as

emergency plans. The designer can also specify

template-based messages to be sent during the

process execution. Once the context model is

described with OntoUML, the situations with SML

and the behaviour with BPMN, the designer can

make use of the cyclic verification and validation

approach introduced in (Sobral et al., 2015). This

approach allows syntax verification of the context

model and visual validation through simulation (also

called situation assessment). This approach uses the

Alloy logic language to automatically generate

possible instances of the STs and their participants

as object diagrams that the user may inspect to find

whether the model represents intended or unintended

state of affairs. This is supported by MDE

transformations from OntoUML/SML to Alloy. The

software designer uses this validation process

repeatedly, until sufficient confidence in the models

is achieved. Then, SCENE code can be

automatically generated from the SML model.

Here we emphasize the fundamental role of SML

in our framework. Although the initial version of

SML (1.0) covered temporal reasoning and

existential rules, it did not possess the necessary

expressiveness. SML 1.0 did not consider

foundational aspects, as the explicit distinction of

context, situation and event, as well as their

association (including causal) relationships. These

distinctions are necessary to properly formalize the

measures of association and causality. (Sobral et al.,

2015) discusses the verification and validation of

SML with OntoUML, but SML 1.0 still fails to

represent, for example, the mutability and

cardinality of participants, relationships and self-

reference of STs, functions and aggregation

functions of collections. These elements are also

required for the definition of the measures of

frequency within epidemiological studies. These

limitations have inspired our SML redesign.

3 SML REDESIGN

The SML redesign is discussed in terms of

semantics (meaning of the elements) and its

metamodel, as suggested in (Brambilla et al., 2012).

3.1 Foundations (Semantics)

In colloquial language, the term situation is often

used with the same meaning as context. The term

situation is also often used as a synonymous of

event, especially in the CEP community. Experience

with SML 1.0 showed that this overloaded and

ambiguous use of these terms brings difficulties in

understanding the language. In this work we

explicitly distinguish among these concepts. As

foundations to support the choices made here we

studied the Barewisean situation semantics theory,

its extension for the GFO ontology (the Situoid

theory), the perdurantism theory behind UFO and

the Endsley`s SA theory for human factors (Moreira

et al., 2015). Formal ontology and human factors

communities discuss whether the situation concept

is an endurant or a perdurant. Following

endurantism, a situation is a “snapshot” of the

observed world, a static object. In contrast,

following perdurantism, a situation has a temporal

and spatial extent. In particular, the situoid theory

(Herre and Heller, 2005) deals with this issue by

including the concept of situoid, a perdurant

MODELSWARD 2017 - 5th International Conference on Model-Driven Engineering and Software Development

470

element that is the composition of snapshots

(situations) in time. Situoid is a combination of

infinitesimal slices, where each slice is a situation

framed by its initial and final timestamps, i.e. a

situation is a projection of a situoid on time

boundaries. A Situoid is composed by situations that

satisfy the configuration of a set of rules among

events changing the contextual elements.

Definition (Contextual Element): In this work we

refer to a structural entity (an endurant) that

participates in a context as a contextual element to

avoid the overloaded use of the term entity, having

similar meaning to an infon. A contextual element

can be perceived as a complete concept even if we

were able to freeze time. A context is composed of

one or more contextual elements.

Definition (Context): Context is “what can be said

about an entity in its environment, i.e. context does

not exist by itself. The context of an entity may have

many constituents, called context conditions.

Examples of context conditions of a person are the

person’s location, mental state, and activity.

Together, these context conditions form the entity’s

context.” (Costa, 2007). Context refers to the data

elements being perceived from the observed world,

i.e. the first level of the SA theory (Wickens, 2008).

A context can be represented using a structural

modelling language, i.e. context model is a

conceptual model of context. Analogous to the

ontological categories of moment in UFO (intrinsic

and relational), we define two main categories of

context: intrinsic context and relational context. An

intrinsic context belongs to the essential nature of a

single entity, not depending on relationships with

other contextual elements, e.g. the body temperature

of a person. On the contrary, a relational context

depends on the relation between different entities.

Definition (Event): An event (or perdurant) is an

occurrence. An event can be atomic or complex. An

atomic event occurs in a moment in time, does not

last, i.e. its begin and end time points are the same.

A complex event is an individual composed of other

event instances, i.e. it accumulates temporal parts,

extending in time (Guizzardi et al., 2013). An event

is responsible for the transition between states of a

contextual element. Examples of events are: an

increase of a patient’s body temperature, the arrival

of an ambulance, a recommendation of

hospitalization and a bite of a mosquito. Events are

responsible for the lifecycle of a relationship (or

relator in OntoUML), i.e. a relationship exists

according to the occurrence of events. For example,

the event of a health professional giving first aid to a

patient can initialize an instance of a treatment

relationship between the patient and a health unit.

This instance depends on other events to keep their

existence, such as the patient taking medicine.

Definition (Situation): “A situation is a special

configuration which can be comprehended as a

whole and satisfies certain conditions of unity

imposed by certain universals, relations and

categories associated with the situation” (Herre and

Heller, 2005). Comprehension refers to the second

level of the SA theory, i.e. the understanding on how

the data elements perceived (contextual elements)

relate to each other in a way that it can be

recognized as a whole. Therefore, comprehension is

the reasoning capability of the phenomena observer

– in SML context the software designer and domain

experts. Universals refer to the homonymous

category in UFO. A situation is a part of the reality

that can be comprehended as a whole, has duration

and can be past (we use the term deactivated) or

current (actived). Therefore, a situation is an

individual that is composed by other individuals,

including the states of contextual elements and other

situations, constrained by formal and temporal rules.

For example, in the situation of “John being infected

with Zika”, the contextual elements are John, Zika

and (possibly) an Aedes mosquito. The event that

triggers this situation is the bite of the mosquito on

John skin, followed by the event of the virus

entering in John’s blood flow.

We classify simple and complex situation as

fundamental categories, avoiding the use of the term

type to reflect the abstraction of a domain-related

situation, the initial idea of ST. In order to address

this abstraction of types of events, types of situations

and their causal relations, we employed the multi-

level theory embedded in UFO-MLT (Carvalho et

al., 2015). We say that situation and event are

individuals, while situation type and event type are

first-order types. Thus, an event is instance of (iof)

an event type, while a situation iof a situation type.

“A mosquito biting a person” is an instance of an

event type. “A mosquito biting John’s skin in a

specific moment” is an instance of an event and, by

transitivity, is iof that event type. Similarly, the

situation type of “possible contagion of Zika within

a geographical area” is a powertype of the situation

of “a contagion of Zika in Brazil in 2016”. With this

distinction, we can abstract causal relationships

among types of event and situation, e.g. the event

type of “a mosquito biting a person” can cause the

situation type of “possible contagion of Zika”.

Ontology-Driven Conceptual Modeling for Early Warning Systems: Redesigning the Situation Modeling Language

471

When developing a situation-aware application

that relies on data events we need to differentiate an

event observed in the real world and its

representation as a digital data. For example, the

increase of a body temperature of a patient is an

event observed in real world, while the specific

measurement collected by a thermal sensor attached

to the patient is the representation of this event.

Therefore, a situation creation trigger event is a

simple event (observed in real word) that evidences a

situation creation event that activates a situation in

the software system. The same approach is used for

the situation deactivation trigger event, an event

perceived from real world (e.g. body temperature

decreases to 37) that evidences the situation

deactivation event (the digital data collected by the

sensor and sent to the software), which is

responsible for deactivating the situation instance

(e.g. fever) in the software system.

Figure 2: Situations x Events.

Figure 2 illustrates the metamodel with these

most fundamental concepts, considering the

properties of an event and the derived properties of a

situation. An event has a begin and end time points.

The end time point of an event is equal to the begin

start time point of the situation triggered by this

event. For example, the situation of “Zika

transmission in one person” is triggered by the

Aeedes mosquito bite (a situation creation trigger

event), therefore the begin time point of this situation

is equal to the end time point of the event. The

temporal properties of the event are also responsible

for the current property of a situation. When an

event triggers a situation, the situation instance is

created and the current property is set to true as

default. When a situation deactivation event occurs,

the current property of the situation is set to false

and its end time point is set to the same value of the

end time point of the event, i.e. the situation instance

is deactivated and made immutable. This enables to

design a situation composed of deactivated

situations (historical data), such as “intermittent

fever” situation, which is a situation of repeatable

deactivated “fever” situations within a time period.

An atomic event depends on a contextual

element, i.e. a structural object plays the role of a

participant of an event. For example, a person

participates in the event of “body temperature

changing” while the “body temperature changing”

existentially depends on a person. This definition is

aligned with UFO-B in terms of the existential

dependency between objects and events (Guizzardi

et al., 2013). Similar to the properties of situation

derived from event, the participants of a situation are

derived from the contextual elements participating in

the events related to the situation (the creation and

deactivation trigger events). This property is

transitive for composite situations, i.e. participants

of a composite situation are derived from the

participants of the situations within the composite

situation.

3.2 SML Metamodel

The abstract syntax of our SML redesign is

described here in terms of its metamodel by

discussing the improvements necessary to address

the requirements of the epidemiology scenario.

OntoUML improves the expressiveness of the

context models. For example, OntoUML allows the

dynamic classification of entities through modality

types, one of the main benefits of an ontological

language (Guizzardi et al., 2015). This is achieved

by the clear distinction of rigidity: the instance must

instantiate a class while it exists, and non-rigid

concepts: the instance can instantiate a class

contingently. Table 1 summarizes the mappings

between the SML and OntoUML metamodels. In

OntoUML a Quality is related to a substantial, while

a Property is an attribute of a substantial (as a class

property). Intrinsic context in the former SML was

used to represent both qualities, i.e. properties that

can be directly evaluated (e.g. temperature and

location), and modes, i.e. properties that cannot be

directly evaluated (measured) in terms of a single

space (e.g. a person’s headache). Since OntoUML

provides this distinction, we absorb it by mapping

the AttributeReference (SML) to Quality or Property

(OntoUML), and by introducing ModeReference

(SML), mapping to Mode of OntoUML. In addition,

a relational context in SML is derived from the

notion of Relator in OntoUML, which is a moment

representing objectifications of relational properties.

SituationTypeComplex specializes SituationType

MODELSWARD 2017 - 5th International Conference on Model-Driven Engineering and Software Development

472

and allows the use of SituationParticipant in its

diagram, which is a reference for a SituationType

and has properties isCurrent, beginTime and

endTime. The property isCurrent reflects the

TemporalKind enumerator, characterizing whether

the SituationType is present or past. In addition,

SituationParticipant has the ability of exposing each

of the EntityParticipant that composes the

referenced SituationType. A SituationTypeComplex

also allows the use of each specialization of

AllenLink (e.g. before, during, overlaps) to relate

two different SituationParticipant. A

SituationTypeSimple also specializes SituationType,

but does not allow the use of SituationParticipant

neither AllenLink.

Table 1: Mappings from OntoUML to SML metaclasses.

SML OntoUML

ContextModel Model

AttributeReference Quality/Property

ModeReference Mode

QualityLiteral ReferenceStructure/DataType

TypeLiteral SubstantialClass

EntityParticipant SubstantialClass

RelatorParticipant Relator

ContextFormalLink FormalAssociation

CharacterizationLink Characterization

MediationLink Mediation

A SituatioTypeElement can be a

ReferableElement, a Literal or a

SituationTypeAssociation. A ReferableElement can

be a ModeReference, an AttributeReference, a

Function or a Participant. ModeReference and

AttributeReference are derived from OntoUML

(mode and quality, respectively), while Function

represents a relation between elements not defined in

the context model for some reason (e.g. scope

limitation). A Function can be a calculus or

derivation function, i.e. a user-defined operation,

between one or more AttributeReference and

QualityLiteral, which play the role of parameters of

the function. A Participant can be a

SituationParticipant, an EntityParticipant or a

RelatorParticipant. A Literal can be a QualityLiteral

or a TypeLiteral, derived from OntoUML. A

QualityLiteral can be used only when related

through an OrderedComparativeLink or EqualsLink.

A TypeLiteral can be used only when related

through an InstanceOf to an EntityParticipant.

Mutability was addressed in SML 1.0 to specify

whether exists (or not) the occurrence of a

SituationParticipant element in a ST with the exists

logical quantifier, representing the “at least one”

instance rule. In SML 2.0 we extended this

capability to EntityParticipant and

RelatorParticipant so it is possible to specify if an

entity (or relator) exists in a ST. This is defined by

the meta-attribute immutable in the metaclass

Participant. Mutability is defined apart from

multiplicity (through cardinality) because the goal of

cardinality is to allow the specification of minimum

and maximum numbers of instances. Participant

cardinality is addressed through the meta-properties

max and min ([1..1] as default) in the metaclass

Participant. With mutability it is possible to specify

an ILI situation for example, i.e. the model of the

rule “if exists fever and coughing within the past 10

days”. With cardinality it is possible to specify an

intermittent fever. SML also allows the specification

of descriptive multivariate functions, i.e. user-

defined operations, through the Function metaclass.

Examples of these functions are sum, square,

division or even more complex functions, as

multivariate polynomial integrals and derivations.

Figure 3 illustrates the main elements of the SML

2.0 discussed here.

Figure 3: Main elements of SML 2.0 metamodel.

SituationTypeAssociation represents all possible

relationships in a SML model. To create a

relationship in MDG technology a stereotype needs

to extend a metaclass named of Association, having

the properties of the relation described as attributes

of this metaclass. For example, the style of the line

is described by the attribute _lineStyle, source and

target multiplicities are set by _SourceMultiplicity

and _TargetMultiplicity, respectively. Direction,

reflexivity, symmetry, cyclicity and transitivity are

defined through attributes too.

In OntoUML a Quality, Property and Mode are

dependent on a relationship to a substantial (e.g. a

Kind or a Role). This substantial is mapped to SML

Ontology-Driven Conceptual Modeling for Early Warning Systems: Redesigning the Situation Modeling Language

473

as an EntityParticipant (or a RelatorParticipant) and

an AttributeLink relationship is created to link them.

Besides AttributeLink, SML presents other two

relationships that are derived from OntoUML: (i)

CharacterizationLink, which reflects the

characterization relationship between a mode and an

entity (links a ModeReference to an

EntityParticipant); and (ii) MediationLink, which

reflects the mediation relationship between a relator

and an entity (links a RelatorParticipant to an

EntityParticipant).

SML 2.0 also allows the creation of comparative

relationships through (i) domain-specific formal (as

OntoUML formal relationship), (ii) primitive

(equals, greater than, less than) and (iii) Allen

(before, overlaps, meets, etc.) relationships. This is

addressed by, respectively: (i) the

ContextFormalLink, (ii) the EqualsLink and the

specializations of the OrderedComparativeLink and

(iii) the specializations of the AllenLink. An example

of using a primitive relationship to specify the rule

of the fever ST is: person (EntityParticipant) body

temperature (AttributeReference) greater than

(OrderedComparativeLinkGreaterThan) 37 degrees

(QualityLiteral). The main difference to SML 1.0 is

that each Allen relator was introduced as a metaclass

specialization of AllenLink, improving the

expressiveness of SML.

Regarding the composition of STs, an issue of

SML 1.0 was to limit the relation of

EntityParticipant (or RelatorParticipant) within a

SituationParticipant to EntityParticipant (or

RelatorParticipant) within the complex ST only

with the equals relationship, which had the side-

effect that the EntityParticipant (or

RelatorParticipant) should be bounded by the

temporal space in which the complex ST occurs. To

address this limitation, SML 2.0 allows exposing not

only the EntityParticipant (or RelatorParticipant) of

the SituationParticipant, but also the attributes

(AttributeReference) of these participants. Therefore,

in addition to equals relation from a participant of a

ST, the new metamodel allows to use of

AttributeLinks of a participant to be exposed in a

SituationParticipant.

Dynamic classification was partially covered by

SML 1.0, by using past situations to indicate

whether a situation participant was an instance of

some type in the past. This is a limited solution

because it is not possible to address the cases in

which a participant is no longer an instance of this

same type in the present. For example, a ST of

“healed patient” is composed by a past situation of

“has any ongoing treatment”, which exposes the

reference to the patient instance, used to link to the

person entity (through a comparative relation equal).

This model allows an instance of a person that is not

being treated anymore but can still be a patient,

which is an unwanted instance. To address this

issue, we introduced the instance of relationship in

SML, indicating if contextual element is instance of

a type or not. In the example scenario, the negation

of this relation can be applied to the person entity,

characterizing that in the current situation the person

is not an instance of patient anymore. This approach

also solves the problem of representing “past

specialization”, a common issue in conceptual

modelling.

Another issue of SML 1.0 regards that specific

formal relations were needed to specify that a past

situation (of the same ST) occurred at some point in

the past. For example, in (Costa et al., 2012) the

formal relation “within the past” had to be created in

the example of AccountUnderObservation ST,

embedding the semantics of the Allen relation

before, which is unwanted since it is considered a

workaround. To address this problem, the

SelfParticipant was added as a specialization of

SituationParticipant with temporality set to current,

allowing the designer to use AllenLinks to relate a

ST reference to the ST being composed (the “self”,

which is always the current situation).

Regarding the use of Function, the function

parameters (input(s)) are represented by the

FunctionParameterLink, which is a relationship

from an AttributeReference or a Literal to the

Function. The FunctionResultLink is the relationship

used to specify the output(s) of a Function.

OrderedComparativeLinks and EqualsLink can be

used from a Function when the Function has only

one output and the designer wants to specify a

comparative relation with this output. For example,

the ST “High Body Mass Index (BMI)” uses the

“square” Function with a person`s height as input,

producing the squared height (as output), which is

used as input, along with the person`s weight, for the

function “division”. The greater than relationship is

used from the output of “division” to the literal “25”

(if it is greater than 25, it characterizes a high BMI).

3.3 Validation and Discussion

An initial validation to measure whether SML 2.0

satisfies the requirements listed in section 2.2 was

performed by using the example scenario of Zika

outbreak. Figure 4 illustrates the “possible Zika

contagion” ST (ST1). On top, the “High risk Zika

infection” ST is based on (Rocklöv et al., 2016),

MODELSWARD 2017 - 5th International Conference on Model-Driven Engineering and Software Development

474

relating three STs in the same geographical region

(within 50 km): (ST2) “adequate conditions for

Aedes proliferation”, dependent on temperature,

precipitation and hydrology of the location; (ST3)

“Aedes mosquitos reported in the past year”; and

(ST4) “Person travelling from infected area”. On

bottom, “Person diagnosed with Zika” (ST5) is

activated if there is a positive result of specific

exams for a person. The temporal relation

characterizes that ST1 overlaps ST5, while the

location of ST1 must be near the location of the

infected person from ST5.

Figure 4: “Possible Zika contagion” designed with SML.

SML graphical notation was designed according

to the easiness of assimilation and comprehension of

the chosen symbols (Moody, 2009). Visual cognitive

principles were used in this evaluation, such as: (i)

semiotic clarity: correspondence between a semantic

constructor and a graphical symbol achieved by the

one-to-one mapping from the concrete syntax

classes to the graphical elements; (ii) perceptual

discriminability: the easiness and precision in which

graphical symbols can be different to each other is

achieved by distinct textures, forms, colours and

brightness of the graphical elements of SML. For

example, a SituationParticipant is a diagram

reference of a ST, when having the tagged value

isCurrent set to true the label background is greyed,

otherwise white; (iii) visual expressivity: the number

of visual variables among the elements is addressed

by distributing the use of these variables among the

graphical vocabulary to optimize the cognitive load

required for interpretation. The symbols in SML are

differentiated by two variables at least. For example,

ModeReference and TypeLiteral are distinct by

colour and shape; (iv) graphic economy: the control

of the number of distinct graphical symbols is

addressed by a balanced approach between adding

new symbols or only differentiate them by minor

details, such as textual differentiation. Although

using text is not considered a good practice for the

perceptual discriminability, this choice was made for

the existential composition of situations, to avoid the

increase of the language complexity. When the

immutable attribute of the specializations of

Participant are set to false, the existential symbol

(∃) is added before the name of the Participant.

Negation (∃!) is also possible with negationMutable.

The concrete syntax of SML indicates that the

improvement of common understanding of rules can

be achieved by the higher expressivity of the

language supported by the graphical symbols.

Temporal reasoning specification is addressed by

Allen`s operators among STs. Temporal existential

rule is addressed by mutability in SML (the

existential symbol). It is clear that, although we used

SCENE technology to support the SML redesign,

SML 2.0 is independent of any implementation

technologies and is independent of a specific

domain. SML`s semantic and syntax address the

epidemiology and CEP requirements without

referring exactly to them, therefore being general.

4 RELATED WORK

There is a number of MDE approaches in healthcare

(Martínez-García et al., 2015), but only few address

epidemiological studies. Kendrick language is a

DSL that supports epidemiological modelling,

through mathematical models for deterministic,

stochastic and individual-based simulation (Bui Thi

Mai et al., 2015). It differs from ours because it is a

DSL that supports epidemiologists on the creation of

their studies, while our approach emphasizes on

using the findings of epidemiological studies to

EWS for real-time epidemiological surveillance. The

TRANSFoRm (FP7 project) Query Workbench (Lim

Choi Keung et al., 2015) is a MDE platform

enabling search of patient`s electronic health records

from distributed clinical data repositories.

Equivalent to Kendrick, it supports epidemiologic

studies, but enabling health data extraction from

Ontology-Driven Conceptual Modeling for Early Warning Systems: Redesigning the Situation Modeling Language

475

disparate sources. It is founded on the Clinical Data

Integration Model, a core ontology grounded on the

Basic Formal Ontology (BFO). Another ontology-

based approach for epidemiological data integration

was described in (Pesquita et al., 2014) (Epiwork

FP7) with the Epidemiology Ontology (EPO). EPO

describes epidemiology-specific terms and their

relations, supported by the OBO Foundry guidelines,

grounded on BFO and based on other sources, as the

Infectious Disease Ontology, SNOMED-CT and

Unified Medical Language System.

Developing interoperable context-aware

applications for e-Health is explored in (Cardoso de

Moraes, 2014), a survey of the most common

standard-based approaches for healthcare ubiquitous

computing, e.g. HL7, Multilevel Healthcare

Information Modelling, EN13606 and openEHR.

Although none of these standards are specific for

epidemiology, they support the communication of

upstream data to EWS. In the context of MDE to

support disaster risk reduction, the approach of (De

Nicola et al., 2012) introduces the Crisis and

Emergency Modelling Language (CEML), a

graphical language to describe emergency scenarios

with critical infrastructures. It includes modelling

constructs as security, rescue and expert teams,

emergency vehicles and equipment. It extends

SysML and uses OCL. CEML does not address

temporal reasoning and comparative relations.

MEdit4CEP (Boubeta-Puig et al., 2015) addresses

this gap through Model4CEP language by

abstracting CEP constructs. It introduces a graphical

notation with simple and complex events, pattern

timers and operators, data windows and aggregation

operators. SML has a higher abstraction level (for

domain experts) than Model4CEP (for software

programmers); therefore, they are a complementary

approach. Another complementary approach is

RuleML, a XML standard used for forward and

backward rules description, supporting the exchange

of rules among CEP platforms.

5 CONCLUSIONS

In this paper we described our MDE framework for

the development of situation-aware applications that

compose public health surveillance early warning

systems (EWS). In particular, we described the

evolution of the detection specification element, the

Situation Modelling Language (SML), for modelling

rules resulted from epidemiological studies,

improving common understanding among

stakeholders. According to (Brambilla et al., 2012),

“semantics is often neglected in the definition or in

the usage of the language. (…) It doesn’t make any

sense to define a language without fully specifying

the conceptual elements that constitute it and their

detailed meaning”. Therefore, we gave emphasis to

the description of SML semantics and the

consequences on the abstract and concrete syntaxes,

grounded on the theories of situation-awareness and

situation logics and a foundational ontology (UFO).

The current version of SML was initially

validated by designing examples of situation types

based on epidemiological studies of Zika, Influenza

and Tuberculosis. As result of this validation, the

main requirement to clearly specify the knowledge

generated by epidemiological studies for software

development was addressed, due to SML`s higher

expressivity and visual cognitive principles. The

main limitations identified are the lack of

aggregation functions and Boolean operators for

composing comparative relations. Future work

comprises addressing these limitations and an

extended validation of SML in outbreak scenarios in

Europe, considering the transformations from SML

to SCENE and to EDXL. Future work includes (i)

formalization of the SML elements in UFO with first

order logic, a common approach in formal ontology

community; (ii) formal representation of causation

relationship between situation and event, based on

an epidemiological causality model, as the Bradford

Hill Criteria; (iii) transformations between SML and

Model4CEP; (iv) use of RuleML to share the

situation identification rules; (v) integration with

existing scientific workflow engines.

Although SML was redesigned to realize the

requirements of epidemiological surveillance EWS,

it can be used by non-IT experts for other types of

applications that require situation identification

modelling. SML is general enough to specify the

detection of other types of emergencies, as floods,

landslides and wildfire, as well as logistics

monitoring and crime detection. We believe that our

framework is suitable for the development of

applications to support the aforementioned domains.

ACKNOWLEDGEMENTS

Our thanks to CAPES (process BEX 1046/14-4).

REFERENCES

Al-Khudhairy, D., Axhausen, K., et al. 2012. Towards

integrative risk management and more resilient

MODELSWARD 2017 - 5th International Conference on Model-Driven Engineering and Software Development

476

societies. The European Physical Journal Special

Topics.

Bello-Orgaz, G., Hernandez-Castro, J., et al. 2015. A

Survey of Social Web Mining Applications for

Disease Outbreak Detection. Intelligent Distributed

Computing VIII.

Boubeta-Puig, J., Ortiz, G., et al. 2015. MEdit4CEP: A

model-driven solution for real-time decision making in

SOA 2.0. Knowledge-Based Systems.

Brambilla, M., Cabot, J., et al. 2012. Model-Driven

Software Engineering in Practice, Morgan \&

Claypool Publishers.

Bui Thi Mai, A., Stinckwich, S., et al. KENDRICK: A

Domain Specific Language and platform for

mathematical epidemiological modelling. In:

Research, Innovation, and Vision for the Future.

Cardoso De Moraes, J. L. 2014. Methodological support

to develop interoperable applications for pervasive

healthcare. PhD.

Carvalho, V. A., Almeida, J. P. A., et al. 2015. Extending

the Foundations of Ontology-Based Conceptual

Modeling with a Multi-level Theory. International

Conference Conceptual Modeling.

Costa, P. D. 2007. Architectural support for context-aware

applications: from context models to services

platforms. PhD.

Costa, P. D., Almeida, J. P. A., et al. 2016. Rule-Based

Support for Situation Management. Fusion

Methodologies in Crisis Management: Higher Level

Fusion and Decision Making.

Costa, P. D., Mielke, I. T., et al. A Model-Driven

Approach to Situations: Situation Modeling and Rule-

Based Situation Detection. In: Enterprise Distributed

Object Computing Conference.

De Nicola, A., Tofani, A., et al. 2012. An MDA-based

Approach to Crisis and Emergency Management

Modeling. International Journal On Advances in

Intelligent Systems.

Guizzardi, G., Wagner, G., et al. 2015. Towards

ontological foundations for conceptual modeling: The

unified foundational ontology (UFO) story. Journal of

applied ontology.

Guizzardi, G., Wagner, G., et al. Towards Ontological

Foundations for the Conceptual Modeling of Events.

In: International Conference Conceptual Modeling.

Herre, H. & Heller, B. 2005. Ontology of time and

situoids in medical conceptual modeling. Proceedings

of the 10th conference on Artificial Intelligence in

Medicine.

Lai, P.-C., Chow, C. B., et al. 2015. An early warning

system for detecting H1N1 disease outbreak – a

spatio-temporal approach. International Journal of

Geographical Information Science.

Lim Choi Keung, S. N., Khan, O., et al. 2015. A query

tool enabling clinicians and researchers to explore

patient cohorts. International Conference on

Informatics, Management, and Technology in

Healthcare.

Liu, S., Smith, K., et al. 2015. A multivariate based event

detection method and performance comparison with

two baseline methods. Water Research.

Marsh, K., Ijzerman, M., et al. 2016. Multiple Criteria

Decision Analysis for Health Care Decision Making—

Emerging Good Practices. Value in Health.

Martínez-García, A., García-García, J. A., et al. 2015.

Working with the HL7 metamodel in a Model Driven

Engineering context. Journal of Biomedical

Informatics.

Moody, D. 2009. The "Physics" of Notations: Toward a

Scientific Basis for Constructing Visual Notations in

Software Engineering. IEEE Transactions on Software

Engineering.

Moreira, J. L. R., Ferreira Pires, L., et al. 2015. Towards

ontology-driven situation-aware disaster management.

Journal of applied ontology.

Moreira, J. L. R., Ferreira Pires, L., et al. 2016. Improving

semantic interoperability of big data for

epidemiological surveillance. I-ESA, BDI4E

workshop.

Pesquita, C., Ferreira, J. D., et al. 2014. The epidemiology

ontology: an ontology for the semantic annotation of

epidemiological resources. Journal of Biomedical

Semantics.

Rocklöv, J., Quam, M. B., et al. 2016. Assessing Seasonal

Risks for the Introduction and Mosquito-borne Spread

of Zika Virus in Europe. EBioMedicine.

Sobral, V. M., Almeida, J. P. A., et al. Assessing situation

models with a lightweight formal method. In: 2015

IEEE International Multi-Disciplinary Conference on

Cognitive Methods in Situation Awareness and

Decision.

Wächter, J. & Usländer, T. 2014. The Role of Information

and Communication Technology in the Development

of Early Warning Systems for Geological Disasters:

The Tsunami Show Case. Early Warning for

Geological Disasters: Scientific Methods and Current

Practice.

Wickens, C. 2008. Situation awareness: Review of Mica

Endsley's 1995 articles on situation awareness theory

and measurement. The Journal of the Human Factors.

Ontology-Driven Conceptual Modeling for Early Warning Systems: Redesigning the Situation Modeling Language

477