SATISFYING USER EXPECTATIONS IN ONTOLOGY-DRIVEN

COMPOSITIONAL SYSTEMS

A Case Study in Fish Population Modeling

Mitchell G. Gillespie, Deborah A. Stacey

School of Computer Science, University of Guelph, 50 Stone Road East, Guelph, ON, Canada

Stephen S. Crawford

Department of Integrative Biology, University of Guelph, 50 Stone Road East, Guleph, ON, Canada

Chippewas of Nawash Unceded First Nation, RR #5, Wiarton, ON, Canada

Keywords:

Ontology-Driven Compositional System (ODCS), Ontology, System composition, User expectations, Fish

population, Modeling.

Abstract:

Ontology-Driven Compositional Systems (ODCSs) are designed to assist a user with semi- or fully automatic

composition of a desired system. Current research with ODCSs has been conducted around the discovery and

composition of web services and alternatively a bottom-up resource management approach to automatic sys-

tem composition. This paper argues that current ODCSs do not truly satisfy user expectations as the semantic

knowledge required to make proper discovery, decision-making and composition has not been fully repre-

sented. The authors introduce the beginning of their work of utilizing the inheritance of multiple ontologies

to fully represent the functional, data, quality & trust, and execution of compositional units within an ODCS.

Furthermore, a case study of ﬁsh population modeling is presented.

1 INTRODUCTION

For many years, stakeholders have utilized previously

implemented algorithms and packages to minimize

the work for software developers when designing a

software system. One example is the Open Source

community that supports software developers with

the ability to manually integrate previously composed

modules and systems (Feller and Fitzgerald, 2002).

Similarly, an increasing number of Web Services al-

low developers to manually connect to remote ser-

vices via their own distributed system (Meng et al.,

2006). Recent research in this area has shifted fo-

cus to understand the requirements and processes of

Compositional Systems (Cardoso and Sheth, 2005).

Systems that could provide assistance with automatic

or semi-automatic system composition using a col-

lection of previously developed algorithms, modules,

services and packages.

To comprehend the previously implemented soft-

ware available in a Compositional System, a semantic

knowledge of the various components must be pro-

vided. Ontologies are explicit speciﬁcations of a con-

ceptualized body of knowledge, and commonly uti-

lized to understand the sharing of knowledge among

people and/or software agents (Gruber, 1993). Thus,

ontologies are recognized as a appropriate tool to rep-

resent the semantic knowledge that drives a Compo-

sitional System (Arpinar et al., 2005; Hlomani and

Stacey, 2009). Current research from Apinar (2005)

focused on a Ontology-Driven Web Service Composi-

tion System, and Hlomani and Stacey (2009) shifted

their focus to incorporate both distributed and non-

distributed components. Overall, the research of

Ontology-Driven Compositional Systems (ODCSs)

is in its youth and many gaps still need to be ap-

proached. Thus far, ODCS research has focused pri-

marily on the discovery and composition of previ-

ously developed software, and rarely considers rank-

ing and selection previously developed software to

decide which to use. A ranking and selection process

requires an understanding of the end-user’s expecta-

tions to evaluate whether or not a resulting composed

system form the ODCS is satisfactory.

Through collaboration with the Integrative Biol-

ogy Department at the University of Guelph and the

133

G. Gillespie M., A. Stacey D. and S. Crawford S..

SATISFYING USER EXPECTATIONS IN ONTOLOGY-DRIVEN COMPOSITIONAL SYSTEMS - A Case Study in Fish Population Modeling.

DOI: 10.5220/0003103801330143

In Proceedings of the International Conference on Knowledge Engineering and Ontology Development (KEOD-2010), pages 133-143

ISBN: 978-989-8425-29-4

 2010 SCITEPRESS (Science and Technology Publications, Lda.)

Chippewas of Nawash Unceded First Nation ﬁsheries

management program (Crawford et al., 2008), the au-

thors isolated a case study where they can measure

user expectations. Why ﬁsheries population model-

ing? In most cases, ﬁsheries biologists lack the math-

ematical and computational knowledge to model ﬁsh

populations properly (Megrey and Moksness, 2009)

and ODCSs could provide the appropriate knowledge

to allow the successful construction and execution of

accurate population models. Furthermore, as a ﬁsh-

eries biologist’s knowledge is specialized in popula-

tion dynamics, their expectations of a composed pop-

ulation modeling system will be more complex than

whether or not the computer population models com-

pile and execute. Domain-speciﬁc metrics of qual-

ity and trust could affect a ﬁsheries biologist’s accep-

tance of certain model components over another.

Utilizing the case study in ﬁsheries population

modeling, the authors will illustrate how satisfying

user expectations in an ODCS accommodates various

research initiatives. Explicitly, compositional systems

researchers (Cardoso and Sheth, 2005; Arpinar et al.,

2005; Hlomani and Stacey, 2009) will further under-

stand the dynamics involved in the creation of prac-

tical applications for ODCS. Also, ﬁsheries manage-

ment experts will be introduced to an innovative tool

to assist with the composition of population mod-

els. Implicitly, these population modeling experts will

also be embracing a framework for representing in-

stances of population models.

In this paper, current implementations of

Ontology-Driven Compositional Systems are as-

sessed, followed by a presentation of work being

conducted by the authors. Section 2 introduces

the deﬁnitions of System Composition and Compo-

sitional Systems and includes a framework of the

semantic knowledge and processes required within a

Compositional Systems. Section 3 assesses current

implementations using the deﬁnitions and framework

from Section 2. Section 4 presents aspects of User

Expectations desired in the design of an Ontology-

Driven Compositional System to enhance the quality

and trust of the resulting system’s outputs. Using the

desired user expectations, Section 5 introduces the

current work by the authors, including the ﬁsheries

population modeling case study. Finally, discussion

of an evaluation technique for Ontology-Driven

Compositional Systems and future work is presented.

2 SYSTEM COMPOSITION &

COMPOSITIONAL SYSTEMS

System Composition is the process by which two

or more previously implemented compositional units

(e.g. algorithms, packages, services) are constructed

together to create a holistic functional system. This

deﬁnition can refer to both manual construction by a

software developer (i.e. without the aid of any compu-

tational tools) or any type of computer assisted tech-

nique. Compositional Units are the functional ”black-

box” algorithms, packages or services that receive a

given input, that provide a function or service and

send a calculated output.

Computer-Assisted System Composition (CASC)

is a process by which a human user utilizes the aid

of a Compositional System (CS) to assist in the con-

struction of a Resulting System. Hlomani and Stacey

(2009) deﬁne a Compositional System as ”a system

that allows its components to be put together in a sys-

tematic manner to achieve a common goal or to de-

rive yet another functional application”. A Resulting

System is the ﬁnal output of interconnected Composi-

tional Units to be executed by the user. Hlomani and

Stacey (2009) refer to the Resulting System as ”yet

another functional application”. With CASC, the de-

gree of user interaction when conducting CASC could

range from requiring user input at every stage of the

System Composition to no interaction in a completely

automatic system composition.

2.1 Semantic Knowledge Requirements

All compositional systems hold some form of seman-

tical knowledge about the compositional units they

wish to utilize (Srivastava and Koehler, 2003; Car-

doso and Sheth, 2005). For web services speciﬁcally,

Cardoso (2005) described four different types of se-

mantical knowledge to be considered by web service

compositional systems. This paper adapts the work of

Cardoso (2005) to provide ﬁve different types of se-

mantics to be considered in all forms of compositional

systems, ontology-driven or otherwise:

1. Functional Semantics: knowledge about the

function, features and compositional purpose of

the compositional units and input/output data.

2. Data Semantics: knowledge about the order, for-

mat and structure of the input/output for a given

compositional unit.

3. Quality & Trust Semantics: knowledge that

characterizes how well a given compositional unit

performs and other metrics which distinguish if

certain compositional units are suitable for a given

KEOD 2010 - International Conference on Knowledge Engineering and Ontology Development

134

system composition (adapted from Cardoso and

Seth’s ”QoS Semantics” ).

4. Execution Semantics: knowledge about loca-

tions, requirements and environment dependan-

cies that need to be utilized or satisﬁed for a com-

positional unit to be executed in a resulting sys-

tem.

5. Timeline Semantics: knowledge about the com-

positional and chronological requirements to cre-

ate sections of or full implementations of a result-

ing system.

The Functional, Data, Quality & Trust, and Exe-

cution semantics represent the knowledge about each

given instance of a compositional unit, whereas the

Timeline semantics are utilized for describing appro-

priate combinations of compositional units. The col-

lection of the ﬁve semantics allow for unique speci-

ﬁcation of each compositional unit (functional, data,

quality & trust, and execution), and general combina-

tions

Let’s investigate an example. A computer popula-

tion model could be an instance of a compositional

unit (CU). This CU could be described to have a

”catch-at-age” functional purpose (Function Seman-

tics), and receive a temporal harvest vector as input

data (Data Semantics). However, the harvest vector

requires data aggregation from month to year, there-

fore another CU is required to execute that function

successfully. The Timeline Semantics would provide

the general combination of a population model CU

and data aggregation CU, while the speciﬁc semantics

would provide the ability to match the correct instance

of a data aggregation CU (i.e. a vector aggregation

tool from month to year) to the population model CU.

2.2 Discovery, Decision-making and

Composition Processes

Within computational systems three general pro-

cesses occur: discovery, decision-making and actual

composition (adapted from Cardoso (2005)). Discov-

ery refers to the process of using the semantic knowl-

edge to match compositional units that meet the re-

quirements for a desired resulting system. Decision-

making refers to the process of ranking and selecting

the discovered compositional units for the system, and

ﬁnally Composition is the process of automatically

constructing the compositional units together to form

the actual resulting system to be executed. Overall,

Figure 1 illustrates how a compositional system uti-

lizes the semantic knowledge of compositional units

to conduct discovery, decision-making and composi-

tion of a resulting system.

Figure 1: Discovery, Decision-Making and Composition

processes in a Compositional System utilizing the seman-

tic knowledge of compositional units to create a resulting

system for the user to execute. Compositional units may or

may not be locally stored and/or managed (i.e. a download-

able package would be locally stored by the compositional

system, whereas a web service would not).

All compositional systems require their seman-

tic knowledge represented in one relational form or

another. As more functions and features of compo-

sitional systems are considered, the relationship be-

tween all the different facets of knowledge becomes

more difﬁcult to maintain. Enter Ontologies!

3 ONTOLOGY-DRIVEN

COMPOSITIONAL SYSTEMS

Hlomani and Stacey (2009) deﬁne an Ontology-

Driven Compositional System as ”a prototype system

[...] that demonstrates the power and suitability of us-

ing ontologies as the main driver for a compositional

system”. Using the deﬁnitions from Section 2, this

paper states that an Ontology-Driven Compositional

System (ODCS) is a compositional systems which uti-

lizes ontologies to perform computer-assisted system

composition for a user, where the ontologies represent

the functional, data, quality & trust, execution and

timeline semantic knowledge of the included composi-

tional units. The ODCS conducts semantic reasoning

on the ontologies for the discovery, decision-making

and composition of a resulting system. Figure 2 (an

adaptation from Figure 1) illustrates the ontological

representation of semantic knowledge.

Certain computational characteristics of the com-

positional units (e.g. a web service or downloadable

package) affect the speciﬁc focus of a compositional

system. Two variations of ODCSs are presented:

Ontology-Driven Web Service Composition Systems

SATISFYING USER EXPECTATIONS IN ONTOLOGY-DRIVEN COMPOSITIONAL SYSTEMS - A Case Study in

Fish Population Modeling

135

Figure 2: Adaptation of Figure 1 to illustrate how ontolo-

gies are utilized by ODCSs during the process of system

composition.

(Meng et al., 2006; Arpinar et al., 2005) and a ”re-

source management”-based Ontology-Driven Com-

position System (Hlomani and Stacey, 2009).

3.1 Ontology Driven Web Service

Composition

With the growing popularity of distributed web ap-

plications, there is a strong focus on delivering ”Se-

mantic Web Services” (Cardoso and Sheth, 2005).

Semantic Web Services are deﬁned as instances of

web services that ﬁt within a semantic representation

for users and/or developers to discover and utilize in

their distributed software systems. Cardoso and Sheth

(2005) emphasized the importance of ontologies to

represent the semantic knowledge to allow an artiﬁ-

cial intelligence to reason and discover available web

services. Many research foci have expanded from this

initiative by proposing the use of ontologies to assist

with semi- or fully-automatic web service composi-

tion(Meng et al., 2006; Arpinar et al., 2005).

Arpinar et. al. (2005) considered all processes pre-

sented in Section 2.2 with speciﬁc application to a

winery, web service case study. The system architec-

ture of their ODCS utilized three ontologies to drive

the composition: a Domain ontology, a Web Services

ontology, and a Process ontology (Figure 3). The Do-

main ontology was only utilized to represent the data

input and output, while the actual web services were

represented in a general Web Services ontology.

Arpinar et. al. (2005) referenced work by Zhang

(2004) where the deﬁned Web Services ontology

held entity titles like ’Wine-Searcher’ (sub-class of

’Searching Service’) representing instances of web

services utilized to ﬁnd different types of wines. The

Figure 3: System Architecture Diagram of the Ontology-

Driven Web Service Composition Platform (Arpinar et al.,

2005).

Domain ontology listed properties such as wine name,

vintage, winery, etc. which allowed the user to deﬁne

ranges and ﬁlters to the ’Searching Service’.

Finally, the Process Ontology represented se-

quences of services that were connected to one an-

other if their interfaces were semantically matched.

The work by Arpinar et. al. (2005) focused most of

their work on the interface input/output matching be-

tween web services (i.e. compositional units) and the

execution of the ﬁnal resulting system. The quality

of web services was a consideration through a set of

quality criteria adopted from Zeng et. al. (2003), how-

ever Arpinar et. al. (2005) recognized quality as a fea-

ture that had to be further developed and researched

for their system.

Ontology-Driven Web Service Composition has

gained large amounts of popularity as many stake-

holders hold an interest in the growing web services

economy and are willing to commit funds and ef-

fort (Cardoso and Sheth, 2005). However, most of

the work on Ontology-Driven Web Service Composi-

tion is focused heavily on the discovery, matching and

composition of compositional units, and less on their

ranking and selection (decision-making).

3.2 Resource Management Focused

Ontology-driven Compositional

System

Hlomani and Stacey (2009) approached the design of

an Ontology-Driven Compositional System in a dif-

ferent manner. Their compositional system was de-

signed (in theory) to accept compositional units im-

plemented with distributed or non-distributed func-

tionality. Furthermore, the ODCS contained a phys-

ical copy of all non-distributed compositional units

giving the ODCS a ’resource management’ focus.

KEOD 2010 - International Conference on Knowledge Engineering and Ontology Development

136

Figure 4: An Illustration of the Functional Semantics Rep-

resented in the Algorithm Ontology created by Hlomani and

Stacey (2009).

The overall goal of Hlomani and Stacey (2009) was to

create a plug and play hub of compositional units to

allow users to explore the discovery, decision-making

and composition of a variation of resulting systems.

The ODCS for Hlomani and Stacey (2009) uti-

lized two ontologies to drive their compositional sys-

tem: an Algorithm ontology and a Execution Time-

line ontology. Rather than web services, the composi-

tional units in Hlomani and Stacey’s ODCS were al-

gorithms, thus the Algorithm ontology (Figure 4) pro-

vided an extensive standard description of algorithms

(i.e. functional properties and the classiﬁcation of in-

put and output). Similar to the Process ontology in

Arpinar et. al. (2005), the Execution Timeline ontol-

ogy provided semantic descriptions to assist with the

composition of events.

To date, Hlomani and Stacey (2009) is the only

ODCS research outside the web services focus. Sim-

ilar to Arpinar (2005) the ODCS focused on the func-

tional, data and execution semantic knowledge, how-

ever they did note the importance of ranking the qual-

ity and trust of compositional units. One unique

strength noted from their work, was the explicit dis-

tinction between compositional units that were ac-

tual computational components (i.e. Algorithm Com-

ponents) and ”helper” compositional units which as-

sisted to glue together inputs and outputs of computa-

tional units that could otherwise not be connected (i.e.

Architectural Components).

3.3 Overall Assessment of Current

Implementations

After a brief investigation of recent implementa-

tions of an Ontology-Driven Compositional Systems,

the authors identiﬁed important strengths and weak-

nesses:

Strength: Methods of Discovery and Composition

Both Arpinar et. al. (2005) and Hlomani and Stacey

(2009) focused most of their efforts on discovery and

composition. As ODCSs are in their youth, a proto-

type with minimal complexity is optimal. Both OD-

CSs utilize a unique ontology to assist with deﬁn-

ing the combination of compositional units required

for a certain resulting system requested by the user.

The Input/Output matching algorithms have been im-

plemented successfully, however a larger knowledge

base of compositional units, input/output and result-

ing systems would provide opportunities for more

robust matching. Hlomani and Stacey (2009) ac-

cepted this recognition by adding functional seman-

tic knowledge that described ”architectural” composi-

tional units to act as glue to match more input/output.

Weakness: Representation of Domain-Speciﬁc

Functional Semantic Knowledge

The functional semantic knowledge of compositional

units and their input/output has been limited to the

use of a small domain ontology, or a minor inheri-

tance relationship in one ontology. Creating a more

robust functional semantic knowledge representation

of compositional units has been proposed, but not in-

vestigated (Arpinar et al., 2005). Furthermore, as

the number, variability and speciﬁcation of different

types of compositional units increase, more attributes

and descriptors will be necessary to distinguish the

difference between which compositional units are de-

sired by the user.

Weakness: Quality & Trust of Compositional

Units

Currently, only minor efforts have been attempted to

rank and select compositional units in the ”Decision-

Making” process in an ODCS (Hlomani and Stacey,

2009; Arpinar et al., 2005). The ranking of quality

was focused mostly on computational performance-

based metrics and did not consider domain-speciﬁc

and qualitative-like performance metrics. Research

within the ”Semantic Web Services” (Cardoso and

Sheth, 2005) domain has focused on improving rank-

ing of quality (Zeng et al., 2003; Tran, 2008; Wang

et al., 2006), however that research has not yet been

implemented into an ODCS. The semantic represen-

tation of ”Trust” was considered only by Arpinar et.

al. (2005) with a brief mention of reputation, however

many more elements should be considered.

SATISFYING USER EXPECTATIONS IN ONTOLOGY-DRIVEN COMPOSITIONAL SYSTEMS - A Case Study in

Fish Population Modeling

137

3.4 Designing an ODCS from User

Expectations

The weaknesses identiﬁed are focused mostly around

the satisfaction of a user’s expectations of a result-

ing composed system. A user expects to utilize and

match as many compositional units as possible and

be provided with a robust and holistic view of the

compositional units. Furthermore, a user should ex-

pect to rank compositional units based on aspects of

performance and/or non-performance quality metrics

to assess which compositional units he/she or other

users/experts trust.

4 DEFINING USER

EXPECTATIONS FOR AN ODCS

User satisfaction is driven by a user’s initial expec-

tations. Satisfying user expectations of an ODCS is

therefore deﬁned as ”creating a composed system that

satisﬁes the initial expectations of a user to the point

where he/she is able to accept the resulting composed

system”. One aspect of the use of ODCSs is that

users need to specify input/output requirements be-

fore the resulting system is composed. Therefore, a

user’s expectations can be retrieved at this time as

well. This paper introduces three user expectations

to drive the operation of ODCSs: leverage and ac-

quisition of knowledge, acknowledgment of trust and

satisfaction of performance. Obviously, other user ex-

pectations could be considered, but for the context of

this paper these three are examined.

4.1 Leverage & Acquisition of

Knowledge

Users are more satisﬁed with a system when the

amount of their cognitive processing and knowledge

requirements are decreased. (Zhang and von Dran,

2002). ODCSs could be utilized by users who have

varying sets and levels of predetermined knowledge.

For example, one user with comprehensive knowl-

edge of programming and software engineering tech-

niques will understand the dynamics of matching in-

put and output. However, a second user without that

knowledge will need to further leverage the assis-

tance of the ODCS. Users may also not comprehend

domain-speciﬁc aspects of a compositional unit rele-

vant to the resulting system they wish to create. (e.g.

a ﬁsheries biologist may not understand the dynamics

of statistical modeling in the running of population

models). As stated in Section 2.1, compositional sys-

tems must hold different types of semantic knowledge

and the user expects to utilize that network of expert

knowledge in the ODCS.

Using the ﬁve different types of semantic knowl-

edge in compositional systems (Section 2.1), three

facets of expert knowledge are described: compu-

tational system, domain-speciﬁc and task-oriented.

Computational system expert knowledge is the

system-speciﬁc function, data, quality and execution

knowledge in a compositional system. This would be

very useful for a software developer, but not for in-

dividuals without some sort of programming knowl-

edge. The domain-speciﬁc expert knowledge is the

”real world” function, data, quality, and execution

knowledge of the compositional units being used

to make the resulting system. This information is

usually embedded within a program and not explic-

itly differentiated. Finally, the task-oriented expert

knowledge the user expects to leverage is the task and

goal semantic knowledge of the compositional sys-

tem.

Deconstructing user expectations further, as a user

is leveraging different facets of expert knowledge,

s/he will also be consciously or unconsciously ac-

quiring that knowledge. Acquisition of computa-

tional system and/or domain-speciﬁc expert knowl-

edge could be obtained by exploring and experiment-

ing with different selections of compositional units

(e.g. a ﬁsheries biologist could research more com-

puter population models by including different com-

binations in a resulting composed system from a

ODCS to gain more understanding why a given pop-

ulation model is trusted more than another).

4.2 Deconstructing Quality & Trust

As emphasized previously, the end-user of a ODCS

does not only leveraging and aquire knowledge. S/he

also considers the quality of the compositional units

to be a important factor of the resulting system. The

following two user expectations relate directly to the

”Quality & Trust” semantic knowledge presented in

Section 2.1.

4.2.1 Acknowledgment of Trust

An increasing number of software developers and

human-computer interaction specialist acknowledge

that if a user does not trust aspects of software sys-

tems or interfaces, than s/he automatically expects

to be satisﬁed less. Duez et. al. (2006) states, ”If

[users] do not trust the new automated tools, they will

not use them no matter how useful or efﬁcient they

KEOD 2010 - International Conference on Knowledge Engineering and Ontology Development

138

might be”. In an ODCS, a user expects to be pro-

vided with compositional units that are tested and ac-

cepted by various software and domain-speciﬁc ex-

perts. Also, each user is unique and s/he may hold

higher regard for certain software developers or do-

main experts, and respect that those experts have the

knowledge to understand certain compositional units

more clearly. Thus, a higher ranking should be as-

signed to the compositional units that the certain ex-

perts have either programmed or trusted as the user

will be more satisﬁed.

As explained, the knowledge of different types of

experts is leverages, therefore attributes of trust that

a software developer would consider is program cre-

ator, number of times utilized, reputation from other

programming experts (especially security reputation),

minimal number of errors during testing, minimal

number of bugs etc. Whereas, a domain-speciﬁc ex-

pert would trust similar and inherited features, yet

each unique domain would have separate features that

also must be considered (e.g. a population modeling

expert would consider the number of simulation stud-

ies to be an appropriate measure).

4.2.2 Satisfaction of Performance

Ultimately, the resulting system generated from a

compositional system is still a piece of computational

software. Therefore, most users will still consider

performance metrics to be extremely important. Pro-

gramming speciﬁc features like estimated time of ex-

ecution, amount of memory required or, distance of

distributed machines will always be important. Sim-

ple metrics and protocols like QoS can be utilized.

Domain-speciﬁc performance proves slightly more

difﬁcult, as the quality of performance could be mea-

sured by different numerical representations. For ex-

ample, many ﬁsheries management population mod-

els may only produce a successful modeling simula-

tion some of the time due to factors such as data avail-

ability, data quality, or model complexity (Megrey

and Moksness, 2009).

5 FIRST ATTEMPT AT AN ODCS

MOTIVATED BY USER

EXPECTATIONS

After the investigation of current ODCSs in Section 3,

and a deﬁnition of user expectations in Section 4, an

ODCS motivated by user expectations can be inves-

tigated. This paper utilized work conducted by Hlo-

mani and Stacey (2009) with a focus on the discovery

and decision-making processes. Figure 5 illustrates

how the authors have begun to adapt the ontologies

and processes to satisfy the user expectations deﬁned

in the section above. The ﬁgure generalizes how a

domain ontology represents a unique set of semantic

knowledge, yet through ontology merging and inher-

itance all required semantic knowledge can be repre-

sented as a whole.

Currently, the authors are focused on the holis-

tic capture and development of a Compositional Unit

ontology (CU ontology), a Statistical modeling CU

ontology, and a Fish Population modeling CU ontol-

ogy to improve the discovery (e.g. matching) and

decision-making (e.g. ranking and selection) pro-

cesses for the end-user. Section 5.1 describes the

adapted compositional unit ontology previously uti-

lized by Hlomani and Stacey (2009). Section 6 fol-

lows by providing an example of how domain-speciﬁc

ontologies (Statistical modeling CU ontology, and

Fish Population modeling CU) will inherit the CU

ontology (and each other) to fully represent the case-

study domain in a holistic fashion.

Figure 5: Adaptation of Figure 2 to illustrate how this

paper’s ODCS utilizes inheritance and merging of multi-

ple ontologies to satisfy the user expectations of leveraging

knowledge and incorporation of trust.

5.1 CU Ontology

The CU ontology was adapted from Hlomani and

Stacey (2009) by evolving their Algorithm ontology

into what this paper deﬁnes as the Compositional

Unit ontology (CU ontology). The change of name

to ”Compositional Unit” was necessary since the

term ”Algorithm” does not semantically deﬁne all

possible units of composition (see Section 2 for

SATISFYING USER EXPECTATIONS IN ONTOLOGY-DRIVEN COMPOSITIONAL SYSTEMS - A Case Study in

Fish Population Modeling

139

Figure 6: A few of the essential entities in the prototype Compositional Unit Ontology (CU ontology). The CU Ontology is

proposed to be the base system ontology that is utilized along with the Timeline ontology to compose systems. As Figure 5

illustrates, domain-speciﬁc ontologies inherit and merge with the CU ontology to allow the classiﬁcation of speciﬁc domain

knowledge. Section 6 presents a case study of this paradigm.

the deﬁnition of a compositional unit). Similar to

Hlomani and Stacey’s Algorithm ontology, the CU

ontology represents the functional, data, quality &

trust, and execution semantic knowledge of single

compositional units. The Timeline ontology has

not yet been adapted by the authors, but will be

approached in the future. Figure 6 provides an

illustration of some of the entities within the CU

ontology. As the focus of this paper is on satisfying

user expectations in an ODCS, the execution and

timeline semantic knowledge will not be presented

because it is not a high priority concern for the

end-user.

Functional Semantic Knowledge. The functional

semantic knowledge represented in the CU ontology

attempts to provide an explanation of the features,

elements, and compositional purpose of the composi-

tional units. The term ”features” refers to high-level

characteristics of a given instance of a compositional

unit and the ”elements” refers to functional charac-

teristics within an instance of a compositional unit.

Finally, ”compositional purpose” refers to the reason

why compositional units are presented. Similar to

Hlomani and Stacey (2009), most compositional

units are ”computational” components that provide

services described by the ”features” and ”elements”,

however some act like glue and are semantically

represented as ”architectural”.

Data Semantic Knowledge. Most of the data

semantic knowledge is utilized during the matching

of input/output (i.e. discovery) of the compositional

units for a resulting system. Both the input/output

could have deﬁned format, structure, and/or type;

format refers to string-like protocols that are followed

(i.e. regular expressions, or string date formats), and

structure refers to the method by which the data is

represented (e.g. matrix, vector, etc.). Furthermore,

input/output of the compositional units will be

received/sent via a given data stream which could be

local or remote.

Quality & Trust Semantic Knowledge. The qual-

ity & trust semantic knowledge in the CU ontology

focuses on characteristics that provide the ability to

rank and select the compositional unit. Informa-

tion like positive and negative feedback could dictate

whether an end-user would select the given composi-

tional unit. Alternatively, an end-user may not trust a

certain expert or developer, therefore any feedback or

tests by that expert or developer should be ignored in

a ranking process.

6 CASE STUDY: FISHERIES

POPULATION MODELING

Fisheries managers understand that the anthropogenic

effects of harvesting must be studied in a transpar-

KEOD 2010 - International Conference on Knowledge Engineering and Ontology Development

140

Figure 7: A case study of ﬁsheries population modeling: an example of how multiple ontologies could inherit the CU ontology

and further describe the functional, data, and quality & trust of the compositional units within their own domain.

ent, accountable and scientiﬁcally-defensible man-

ner (Walters and Martell, 2004). Fisheries biologists

are speciﬁcally responsible for collecting and analyz-

ing data to estimate ﬁsh population abundance, as-

sess human and non-human sources of mortality, and

to ensure sustainable harvest levels (Quinn and De-

riso, 1999). Unfortunately, many ﬁsheries manage-

ment decisions are based largely on qualitative indica-

tors or expert opinion of ﬁsh population status, rather

than estimates derived from more complex mathemat-

ical modeling tools developed over the past 30 years

(NRC, 2005; Methot, 2009). In some cases, the biol-

ogists are constrained by the quantity and/or quality

of data required for the mathematical models, while

in other cases the biologists are themselves limited

in their understanding of how these models function

(Stringer et al., 2009).

Through a unique ﬁsheries research collaboration

between the Chippewas of Nawash Unceded First Na-

tion and the University of Guelph, the authors have

investigated the domain of ﬁsheries population mod-

els to understand how ﬁsheries biologists could: [1]

leverage and acquire the computational, mathemati-

cal, and statistical knowledge to appropriately employ

the population models, and [2] enhance their level of

trust in the quality and performance of the generated

system. The following two sub-sections present how

the authors ODCS could be utilized to achieve the

stated objectives. The ﬁrst sub-section introduces an

example of how ﬁsheries managers could leverage a

network of expert knowledge through the inheritance

and merging of the multiple domain ontologies. The

second sub-section provides a speciﬁc focus on qual-

ity and trust semantic knowledge and how it has been

considered through the multiple domain ontologies as

well.

6.1 Leveraging Knowledge for Fish

Population Models

Fisheries population models are mathemati-

cal/statistical models implemented as computer

algorithms speciﬁcally to focus our understanding

on important factors that determine population

abundance and condition (Hilborn and Walters,

1992). Therefore, two levels of domain knowledge

need to be considered past the semantic knowledge

represented in the Algorithm Ontology. As shown in

Figure 5, the ﬁrst domain of knowledge considered is

Statistical Modeling, followed by the domain of Fish

Population Modeling itself. Considering functional

semantic knowledge in particular, the Statistical

Modeling ontology describes high level functional

features such as spatial/temporal dimensions or

stochastic/deterministic processes and similar func-

tional features of data, such as whether or not given

parameters are sampled using normal, log-normal or

uniform distributions. The Fish Population Modeling

ontology would hold functional features such as mor-

tality and/or recruitment implementations utilized,

and functional features of data such as the functional

purpose of its input (i.e. mortality, recruitment,

etc.). Figure Figure 7 provides an example of how

the two domain ontologies in this case study inherit

knowledge and merge with the CU ontology.

6.2 Incorporating Quality & Trust

Within each domain presented in this case study, dif-

ferent types of metrics need to be considered to mea-

sure and rank aspects of quality and trust speciﬁcally.

As shown in Figure 7, statisticians would trust a given

population model CU more if simulation studies had

SATISFYING USER EXPECTATIONS IN ONTOLOGY-DRIVEN COMPOSITIONAL SYSTEMS - A Case Study in

Fish Population Modeling

141

been conducted and ﬁsheries managers may wish to

trust population models that have gotten feedback

from expert population modelers whom they trust.

More complex metrics of trusting quality and perfor-

mance are also included, however, these are not con-

sidered in the examples provided here.

7 DISCUSSION OF FUTURE

WORK

There are many gaps that must be investigated in

ontology-driven compositional systems. Currently

two facets of future work are being considered: fur-

ther development of domain-speciﬁc ontologies and

investigation of evaluation techniques.

7.1 Further Domain-speciﬁc Ontology

Capture

Aside from ongoing adaptations of the ODCS of

Hlomani and Stacey (2009) to the CU ontology,

the domain-speciﬁc ontologies (Statistical modeling

CU and the Fish Population modeling CU) still re-

quire further development. The authors have orga-

nized a series of three focus group sessions with

the Integrative Biology Department and the Saugeen-

Ojibway/Nawash First Nations ﬁsheries managers to

approach this. The group sessions will continue to

focus on aspects of functionality, data, and quality &

trust.

7.2 Investigation of Evaluation

Techniques

Because ODCS research is still very new, no attempts

at deﬁning a ”golden standard” have been made.

Like similar research in ontology capture, engineer-

ing and validation it is very difﬁcult to argue whether

or not a given ontology is properly representing the

given knowledge domain (Uschold and King, 1995;

Gangemi et al., 2005). This argument is dynamically

dependent on the ”eye of the beholder”, as separate

individuals will have different opinions on whether or

not a given ontology represents the state of knowl-

edge. In the future, the authors will investigate how

a ”golden standard” would be deﬁned for ontologies

within an ODCS with a speciﬁc focus on classifying

how well the ontologies represent the different tenets

of semantic knowledge described in Section 2.1. This

”golden standard” would then have to be adapted to

facilitate the various domain-speciﬁc ontologies, as a

certain domain of semantic knowledge may require

different standards.

Expanding further, the beneﬁt of this paper’s re-

search is the overall goal: to satisfy user expectations

of a computer assisted system composition process

within an ODCS. Therefore, the authors will run user

interviews with domain users of varying levels/sets of

knowledge (i.e. statisticians, population modellers,

software developers, ﬁsheries biologist, etc.) to in-

vestigate how well that goal is satisﬁed.

ACKNOWLEDGEMENTS

The authors would like to acknowledge all of the col-

laboration with the Integrative Biology Department

at the University of Guelph and the Chippewas of

Nawash Unceded First Nation, speciﬁcally Chief &

Council, Scott Lee, Ryan Lauzan, Dan Gillis, and

Jasper Tey. Mitchell Gillespie would like to acknowl-

edge the patience, kindness, assistance and opportu-

nities that Dr. Deborah Stacey has provided during

his work as a graduate student. Last, but not least,

he acknowledges all of the support he receives from

his partner (Kathryn Marsilio), and his family (Katie,

Tom, and Mugz). Deborah Stacey wishes all of her

graduate students were so dynamic and productive.

REFERENCES

Arpinar, I. B., Zhang, R., Alemen-Meza, B., and Maduko,

A. (2005). Ontology-driven web services composition

platform. Information Systems and e-Buisness Man-

agement, 3:175–199.

Cardoso, J. and Sheth, A. (2005). Introduction to seman-

tic web services and web process composition. In

First International Workshop on Semantic Web Ser-

vices and Web Process Composition, Lecture Notes in

Computer Science, pages 1–13. Spinger.

Crawford, S., Gillis, D., and Rooney, N. (2008). A review

of population level ecological risk assessments for the

candu owners group. Technical report, CANDU Own-

ers Group, Toronto, ON, Canada.

Crawford, S., Muir, A., and McCann, K. (2001). Ecological

basis for recommendation of 2001 saugeen ojibway

commercial harvest tacs for lake whiteﬁsh (coregonus

clupeaformis) in lake huron, report prepared for the

chippewas of nawash ﬁrst nation. Technical report,

University of Guelph, Wiarton, ON, Canada. (revised

with references 11 July 2002; revised with response to

OMNR comments 02 January 2003).

Duez, P. P., Zuliani, M. J., and Jameison, G. A. (2006). Trust

by design: Information requirements for appropriate

trust in automation. Technical report, IBM Canada

Ltd.

KEOD 2010 - International Conference on Knowledge Engineering and Ontology Development

142

Feller, J. and Fitzgerald, B. (2002). Understanding Open

Source Software Development. Addison-Wesley, Lon-

don.

Gangemi, A., Catenacci, C., Ciaramita, M., and Lehmann,

J. (2005). A theoretical framework for ontology eval-

uation and validation.

Gillis, D., Tey, J., Gillespie, M., and Crawford, S. (2009).

Do ﬁsheries biologists have appropriate tools for as-

sessing dynamics of harvested ﬁsh populations? Not

yet submitted to Journal.

Gruber, T. R. (1993). A translation approach to portable on-

tology speciﬁcations. Knowledge Acquisition, 5:199–

220.

Hilborn, R. and Walters, C. J. (1992). Quantitative ﬁsheries

stock assessment: choice, dynamics and uncertainty.

Chapman and Hall, New York.

Hlomani, H. (2009). A bottom-up approach to system com-

position using ontologies. Master’s thesis, University

of Guelph, Guelph, ON, Canada.

Hlomani, H. and Stacey, D. (2009). An ontology driven ap-

proach to software systems composition. In Proceed-

ings of the 2009 International Conference of Knowl-

edge Engineering and Ontology Development. IN-

STICC.

Majithia, S., Ali, A. S., Rana, O. F., and Walker, D. W.

(2004). Reputation-based semantic service discov-

ery. Enabling Technologies, IEEE International Work-

shops on, 0:297–302.

Megrey, B. A. and Moksness, E. (2009). Past, present, and

future trends in the use of computers in ﬁsheries re-

search, pages 1–30. Computers in Fisheries Research,

2nd Edition. Springer Science.

Meng, X., Junliang, C., Yong, P., Xiang, M., and Chuan-

chang, L. (2006). A dynamic semantic association-

based web service composition. In Proceedings of

the 2006 IEEE/WIC/ACM International Conference

on Web Intelligence. IEEE.

Methot, R. D. J. (2009). Stock assessment: operational

models in support of ﬁsheries management, pages

137–165. Fish & Fisheries Series 31. Springer Sci-

ence, Netherlands.

NRC (2005). Improving ﬁsh stock assessments.

Quinn, T. J. and Deriso, R. (1999). Quantitative ﬁsh dynam-

ics. Ofxord University Press, New York.

Srivastava, B. and Koehler, J. (2003). Web service composi-

tion – current solutions and open problems. In ICAPS

2003 Workshop on Planning for Web Services, Trento,

Italy.

Stringer, K., Clemens, M., and Rivard, D. (2009). The

changing nature of ﬁsheries management and impli-

cations for science, pages 97–111. Fish & Fisheries

Series 31. Springer Science, Netherlands.

Tran, V. X. (2008). Wsqosonto: A qos ontology for web

services. In 2008 IEEE International Symposium on

Service Oriented System Engineering, pages 233–238.

IEEE.

Uschold, M. and King, M. (1995). Towards a methodol-

ogy for building ontologies. In In Workshop on Ba-

sic Ontological Issues in Knowledge Sharing, held in

conjunction with IJCAI-95.

Walters, C. J. and Martell, S. J. D. (2004). Fisheries Ecol-

ogy and Management. Princeton University Press,

New Jersey.

Wang, X., Vitvar, T., Kerrigan, M., and Toma, I. (2006).

A QoS-Aware Selection Model for Semantic Web Ser-

vices, pages 390–401. Lecture Notes in Computer Sci-

ence. Springer, Berlin, Germany.

Wang, Y. and Vassileva, J. (2007). Toward trust and rep-

utation based web service selection: A survey. In

Proc. Intl. Transactions on Systems Science and Ap-

plications, volume 3, pages 118–132.

Zeng, L., Benatallah, B., Dumas, M., Kalagnanam, J., and

Sheng, Q. Z. (2003). Quality driven web services

composition. In Proceedings of the 12th International

conference on the World Wide Web, pages 411–421.

ACM.

Zhang, P. and von Dran, G. M. (2002). User expectations

and rankings of quality factors in different web site

domains. International Journal of Electronic Com-

merce, 6(2):9–33.

SATISFYING USER EXPECTATIONS IN ONTOLOGY-DRIVEN COMPOSITIONAL SYSTEMS - A Case Study in

Fish Population Modeling

143