TOWARDS A GENERIC AND CONFIGURABLE MODEL OF AN

ELECTRONIC INFORMER TO ASSIST THE EVALUATION OF

AGENT-BASED INTERACTIVE SYSTEMS

Chi Dung Tran, Houcine Ezzedine and Christophe Kolsky

LAMIH – UMR CNRS 8530, University of Valenciennes and Hainaut-Cambrésis

Le mont Houy, 59313 Valenciennes cedex 9, France

Keywords: Human-computer interaction, supervision, interactive system architecture, evaluation of interactive systems,

agent-based interactive systems, electronic informer.

Abstract: This paper presents a generic and configurable model of an electronic informer to assist the evaluation for

agent-based interactive systems. In order to propose this model, the current state of the art concerning

architectures used for traditional interactive systems is presented, along with that for agent-based interactive

systems. In this article, we propose an agent-oriented architecture that is considered as being mixed (it is

both functional and structural). By using this architecture as a basis, we propose a generic and configurable

model of an evaluation tool called “electronic informer” which assist evaluators in analysing and evaluating

interactive systems of this architecture.

1 INTRODUCTION

Nowadays, in spite of the existence of several

methodologies for the development of interactive

systems, designing, developing and assessing, in

terms of utility and usability (Bastien and Scapin,

1995; Nielsen, 1993; Shneiderman, 1998) an agents-

based interactive applicative system is still a difficult

task. It is therefore necessary to provide methods,

models and evaluation tools to make it easier. A lot

of research work has dealt with the specification and

design of object-oriented interactive systems as well

as agent-based interactive systems. However, the

choice of an architecture which is adequate for the

system is not on easy task. Furthermore, in the

majority of cases, the current evaluation methods do

not take into account the specific architecture of an

agent-based interactive system (Trabelsi et al., 2004)

and there are rarely propositions concerning the

coupling between the architecture and evaluation

phase (Trabelsi, 2006).

The section 2 presents a brief state of the art

concerning the architectures for traditional

interactive systems as well as for agent-based

interactive systems. The section 3 proposes an

architecture which is both functional and structural,

and which provides a separation into three functional

components (Ezzedine et al., 2003). By using this

architecture as a basis, in the section 4, we propose a

generic and configurable model of an evaluation tool

called “electronic informer”; its aims at assisting the

evaluation of agent-based architecture systems.

2 INTERACTIVE SYSTEM

ARCHITECTURES

Architecture of an interactive system supplies the

designer with a generic structure from which he/she

can build an interactive application. It is a set of

structures that include: components, the outside

visible properties of these components and the

relations between them. Researchers have proposed

several architecture models over the past twenty

years. Two main types of architecture can be singled

out: functional (Seeheim, Arch) and structural

models (PAC proposed by J. Coutaz, and its

variations, such as PAC Amadeus, MVC and its

variations, such as MVC2). The functional models

split an interactive system into several functional

components. For ex., the Seeheim model is made up

of 3 logical components (Presentation, Dialogue

Controller, Application Interface); the Arch model

defines a functional breakdown of an interactive

system into 5 components in which: both the

presentation and interaction components are a

290

Dung Tran C., Ezzedine H. and Kolsky C. (2007).

TOWARDS A GENERIC AND CONFIGURABLE MODEL OF AN ELECTRONIC INFORMER TO ASSIST THE EVALUATION OF AGENT-BASED

INTERACTIVE SYSTEMS.

In Proceedings of the Ninth International Conference on Enterprise Information Systems - HCI, pages 290-293

DOI: 10.5220/0002377702900293

 SciTePress

decomposition of the presentation of the Seeheim

model, the functional kernel component, the domain

adapter component and the dialogue controller

component. Seeheim and Arch provide canonical

functional structures with big grain; they are useful

as a structural framework for a design or a rough

analysis of the functional decomposition of an

interactive system (Trabelsi et al., 2004). These

decompositions are generally not enough to complex

applications; the functionalities are mixed in the too

macroscopic components (Tarpin-Bernard and

David 1999). The structural models aim at a finer

breakdown by using structural components, and in

particular those said to be distributed or agent

approaches, suggest grouping the functions together

into one unit, the agent. The agents of this type of

architecture are then organized in a hierarchical

manner according to principles of composition or

communication. For example, a MVC agent is made

up of three facets: Model, View and the Controller.

The PAC model defines an agent using three facets:

the Presentation, the Abstraction and the Control.

These architecture models recommend the same

principles, based on the separation between the

functional core of system (application) and the

human-machine interface. The architecture therefore

has to define a distribution of the interface services

and to define an exchange protocol (Hilbert and

Redmiles, 2000). This separation makes

modifications easier; it allows modifying the

interfaces without affecting the application. In the

next part, we present an agent oriented architecture.

3 AN AGENT ORIENTED

ARCHITECTURE

Our approach is intended to be mixed as its

principles borrow from both types of model; it is

both functional and structural. In our architecture

(Grislin-Le Strugeon et al., 2001; Ezzedine et al.,

2003), we suggest using a division into 3 functional

components recommended in the Seeheim model

which we have called respectively: interface with the

application (connected to the application), dialogue

controller, and interface or presentation (this

component is directly linked to the user). Each of

these components can be broken down further in a

structural approach in the form of agents. These

components are built like three multi-agent systems

and they are considered as working in parallel, at

least, at a theoretical point of view.

Figure 1: Agent oriented architecture.

The application agents, manipulating the field

concepts of the application, cannot be directly

accessed by the user. One of their roles is to ensure

the real time transmission of the information

necessary for the other agents to perform their task.

The interface agents are in direct contact with the

user (they can be seen by the user). These agents co-

ordinate between themselves in order to intercept the

user commands and to form a presentation which

allows the user to gain an overall understanding of

the current state of the application. The control

agents in the Dialogue Controller component

provide services for both the application and the

interface agents in order to guarantee coherency in

the exchanges emanating from the application

towards the user, and vice versa. Their role, in

particular, is to link the two other components

together by distributing the user commands to the

application agents concerned, and by distributing the

application feedback towards the interface agents

concerned. All these agents communicate amongst

themselves in order to answer the user actions. This

communication can be considered as being services

among agents and can be represented using high

level Petri Nets (PN) (Ezzedine et al., 2006).

4 PROPOSITION

An "electronic informer" is a software tool that

ensures the automatic collection, in a real situation,

of users’ actions and their repercussions on the

system. The collection of information is done in a

discreet and transparent way for the user, who must

not at any time feel hampered by the presence of the

informer. This is an advantage of such a tool.

Objective data collected through interactions can be

processed, analyzed and shown in a synthetic shape

to the evaluator. This facilitates the analysis of the

results. The reader can find in (Hilbert and Redmile,

2000) state of the art concerning tools of this type.

We propose a generic and configurable model of

an “electronic informer” which assists the evaluator

in analysing and evaluating agent-oriented

interactive systems. Such tool takes into account the

specific architecture of these systems. It also

proposes explicitly a coupling between the

TOWARDS A GENERIC AND CONFIGURABLE MODEL OF AN ELECTRONIC INFORMER TO ASSIST THE

EVALUATION OF AGENT-BASED INTERACTIVE SYSTEMS

291

architecture and evaluation phase. At the present

time, the first version of an electronic informer has

been studied and developed (Trabelsi, 2006).

However, it is not a generic tool but only a specific

tool to evaluate a specific agent-oriented applicative

system that is intended to supervise the passenger

information on a public transport system. It cannot

be used to evaluate other agent-oriented systems

because it depends on the number of agents, the

structure and the contents of such systems.

Furthermore, it shows some inconveniences and

shortcomings. We solve such problems by a generic

and configurable model of an “electronic informer”.

It is made up of 7 main modules (Fig. 2).

Module 1 (M1): collecting events in user

interface and service level from all agents and users

of the concerned interactive system.

M2: associating events in intermediate level

(user interface and service events) with each

application task. Several events in intermediate level

can be realized to obtain a certain application task.

For ex., 3 user interface events TabDriver_click,

TextBoxMessage_OnChange and buttonOK_Click

and 2 services with the same name “Send a message

to the driver” of the agent interface Vehicule and of

the agent application Vehicule associated with the

application task “Send a message to the driver” of a

system intended to supervise the passenger

information on a public transport system.

M3: processing collected data of a chosen agent

in a certain period of time and showing results in

comprehensible forms. Here are examples of

calculations and statistics: response time for

interactions between services; time for a certain user

interface event (time for loading an interface agent

or for typing an text box…); time for completing a

service and furthermore, an application task; time

for consulting help or unproductive time (help time

+ snag time + search time (Bevan and Macleod,

1994; Chang and Dillon, 2006)) that user takes to

complete a certain application task, the percentage

of services accomplished and furthermore, of

application tasks accomplished, the error’s

percentage, the help’s use frequency, the percentage

of services and furthermore, of application tasks

achieved per unit of time, the ration of failure or

success for each interaction between services, the

ration of appearance of each user interface event of a

certain interface agent, the percentage of use for

each service of a certain agent, the average number

of user interface events per unit of time, and so on.

M4: generating the Petri Nets (PN) to describe

activity process of agents and users in the system

from collected data and BSA (Specification Base of

Agents). Indeed, it describes process of interactions

between services of different agents as well as

process of activity of user to complete application

tasks. We call them “observed” PN. Generating PN

facilitates evaluators because it provides them with

the visual views of all the activities of the user and

the concerned system.

M5: comparing observed PN created above with

the PN that system designer has intended before to

complete application tasks. This comparison assists

the evaluators in detecting use errors; for ex., the

evaluator can perceive that the user has passed

redundant state, has realized useless manipulations

or takes more time than the one predicted by

designer to complete an application task. M5 can

also be used to assist the evaluator in comparing the

ability of different users to use a system.

M6: using results of processes from M3, the PN

generated by M4, the comparison of two PN from

M5 and usability characteristics as well as

ergonomic criteria as a basis, M6 is responsible of

assisting the evaluator in criticising concerned

system and advising the designer to improve it.

Although the term “usability” has not been defined

homogeneously, it exists several definitions (Dix et

al., 1993; Nielsen, 1993; ISO/IEC 9126-1); in

general, it refers to a set of multiple concepts, such

as execution time, performance, user satisfaction and

ease of learning (“learnability”), effectiveness,

efficiency, taken together (Abran et al., 2003). There

are also several sources from different authors and

organisations. M6 assists the evaluator in evaluating

concerned system on the basis of criteria from

several different sources such as the ergonomic

criteria of (Bastien and Scapin, 1995), the quality

attributes of (Lee and Hwang, 2004) and the

characteristics of the consolidated usability model of

(Abran et al., 2003). The results of processes

(calculations and statistics) from M3 provide the

necessary measures for the evaluation of these

ergonomic criteria, quality attributes and

characteristics. M6 is not yet realized.

M7: configuring electronic members to evaluate

different agent-oriented systems. It allows entering

the BSA (Specification Base of Agents) that

describes the evaluated system, the PN that system

designer has intended and some configuration

parameters of evaluated system.

5 CONCLUSION - PERSPECTIVE

We have presented a brief state of the art concerning

interactive system architectures, and proposed a

mixed architecture as well as a generic and

configurable model for assisting the evaluation for

ICEIS 2007 - International Conference on Enterprise Information Systems

292

Figure 2: A Generic and Configurable Model.

agent-oriented interactive systems. We intend to

combine this “electronic informer” method with

other methods (questionnaire, interview…). It needs

to combine data collected from “the electronic

informer” and data collected from the other methods

to evaluate more efficiently such systems.

ACKNOWLEDGEMENTS

The present research work has been partially

supported by the “Ministère de l'Education

Nationale, de la Recherche et de la Technologie»,

the «Région Nord Pas-de-Calais», the FEDER

(MIAOU, EUCUE, SART), the ANR ADEME

(Viatic.Mobilité), the PREDIM (MouverPerso).

REFERENCES

Abran, A., Khelifi, A., Suryn, W., Seffah, A., 23–25April

2003. “Consolidating the ISO Usability Models”. In:

Proc. of 11th Int. Software Quality Management

Conference, Springer, Glasgow, Scotland, UK.

Bastien J.M.C., Scapin D.L., 1995. Evaluating a user

interface with ergonomic criteria, In: Int. Journal of

Human-Computer Interaction, vol. 7, p. 105-121.

Bevan, N., Macleod, M., 1994. Usability measurement in

context. In: Behav.Inf. Technol., vol. 13, no. 1/2, pp.

132–145.

Chang, E., Dillon, T., March 2006. A Usability-Evaluation

Metric Based on a Soft-Computing Approach. In:

IEEE Trans. On SMC—Part A: Systems and Humans,

Vol. 36, N. 2.

Ezzedine, H., Trabelsi, A., Kolski, C., 2003. Modelling of

Agent oriented Interaction using Petri Nets,

application to HMI design for transport system

supervision. In: Proceedings CESA, Lille, France.

Ezzedine H., Trabelsi A., Kolski C., 2006. Modelling of

Agent oriented Interaction using Petri Nets,

application to HMI design for transport system

supervision. In: Mathematics and computers in

simulation, 70, pp. 358-376.

Grislin-Le Strugeon, E., Adam, E., Kolski, C., 2001.

Agents intelligents en interaction homme-machine

dans les systèmes d'information. In: Kolski C. (Ed.),

Environnements évolués et évaluation de l’IHM, pp.

207-248. Paris: Éditions Hermes.

Hilbert D.M., Redmiles D.F., 2000. Extracting usability

information from user interface events. In: ACM

Computing Surveys, 32 (4), pp. 384-421.

Lee, S.K., Hwang, C.S., 2004. Architecture modelling and

evaluation for design of agent-based system. In: The

Journal of Systems and Softwares, 72, 195-208

Nielsen, J., 1993. Usability Engineering, Academic Press,

Boston.

Shneiderman, B., 1998. Designing the user interface :

strategies for effective human-computer interaction.

Addison-Wesley.

Tarpin-Bernard, F., David, B., Mai 1999. AMF : un

modèle d'architecture multi-agents multi-facettes, In:

TSI, Vol. 18, No. 5, 555-586.

Trabelsi, A., Ezzedine, H., Kolski,. C., October 2004.

Architecture modelling and evaluation of agent-based

interactive systems. In: Proc. IEEE SMC 2004, The

Hague, pp. 5159-5164.

Trabelsi, A., 2006. Contribution à l’évaluation des

systèmes interactifs orientés agents. Application à un

poste de transport urbain. Ph.D. Thesis, University of

Valenciennes, France.

TOWARDS A GENERIC AND CONFIGURABLE MODEL OF AN ELECTRONIC INFORMER TO ASSIST THE

EVALUATION OF AGENT-BASED INTERACTIVE SYSTEMS

293