2Faculty ofScience ofSfax, Data processing and Communications Department, BP 802 - 3018Sfax, Tunisia
3LAMIH - UMR CNRS 8530, University of Valenciennes and Hainaut-Cambresis, Le Mont Houy,F-59313 Valenciennes cedex 9, France, [email protected]
AbstractIn this paper we propose an approach aiming tointegrate Human-Computer Interaction (HCI) aspectsin decision support system (DSS) development. Wepropose an approach combining two methods: oneissued from software engineering field (the RationalUnified Process) and the other one from the HCI field(the U model). We have tested our approach in a DSSset up in the healthcare domain: the supervision ofnosocomial infections in an intensive care unit.Key words: Decision Support System, HumanComputer Interaction, Software Engineering, UnifiedProcess, U model.
1. IntroductionThis paper comes within the scope of researches on
Decision Support Systems (DSS) development. Theevolution of data processing, the nature and theincreasing complexity of the tackled problems havebrought out new social, technical and political realities.In this context, the companies are adopting more andmore technological solutions in their decision processstrategy. The decision-making allows choosingbetween various possible alternatives to solve aproblem by selecting the best solution or compromise.The DSS deals with the problem according to thehuman knowledge and aims to help users in the DSSprocess from beginning to end. So the HumanComputer cooperation is essential throughout thedecision process. It consists in sharing tasks betweenthe human and the computer and defining "Who doeswhat?". In this way, the Human-Computer Interaction(HCI) is a crucial aspect for the interactive decisionsupport systems and its design leans on a user-centredapproach. The evaluation ofHCls allows validating thequality, the usefulness and the usability of the supportsystems. In this scope we propose to combine andintegrate methods coming from Software Engineering
(SE) on one hand with methods coming from HCI onthe other hand, in order to accurately design a DSS. SElifecycles aim to improve reliability, evolutioncapacity, re-usability and portability of software.However these models are often limited when thestudied system is highly interactive because they do notimplicate explicitly and systematically the user [10].For several years, several researchers in the HCI fieldhave worked on the enrichment of software lifecyclesby considering the human participation [10]. It seemsalso possible to propose adapted DSS developmentapproaches combining both fields: SE and HCI.
The second section starts with a brief state of the artabout interactive DSS. Then in the third one we presentan overview of the most common developmentprocesses, in the field of SE as well as in HCI. In thefourth section we present our approach that takesadvantage from the contribution of each field (SE andHCI) for DSS development. The last section lays out anapplication of our approach in the healthcare domain inwhich Intensive Care Unit (ICU) physicians have tosupervise nosocomial infections.
2. Interactive Decision Support SystemsSimon [22] stipulates three phases in the decision
making process: (1) Information research: theidentification of the problem requires looking for therelevant information according to the user's needs. Thisphase leads to expressing the problem to be solved; (2)Design: i.e. creation, analysis and development ofpotential solutions; and (3) Choice: the user has todecide among the different solutions identified. Thisphase includes the research, evaluation andrecommendation of the appropriate solution(s).
We consider the interactive DSS like systems which,through a human-computer dialogue, one or severalusers have to identify, study and solve problems. Theconcept of interactivity in a DSS returns to the essential
895
part of the human in its operation, non-passive partwhich underlies the term "Decision-making support",but also to the quality of the integration of the variouscomponents of the system and to the nature of the HCIplaying the part of the decision maker's assistant.
It is a question of designing a cooperative systemallowing an evolutionary distribution of skills betweenthe user and the computer (dependent on the problemto be solved) and offering a good integration of the twotypes of actors (human being, computer) in thedecision-making process [11].
3. Development approachesFinding the appropriate DSS development process
is the main topic of this paper. For this reason, a criticalrecall of the most traditional development modelsproves to be necessary (we mainly focus our study onthe available models in SE and HCI, the DSS fieldbeing poor on this subject) [14].
3.1. Development models provided by the SEThe software development models allow indicating
the logical or temporal order in which the stages of thesoftware production occur. Several models (andvariants) are available.
From the SE field, we can talk about sometraditional and recent development cycles such as thewaterfall model [20], the V model [17], the Spiralmodel [4], or also the Unified Process (UP) [7, 9].These models aim at the production of quality systems.But the most traditional models are too often directedtowards the technical part and not towards the user,except for UP, until a certain level. For this reason, wefocus on the Unified Process (UP) which is based on aspecific methodological process. It is (1) based on theuse cases representing the functional needs of thesystem, (2) architecture centred, which provides thestructure that will be used as framework to the workcarried out during iterations, (3) iterative andincremental; it reduces the complexity by controlling it.A project can be also seen as split up into underprojects which each of them represents iteration. Theiterations indicate stages of the activities sequence,while the increments correspond to stages of theproduct development.
The aspects of modelling in analysis and design arebased on Unified Modelling Language (UML) [12,19]. UP includes 4 phases: (1) the initialization todefine the extent of the project and the feasibility study;(2) the development to define the needs and specifyarchitecture; (3) the construction during which thesoftware is built by means of several iterations andvarious versions of the system; (4) the transition todeliver the system to the end-users with the setting into
service and to train and support them [7, 12]. However,Lemieux and Desmarais showed many limits of UPconcerning the taking into account of the end-users[13]. Considering the development of an interactivesystem, the characteristics of the user must be stronglytaken into account. Insofar as the developmentprocesses from SE are considered to be insufficient forthe taking into account the user, there was animprovement of traditional models under the HeIangle, which is the subject of the following section.
3.2. HCI-enriched development modelsThe improved models under HCI consideration add
concepts missing in the SE models [10] such as thehuman tasks modelling, the iterative development bybuilding up prototypes and the interactive systemevaluation. Among these models, we can quote the starmodel [6], the improved V model [3], or also the Umodel [2,14,19]. These improved models showevolutions brought by the HCI to the SE. Nevertheless,none of these HCI-enriched models is perfect. Some ofthem present insufficiencies, such as the iterativedevelopment which can remain limited. As a result, wecan note that no model makes equivalence between theadvantages of the SE models and the HCI models.Moreover none of these improved models is directlylinked to DSS. So, we suggest integrating at the sametime SE and HCI dimensions. Under this angle, wehave chosen the UP (SE model) in the precedentsection. It is time now to choose a HCI-model. In thiscontext we opt for the U model because it allowsplacing fundamental stages necessary to the design andthe evaluation of interactive systems. It is a modelwhich can be adapted according to the applicationcharacteristics and which already showed itsapplicability in various complex fields (air control,chemical, railway applications ... ) [10]. The U model isstructured in two phases: (1) a descending phase ofinteractive system specification and design which leadsto implementation; (2) an ascending phase made up ofthe system evaluation, according to effectiveness andergonomic criteria. The validation consists incomparing the model of the prescribed (theoretical)tasks, obtained in the descending phase, with the realtasks discovered in the ascending phase (according tothe original principles suggested by [1]). The result ofthis confrontation permits either to validate theinteractive system or to demonstrate its deficienciesand then to refine it gradually. The final modelresulting from the assessment allows the users' specificbehaviors to be generalized under particular workconditions, and can be reused in situations dealing withsimilar systems [14]. The U model can be adaptedaccording to the application characteristics; e.g., in
896
[14], it was adapted for particular types of DSS inprojects where the elicitation of the experts' expertiseis central. This elicitation aims to design correspondingsoftware components aiming to assist the expertactivities. This model is centred on HCI aspects;nevertheless, it does not present clearly the iterativeand incremental development of the interactive systemto be realized.
4. Proposed Approach: UPIUThe interactive system design as it is done in HCI
and the processes from SE are generally carried outseparately. And when collaboration exists, it isessential to study the points of anchoring privilegedbetween the practices of the HCI and those of the SE[5]. So we propose the participation and the linking oftwo models of both fields (HCI and SE): the U modeland the Unified Process (UP), two models which arecomplementary according to us. In fact, we presentthree points of complementarities between UP and theU model: (1) the UP is generic (as explained above).However, the U model is specific for interactiveapplications, it is enriched from the HCI point of view;(2) UP is iterative and incremental, based on the usecases, and thus the users needs, centred on architecturebut not on the user. Nevertheless, in the U model, thehuman factors are taken into account by thedevelopment team; it is user centered (theconfrontation between theoretical and real tasks for thevalidation and the refining of the system); and finally(3) the design activities of the UP can be enriched bythe design activities of the U model under the HCIangle. So, the development approach we propose mustallow: on one hand to take into account the user(contribution of the U model) and on the other hand tofocus on the prototyping, the explicit positioning of theactors' activities in the development process and theanalysis of the activity (contribution of the UP).
For the DSS modeling, we use UML [12, 21]. Infact, UP is based on UML [7]. In addition it is alanguage that allows models to be represented withoutdefining their development processes. So it can be usedtransparently with any other software developmentprocess, allowing using this language with the U modelthat does not define an obligatory modeling language.
4.1. Development ApproachFor highly interactive DSS, we have adopted the U
model, which needs to be adapted to the UML contextto be used for DSS development [15].
Adaptation of the U model. Why shall wemodifying the U model? The U model splits up intoseveral stages which clearly show the integration of theHCI in the development process. Nevertheless, this
model can be adapted in our context. Thus, when theexisting system is studied, it is important that thedecision maker (potential user) describes his/herfunctional needs and evaluate and validate the first HCIversions (mock-ups, prototypes) in order to show theway he/she wants the future application interfaces tobe. All this information can be used "to model" thedecision maker (characteristics, preferences,strategies ...) [20]. Moreover, for the DSS, thedefinition and the allocation of the tasks are veryimportant and must be well defined. The initial Umodel [2,14,18] does not clearly present the order ofthe design activities (needs assessment, analysis,design ...). We think that such a precision is necessaryto use the U model with UP. The stages of U model aregenerally interested in the specification and theevaluation of the HCI but do not mention much aboutthe applicative side of the interactive system, whereasthis aspect is significant for the DSS. The adapted Umodel is presented in figure 1. Its description is asfollows.
Descending phase [15]: this phase starts with twoessential simultaneous stages that show the beginningof the project: (1) the analysis of the application fieldgenerally allows the first functional and structuraldescription of it; (2) as soon as possible in the project,we have to build first prototypes to imply rapidly thefuture users in giving them an outline of possiblesolutions. These two stages are designed to provide astructuring framework, with regard to the futureactivities as well as the technical solutions. After thesetwo stages, a definition of a process model of work isissued (figure 1). That allows to define the list of tasksto be carried out for the operation of the future DSS torealize. They can correspond to functional and nonfunctional needs: for example the need to make the useof the DSS very easy corresponds to a non-functionalessential need. Preceding work can be cyclic, as thearrows suggest it on the model. The prescribed tasksfollowing the execution of the preceding stages must bemodelled [1]. One uses UML to define the tasksthrough use cases diagrams with their detaileddescription. Following the use cases construction, it isnecessary to carry out the tasks-oriented analysis anddesign both in terms of HCI and applicative aspects.Each task in the interactive system has a degree ofinteractivity. Three main categories of tasks can beidentified (these categories are traditional in SE andHCI since the Eighties): (1) the tasks in which only theuser is implied, called manual tasks, (2) the tasks inwhich only the applicative aspect is implied, calledsystem or automatic tasks, (3) tasks implying both theuser and the system, called interactive tasks. The UML
897
collaboration and sequence diagrams allow specifyingand designing the HCI for each task (while associatingthese models increasingly precise representations interms of screen pages). In that purpose, we have torefer to the probable behaviours of the various types ofuser: it corresponds to a user model in the large sense[20]. Some knowledge on the users are general andcome from the literature, others are specific and comefrom the field (interviews, observations ...).
According to the field specificities, it is a questionof analyzing the various tools for Decision Support(DS) in order to define the most adapted ones to thecontext of the system to realize. The last stage of thedescending phase leads to the implementation of thecomplete DSS or its prototyping. This stage ofimplementation transforms the concrete HCIcharacteristics in a representation that can beimplemented directly by using a graphic toolbox or aHCI generator.
Ascending phase [15]: the evaluation of theinteractive system consists in testing if the users canachieve their tasks by using it. Two properties areusually explored for such evaluations: utility andusability, themselves divided into a whole of wellknown criteria in HCI. There are a great number ofmethods usable for HCI evaluation [23]: observations,interviews, traces analysis, questionnaires, etc. In thisphase, we generally concentrate on the execution of thetasks linked with the visible elements of figure 1: (1) onthe one hand according to the user behaviour during theinteraction with the system (difficulties, necessary timeto achieve a task, numbers and types of human errors,etc), (2) and on the other hand, according to the systemin terms of differences between the objectives andresults obtained. This ascending phase starts with therigorous definition of the experimental protocols(subjects, situations and tasks concerned, implied HCIand assistances, data to collect. ..) [2]. Once collected,the data is treated according to the worked outoperational principles of the preceding stage. It is aquestion of correlating the data obtained with thehuman activity which was observed, and this for theunit of the decision-making process. Operatingsequences are thus showed. This work allows graduallyreconstituting the real tasks. Experience shows thatthese tasks can be very different from the prescribedtasks (initially described in the descending phase). Afundamental principle of the U model is precisely theconfrontation between the real tasks and the prescribedtasks. The result of the confrontation leads, either tovalidate the complete system (HCI and support), or toreveal its limitations to improve it gradually. Feedbackstowards various stages of the descending step are thus
necessary, according to the extent of the improvementsto be brought to the system.
4.2. Total methodology suggested (UPIU)The HCI has a dominating place in our approach
which aims at the realization of DSS. The approachthat we propose (figure 1) redefines the user role sothat he/she can intervene at any time of the decisionmaking process. To do so, we used the principle ofiterative and incremental development of UP, thatallows an early evaluation of each task carried out fromthe first iterations of the development process. So ourapproach consists in proceeding in several iterationsfrom inception to the transition. We have applied theadapted U model in each iteration. In fact, the fiveactivities of the original UP (needs assessment,analysis, design, implementation and tests) modelneither the users nor their interactions with themachine. Our approach thus recommends the presenceand the constant user participation throughout theproject. The objective is to get the interaction betweenthe user and an intuitive, ergonomic and adapted DSSin order to increase the effectiveness of the humanmachine couple and to prepare, as soon as possible, theacceptance of the system by the users. In figure 1: A,BCD and E are the new activities of UP. Eachactivity is divided into sub-activities that model theHCI of the aimed DSS.
5. Application in the healthcare domainIn the preceding section we presented our globalsolution for DSS development. This approach wasapplied to a concrete case in the healthcare domain. Itaims to help physicians (users of the system), tounderstand and prevent nosocomial infections (i.e.infections contracted by the patients during theirhospitalization). It is currently under use in the ICU ofthe Teaching hospital Habib Bourguiba in Sfax,Tunisia. The design and the realization of this DSSwere carried out according to the four phases of UP(inception, development, construction and transition).Each stage of these iterations (needs assessment (A),analysis (B) and design (C), implementation (D) andtest (E)) was carried out while following the actionssuggested in the U model (figure 1) (needs assessment,definition of the tasks, users modeling, and prototypingof the interfaces and application, evaluations andvalidations). These actions are not carried out with thesame "intensity" in each iteration. Let us take theexample of the action "user modeling"; this actionbegan "quietly" at the beginning of the project (phaseof inception of UP), became dominant from the firstiteration of the development phase, it is reduced toalmost nothing at the last iterations of the construction
898
and transition phases. At the moment we are testing the9th version of the applicat ion. This gives an outline ofthe difficulties related to this application field, andespecially of the need to go ahead step by step in veryclose link with the users suggesting gradually of new
improvements and tests, within sight of the resultsobtained at the time of the preceding iterations.Because of the lack of place in this article, it is notpossible to explain each implementation step of thesuggested approach, but that will be the subject offuture articles.
cted
eria
r------------------------Test and evalua tion (E)O b j ec tives
------ ! --------I 1I 1I 1
D ev elo p Analyze t h e ~Iodel the user I 1
interface application I+ - ~- - ~ - ; V a lid a ted D S Sp r o to types D Ol113 in
Specify and design S pecify a nd design the er o f r o n t a ti o n1 1
the automatic D S HCI and interactive D S 1 1 D ata to be coUem odules modules !+----- E l-gopomic
I . . E valua ti o n critcfltena
a n d methodsnatysts and d esfg n 1 1
(B) and (C 1 1
1 1 D efin e experin lentalSpecifica ti o n and 1 1 protocols
__ ~~i,gn_n.!.o!!e!s__ 1 1------- ---------- 1-------- ----------- ---------- , 1 D S S versrou to1
I Create and integrate the D S S (HCI and D S ~oo"')be evaluated
1111
1 Implemen tation (D) 1L 1 _
Pha s e s
A
Needs assessmen t
In ruauzanon I Dev el opm en t I
(A) I -----l - I__r : t'--
Con stru el i on Tra n ~.lti on
Analy s i s
D e s ign
Implementa t io n
T e stin g
I
'" 'l'"(B) .:-- -I I
(C)I I1 ..--- I
I I
(0) I ....i-f,--
I I
(E) I I
I :'-;- 1P co j i rn i n .. y I I t e r . rt e rIt 13' r;;t t r o n (5) # 1 #n
,,: : I
,~
:, n e r :1 i t a f . I' n e t
I # n-+l I #n-+:! # m
"
i t e ,# m + 1
This article is the result of a work proposing anHCI-enriched approach for DSS development. We are
6. Conclusion
I t e r a ti o nsFigure 1: the UP/U approach
interested in highly interactive DSS. Indeed, thedecision-making requires a Human-ComputerInteractivity to be carried out. A modelling of such asystem must thus call upon two specialities: SE and
HCI. That is why we propose an approach using the Umodel coming from the HCI field, and UPmethodology coming from the SE one. They arecomplementary; so it is possible to carry out a DSS thattakes advantage from the two fields. In order tovalidate this approach, we currently develop a DSS inthe healthcare domain. This system aims at thesupervision of the nosocomial infections, contracted ina hospital Intensive Care Unit. The realization of theproject is done in constant connection with thephysicians, users of the system: needs assessment, tasksdefinition, identification of the users' characteristics,and prototyping, evaluations and validations with them.The finalization of our DSS is not totally completed.Further works will be held to improve and validate it.We also intend to use techniques of data mining toextract knowledge from data. Knowledge resultingfrom each of these techniques helps users to makedecisions. We also wish to propose a methodology forthe evaluation of DSS taking into account the methodsand techniques of evaluation used in two fields: HCIand visual data mining [8] [16].
7. References[1] M. Abed, 1.M. Bernard and 1.C. Angue, "Task analysis
and modelization by using SADT and Petri Networks",Proc. Tenth European Annual Conference on HumanDecision Making and Manual Control, Liege, 1991.
[2] M. Abed, Methodes et Modeles formels et semi-formelsde conception et evaluation des systemes hommemachine, HDR Thesis, University of Valenciennes,France, 2001.
[3] S. Balbo, Evaluation ergonomique des interfacesutilisateur: un pas vers I 'automatisation, PhD Thesis,Grenoble, France, 1994.
[4] B. Boehm, A Spiral Model ofSoftware Development andEnhancement, 1988
[5] S. Garlatti, "Multi-media and interactive decision supportsystems in complex situation", In 43th meeting of theEuropean working group "Multicriteria Aid forDecisions", 1996
[6] D. Hix and H.R. Hartson, Developing to use interface:ensuring usability through product & process, Wiley,New York, 1993.
[7] I. Jacobson, G. Booch and 1. Rumbaugh, The UnifiedSoftware Development Process, Addison WesleyLongman, 1999.
[8] D. Keirn, Information Visualization and Visual DataMining. IEEE Transactions on Visualization andComputer Graphics, Vol. 8(1), pp.I-8, 2002.
[10] C. Kolski, H. Ezzedine and M. Abed, "Developpementdu logiciel : des cycles classiques aux cycles enrichissous l'angle des IHM". In Kolski C. (Ed.), Analyse etConception de l'IHM. Interaction Homme-machine pourles SI, vol. 1, Hermes, Paris, pp. 23-49, 2001.
[11] M. Lajnef, M. Ben Ayed, and C. Kolski, "Convergencepossible des processus du data mining et de conceptionevaluation d'IHM : adaptation du modele en U".Proceedings of IHM 2005, International ConferenceProceedings Series, ACM Press, Toulouse, pp. 243-246.
[12] C. Larman, UML 2 and Design Patterns, Eds PearsonEducation, 2005.
[13] F. Lemieux and M.C. Desmarais, "RUP et conceptioncentree sur I 'utilisateur: une etude de cas", Proc. ERGOIA' 06, Biarritz, 2006.
[14] S. Lepreux, M. Abed and C. Kolski, "A human-centredmethodology applied to decision support system designand evaluation in A railway network context", CognitionTechnology Work, vol. 5, pp. 248-271, 2003.
[15] H. Ltifi, M. Ben Ayed, C. Kolski, and A-M Alimi,"Prise en compte de l'utilisateur pour la conception d'unSIAD base sur un processus d'ECD", In Proc. ERGO'IA08, Biarritz, pp. 85-92, 2008.
[16] H. Ltifi, M. Ben Ayed, S. Lepreux, A-M. Alimi, "Surveyof Information Visualization Techniques for Exploitationin KDD", In the 7th ACS/IEEE International Conferenceon Computer Systems and Applications, Rabat, Morocco.May 10-13,2009.
[17] J. McDermid and K. Ripkin, "Life cycles support in theADA, environment", Cambridge University Press,Cambridge, the U.K., 1984.
[18] P. Millot, "Cooperation homme-machine : exemple dela teleoperation", Actes des Journees du GRAutomatique, Strasbourg, 1990.
[19] 1. Rumbaugh, I. Jacobson, G. Booch, The UnifiedModeling Language Reference Manual, Addison-WesleyProfessional, 1999.
[20] 1.M. Robert, "Que faut-il savoir sur l'utilisateur pourrealiser des interfaces de qualite'l", Ingenieric cognitive,IHM et cognition, Hermes, pp. 249-283, 2003
[21] 1.R. Ruault, "interactive UML and systems, another stepforward", In Computer-aided Design of User InterfaceIII, Kluwer Academic Publishers, pp. 243-256, 2002.
[22] H.A. Simon, The new science ofmanagement decision,Prentice Hall, New Jersey, 1960.
[23] 1.R. Wilson and E.N. Corlett (eds.), Evaluation ofhuman works: a practical ergonomics methodology (2nd
edition), Taylor & Francis, 1996.
Acknowledgements. The present research workhas been partially supported by the "Ministere del'Education Nationale, de la Recherche et de laTechnologie», the «Region Nord Pas-de-Calais», theFEDER, l' ANR. The authors gratefully acknowledgethe support of these institutions.
