9512.net
甜梦文库
当前位置:首页 >> >>

Automatic personalization of the human computer interaction using temperaments

Automatic Personalization of the Human Computer Interaction Using Temperaments ?
Hector G? mez-Gauch?a, Bel? n D?az-Agudo and Pedro A. Gonz? lez-Calero o ? e ? a
Dep. Sistemas Inform? ticos y Programaci? n a o Universidad Complutense de Madrid {hector,belend,pedro}@sip.ucm.es

Abstract
In this paper we model a personalization system which takes into account the user interaction styles regarding a software artifact and the user temperament. We propose the use of Knowledge Intensive CBR where there are cases that represent speci?c variations of a given software artifact, and there are ontologies for the static knowledge. Our model is generic and reusable. In this paper we exemplify it with a system to personalize the Linux operating system environment.

Introduction
There are many efforts in research and industry to design friendly GUI’s. But what is the meaning of friendly? It is the same meaning for all kind of users? Obviously not. But how can the different perceptions of the users be determined? A solution may be the set of characteristics which users show when they interact with the computers. These characteristics are the result of the user’s temperament. The concept of Interaction Styles refers to all the ways the user can communicate or otherwise interact with the computer system. The concept belongs to the realm of Human Computer Interaction (HCI). In HCI textbooks, such as Shneiderman (Shneiderman 1997) and Preece et al. (Preece 1994), the types of interaction styles mentioned are usually in relation to the different computer interfaces: command language, form ?lling, menu selection, and direct manipulation. The Interaction Styles are in?uenced directly by the user’s temperament, there are studies that de?ne these relations (Berens 2001). From these studies we can extract what may appeal as friendly and comfortable for users with a speci?c temperament. To be able to do this we needed to build a very ?exible model where we can test many re?nements in order to ?nd out the adequate conditions for each temperament. We found the ?exibility in the knowledge based systems, where the conditions were independent to the software. The model we are thinking about requires a lot of general knowledge to classify each feature and reason with it for each temperament. The knowledge intensive Case-Based Reasoning (KI-CBR) has these characteristics (Aamodt 1990;
? Supported by the Spanish Committee of Science & Technology (TIN2005-09382-C02-01) Copyright c 2006, American Association for Arti?cial Intelligence (www.aaai.org). All rights reserved.

D?az-Agudo & Gonz? lez-Calero 2000). This approach uses ? a ontologies, which allows us to perform the mentioned tests with ?exibility and reasoning about the topics that participate in the temperaments and in the possible variations of the system. These variations represent the different aspects of a certain system that are adequate to be personalized. The model we present is a personalization system independent of the speci?c software artifact. To apply it to a certain software, we only have to create the speci?c variations to populate the case base. In this article the domain of our example is to personalize the Linux operating system in general. To do so there is an ontology and a case base that describe the speci?c variations of Linux. In the next section we describe the three main components of our model: temperaments, variations and users. For each one of them we have formalized one ontology that is specialized in a case base. We then de?ne how to use the mapping to de?ne how a variation affects to a temperament. Then we describe the KI-CBR reasoning cycle and the architecture where we divide the main tasks in two sets: the Knowledge Engineer tasks and the user tasks. Before conclusions, there is a brief description of a domain case study where we are implementing a prototype of the model.

The main ontologies and case bases
This model relies heavily on ontologies. The static knowledge of each aspect of the design is described by terms in an ontology. Ontology is a term borrowed from philosophy that refers to the science of describing the kinds of entities in the world and how they are related. To formalize our ontologies we use the Ontology Web Language (OWL). An OWL ontology may include descriptions of classes, properties and their instances. Given such an ontology, the OWL formal semantics speci?es how to derive its logical consequences, i.e. facts not literally present in the ontology, but entailed by the semantics. These entailments may be based on a single document or multiple distributed documents that have been combined using de?ned OWL mechanisms 1 . The OWL reasoning capabilities relies on the Description Logics paradigm. We de?ne cases as individuals of a concept which belongs to the ontology. A case has several slots and facets, repre1

http://www.w3.org/TR/owl-guide/

sented as properties and sub-properties. Each of them may contain individuals of other objects of the same ontology or other imported ontologies. The advantage of this approach is that a reasoner may check the consistency and may classify the kind of individuals. We use very narrow ontologies and cases to simplify the design and updates. This model is embedded in another model (G? mez-Gauch?a, D?az-Agudo, o ? ? & Gonz? lez-Calero 2005), and both share the ontologies and a case bases. This is the reason because the model do not use some of the classes mentioned in the ontologies.

Temperaments: TEMPonto and the case base
We use the temperament theory (Keirsey 1998), which is widely applied in psychology and in companies to interview job candidates. Keirsey’s theory is centered in the long-term behavior patterns, i.e., what people do. It is an interpretation of the Myers-Briggs and Carl Jung’s writings on personality types. These authors were more interested in what people think. The theory de?nes four basic temperaments. Each person has a unique proportional combination of the four temperament types. In this article we use an example where we consider an unique proportional combination as the description of a new case: Artisan 10 %, Guardian 10 %, Idealist 30 % and Rational 50 %. Normally one of the temperaments is predominant, rational in the example. This means that the person will behave most of the time like that temperament. We describe each temperament type by a set of traits. In Figure 1 there is a complete set of traits for the rational temperament. Each trait characterizes the way people behave in relation with a particular aspect. For our example we consider the trait ”value”, that means what people will value most in relation with the different aspects of that trait. The aspects of the ”value” trait are shown in Figure 1. One of them is being, that means how people appreciate to be. Rational people value being calm. The word ”calm” is the value of the aspect ”being”. To represent it in the ontology, shown partially in Figure 2, we create one individual temp:ValueBeingCalmX of Trait class. In contrast with the previous example, artisan people value most being excited. Another example is the trait Language which has a rhetorical aspect. Rational people use a heterodox kind of rhetorical aspect. In contrast, artisans prefer a technical kind of rhetorical aspect. We believe that if the model may use different aspects of language adapted to the language of the user’s temperament, the user will understand better the messages and he will feel more comfortable working with the system. We personalize the system building variations which affect some speci?c aspects of a trait.

Character: RATIONAL Communication: abstract Implementation: utilitarian DESCRIPTION : Language: deductive - Referential: categorical - Syntactical: subjunctive - Rhethorical: technical Intellect: strategic Directive role: coordinator - Expresive role: fieldmarshal - Reserved role: mastermind Informative role: engineer - Expresive role: inventor - Reserved role: architect Interest - Education: sciences - Preoccupation: technology - Vocation: systems Orientation - Present: pragmatism - Future: skepticism - Past: relativism - Place: intersections - Time: intervals Self-Image - Self-Esteem: ingenious - Self-Respect: autonomous - Self-Confidence: resolute Value - Being: calm - Trusting: reason - Yearning: achivement - Seeking: knowledge - Prizing: deference - Aspiring: wizard Social Role - Mating: mindmate - Parenting: individuator - Leading: visionary

Figure 1: A basic temperament and its traits classes that are necessary to de?ne the variation class. Properties names are on top of the text boxes in Figure 2, their are in thick font in grey color. The main properties of a variation are: ? affectToTraits has several traits of temperaments which are affected by the variation. Our variation example has the temp:ValueBeingCalmX trait. ? hasExecutionSteps are the necessary actions to execute the variation. In our example we have only one action, ”to change the background color to green”. The command to execute it is: Gconf --set \# 008000 The command is represented by an individual, execStepChangeBackgroundGreen, which belongs to ExecutionStep class (not shown in the ?gure.) This class has two main properties: – hasActivationCommand has the command to execute the step, e.g.: ”Gconf” in our example. This is an Linux utility that changes several aspects of the GUI, depending on the parameters, that is the next property. – hasActivationParameters the parameters of the command. The parameters are of the ExecParam class that

Variations: VARIonto and the case base
We call variations to all the possible actions that are executed to personalize a domain. For example, in our sample domain, the Linux system, we have the assertion ”the green color makes you feel calm” which is illustrated in Figure 2. We create a variation to include this assertion as a possible personalization action. The variation is an individual GuiLookBackgroundColorTest1 of the variation subclass GUILookBackgroundColor. The ontology has several

is explained below. The ExecParam class is specially important for the adaptation of cases. To adapt a case we need a range of ?exibility. We get it by the declaration of possible ranges. Each one refers to one of the basic temperament. In Figure 3 it is the execParamBackgroundGreenSet individual, which is the parameter of the ExecutionStep described before. The main properties of the ExecParam class are (see in Figure 3, thick font and gray color): ? hasExecParamName has the literal name of the parameter to be executed by the command. In our example ”–set”. ? hasExecParamValue has the value of the param. In our example is ”#008000”. ? possibleAdaptationRange has four subproperties: idealistRange, artisanRange, rationalRange, guardianRange. It indicates how strong is the effect for each basic temperament Each one has a list of values. They describe the distance of the value to that speci?c temperament. The ?rst value in the list is the nearest with the strongest effect and the last value is the farthest with the weakest effect. An example of the lists is in Figure 3. ? hasExecParamRelevancy indicates the general relative importance in the adaptation process. It contains the corresponding four subproperties. Each subproperty is the relative importance speci?cally for each temperament. For example, a parameter has a very low value in guardianRelevancy when it is very little related with that temperament, e.g.: the background color is not relevant for guardian people because they do not care much about colors. This property does not appear in the ?gure.

? hasUserType that is the best match of user types. ? hasUserAdaptations with adaptations of that UserType to the speci?c user. This property is empty when the UserType Matches exactly or very near with the user’s Temperament proportions. The adaptations are obtained from the ranges of each variation that were explained in the VARIonto. The UserType case represents the unique proportion of the four basic temperaments and the variations generated and stored for it. Each case in the case base is one of these combinations. To avoid an in?nite number of cases, a minimum gap between cases is de?ned a priori. The name of the example UserType case is UserType-A10-G10-I30-R50. The main properties are: ? hasTemperamentProportions is the same property as in the user case. ? hasUserType with the variations for this user type.

Reasoning Cycle
We are using the previous example to illustrate the general reasoning cycle in our model (Figure 4). We follow the classic CBR cycle (Armengol & Plaza 1994). We query the CBR system with a description based on a form ?lled by the user. Suppose the form gives us the following user temperament proportions: Artisan 10 %, Guardian 10 %, Idealist 30 % and Rational 50 %. The reasoning cycle retrieves the most similar case. Let’s suppose it is Artisan 30 %, Guardian 10 %, Idealist 30 % and Rational 30 %. To measure the similarity between cases we use the following de?nition of distance which is a sum of the differences for each of the four temperaments:
T=Rational TArtisan =

Mapping variations into temperaments
Each parameter is related with speci?c traits of the basic temperaments, not just with the temperament itself. This relation is by the property affectToTraits of a variation class that contains traits of the temperament class. This is depicted in Figure 2. For example, the previous example of background color affects in the value trait of the rational temperament in a very different way to the artisan temperament. This is because the latter has “artistic” as value in the selfEsteem trait. This indicates that the artisan people appreciate the colors from the artistic viewpoint which is very different to the guardian people that appreciate colors that produce calm. ValueBeingCalm is an aspect of the trait ”value” represented as a subclass of the ”value”. This is possible because the GuiLookBackgroundColorTest1 variation of our example has in affectToTraits both traits, ValueBeingCalm and selfEsteemArtistic. And the execution parameters have different ranges for each of both temperaments.

?

( %NewCaseT - %retrievedCaseT ) * CorrectionFactorT

User Types and users: USERonto and the case base
In the USERonto there are two kinds of cases, the UserType and the User. The User case describes the user knowledge needed by the model and has these main properties: ? hasTemperamentProportions with the percentage of each of the four basic temperaments.

The %NewCase is the proportions of the query. There is a correction factor that is proportional to the previous difference. The correction factor is very high when the retrieved case is very far for one of the temperaments with a high percentage, because this retrieved case is not a good case even if the other percentage temperaments are similar. The system reuses the retrieved case by adapting the retrieved case output slots, i.e.: a set of variations for each of the four basic temperaments. Variations are in its own case base. The adaptation actions are calculated by the difference between the proportions in the query and in the retrieved case; in our example the adaptation actions are: make Artisan decreasing 20 %, Guardian stays unmodi?ed, make Rational increasing 20 % and let Idealist without modi?cations. After the adaptations, the system executes the modi?ed variations. These variations represent the solved case. The system performs the adaptation of each variation using the possibleAdaptationRange property that has the possible ranges for each temperament as described before. In Figure 3 is our variation example; we see in it that the range for each temperament is a list of values,e.g.: rationalRange. The total number of values is one hundred percent. To make

--> The assertion: “the color green makes you feel calm” VARIonto TEMPonto --> Variation: change the background color to Green
Character: RATIONAL Communication: abstract Implementation: utilitarian DESCRIPTION : Language: deductive - Referential: categorical - Syntactical: subjunctive - Rhethorical: technical Intellect: strategic Directive role: coordinator - Expresive role: fieldmarshal - Reserved role: mastermind Informative role: engineer - Expresive role: inventor - Reserved role: architect Interest - Education: sciences - Preoccupation: technology - Vocation: systems Orientation - Present: pragmatism - Future: skepticism - Past: relativism - Place: intersections - Time: intervals Self-Image - Self-Esteem: ingenious - Self-Respect: autonomous - Self-Confidence: resolute Value - Being: calm - Trusting: reason - Yearning: achivement - Seeking: knowledge - Prizing: deference - Aspiring: wizard Social Role - Mating: mindmate - Parenting: individuator - Leading: visionary

->“Rational Temperament values being calm”

Figure 2: An assertion as a variation related to a speci?c trait of a temperament the retrieved case 20 % more rational we calculate that proportion in the list and we choose the element which occupies the position corresponding to the obtained proportion, i.e.: 50%. This is based in the fact described previously that the ?rst elements in the list are nearer to the temperament, i.e.: in our example those ?rst elements are more ”rational”. The current mood ?lters the adaptation, increasing or decreasing some of these adaptations. The next task is to revise the solved case using the user’s feedback. The user may agree or complain about the performed variations. If there are complains, the system repairs the case to create the repaired case. Once the user agrees with the variations, the result is a new unique case that is remembered, i.e.: stored in the case base. To avoid having too many cases that are very similar each other, the store task is done only if there is at least the minimum allowed distance between the new case and the old cases. This threshold keeps a balance between having to compute a solution from scratch for each new query or to have thousands of very similar cases stored that would prevent a good case retrieval performance.
Problem
Learnt case
It stores the Repaired case
Ontologies
(Structure) Description Logics

User’s profile: -Temperament -Current mood

RET RIE VE

Retrieved case
(Contents) Variations: - GUI - Speed Emotions Moods Mapping

Set of system variations for personalization

EMBER REM

.....

Temperament

Case Bases

Ways of : - Communicating - Thinking - Using time - Using tools

It modifies the system variations with the user feedback

Repaired case

Solved case

It executes the set of system variations

R E V IS E

Confirmed solution

Suggested solution

The Architecture: Main Tasks
In this section we brie?y explain how we organize our model in terms of tasks, i.e., pieces of work required to be done and that are treated as basic units. As it is shown in Figure 5 our model organizes these tasks into two types: knowledge engineer and user tasks. To solve these tasks our model relies in the described ontologies, that includes the common vocabulary and general descriptions, and the DLs reasoner. Solving the knowledge engineering tasks means de?ning the ontologies, making the mappings with the speci?c domain variations and verify them. Once these tasks are solved the reasoning cycle described previously is in charge of solving the user tasks, composing their results, and performing the corresponding variations.

Figure 4: Overview of the reasoning cycle with ontologies

User Tasks
? Interviewer The ?rst time user logs in(or when ever he wants to revise his extended pro?le), he ?lls three types of information: – a questionnaire. He has two options, a short questionnaire of sixteen questions or a long one of seventy questions. The ?rst one, the so called four type sorter, classi?es the user temperament giving a unique proportion for each of the four basic temperaments. The long one, the so called temperament sorter II, assigns one of the

RE USE



Figure 3: ExecParam of a variation class and ranges to be adapted sixteen temperaments based on the four pairs of features represented with eight letters. The user may ?ll in one of the questionnaires or both. In this last situation the system has more elements to reason more accurately. In this article we focus on the model that uses the four basic temperament types. – Background knowledge topics he ?lls in the extended pro?le are: Computer skills, Domain expertise, ... – The last topic the interviewer asks is the current mood or emotional state. There is a set of icons re?ecting several general emotional states such as happy, sad, apathetic, ... or angry. The user may choose the descriptors that better de?ne his current mood. – After the previous step, or when the user logs into the system for second time, it asks him to choose a mood that better ?t in his current state. The result is a query with the description of the user temperament. The personalizer uses it in the CBR cycle. ? Personalizer There are two types of personalization: Manual, where the user chooses the state of each of the possible variations of the system; Automatic , where the system reasons to decide the appropriate variations according to the user’s temperament and his current mood. This is performed by the Case-based reasoning cycle described in the previous section. These are the functional steps: 1. The personalizer retrieves the most similar case to the query. The similarity is an heuristic distance between the proportions for each temperament of the query and the description of cases in the case base of users. 2. The personalizer applies the heuristic distance to adapt the variations of the retrieved case to the description of the query. The result is the solved case. After the adaptation, the personalizer executes the variations. 3. The personalizer calls the feedbacker task to revise the solved case. The user may complain about the variations supplying a set of ?xes. The personalizer will modify the solved case variations using the ?xes. The result is the repaired case. 4. The personalizer stores the repaired case, called learnt case, only if it has the minimum threshold of heuristic distance to other stored cases. If it does not reach the threshold, the personalizer stores the ?xes as a particular personalization of that user. ? Feedbacker This task is optional. The user activates it when he disagrees with the variations. The user creates ?xes, which indicate which aspects, the variation and its parameter values, he dislikes and he proposes new ones. These ?xes are passed back to the personalizer.

Knowledge Engineering Tasks
? Editor The ?rst task is to build the ontologies and the case bases. The knowledge engineer describes the model in terms of classes, properties and its instances. To create the case bases he uses the ontology terms. The editor uses a reasoner (Pellet2 or Racer3 ) to verify the consistency of cases and ontologies. We use Protg as the ontology editor 4 . The other tasks use the reasoner to classify new individuals of cases and to update the ontologies and case bases. ? Mapper The model is independent to the personalized domain because the variations are in a separate ontology,
2 3

http://www.mindswap.org/2003/pellet/index.shtml http://www.sts.tu-harburg.de/ r.f.moeller/racer/ 4 http://protege.stanford.edu/

Knowledge Engineer Tasks Editor Mapper Ontologies, & case bases VARIonto
Variations case base
temperaments case base

Conclusions and Further Work
Since the meaning of friendly GUI is not the same for all kind of users, we propose to personalize systems based on characteristics which users show when they interact with computers. These characteristics are the result of the user’s temperament. We propose a model to personalizing software systems taking into account the user’s temperament. It is independent of the domain and ?exible because we use knowledge based systems to keep separated the knowledge of the domain from the reasoning process. We use the KI-CBR approach to model the static knowledge, which we represent with ontologies. The core ontologies are: variations of the system, which are the possible elements to be personalize; temperaments with traits, which are the aspects of a temperament affected by the variations; and users, which has a temperament represented by a unique proportion of the four basic temperaments. To apply the model to a domain, i.e.: a program or operating system, is only necessary to create the speci?c variations to populate the case base. In this article we describe brie?y a case study in the domain of Linux personalization. Currently we are developing a prototype for the Debian distribution and Gnome windows manager.

Verificator
User’s Tasks

TEMPonto

Interviewer
Personalizer: - Automatic -Manual Feedbacker

OWL-DL Reasoner

users case base

USERonto

Figure 5: Tasks in the architecture of the model

VARonto. We can put variations of a new domain application in this ontology and exchange the mapping of VARonto to the temperament ontology. With only these changes the model will personalize the new domain application. This task is an editor that helps to create the relations between the variations and the traits affected by them: – It helps to create a variation. – It helps to create the temperaments. – It helps to create the mapping between variation and the temp. The most complex is to decide which traits of which temperament are affected by the variation. We found useful to ask for help to a psychologist. The positive side is that once this mapping is performed, it is not necessary to change unless a complain from the feedback is ?led.

References
Aamodt, A. 1990. Knowledge intensive case-based reasoning and sustained learning. In Proceedings of the ninth European Conference on Arti?cial Intelligence – (ECAI90), 1–6. Armengol, E., and Plaza, E. 1994. A knowledge level model of knowledge based reasoning. In Wess, S.; Althoff, K. D.; and Richter, M. M., eds., Proceedings of the 1st European Workshop on Topics in Case-Based Reasoning, Kaiserslautern, Germany - EWCBR’94. Berlin: Springer– Verlag. 53–64. Berens, L. V. 2001. Understanding Yourself and Others: An Introduction to Interaction Styles. Telos Publications. D?az-Agudo, B., and Gonz? lez-Calero, P. A. 2000. An ? a architecture for knowledge intensive CBR systems. In Blanzieri, E., and Portinale, L., eds., Advances in CaseBased Reasoning – (EWCBR’00). Berlin Heidelberg New York: Springer-Verlag. G? mez-Gauch?a, H.; D?az-Agudo, B.; and Gonz? lezo ? ? a Calero, P. 2005. COBBER, Towards an Affective Conversational KI-CBR Framework. In In Proceedings of 2nd Indian International Conference on Arti?cial Intelligence. Keirsey, D. 1998. Please Understand Me II: Temperament Character Intelligence. Prometheus Nemesis Book Company. Preece, J. 1994. Human-Computer Interaction. Addison Wesley. Shneiderman, B. 1997. Designing the User Interface: 3 edition. Addison Wesley.

A case study: personalization of Linux
The main ontologies are created. To applied the model to the Linux system there are necessary three steps: ? To de?ne and implement a set of variations for Linux. ? A mapping between the set of variations and the temperament ontology to indicate how each variation affects each temperament trait. ? De?ne a period of re?nement using the feedback of the users when they do not like the personalization. Our users are a set of students of the last year of their Degree on Computer Science. We are developing a prototype for the Debian distribution using Gnome as window manager. It allows variations at several levels of abstraction, from a very detailed level, such as the ”Gconf” utility of our example, to very generic level, such as ”themes”. Themes offer a complete unique look for the Gnome. We create a case base of themes, classify them according to the traits that affect to the temperaments and feed them as variations of the system.



更多相关文章:

All rights reserved Powered by 甜梦文库 9512.net

copyright ©right 2010-2021。
甜梦文库内容来自网络,如有侵犯请联系客服。zhit325@126.com|网站地图