From Classic User Modeling to Scrutable User Modeling

Giorgio Gianforme

, Sergio Miranda

, Francesco Orciuoli

and Stefano Paolozzi

Dipartimento di Informatica e Automazione

University of ROMA TRE, Rome, Italy

Dipartimento di Ingegneria dell’Informazione e Matematica Applicata

University of Salerno, Fisciano, Italy

Abstract. User Modeling still represents a key component for a large number of

personalization systems. Maintaining a model for each user, a system can suc-

cessfully personalize its contents and use available resources accordingly. On-

tologies, as a shared conceptualization of a particular domain, can be suitably

exploited also in this area. In this paper we explain some concepts about user

modeling, particularly focusing on scrutability and its importance in ontology-

based user modeling systems.

1 Introduction

In a world where information increasing constantly, the challenge is not only to make

information available to people at any time, at any place, and in any form, but specif-

ically to say the “right” thing at the “right” time in the “right” way. User modeling

researches try to address and solve these issues.

User modeling is one of several research areas that intuitively seems to be winning

propositions based on their obvious need and potential “return of investment”. Indeed

as systems have became increasingly more complex in terms of large amounts of infor-

mation to be conveyed and a wide range of task structures, it is important that a system

should be able to provide effective guidance and assistance to enable the user to make

full use of the available functionality. Moreover, given the variety of users of such sys-

tems, the fact that different users will not all have the same problems or needs, and that

a user’s level of expertise is likely to change over time, it is desirable that interfaces

to complex systems should be able to adapt to and support the requirements of indi-

vidual users. Thus an information-providing system could tailor the form and content

of information provided to users according to its assessment of what the users need or

wish to know, while a dialogue module could construct the human-computerinteraction

according to the users’ preferences.

In this paper we analyze the characteristics of actual user modeling systems and

their importance in human-computer interactions. Moreover, we explain the details of

a fairly new concept in user modeling called scrutability: the ability of analyzing and

eventually modifying a user model by its user. Finally we discuss about the role of

ontologies in user modeling also in according to the scrutability concept.

Gianforme G., Miranda S., Orciuoli F. and Paolozzi S. (2009).

From Classic User Modeling to Scrutable User Modeling.

In Proceedings of the 1st International Workshop on Ontology for e-Technologies OET 2009, pages 23-32

DOI: 10.5220/0002222200230032

 SciTePress

The rest of the paper is structured as follows. Section 2 explains the main concepts

of modern user modeling, describing the characteristics of a user model. Section 3 de-

scribes the importance of user modeling in the human-computer interaction ﬁeld. In

Section 4 the concept of scrutability and its importance in user modeling are described.

Section 5 are devoted to the description of the ontology role in scrutable user modeling.

Finally Section 6 concludes.

2 User Modeling Analysis

The term user modeling has been applied to the process of gathering information about

the users of a computer system and using this information to provide services or infor-

mation adapted to the speciﬁc requirements of individual users (or groups of users).

A “perfect” user model would include all features of the user behavior and knowl-

edge that affect their learning and performance [20]. The information, contained in the

user model determines the adaptability of the system. Constructing this kind of model

is quite complex task, even for people not only for machines, thus simpliﬁed, partial

models are used in practice.

The term user model has been used in several different ways in Artiﬁcial Intelli-

gence (AI) and Human-Computer Interaction (HCI) and it is important to clarify these

differences at the outset. According to software design terminology, the term ’user

model’ usually refers to the designer’s model of the users of the system, either in fairly

gross terms, such as their expected level of proﬁciency, or more speciﬁcally in terms of

the models users may have of the system and of the procedures involved when interact-

ing with the system.

These models can be identiﬁed as the users’ system model and the users’ task

model [10]. Generally speaking, designers models of users are implicit and thus are not

represented explicitly within a computer system. This type of modeling is usually done

once only, at the design stage of a system, so that any unanticipated user characteris-

tics or any changes in user characteristics over time cannot be accommodated without a

complete re-design of the system. A more reﬁned system is one which is adaptable, that

is system where a user can make choices among various options that affect the system’s

behavior and can save these choices for future reference in a user proﬁle. However, the

most complex (or even sophisticated) kind of system, is an adaptive system which au-

tomatically acquires knowledge about its users, updates this knowledge over time, and

uses the knowledge to adapt itself to the different users requirements.

The two main features of a user model in an adaptive system can be summarized as

follows. First of all, the user model can be viewed as a knowledge source of information

about users, which contains explicit assumptions on those aspects of a user that might

be relevant to the interface of the system. Secondly, a user model could also be used

to simulate the user activity, for example, to predict the effect of an action of the user.

User models have been employed in this way in intelligent tutoring systems and also to

provide anticipation feedback in natural language dialogue systems [5].

2.1 User Model Characteristics

Three main characteristics can be considered for the deﬁnition of a user model. Namely

they are:

– what information about the user is included in the model and how it can be obtained

– how we can model the user representation in the system

– how we can distinguish between a model for individual user and a model for classes

of users

We described these aspects in the following.

Information About the User. Information about the user, which is usually saved in

the user model, can be divided in static (i.e they remain constant through the learning

process) and dynamic (i.e. they can change during the learning process).

The static part of the user model includes user’s personal characteristics (e.g. age,

gender, level of instruction, etc.), user’s capabilities (e.g. background knowledge, cog-

nitive and non-cognitive abilities, etc.) and user’s preferences. These characteristics are

usually analyzed at the beginning of the learning process, using interviews (e.g. online

questionnaires) and different tests [6].

The dynamic part of the user model includes user’s knowledge, concepts and skills,

learning style, motivation, viewpoints, current goals, plans and believes, learning ac-

tivities that have been carried out, etc. The information for devising this part of the

user model are gathered directly from users (e.g. speciﬁcation of current goal), through

tests and practice (e.g. test results, user history of responses and problem solving be-

havior), or user’s actions (browsing behavior, visited concepts, time spent on page, total

session time, selection of links, searching for further info, queries to the help system).

This information is constantly collected during the learning process and is also used for

updating the user model.

It is also possible to distinguish between explicit information acquisition (EIA) and

implicit information acquisition (IIA) in user modeling. EIA usually refers to the ad-

dition of new information to the user model on the basis of information gathered from

an external source. This external source can be represented, for example, by a user

interacting with the system or a user such as a team manager who is reporting the char-

acteristics of an employee or groups of employees to the system. IIA refers to inferring

user model information on the basis of the knowledge available within the user model

and rules of inference that are applied to this knowledge. This kind of information is

usually acquired incrementally. Examples of IIA and EIA are given by user-driven ac-

quisition (in which the system passively acquires information from user actions) and

system-driven acquisition (in which the system identiﬁes what information is needed

and seeks to acquire it), respectively.

Representation of the User’s Information. Different methods can be used for con-

structing the user model. They may include:

– Bayesian methods,

– Machine learning methods (rule learning, learning of probabilities, instance-based

learning),

– Logic-based methods (ﬁrst order predicate calculus), overlay methods,

– Stereo-type methods [17] [7],

– Speciﬁcally developedcomputational procedures (user’s expertiseis calculated from

their navigational actions or time spent on documents),

– Speciﬁcally developed qualitative rules and procedures (special rules regarding

user’s properties or behavior),

– Other general methods (plan recognition)

In real application a combination of two or more methods is frequently applied,

especially when different methods are used for initializing and maintaining the user

model. This grants more accurate modeling and allows better exploitation of gathered

information.

Constructing and Updating the User Model. The methodology for exploiting the

user model is, in some way, similar to the user modeling construction methodology.

Moreover, we can use well-known methods from artiﬁcial intelligence and machine

learning (e.g. Bayesian networks, rule learning, instance-based learning, learning of

probabilities, logic-based and heuristic), as well as other general techniques and princi-

ples (e.g. plan recognition), speciﬁcally developed computational procedures or speciﬁ-

cally developed qualitative rules (e.g. rules for selecting and evaluating examples, rules

for choosing adaptation type and rules for choosing questions).

Classes of Users vs Individuals. Grouping together classes of users, according to

shared characteristics, has been a widely applied technique in user modeling and it is

related to the topic of stereotypes [17] [15]. Stereotypes provide a way of reducing to

some extent the problem of the unique modeling of each individual user as we can use

stereotypes to ﬁll in the background information on a person when we know only few

details about them [16].

We can also reﬁne the distinction between individual users and group of users in

terms of the number of agents to be modeled and the number of models. Generally, we

assume that the user model refers to a model of the person who really uses the system.

However, a system may also be required to model other persons at the same time [19].

For example a manager may want to ask a system about characteristics of employees.

3 User Modeling in Human Computer Interaction

We have already mentioned that the user modeling problem involved two main research

areas, i.e. artiﬁcial intelligence and human-computer interaction (HCI). For the purpose

of this paper we are more interested in HCI.

A key objective of HCI researches is to make systems more usable, more useful, and

to provide users with experiences which are strictly related to their speciﬁc background

knowledge and goals. The challenge is not only to make information available to people

at any time, at any place, and in any form, but speciﬁcally to say the “right” thing at the

“right” time in the “right” way [18].

Some HCI researchers have been interested in user modeling because there is the

potential that user modeling techniques will improve the collaborativenature of human-

computer systems [2]. In HCI the different kinds of information represented in a user

model are usually concerned with the user’s cognitive style and personality factors.

The main objective in HCI is to provide a “dialogue” adapted to the requirements

of the user and to the particular task which the user is currently engaged in. Thus, in

addition to a user model, a task model is required which describes both the physical and

the conceptual aspects of a domain [1]. While the task model is relativelystable, the user

model, which includes the user’s mental model of the domain and its task structure, is

likely to change dynamically as the user interacts with the system. A requirement on

user models in HCI is that they should be psychologically relevant, and this aspect is

generally validated empirically in experimental studies. This information about users

may be used in a predictive way to assist system designers to design a system with

user requirements in mind; however, recent work in HCI has been concerned with the

automatic adaptation of computer systems to their users and has thus converged with

one of the main concerns of AI.

Recent studies relate user modeling with the new research area of “multimodality”.

By deﬁnition, “multimodal” should refer to the use of more than one modality, regard-

less of the nature of the modalities. However, many researchers use the term “multi-

modal” referring speciﬁcally to modalities that are commonly used in communication

between people, such as speech, gestures, handwriting, and gaze. Multimodality seam-

lessly combines graphics, text and audio output with speech, text, and touch input to

deliver a dramatically enhanced end user experience. When compared to a single-mode

interface in which the user can only use either voice/audio or visual modes, multimodal

applications give them multiple options for inputting and receiving information. The

term “modality” is used to describe the different ways of operation within a computer

system, in which the same user input can produce different results depending of the

state of the computer. It also deﬁnes the mode of communication according to human

senses or type of computer input devices. In terms of human senses the categories are

sight, touch, hearing, smell, and taste. In terms of computer input devices we have

modalities that are equivalent to human senses: cameras (sight), haptic sensors (touch),

microphones (hearing), olfactory (smell), and even taste. In addition, however, there

are input devices that do not map directly to human senses: keyboard, mouse, writing

tablet, motion input (e.g. the device itself is moved for interaction), and many others.

In the ﬁeld of HCI with reference to user modeling it is important to distinguish

between the concepts of adaptive and adaptable. An interesting comparison between

these concepts are sketched in [2] and reported in Figure 1 (see also [13] for more

details).

4 Ontology and User Modeling: Scrutable Models

In the last years the concept of “scrutability” arises, connecting ontologies and user

modeling more deeply. A user modeling system is called scrutable if users can not

only examine the data in their user models, but also the processes that use that data for

personalization.

Weakness

Strengths

Application

Domains

Mechanisms

Required

Knowledge

Definition

AdaptiveAdaptable

Weakness

Strengths

Application

Domains

Mechanisms

Required

Knowledge

Definition

AdaptiveAdaptable

Fig.1. Comparison between Adaptable and Adaptive systems.

Scrutability means that the user should be able to explore the user model and model-

ing processes. It is fairly clear that the most important reason for scrutable user models

is the possibility for users to know what information a system maintains about them.

An important characteristic of this kind of user modeling is that a scrutable user

model enables the user to check and correct the model itself. This is a critical argument

for scrutability, and it relates to the previous one stating the users right to know what

the system stores about them. Often, it is simple to gather user modeling information

from the analysis of that part of the users behavior which is observable by the machine.

To understand the importance of this models feature, let us consider the following

scenario: a user allows another person to use his/her system account: the inferences are

no longer about the individual we intended to model. The models accuracy can also be

affected by people giving the user advice: someone might direct the user to type a series

of complex commands. This could result in a user model with an inaccurate picture of

the users behavior. If users can easily check the user model, they can correct it.

Making a model scrutable ensures that the user can see and control the bases of the

systems inferences about them. Anyway, there is a potential problem in allowing users

to see and alter the user model. Indeed, they might just decide to lie to the system and

create a model of themselves that they would like to be accurate: perhaps they would

like the system to show them as experts. Another typical situation is the presence of a

“curious” user who might simply like to see what happens if he/she modiﬁes his/her

user model. It is clear that there is a concrete possibility that users may corrupt their

user model, therefore it must be taken into account in designing the representation and

support tools of a system.

However this possibility of user models corruption, it is not a sufﬁcient reason to

prevent users from controlling the systems model of them.

There are, obviously, differences between machines and people. These differences

introduce the possibility for user-adapted systems to be especially effective in some in-

teractions. In particular, the processes controlling an user-adapted computer system are

normally deterministic. On the contrary, in human-to-human interaction, the processes

behind actions may not be accessible. Although the machine cannot really know the

users beliefs, it should be possible for the user to know the machines beliefs, especially

the beliefs contained in a explicit model of the user. In collaboration (or co-operative)

systems, the scrutability of the user model can play an important role in helping the

user understanding the systems goals and viewing the interaction. Moreover, we can

reduce user misunderstandings, due to expectations and goals held by the system and

not appreciated by the user.

Scrutability can be the basis for adaptable systems. Indeed, we can expect to en-

hance the quality of the collaboration if both the user and system can be aware of the

beliefs each holds about the other: an accessible user model can help the user be aware

of the machines beliefs.

In the case of teaching systems, there are additional reasons for scrutability. Indeed,

we can observe that it is possible to change the user model (usually called a student

model in this context) to become a valuable basis for devising deeper learning [9]. In

this context, the important role of a scrutable user model is to represent what the learner

knows and does not know at any stage. If the model is externalized by the system, pre-

sented in a way that enables the student to reﬂect on and evaluate their own knowledge,

the student is more motivated in developing a deeper understanding. Then the user may

be motivated to learn more. The discussion to this point has emphasized the possibility

of a teaching system making an externalized user model part of its teaching. There is

also a learner-initiated view.

It can be observed that in a not-scrutable system, the users exploration and curiosity

always corrupt the user model. This means that scrutability might enable the user to

explore the system without compromising the accuracy of the user model.

Finally, scrutability becomes especially important when user models are very large,

i.e. consisting of hundreds or even thousands of components (such as large ontologies

describing user characteristics and behavior). Inference can be something that becomes

very complex due to the number of components involved, creating considerable chal-

lenges to maintain the scrutability of the user model. There is the need to explore ways

to structure and visualize user models to make the user modeling process in these situ-

ations more explicit and understandable.

5 The Role of Ontologies in User Modeling

Ontologies, deﬁned by Gruber as “a shared conceptualization of a particular domain”

are a fundamental element of the new generation Web, the Semantic Web [3]. Although

ontologies have their foundations in the ﬁeld of artiﬁcial intelligence, they have also

attracted the attention of other ﬁelds such as knowledge management, information re-

trieval, and e-Learning.

Ontologies have also been recently applied to the ﬁeld of user modeling (as proved

by several researches [14] [4]) In these systems, ontologies provide an important link

between the domain content, user models, and adaptation.

There is also a role for ontologies in providing structure to a user model. In fact,

when the user model does not have a suitable structure to organize the components in

the model, an ontology should be extremely useful. This is true essentially for two rea-

son: ﬁrst of all, ontology can provide such a graph structure and, second, that structure

should make sense to the user, in terms of the meanings of the concepts modeled [8].

Ontologies can also be used for reasoning. Indeed, moving from the knowledge of a

subset of facts within the domain, new facts can be produced by inference. We can use

inference on user models with an ontological structure to ﬁnd out new facts about the

user.

As stated in the previous sections, user models typically consist of different compo-

nents, each representing a single piece of information or belief about the user. Although

the user model may be very large with hundreds or even thousands of components, it

is very important that the system maintains scrutability, particularly in reasoning about

the data, whether it is performed by the user modeling system or other users who might

have access to data from the user model.

Therefore, as previously stated in Section 4 it would be greatly useful for users to

be able to explore possible inferences that can be made about them by an agent from

their user model or better a partial version of this model. This is important because even

though the partial user model may only contain a subset of data, many inferences can

be made about the user on the basis of an underlying ontology.

This is true especially in case a user wants to keep private some elements of his

model (e.g. personal details and contact information). A system can still provide an

adequate picture of the user by making inference, even with partial information about

the user himself. A similar situation is sketched in Figure 2. In this case, the user has

chosen only a portion of his model to be shared (top right of the ﬁgure). On the other

side the system may allow the user to visualize possible inference that can be made

about him from this information (bottom right of the ﬁgure).

System OntologySystem Ontology

User ModelUser Model

User DataUser Data

User’s Public ModelUser’s Public Model

User ModelUser Model

Fig.2. An illustration of a user model.

Ontological inference is particularly important in environments where evidence

feeds into the user model at very different granularities. A typical example can be

found in the e-Learning systems, where course materials typically teach ﬁne-grained

concepts that contribute to higher level learning goals. It is important to be able to

model both coarse and ﬁne grain concepts in the learner model, so that learners can see

their overall learning progress and goals through the coarse grained concepts, and also

determine what elements of work contribute to these high level goals through the ﬁne

grained concepts [12].

In this way, we can exploit ontologies for both scrutability of the user model and also

reasoning about users. Using ontologies for scrutable modeling raises several issues to

be considered (as recently observed [11]):

– Construction. Creating ontologies is not a trivial activity. In fact, we require a

source of knowledge of the domain we are trying to model (either through do-

main experts or documentation). By this process we can identify the concepts and

relationships that make up the ontology, and a way to represent the ontology.

– Enhancement. When we create an ontology, it may not contain all the concepts and

relationships necessary for the the particular objective. It is imperative that ontolo-

gies were easily and quickly updatable; this, however, introduces the challenge of

being able to update (enhance) the ontology without compromising its integrity.

– Interface. An effective interface for viewing and exploring the ontology is obvi-

ously required. Moreover we need convenient tools for aforementioned construc-

tion and enhancement processes.

6 Conclusions

As a demonstration of the growing interest in ontology-based user modeling, a lot of

researches and tools have been developed in the last years. Therefore, User Modeling

still represents a key component for a large number of personalization systems. In this

paper, after a review of the main concepts of user modeling, we have discussed about

the importance of ontologies in building user models, reasoning about them and ﬁnally

supporting scrutable user modeling system.

References

1. D. Benyon, J. Kay, and R. C. Thomas. Building user models of editor usage. In E. Andre,

R. Cohen, W. Graf, B. Kass, C. Paris, and W. Wahlster, editors, Third International Workshop

on User Modeling - UM92 (Proceedings), pages 113–132. IBFI (International Conference

and Research Center for Computer Science), 1992.

2. Gerhard Fischer. User modeling in human–computer interaction. User Modeling and User-

Adapted Interaction, 11(1-2):65–86, 2001.

3. Thomas R. Gruber. Toward Principles for the Design of Ontologies used for Knowledge

Sharing. Formal Ontology in Conceptual Analysis and Knowledge Representation, 1993

4. Dominik Heckmann, Eric Schwarzkopf, Junichiro Mori, Dietmar Dengler and Alexander

Kr¨oner. The User Model and Context Ontology GUMO Revisited for Future Web 2.0 Ex-

tensions. In Proceedings of the International Workshop on Contexts and Ontologies: Repre-

sentation and Reasoning (C&O:RR), Denmark, 2007.

5. Anthony Jameson. Impression monitoring in evaluation-oriented dialog. the role of the lis-

tener’s assumed expectations and values in the generation of informative statements. In

Proceedings of IJCAI’83, Karlsruhe, 1983. International Joint Conferences on Artiﬁcial In-

telligence.

6. Anthony Jameson. User-Adaptive Systems: An Integrative Overview. Tutorial presented at

UM99, International Conference on User Modeling, Banff, Canada, 1999.

7. Aleksandar Kavcic. The role of user models in adaptive hypermedia systems. In Electrotech-

nical Conference, 2000. MELECON 2000. 10th Mediterranean, vol. 1. 2000.

8. Judy Kay and Andrew Lum. Exploiting Readily Available Web Data for Scrutable Student

Models. In Proc. Proceedings of the 12th International Conference on Artiﬁcial Intelligence

in Education, AIED 2005, pages 763–765, Taiwan, 2006.

9. Judy Kay and Lichao Li. Scrutable Learner Modelling and Learner Reﬂection in Student

Self-assessment. In Intelligent Tutoring Systems. ITS 2006, pages 338–345, Amsterdam,

The Netherlands, 2005.

10. Muneo Kitajima and Peter G. Polson. A computational model of skilled use of a graphical

user interface. In CHI ’92: Proceedings of the SIGCHI conference on Human factors in

computing systems, pages 241–249, New York, NY, USA, 1992. ACM.

11. Andrew W K Lum. Light-weight Ontologies for Scrutable User Modelling. PhD thesis,

2007.

12. G. McCalla and J. Greer. Granularity-Based Reasoning and Belief Revision in Student Mod-

els. In Student Modelling: The Key to Individualized Knowledge-Based Instruction., Berlin

Heidelberg Springer-Verlag. 25, pages 39–62, 1994.

13. Andreas Nrnberger and Tanja Falkowski. Adaptive user support in information retrieval

systems. In Information-Mining und Wissensmanagement in Wissenschaft und Wirtschaft,

pages 111–125, 2004.

14. Liana Razmerita and Miltiadis D. Lytras. Ontology-Based User Modelling Personalization:

Analyzing the Requirements of a Semantic Learning Portal. In Emerging Technologies and

Information Systems for the Knowledge Society, First World Summit on the Knowledge Soci-

ety, WSKS 2008, vol.1, pages 354–363, Athens, Greece, 2008.

15. Elaine Rich. Stereotypes and user modeling. In A. Kobsa and W. Wahlster, editors, User

Models in Dialog Systems, pages 35–51. Springer, Berlin, Heidelberg, 1989.

16. Elaine Rich. Users are individuals: individualizing user models. International Journal of

Man-Machine Studies, 18:199–214, 1983.

17. Elaine Rich. User modeling via stereotypes. pages 329–342, 1998.

18. Cristian Rusu, Virginia Rusu and Silvana Roncagliolo. Usability Practice: The Appealing

Way to HCI. In Proceedings of First International Conference on Advances in Computer-

Human Interaction, ACHI 2008, pages 265–270, Sainte Luce, Martinique, France.

19. K. Sparck Jones. Realism about user modeling. In A. Kobsa and W. Wahlster, editors, User

Models in Dialog Systems, pages 341–363. Springer, Berlin, Heidelberg, 1989.

20. Etienne Wenger. Artiﬁcial intelligence and tutoring systems: computational and cognitive

approaches to the communication of knowledge. Morgan Kaufmann Publishers Inc., San

Francisco, CA, USA, 1987.