Proceedings of the 2nd Workshop on Adaptive Hypertext and Hypermedia,
HYPERTEXT'98, Pittsburgh, USA, June 20-24, 1998

Building a User Model for a Museum Exploration and Information-Providing Adaptive System

Marcello Sarini and Carlo Strapparava
Istituto per la Ricerca Scientifica e Tecnologica,
I-38050 Povo/Trento, Italy
e-mail: {sarini | strappa}

Abstract: Hyperaudio is a system able to organize the presentation of a museum contents taking into account the visitor's needs and the layout of the physical space. The system is able to integrate a physical space with a virtual space in order to build a more general notion of augmented space: not only can the system provide the visitor with information tailored on his own interests and interaction history, but it can also support the visitor in his own exploration of the physical space, helping him to find what he is looking for and suggesting new interesting physical locations.
In this paper we describe the developing of the User Model components for this type of system: we found helpful to separate distinct functionalities about user modelling. These components are useful in planning the content presentations to the visitors. A presentation could take into account what the user already knows (to avoid boring repetitions, to make new things easier to understand by making comparisons, etc...), what the user is interested to (to propose new concepts and/or location to go etc...) and the interaction way the user seems to prefer.

Keywords: Adaptive System, User Modelling, Augmented Space.

1. Introduction

The ideal guided visit to a ``cultural'' space (such as an exhibition, a museum, an open-air exposition, an archaeological site and so on) allows the visitors to organize the tour through the different areas according to their own interests or preferred criteria. On the other hand, it may happen that the default physical organization of the exhibition (chosen by an architect or by a museum curator according to a ``default'' perspective of information presentation) does not meet directly the visitor's expectations, possibly making it difficult to build a personal route.
Virtual museums (implemented for example as hypertextual resources browsable from the World Wide Web) may offer a more flexible (dynamically computed) object display determined by the visitor's individual preferences. The visitors can play around a clickable representation of the museum rooms and objects, getting informative cards on what they are most interested in. However, with purely virtual spaces the visitor may perceive objects differently (e.g., different dimensions and colours) and miss the emotional involvement you get in experiencing with real objects (e.g., being in front of Mona Lisa at the Louvre is quite a different experience from looking at its reproduction on the Web).
Hyperaudio [*] is a research project for developing a museum exploration and information providing adaptive system, able to integrate a physical space with a virtual space in order to build a more general notion of augmented space: not only can the system provide the visitor with information tailored on his own interests and interaction history, but it can also support the visitor in his own exploration of the physical space, helping him to find what he is looking for and suggesting new interesting physical locations (see [Not et al. 1997]).
Each visitor is equipped with a palmtop computer endowed with headphones, on which an infrared receiver is mounted. Each meaningful physical location has a small (power-autonomous) infrared emitter, sending a code that uniquely identifies it (see Figure 1).
Exploiting the infrared signals, the system is able to identify when the visitor reaches a certain physical location and can activate a relevant portion of the information repository loaded on the palmtop. Meaningful information are selected and organized to be played as audio messages or displayed as follow-up links on the palmtop screen. Adaptive and dynamic hypertext technology (see for example [Knott et al. 1996]) can be exploited to tailor a presentation according to the visitor interests, the actual context of the visit and so on.
Fundamental in such a system are the user modelling components that take into account the user's knowledge, interests, preferences etc...In this paper we describe the peculiarities of User Model components for an adaptive system in an augmented space.


Figure 1: A visitor exploring an augmented room in a museum.

2. Different Functionalities in User Modeling

In an adaptive system, the user model represents the system's assimilation of the interaction and contains information about the user and the current context that can increase the system's ability to exhibit pragmatically correct behavior and, more generally, to engage in effective communication [Stock et al. 1995].
Ideally, presentations provided by a museum exploration system should contain the information that will be most helpful to the user. But since not all users are alike, achieving such behavior requires that the system has a model of the particular user with whom it is currently interacting. This model could include information about the user's knowledge, beliefs, abilities, attitudes, preferences and possibly goals and plans for achieving these goals.
Taking into account our experience in building multimodal natural language dialogue systems [Stock and Team1993], [Stock et al. 1995], it could be useful to separate two distinct functionalities about user modelling (what the user has been exposed to, linguistically or through images, or is assumed to know, and what the user seems to be interested in). Moreover, a third functionality could take into account user preferences with respect to presentation and interaction style. This is particularly useful in a system able to generate many alternative multimodal presentations.
This kind of information is represented in three modules (see Figure 2) and in a user profile.
The user's knowledge model, or UK, is based on an initialization (see section 4) and on a modelling of what the user has become aware of so far. User knowledge could be implemented as a set of facts, without involving advanced inference capabilities (at least for the first prototypes).
The user's interest model, or UI, provides a model of the potential interest of the user and could consist on an activation/inhibition weighted network whose nodes (the interest areas) are associated with ordered sets of entities. Each time a certain entity is ``taken into consideration'' by the user (by clicking a link while navigating in the hypertext) or is presented by the system, the areas to which that entity is associated receive an activation impulse. Impulses are propagated through the network, decreasing in intensity according to the weights of the links traversed. The interest model evolves and becomes more and more focussed during the interaction between the user and the system. The status of the interest model provides a criterion for the computation of the relevance of the exhibits with respect to the current interaction context.
The user's preferences about presentations and interaction style could be also implemented as an activation/inhibition weighted network like the user's interest model in the case it is important to have an evolving model of user's preferences. Otherwise it could be modelled as a static characteristic of the user.
These components are useful in planning the content presentations to the visitors. A presentation could take into account what the user already knows (to avoid boring repetitions, to make new things easier to understand by making comparisons [Milosavljevic1997], etc...), what the user is interested to (to propose new concepts and/or location to go etc...) and the interaction way the user seems to prefer.
A vital statistics user profile (age, education level etc...) is useful, for example, to plan more use of technical language (in the case of an expert) in the content of the presentations.

3. User Model for Museum Explorations

In the literature, there have been many empirical studies on how visitors typically behave in museums. In many cases, the substantial amount of collected results has allowed curators to identify visitors' expectations and preferences, therefore allowing the improvement of museum exhibitions (in terms of quality of layout, lighting, labels, information services, etc...).
In particular, psychological studies suggest some general statements (for a complete list see that could be useful to characterize a system for museum explorations. Some statements help identifying parameters and features that influence how the User Model evolve and could be organized.

Examples of these statements are:

Starting from the statements provided by psychologists and from the findings emerging from questionnaires, workshops and observations about users, a set of parameters and dynamic features can be identified as key elements for modelling users' behaviour.

Two major classes of parameters/features can be distinguished:

These parameters tend to be more effective in an augmented space than in a purely virtual space. It could also be possible to take into account some ``physical'' aspects like evolving time, tiredness of the visitors, and in general it is more sensible to give value to the emotional involvement in experiencing with real objects.

3.1 Parameters related to the Exhibits

Many museum studies highlight the fact that there are many ways in which layout, lighting, object dimensions, physical structure of the rooms, ... may affect the visitor's desire to approach and study an exhibit (see [Boisvert and Slez1995] for an overview on this topic). When building a portable information system for a museum (from simple tape-based audio guides, up to more sophisticated adaptive electronic guides), developers have to keep in mind that information has to be provided in the context of the exhibition ([Serrell1996]): this means that all the messages to be conveyed to the user need to take into account how the objects are placed, how each of them contributes to the overall topic of the exhibition, how they attract and keep visitors' attention. This because all the aspects of the museum experience (perceptual and cognitive) should contribute synergically to an increased enjoyment, understanding and learning on the part of the visitor. Attracting and holding power should be considered in modelling the parameters related to the objects in an exhibition ([Boisvert et al., 1995]):

The model of the exhibition is therefore a very important piece of knowledge when setting up an electronic guide. More importantly, an adaptive information system can also exploit the model of the exhibition to interpret visitors' movements to dynamically update the user model. For example, if the visitor does not stop in front of an object with a high attracting power, the system could assume that he is not interested in the particular item and could update the user model accordingly.



Figure 2: User Model Components

3.2 Parameters/Features related to the Visitor

Similarly, we try to find a significant set of parameters/features more related to the visitors.
Obviously, more features might emerge from the literature (either in the field of psychology and of user modelling) or from workshops with users and the analysis of questionnaires.

The attention is figured as soon as the visitor comes in front of the exhibit, while engagement is evaluated during and/or after the visitor is enjoying the object.

4. An example of evolving UM



Figure 3: Example of UM evolution

By means of a questionnaire the user model is initialized in its profile part (e.g. visitor's age, handicaps, etc...), which refers to user's characteristics. The questionnaire could also contain a set of questions (e.g. about competence in some sectors, particular interest in some topics etc...) that gives initialization to the three user model components (e.g. ``default'' facts in the user's knowledge, initial impulses to the interest model, initial interaction preferences).
During the visit and taking note of the user behaviour, the system will refine these initializations that could be not accurate for some aspects.
As an example of how the components of User Model evolve during a visit, consider Figure 3.
After the visitor has spent some time (now Ti) in the exhibition (eventually increasing his knowledge about displayed objects and showing interest in some of them) the visitor comes to the ``Hunting Lion'' showcase. The exhibit has a particular holding power and attractive power and has associated a ``relevant fact''. This fact is, for example, what the exhibition curator would like to communicate to the visitors.
The system proposes to the visitor a user's tailored presentation that contains a sequence of audio files which will be played on the headphones, a set of anchors for further elaborations which will be depicted on the display, and possibly a properly oriented map with the visitor's current position.
During the interaction time, the visitor is focussed on the exhibit enjoying the audio files and the navigation in the information space. The system detects the visitor engagement considering how much the user ``navigates'' to get more explanations about the object. For the moment this is the only possible way to check the engagement. More sophisticated verification could be realized providing, for example, an eye-movement tracking or other behaviour recognition systems.
Another capability the system has to provide, during the interaction time, is the tracking of the knowledge acquired by the visitor hearing the audio files. This knowledge is a priori associated to the information the user is exposed to.
At the end of the interaction (at Ti+1 time) the user interest model is updated by the system considering through an ``impulse'' function of attention, variance on object holding power and object attracting power, and visitor engagement.
The update of user knowledge is made by adding the knowledge recorded during the interaction time. (this is obviously a simplification: this doesn't guarantee that the user really understood the ``new'' knowledge)

5. Conclusion

We have described the developing of User Model for a museum exploration and information-providing adaptive system.
We found helpful to have separate modules that model different aspects about user modelling: what the user has been exposed to, linguistically or through images, or is assumed to know, what the user seems to be interested in and user preferences with respect to presentation and interaction style.
These modules are useful in planning the content presentations to the visitors to avoid boring repetition, to propose new concepts and/or location to go and considering the interaction way the user seems to prefer.
As future development, we are investigating the possibility to introduce the concept of visitor communities, built from dynamically clustering single visitor's user models. A community groups visitors with similar characteristics during their museum explorations. Information about communities could be used for planning group tailored educative strategies, for sending messages to a community about museum events etc...


[Boisvert and Slez1995] Dorothy Lozowski Boisvert and Brenda Jochums Slez. The relationship between exhibit characteristics and learning-associated behaviors in a science museum discovery space. Science Education, 79(5), 1995.

[Canestrari1994] R. Canestrari. Psicologia Generale e dello Sviluppo, volume 1. CLUEB, 1994.

[Falk1982] John H. Falk. Time and behaviour as predictors of learning. Science Education, 67, 1982.

[Falk1985] John H. Falk. Predicting visitors behaviour. Curator, 28(4), 1985.

[HIP] WWW home page for the HIPS project:

[Knott et al. 1996] Alistair Knott, Chris Mellish, Jon Oberlander, and Mick O'Donnell. Sources of flexibility in dynamic hypertext generation. In Procof the 8th International Workshop on Natural Language Generation, Herstmonceux Castle, UK, 13-15 June 1996.

[Korn1995] Randy Korn. An analysis of differences between visitors at natural history museums and science centers. Curator, 38(3), 1995.

[Milosavljevic1997] Maria Milosavljevic. Content selection in comparison generation. In Proc. of the 6th European Workshop on Natural Language Generation, Duisburg, Germany, 1997.

[Not et al. 1997] E. Not, D. Petrelli, O. Stock, C. Strapparava, and M. Zancanaro. Person-oriented guided visits in a physical museum. In Proc. of the 4th International Conference on Hypermedia and Interactivity in Museums, Louvre, Paris, September 1997.

[Serrell1996] Beverly Serrell. Exhibit Labels. An Interpretative Approach. AltaMira Press, 1996.

[Stock and Team1993] O. Stock and The ALFRESCO Project Team. ALFRESCO: Enjoying the combination of NLP and hypermedia for information exploration. In M.T. Maybury, editor, Intelligent Multimedia Interfaces. AAAIPress/MIT Press, Menlo Park CA/Cambridge MA, 1993.

[Stock et al. 1995] O. Stock, C. Strapparava, and M. Zancanaro. Explorations in a natural language multimodal information access environment. In Workshop on Intelligent Multimedia Information Retrieval, Montreal, 1995. IJCAI 95.

Hyperaudio is a project developed at IRST in collaboration with Civic Museum of Natural Science in Rovereto (Italy). Hyperaudio will contribute to more advanced research and scenarios inside HIPS European Project [HIP] whose consortium includes: University of Siena(coordinating partner), CB&J (France), GMD (Germany), IRST (Italy), SIETTE-Alcatel (Italy), SINTEF (Norway), University of Dublin and University of Edinburgh.