Augmen

ting a Museum Visitor’s Tour with

a Context Aware Framework

Andry Rakotonirainy

and Nicholas Lehman

Centre for Accident Research and Road Safety - Queensland (CARRS-Q)

Queensland University of Technology 4034

School of Information Technology and Electrical Engineering

University of Queensland 4072,

Australia

Abstract. We present a context awareness framework to assist a Mu-

seum visitor’s learning process by providing an enriched tour exp erience.

We adapt the information a visitor receives about an artwork based on

their interests and knowledge. We show how this is achievable by imple-

menting a system that passively and unobtrusively gathers information

using physical and virtual context. We also allow the visitor to visit vir-

tually museums in diﬀerent places while still beneﬁting from information

adapted to their interests and knowledge.

1 Introduction

Museums give people the chance to look at artworks of many diﬀerent forms,

from prehistoric dinosaur bones to modern abstract sculptures, with the advent

of the ”digital tourist,” [1–3, 5] museum visitors can look at an artwork and

instantly have an abundance of information at their ﬁngertips. More often than

not this information will not change, additions can be made by museum staﬀ,

web-site administrators and in some cases other museum visitors [3, 5, 6] but

the content displayed will still be the same for each visitor even though their

level of knowledge and degree of interest are diﬀerent. This paper presents a way

to adapt the information received about an artwork based on what the visitor

knows and where the visitor’s interests lie. It describes a framework supporting

diverse information requirements of museum visitors potentially ranging from

children to senior researchers, and so to that end we present a way of adapting

the information received about an artwork based on the visitor’s interests and

knowledge. The framework provides functionalities described in the following

scenario.

2 Museum visitor’s tour scenario

Upon entering a museum the visitor logs onto the museum network and their

PDA downloads the physical layout of the museum. The visitor’s itinerary and

calendar for the museum are updated and the visitor is also notiﬁed of any events

the system deems they would be interested in.

Rakotonirainy A. and Lehman N. (2004).

Augmenting a Museum Visitor’s Tour with a Context Aware Framework.

In Proceedings of the 1st International Workshop on Ubiquitous Computing, pages 104-112

DOI: 10.5220/0002669401040112

 SciTePress

A list of tour options is provided for the visitor, these tours are based on

the visitor’s interests, the visitor may select a tour or opt to proceed on their

own. If a tour is selected an interactive map shows the visitor where to go, a

digital assistant also provides directions if needed. The visitor can click on rooms

displayed on the map to learn about the room’s contents. During the tour the

visitor can point their mobile device at a piece of artwork, this causes a browser

window to open containing a small replica of the artwork. These replicas can

be marked for printing as posters or postcards; the visitor picks up and pays

for these after the tour. Textual and audio information regarding the artwork is

provided in addition to links to similar artworks.

The visitor can choose to view additional information based on their proﬁle.

Their level of knowledge and degree of interest is taken into account when deter-

mining what extra information is displayed, for example, a visitor who considers

themselves unversed in a particular form of art would receive information about

the artist, the artwork’s history and a general overview of similar artworks.

A user who considers themselves well versed in the same form of art would

receive a detailed description of what other visitors of similar interest have no-

ticed about the art form. Each of these techniques aims to enhance the visitor’s

experience by selectively adapting information to suit their proﬁle. When the

visitor has ﬁnished viewing a piece of art, the list of artworks to see in the vis-

itor’s tour is updated. The visitor can also leave comments about the artwork

for future visitors as well as staﬀ to read.

During the visit the visitor may deviate from the tour to look at other art-

works, after examining similar artworks the digital assistant informs the user

where there are similar artworks that previous visitors with similar interest and

knowledge also found interesting. The visitor is given the option of reading why

the previous visitors found the other forms of art interesting. The visitor can

opt to proceed with the tour, keep browsing or look at the other forms of art

the system has nominated.

When closing time is drawing near the digital assistant unobtrusively informs

the user that time is running short, this allows the visitor to prioritize the re-

mainder of their tour. After the visitor has ﬁnished their tour they can pickup

any replicas marked for printing, leave comments for future visitors to peruse,

they then log out.

3 Information model

Before gathering contextual information to deduce a visitor’s interests and knowl-

edge we establish a framework that addresses the way artworks and the infor-

mation about those artworks are classiﬁed.

3.1 Classifying Artworks’ information

Classifying artworks is a challenging problem, too little information and we can-

not deduce co-locality or commonality across a set of artworks, too much infor-

mation and a visitor gains no beneﬁt from the service as seemingly unrelated

105

artworks overlap. We use a hierarchical approach. Storing the information hi-

erarchically provides us with a great degree of versatility, as visitors can be

interested in artwork aspects that are both broadly and narrowly deﬁned.

Information about artworks takes on many forms of varying complexity. The

museum is responsible for classifying the information held about each artwork,

by ensuring an accurate mapping between a visitor’s proﬁle and an artwork’s

subjects we can take advantage of the semantic layer’s grouping facilities to

deduce a visitor’s interests.

By classifying the information about the aspects of an artwork, we can ac-

curately determine what type of information the visitor is looking for and what

types of information they have accessed.

3.2 Classifying Interests

A visitor’s interests must match the allowable artwork information classiﬁcations,

the developed prototype updates a visitor’s proﬁle based on the classiﬁcations

of the artworks seen, section 6 details how the visitor’s interest and knowledge

vectors are updated.

Upon entering the Museum, the user ﬁll a registration form to explicitly state

their interests and self-assessed knowledge, the form is created by the museum

and is displayed as a tree-like structure, the visitor is free to select interests as

broad or speciﬁc as they desire. The level of interest will be updated according

to the visitor’s behavior.

3.3 Classifying Knowledge

We classify the areas of interest about an artwork with the preferable level

of knowledge or understanding required to gain beneﬁt from the information.

Figure 1 shows a knowledge tree as a means of representing diﬀerent level of

knowledge which are basic, medium and advanced.

Basic

knowledge

Advanced

knowledge

Medium

Fig. 1. A knowledge tree

We make use of this information hierarchy to select information that is only

relevant to the current visitor and to prune information that the visitor would

ﬁnd trivial or nonsensical. The tree is traversed when information is requested,

106

the search algorithm takes each of the visitor’s interest and knowledge vectors as

inputs. For each category within the artwork’s proﬁle a depth ﬁrst search is per-

formed, each level of the tree contains more specialized or detailed information

than the previous layer.

4 Conceptual Design

As a way of separating concerns and supporting the sequence of events described

in Section 3.3, it was decided to adopt a 3-layered. The three layers are called

physical, virtual and semantic layer as described in Figure 2.

Fig. 2. The 3-layered approach

Figure 2 shows that the three artworks share the same topic. If a visitor

were interested in two of the artworks then they would probably be interested in

the third because we can infer that they are interested in the topic the artworks

depict. We can determine what a visitor is truly interested in by their interest

vectors, since each vector is updated when the visitor accesses information or

looks at an artwork the subject of most interest to the visitor will have the

highest vector as it is common across all instances.

4.1 Physical Layer

The physical layer is responsible for keeping track of the objects within a museum

and encompasses the physical layout of a museum including all the rooms and

facilities. The primary task of the physical layer is to store the physical location

of each artwork.

The physical layer is usually static, museums typically organize artworks

according to a ﬁxed criteria, for example, paintings may be located in a single

section of the museum and arranged in chronological order.

In our model each artwork is aﬃxed with an infrared transmitter that contin-

uously emits a unique identiﬁer, this identiﬁer correlates the artwork’s physical

location with its corresponding location in virtual space. Section 6 describes

the mechanisms for gathering and processing physical contextual information to

deduce a visitor’s interests.

107

4.2 Virtual Layer

The virtual layer is responsible for describing the physical layer; it contains

information about the aspects of an artwork that can include the artist, medium,

description and links to additional information about the artwork.

The virtual layer changes as information is added remove or altered by mu-

seum staﬀ, web-site administrators and other visitors. The pages accessed by

the visitor are dynamically created according to the visitor’s knowledge and

interests.

4.3 Semantic Layer

The semantic layer describes the relationship b etween artworks and is responsi-

ble for mapping the correlation between what a visitor sees in the virtual and

physical layers and what their actual interests and knowledge are. The seman-

tic layer can logically be thought of as all the ways to group artworks by their

aspects.

The semantic layer works by categorising groups of artworks. Artworks within

the same group are said to be semantically co-located. An artwork can be se-

mantically co-located with many artworks for example, Picasso paintings are

semantically co-located as they share the same artist while a subset of these

artworks are semantically co-located with a subset of Braque paintings as they

share the same type - cubism.

The semantic layer facilitates the processing of context gathered by the physi-

cal and virtual layers by providing an easy way to determine how a set of artworks

or artwork information are related.

5 Participating Entities and Roles

Our design encompasses the roles performed by three distinct entities, the visi-

tor’s PDA, the museum and the global repository.

The visitor’s PDA (equipped with 802.11 and IrDA) is responsible for keep-

ing track of the visitor’s knowledge and interest vectors. When a PDA is pointed

at certain artworks, the infrared transmitter send a URL (website address) to

the device, causing a browser to appear. This allows the user to select from a

range of options which might include: biographical information about the artist;

a description of the process or motivation behind the artwork; or suggestions

for where the visitor might like to go next. Additionally, by allowing visitor’s

to explicitly state their interests and self-assessed knowledge we can adequately

cater for visitors wishing to expand on existing interests as well as visitors wish-

ing to learn about a particular topic, this provides the visitor with an enriched

museum tour experience. Visitor’s who feel they are suﬃciently knowledgeable

about certain areas gain beneﬁt faster and we overcome any frustration caused

by an enforced learning curve.

The museum is concerned with the information management of the physi-

cal and virtual layers as well as the implementation of the semantic layer. The

108

museum database should at all times maintain a consistent and accurate rep-

resentation of the museum and each artwork located within the museum. The

types of structures used to store the museum’s physical representation are outside

the scope of this paper. The museum is responsible for retrieving and display-

ing adaptable information about an artwork to the visitor.The museum acts as a

’middleware’ layer for the visitor. The museum periodically sends updates to the

global repository containing each visitor’s proﬁle as well as the artworks and in-

formation they found interesting. This allows museum visitors around the world

to beneﬁt from the experiences of anyone with similar interests and knowledge

instead of only local museum visitors.

The global repository is concerned with the information management of

the semantic layer and contains visitor proﬁles together with information about

what artworks they found interesting, information accessed and any additional

comments made. The repository (multi-database) makes it possible to compare

visitor proﬁles across museums. It accepts queries from participant museums.

These queries are either search requests that return the artworks seen by visitors

with similar interests and knowledge or updates to the stored data.

(location)

Multimodal Interaction

to User

Elvinhttptcp/ipIrDa

Context

Manager

IrDA

IRX2.2

Serial cable (IrDa)

Protocols

Communication

Internet

Stationary Server

802.11

(Adapted

Information)

Ethernet

Adaptation

Manager

Notes

Profiles

Users

Database

Museum

Fig. 3. Context Aware Architecture

6 Context Gathering

Our architecture is shown in Figure 3. The central piece of the architecture is

the context and the adaptation managers located in the stationary network as a

server. The context manager gathers and stores all information required for the

adaption manager to customize the information to be delivered to the visitors.

109

6.1 Catering for four types of Visitor

Our system supports a diverse range of visitors. Visitors can have many or few

interests, some are more knowledgeable than others. Table 1 summarizes the 4

types of visitors.

Table 1. Four Types of Museum Visitor

Low Interest High Interest

Low Knowl-

edge

Visitors not wanting to learn

or have not heard of the

topic.

Visitor wanting to learn and

explore their interests to

develop their knowledge.

High

Knowledge

Visitors with changed

(diminished) interests.

Researchers within a specific

field.

6.2 Gathering Context from the Physical Layer

In our prototype each artwork is aﬃxed with an IRX2.2 infrared transmitter,

the transmitter continuously emits an identiﬁer that is received by the infrared

port of a PDA. Since the interval b etween successive transmissions is a known

constant we can know the amount of time a visitor has been pointing their device

at an artwork.

6.3 Gathering Context from the Virtual Layer

Context gathered from the virtual layer represents a more accurate measure-

ment of a visitor’s interests and knowledge since the visitor is actively seeking

additional information. By looking at the types of information a visitor accesses

we can passively infer various things about their interests and knowledge.

When a visitor accesses dynamic information we store the current time and

wait until the visitor clicks on a link or exits the page, when this occurs the stored

time is forwarded together with the visitors identiﬁer and requested action (e.g.

exit, access information, logout) to an intermediary responsible for processing

the virtual context, the algorithm used to process this information is provides

in the following sections.

7 Context Processing

By utilizing the contextual information gathered from the physical and virtual

layers described in section 5 we now present the process used to determine a

visitor’s interests and knowledge.

At the physical layer,we can measure the amount of time a visitor points

their PDA at a particular artwork by counting the number of times the artwork’s

identiﬁer is received. When the PDA stops receiving the identiﬁer we retrieve

the subjects associated with the artwork from the museum’s database.

110

At the virtual layer, when a visitor accesses additional information about

an artwork the information displayed is adapted to their proﬁle of interests and

knowledge, as a consequence we receive positive reinforcement of the visitor’s

interests.

This paper emphasizes using virtual context to deduce a visitor’s interests,

the algorithm used to update the visitor’s interest vectors are weighted more

heavily than those used when gathering physical context, this reinforces the

theme that the quantity and quality of information a visitor accesses is a better

indication of a visitor’s interests and knowledge than physical context gathering.

By examining the knowledge classiﬁcations of the information accessed we

can roughly determine the amount of knowledge a visitor has about a particular

subject. Then we adapt the next set of information to build on the concepts

the visitor has gained by classifying the knowledge content of the information

displayed. We assume that a visitor accessing a dynamic page for seven seconds

or more gains the relevant knowledge content and as such their knowledge vectors

are updated to reﬂect this, we set the visitor’s knowledge vector to the threshold

value of the next knowledge classiﬁcation.

The original prototype determined the number of seconds a visitor accessed

information about an artwork and added this numb er to their proﬁle for each

relevant subject, this method was found to produce erroneous results as a visitor

reading a single page for several minutes would have the same interest vector as

a visitor reading many similar pages for a few seconds. To avoid this anomaly

it was decided to classify interest periods, for instance, it might take a visitor

three seconds to decide if they are interested in the accessed information, a

subsequent thirty seconds will determine if they are keen on learning more and a

further two minutes might elapse before we know a visitor enjoys reading about

a particular topic. Using threshold values based on a time/interest curve we get

a more accurate heuristic.

This value also enables us to determine when a visitor is no longer interested

in a subject. Our basic premise is that when people are interested in something

they tend to seek additional information about it, similarly when people lose

interest in something they generally do not continue to actively increase their

knowledge about the subject or do not tend to perform the activity any more.

By maintaining a list of the artworks a visitor has seen we can make further

inferences about a visitor’s interests when they refer back to a piece of artwork

or information. We keep a count of the number of times a visitor reviews an

artwork or information. At present this context is only used to create a list of

artworks and information the visitor is particularly interested in, the prototype

records the location of any artworks or information a visitor has visited ﬁve times

or more.

8 Conclusion

This paper presents a framework that adapts the information a visitor receives

about an artwork based on their interests and knowledge. We addressed the

111

issues involved with the gathering and subsequence processing of contextual in-

formation to deduce a visitor’s interests and knowledge, we presented our design

and discussed its limitations. The system unobtrusively gathers information us-

ing physical and virtual context, the goal to provide a passive framework has

been achieved, after the visitor has registered they are free to browse normally

while still gaining beneﬁt from the service.

The ﬁndings of this paper are intended to be used in conjunction with pre-

existing museum tour guides. Whereas the majority of the research papers ex-

amined the physical context of a portable digital assistant (PDA) this paper

employs a context-gathering framework inspired by [7], we explore the concept

of examining the virtual context or the information accessed by the visitor in ad-

dition to the physical context to determine their interests and level of knowledge.

We currently evaluate the usability aspect of the system.

References

1. Oppermann, R., Specht, M. 1999 ’Adaptive Mobile Museum Guide for Informa-

tion and Learning on Demand’ Edited by Hans-Jrg Bullinger and Jrgen Ziegler

Pro ceedings of the 8th International Conference on Human-Computer Interaction:

Munich: 642-646

2. Dix, A., Rodden, T., Davies, N., Trevor, J., Friday, A. and Palfreyman, K. 1999

’Exploiting space and location as a design framework for interactive mobile sys-

tems’ Department of Computing, Lancaster University: Lancaster

3. Pennywitt, K. and Abowd, G. ’Cyberguide Project’

http://www.cc.gatech.edu/fce/cyberguide/ April, 1997.

4. Kindberg, T., Barton, J., Morgan, J., Becker, G., Caswell, D., Debaty, P.,

Gopal, G., Frid, M., Krishnan, K., Morris, H., Schettino, J., Serra, B. and

Spasojevic, M.”People, Places, Things: Web presence for the real world”

http://cooltown.hp.com/dev/wpapers/webpres/WebPresence.asp 2001.

5. Arms, W., Lagoze, C., Frei, M., Gleason, S. and Samant, S. ’BOOM:

An Experiment in Personal Information Spaces in the Museum Context’.

http://www.cs.cornell.edu/bo om/2001/Frei/ September, 1999.

6. Burrell, J and Gay, G, ’E-graﬃti: evaluating real-world use of a context-aware

system’. http://testing.hci.cornell.edu/context/ April, 2001.

7. Gellersen, H., Schmidt, A. ’Look, Who’s Visiting? Supporting Visitor Awareness

in the Web’ TecO, University of Karlsruhe, Germany

8. Suchman, L. 1987 Plans and Situated Actions: The problem of human-machine

communication. Cambridge University Press: Cambridge.

112