Non-Intrusive User Modeling for a Multimedia Museum Visitors Guide by guy24

VIEWS: 8 PAGES: 6

									            Non-Intrusive User Modeling for a Multimedia
                  Museum Visitors Guide System

              Tsvi Kuflik, Charles Callaway, Dina Goren-Bar, Cesare Rocchi,
                          Oliviero Stock and Massimo Zancanaro

                        ITC-irst, via Sommarive 18, 38050 Povo, Italy
          {kuflik,callaway,gorenbar,rocchi,stock,zancana}@itc.it



         Abstract. A personalized multimedia museum visitor's guide system may be a
         valuable tool for improving user satisfaction in a museum visit. Personalization
         poses challenges to user modeling in the museum environment, especially when
         several different applications are supported by the same user model, where it is
         required to operate in a non-intrusive manner. This work presents the PEACH
         experience of non-intrusive user modeling supporting online dynamic multime-
         dia presentation production and additional applications such as visit summary
         report generation.




1 Introduction

A museum visit is a personal experience encompassing both cognitive aspects (such as
the elaboration of background and new knowledge) and emotional aspects, which may
include the satisfaction of interests or the fascination with the exhibit itself. Despite
the inherently stimulating environment they create, cultural heritage institutions often
fall short of successfully supporting conceptual learning, inquiry-skill-building, analyt-
ic experiences or follow-up activities at home or at school [7]. The value of multime-
dia for a museum mobile guide is discussed in [4] with an extended user study con-
ducted at Modern Tate in 2002. Yet the optimal multimedia tourist guide should sup-
port strong personalization of all the information provided in a museum, in an effort to
ensure that each visitor can accommodate and interpret the visit according to his or
her own pace and interests. Simultaneously, a museum guide should also provide the
appropriate drive to foster learning and self-development so as to create a richer and
more meaningful experience.
   In the context of the PEACH1 project, we are building and evaluating a number of
prototypes aimed at providing the visitor with a personalized experience. Common to
all these prototypes is a user model that gathers information about the visitor and
guides the adaptation of information presented to the user.
The PEACH museum visitors' guide consists of a Dynamic Presentation Composer
that generates personalized presentations seen by the user, detailed in [6] and currently

1   http://peach.itc.it. The PEACH project is funded by the Autonomous Province of Trento.
a Report Generator that generates a personalized visit summary for the visitor, as
detailed in [1]. Both components employ a common Domain Knowledge Base (KB)
and are supported by a Dynamic User Modeler (UM) for generating personal presen-
tations for the visitor.
    There are two unique challenging aspects for personalization in the context of the
PEACH project. The first challenge is that user modeling is required to be "non-
intrusive"; hence visitors are not required to provide any personal information and
user modeling is based solely on users' behavior. The second challenge is that user
modeling component needs to support different applications, with different require-
ments: the main one is online production of personalized presentations delivered to
the visitor during the visit; another is supporting a personalized visit summary report.



2 User Modeling Challenges in the PEACH Scenario

Non-Intrusive User Modeling means that the model is built solely by observing the
visitor’s behavior. The information that is available for modeling includes the se-
quence of visitor's positions (exhibits) and time spent at each position, presentations
presented to the user, and an enjoyment feedback from the user (the user is able to
respond to and rate the presentations delivered). This information then drives the
inference mechanism for assessing user interests.
   Dynamic Presentation Generation requires a lot of personal and contextual infor-
mation: for example, spatial information (current user position and whether the user
has already been here, is in front of an artwork, or just near it), visitor interests with
respect to the current exhibits, discourse history (what particular presentations were
delivered to the visitor)
   These attributes can be used in set of rules guiding the dynamic generation of per-
sonalized presentation for the user during her visits, as presented in [6]. In addition to
specific details regarding the current visit, which consist of the visitor's path through
the museum, presentations delivered and visitor's feedback, there is a need for a more
abstract representation of user interests to guide future generations of presentations.
Several works have dealt with adaptable guides. For instance in HIPS [4] work on
adaptation has included the classification of users’ patterns of movements in the
course of the visit. Here the situation is different: our UM must support dynamic
multimedia, including video generation, seamless presentations on mobile and statio-
nary devices [5] and additional applications as discussed below.
Visit Summary Report Generation requires the consideration of various different
aspects; factual aspects of the visit (such as exhibits visited, the visit sequence, the
time spent at different locations, the presentations delivered to the visitor, and visitor's
actions), cognitive aspects related to the exhibitions (such as interest for themes, plea-
sure, boredom etc), extra subject-centered aspects2 (such as persons met, discussions
held and additional events that occurred), attention-grabbing elements and hints for
subsequent reading and visiting and the appearance of the report (combining text,


2   at least in principle, not implemented so far
images and possibly additional forms of media, in a personalized manner either on
paper or in an electronic form)
   The quality of the report is crucial: it should be a memory aid for later consultation,
something one can share with others, and an entry point for getting deeper into a sub-
ject. It should be short but readable and concrete. Detailed descriptions are important,
but only when relevant to the specific visitor [1]. As such the user model that supports
report generation should provide detailed descriptions of information presented to the
user that seemed specifically interested in it.



3 User Modeling in PEACH

The PEACH Dynamic User Modeler (UM) works in a "non-intrusive" manner. Cur-
rently there is no initial information about the visitor when starting each visit, and as a
result, the model is built solely by observing the visitor’s behavior. We are studying
ways of importing and adapting pre-existing models of the specific visitor obtained
from other applications.
   The user is tracked during the visit by recording the visitor's positions (in terms of
the visited exhibits) and the time spent at each position are recorded by the UM, as
well as the presentation delivered. User interests are defined in terms of domain con-
cepts, which are associated with individual presentations. These concepts provide a
description of the content of the presentation, thus representing its theme. The con-
cepts are drawn from a domain knowledge-base that is primarily designed for natural
language generation for visit summary reports. Since there is no prior knowledge
about the user, the knowledge base and the concepts associated with the individual
presentation are the only source of information for user preferences with respect to the
exhibits visited and presentations delivered in the current museum visit.
   In addition to the various events recorded, the UM also contains inferred informa-
tion about the level of interest the visitor has in the concepts associated with the pres-
entations delivered to her. In addition to the specific concepts associated with the
presentation, an inference mechanism that follows ontological links in the KB, from
the specific concepts associated with the presentations to related concepts augments
the user model with additional, more abstract concepts extending the UM to categories
of concepts beyond those that were associated with the presentation seen by the visi-
tor. For example, if a presentation that seems to be of interest to the visitor is
represented by the concept "knight", this concept is added to the UM, with an initial
value, the interest in "knight" is now propagated to the more abstract concept "aristo-
cracy", which is added to the UM.
   The visitor's interests in the various concepts are defined in a 5-level scale. Explicit
and implicit visitor's feedbacks are used to infer user interest in the various concepts
associated with the presentations delivered to the visitor. Explicit user feedback is in
the form of pressing a “More” button (for positive reaction) or an “Enough” button
(for negative reaction). Implicit positive feedback is the completion of a presentation
delivery to the user, without objection (e.g. no "Enough" button pressed or position
changed). Explicit feedback has a higher priority than implicit feedback in the sense
that explicit feedback is more reliable so it drives an immediate change in level of
interest in the concepts associated with the delivered presentation, while implicit feed-
back requires accumulation of evidence for every concept (several implicit responses)
before changing a visitor’s interest level in that given concept. Several implicit res-
ponses are required for updating a level of interest. Whenever a new concept is added
to the list of interests, it gets a neutral value – "interested a little". As mentioned
above, the interest level is propagated to a more abstract concept related to them,
following ontological links among the concepts as represented in the system KB. The
level of interest associated with the concepts that are related to the original concepts
decays as a function of the distance from the initial concept.
   The information stored in the UM includes both recorded information of all events
that happened during the visit and inferred information regarding level of interest is
dynamically updated and used during the visit to help prepare the presentation for the
visitor. This is done by tailoring the presentations to the current visit context – what
the user has already seen, visitor's current location and specific interests. Finally, the
UM drives generation of a visit summary report. This report includes details about the
visit, and suggestions for future activities for future visits to this and other museums



4 Planned Evaluations

We are performing several user studies aimed at evaluating the PEACH guide user
interface and the adaptive report. We will address in short to both of them.
   1. User studies on PEACH guide user interface. In order to assess if the user
perceive the adaptive dimensions depicted in the guide we conducted a simulation
study assessing four dimensions of adaptivity: location awareness, follow-ups, content
adaptation with respect to user interests, and content adaptation with respect to history
of interaction. The results of this study are reported in [2]. Currently we are assessing
real visitors in the museum using the implemented visitor guide. At the end of the visit
users are interviewed about the same four adaptivity dimensions. We resort to an ac-
tion-protocol and retrospective-interview qualitative study; in particular, we target the
expression of the affect and the delegation of control paradigm implemented that are
the main events that affect the UM component. The results of this iterative evaluation
cycle will lead to the final design of the PEACH guide interface. We expect that users
will be able to properly carry out the task with a reasonable understanding the concep-
tual model of the system. Furthermore, we expect that the interface is easy to use and
that their expectations about the interest model will be fulfilled.
   2. User studies on PEACH adaptive report. In order to assess if the user perceive
the adaptive dimensions depicted in the summary reports we are currently conducting
a simulation study. We compare three types of summary visits: Adaptive Sequential
Report, Adaptive Thematic Report and a Non Adaptive Generic Report that differ on
the following adaptivity dimensions: (1) Sequential vs. thematic (2) Personalized vs.
generic (3) Reference to related topics in unseen frescos (4) Reference to related top-
ics in other museums and sites (5) Reference to the most interesting scene (6) Compar-
ison between topics within and between frescos. At the end of the visit using the
PEACH multimedia guide the experimenter give the three simulated summary visit
reports in different order and let the users read them all. Then the visitors are inter-
viewed in order to assess all six dimensions. The preliminary results indicate that
visitors perceive the differences between the sequential and the thematic reports; they
like more the personalized over the non personalized reports, rather because of the
personal reference present in the adaptive versions than for the adaptive contents of
them. The full results of the current user study will be reported in the future.



5 Conclusions and Future Work

   This paper presented user modeling challenges in the PEACH scenario, where
"non-intrusive" user modeling is required to support personalization of dynamic pres-
entation generation and other applications, including summary visit reports for mu-
seum visitors. Concepts drawn from a domain KB, used primarily for natural language
generation are associated with presentations delivered to the visitor and used to
represent visitor's interests. User behavior is used to determine the level of interest the
visitor has in the various concepts. In the next phase we are going to apply different
user modeling techniques working in parallel (as competing user modeling agents).
We are also studying ways of importing and adapting pre-existing models of the spe-
cific visitor obtained from other applications. Finally, research will focus on evalua-
tion of the user modeling activities as part of the whole system evaluation.



References

1. Callaway C., Kuflik T., Not E., Novello A., Stock O. and Zancanaro M. Personal
  Reporting of a Museum Visit as an Entrypoint to Future Cultural Experience. In
  Proceedings of Intelligent User Interfaces IUI’05,San Diego,275-277, January 2005
2. Goren-Bar D., Graziola I., Pianesi F. and Zancanaro M. Dimensions of Adaptivity
  in Mobile Systems: Personality and People’s Attitudes. In Proceedings of Intelligent
  User Interfaces IUI’05. San Diego,223-230, January 2005
3. Marti P., Rizzo A., Petroni L., Tozzi G., and Diligenti M. Adapting the Museum: a
  Nonintrusive User Modeling Approach. Proceedings of the Seventh International
  Conference on User Modeling (UM99), 311-313, Banff, 1999
4. Proctor N. and Tellis C. The State of the Art in Museum Handhelds in Proceedings
  of Museums and the Web Conference, 227-237, 2003
5. Rocchi C., Stock O., Zancanaro M., Kruppa M. and Krueger A. ‘The Museum
  Visit: Generating Seamless Personalized Presentations on Multiple Devices’ Pro-
  ceedings of Intelligent User Interfaces IUI’04, 316-318, Madeira, 2004.
6. Rocchi C. and Zancanaro M. Rhetorical Patterns for Adaptive Video Documenta-
  ries. In Proceedings of Adaptive Hypermedia and Adaptive Web-Based Systems AH-
  2004, Eindhoven, 324-327, 2004.
7. Semper R. and Spasojevic M. The Electronic Guidebook: Using Portable Devices
  and a Wireless Web-Based Network to Extend the Museum Experience. In Proceed-
  ings of Museums and the Web Conference, 18-20, Boston, 2002

								
To top