multimedia @ VU
In this paper we present a first version of
a recommender system in the domain of cultural
heritage based on semantic web technology.
Our recommender system gives advice on categories and topics
of interest, based on the users' judgement,
indicating like or dislike,
of a small sample of artefacts from the collection of the museum.
A first user study indicates that our recommender system
actually helps novice users, having little knowledge
of the history of art or the collection of artefacts in the museum,
to select categories and topics
of interest from the otherwise confusing catalogue of the
Based on the apparently promising results of our user study,
which included the assessment of a personal profile, as well
as an evaluation of the usability of our prototype,
we will further explore the development of
a personalized recommender system
based on semantic web technology,
following well-established methods of user-centered design.
In the last years, dedicated recommender systems have gained
and became more and more
established practice in online commerce,
for example for the purchase of books and DVDs.
In our project, CHIP, we
aim to develop recommender systems in a more generic way,
based on semantic web technology,
to allow for the disclosure of rich collections
of information in the public and cultural domain,
aiding the user in navigation and interaction.
In this paper, we present a user study on a first prototype
embodying our approach, to gain insight
in the way users understand the purpose of such systems
and how they may profit from them.
The case study concerns a recommender system
that proposes categories and topics of interest
based on a preference rating of a small sample of artefacts
from a museum collection, in our case
the Rijksmuseum, Amsterdam.
The actual user study consists of four questionaires
and one trial with our system,
resulting in a list of potentially interesting
categories and topics.
Two questionaires asked the user
explicitly to indicate his/her interest in particular
categories and topics,
one taken before and the other after the trial.
In addition, one questionaire was meant to determine
the level of expertise of the user, that is his/her
familiarity with art, the history of art, and the collection
of the museum, and one questionaire, taken at the end of the session,
was used to evaluate the overall usability of our system.
In this way we wished to obtain feedback on
our approach, that is more specifically, whether we could realize
a recommender system in the domain of cultural heritage,
using generic semantic web technology,
to give access to the collection of artefacts of the museum.
As such, if our assumptions hold, our prototype
may be considered a showcase for a new approach to recommender
systems, using emerging semantic web standards for interesting
ways of feedback and user guidance.
Our actual system is meant to be the first in a sequence,
to be followed by a system that generates personalized
guided tours, based on a user profile generated
from a preference rating of a small sample of artefacts or examples.
In order to obtain real information we decided
to perform the user study with real visitors
from the Rijksmuseum, selected randomly at the entrance
of the museum, instead of a potentially more clean
study in an academic environment, using students as participants.
Our working hypothesis in this study was that novice
users would profit from our recommender system, whereas
for expert users, given the simplicity of this prototype,
it would not have much to offer.
To test this hypothesis, we compared the results
of the questionaire taken before the trial with the results
of the questionaire about categories and topics taken after the
or more precisely, we compared the respective relationships
between the outcome of the trial
and the results of the questionaires.
Our results look promising, confirming the trend
we hoped for, in particular when we define
level of expertise on a sufficiently wide range
from absolute novice to established expert.
Equally important, though, are perhaps the findings
we obtained from the usability evaluation questionaire,
which gives clear indications what factors
do influence the acceptance of the recommender system,
and what aspects may confuse the user.
These insights will govern our subsequent design
of recommender systems, which will offer more functionality,
following well-established practices of user-centered design.
The structure of this paper is as follows ...