background image
DigiCULT 31
of ebony, such as black ebony, marblewood, or kaki (Japa-
nese ebony). The thesaurus structure specifies the pre-
cise nature of the relationship between each of these
terms, allowing a search process to expand the initial
query automatically to include closely related terms.
Having determined a degree of closeness between
thesaurus terms, we then determine a suitable result
set by comparing the expanded query with index-
ing terms. The results are displayed as a ranked list in
order of decreasing relevance to the initial query. Fig-
ure 2 illustrates the Results window where a particu-
lar result (with overall match of 56 per cent) has been
double-clicked to show the degree of match of indi-
vidual terms in the query (armchairs, brocading, mahog-
any, Edwardian). Note that no query term matched
exactly but all had partial matches to semantically
close index terms. Relevance to the searcher will
depend on context. The point is to provide a seman-
tic expansion option for the user when exact matches
are not available.
Various standalone prototype systems were devel-
oped in an iterative design and evaluation cycle.
Evaluation sessions were conducted on two major
standalone prototypes. Data gathered included tran-
scripts of think-aloud sessions, screen capture videos,
user action logs and observation notes. The sessions
involved 23 users (the vast majority being museum-
related, including cataloguers, collections management
and curators). Participants were given an introduc-
tion which walked them through a training task.
They were then asked to carry out a number of addi-
tional search tasks, corresponding to typical muse-
um enquiries. Our focus was on the process of user
interaction and searching behaviour with the system.
Evaluation of the first prototype illuminated impor-
tant design issues, including how best to support the
controlled vocabulary search process. Major enhance-
ments to the interface followed.
The standalone system focused on the collections
from the National Railway Museum,
which is part
of NMSI. The NRM Furnishings collection, which
includes objects from the `Palaces on Wheels' Roy-
al Train collection (such as Queen Victoria's saloon),
offered rich detail for indexing with the AAT. For
example, the general description of one such object
reads: `Carver chair, Oak with oval brocade seat. Prince of
Wales crest on back from Royal Saloon of 1876'. Analy-
sis of evaluation data from a second prototype is still
ongoing but has fed into the design of the Web dem-
This was one of the final outcomes of
the project and explored how the techniques from
the standalone systems could be employed with-
in a Web environment as dynamically generated
Web components. The interface does not rely on pre-
built static HTML pages; thesaurus content is gener-
ated dynamically.
We also experimented with a number of smaller
specialist thesauri, including draft versions of the
mda's Railway Object Names Thesaurus and Water-
ways Object Names Thesaurus, together with the
Alexandria Digital Library's Feature Type Thesaurus.
Through participating in the work of the Railways
Terminology Working Group during the develop-
ment of the thesaurus and in its peer review proc-
ess, we came to appreciate some of the immense work
involved in developing even a smaller thesaurus. This
includes reaching consensus
both on choice of ter-
minology and on how it should be organised in hier-
In addition to providing the AAT, the Getty Vocab-
ulary Program's
Web interface and User's Guide to
the AAT Data Releases (P. Harpring ed.) were useful
sources of ideas for developing FACET's own inter-
face, as was CHIN's experience in developing Arte-
facts Canada,
one of the first Web-based `virtual
museum' applications with an interface incorporating
a major thesaurus, such as the AAT.
Collaboration with various NMSI staff and their
experience with the collection management data-
base and public inquiries proved extremely useful
throughout the project. One motivational example for
the research was a (1997) public request to the Sci-
ence Museum for information on eighteenth centu-
ry European celestial navigation instruments. At the time,
this request highlighted some difficulties with existing
techniques since initial query terms did not generate
easy matches. Multiple queries to several fields in the
Figure 2: Example from FACET standalone system
showing a result from semantic expansion
Details of the algorithm
are given in http://www.
National Railway
Museum, http://
FACET Web interface
From the launch press
release: `The mda Railway
Terminology Working
Group has wide representa-
tion from organisations
with an interest in railway
collections and information.
The group formed in
1996 following the mda
Terminology Conference
and includes staff from
the National Railway
Museum, York; London's
Transport Museum;
Beamish, North of England
Open Air Museum;
the Historical Model
Railway Society; Heritage
Railways Association;
English Heritage, National
Monuments Record
Centre; and mda.'
Getty Vocabulary
Program, http://www.getty.
CHIN, Artefacts
Canada, http://www.chin.