Information Technology Reference
In-Depth Information
tag clouds. We have illustrated its use based on the example of browsing a pub-
lication collection. Such interfaces could support a web search service either of a
single research group's publications or over an entire digital library—simply by
adapting the specification. While there are faceted search interfaces to publica-
tion collections, such as DBLP, our approach is much more flexible, since it not
only supports searching for publications, but users can also shift their search fo-
cus to other entities of interest, such as authors or conferences.Furthermore,the
selection of the visualised entities, their relationship and alternative tag cloud vi-
sualisations are configurable based on a combination of user selection, developer
specification and default system behaviour.
Our approach is not dissimilar to the one taken by [2], where they provide
a domain-specific tool for searching semi-structured clinical trial data where a
set of predefined categories are represented using tag clouds. As with standard
faceted browsing, users can start with a keyword search and the number of rel-
evant documents are returned as a list, which can be further refined using the
tag clouds. The selection of a tag in one dimension triggers the synchronisation
of the tag clouds representing all other dimensions, as well as the filtering of
the search result. While our approach could be seen as a generalisation of their
work as we propose an augmented data model and a framework that supports
the configuration of search interfaces for a domain of choice, it is also impor-
tant to highlight the differences. Their interface consists of a set of predefined
facets represented as a tag cloud, while we offer configurability at the interface
level through dropdown menus that allow the selection of other tag cloud rep-
resentations of the same entity. Furthermore, their data model corresponds to
a typical data model underlying faceted browsing that is often based on star
or multi-dimensional schemas, while our synchronised tag clouds do not evolve
around a particular pivot entity. This means that there is no central entity that
all other dimensions depend upon. In addition, with our approach, the tag size
can be configured to represent dependencies to other entities of interest or simply
the occurrence of a specific term, while with their approach the tag size always
refers to the number of occurrences of a terminrelationtotheentityofinterest,
which in their case is clinical trial data. However, there are also some restrictions
to the database schemas we support. The schema has to be a connected acyclic
graph in order for our propagation algorithm to calculate the tags for each viewer
correctly. While with cyclic structures, the propagation algorithm simply uses
a shortest path approach, we could extend our framework so that a developer
could configure the algorithm to achieve a different behaviour, if desired.
We note that our current implementation follows a stateless approach. This
has some implications on system performance. Users can always choose to navi-
gate to a breadcrumb, which is a bookmark to an individual search and allows
a user to continue from there. With our current approach, these queries are
executed again, invoking the propagation algorithm to adapt all adjacent tag
clouds, while with a stateful approach these views could simply be cached. How-
ever, such an approach would be memory-intensive since it requires these views
to be materialised.
