Faceted browsing Faceted browsing is a superior exploration technique for large structured datasets [4], useful when users do not know the data that they are looking for. In faceted browsing, the information space is partitioned using orthogonal conceptual dimensions of the data (called facets). Each facet has multiple restriction values; users select a restriction value to constrain relevant items in the information space. Existing approaches such as Flamenco [4] and Ontogator [1], cannot navigate arbitrary datasets but are limited to manually defined facets over predefined data structures. We have defined an approach for automatic construction of facets in semi-structured RDF data [3]. We extend the notion of faceted browsing to (graph-based) RDF data and define metrics for automatic ranking of facet quality. Our technique works for arbitrary RDF data, without the need to conform to any schema. In this paper we present the accompanying prototype implementation.
The prototype implementation is available at http://browserdf.org. A screenshot, showing an RDF dataset1 representing the FBI most-wanted fugitives, is shown in Fig. 1. The interface is automatically generated for this data, and works similarly for arbitrary data.
On left-hand side of Fig. 1 we see the various facets in the dataset, e.g. alias, eye-color, hair-color, or nationality. The user can select any facet and restrict it, either by directly choosing restriction values (for literal values) or by indirectly (for resources which themselves have properties that are facets). At the top the currently applied constraints are shown, and the main part displays the resources that conform to the current constraints.
The user is prevented from following dead-ends: at each step, only those facets and restriction values that have non-empty results are available. Searching with keywords is also available, either within the current selected resources or in the whole set.
The screenshot shows a simple restriction-value over the weight of a person. The prototype offers several other types of constraints, such as arbitrary joins, arbitrary selection focus, property inversion, and intersection of these.
Implementation
The prototype is implemented using the web application framework Ruby-on-Rails2
and the object-oriented RDF API ActiveRDF3 [
We are currently performing an comprehensive evaluation, about which we will report in future work. Preliminary evaluation results look promising: the metrics give priority to the most important predicates, the automatically constructed interface gives a usable overview of the dataset and users find the interface highly usable.
Acknowledgements This material is based upon works supported by the Science Foundation Ireland under Grants No. SFI/02/CE1/I131 and SFI/04/BR/CS0694.