Publishing Linked Data on the Web has become much easier with tools such as Drupal. However, consuming that data and presenting it in a meaningful way is still di cult for both Web developers and for Semantic Web practitioners. We demonstrate a module for Drupal which supports visual query building for SPARQL queries and enables meaningful displays of the query result.
For illustration of the tool and its usefulness, we imagine a research institute Web site such as http://deri.ie/. The site includes pages for researcher pro les. A researcher's pro le page pulls the researcher's publication list from an dataset such as DBLP3. The pro les will be set up and administered by a Webmaster who is unfamiliar with SPARQL, but knows a small amount of PHP and is comfortable with HTML.
We walk through this use case in a video demonstration available at http: //lin-clark.com/iswc. We give an analysis of the challenges faced in building such a Web site and our solutions to those challenges below. 3 3.1
Inexperienced users are overwhelmed when given too many interaction options
or, even worse, when given a blank slate where they must enter a command
language like SPARQL [
Interaction strategy|To assist users in nding where to start, we initially
o er one single point of interaction. We make the assumption that most queries
are centered around predicates, and we use the drag-and-drop predicate list as
the rst means of interaction. We then animate the addition of subject and
object to that predicate in order to guide the user to the next step, using the
strategy of sequential a ordance[
When accessing an arbitrary endpoint, end users have little support in understanding what data is contained in the dataset and how it is linked.
Interaction strategy|The endpoint is queried to determine which predicates are used in the dataset. Only predicates that are present in the dataset are displayed. An autocomplete search box allows the user to lter to appropriate terms. 3.3
Declaring pre x mappings requires knowledge of both syntax and of standards and inexperienced users can have a hard time nding the appropriate namespace for a vocabulary.
Interaction strategy|The pre x declaration is automatically generated for the user from the predicates that are used. If a namespace mapping is not available in the system (which uses pre x.cc as it's source), then the full URI is used in the predicate display and the query. 3 http://dblp.l3s.de/d2r/sparql
SPARQL Views: A Visual SPARQL Query Builder for Drupal
While conjunction and disjunction based logic is ubiquitous in everyday life and is thus intuitive to a large degree, the syntax of SPARQL obscures that logic for the inexperienced user.
Interaction strategy|Where SPARQL syntax requires mental
computation of the logic, visual representation of queries allows perceptual inference of
the underlying logic[
SPARQL results are not useful in their raw format, but need to have tailored displays in order for the pattern of information within the data to be communicated. Semantic Web practitioners often create bespoke systems for this display, leading to duplication of e ort within the Semantic Web community.
Interaction strategy|We integrated the query building tool with a widely deployed Drupal module, Views. This module o ers a pluggable system for displaying query results. Switching from an HTML table view, to a Google API chart view, to a JavaScript Exhibit of the data is easy to do through the Views user interface. Because SV is integrated with this pluggable system, any style and display plugins that the Drupal community develops to display various kinds of data can also style SPARQL results. 4.2
Information is most useful if it is tailored to the page where it is displayed. For instance, in the use case above, when a visitor visits a faculty member's pro le, the query should return only publications authored by that faculty member.
Interaction strategy|We provide support for Drupal's Token API. This allows users to easily register tokens|small placeholder variables|and use these in their queries. The token is then evaluated at page load, so context, such as which user's pro le is being viewed, can be assessed when creating the query.
We currently use a DISTINCT query to get the predicates from a dataset. For certain endpoints, such as the one at http://dbpedia.org/sparql, this query will timeout and not return results. We plan to create a Web service that can extract predicates from crawls preprocess the predicates for di erent endpoints. This service could provide additional data, such as the predicate de nition and a ranking based on predicate usage within the dataset. 5.2
A barrier we have not addressed in the current work is the selection of appropriate datasets. We plan to create tools to guide users in nding an appropriate dataset. 5.3
Currently, there can only be one endpoint or dataset de ned per view. However, one of the most exciting potentials of the Semantic Web is to mix data from multiple sources and we plan to explore how this can be supported in SV. 5.4
We believe that it is possible to support and encourage the user's increasing understanding of SPARQL by lowering the barrier to entry and focusing on learnability. We plan to evaluate this assertion, to see whether interaction with SPARQL Views increases the acceptance of SPARQL and the understanding of the syntax itself, enabling users to incrementally improve their understanding of the query language.
Acknowledgments. This work was funded by Google Summer of Code, LOD Around-The-Clock (LATC) ICT-256975, and by Science Foundation Ireland under Grant No. SFI/08/CE/I1380 (Lion-2). Thank you to Drupal contributors Daniel Wehner and Laura Scott for their guidance and feedback.