The wealth of information in the Linked Open Data cloud (hereafter LOD) is large and growing, enabling comparisons between previously isolated datasets to be made [3]. However, exploring the linked data cloud is difficult for users unfamiliar with semantic web concepts such as SPARQL and RDF. Webbased visualization tools such as IBM's Many Eyes have shown promise in allowing collaborative data exploration [7]. We have developed a tool that allows users to similarly plot quantitative data found on the LOD cloud, with minimal knowledge of semantic web syntax. This tool enables users to explore, share, and expand upon the data found in the LOD cloud.
Finding data on the LOD cloud using openChart 1 consists of identifying an entity of interest, choosing two of its quantitative properties, and selecting a peer group with which to compare values. To enable entry into the semantic web, we use Wikipedia's autosuggest API to determine an entity's Wikipedia address. This is then matched with a corresponding semantic web resource using a SPARQL query on the DBpedia database, which is a central hub of the LOD cloud with many owl:sameAs linkages to other sources of data [2]. While other endpoints could be used with the openChart framework, DBpedia has a high number of links to other LOD sources, making it useful for a general purpose tool.
.
Following the identification of an entity of interest, for example Bangladesh, we find the quantitative properties from the RDF resource by using regular expressions to remove non-quantitative information. Two of these are selected by the user, for example hdi and population density. Then, peer groups are found through a SPARQL query that looks for distinct rdfs:type that contains objects with both of the quantitative properties. These peer groups may or may not contain the users' original search term-but the selection of one, for example Country, will display a scatter-plot of the two variables. This peer group feature is an important aspect of our application because it allows the user to branch out when navigating information on the semantic web rather than fixate on answering one question in particular. . At all levels of the exploration, the data is locally cached in a MySQL server. The frontend is written using JavaScript and the jQuery library, while the backend is written in PHP 5 and the ARC2 library. The plotting is done using the Protovis JavaScript
A focus on a simple user interface has made openChart an easy to use introduction for WWW users unfamiliar with Linked Open Data. By focusing on peer groups and not only information directly relevant to a users' query, the openChart tool emphasizes a broad exploration of available data rather than merely answering a specific question. Additionally, we incorporate a social component into openChart, where interesting relationships between concepts can be shared. This is new knowledge that is being created, and eventually will be integrated into the LOD cloud itself by defining such shared charts as RDF objects.
An additional benefit of displaying data visually in openChart is the ability to quickly identify errors within the data contained in the LOD cloud. In isolation, it is often difficult to see errors in scale or other such mistakes-displaying them as outliers enables mistakes to be rapidly identified. These data points can then be flagged for review in order to improve the quality of the source data, or any scripts that are used to parse the data into the RDF format in the first place. Such flagging could be achieved by defining a quality ontology and publishing triples for user identified errors.
When searching the LOD cloud through a SPARQL query, it would be economical to restrict SPARQL queries to retrieve only properties with ranges limiting them to numerical values. However, we found that many of the properties lack associated rdfs:range and/or rdfs:domain values. This resulted in a need to retrieve all results and then parse them using regular expressions, increasing the overhead of the application. Thus, we suggest that RDF authors take the time to specify rdfs:range and rdfs:domain values such as xsd:integer and xsd:decimal to facilitate statistical work using Linked Open Data.
Another aspect often missing from data sources, especially from DBPedia, is units of measure. Particularly when comparing across endpoints, it is imperative that the units of measurements are understood, in order to prevent scaling errors when comparing data from different sources. We suggest that creators of RDF data take the time to include unit specifications, either through ontologies such as Quantities, Units, Dimensions and Data Types in OWL and XML [6], or by agreeing on standardized unit abbreviations and distributing unit-aware parsers.
Since the data in openChart is coming from multiple sources, tracking the sources of a chart's data would be important in enabling the use of the charts in research. As a result, we plan to implement a feature by which the origins of the data contained within a chart will be displayed concurrently with the chart.
Although RDF quadruples (such as [4]) would allow this to be easily implemented, methods that determine authorship based on particular endpoint characteristics could be implemented currently.
There exist many existing browsers of semantic web data, such as Tabulator, which offer capabilities similar to our system [1].
Although openChart is easier to use than these programs, due to the restrictive nature of the queries permitted on our system, we are working to enable the switching back and forth between Tabulator and openChart, to allow more technical users to experience the full potential of the semantic web, using openChart as a starting point.
As mentioned previously, the information gleaned from openChart can be published for others to access. Statistical relationships can be described using the SCOVO ontology, which allows the specification of statistics with reference to a particular dataset over a range of time [5]. Care must be taken to ensure the completeness of the data, however, since the statistics generated only represent the data published to the LOD cloud. Two groups of statistics are generated -one describing the local cloud itself, such as describing the number of triples, and another describing the data contained therein.
We would like to thank Tim Berners-Lee, K. Krasnow Waterman, Reed Stuyvesant, Ian Jacobi, Oshani Seneviratne, and everyone else who participated in and organized MIT's Linked Data week in January of 2010.