Exploration and Visualisation of Converted Data
In the following we describe two scenarios to showcase benefits of the presented framework. The first one is about statistical data discovery using CubeViz. The second scenario is about discovering geospatial information by the use of Facete.
Statistical Data Exploration
CubeViz, the RDF Data Cube browser, depicted in Fig. 6 allows to explore data described by RDF Data Cube vocabulary [2]. CubeViz generates facets according to the RDF Data Cube vocabulary artefacts such as Data Cube DataSet, Data Cube Slice, a specific measure and attribute (unit) property and a set of dimension elements that are part of the dimensions.
Fig. 6. Screenshot of CubeViz with faceted data selection and chart visualization component.
Based on the selected facets, CubeViz retrieves data from a triplestore and suggests possible visualizations to the user. Users or domain experts are able to select different types of charts such as a bar chart, pie chart, line chart and polar chart that are offered depending on the selected amount of dimensions and respective elements.
Geospatial Data Discovery
Facete, depicted in Fig. 7, is a novel web application for generic faceted browsing of data that is accessible via SPARQL endpoints. Users are empowered to create custom data tables from a set of resources by linking their (possibly nested) properties to table columns. A faceted filtering component allows one to restrict the resources to only those that match the desired constraints, effectively filtering the rows of the corresponding data table. Facete is capable of detecting sequences of properties connecting the customized set of resources with those that are suitable for map display, and will automatically show markers for the shortest connection it found on the map, while offering all further connections in a drop down list. Facete demonstrates, that meaningful exploration of a spatial dataset can be achieved by merely passing the URL of a SPARQL service to a suitable web application, thus clearly highlighting the benefit of the RDF transformation.
Fig. 7. Screenshot of Facete showing data about allotments in South East London.
Drill-Down Choropleth Maps
Import and export statistics of Poland collected in INSIGOS HZ/GEO dataset are best visualised on the globe. The globe itself is part of D3 library[1]. Some work was, however, necessary in order to allow display of data from triple store. Several parameters are defined in the graphical interface, and based on it SPARQL queries are prepared. Then, the legend is defined in such a way that colours are more or less equally distributed. Normally the numbers for import and export are subject to power law, therefore the legend scale cannot be linear. The map is coloured according to values assigned to selected countries. A sample map is presented in Fig. 8 shows 20 countries with the greatest value of export in 2012 expressed in PLN (Polish currency), with the unit being millions.
Not only technical communication with Virtuoso had to be solved. We first needed to integrate data on semantic level, i.e. map of the world in D3 had
Fig. 8. Export statistics of Poland in 2012 presented on a globe
country codes consisting of three letters. Countries in INSIGOS dataset had just names, and therefore additional mapping was necessary. It should be noted that list of countries changed for the period analysed but the map has not been updated.
More popular is visualisation of data on the country level. For this purpose we need a map of country available in SVG and compatible with D3 library. It can also be derived from open data. For example, a map of Poland has been prepared based on OpenStreetMap and administrative division included there.
On the map of Poland we visualise data concerning public contracts. Several measures can be visualised like: number of contractors, number of contracts or value of contracts. All measures are sliced by geographical dimensions reflecting administrative division of Poland. There are 16 voivodeships (wojewo´dztwo) and 380 districts (powiat ) in Poland. Showing 380 entities on the map is not very useful for interpretation. Therefore we have applied the drill-down approach. First, a user is presented with a map of the whole Poland with 16 voivodeships.
Then, after clicking on selected region, the user goes to detailed map of districts in a given region. There is also another administrative level – county (gmina) – which can be included when needed. Analogous maps can be prepared for other countries as well.
Drill-Down Tables
In terms of the drill-down functionality, we need to remember that datasets can be aggregated on various level of detail and very often they are offered as the same package. Geography is not the only dimension. There are several others that cannot be visualised within the map, hence the need to develop a drilldown table. Some examples include (in the case of Polish data): time dimension; Polish classification of activities (PKD, NACE): sections and chapters; common nomenclature (CN): several levels; various economic indicators in energy-related data (e.g. production total → from renewable sources → from solar plants).
Due to required integration with triple store we prepared our own drill-down table from scratch. The prerequisite is that the dimension to be used for drilldown is described as SKOS concept scheme. It is an industry standard and allows to represent hierarchies conveniently. It has also mechanism for labels in various languages. Alternative labels make mapping to this headers more flexible when heterogeneous sources are considered. All vocabularies, including time dimension, were prepared with this prerequisite in mind.
There are in fact three queries necessary to prepare a drill-down table. The approach is thus similar to multidimensional queries against OLAP cubes in MDX[2]. First, we need to get headers of rows and columns, then data itself. Not only labels for headers are necessary but also interdependencies between headers and their level. When a drill-down table is first loaded, rows and columns are collapsed, so that only most aggregated data is shown. It is then possible to click on the respective row or column marked with 'plus' sign to expand one level. Figure 9 presents expanded columns.
Fig. 9. Drill-down table with expanded columns
