Keywords

1 Introduction

Statistical datasets are widely spread and published on the Web. However, many users’ information need is not the mere consumption of statistical data as such, but the search for patterns and explanations. As shown in previous works [3], information from the Linked Open Data (LOD) cloud can serve as background knowledge for interpreting statistical data, as it covers various domains, ranging from general purpose datasets to government and life science data [7].

In this paper, we present the Web-based tool ViCoMapFootnote 1, which allows automatic correlation analysis and visualizing statistical data on maps using Linked Open Data. The tool automatically enriches statistical datasets, imported from LOD, RDF datacubes, or local datasets, with information from Linked Open Data, and uses that background knowledge as a means to create possible interpretations as well as advanced map visualization of the statistical datasets. To visualize geospatial entities on a map, we use GADMFootnote 2, a LOD database of polygon shapes of the world’s administrative areas.

So far, many tools for visualization of LOD and statistical data have been developed [4]. In particular, for RDF data cubesFootnote 3 exposing statistical data, different browsers have been developed, such as CubeVizFootnote 4 or PayolaFootnote 5 [1]. The CODE Visualisation WizardFootnote 6 [2] also features different chart and map based visualizations. Tialhun et al. [8] have developed a LOD-based visualization system for healthcare data. The system visualizes different healthcare indicators per country on a map, and is able to perform correlation analysis between selected indicators, which can later be visualized as a chart.

The direct predecessor of ViCoMap is Explain-a-LOD [3], which is one of the first approaches for automatically generating hypothesis for explaining statistics using LOD. The tool enhances statistical datasets with background information from DBpediaFootnote 7, and uses correlation analysis and rule learning for producing hypothesis which are presented to the user.

ViCoMap, presented in this demo, combines map-based visualizations on the one hand side, and mining for correlations using background knowledge from LOD on the other. As such, it opens new ways of interpreting statistical data.

Fig. 1.
figure 1

ViCoMap architecture

Full size image

2 The ViCoMap Tool

The architecture of the ViCoMap tool consists of three main components, as shown in Fig. 1. The base component of the tool is the RapidMiner Linked Open Data extension [5]. The extension hooks into the powerful data mining platform RapidMinerFootnote 8, and offers operators for accessing LOD in RapidMiner, allowing for using it in sophisticated data analysis workflows using data from LOD. The extension allows for autonomously exploring the Web of Data by following links, thereby discovering relevant datasets on the fly, as well as for integrating redundant data found in dfferent datasets, and wraps additional services such as DBpedia LookupFootnote 9 and DBpedia SpotlightFootnote 10.

All processes built within the RapidMiner platform can be exposed as Web services through the RapidMiner Server, which can be consumed in a user Web application. We use such a setup to integrate the functionalities of the RapidMiner LOD extension, as well as the functionalities of RapidMiner built-in operators, in the ViCoMap Web application. The ViCoMap Web application offers three main functionalities to the end-user: Data Import, Correlation Analysis, and Visualization on Maps.

2.1 Data Import

There are three options to import data, i.e.,import a dataset published using the RDF Data Cube vocabulary, import data from a SPARQL endpoint, and import data from a local file.

Import RDF Data Cubes: To import a dataset published using the RDF Data Cube vocabulary, the user first needs to select the data publisher source, and a dataset that will be explored. Currently, we provide a static list of most used RDF Data Cube publishers, like WorldBankFootnote 11. After selecting a data cube and the dimensions to be analyzed, the dataset is loaded into RapidMiner by the LOD extension.

SPARQL Data Import: To import data from a SPARQL endpoint, the user first needs to select a SPARQL endpoint and to provide a SPARQL query. The tool offers a SPARQL query builder assistant, which helps the user formulate queries such as Select the number of universities per federal state in Germany, by selecting a set of spatial entities (states in Germany) and a subject entity (universities).

Local Dataset Import: The user can import data from a local CSV file.

2.2 Correlation Analysis

Once the data is loaded, the user can select a column that will be used for correlation analysis. The user can choose the LOD sources that will be explored to find interesting factors that correlate with the target value at hand. To link the data at hand to remote LOD datasets, the tool exploits existing owl:sameAs links, and it automatically creates additional links, e.g., via DBpedia Lookup for non-linked datasets, such as CSV files. From the data retrieved from the additional LOD sources, a simple correlation analysis is performed to find simple correlations of the generated features and the target value under examination. The discovered correlations are sorted by confidence and presented to the user.

Visualization on Maps. After the correlation analysis is completed, the user can visualize any correlation on a map, using the Google Maps APIFootnote 12 and displaying two maps for the correlated values side by side. The shape data of the geographical entities is retrieved from GADM. DBpedia provides external links to the GADM dataset, which were created using different heuristics based on the label and coordinates of geographical entities [6]. DBpedia 2014 contains 65, 616 links to the GADM dataset, for entities on different administration level, e.g., municipalities, regions, states, departments, countries, etc. Such links allow us to visualize spatial entities on any administrative level.

Fig. 2.
figure 2

German states use case workflow

Full size image

3 Use Case: Number of Universities per State in Germany

In this use case, we analyze which factors correlate with the number of universities per state in GermanyFootnote 13. To import the initial data we use the query builder assistant from the SPARQL data import tab (Fig. 2a). After executing the query, the data is presented in a table with two columns, i.e., the DBpedia URI for each state, and the number of universities per state (Fig. 2b). By pressing the button Find Correlations, we can select the LOD sources that will be included in the correlation analysis (Fig. 2c). Next, the discovered correlations are presented in a new table with two columns, i.e., a column with the factor label, and the a column with the correlation confidence (Fig. 2d).

We can see that in this case, as shown in Fig. 3, the highest positive correlation is the RnD expenses of the states (\(+\)0.84), which is retrieved from Eurostat. The highest negative correlation is the latitude of the states (\(-\)0.73), which is retrieved from GeoNames, which reflects the north-south gradient of the wealth distribution in Germany.Footnote 14

Fig. 3.
figure 3

Correlations visualized on a map using GADM geographical shape data

Full size image

4 Conclusion

With this demo, we have introduced the web-based ViCoMap tool, which allows the users to analyze statistical data, and visualize it on maps using external knowledge from Linked Open Data. While at the moment, we use only literal data properties and types for finding correlations, we aim at a more intelligent exploration of the feature space, e.g., by automatically finding meaningful aggregations of different measures.