We present the Dataset Dashboard, a SPARQL endpoint exploration tool that helps to understand the dataset structure, as well as relationships to other datasets. The tool is based on the notion of a dataset descriptor which describes some characteristic of a dataset. Currently, the tool offers descriptors for basic class/property statistics, spatial information, temporal information, as well as advanced dataset summarization. The tool currently registers over 200 SPARQL endpoints and named graphs inside the SPARQL endpoints.
When trying to understand unknown RDF datasets1, linked data consumers needs to judge suitability of the dataset for his/her scenario in an efficient way. However, the dataset exploration itself is a non-trivial and time-consuming task for many reasons. First problem lies in the data – although many RDF datasets are cataloged, the metadata gathered about these datasets are typically not put in sync with the actual dataset content on the regular basis, or describe provenance metadata and do not reflect the dataset content at all. Second important problem lies in the accessibility of the data, which is often limited by (un)availability/latency of SPARQL endpoints, or efficiency of the underlying hardware.
In this paper we aim at the problem of dataset exploration by users with technical background who need to become familiar with unknown datasets (e.g. linked data designers who link to external sources, data journalists who explore a new linked open data source, etc.). As an example, let’s consider a linked data designer who needs to find out, whether a new and unknown dataset is suitable for integration with another dataset (s)he is designing. For this purpose, a bunch of exploratory SPARQL queries need to be posed by the designer to become familiar with the vocabulary used in the dataset, mutual interlinks between classes, their significance, as well as the spatial (e.g. which spatial area it covers) or temporal (which temporal interval it spans) scope of the dataset.
The introduced Dataset Dashboard tool aims at making the process of understanding an unknown RDF dataset more efficient. For this purpose, the tool shows several 1 In this paper, we only consider RDF datasets available as subsets of actual snapshots of SPARQL endpoints content. interactive visualizations of general content-level characteristics of the dataset, including a filterable dataset summary, basic class and property statistics, as well as spatial and temporal range of the dataset. 2
Please refer to the RDF(S) specifications [
Let’s introduce basic features of the Dataset dashboard2, which are later presented in Section 3 on an example.
Summary Schema Widget The basic summarization descriptor provided by the Summary Schema Widget is the SPO Summary, schematically described as
δSP O(g) = {hc1, p, c2i|{hx, rdf : type, c1i, hy, rdf : type, c2i, hx, p, yi} ⊆ g} Additionally, frequencies of the summary triples are computed and visualized by the weight (thickness) of the corresponding graph edge. The widget provides also an enhanced version of the SPO Summary descriptor which shows instance-level links to other datasets. The technique to compute enhanced descriptors is currently under review at the Journal of Web Semantics.
To explore large summaries, two filtering options are offered – content-based filtering allows to exclude concrete classes/properties from the summary graph, while weight-based filtering allows to exclude all edges with weight below a threshold, i.e. filtering out triple patterns which are not frequent enough in the dataset.
Spatial Widget Spatial part of the data is represented by geographical objects, using GeoSPARQL3. Spatial Widget extracts GeoSPARQL geometries and displays them in a map, schematically δGeoSP ARQL(g) = {ho, p, gi|ho, p, gi ∈ g, for p ∈ {geosparql : asWKT, geosparql : asGML}} The widget recognizes both Well-Known Text (WKT) geometry and Geography Markup Language (GML) geometry. All types of features found in the dataset are listed in the pop-up menu next to the map, where the user may choose which features to show in the map. Point, line and polygon geometries are supported. Moreover, points are represented as markers with pop-up bubble containing the link to the RDF resource representing the feature. 2 Available on-line at https://onto.fel.cvut.cz/dataset-dashboard 3 http://www.opengeospatial.org/standards/geosparql, cit. 1.6.2018
Temporal Widget Temporal widget shows the temporal coverage of the dataset,
computed as minimum and maximum time points occurring in the dataset. The computation
itself considers the structured temporal information as well as temporal information
extracted from non-structured texts. For the unstructured temporal information, it retrieves
the textual literals and performs a natural language processing analysis to extract the
time information. To perform this step, the Stanford temporal tagger (SUTime)4 which
is part of the StanfordCoreNLP5 tool is used. The pipeline for computing the temporal
descriptor can be found in [
Dataset Dashboard Bookmarking In order to share a particular dataset dashboard across the community, bookmarkable dataset dashboard URIs are offered. 3
Let’s consider a dataset gexp about parcels, buildings, floors and land use which is a part of the dataset maintained by the Prague Institute of Planning and Development6. Comparing to the original, gexp is limited to the data for Prague Centre, which results in approx. 1.8 mil. triples.
Basic capabilities of the dataset dashboard are shown in Figure 1. The figure shows an overview of gexp in terms of the SPO summary provided by the Summary Schema widget. The full SPO summary contains 13 RDFS classes and 25 RDF properties. The dataset dashboard allows to filter out some classes/properties from the schema. Additionally, edges can be filtered according to their weight. Thus, the figure excludes two properties from the view (rdfs:label, ddo:has-published-dataset-snapshot) and shows only edges with weight at least 12000 (adjustable by the slider control above the SPO summary graph).
Spatial and temporal context of the dataset is shown in Figure 2. The GeoSPARQL descriptor shows geometries attached functional land use (instances of the RDFS class town-fvu:VyuzitiPloch). Dereferencable link of one of the geometries7 is shown in a tooltip. The temporal descriptor shows the temporal range of the dataset. This dataset provides temporal information only inside the properties town-parcely:dat vznik and town-budovy:dat vznik, town-budovy:dat zmena, which denote the creation/change dates of the data change record. So in this case, and contrary to the GeoSPARQL descriptor, the extracted temporal knowledge refers rather to the creation of the data than to the actual content. The dashboard of this example is shown at the following URL https://onto.fel.cvut.cz/dataset-dashboard/#/dashboard/online? endpointUrl=http://onto.fel.cvut.cz:7200/repositories/ipr_ datasets&graphIri=http://onto.fel.cvut.cz/ipr-datasets/resource/ town-plan 4 https://nlp.stanford.edu/software/sutime.html 5 https://stanfordnlp.github.io/CoreNLP/ 6 http://en.iprpraha.cz, cit. 1.6.2018 7 http://onto.fel.cvut.cz/ontologies/town-plan/pvp_fvu_p/ geometry/70/2018-01-29T14:36:24.178617 4
Dataset dashboard is a client-server application8, as shown in Fig. 3a. The client side serves mainly as a container for various widgets. Each widget is a separate React component, which is able to visualize a dataset descriptor of one or more types (see Section 4.1). The server takes care of proper creation and consumption of Dataset Descriptor Ontology metadata (see Section 4.1). In order to implement a new visualization, one needs to implement a visual widget component and just insert it in the DatasetDashboardController, along with the other widgets. Upon request, the widget gets from the Server the result of one of the predefined SPARQL queries. For more complicated scenarios, one needs to also implement an SPipes (see Section 4.2) pipeline that computes the particular descriptor.
(a) Dataset Dashboard Architecture (b) Dataset Description Ontology Core 4.1
Dataset Descriptor Ontology (DDO) [
SPipes is a custom implementation of the SPARQLMotion engine10. It allows for
creation RDF processing pipelines, consisting of modules (nodes) and their edges
dependencies (edges). A dependency says that first the source module is executed and its
resulting RDF graph together with global variable bindings is passed to the target
module. Then the target module is executed. The language is very well integrated with
SPARQL[
An overview of exploration tools can be found in [
LODeX [
LODSight [
Linked Data Visualization Wizard [
Linked Geo Data Browser [
Facete [
Furthemore, comparing to the all mentioned tools, Dataset Dashboard provides a comprehensive view of the dataset under exploration by combining different descriptors into a single dashboard. Also, it stores the computed dataset descriptors14 and provide persistent identifiers, for efficient sharing of the view over the dataset. 6
We conducted a preliminary survey about usefulness of the tool among three IT experts (a PhD student in the semantic web field, a linked data expert and a semantic web developer) not involved in the system design and development. As a part of the survey they had to explore three datasets (a SKOS vocabulary about labor law15), a complex dataset about EU television content16), a dataset with geospatial and temporal information about urban development in Prague17) not known to them before and judge the benefits of using tool for their use-case. Although the experts were not provided with any information about the tool beyond its SPARQL endpoint URL, all of them were successful in describing what the topics of all three datasets are (mainly using the Summary Schema Widget). Two of them see the main advantage of the tool in support for subsequent SPARQL query formulation to the particular SPARQL endpoint. The third one sees its advantage in providing visualization revealing the complexity of the dataset. Two of the experts find dataset dashboard persistent links useful for sharing information about the datasets. On the other hand, for the purpose of dataset exploration, the experts miss visualization of data samples and more advanced data statistics. 7
We presented the Dataset dashboard – a tool for dataset exploration using different dataset descriptors. The tool currently registers over 200 SPARQL endpoints and named graphs inside the SPARQL endpoints and is currently used in two national research projects. Initial feedback by IT experts is motivating, revealing that the tool is useful for dataset exploration, as well as providing suggestions for future work.
In future, we aim at providing history tracking for computed descriptors, as well as introducing new descriptor types (e.g. data samples, as suggested by the survey) for spatial and temporal widgets.
Acknowledgements This work was partially supported by grants No. GA 16-09713S Efficient Exploration of Linked Data Cloud of the Grant Agency of the Czech Republic and No.SGS16/229/OHK3/3T/13 Supporting ontological data quality in information systems of the Czech Technical University in Prague. 14 Currently only for SPO summary and temporal descriptors. 15 http://vocabulary.wolterskluwer.de/PoolParty/sparql/ arbeitsrecht, cit. 10.7.2018 16 http://lod.euscreen.eu/sparql, cit. 10.7.2018 17 The dataset presented in Section 3