The Linked (Open) Data cloud has been growing at a rapid rate in recent years. However, the large variance of quality in datasets is a key obstacle that hinders their use, so quality assessment has become an important aspect. Data pro ling is one of the widely used techniques for data quality assessment in domains such as relational data; nevertheless, it is not so widely used in Linked Data. We argue that one reason for this is the lack of Linked Data pro ling tools that are con gurable in a declarative manner, and that produce comprehensive pro ling information with the level of detail required by quality assessment techniques. To this end, this demo paper presents, Loupe API, a RESTful web service that pro les Linked Data based on user requirements and produces comprehensive pro ling information on explicit RDF general data, class, property and vocabulary usage, and implicit data patterns such as cardinalities, instance ratios, value distribution, and multilingualism. Pro ling results can be used to assess quality either by manual inspection, or automatically using data validation languages such as SHACL, ShEX, or SPIN.
The Linked (Open) Data cloud has been growing at a rapid rate in recent years.
Some portions of it come from crowd-sourced knowledge bases such as Wikipedia,
while others come from government administrations, research publishers, and
other organizations. These datasets have di erent levels of quality [
Juran and Godfrey describe quality using multiple views [
Detailed data analysis is one common preliminary task in quality assessment,
and data pro ling is one of the most widely-used techniques for such analysis
[
In this paper, we describe a Linked Data pro ling service, the Loupe API,
which provides access to the Loupe tool via a RESTful interface. The Loupe API
may be con gured to specify the source data as well as the pro ling activities it
should perform. As a consequence it can be used for di erent purposes. In the
recent years, Loupe has been used to assess the quality of datasets in several
projects such as DBpedia [
The Loupe API2 is a con gurable Linked Data pro ling service. The three main phases in Linked Data pro ling are (1) speci cation of input, (2) execution of data pro ling, and (3) representation of pro ling results. Listing 1a shows an example input of a Loupe API pro le request. Users can specify their requirements (i.e., which pro ling tasks to execute) and other con guration details such as how to access the data source (and which data to pro le), or whether to persist the pro ling results in the Loupe public repository (i.e., they will be available via search). The pro ling tasks are grouped into four categories: { summary - provides generic statistics on an RDF data source related to its size and the type of content it has, for example, typed entity count or distinct IRI object count. { vocabUsage - provides information on the implicit schema of the data by analyzing how vocabulary terms such as classes and properties are used, their domains and ranges, cardinalities, uniqueness, among others. { languagePartitions - provides information on multilingual content by analyzing the frequency of each language in language tagged strings. 1 http://demo.mappingpedia.linkeddata.es/ 2 http://api.loupe.linkeddata.es/ { valueDistributions - provides information on the value distribution of a given property.
The results of pro ling are represented in RDF using the Loupe ontology3. The main elements of the pro ling results are illustrated in Figure 1b; the complete results in RDF are available4; we also provide a set of cURL examples5 for invoking the service.
(a) Input Con guration (b) Output Elements
These pro ling results can be used for validating the quality of a dataset
either by manual inspection or by specifying the validation rules in a language such
as SHACL6, ShEx7, or SPIN8. Data pro ling facilitates the manual inspection
by providing a high-level summary so that an evaluator can adapt techniques
such as exploratory testing [
Nevertheless, automatic validation is needed when a large amount of data is
present and it is feasible in most situations. For example, data model constraints
such as uniqueness of values, expected cardinalities, domains and ranges,
inconsistent use of duplicate properties [
Further, pro ling results enable the analysis of a dataset over a period of time by periodically pro ling data and performing the analysis on multiple pro ling results. For instance, expected deletions of data or undesired changes can be detected by analysing the changes in the dataset pro les.
The Loupe API is implemented as a RESTful service and currently three
operations are available as illustrated in Figure 2.
3 http://ont-loupe.linkeddata.es/def/core#
4 https://git.io/vy1tO
5 https://github.com/nandana/loupe-api/wiki/examples
6 https://www.w3.org/TR/shacl/
7 https://shexspec.github.io/spec/
8 http://spinrdf.org/
Zaveri et al. [
Tools that provide statistics on the Linked Open Data Cloud include Aether
[
Di erently from the other tools mentioned, the Loupe API is available as a RESTful web service where users can con gure and generate Linked Data pro les in RDF using the Loupe ontology. Further, Loupe provides summarized information not only on explicit vocabulary, class and property usage as the other tools but it also facilitates the analysis of implicit data patterns by providing a ner grained set of metrics compared to existing tools, such as instance ratio (ratio of instances of a given class to all entities) and property cardinalities. Low granularity metrics and other capabilities of Loupe have been applied to the analysis of redundant information, consistency with respect to the axioms in the ontology, syntactic validity and detection of outliers. 4
This paper presents the Loupe API, a con gurable RESTful service for pro ling Linked Data, where results can be used for quality assessment purposes. The paper illustrated its use, and motivated it with a discussion on how it can be integrated to the quality assessment process.
Nevertheless, there are several challenges in pro ling large datasets using a service compared to a standalone tool. Thus, Loupe API is mostly suitable for pro ling small datasets. Large datasets (e.g., DBPedia) could take a long time to pro le and the requests might timeout. In the future, we plan to provide support for asynchronous executions for such cases.
Another challenge is to detect the capabilities and limitations of the SPARQL endpoint and to adapt to those capabilities. Loupe API uses SPARQL 1.1 features and some metrics are omitted if an endpoint only supports SPARQL 1.0.
In the future, we also plan to extend the pro ling service to other Linked Data sources such as RDF dumps, SPARQL construct queries, and LDF endpoints. Further, we plan to allow users to specify their quality requirements in a declarative manner using formal languages such as SHACL, ShEX, SPIN or using an editor with common validation rules. This will allow Loupe API to generate quality assessment reports based on those requirements.