There are several ways to publish data on the web, from plain or GitHub web pages, to linked open data, and online catalogs. However, since each data owner can select a diferent way to publish his data, it is challenging to discover all the available datasets that are represented through a particular ontology O. The rationale for building a catalog for datasets expressed with respect to a particular ontology O (or specializations of O), is that the community which is interested in that ontology has various incentives to publish its datasets in such a catalog: for inspecting the use of the ontology (for spotting errors and/or for guiding the use and evolution of the ontology), for finding other datasets that could be easily integrated, and others. In this paper we focus on such catalogs, and in particular we showcase a catalog for datasets expressed with respect to CIDOC-CRM. Even if there are dozens of datasets that are represented using CIDOC-CRM, there is not any online resource that contains even a simple textual list of all these datasets. To fill this gap, we present an interactive portal that contains ontology-based descriptions for 30 CIDOC-CRM datasets. Through this portal, the user can browse all CIDOC-CRM datasets (and their statistics), can find all datasets that use a particular CIDOC-CRM property/class, can see the most frequent properties and classes, can check the commonalities between diferent datasets, and can enrich the catalog with new datasets. Finally, we provide indicative measurements over the 30 collected CIDOC-CRM datasets and for the properties/classes of the current version of CIDOC-CRM.
There are numerous ways for publishing data on the web, such as in a web page, in GitHub, in
Zenodo, as Linked Open Data (e.g., in a SPARQL endpoint or through a data dump) or/and by
uploading them to an online catalog, which can ofer more services comparing to the previous
ones. In particular, catalogs can ofer the following services (upper part of Fig 1): S1) hosting of
datasets descriptions [
The objective for T3 catalogs follows: i) the community that is interested in the focused
ontology has incentive to publish their datasets in that catalog, and ii) it is more sustainable to
achieve completeness for one ontology, than being complete for all the available ones. However,
it is not trivial to create such a catalog; due to the numerous ways to publish a dataset, it is quite
challenging even to discover all the datasets using a popular model, such as CIDOC-CRM. Indeed,
the lack of even a simple list that includes information of all the CIDOC-CRM datasets (except
for a table in a GitHub page [
Concerning our contribution, we present an ontology-specific portal (or catalog), by focusing
on CIDOC-CRM. The objective is to make it feasible the CIDOC-CRM users, dataset owners and
experts, to discover the available CIDOC-CRM datasets and to browse statistics/visualizations
about them, since it can be important for several use cases including data discovery, data
integration, ontology evolution and others. Specifically, we first collect 30 CIDOC-CRM datasets
and we compute ontology-based descriptions using VoID vocabulary [
In the rest of this paper, §2 discusses the related work, §3 shows the use cases and §4 presents the steps for creating the ontology-based descriptions. §5 presents the functionality of the portal and §6 provides measurements over the collected datasets. Finally, §7 concludes the paper.
We discuss dataset catalogs, CIDOC-CRM based services and a comparison with related work.
Dataset Catalogs. Concerning global catalogs, in Datahub and lod-cloud.net [
Services using CIDOC-CRM. There are many services that use the CIDOC-CRM model for
various tasks, including Entity Recognition [
Placement and Novelty. The portal of this paper belongs to the scope T3, and mainly
ofers S2 services. Comparing to similar catalogs that compute VoID statistics [
We present the use cases where the portal can be exploited (and the corresponding users). First, the users are divided in 3 categories: a) simple (RDF) users, i.e., users familiar with Semantic Web technologies, b) CIDOC-CRM dataset owners, i.e., users that have published at least one dataset by using the CIDOC-CRM model, and c) CIDOC-CRM experts/researchers, i.e., the experts of the CIDOC-CRM community. Below, we provide 4 use cases (UC) and example user queries that we desire to support. The use cases (see Fig. 2) are the following: UC1) dataset discovery and selection, UC2) data publishing, UC3) data integration and UC4) ontology evaluation.
∙ UC1. Dataset Discovery and Selection. The objective is any user to be able to discover
all the available CIDOC-CRM datasets (see Query Q1 in Fig. 2), i.e., each dataset is a kind of
example of how to use CIDOC-CRM, thereby it can aid the adoption of the ontology. Moreover,
one can find datasets having specific properties or/and classes (see Query Q2 in Fig. 2), such
as datasets describing places, events, etc. In this way, the user can select the most appropriate
dataset(s) for creating an application, such as for Question Answering [
∙ UC2. Data Publishing. Since there are several ways to publish a dataset, and given the large number of published datasets, the key notion is a dataset to be easily discoverable and reusable by interested users. Therefore, by having a single portal including all the datasets of a specific model (such as CIDOC-CRM), we expect that it will be more discoverable from users that are interested in the given ontology or/and domain (i.e., in our case Cultural Heritage).
∙ UC3. Data Integration. In many cases, the data owners desire to integrate their data with
existing datasets for enriching their information, i.e., for creating larger and more complete
datasets. By ofering services for all the datasets described through a specific model (e.g.,
CIDOC-CRM), the dataset owners will be able to discover the datasets having the most common
properties and classes with their datasets for selecting them for semantic data integration [
∙ UC4. Ontology Evaluation. Since some ontologies are widely used, the experts of such models usually desire to evaluate which classes/properties are used and how, to detect problems more easily (e.g., using the ontology in a wrong way) and to think about possible extensions. We expect that queries like Q5 and Q6 of Fig. 2 will be quite useful for CIDOC-CRM experts, since there is a very active community through the CIDOC-CRM Special Interest Group (SIG), where many organizations and researchers participate (https://www.cidoc-crm.org/sig-members-list). This group is associated with several management activities: ontology versions, mappings, translations, compatible models, use cases, issues, best practices, meetings, and others.
Here, we present the steps (i.e., see Fig. 3) that are followed for collecting CIDOC-CRM datasets
and for producing statistics using the VoID vocabulary [
Step 1. Collecting CIDOC-CRM datasets. We tried to collect all the available CIDOC-CRM
datasets that ofer either an online SPARQL endpoint or an RDF data dump. We used the list
of 18 datasets provided in a GitHub page [
Step 2. Computing the Ontology-Based Descriptions using VoID. For the computation
of the statistics we send queries to SPARQL endpoints, however there are datasets where a
SPARQL endpoint is not provided. For these datasets, we downloaded the data dumps and we
uploaded them to our SPARQL endpoint for performing the computations. The mentioned
process was time consuming in some cases, due to i) the large size of some datasets and ii)
syntax errors in some RDF files. Concerning the computation of statistics (Step 2 of Fig. 3), we
use some basic SPARQL queries by exploiting the VoID vocabulary [
Example. Fig. 4 shows the file for the dataset ‘OpenArchaeo" [
Important Note. Concerning the CIDOC-CRM model, in this paper when we refer to “CIDOCCRM properties and classes", we refer to all the properties and classes of the RDF file of CIDOC-CRM version 7.1.2 (https://cidoc-crm.org/rdfs/7.1.2/CIDOC_CRM_v7.1.2.rdfs) which contains 309 CIDOC-CRM properties (including inverse properties) and 76 CIDOC-CRM classes, and not in properties and classes that extend the mentioned CIDOC-CRM properties and classes.
Step 3. Upload the Ontology-based Descriptions to a SPARQL Endpoint. The produced ifles (see Step 3 of Fig. 3) of all the datasets are uploaded in an online SPARQL Endpoint (i.e., 30 “Turtle" files). For describing all these (VoID) statistics for these datasets, 23,195 triples were created. The key notion is the endpoint to be used at real time from the portal for enabling a) the visualization of the already computed statistics, b) the computation of even more statistics through more SPARQL queries and c) the easy addition of any CIDOC-CRM dataset.
Here, we provide some details about the architecture of the web portal and then we present the functionality of the portal and we explain how it corresponds to the users and use cases of §3.
The portal is available in https://demos.isl.ics.forth.gr/CIDOC-CRM_Portal/ and ofers real time interactive browsing and visualizations. It runs on a server with 4 GB main memory, 8 cores and 60 GB disk space. Its architecture is shown in Fig. 5; the frontend has been designed by using the Angular framework (https://angular.io/) and the backend ofers a REST API by using the Spring Boot Framework (https://spring.io/). When a user selects a mode, a request is sent to the REST API, which is connected to the public SPARQL endpoint that uses Virtuoso Openlink Software (https://virtuoso.openlinksw.com/) that contains the ontology-based descriptions. Afterwards the response is sent back to the frontend in JSON format, and it can be presented to the user through diferent types of visualizations: a) HTML tables, b) Bar and Rose charts by using the NGX-Echarts library (https://xieziyu.github.io/ngx-echarts) and c) Chord charts through the D3.js library (https://d3js.org/d3-chord).
Code and Queries. The code for all the components of the portal and all the SPARQL queries for all the modes are available in https://github.com/mountanton/CIDOC-CRM_Portal.
The webpage ofers five interactive modes: i) Datasets Browsing, ii) Global Search, iii) Commonalities, iv) Most Frequent Elements, and v) Add Dataset. For each mode, Fig. 6 shows a screenshot with an example. Finally, a tutorial video can be accessed in https://youtu.be/ar8JEty94_w.
∙ Mode A. Datasets Browsing. This is the default mode and the user can browse statistics and visualizations for all the datasets (see the upper side of Fig. 6). In particular, one can browse ranking lists (using HTML tables) and visualizations through charts. By clicking on a single dataset, more information are shown for that dataset, including a description, a URL, its statistics and also the list of its properties and classes (including dedicated lists for CIDOC-CRM).
Use Cases. It mainly corresponds to the UC1 and can be useful for any user for discovering the most appropriate datasets for their needs. Secondarily, it is connected to UC4, since the CIDOC-CRM experts can exploit all the statistics for evaluating how the CIDOC-CRM model is used, e.g., to check about distributions of CIDOC-CRM properties and classes.
∙ Mode B. Global Search. The user can search for any property/class, and the portal returns all the datasets containing the desired property/class and the number of triples that they appear. For aiding the user, we provide autocomplete services and a drop-down list including all the CIDOC-CRM properties and classes. For instance, Fig. 6 shows: i) the datasets describing places (class “crm:E53_Place") and ii) the datasets including the property “crm:P52_has_current_owner".
Use Cases. It corresponds to the UC1 and can be useful for discovering datasets containing entities of a desired class (e.g., places, births) or property (e.g., “took place at", “carried out by").
∙ Mode C. Commonalities. The user can discover all the common properties and classes
between any pair of datasets, e.g., Fig. 6 shows the common CIDOC-CRM classes and properties
between the datasets “Sealit" [
Use cases. It refers to the UC3, and can be useful for the dataset owners to discover which datasets have the most commonalities with their dataset, i.e., for creating an integration service.
∙ Mode D. Most Frequent Elements. Here, the objective is to find the most frequent properties and classes according to a) the number of datasets or b) the number of triples that they appear, e.g., see the example in Fig. 6 including the most popular CIDOC-CRM classes.
Use cases. It corresponds to the UC4, since it can be useful for the CIDOC-CRM experts for analyzing the distribution of the properties and classes.
∙ Mode E. Add Dataset. The objective is to enable the addition of new datasets for any dataset owner. Indeed, one can fill and submit a form (see the lower right part of Fig. 6) including some very basic details of the dataset. For avoiding spamming issues, the form is first evaluated by the administrators of the web portal and then the process of Fig. 3 is performed.
Use cases. It corresponds to the UC2 (Data Publishing), since any CIDOC-CRM dataset owner can fill the form for requesting to add their dataset to the portal.
Total Number of Collected (CIDOC-CRM) Datasets Triples Entities Triples with a CIDOC-CRM property Triples with a CIDOC-CRM instance
Here, we provide indicative statistics and measurements concerning the 30 collected datasets, which can also be browsed and visualized in https://demos.isl.ics.forth.gr/CIDOC-CRM_Portal/.
First, Tables 1 and 2 provide some general statistics (total and average numbers) about the 30 collected datasets, based on the computed ontology-based descriptions. From the 560M triples of all the datasets, 168M triples contain a CIDOC-CRM property (approximately 30% of all triples) and 300M triples a CIDOC-CRM instance (approximately 53.5% of all triples).
Properties and Classes. Table 2 shows that each dataset uses on average 37.7 CIDOC-CRM properties and 19.3 CIDOC-CRM classes. Fig. 7 shows the exact number of CIDOC-CRM properties per dataset. Indicatively there are 7 datasets using ≥ 60 CIDOC-CRM properties, whereas only 3 datasets use ≤ 10 CIDOC-CRM properties. As regards the classes, Fig. 8 shows that 25 datasets use ≥ 10 CIDOC-CRM classes, while 12 datasets use ≥ 20 CIDOC-CRM classes.
Percentage of Triples (per dataset) using CIDOC-CRM properties and Instances. Fig. 9 shows for each dataset the percentage of triples containing a CIDOC-CRM property; indicatively 20 datasets use CIDOC-CRM properties in at least 30% of their triples. Concerning the instances, Fig. 10 depicts for each dataset the percentage of triples containing a CIDOC-CRM instance (i.e., an entity that is a member of a CIDOC-CRM class), and we can observe that half of the datasets (15 out of 30) include a CIDOC-CRM instance in at least 80% of their triples.
Here, we provide some indicative measurements about the CIDOC-CRM properties and classes.
Distribution measurements. The distribution of CIDOC-CRM properties and classes (according to the number of datasets that they appear) is shown in Figures 11 and 12, respectively. Concerning the properties, we observe a power-law distribution, i.e., some few CIDOC-CRM properties are used from many datasets, whereas most of them in a few datasets. Indicatively, only 29 CIDOC-CRM properties are used from ≥ 10 datasets, whereas 100 properties are used by two or a single dataset. Regarding the classes, most of them are also used by a low number of datasets, i.e., see Fig. 12. Finally, from the 309 properties and the 76 classes of the current CIDOC-CRM version, there are 100 properties (59 of them are inverse properties) and 12 classes that are not used by the collected datasets, i.e., the 32.3% of properties and the 15.7% of classes.
Most Popular CIDOC-CRM Properties and Classes. We show the most popular CIDOCCRM properties and classes according to the number of a) datasets and b) triples, that they appear. Fig. 13 shows that the most popular properties are “crm:P14_carried_out_by" and “crm:P2_has_type" that appear in 25 datasets. Concerning the number of triples, i.e., see Fig. 14, again the property “crm:P2_has_type" is the top one, appearing in 88M triples. Regarding the CIDOC-CRM classes, the most frequent one is the “crm:E53_Place", i.e., see Fig. 15, which appears in 23 datasets, whereas the class occurring in the highest number of triples (i.e., having the most instances) is crm:E54_Dimension with 11M triples (see Fig. 16).
We presented a portal that focuses on the ISO Standard CIDOC-CRM, for enabling the
browsing and visualization of ontology-based descriptions of any CIDOC-CRM-based dataset. We
described several use cases, all the details about how the statistics are computed, and the modes
of the portal. We ofered measurements about 30 real CIDOC-CRM datasets, which revealed a
power-law distribution; some few CIDOC-CRM properties and classes are widely used, whereas
most of them are used by a few datasets. As a future work, we plan to a) compute/visualize more
complex statistics (e.g., triple/path patterns since they can be exploited for Question Answering
[
Acknowledgments. This work has received funding from the European Union’s Horizon 2020 coordination and support action 4CH (Grant agreement No 101004468).