During the past few years a significant number of agriculture-related research and development projects have been implemented by receiving funding from the European Commission. All these projects have aimed to address specific problems and have produced solutions documented in the digital objects created in their context. However, the uptake of the existing information and knowledge by the concerned stakeholders is not yet adequate. On the other hand, the divergence of the available digital objects in terms of their types and formats poses significant challenges in homogeneously describing them with metadata. The Horizon 2020 EUREKA project aims to make a contribution towards this direction by developing a FAIR digital repository based on a semantic data model. The present document focusses on the design decisions related to it, as well as the rationale for the need to develop FAIR digital repositories based on formal data models.
In the seven years of implementation of the Horizon 2020 Framework Programme, the European Commission invested nearly one billion euros in projects related to agriculture, forestry, and rural development1. These projects have created a large number of digital objects conveying valuable information and knowledge about best and innovative practices. However, the uptake and re-use of these digital objects has not yet been realised to an adequate extent. In addition, most of these digital objects are no longer available after the end of the projects. As a result, there is little potential for the various agricultural stakeholders to have access to the available knowledge for further research and solution development. This is the context in which the EUREKA project (https://www.h2020eureka.eu/) is being situated. More specifically, EUREKA’s goal is to build a meta-repository (i.e., the “FarmBook”) of the digital objects created by the EU-funded, Horizon 2020 multi-actor projects. By this way, EUREKA aspires to make a contribution towards ensuring the longevity of the knowledge generated in those “source” projects and therefore, allowing it to be further circulated and used as the basis for future developments. However, in this efort, there are significant challenges to address mostly relating to the heterogeneity of the digital objects to be integrated into the meta-repository.
Semantic web technologies and standards [
In the rest of the article, an overview of the issues related to the adoption of the FAIR principles for developing FAIR digital repositories is provided. Then, our methodology and the semantic data model are presented. Finally, some concluding remarks are provided together with accounts of the steps to be taken next in EUREKA.
The FAIR data principles have been developed as guidelines for good data sharing practices,
initially focussing on the publication and re-use of scientific data sets [
To turn these ideas into reality, we need to establish a FAIR data ecosystem [
The FarmBook data model’s design has been based on the methodology of Noy and McGuiness
[
STEP 1 - Determine the domain and scope of the ontology: The overarching goal of the EUREKA FarmBook’s semantic data model is to provide formal descriptions of the concepts, the concept relations, and the properties of concepts needed to be considered for defining the metadata to homogeneously characterise the breadth of digital objects available from the EU-funded, Horizon 2020 multi-actor projects. The design of the data model has been driven by the following, indicative, competency questions: • What are the subjects (topics) of the digital objects available from project X? • Which are the geographic locations related to the content of the digital objects of type X? • What are the digital objects of type X on subject Y, related to the geographic location Z (e.g., the videos on pig farming in oak forests in Spain)?
STEP 2 - Consider re-using existing ontologies: The external ontologies that have been considered, in terms of re-using definitions of concepts, relations, and concept properties, for the design of our semantic data model, are Schema.org, FRAPO, FOAF, and DCMI . These are generic ontologies selected with the aim to help us model basic descriptive features of a digital object relating, among others, to the format in which information is conveyed (text, audio, video, etc.), as well as provenance (e.g., the digital object’s creator and source multiactor project). We are indeed familiar with agriculture-related controlled vocabularies (e.g., AGROVOC) and ontologies (e.g., FoodOn, AgrO), and deeply acknowledge the added value of re-using their components as part of the development of a formal data model. These structures 2http://obofoundry.org/principles/fp-000-summary.html will be considered next in EUREKA for the needs of creating a flexible model of the subject(s) addressed in the various digital objects.
STEP 3 - Define the classes and class hierarchy and STEP 4 - Define the properties of the classes: The details of the definition of the model classes, their hierarchy, and the class properties are presented in the next section.
The FarmBook’s semantic data model is shown in Figure 1 below. The digital object available by a multi-actor project, the project providing it, its creator, and the geographic location mentioned in its content are key concepts needed to be captured in the model. A digital object’s subject may be a term defined in (and thus, imported from) agriculture-related controlled vocabularies (e.g., AGROVOC) or ontologies (e.g., FoodOn, AgrO), and is denoted with a rectangular shape following a modelling convention similar to that in FoodOn. We consider more than one such structures because of the divergence in the "source" projects’ focus.
Table 1 shows the alignment between the concepts captured in our model and the components re-used from existing ontologies. Table 2 provides information about the relations captured in our model and the ontologies they have been imported from.
To describe the full extent of the digital object categories needed to be considered for the FarmBook’s design, a set of relevant concepts, organised in appropriate class hierarchies, have been imported into our semantic data model from the Schema.org ontology. These concepts are: (i) TextDigitalDocument and PresentationDigitalDocument (being subclasses of the DigitalDocument class); (ii) AudioObject, ImageObject, and VideoObject (being subclasses of the MediaObject class); (iii) SoftwareApplication; and (iv) Dataset. DigitalDocument, MediaObject, SoftwareApplication and Dataset are all subclasses of the CreativeWork class, which has, in turn, been modelled as a subclass of the frapo:Output class.
This is a decision based on the fact that there are project outputs not considered as “creative work” (e.g., the network of partners established in the project). This hierarchy is illustrated in Figure 1 by viewing the diagram from right to left (i.e., moving from a lower to a higher level of abstraction in regard to the definition of our model’s classes). The import of specific OWL classes and properties (namely, owl:Restriction, owl:Class, owl:onProperty, and owl:minCardinality), has allowed to infuse “heavier” semantics into our model and explicitly define cardinality restrictions over specific relations (e.g., the dct:creator relation as shown in the code snippet below).
By capturing a number of digital object properties, modelled with the help of the schema: {description, keywords, inLanguage, potentialAction, url, fileSize, dateCreated, license} and the dct: {title, format, type} properties, as well as properties of the “source” project modelled with the help of the schema: {name, alternateName, url} properties, it has been made feasible to assemble the FarmBook’s semantic data model and, finally, come up with a set of metadata considered both rich and flexible enough (including administrative, structural, and provenance information) to efectively cover the needs of the EUREKA FarmBook’s target end-users (namely, farmers, foresters, and advisors).
Despite the considerable investments of the European Commission in agriculture-related research and development projects, the concerned stakeholders cannot easily find and access relevant information and knowledge. In addition, there is a need for good data sharing practices. The broad spectrum of formats in which data and information is conveyed makes imperative a shift of focus from "FAIR datasets" to “FAIR digital objects” and “FAIR repositories”. The EUREKA FarmBook’s data model, designed upon well-acknowledged semantic web standards, is an initiative that paves the way towards the development of FAIR repositories of heterogeneous, agricultural digital objects. The semantic data model proposed will be used to define a set of metadata for the annotation of the digital objects, currently available and collected from the various multi-actor projects, which will be housed in the FarmBook. Moreover, it is also intended to serve as a reference model, adopted by future projects, for the description of the variety of the data and information that will be generated in their context and become available through their own, custom repositories. This will facilitate a uniform, standards-based approach in the characterisation of agricultural digital objects and enable their broader uptake and re-use.
The next steps in EUREKA involve the development of a graph-based structure for modelling the subjects addressed in the FarmBook’s digital objects, based on the re-use of components of agricultural ontologies (e.g., FoodOn, AgrO) and controlled vocabularies (e.g., AGROVOC). Furthermore, the procedures for ingesting digital objects in the FarmBook will be established. This will allow to test in practice the utility of our model and the metadata set derived from it.
The work presented in this documented has taken place in the Horizon 2020 EUREKA project receiving funding from the EU under the No 862790 Grant Agreement.