Scholarly article on topic 'Moving from dataset metadata to semantics in ecological research: a case in translating EML to OWL'

Moving from dataset metadata to semantics in ecological research: a case in translating EML to OWL Academic research paper on "Computer and information sciences"

CC BY-NC-ND
0
0
Share paper
Academic journal
Procedia Computer Science
OECD Field of science
Keywords
{"ecological datasets" / EML / OWL / OBOE / ontology / CERIF}

Abstract of research paper on Computer and information sciences, author of scientific article — Elena Mena-Garcés, Elena García-Barriocanal, Miguel-Angel Sicilia, Salvador Sánchez-Alonso

Abstract The Ecological Metadata Language (EML) is an XML-based metadata specification developed for the description of datasets and their associated context in ecology. The conversion of EML metadata to an ontological form has been addressed in existing observation ontologies, which are able of providing a degree of computational semantics to the description of the datasets, including the reuse of scientific ontologies to express the observed entities and their characteristics. However, a number of practical issues regarding the automated translation of the available EML datasets to a representation with formal semantics and its subsequent integration into Research Information Systems (RIS) require separate attention. These issues include expressing meaning by using existing terminologies, the mapping of EML with models of research information and the mapping with mainstream metadata schema. This paper describes the approach taken for that purpose in the VOA3R project, describing the main mapping and translation decisions taken so far and some common pitfalls with metadata records as they are currently available through the Web.

Academic research paper on topic "Moving from dataset metadata to semantics in ecological research: a case in translating EML to OWL"

Available online at www.sciencedirect.com

ScienceDirect

Procedía Computer Science 4 (2011) 1622-1630

International Conference on Computational Science, ICCS 2011

Moving from dataset metadata to semantics in ecological research: a

case in translating EML to OWL

Elena Mena-Garcésa, Elena García-Barriocanala, Miguel-Angel Siciliaa*a and Salvador

Sánchez-Alonsoa

a University of Alcalá, Computer Science Department, Ctra. Barcelona km. 33.6 - 28871 Alcalá de Henares, Spain

Abstract

The Ecological Metadata Language (EML) is an XML-based metadata specification developed for the description of datasets and their associated context in ecology. The conversion of EML metadata to an ontological form has been addressed in existing observation ontologies, which are able of providing a degree of computational semantics to the description of the datasets, including the reuse of scientific ontologies to express the observed entities and their characteristics. However, a number of practical issues regarding the automated translation of the available EML datasets to a representation with formal semantics and its subsequent integration into Research Information Systems (RIS) require separate attention. These issues include expressing meaning by using existing terminologies, the mapping of EML with models of research information and the mapping with mainstream metadata schema. This paper describes the approach taken for that purpose in the VOA3R project, describing the main mapping and translation decisions taken so far and some common pitfalls with metadata records as they are currently available through the Web.

Keywords: ecological datasets, EML, OWL, OBOE, ontology, CERIF

1. Introduction

Ecology as a science evolves by testing hypotheses about natural processes and patterns against observation data. The resulting datasets are increasingly being shared openly and the practice of describing them though metadata is becoming widespread among researchers and institutions. However, the meta-description of the datasets is in many cases not formal enough for semantic discovery and automated integration (Madin et al., 2007). Observation or analysis replication can be done automatically in some cases in which "high quality" metadata is available but some pitfalls and inconsistencies in usage of metadata schemas prevent this to be consistently supported (San Gil, Vanderbilt and Harrington, 2011). This has raised the need for data schemas with stricter requirements and semantics, including the use of formal models as those represented with ontologies based on description logics - a relevant example is the OBOE ontology (Madin et al., 2007).

a Miguel-Angel Sicilia. Tel.: +34 91 885 66 40. E-mail address: msicilia@uah.es

1877-0509 © 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of Prof. Mitsuhisa Sato and Prof. Satoshi Matsuoka doi:10.1016/j.procs.2011.04.175

Madin et al. (2007) described the role ontologies as formal representation of concepts could play in the advancement of the science of ecology. In their words "terminological ambiguity [that] slows scientific progress, leads to redundant research efforts, and ultimately impedes advances towards a unified foundation for ecological science". These authors conclude considering that ontologies are able to fill an important gap "by facilitating powerful data discovery based on rigorously defined, scientifically meaningful terms. By clarifying the terms of scientific discourse, and annotating data and analyses with those terms, well defined, community-sanctioned, formal ontologies based on open standards will provide a much-needed foundation upon which to tackle crucial ecological research". However, while ontologies are available for the just mentioned aims, the mainstream used models for sharing data descriptions on the Web are based on metadata schemas that are less stringent in the way the descriptions are provided.

The open Ecological Metadata Language (EML) is a metadata schema that provides a vehicle for sharing datasets that is widely used in ecological research. Further, the OBOE ontology (Madin et al., 2007) allows for a formal representation of observations that has already been used for the practical purpose of representing EML-described datasets. These provide the required schema for expressing observational data, and connect to other life science ontologies that serve as vocabularies to express the observed entities and the concrete attributes measured (e.g. "crop kernel/seed harvested at crop harvest in kg/ha"). The translation of EML to OBOE can be done by some mapping rules that could be subject to automation, but there is not a complete and unambiguous translation between the two schemas that preserves all the semantics. Indeed, many EML records use text sentences in many of its elements that cannot be translated automatically to a formal representation, as the semantics are implicit in natural language. For example, an entity description as "2000 chemical data for all AK Lakes" in a dataset describing different lake parameters should be mapped to an ontology element representing lakes as Lake in the SWEET 2.1 set of ontologiesb. This is a difficult task as the descriptions are not normalized and text or natural language matching algorithms may work only to a level of effectiveness. In practice, many EML records available are not using controlled vocabularies or taxonomic standards, but providing values as text without normalization, which hampers the possibility of semantically integrating and contrasting data from datasets produced at different locations with different conventions.

In consequence, integrating datasets into a Research Information System (RIS) with formal semantics requires a number of tasks that go beyond the translation of EML records to instances using the OBOE schema or another similar ontology. These tasks include the automatic translation of EML to OBOE data, the association of EML information elements with existing terminologies or ontologies and the mapping of the datasets with existing RIS models. OBOE has been demonstrated to be flexible enough to represent various EML schemas, especially due to its flexibility in modeling observations that serve as "context" for other observations. However, the interpretation of what contexts are and how they are structured is for many EML records impossible to derive via automated means, even though humans can in most cases perfectly derive it manually.

This paper reports on the experience on integrating EML datasets into a complete RIS based on the CERIF model that has been deployed in the context of project VOA3R, an EU funded effort aimed at integrating scholarly content including datasets. While the project focuses on agriculture and aquaculture, which are not the most common domains for EML datasets available today, the technical issues described are of general applicability to other systems more focused in ecological information. The approach described in this paper is that of trying to reuse the existing mass of EML-described datasets via automated ingestion into a central aggregation system. In consequence, the way of translating EML and linking with other models are only "best effort" and cannot be guaranteed to be semantically complete or correct for all the cases, for reasons as the ones described above. However, they represent a starting point for refining them and allow for analyzing how an automated translation works and which additional information should be provided in EML to make that translation more accurate.

The rest of this paper is structured as follows. Section 2 provides background information on metadata for datasets containing ecological information and ontologies that can be reused to formally represent that metadata. Section 3 explains the translation process from EML records taken from the LTER (Long Term Ecological Research) sites to the OBOE ontology model establishing a generic mapping between elements from EML and elements from the OBOE ontology. Then, Section 4 describes a concrete case study of application in the field of

b http://sweet.jpl.nasa.gov/

agriculture and aquaculture within the context of the ICT PSP Project VOA3R. Finally, conclusions and outlook are provided in Section 5.

2. Background

In the field of ecological research there is a need of getting a more powerful and flexible way to capture the semantics of complex datasets, as well as the possible relationships among them. In this context the efforts made by the LTER Network focused on providing to the scientific community a common basis to represent synthesized data through the 26 LTER sites available online, each of them in charge of different research approaches and ecosystems but conforming a network with a common mission. In order to solve this disintegration in the representation of the metadata of each site, LTER adopted the EML schema for the entire network. The EML was developed thanks to the efforts of researchers and information managers from several institutions aiming at producing a standard for describing ecological long-term research. In 2003 the LTER network started to use EML as the standard for the LTER metadata. Therefore, EML became the vehicle for interoperability through the different metadata providers inside LTER network.

EML is a metadata specification implemented as a set of XML (Extensible Markup Language) modules that enables the documentation of ecological data in a modular and extensible way. This specification is made up of several modules; each of them is defined to describe a part of the essential information for describing ecological data, as well as their recommended format. EML is intended to avoid or reduce the ambiguity by formalizing this kind of information into a comprehensive and standardized set of terms specifically selected for ecological data. The key principles of EML include the following: modularity, detailed structure, compatibility with other metadata standards, strong typing and ruling. However, as with many metadata schemas, flexibility and general applicability are reflected in the accommodation of different alternatives for expressing the same information and in a degree of optionality and recourse to textual descriptions that results in records of a very heterogeneous nature and thus lacking normalization.

The Science Environment for Ecological Knowledge (SEEK) project0 has developed and formalized critical aspects of EML in the Extensible Ontology for Observation (OBOE). The main approach followed was to extend EML ideas to allow the semantic annotation of ecological data sets using ontologies. OBOE has also been developed within the Semantic Tools for Data Management (SEMTOOLS) projectd for describing a wide range of ecological datasets stored within the Knowledge Network for Biocomplexity (KNB) as well as extensions for ontology based data annotation and discovery within the MetaCat software infrastructuree. OBOE is an extensible set of ontologies represented in a formal ontology language (OWL-DL) that serves as a way to describe scientific observations and opens up the possibility of sharing, integrating and discovering all the datasets even though their context are different, because OBOE is not domain specific. The major benefit of using OWL as the base language is the formalization of ecological terms through named classes where each member of a set is an instantiation of a class.

OBOE defines the scientific observation through the following core ontology classes: Observation, Measurement, Entity, Characteristic and Measurement Standard. This is a well-structured and hierarchical but generic approach. Moreover, as is depicted in the Figure 1 the original ontology core structure is based in some properties that connect the main concepts mentioned before.

OBOE was mainly developed to describe scientific observations and measurements emphasizing on getting the observational context that is really important when the dataset information is used in searching processes. For that reason, it provides the required level of structure that is needed for systems integrating heterogeneous datasets and other kinds of scholarly content. It is not covering the full range of metadata included in EML, and instead focuses on the representation of the observational data.

c http://seek.ecoinformatics.org/ d https://semtools.ecoinformatics.org/ e http://knb.ecoinformatics.org/knb/metacat

Figure 1: The main classes (ellipses) and properties (arrows) of the Extensible Observation Ontology (OBOE) taken from (Madin et al., 2007). Grey ellipses represent the extension point for domain-specific ontologies and the square ones simple data types.

3. The translation process

The reuse of the mass of data available described with EML into systems using ontology-based semantics requires a previous mechanism for the translation of metadata. The translation process described here starts from the parsing of EML documents, and a subsequent process of translation involving a relational database and, in a second stage, translation into ontological form. These EML documents can be found online for example by searching through the different LTER sites, and can be downloaded to act as example input data in the translation process. A relational database is intended to provide a centralized repository for the datasets into a common data format. Only data table entities are considered in the current version, spatial rasters or vectors and other data types are currently not supported. Further, not all the elements in the current EML 2.1 version are considered, only the ones mentioned in what follows are captured in the translation process. Figure 2 depicts the main elements involved in the process.

Figure 2: Architecture of the solution for ingestion and transformation of EML records

After parsing the EML files, a relational database representation is used to store the selected EML data values and describing metadata in different tables, e.g. general, person, attribute and dataset_values. This task is implemented by the EMLFileProcessor Java class that takes a set of EML files and its associated dataset files

as input and stores the translation of the dataset in the relational database. Then, the OBOEConversion Java class performs the queries needed to obtain the information about each identified Observation to be added to the OBOE ontology as a new instance with its related values. This is implemented by using the OWL-APIf. The use of an intermediate relational database serves the purpose of being able to include datasets that were not described in EML, so that the OBOEConversion utility is independent of the schema. The approach to an automated translation proceeds with the following general rules:

• An Observation instance is created to represent the dataset as a whole.

• An Observation instance is created for each <dataTable> element and these are connected to the instance representing the dataset using the hasContext property.

• The individual rows in each data table are mapped to individual Observation instances which are connected to the instance representing the data table using the hasContext property.

• Each of the attribute values becomes an instance of Measurement that is connected to the information representing the row using the hasMeasurement property.

This provides a basic, generic mapping mechanism that is sufficient for many cases. However, there are other cases in which further interpretation is required to come up with a representation that faithfully captures the details of the research carried out. For example, some multi-table datasets include a <eml-constraint> section that defines integrity constraints between entities (e.g., data tables) as they would be maintained in a relational management system. In case these are provided, a different translation behavior is required. For example, if two data tables are related by a foreign key constraint, then a join operation in the two tables would be used to produce a single observation for each of the rows in the relation resulting from the join.

An example of the translation process is depicted in Figure 3, which presents an excerpt of the translation of the dataset. The dataset translated is the "Agronomic Yields (1989 to present)" with DatasetID KBS020 that can be found in the LTER Data Catalog. Concretely, it represents an excerpt of one of the values in the second data table with <entityName> "Agronomic Alfalfa Yields". The automated ingestion process takes words in entity names and attempts to find entities in the sub-vocabulary "common names for plants" in the AGROVOC thesaurus maintained by FAO and used in the project. "Alfalfa" corresponds to the non-descriptor code 8 7 91, and following the use relation, the descriptor "4693 - Medicago sativa (EN) " is found. The same procedure produces mappings with the names of the other crop types in the first of the data tables in the same dataset. The metadata contained in VOA3R includes any type of scholarly content, which include references to papers. The AGROVOC thesaurus is used for categorizing all the content items (including automatic classification for papers that are lacking these in the metadata). Even though AGROVOC is not a formal ontology (Sini et al., 2008) and thus does not provide full formal semantics, it is widely used and comprehensive in the field of agriculture addressed by VOA3R, and other ontologies can be mapped to their terms (Sánchez-Alonso, 2009). In consequence, the ingestion process attempts to identify AGROVOC terms as particular kinds of OBOE Entity and Characteristic instances. While the automated mapping is based on text-matching and thus can not provide reliable computational semantics, it is useful for searching datasets based on these terms and for relating datasets among them or with other kinds of scholarly content. In Figure 3, an instance of AgrovocEntity (a subclass extending OBOE's Entity class) is used to represent the AGROVOC mapping. The case of the measurements is different, as AGROVOC is not defining sub-vocabularies for characteristics, and the mappings based on text-matching are less likely to be correct. The

Name subclass of Characteristic is used for attributes containing values that are found in taxonomic systems, as is the case of the crop attribute in the example. For the general case in which there is no possibility to identify the kind of attribute being referred, instances of Characteristic are used. The case of the attribute "dry yield in metric tons per hectare " is one of these latter undetermined characteristics, as with the current vocabularies used it is not possible to map the textual description neither with physical characteristics in OBOE (as MassDensity) nor to elements in AGROVOC that are close to the intended meaning. This calls for the development of vocabularies that normalize types of characteristics common in some domains.

f http://owlapi.sourceforge.net/

Figure 3: Excerpt from an example observation in the "Agronomic Alfalfa fields" dataset.

4. Case study: in integrating with research in agriculture and aquaculture

Datasets are a particular form of scholarly content that requires an integrated but differentiated treatment from other kinds of content, and typically from bibliographical resources. Current approaches in integrating scholarly content rely on the harvesting of content metadata followed by a process of mapping or reconciliation of metadata schemas. Notably, this is the approach taken in the data model of the Europeana digital library (Doerr et al., 2010). The VOA3R project uses that kind of harvesting approach combined with the OAI PMH (Van de Sompel et al., 2004) protocol as a standardized mechanism, and translating metadata records into an ontology representation that allows for using enhanced computational semantics. In this section, the main issues of the integration of observation metadata coming from EML records into the model used in VOA3R are briefly described.

The first step in the mapping process is representing each EML instance in the form of one of the generic metadata models inside VOA3R which are able to ingest Dublin Core (among other schemas). This can be easily done as there is a proposed EML to Dublin Core mapping availableg, even though it is generic and provides only a kind of "approximate mapping". In addition, the VOA3R system is implementing the CERIF model. CERIF (Common European Research Information Format) is a formal model to setup Research Information Systems and to enable their interoperation (Jeffery, 2010), maintained by EuroCRISh and recommended by the European Union to its Member States. The CERIF 2008 version is organized around three core entities: Person, Project and OrganizationUnit, and three result-related entities: ResultPublication, ResultPatent and ResultProduct.

Table 1 summarizes the mapping decisions from EML to Dublin Core taken beyond what was explicitly specified in the proposed mapping.

g http://knb.ecoinformatics.org/software/eml/eml-2.0.1/eml-dublinCore.html h http://www.eurocris.org/

Table 1. Mapping of EML to Dublin core (additional mappings)

Dublin Core element EML equivalent

Additional details on the mapping

dc:creator

dc:subj ect

eml-resource:creator

eml-resource:keywordSet

dc:publisher eml-dataset:publisher

dc:contributor eml-

dataset:associatedParty

In VOA3R a subset of CERIF model will be used to identify users with different roles in the system and they will be assigned to the DC metadata element as an URL that matches with an existing entity in our CERIF model, this approach matches correctly with the EML data type ResponsibleParty.

This metadata field will link with other classification thesaurus such as AGROVOC and other domain ontologies such as a "research method" ontology developed within VOA3R, as well as keywords sets that reflect or provide information about the context of the dataset.

Same case as in creator.

Same case as in creator.

After the metadata is translated, there is an internal process of detecting candidate mappings. For example, the eml-resource:creator specified with the ResponsibleParty data type might be considered to be matched with a VOA3R user (e.g. by matching his/her email address). In such a case, the string description of the creator will be internally substituted by a reference to the RDF representation of the particular user, e.g. http://voa3r.eu/users/32. This serves as a form of authority control, enables internal navigation, filtering and in general, provides computational semantics to the metadata coming from disparate EML files. It is also providing the necessary URI identification of entities to export information as linked data (Bizer, Heath and Berners-Lee, 2009).

In addition to the above, there are other EML modules and concrete elements that have an interpretation inside the VOA3R system.

Table 2. Mapping of EML modules with elements in the VOA3R representation

EML module VOA3R element Details

eml-literature Citations ingested in VOA3R from EML is following EndNote conventions, which can also

institutional repositories using Dublin Core be mapped to existing general metadata schemas as

and MODS. Dublin Core or bibliographic schemas as MODS.

eml-protocol A research method or research procedure as EML includes protocol and methods sections. The

described in the VOA3R ontology. distinction is that the use of the term "protocol" is used

in the "prescriptive" sense, and the term "method" is used in the "descriptive" sense, the specification says that "This distinction allows managers to build a protocol library of well-known, established protocols (procedures), but also document what procedure was truly performed in relation to the established protocol."

eml-project Maps to the CERIF core entity Project. In CERIF, instances of Project can be connected to

Person or OrganisationUnit, which corresponds to the personnel element in EML. The rest of the EML project elements have also equivalents in the CERIF model.

eml-coverage Maps to existing internal taxonomic A common identification schema is used for formal

(taxonomic) identifiers. taxonomic names.

The approach to a common identification is based on using public identifiers. The Encyclopedia of Life (EOL) (Wilson, 2003) provides an identification of species and taxa that is linked to other classification systems that are widely used, as the NCBI Taxonomy and the Integrated Taxonomy Information System (ITIS) among others. As the EML is providing an structured schema for taxonomic coverage, the mapping can be done at several levels. For example, the description of the LTER dataset "Whole Lake Manipulations: Rainbow Smelt Removal" covering the Sparkling Lake (Vilas County, WI) includes full taxon values for the species that are subject to observation, in this case Osmerus mordax. The species taxon can be mapped to the one in EoL, as shown in Figure 4. In that case, the EoL identifier 357 054 for the page can be used to uniquely identify the coverage and to access additional information. This approach works for all the properly annotated EML records that contain taxonomic coverage.

Figure 4: View of the Osmerux mordax multi-classification and content in the EoL.

5. Conclusions and outlook

The combination of the EML schema with formal ontologies like OBOE provide the required baseline for building research information systems with formal semantics that allow for computational tasks entailing analysis or interpretation of datasets produced at different locations. However, the translation from EML to an ontological form cannot in all the cases be done in a way that all the information included in the original markup are translated into the formal representation. This is due to aspects as the interpretation of context that are not explicitly represented and to the lack of a standardization of vocabularies and ontologies for describing entities and attributes.

This paper has reported a practical way of ingesting EML metadata into a research information system using formal representations. The translation mechanisms have been described, and the main elements pertaining to the integration with the models of research context have been discussed, including specifics of mapping projects with the CERIF model and other elements as taxonomic coverage, protocols or connections to existing thesauri. Future work should deal with the problem of obtaining richer automated translations from EML, but this may require some additional conventions on the actual usage of EML, possibly including additional elements that help automated translators in the task of correctly interpreting the semantics of the different data pieces that are reflected in the data tables.

Acknowledgements

The work leading to these results has received funding from the European Commission under grant agreement n° 250525 corresponding to project VOA3R (Virtual Open Access Agriculture & Aquaculture Repository: Sharing Scientific and Scholarly Research related to Agriculture, Food, and Environment), http://voa3r.eu.

References

I. C. Bizer, T. Heath, and T. Berners-Lee. Linked data - the story so far. International Journal on Semantic Web and Information Systems (IJSWIS), 5(3): 1-22. (2009).

2. M. Doerr, S. Gradmann, S. Hennicke, A. Isaac, C. Meghini and H. van de Sompel. The Europeana Data Model (EDM). World Library and Information Congress: 76th IFLA general conference and assembly (2010).

3. K. Jeffery. The CERIF Model As the Core of a Research Organization. Data Science Journal, 9 (2010).

4. J. Madin, S. Bowers, M. Schildhauer, S. Krivov, D. Pennington and F. Villa. An ontology for describing and synthesizing ecological observation data, Ecological Informatics 2: 279-296 (2007).

5. J. Madin, S. Bowers, M. Schildhauer, and M. Jones. Advancing ecological research with ontologies. Trends in Eco. and Evol., 23 (3):159-168 (2008).

6. I. San Gil, K.Vanderbilt and S. Harrington. Examples of ecological data synthesis driven by rich metadata, and practical guidelines to use the Ecological Metadata Language specification to this end. International Journal of Metadata, Semantics and Ontologies 6(1). (2011).

7. S. Sánchez-Alonso: Enhancing availability of learning resources on organic agriculture and agroecology. The Electronic Library 27(5): 792-813 (2009).

8. M. Sini, B. Lauser, G. Salokhe, J. Keizer and S. Katz. The AGROVOC Concept Server: rationale, goals and usage. Library Review, Vol. 57 Iss: 3, pp.200 - 212 (2008).

9. H. Van de Sompel, M. L. Nelson, C. Lagoze and S. Warner. Resource harvesting within the OAI-PMH framework. D-Lib Magazine, 10(12), 1082-9873 (2004).

10. F. Villa, I. Athanasiadis, A. Rizzoli. Modelling with knowledge: A review of emerging semantic approaches to environmental modelling, Environmental Modelling & Software, 24(5):577-587 (2009).

II. E. O. Wilson. The Encyclopedia of Life. Trends in Ecology and Evolution, Vol. 10, pp-77-80 (2003).