Testbed-12 Semantic Enablement Engineering Report

The optimal provisioning of metadata is still subject to debate. A number of standards are available to express metadata, all of them focusing on particular requirements, serialization models, or encodings. The virtual goal is the development of a metadata structure that allows the discovery of all other objects, services, and metadata that are associated to any single object. This approach requires a graph structure with associations between individual objects that are themselves first class citizens, i.e. objects with properties and potentially further links to other objects (see previous chapter on links between resources in addition to the Testbed-12 report on the General Feature Model).

Traditionally, metadata standards in the Geospatial domain have focused on either data or service descriptions, but rarely addressed links between objects or associations to particular object portrayal or other processing services. To overcome these shortcomings, Testbed-12 analyzed a number of metadata standards including W3C standard DCAT, DCAT-AP, GeoDCAT-AP, ADMS, Project Open Data 1.1, Dublin Core, ISO 19115, and ISO 19119 to identify the common and relevant metadata information needed for search and discovery. Testbed-12 also identified gaps in existing standards to provide a complete dataset description, including service, portrayal information, schema and schema mapping. These efforts resulted in the development of the Semantic Registry Information Model (SRIM), and the Semantic Registry Service. SRIM is defined as superset of DCAT and its existing application profiles (DCAT-AP, GeoDCAT-AP, ADMS) and introduces a superclass of dcat:Dataset called srim:Item. The reason not to use DCAT exclusively is the strict focus of DCAT on data; excluding other elements such as e.g. services or map layers. SRIM enables the integration of different metadata providers (CSW, CKAN, POD, WMS, WCS) by providing a common core vocabulary to describe resources (data, services, vocabularies, maps, layers, schemas, etc.) and to accommodate specificities of each source by leveraging the built-in extensibility mechanism in OWL. SRIM has been defined in UML and serialized in RDF and JSON. The JSON serialization has been closely aligned with the Project Open Data metadata schema 1.1 standard. However, to accommodate some of the requirements needed by the Semantic Registry Service it has been extended and partly modified.

The Semantic Registry Service implements the Semantic Registry Information Model and allows acting as a proxy between a client and any number of catalog services. Passing on clients' search requests to OGC CSW instances requires the dynamic query-rewriting to bridge between the various service metadata encodings models and formats and the Semantic Registry Service JSON encoding supported by the service’s REST API. This aspect needs to be implemented and further explored in future testbeds. The Testbed-12 implementation focused on a harvesting approach, where metadata elements from CSW services have been harvested and locally converted in order to fit the Semantic Registry Information Model.

The initial implementation of the Semantic Portrayal Service during the OGC Testbed-11 focused on defining styles, portrayal rules, point-based symbols and graphics to enable Web Processing Services to produce SLD documents. Testbed-12 now broadened the focus to include in particular symbology styles for lines, areas, and texts based on existing standards such as Symbol Encoding and Styled Layer Descriptor, SVG, ISO 19117, and KML. All portrayal information is captured in a set of microtheories, including a style ontology, a symbol ontology, a graphic ontology, and a portrayal catalog ontology. All semantic portrayal information (i.e. styles, rulesets, symbol-sets, and symbols) has been made available through a hypermedia-driven REST-based API.

The semantic mediation work in Testbed-12 was closely related to the semantic portrayal work described above and built on the achievements from Testbed-11. OGC 15-058, Symbology Mediation Engineering Report from Testbed-11 describes the basic principles of semantic mediation using the example of two portrayal ontologies that need to be aligned in an ad hoc manner. Testbed-12 now focused on the usage of a schema registry to store information about schemas and schema mappings to support ad hoc transformations between source and a target schemas, see chapter semantic registry. Schema mappings can be considered a simple form of semantic mediation, but go without explicit formalization of the underlying semantic knowledge required to map from one schema to another. For that reason, the idea was to design Semantic Mediation Service REST API and integrate it with the Semantic Registry and CSW ebRIM profile for Schema Registry.

The OGC approach to support data and service discovery is the Catalog Service for the Web (CSW). The current versions of the CSW are supporting syntactical and schematic interoperability. CSW support searches by temporal and spatial dimensions, keywords, and well defined terms organized in the service taxonomy (e.g. search for a specific service type); an approach that is insufficient for automatic service discovery based on data contents. Service brokers that mediate between a client and catalogs or other services can only partly mitigate that problem, despite being loaded with additional knowledge (which is a very complex and cumbersome task).

In particular the keyword-based search approach, which uses a lexical comparison between search and target terms, often leads to poor discovery results because the keywords in the query may be semantically similar but syntactically different, or syntactically similar but semantically different from the terms in a Web service description. Thus, traditional keyword-based search approaches are inherently restricted by the ambiguities of natural languages.

One of the main shortcomings in automated service discovery and invocation of OGC compliant Web services is the lack of shared semantics about any specific service. A WMS provides maps with any number of layers, but each layer is described very briefly (if at all) using free text. If a user cannot interpret the layer named jõed or the short description suurimaid jõgesid Eestis, it is not possible to derive any further knowledge from that service. Whereas maps consist of a number of layers and might still serve some meaningful purpose even though semantically weak, other services such as WPS that interface literally any type of geospatial processing are not usable without clear semantics.

Semantic annotation was subject of research roughly 10 years ago, with Roman et al (2007), Tanasescu et al (2007) or Lutz (2007) suggesting ontologies to describe OGC Web services. All these approaches forwent using information provided directly by the Capabilities-document. Others such as Maué (2008) have addressed the aspect of missing semantics by adding links to the capabilities document. These links annotated data objects with semantics from a domain ontology and functional aspects with semantics from Web service process ontologies. In contrast to these approaches, Stock et al - and reused in the Aviation Semantics work - suggest to store and augment the capabilities content in a Web service ontology to assist discovery, execution, and orchestration of these services.

The W3C specification OWL-S is used to describe Web services with strong semantics that allow discovery and reasoning. OWL-S specifies ontologies to generically describe any Web service, but requires extensions to cover domain-specific aspects, e.g. the geospatial characteristics of OGC Web service interfaces. This is complementary to the existing OGC approach using Capabilities-documents to describe technical aspects and limited semantics of Web services without much room for semantic reasoning.

The Aviation Semantics Engineering report takes on previous work done by Stock et al and suggests the representation of OGC-compliant geospatial Web services using ontologies. The general idea is to populate the ontology with data from the Capabilities-document provided by OGC Web services.

It has to be noted that the previous research referenced before describes OWL-S as a cumbersome way to describe Web services with significant repetition. The OWL-S model with its three main parts: the service profile, the process model and the grounding has been proven onerous in practice, which is particularly true for the input and output parameter descriptions. The suggested solution limits their full description to the process model with references in other places. This far more practical solution is technically not correct, as it violates the OWL-S formal model. Further issues are based on the nature of OWL and its limited means to express cardinality without massive proliferation of properties, which is particularly relevant to non-atomic parameters, e.g. Query. All these experiences lead to a mixed picture of the suggested approach. It seems that it might make more sense to restrict ontology-based concepts to the discovery phase and use alternative methods to describe parameter details.

The aviation thread made good experiences extending an existing ontology with GeoSPARQL classes and properties to allow for reasoning on geospatial characteristics. The goal was to evaluate if OGC specific elements such as GetCapabilities elements can be added to existing ontologies. The thread successfully developed a method for the semantic representation of OGC-compliant geospatial web services using an equivalent OWL ontology as a potential extension to the WSDOM ontology. WSDOM is a formal Web Service description ontological model that is being developed by FAA for use in building applications that process and exchange web service information. WSDOM was designed based on the OWL-S service ontology and the service taxonomies used by FAA. The WSDOM 1.1 release consists of six OWL files and three RDF files. For further details see OGC 16-039.

Catalogue services support the ability to publish and search collections of descriptive information (metadata) for data, services, and related information objects. Metadata in catalogues represent resource characteristics that can be queried and presented for evaluation and further processing by both humans and software. Catalogue services are required to support the discovery and binding to registered information resources within an information community. However, the current OGC catalogue service specification only supports taxonomies, which means it is not suited for semantic based service discovery.

In order to understand the needs for service discovery in the domain of aviation, the aviation thread developed a number of use cases that include aviation services to be described and discovered using semantic technologies. The use cases are related to common capabilities provided by traditional aeronautical information systems. The major difference is in a new assumption regarding the System Wide Information Management (SWIM) - inside SWIM, numerous aviation services might be provided by various authorized service providers. Such diversity requires the services to be properly described in order to be successfully discovered and consumed. With highly flexible SWIM anticipated, the aeronautical data retrieval becomes more complex. Many aviation information services will be available in SWIM and the complete knowledge about characteristics and capabilities they provide (operational, data types, spatial coverage), as well as the implementation details, will not always be explicitly available. In other words, the availability of service metadata consolidated in a centralized SWIM catalogue/registry for the purpose of service discovery is an important precondition for successful service consumption. The aviation thread settled for use case development only. Future activities need to address this topic in more detail.

With the deployment of the Semantic Web and Linked Open Data we have not only multiplied the data sources but also the machine-processable controlled vocabularies that structure and constrain the interpretations of these data. These controlled vocabularies can be ontologies (RDF Schema, OWL), codelists, taxonomies, thesauri (SKOS) sometimes augmented with additional rules (SPARQL/SPIN rules, SWRL, RIF), and constraints (SHACL). Controlled vocabularies provide a better way to organize knowledge for subsequent retrieval.

.image source semanco-tools.eu

Vocabulary directories now exist (e.g. Linked Open Vocabularies, LOV), but there is an ever-increasing demand for environments that simplify searching, editing and collaborative contributions to the vocabularies by non-experts of the Semantic Web. This creates a tension between state-of-the-art, very rich formalisms and methods for modeling vocabularies and a need to democratize and decentralize participation in the life cycle and usage of controlled vocabularies.

Vocabularies are most likely to be adopted and shared if they are made available easily. Nevertheless, despite successes in the use of SKOS for encoding vocabularies, current standards provide only low-level interfaces to vocabulary data. For example, many vocabularies are published as an RDF document for download. However, if the vocabulary is large, then the download will be commensurately large; if the user only wants to retrieve a single vocabulary term or select a few terms, this option requires processing on the client side. Alternatively, access to vocabularies is often provided at a SPARQL endpoint. SPARQL is the generic RDF query language. While it is powerful, it is also considered a low-level language similar to the relational database query language SQL that is normally only used by database administrators.

Some SKOS vocabularies are published via other HTTP interfaces. However, each implementation uses different protocols and supports a varied set of features (e.g. content-negotiation provided by the GEMET REST interface and NERC Data Grid’s Vocabulary Server SOAP interface). In some cases, one or both of human-readable formats and machine-readable formats are not available. Thus, discovery and access across vocabulary endpoints becomes challenging and ad-hoc.

There is a clear opportunity here to design an API to match the SKOS and OWL vocabularies, taking advantage of the fact that most modern vocabulary content is structured using SKOS and OWL classes and predicates. This API can then be used as the basis for various higher-level vocabulary applications (NLP applications, Concept Recommender, Semantic Enricher, etc.) that can be used to enrich for example ISO 19115 metadata and other OGC services using controlled vocabularies in their metadata.

As stated above, experiments with real world models and ontologies need to be expanded in future testbeds. Tests run as part of the ShapeChange UML-RDF/OWL/SKOS work produced very promising results using the NATO Geospatial Real World Object Index.

Further on, current experiments with tools such as ShapeChange have shown a number of issues concerning UML to ontology mapping that need further research. This includes

.image source: eazysafe.com

Multilingual environments often struggle with common semantics due to different usage of terms in the various languages. Though technologies such as RDF support multilingual properties already, the feature is currently not used very often. In particular in infrastructures such as the Arctic Spatial Data Infrastructure, where many data sets are available on the national scale at the eight adjacent nations, represent ideal test cases to further experiment with multilingualism and semantic mapping.

Constraints-languages such as OCL allow constraining the mapping from one model representation/serialization to another. During the mapping process from UML to OWL, OCL could for example be used to constrain OWL constructs. These technologies should be added to UML converter tools such as ShapeChange to improve interoperability and reasoning capabilities. This applies to constrained property types, values, geometries, or times.

Experiments have been started with RDF and JSON-LD, though no full comparison of both approaches has been performed yet. JSON-LD is a lightweight syntax to serialize Linked Data in JSON. As the name suggests, it supports linking in and between datasets, much like Xlinks in GML data.

A key aspect that JSON-LD adds to JSON is semantic tagging of data elements. Like in XML, each element can be assigned to a namespace. This allows clients to identify the exact meaning of an element, especially if there are multiple elements that happen to have the same name. JSON-LD context documents are used to provide the necessary information. A context document typically references terms from one or more vocabularies or ontologies. These terms then provide the semantics of elements in the actual JSON data. In this context, a number of questions arise that need additional research and experimentation, such as:

With Hypertext Application Language (HAL), JSON-LD, Hypermedia-Driven Web APIs (Hydra), or Siren: a hypermedia specification for representing entities, there exist a number of hypermedia formats that allow linking resources. Future IP initiatives should investigate these formats and develop recommendations on how to integrate which type into the OGC meta architecture.

Precise semantics of association types between resources are essential for any set up of resource servers that use links between resources. Testbed-12 partly addressed this topic as part of the REST User Guide and REST Architecture Engineering Report. To establish any semantically sound data serving infrastructure, a reliably available and clearly defined set of link types that is aligned with other initiatives such as IANA link relations is absolute essential. The OGC, under the supervision of the OGC Naming Authority, should establish a registry of typical link types for geospatial applications.

Lots of base work has been done by the joint OGC-W3C Spatial Data on the Web Working Group. Future testbeds should identify a number of key elements and progress the work of the working group with implementations and further research. The important elements include RDF Data Cube to define thematic, spatial and temporal dimensions, the the Vocabulary of Interlinked Datasets, VoID as an RDF based schema to describe linked datasets, and their integration with shared vocabularies and the various OGC Web service interfaces as discussed above.

Ontologies describe semantics. Reasoning on RDF data can be performed based upon the information found in ontologies. This can lead to new information and knowledge. Pure validation of RDF data is another use case. Validation can verify that a given dataset is compliant to a specification. This is of interest whenever a data publisher and consumer have agreed to exchange information compliant to a certain specification. This is even more important when there are many publishers and consumers (think of information exchange on and between the communal, regional, national, and international level).

SHACL, the Shape Constraint Language, defines the shape of the graph and allows integrity checks of graph-based structures. SHACL has been identified in Testbed 12 as a key candidate to support the missing validity and integrity checks for OWL/RDF based serializations of information models.

ISO 19115 uses a hierarchical data model that is not perfectly suitable for graph-based technologies. Future activities should investigate how ISO 19115 could be better aligned with the current best practices for Linked Data publication in general and the SRIM in particular.

The Testbed-12 Semantic Registry Service implements the Semantic Registry Information Model and allows acting as a proxy between a client and any number of catalog services. Passing on clients' search requests to OGC CSW instances requires the dynamic query-rewriting to bridge between the various service metadata encodings models and formats and the Semantic Registry Service JSON encoding supported by the service’s REST API. This aspect aspect needs to be implemented and further explored in future testbeds. The Testbed-12 implementation focused on a harvesting approach, where metadata elements from CSW services have been harvested and locally converted in order to fit the Semantic Registry Information Model. In future, support for arbitrary OGC services (providing access to e.g. feature collections, maps, map layers, coverages, and other objects) should be added and tested. This requires extensions to the Semantic Registry Information Model, see below.

Investigate a profile for Layer and Map that extends the RegisteredItem and relates to Datasets, Services and Portrayal Information developed for the Semantic Registry and Semantic Portrayal Service.

For this testbed, the Semantic Registry harvested information from a federation of CSW services as the focus was to exercise the Semantic Registry Information Model (SRIM) and the REST API. For the next testbed, efficiency could be improved by investigating the publish/subscribe protocol and versioning management of the register items in the Semantic Registry. This approach is complementary to the dynamic query rewriting discussed above. Both approaches have their advantages and disadvantages that need to be explored in further detail.