Can data shape the future of mental health support? – The Guardian 20160907

From: Can data shape the future of mental health support? – The Guardian 20160907

Open data is being used to design resources for people with mental health conditions to help them find the right support

Head on digital screen.

If you’re experiencing a mental health issue, one of the people you probably least want to speak to about it is your employer. Disclosing depression or anxiety has long been seen as the last workplace taboo, for fear of repercussions. This is despite the existence of the Equality Act 2010, which protects employees with physical and mental disabilities from discrimination.

But just over a third of workers with a mental health condition discuss it with their employer, according to a survey of 1,388 employees carried out by Willis PMI Group, one of the UK’s largest providers of employee healthcare and risk management services. The research found that 30% of respondents were concerned that they wouldn’t receive adequate support, 28% believed their employer wouldn’t understand, and 23% feared that disclosing it would lead to management thinking less of them.

A culture of fear and silence can have a huge impact on productivity – the charity Mind estimates [pdf] that mental ill health costs the economy £70bn a year. The challenge is that seeking help involves taking ownership of the problem, says Mark Brown, development director of social enterprise Social Spider and founder of the now defunct mental health and wellbeing magazine One in Four. And finding support online can be a time-consuming and frustrating experience.

“Just serving up ever great slabs of information – the internet is awash with it – isn’t going to help anyone to know what to do,” says Brown. “We often confuse the provision of information with the solving of problems. Knowing information is different from knowing how to put that information into action.”

Brown believes that bringing together information with public and open data into a single digital space is one way that could innovate how advice is delivered.

Plexus is aiming to achieve just this. Built by the digital studio M/A, with funding from the Open Data Institute, the knowledge base is being used to design resources for people with mental health conditions, their families, and even employers, to find support available in local areas, seek advice on how best to cope with returning to work after a period off and understand employee rights and employer responsibilities.

Plexus has pooled data from a couple of dozen organisations including NHS Choices, Department for Work and Pensions, the Office for National Statistics and Citizens Advice. In some cases the information has been pulled from APIs; in other instances it has been scraped using web data platform

The first tool Plexus developed is a chatbot called Grace, which is currently in beta testing. It enables users to record thoughts and feelings anonymously, receive feedback in the form of a newsletter and log in to an online dashboard to see a more detailed analysis, including whether there are any patterns in mood emerging over a period of time. The tool also offers guidance from the various governmental and charity websites under easy-to-navigate sections, such as legal rights and preparing for work.

“Through machine learning, Grace will intuitively know when our users are mostly likely to want to speak with us, be able to see the positive and negative nature of the user’s reply, and adapt the questions to encourage more positive responses,” explains Martin Vowles, creative director and co-founder of M/A. “We’re hoping this approach will allow us to offer a unique tool to each user which helps them understand and develop their mental wellbeing.”

Brown says that the potential for machine learning to tailor information and services is exciting. “It’s very good at looking at big piles of data for patterns. When we know certain things to be correct from one dataset, it can begin to make guesses about lots of other things based on what the machine is being fed.”

The sensitive nature of data being submitted by users on a platform like Grace, though, means many people are likely to be uneasy about their data being made accessible. To get round this, Plexus allows users to decide how their data is shared, with data licences lasting between 13 and 26 weeks. Vowles hopes that “as users become more trusting of Grace and what it can do for them, they’ll become more trusting with [us] using their anonymised personal data.”

Plexus aims to release a series of open datasets, including qualitative, quantitative and information on resources accessed by Grace users, to enable NGOs and local authorities to understand the country’s mental health provision. It’s hoped that they’d then use the knowledge to devise new strategies and ensure targets are met and resources and services available in local areas are of an acceptable standard.

There are also plans to make certain data available to employers, but “this has to be on the employee’s terms”. Vowles imagines that involving employers in the process of receiving support could allow them to get a clearer picture of mental health in the workplace. They could then adapt to make employees feel more comfortable and ensure their business has adequate support in place.

The potential to use open data to shape how future mental health support is delivered is an area that has been underexplored. At the end of last year, the Royal Society of Arts launched an interactive platform with Mind that allows members of the public to find out how well local health providers are looking after people with mental health conditions. The full dataset is available to download and includes data extracted from Public Health England, as well as metrics such as percentage of people with a mental health condition in employment in local areas. Plexus, however, is the first tool to use open data with the aim of providing people with a holistic view of their mental wellbeing.

Brown supports the idea of using open datasets and combine them, but stresses that any tool or platform has to benefit users. The data and information must be digestible and it needs to help them understand and take away from it what they need.

“It’s often extremely easy to forget that people with mental health difficulties are people first and foremost – not objects or problems.”

Linked USDL – Introduction

After Pedrinaci, C, J Cardoso, and T Leidig – Linked USDL – A vocabulary for web-scale service trading – 2014


The importance of real-world services, that is business activities of a mostly intangible nature (e.g., life insurance, consulting), has grown over the last 50 years to dominate economic activity [1]. Because of their intangible nature, services can often be bundled, adapted, and traded in an automated manner. In an attempt to exploit the Web as a service trading platform a number of service marketplaces have emerged, ranging from purely technical registries like UDDI [2], to business-oriented marketplaces such as Google Helpouts. Technical registries have for the most part focused on the computer science aspects of services which is limiting as it ignores fundamental characteristics of services including the economic, social, and business contexts [3]. Business-oriented marketplaces on the other hand have focussed on providing silos that offer limited search and comparison capabilities through an essentially human-oriented storefront [4]. As a result, common and essential economic activities in the service sector such as the generation of customised offerings, the creation and trading of possibly cross-domain and multi-provider service bundles, or simply the communication between customer and provider remain largely manual activities [4].

Supporting the trading of services over the Web in an open, scalable, and automated manner enabling the opportunistic engagement of multiple crossdomain providers requires a shared means for capturing and reasoning upon the economic, social, and technical aspects governing service exchanges [1,3,4]. The Unified Service Description Language (USDL) is the most comprehensive attempt in this direction but it has received limited adoption due to its complexity, while it also exhibited limitations with respect to the level of extensibility and automation supported. In this paper we present Linked USDL, a new vocabulary which builds upon the results and experience gained with USDL combined with prior research on Semantic Web Services, business ontologies, and Linked Data to better support Web-scale automated service trading. We present the methodology and main decisions adopted for transforming the complex USDL specification into a network of vocabularies that is anchored on simplicity as well as on vocabulary and data reuse. The resulting vocabulary is thoroughly evaluated in terms of domain coverage, suitability for purpose, and its current level of adoption.

Related Work

Service Science aims to reach a better understanding of services, service networks, value co-creation and service innovation, to name a few of the main research topics [1]. These efforts, which encompass several disciplines, are geared towards establishing solid foundations for advancing our ability to design, create, and analyse service systems with both business and societal purposes in mind.

Relevant work in Computer Science includes service-oriented systems, which approach the development of complex applications by integrating networked software components called Web services [2]. This area has been prolific in terms of both tooling and specifications including a number of approaches for describing technical services semantically, e.g., OWL-S, SAWSDL, and WSMO [5,6]. Although (semantic) Web services work provides advanced support for discovering or composing technical services, it disregards the fundamental socio-economic context of real-world services (e.g., value chains and offerings), and does not cover the widespread manual services (e.g., consulting) [3]. Complementary work on Workflow and Business Process Management has focussed on the operationalisation of the processes within enterprises [2,3,5], which has more recently also incorporated human activities [7]. This work is, however, centred on a procedural view on how activities are carried out within an organisation which is orthogonal to the business characteristics of the services offered (e.g., speed of internet connection offered) which are essential to service trading.

The most notable effort able to represent and reason about business models, services, and value networks is the e3 family of ontologies which includes the e3service and e3value ontologies [3,8]. This research has, however, not been much concerned with the computational and operational perspectives covering for instance the actual interaction with services. Likewise, the technical issues related to enabling a Web-scale deployment and adoption of these solutions were not core to this work. GoodRelations [9] (GR) on the contrary is a popular vocabulary for describing semantically products and offerings. Although GR originally aimed to support both services and products, it is mostly centred on products to the detriment of its coverage for modelling services, leaving aside for instance the coverage of modes of interaction, or the support for value chains.

USDL [4,10] is, to date, the most comprehensive approach to supporting the description of services for automated processing. USDL consists of 9 modules modelled in eCore capturing services, interaction interfaces, pricing models, service level agreements, and related legal issues. Despite its comprehensive support, this effort underestimated the need for such an all encompassing model to be widely open, highly flexible and extensible, and yet simple in nature [11]. On the one hand, the rather centralised and controlled nature of the approach led to an overly complex model hard to grasp and apply. On the other hand, eCore exhibited technical limitations towards its extensibility and its use as a lingua franca on the Web where Linked Data and light semantics are currently considered a more adequate technology.

Requirements Analysis

Informed by research carried out on services, including the related work covered earlier, we have elicited a number of requirements that Linked USDL and any other language or vocabulary with such an ambitious purpose should address. This includes notably coverage requirements, which we shall cover first. We also present additional criteria that we identified during the standardisation activities of USDL as potential issues and limitations for its Web-scale adoption [11].

Description Requirements

One of the essential difficulties when dealing with services beyond mere technical interfaces, is the fact that they are at the intersection of many diverse disciplines that range from technical aspects, to operational ones, socio-economic concerns, or even legal issues. Being able to move across each of these domains is essential to support the trading of services online. We detail the main dimensions next.

Functionality Services are business activities that normally take place through (possibly technology mediated) interactions between stakeholders, resulting in benefits to the actors involved. Fundamental to the notion of service is therefore its functionality in terms of what it does, requires, and provides. Given the highly diverse nature of services this should cover the entire spectrum from fully automated provisioning (e.g., Spotify) to those essentially manual (e.g., car repair service). Depending on the stakeholder, the level of abstraction could vary from a detailed operational view (provider), to a high-level one for customers.

Agents and Networks Services delivery engages several stakeholders in (possibly ephemeral) ad-hoc business networks, e.g., banks often engage in partnerships with insurances to provide accounts with integrated travel insurance. The modelling of services should seamlessly support both the emergence and analysis of such networks in order to enable the dynamic co-creation of value through Web-wide service trading. Important aspects to be covered are thus the agents involved in a certain network and the role(s) they play.

Service Relationships Thanks to their intangible nature, services can be combined, repurposed, and adapted to better fulfil customer needs. Services are often related to other services and products. For instance, services can often be enhanced with others, or there can be variations over established types. Services are often bundled, i.e., aggregated and offered jointly in packages like broadband and TV services. And in the case of automated services, services may be composed according to specific data and control flow to achieve a complex objective out of simpler components.

Operational and Delivery The delivery of services is often subject to restrictions or conditions. These may range from geographical concerns (e.g., the insured individual should live in the UK), temporal availability, legal issues, variable pricing, and so on. From a service provider operational perspective, there may well be limitations due to the resources required, e.g., staffing, that need to be tracked as they determine the costs and the capacity for providing a service.

Consumption Services are most often accessed or “consumed” through interactions by means of designated communication channels. For example, making an insurance claim may require the customer to phone the insurance company, or fill up a form online. These communication channels may vary during the service delivery process (e.g., initially claim by phone and check the progress online), and there may exist restrictions on how interactions should take place. For instance a car repair service may require you to bring the car to the garage whereas in other cases the service may take care of sending a mechanic within some geographical boundaries.

Language Requirements

In addition to the aforementioned coverage requirements, research in the area has highlighted further requirements that the language should meet. First and foremost given the complexity of the domain and the fact that the aim is to maximise to the extent possible the level of automation that can be achieved during the life-cycle of services, the modelling of services needs to rely on a conceptual model with formal foundations that can enable automated processing [3, 10]. Nonetheless the language should be modular and extensible in order to be able to accommodate different domains and the many facets of services while minimising the complexity for users and tool developers.

Our subsequent work on standardisation highlighted that although necessary, these requirements did not appear to be sufficient for Web-scale adoption:

An Open Solution To support the engagement of any business entity across any domain the technological approach should be open. It should be open so as to allow anybody to engage and trade services online, as well as towards its evolution in order to cater for new requirements, accommodate new ways of doing business, or support new domains.

A Web-based solution A scenario like the one envisaged requires an approach that can support the engagement of millions of service providers and consumers in exposing, locating, interpreting, and contracting services. This necessarily calls for highly interoperable and scalable solutions in terms of data sharing, data processing, and communication protocols.

Promoting uptake While providing an open solution is likely to have a positive impact on technology take up, adoption will largely be determined by the simplicity with which any business entity could adopt a solution based on these technologies and the compatibility with existing legacy systems.

Linked USDL Vocabulary

Driven by the aforementioned requirements and informed by the drawbacks exhibited by USDL we worked on Linked USDL focussing essentially on reducing the complexity underlying USDL and fostering its wider adoption through the use of Web-centric technologies that are more amenable to extension, modification, and automation at large scale.

Design Decisions

First, due to the success, scale, growth, and current adoption of the Web for worldwide telecommunication and electronic commerce we believe that any technology hoping to enable service trading online should necessarily embrace and build upon the Web principles and technologies [12]. Notably Linked USDL should also embrace principles like i) the establishment of global identifiers, e.g., by using URIs to identify services and providers; ii) the use of links to other resources on the Web to enrich a particular datum with reusable and externally provided information, e.g., pointing to complementary services; iii) the use of HTTP as a simple uniform protocol for supporting interactions; and iv) the decoupling between resources and their representation. Doing so brings a technology stack that has proven to support large scale, efficient, multi-party interactions, as well as it directly provides an integration point with open, standard technologies that are already widely used and supported.

Second, to enable effective interactions at the business level, we need to provide standards that go beyond data transportation and syntactic representation [1].

To this end, Linked USDL embraces the use of formal ontology representation languages to capture the semantics of services such that they are amenable to automated reasoning. Linked USDL goes one step forward in the adoption of Web technologies to embrace the emerging standard approach for data sharing online, namely Linked Data [13]. Adopting these principles enables Linked USDL to capture, share, and interlink data about services of highly heterogeneous nature and domains, in an open, scalable, and uniform manner. Linked Data principles promote and support reuse which in turn helps to reduce the data modelling overhead (e.g., by reusing conceptual models and existing data sets), and maximise the compatibility with existing tooling. Both aspects are major challenges earlier versions of USDL faced which this work aims to alleviate.

Design Methodology

Following common Knowledge Engineering best practices [14], we aimed at creating a modular solution based on well-designed, widely adopted vocabularies that did not introduce substantial ontological commitments away from the core topics of interest. Thus, considerable effort was devoted to identifying and evaluating reusable ontologies.

First, we identified the main topics to be covered given the original USDL specification and determined some core terms characterising each of these topics. Informed by the topics and terms identified, we carried out both a manual and semi-automated search to determine potentially relevant reusable ontologies. On the one hand, we performed a state of the art analysis to identify ontologies that were relevant for the modelling of services, see [11] and Section 2. On the other hand, we used Swoogle [15], Watson [16], LOD Stats [17], and the Linked Open Vocabularies (LOV) engines to search for ontologies covering the main terms identified. For each of the queries asked, we kept the top 10 results. The resulting list was eventually enriched with widely-used general purpose vocabularies such as Dublin Core (DC) and Simple Knowledge Organisation Scheme (SKOS).

Second, for each of the vocabularies identified, we used both LOD Stats and LOV to figure out the number of datasets using these terms, the number of instances of the main concepts of interest present in datasets on the Web, and the number of times the vocabulary is reused elsewhere. The search for reusable ontologies provided us pointers to existing vocabularies of potential interest together with indications regarding their use and popularity. Table 1 shows the results obtained for the vocabularies for which there was at least one instance found on the Web[1]. Indeed, the statistics should not be taken as an exact value of the overall use of these vocabularies (e.g., GR is used more frequently than what is reflected by this analysis), but rather as a relative indication. Indeed we also took into account the properties defined by these vocabularies which are in some cases (e.g., DC Terms) the main constructs reused.

The design of Linked USDL was driven by these statistics, and a manual assessment of the quality, coverage, and potential alignments of the vocabularies.

Topic Vocabulary # Datasets # Instances LOV Reuse
Service GR 6 45 146 0 6
MSM 2 0 41,368 0 0
OSLC 2 0 2 0 0
COGS N/A 5 N/A 0 3
Offering GR 6 8 824 656 4
Location vCard (v3 & v4) 5 0 + 2 3,684 3,686 + 3 0 + 2
WGS84 11 1 3,204 1,7651 1
AKT Signage 18 0 11,789 0 0
DC Terms 1 9 39 39 6 1 5 1
Business Entities 2 4 1,570,778 1,570,778 3
FOAF 60 135 14,613 14,557 29
GR 1 N/A 3,918 N/A N/A
W3C Org. 1,050 11 2 1,050 2
Time W3C Time 9 N/A 236,433 N/A N/A

Table 1. Top Vocabularies per Topic.


Informed by the aforementioned analysis, Linked USDL, which is publicly available together with further examples in GitHub, builds upon a family of complementary networked vocabularies that provide good coverage of necessary aspects and are widely used on the Web for capturing their particular domains. In particular Linked USDL builds upon:

  • DC Terms to cover general purpose metadata such as the creator of a certain description, its date of creation or modification, etc.
  • The vocabulary has been modelled mostly using RDF/RDFS constructs and we have limited the inclusion of abstract foundational concepts, so as to attain a model that is simple enough for its adoption online. The reader is referred to [19] for indications on how this model could be mapped to a foundational ontology.
  • SKOS providing low-cost support for capturing knowledge organisation systems (e.g., classifications and thesauri) in RDF.
  • Time Ontology (Time) for covering basic temporal relations. The ontology allows us to capture temporal relationships such as before and during.
  • vCard vocabulary a vCard 4 compatible vocabulary to support providing location and contact information for people and organisations.
  • Minimal Service Model (MSM) [18] to provide coverage for automated service-based interactions including Remote Procedure Call solutions (e.g., WSDL services) and RESTful services.
  • GR [9] to provide core coverage for services, business entities, offerings, and products.

Linked USDL Core Schema
Figure 1. Linked USDL Core Schema.

As the core and initial module of a set of vocabularies for supporting service trading online Linked USDL Core, see Figure 1, aims to cover four essential aspects: offerings, services, the business entities involved in the delivery chain, and the actual interaction points allowing consumers to contract or trigger the benefits of contracted services.

Linked USDL extends GR which is nowadays the de facto standard vocabulary for publishing semantic descriptions for products. It is worth noting that although services are accommodated within GR, their coverage is rather basic at this stage. Extending GR enables linking services and products descriptions which is particularly useful since many products are often sold in combination with a service, e.g., a repair or replace service. Additionally, it also ensures that an initial alignment with the increasingly popular vocabulary is in place, for GR is already largely aligned to it.

Core Concepts

The most important concepts provided by Linked USDL are:

Service is a refinement of gr:ProductOrService and subsumes all classes describing service types. Examples of subclasses of Service could be “internet provisioning service” and “insurance service”. Instances of Service may define i) prototypical services part of a portfolio, e.g., “BT unlimited broadband service”, as covered by ServiceModel, ii) one-of services custom tailored for a potential customer, or iii) actually contracted services, e.g., “your concrete life insurance provided by AXA”, as covered by gr:ServiceIndividual.

ServiceModel is a refinement of gr:ProductOrServiceModel which specifies common characteristics (e.g., download speed) of a family of services. ServiceModel thus defines families of Services sharing common characteristics, e.g., “BT unlimited broadband services share the characteristic of supporting unlimited download”. An actual service instance shares the properties of its service model. This is a feature that requires non-standard reasoning which specific implementations should take care of.

ServiceIndividual is a subclass of gr:Individual and Service. Instances of ServiceIndividual are actual services that are creating value to a network of business entities. For instance, “your concrete life insurance provided by AXA” is a ServiceIndividual which is providing value to yourself and AXA.

ServiceOffering is a subclass of gr:Offering and represents essentially offerings by a business entity including at least one Service. ServiceOffering may have limited validity over geographical regions or time.

EntityInvolvement is introduced in Linked USDL in order to enable capturing service value networks. In a nutshell, Entity Involvement allows capturing a ternary relationship expressing that a business entity, e.g., “AXA”, is involved in a service, e.g., “basic life insurance” playing a business role, e.g., “provider”. Linked USDL provides a reference SKOS taxonomy of basic business roles that covers the most typical ones encountered such as regulator and intermediary.

InteractionPoint link services to interactions that may be possible or required between the members of a service value network and the service during its life cycle. This allows answering questions such as “what is the sequence of interactions I may expect if I want to make an insurance claim and what communication channels are available to that end?”.

CommunicationChannel is the class of all different communication channels that business entities could use for communication. Linked USDL covers the most widely used channels by means of 2 vocabularies: vCard (e.g., email, phone), and MSM (e.g., Web services, and RESTful services). Communication channels are additionally characterised by their interaction type. Linked USDL provides 2 reference SKOS taxonomies covering the main modes (e.g., automated) and the interaction space (e.g., on-site).

EntityInteraction links interaction points to business entities or types (e.g., provider), and the role they play within the interaction (e.g., initiator). EntityInteraction allows expressing things like “to make a claim, the consumer should first contact the insurance provider and provide the policy number”.

Classifications Classifications or taxonomies of entities are most often used when describing services to capture, for instance, service types, business entity roles, e.g., “provider”, as well as interaction related issues, e.g., “manual vs automated”. We also expect that classifications will be needed in forthcoming modules addressing strategic issues or the internals of delivery chains.

This could be approached directly using subclassing which is directly supported by RDFS. However, the use of a hierarchy of classes establishes strict relationships which may not adequately match existing organisation schemes. For this reason, in Linked USDL we have accommodated the use of SKOS, which enables capturing classification schemes and taxonomies. Indeed, this mechanism does not prevent users from providing their own domain-specific categorisations through subsumption if they wish to. This approach thus enriches Linked USDL with a powerful, yet flexible and extensible means for creating categorisations.

The current version of Linked USDL includes three SKOS schemes with reference categorisations for BusinessRoles, InteractionRoles, and InteractionTypes, see Figure 1. These schemes have been, however, kept as separate modules so that different schemes can be used if necessary.


Coverage Evaluation

Ontologies are often evaluated by comparing them to a gold standard ontology [20]. In our case, we have done such an evaluation by comparing the resulting model to USDL, the most comprehensive model available for describing services. Doing so allows us to get a clear indication of the overall coverage of the domain, and to identify as well the main deviations from USDL.

A fundamental goal of this work is providing a conceptual model that would be easy to grasp, populate, process, and ultimately be adopted for Web-scale use. Thus, out of the 9 modules of USDL we have essentially deferred covering the following modules: Service, Legal, Service Level, and Pricing. Nonetheless, for every module we have checked the coverage of the main concepts defined in order to get an indication of both module-specific and the overall coverage of Linked USDL. [The results of this analysis are summarised in Table 2 of the original paper].

This analysis shows that thanks to integrating an reusing existing vocabularies we have managed to cover the vast majority of USDL, by providing a vocabulary consisting of 12 concepts and 3 complementary SKOS categorisations. In particular, from an original specification with 125 concepts we cover 74%, if we limit ourselves to the specific modules we targeted, and 60% overall, which shall contribute towards reducing the overhead related to understanding and adopting Linked USDL. It is worth noting that out of the concepts not explicitly covered several are sometimes redundant (e.g., Condition is subclassed in many modules), or were seldom properly understood and used (e.g., Functions, Phases of interactions, Service Level Agreements).

Suitability for Tasks and Applications

Given that Linked USDL does not cover all concepts present in USDL it is worth assessing the impact of such decisions. [Again, see Table 2 in the original article]. In qualitative terms, the decisions adopted are such that Linked USDL does not currently provide support for capturing how providers deliver services in terms of resources needed, complex internal workflows, or strategic decisions (e.g., targeted markets). The reason for this is two-fold. First, such aspects are often not automated and when they are, providers already have mechanisms in place to this end. Second, these are private concerns that are orthogonal to the trading of services. Similarly, Linked USDL does not currently include support for conceptualising complex agreements including legal requirements and guarantees as these were barely used or understood by users. Finally, we have opted for a simple mechanism for capturing prices and have deferred to a separate module the modelling of more complex dynamic pricing that are less often used and usually remain private to the provider.

Despite these changes, Linked USDL provides advanced support for modelling, comparing, discovering, and trading services and service bundles. It provides means for tracking and reasoning about the involvement of entities within delivery chains which informs the trading and comparison of services as well as it enables the tracing and analysis of service value networks. It provides advanced support for automating the interactions between actors during the life-cycle of services. Additionally it includes support for capturing service offerings, for combining services and products (e.g., a car often comes with a warranty), and for applying temporal reasoning, which were not previously available. Finally, and most importantly, these activities can be achieved with a greater level of automation benefitting from automated reasoning and they can be performed on a Webscale across Web-sites and service providers thanks to capturing and sharing the semantics of services as Linked Data.

Empirically, the suitability of the language for supporting the automation of key tasks has been evaluated by two main means. On the one hand, we have reused and developed tools that provide support for these tasks, and, on the other hand, we are continuously applying Linked USDL in a number of domains. In terms of reuse, thanks to the adoption of existing Linked Data vocabularies,

Linked USDL benefits from general purpose tooling, e.g., SPARQL engines and RDF stores, but also from vocabulary-specific solutions. This notably concerns existing advanced machinery for discovering, composing, and invoking technical services (i.e., RESTful and WSDL services) described in terms of MSM [18].

Additionally, general purpose infrastructure has been developed specifically for Linked USDL. A Web-based Linked USDL editor is currently available to help providers to easily generate Linked USDL descriptions. There is also an advanced multi-party dynamic and open service marketplace developed in the context of the FI-WARE project, able to gather, combine, and exploit rich service descriptions from distributed providers to help match offer and demand. Notably the marketplace supports consumers in searching for service offerings, comparing them, and contracting them.

Finally, from the perspective of its suitability for supporting service trading across domains, Linked USDL is currently being applied in a variety of domains. For instance, in the field of Software as a Service we have explored the use of Linked USDL in conjunction with TOSCA[21]. Linked USDL was used to formalise, structure, and simplify the discovery and selection of services of the Web-based customer relationship management (CRM) platform SugarCRM, while TOSCA supported the automated deployment and management of the services.

Additionally this work helped us evaluate the extensibility of Linked USDL by integrating it with complementary third party specifications such as TOSCA. In the FI-WARE project Linked USDL is used to support a service infrastructure supporting service ecosystems in the cloud covering both the technical and business perspectives. The FINEST project aims to support the transport and logistics (T&L) ecosystem, in which many service providers collaborate in order to transport goods over what is referred to as a “chain of legs”. Therein Linked USDL is being exploited in the planning of chains of legs to support searching and matching transport service offerings in a transparent, distributed, and multi-party manner.

Across the diverse domains where Linked USDL is being applied (see list of projects next), it has proven to be a valuable resource as a means to provide shared and globally accessible service descriptions integrating both technical and business aspects. The genericity, modularity, and extensibility of the approach has enabled extending the vocabulary with dedicated domain-specific vocabularies in the areas of SaaS and T&L, while generic software infrastructure was easily reused across domains.

Vocabulary Adoption and Use

When evaluating ontologies and vocabularies, one aspect that is often taken into account is their adoption and use. This evaluation may be carried over the ontology itself and/or over the different ontologies that are imported. The former gives an indication of the acceptance and adoption of the ontology in its entirety whereas the latter provides a more granular assessment over the reused ontologies. In this section we mainly address the latter but also provide preliminary indications of the overall adoption of Linked USDL.

The methodology that was followed was centred on the reuse of widely adopted vocabularies. Table 1 presented earlier shows the main vocabularies that were identified through search engines, together with core indicators of their use on the Web. These figures highlight that Linked USDL is based on vocabularies that are the most used in their respective domains of interest. Only two exceptions exist, AKT Signage which was not adopted for it was not dereferenceable, and which is indirectly aligned via GR. This approach in turn reduces the potential overhead one would incur when using Linked USDL: frequently reused vocabularies are likely to have greater acceptance and support by people and existing systems.

Additionally, the availability of datasets with instances in terms of the vocabularies reused guarantees that new descriptions could reuse and link to existing resources, e.g., allowing the reuse of descriptions of companies. Doing so provides clear benefits from the perspective of data acquisition which was one of the main concerns Linked USDL was trying to address. Additionally, by linking to existing instances the data provided is enriched which may in turn enable further advanced processing as well as it may increase the discoverability of services.

Providing a substantial account of the adoption of Linked USDL would require a reasonable wait from its first release, which coincides with this publication. Nonetheless, Linked USDL is currently already in use within more than 10 research projects, namely FI-WARE, FINEST, Value4Cloud, Deutsche Digitale Bibliothek, MSEE, FIspace, FITMAN, FI-CONTENT, ENVIROFI, OUTSMART, SMARTAGRIFOOD, IoT-A, Broker@Cloud, and GEYSERS. These projects are using Linked USDL as the core vocabulary for describing services, contributing to validate the suitability, genericity, and extensibility of Linked USDL for different domains. This also highlights that despite its youth, Linked USDL is already witnessing a promising adoption.


Despite the importance of services in developed economies, the widespread adoption of world-wide electronic commerce over the Web, most service trading is still essentially carried out via traditional and often manual communication means. A fundamental reason for this is the difficulty for capturing the abundant information and knowledge governing services and their related transactions in a way amenable to computer automation. Out of the wealth of work around services, USDL is the most comprehensive solution proposed thus far for enabling (semi)automated service trading. Yet, work on its standardisation highlighted a number of limitations for Web-scale service trading.

We have presented Linked USDL, the next evolution of USDL centred on fostering its wider adoption and better automation support through the (re)use of Linked Data. Linked USDL has been developed following a methodology centred on maximising the reuse of existing vocabularies and datasets and minimising the complexity. The resulting vocabulary has been evaluated in terms of domain coverage, suitability for purpose, and vocabulary adoption.

Despite the good evaluation results obtained, Linked USDL is to be regarded as one step towards enabling Web-scale service trading, albeit a fundamental one. Further work is required for covering aspects such as complex dynamic pricing models and agreements which are common in certain domains such as Cloud services. Additionally, from the tooling perspective, developing advanced mechanisms able to support steps such as the negotiation between service providers and consumers, or the bundling of services would also be necessary. We expect in this last regard to take inspiration and adapt solutions developed for the e3 family of ontologies.

[1] These statistics were last retrieved in November 2013.

Next: Linked Service System USDL (LSS-USDL) – Transparency of services


  1. Chesbrough, H., Spohrer, J.: A research manifesto for services science. Communications of the ACM 49(7) (July 2006) 35.
  2. Papazoglou, M.P., Traverso, P., Dustdar, S., Leymann, F.: Service-Oriented Computing: State of the Art and Research Challenges. Computer 40(11) (2007) 38–45.
  3. Akkermans, H., Baida, Z., Gordijn, J., Peña, N., Altuna, A., Laresgoiti, I.: Value Webs: Ontology-Based Bundling of Real-World Services. IEEE Intelligent Systems 19(4) (2004) 57–66.
  4. Cardoso, J., Barros, A., May, N., Kylau, U.: Towards a Unified Service Description Language for the Internet of Services: Requirements and First Developments. IEEE International Conference on Services Computing (SCC) (2010) 602–609.
  5. Cardoso, J., Sheth, A.: Semantic e-workflow composition. Journal of Intelligent Information Systems (JIIS) 21(3) (2003) 191–225.
  6. Pedrinaci, C., Domingue, J., Sheth, A.: Semantic Web Services. In: Handbook on Semantic Web Technologies. Volume Semantic Web Applications. Springer (2010) 7. Oppenheim, D.V., Varshney, L.R., Chee, Y.M.: Work as a service. In: ICSOC’11: Proceedings of the 9th international conference on Service-Oriented Computing, Springer-Verlag (2011).
  7. Gordijn, J., Yu, E., van der Raadt, B.: e-service design using i* and e3value modeling. IEEE Software 23 (2006) 26–33.
  8. Hepp, M.: GoodRelations: An Ontology for Describing Products and Services Offers on the Web. In: Knowledge Engineering: Practice and Patterns. Springer (2008) 329–346.
  9. Oberle, D., Barros, A., Kylau, U., Heinzl, S.: A unified description language for human to automated services. Information Systems (2012).
  10. Kadner, K., Oberle, D., Schaeffler, M., Horch, A., Kintz, M., Barton, L., Leidig, T., Pedrinaci, C., Domingue, J., Romanelli, M., Trapero, R., Kutsikos, K.: Unified Service Description Language XG Final Report. Technical report (2011).
  11. Jacobs, I., Walsh, N.: Architecture of the World Wide Web, Volume One. W3C Recommendation (2004).
  12. Bizer, C., Heath, T., Berners-Lee, T.: Linked Data – The Story So Far. International Journal on Semantic Web and Information Systems (IJSWIS) (2009).
  13. Suarez-Figueroa, M.C., Gómez-Perez, A., Motta, E., Gangemi, A., eds.: Ontology Engineering in a Networked World. Springer (2011).
  14. Ding, L., Finin, T., Joshi, A., Pan, R., Cost, R.S., Peng, Y., Reddivari, P., Doshi, V.C., Sachs, J.: Swoogle: A Search and Metadata Engine for the Semantic Web. In: CIKM ’04: Thirteenth ACM international conference on Information and Knowledge Management. (2004).
  15. d’Aquin, M., Motta, E.: Watson, more than a Semantic Web search engine. Semantic Web 2(1) (2011) 55–63.
  16. Auer, S., Demter, J., Martin, M., Lehmann, J.: LODStats — an extensible framework for high-performance dataset analytics. In: EKAW’12: Proc. of the 18th international conference on Knowledge Engineering and Knowledge Management, Springer (2012).
  17. Pedrinaci, C., Domingue, J.: Toward the Next Wave of Services: Linked Services for the Web of Data. Journal of Universal Computer Science 16(13) (2010) 1694–1719.
  18. Ferrario, R., Guarino, N., Janiesch, C., Kiemes, T., Oberle, D., Probst, F.: Towards an ontological foundation of services science: The general service model. Wirtschaftsinformatik (February 2011) 16–18.
  19. Sabou, M., Fernandez, M.: Ontology (network) evaluation. In Suarez-Figueroa, M.C., Gómez-Perez, A., Motta, E., Gangemi, A., eds.: Ontology Engineering in a Networked World. Springer (2012) 193–212.
  20. Cardoso, J., Binz, T., Breitenbücher, U., Kopp, O., Leymann, F.: Cloud Computing Automation: Integrating USDL and TOSCA. In: 25th Conf. on Advanced Inf. Systems Engineering (CAiSE 2013). Volume 7908 of LNCS., Springer (2013) 1–16.

Linked Services for the Web of Data

After Pedrinaci, C and J Domingue, Toward the next wave of services: Linked Services for the Web of Data, 2010.


Web Services and Service-Oriented Architecture are lauded as a silver bullet for Enterprise Application Integration, implementation of inter-organizational business processes, and even as a general solution for the development of all complex distributed applications. Despite the appealing characteristics of serviceorientation principles and technologies, their uptake on a Web-scale has been significantly less prominent than initially anticipated [Davies et al., 09]. First and foremost Web services, despite their name, are hardly a Web-oriented technology [Vinoski, 2002] but rather one suited for enterprises which so far have been reluctant to publish functionality on the Web. Secondly, from a technical perspective, current technologies are such that software developers need to devote a significant effort to discovering sets of suitable services, interpreting them, developing software that overcomes their inherent data and process mismatches, and finally combining them into a complex composite process.

Semantic Web Services (SWS) [McIlraith et al., 2001] have long tried to overcome Web services limitations by enriching them with semantic annotations in order to better support their discovery, composition, and execution. Up until now, however, the impact of SWS on the Web has been minimal. In the Web, semantics are used to mark up a wide variety of data-centric resources but are not used to annotate online functionality in any form in significant numbers. In fact, although SWS technologies have already shown their benefits, e.g., in discovery [Pilioura and Tsalgatidou, 2009], research in the area has failed to take into account the socio-economic aspects devoted to the creation and annotation of services. First, research has mostly focused on devising highly expressive conceptual models and has given birth to a number of diverging and largely incompatible solutions. These efforts have essentially glossed over the complexity they introduce, the additional effort demanded of users, and they have brought additional heterogeneity to an already overwhelming stack of specifications. Second, SWS research has for the most part targeted WSDL/SOAP based Web services which are not prevalent on the Web [Davies et al., 09]. As a consequence, SWS is instead a niche technology only accessible to highly trained experts and the benefits obtained are most often not considered worth the additional investment.

In parallel, the Web is currently witnessing a dramatic change with the advent of Web 2.0 [O’Reilly, 2005] and Linked Data technologies [Bizer et al., 2009]. The former is “socialising” the Web, putting individuals at the core of the Web as both data producers and consumers. Web 2.0 technologies have shown that collaboration over the Web can produce outstanding results with a low cost, and it is also encouraging enterprises and institutions to offer their data and services publicly at a previously unprecedented scale and pace [Hendler and Golbeck, 2008, Chi, 2008, Davies et al., 09]. Second, Linked Data technologies, which derive from research on the Semantic Web, have given birth to the Web of Data, “a Web of things in the world, described by data on the Web” [Bizer et al., 2009]. The Web of Data, impelled by the current trend towards an open Web, has recently experimented an outstanding growth and currently provides publicly large amounts of interconnected data concerning a wide range of domains and described in terms of light weight ontologies for supporting automated processing [Bizer et al., 2009].

In this paper we explore the relationship between services and the Web of Data. We identify the potential benefits that can be obtained by adequately integrating these so far rather disconnected worlds. We anticipate that this integration will mitigate the existing limitations of both services and the Web of Data, giving birth to a new wave of services dubbed Linked Services, that will ultimately lead to an explosion in the publication and use of services on the Web. We outline how this integration could take place by using simpler vocabularies for describing services and through the adoption of Linked Data principles for publishing services on the Web. Finally, we outline how Linked Services will be able to provide the additional necessary building blocks for appropriately exploiting the wealth of information exposed in the Web of Data.

The remainder of this paper is organised as follows. First, we present the technological background around services and the Web. We then discuss why, in our opinion, the current situation can give birth to a new wave of services. We then present how the use of light weight semantics can allow us to bring services into the Web enabling their discovery through state of the art Linked Data technologies. Next, we focus on how services can contribute to the Web of Data both generating new data and processing existing one. Finally, we conclude the paper and outline key topics for further research.

Background and Related Work

The current technological landscape is characterised by a number of highly complementary technologies that have so far remained disconnected. In this section we review existing work in the area of Web Services, Web 2.0, Semantic Web, the Web of Data, and Semantic Web Services presenting the main results achieved so far and highlighting the main trends, challenges, and opportunities.

Web Services

The idea of deploying and providing services on the Web has been tightly bound to Web service technologies. Web services are software systems offered over the Internet via platform and programming-language independent interfaces defined on the basis of a set of open standards such as XML, SOAP, and WSDL [Erl, 2007]. Fundamental advantage of this technology lies in the support it brings to developing complex distributed systems maximising the reuse of loosely coupled components. Several languages for Web service composition have been proposed over the years in order to combine services in a process-oriented way, among which the most prominent is BPEL4WS [Andrews et al., 2003]. Additionally, the stack of technologies is completed by a large and rather overwhelming number of specifications dubbed WS-*, which deal with aspects such as security, transactions, messaging, and notification [Erl, 2007]. This stack has brought a considerable level of complexity and yet suffers from the fact that descriptions are purely syntactic. As a consequence discovering, composing, and mediating Web services remains a predominantly manual task.

A fundamental tenet of Service-Oriented Architectures is the notion of service registries for programmatic access and discovery of suitable services. Service publication has therefore been at the core of research and development in this area since the very beginning. The Universal Business Registry part of Universal Description Discovery and Integration (UDDI) [Hately et al., 2004] is perhaps the most well-known effort towards supporting the publication of services on the Web. On the basis of UDDI, large companies like SAP, IBM and Microsoft created a universal registry for enterprise services that could be accessed publicly but it did not gain enough adoption and it was discontinued in 2006 after five years of use.

One of the main reasons for the lack of success of UDDI was the fact that, although these registries are relatively complex, they do not support expressive queries [Pilioura and Tsalgatidou, 2009]. Another fundamental reason is the fact that, as we saw earlier, the work around services has essentially focussed on enterprises which have thus far been reluctant to publish their services on the Web. Today, provides one of the largest indexes of publicly available Web services which currently accounts for 28,500 Web services with their corresponding documentation. The number of services publicly available contrasts significantly with the billions of Web pages available, and interestingly is not significantly bigger than the 4,000 services estimated to be deployed internally within Verizon [Stollberg et al., 2007]. Other academic efforts in crawling and indexing Web services on the Web have found far lower numbers of services [Al-Masri and Mahmoud, 2008].

Web 2.0

The term Web 2.0, commonly attributed to OReilly [O’Reilly, 2005], was first defined on the basis of the technologies used, e.g., AJAX. More recently, however, it is increasingly used to account for the central role users play within these applications [Hendler and Golbeck, 2008, Chi, 2008]. Most successful Web 2.0 web sites are largely based on exploiting user-provided content and on the elicitation and use of the social networks created among them. For instance, Wikipedia and Flickr are largely based on content provided by their users in a rather altruistic manner. This new way of providing content is based on dropping the unnecessarily limiting distinction between providers and consumers, giving birth instead to what is often referred to as prosumers. Additionally, and thanks to the close integration of prosumers in the provisioning process, networks among users are elicited and exploited by sites such as or Amazon to provide highly accurate recommendations.

Impelled by the Web 2.0 phenomenon, the world around services on the Web, thus far limited to “classical” Web services based on SOAP and WSDL, has significantly evolved with the proliferation of Web APIs, also called RESTful services [Richardson and Ruby, 2007] when they conform to the REST architectural style [Fielding, 2000]. This newer kind of services is characterised by the simplicity of the technology stack they build upon, i.e., URIs, HTTP, XML and JSON, and their natural suitability for the Web. Nowadays, an increasingly large quantity of Web sites offer (controlled) access to part of the data they hold through simple Web APIs, see for instance Flickr[1],[2], and Facebook[3]. This trend towards opening access to previously closed data silos has generated a new wave of Web applications, called mashups, which obtain data from diverse Web sites and combine it to create novel solutions [Benslimane et al., 2008]., the most popular directory of Web APIs lists at the time of this writing lists 2,000 APIs and 4,800 mashups. This directory is based on the manual submission of APIs by users and currently provides simple search mechanisms based on keywords, tags, or a simple classification, none of which are particularly expressive. In fact, Web APIs are generally described using plain, unstructured HTML, except for a few that use the XML-based format WADL [Hadley, 2009]. As a consequence, and despite their popularity, discovering Web APIs or developing mashups that integrate disparate services in this manner suffers from a number of limitations similar to those we previously outlined for “classical” Web services, with an increased complexity since most often no machine-processable description is available. Discovering services, handling heterogeneous data, and creating service compositions are largely manual, tedious tasks which result in the development of custom tailored solutions on a case by case basis.

The Semantic Web and the Web of Data

The Semantic Web [Berners-Lee et al., 2001] can be an extension of the current human-readable Web, adding formal knowledge representation so that intelligent software can reason with the information in an automatic and flexible way. Semantic Web research has therefore largely focussed on defining languages and tools for representing knowledge in a way that can be shared, reused, combined, and processed over the Web. This research has led to a plethora of standards such as RDF(S) [Brickley and Guha, 2002], OWL [Patel-Schneider et al., 2004], as well as tools, e.g., ontology editors [Noy et al., 2001], RDF(S) storage systems [Broekstra et al., 2002] and reasoners [Haarslev and Mo¨ller, 2003], to name a few.

The Web of Data is a relatively recent effort derived from research on the Semantic Web, whose main objective is to generate a Web exposing and interlinking data previously enclosed within silos. The Web of Data is based upon four simple principles, known as the Linked Data principles, which essentially dictate that every piece of data should be given an HTTP URI which, when looked up, should offer useful information using standards like RDF and SPARQL [Bizer et al., 2009]. Additionally, data should be linked to other relevant resources therefore allowing humans and computers to discover additional information.

Since Linked Data principles were outlined in 2006, there has been a uptake most notably by the Linking Open Data project[4] through DBpedia [Auer, 2008] and ulterior additions of data about reviews [Heath and Motta, 2008], scientific information and geographical information, to name a few. Large companies like the BBC and governments from countries like the United Kingdom or the United States of America have also joined this initiative and are gradually releasing large amounts of data they have.

This outstanding growth of the Web of Data is urging researchers to devise means to exploit the valuable information it exposes. Among the main applications produced so far there are a number of data browsers that help people navigate through the data like Disco and Tabulator [Berners-Lee et al., 2007]. There are systems that crawl, index and provide intelligent search support over the Web of data like Sindice [Oren et al., 2008] and Watson [d’Aquin et al., 08]. And finally, there are a few domain-specific applications such as or DBPedia Mobile [Becker and Bizer, 2008] that provide domain-specific functionality by gathering and mashing up data. Although useful these applications hardly go beyond presenting together data gathered from different sources leaving the great potential of this massive data space unexploited. It is therefore becoming of crucial importance to devise ways in which smart applications that exploit the Web of Data could be systematically developed.

2.4        Semantic Web Services

Semantic Web services were initially proposed in order to pursue the vision of the semantic Web presented in [Berners-Lee et al., 2001] whereby intelligent agents would be able to exploit semantic descriptions in order to carry out complex tasks on behalf of humans. Early on, however, the research efforts focussed on combining Web services and semantic Web technologies so that tasks such as the discovery, negotiation, composition and invocation of services could have a higher level of automation.

The landscape of semantic Web services is characterized by a number of conceptual models that, despite a few common characteristics, remain essentially incompatible due to the different representation languages and expressivity utilized as well as because of conceptual differences. WSMO and OWL-S adopt a top-down view over services, covering the data models, behavioural aspects, nonfunctional properties, and supporting the definition of processes. The means for describing these are significantly different, though. In contrast, SAWSDL adopts a bottom-up approach and simply provides hooks for linking to particular ontologies and transformation definitions. In practice, the heterogeneity of the existing approaches has prevented their integration, leading to a significant fragmentation in the field and thus harming the adoption of SWS.

On the basis of the aforementioned conceptual models, many researchers have worked on enhancing service registries using semantic technologies, see for instance [Kawamura et al., 2004, Srinivasan et al., 2004], many of which have built upon UDDI. Despite demonstrating the advantages of semantic annotations in discovering services, particularly in terms of accuracy and in dealing with heterogeneous data models, SWS work has downplayed the additional complexity involved in creating semantic annotations for services. Consequently, the Web does not contain a significant body of service annotations: the largest public repository today is probably OPOSSum [Ku¨ster and K¨onig-Ries, 2008] which includes a test collection with approximately 2500 service annotations and provides programmatic access to its content solely through direct access to the database management system [Ku¨ster and K¨onig-Ries, 2008].

Regardless of the differences at the semantic level, the majority of the SWS initiatives are predicated upon the semantic enrichment of WSDL Web services and, as we saw earlier, these have turned out not to be prevalent on the Web. The Web services ecology has recently seen a major evolution with the advent and proliferation of Web APIs and RESTful services [Richardson and Ruby, 2007], and there has not been much progress on, or even concern with, means for providing structured descriptions and discovering these newer kinds of services. Only recently have researchers started focusing on Web APIs and RESTful services, the main examples being hRESTS/MicroWSMO [Kopecky´ et al., 2008, Maleshkova et al., 2009a] and SA-REST [Sheth, 2007].

Services and the Web of Data: An Unexploited Symbiosis

The advent of Web services and related technologies was quickly followed by considerable hype and grandiose expectations with respect to the impact Web services would have for enterprises and the economy in general. It was often assumed that Web services would ultimately lead to the creation of a servicebased economy over the Web. However, Web services are nowadays mostly used within controlled environments such as large enterprises rather than on the Web. One could argue that a reason for this lack of take up is the fact that Web services, despite their name, were not really thought for the Web [Vinoski, 2002]. In fact, the considerable complexity of the WS-* stack did hamper their adoption on the Web as recent practice, based instead on the use of simpler approaches such as Web APIs, shows. Another reason is however the fact that Web services have essentially targeted enterprises, which tend not to publicly publish Web services in any significant numbers.

Research on SWS has managed to alleviate some of the technical drawbacks of existing Web services technologies. Despite the advanced results obtained, none of the approaches devised thus far have gained widespread adoption for three main reasons. First and foremost, all SWS approaches have built upon Web services technologies that are not prevalent on the Web. Secondly, SWS add complex logics to an already complex WS-* stack. SWS require complex architectures, highly advanced reasoning machinery, and rich semantic annotations that, up until now, had to be provided mostly from scratch by highly trained IT staff. Finally, the existing dichotomy between the syntactic level and the semantic level requires devoting significant effort to providing transformation mechanisms between semantic and syntactic representations of information which add further need for manual labour and are highly sensitive to minor variations on data representation.

We believe that the advent of the Web of Data together with the rise of Web 2.0 technologies and social principles constitute the final necessary ingredients that will ultimately lead to a widespread adoption of services on the Web. In the remainder of this paper we shall refer to this new kind of services as Linked Services. The main reasons for this are the existing technical symbiosis between services, semantics, and the Web of Data [Pedrinaci et al., 2010a], as well as the rise of the prosumer and the global movement towards an open Web driven by the current unprecedented sharing of data and functionality openly on the Web.

On the one hand, from a technological perspective, the evolution of the Web of Data is highlighting the fact that light weight semantics yield significant benefits that justify the investment in annotating data and deploying the necessary machinery. This initiative is contributing to generate an outstanding body of knowledge (light weight ontologies and data expressed in their terms) that can help to significantly reduce the effort for creating semantic annotations for services. Furthermore, it also represents a significant use case for the application of services technologies on a Web scale in order to process this wealth of data which remains nowadays largely unexploited. On the other hand, from a socio-economic perspective, the recent evolution around Web 2.0 has shown that collaboration over the Web can lead to large quantities of very useful data with a low cost. Similarly, both Web 2.0 and more recently Linked Data technologies are encouraging enterprises and institutions to offer their data and services publicly at a previously unprecedented scale and pace. This new scenario provides in our view suitable technologies and data, as well as the necessary economic and social interest for the wide application of services technologies on a Web scale.

Linked Services

The vision toward the next wave of services – Linked Services – presented herein is based on two simple ideas: publishing service annotations in the Web of Data, and creating services for the Web of Data, i.e., services that process Linked Data and generate Linked Data. In a nutshell, Linked Services are services described as Linked Data. Therefore, these are service descriptions whereby their inputs and outputs, their functionality, and their non-functional properties are described in terms of (reused) light weight RDFS vocabularies and exposed following Linked Data principles. In fact, as such, Linked Services descriptions represent highly valuable information which is still to be provided in the Web of Data: data about reusable functionality on the Web. Secondly, by virtue of these descriptions, Linked Services are therefore services that, with appropriate infrastructure support, can consume RDF from the Web of Data, and, if necessary, can also generate additional RDF to be fed back to the Web of Data. In other words, Linked Services constitute a processing layer on top of the wealth of information currently available in the Web of Data which remains unexploited.

In the remainder of this paper we shall describe in more detail how this new wave can be supported and promoted technically, we explain which are the essential principles one needs to build upon and, where appropriate, we shall illustrate how our current research is taking us in this direction. Although in this section we present concrete technologies, the reader should note that the vision presented herein could perfectly be achieved by other means. The essential aspects are, however, the publication of service descriptions in the Web of Data for their discovery and reuse, and the provisioning of processing functionality on top Linked Data.

Services on the Web of Data

We previously called attention to the scarcity of publicly available Web services. We highlighted the lack of success of prior service registries on the Web as one of the reasons behind this, and highlighted several aspects that have hampered the adoption of UDDI as a suitable standard for service registries. We also pointed out the fragmentation currently affecting SWS research as well as the proliferation of Web APIs as a simpler and increasingly more popular alternative over “traditional” Web services.

Before any significant uptake of services can take place on the Web, proper mechanisms for creating, publishing and discovering services must be in place. In this respect, our previous review of the state of the art shows that:

  • Semantics are essential to reach a sufficient level of automation during the life-cycle of services,
  • finding an adequate trade-off between the expressivity of the service model used and the scalability from a computational and knowledge acquisition perspective is key for a wide adoption of service technologies,
  • the annotation of services should be simplified as much as possible, and “crowdsourcing” appears to be a particularly effective and cheap solution to this end,
  • on the Web, light weight ontologies together with the possibility to provide custom extensions prevail against more complex models,
  • any solution to deploying services that aspires to be widely adopted should build upon the various approaches and standards used on the Web, including Web APIs, RDF, and SPARQL,
  • Linked Data principles [Bizer et al., 2009] represent nowadays the best practice for publishing data on the Web both for human and machine consumption,
  • links between publicly available datasets are essential for the scalability and the value of the data exposed.

The principles we have just highlighted have an impact in a wide range of activities during the life-cycle of services. Notably, in the remainder of this section we shall tackle how Web services and Web APIs can be annotated, we shall describe how we can better support the annotation of services and finally we described how we are currently supporting the homogeneous publication and discovery of Web services and Web APIs on the Web using light weight semantics.

Annotation of WSDL Services with WSMO-Lite

W3C produced in 2007 the Semantic Annotations for WSDL and XML Schema specification [Farrell and Lausen, 2007], a minimal bottom-up approach to annotating services semantically which has gained further uptake than more ambitious solutions like OWL-S and WSMO. SAWSDL provides simple hooks for pointing to semantic descriptions from WSDL and XML elements. In particular, it supports three kinds of annotations, namely model reference, lifting schema mapping and lowering schema mapping which allow pointing to semantic elements described elsewhere on the Web, or to specifications of data transformations from a syntactic representation to the semantic counterpart and back respectively. SAWSDL does not advocate for a particular representation language for these documents nor does it provide any specific vocabulary that users should adopt.

WSMO-Lite [Vitvar et al., 2008] builds upon SAWSDL overcoming some of its limitations while remaining light weight. In a nutshell, WSMO-Lite provides a very simple RDFS ontology together with a methodology for expressing functional and nonfunctional semantics, and an information model for WSDL services based on SAWSDL model reference hooks. WSMO-Lite makes explicit the intended meaning for model reference annotations without modifying SAWSDL but rather informing users on how they should structure the models their annotations point to.

The WSMO-Lite ontology includes the means for specifying the functionality of a service with respect to a hierarchy of functional categories (e.g., eCl@ss [Hepp, 2006]) through the notion of Functional Classification Root. Additionally, it provides hooks for more advanced definition of non-functional properties as well as Conditions and Effects. The ontology is entirely expressed in RDF(S) and where the expressivity of RDFS is not sufficient (notably for expressing conditions and effects) other languages such as WSML [Fensel et al., 2007] and those produced by the W3C Rule Interchange Format Working Group[5] can be used.

Annotation of Web APIs with MicroWSMO

As we previously introduced, Web APIs and RESTful services are increasingly used on the Web. Therefore any approach to using services on the Web that would disregard them would be unnecessarily limiting. Annotating this kind of service does, however, bring additional complexity given that in most of the cases services are solely described through unstructured HTML pages aimed at humans.

MicroWSMO is a microformat-like[6] notation that forms the basis for our work on describing Web APIs [Kopecky´ et al., 2008, Maleshkova et al., 2009a]. MicroWSMO builds upon the hRESTS (HTML for RESTful services) microformat. hRESTS enables the creation of machine-processable Web API descriptions based on available HTML documentation [Kopecky´ et al., 2008]. As a microformat hRESTS provides a number of HTML classes that allow one to structure APIs descriptions by identifying services, operations, methods, inputs, outputs, and addresses. It therefore supports, by simple injections of HTML code within Web pages, to turn unstructured HTML-based descriptions of Web APIs into structured services descriptions similar to those provided by WSDL.

With the hRESTS structure in place, HTML service descriptions can be annotated further by including pointers to the semantics of the service, operations, and data manipulated. To this end MicroWSMO extends hRESTS with three additional properties, namely model, lifting and lowering that are taken from SAWSDL and have the same semantics. MicroWSMO also adopts WSMO-Lite as the reference ontology for annotating RESTful services semantically.

Supporting Services Annotation

Arguably, one of the main limitations of previous approaches to integrating services in the Semantic Web, has been the difficulty from an annotation perspective. SWS approaches like WSMO and OWL-S, mostly focussed on devising highly expressive frameworks able to capture formally the semantics of services in a considerable detail, overlooked the bottleneck they were introducing with respect to the annotation of services. Indeed, the creation of SWS based on these frameworks requires a significant manual labour devoted to devising domain models, taxonomies, orchestrations, and other rules that can only be created at a slow pace by highly trained IT personnel.

Some effort has been devoted by previous research toward the automation of service annotation, notably [Heß et al., 2004] and [Sabou, 2006]. However, although useful, the support provided still needs to be complemented with substantial manual editing, the creation of ontologies and rules. The use of light weight ontologies as opposed to highly expressive conceptual models reduces considerably the effort involved and the amount of annotations to be provided. Additionally, and more importantly, the Web of Data is significantly changing the environment from an annotation and usage point of view.

On the one hand, the wide range of ontologies and semantic data publicly available on the Web is an increasingly valuable source of knowledge. The Web of Data can be used as background knowledge [d’Aquin et al., 08] in order to provide suitable ontologies that can be used, extended, and combined to create domain ontologies for annotating services in an easier manner as highlighted in [Maleshkova et al., 2009b]. Furthermore, the existence of increasingly large quantities of information expressed in terms of ontologies can effectively be exploited to support the identification of the domain of a service based for instance on its documentation as well as it can, for instance, support the matching of ontologies when creating new domain models or when integrating different services [Sabou et al., 2008].

On the other hand, generating service annotations by reusing existing ontologies directly contributes to increasing services usability and presumably their uptake. For instance services may be classified with respect to well-known service classifications such as the previously mentioned eCl@ss ontology, better supporting their discovery by software and humans aware of that particular ontology. Furthermore, annotating services inputs and outputs with respect to existing vocabularies ensures the direct applicability of services over data already available as well as it allows Linked Data application developers to carry out data driven discovery of services by simply checking the input and output types of services. From a more abstract perspective, this process ensures that services modeled in this way are linked to the Web of Data as encouraged by Linked Data principles. Finally, Web 2.0 applications have highlighted the advantages that the social side of the Web can bring when a significant body of users and data has been gathered [Hendler and Golbeck, 2008]. The same way we can exploit the growing body of knowledge generated by the Semantic Web, we expect that as the number of service annotations grow, we would also be able to exploit them in order to contribute to the overall annotation process by i) ranking the domain models with respect to their popularity thus indirectly contributing to increasing services compatibility; and ii) by refining the identification of the domain of a service based on prior decisions by other users.

We are devoting significant efforts to creating tools that support users in the annotation of services based on the principles introduced above. One such application is SWEET [Maleshkova et al., 2009b] which is, to the best of our knowledge, the first tool that enables the creation of semantic annotations for Web APIs and RESTful services. SWEET provides user support for creating hRESTS/MicroWSMO annotations over any HTML page describing Web APIs, therefore supporting a non-intrusive incremental annotation of existing resources. The tool, assisted by Watson [d’Aquin et al., 08], supports users in browsing the Semantic Web while annotating services so that they can identify suitable vocabularies such as FOAF [Brickley and Miller, 2007], and use them for the annotation. A tool called SOWER, based on the same principles but focussing on the annotation of WSDL services, has also been developed. Currently, the social aspects are not exploited by these tools since it is first necessary to gather a significant body of service annotations.

Homogeneous Publication and Discovery of Services on the Web of Data

Syntactic and semantic descriptions of Web services aim at providing information about services in a way that can automatically be processed by machines. However, at present, these descriptions can only be retrieved through the Web of documents, which is essentially designed for human beings, or through specific interfaces to registries such as UDDI that have failed to gain significant uptake.

A fundamental step for bringing services closer to the Web is their publication based on current best practices. We view service annotations as a particular kind of highly valuable data: data that informs us about existing reusable functionality exposed somewhere on the Web that processes and/or generates data. As such, services should therefore be published on the Web according to current best practices for publishing data – the Linked Data principles – so that applications can easily discover and process their descriptions on the basis of the very same technologies they use for retrieving data.

In order to explore and validate these principles we have developed iServe [Pedrinaci et al., 2010b], a public platform that unifies service publication and discovery on the Web through the use of light weight semantics. iServe builds upon lessons learnt from research and development on the Web and on service discovery algorithms to provide a generic semantic service registry able to support advanced discovery over different kinds of services described using heterogeneous formalisms. The registry is, to the best of our knowledge, the first system able to homogeneously publish and provide advanced discovery support for SWS expressed in several formalisms. It is also the first one to provide advanced discovery over Web APIs and Web services homogeneously.

In the remainder of this section we first outline the conceptual model iServe builds upon and we then present the overall approach implemented by the platform in order to support the homogeneous publication and discovery of services.

Minimal Service Model

In order to publish services on the Web of Data it is necessary to provide a common vocabulary based on existing Web standards able to describe services in a way that allows machines to automatically locate and filter services according to their functionality or the data they handle, and to appropriately support their automated invocation. Additionally, as opposed to most SWS research to date, it is of utmost importance to support the annotation of both “classical” WSDL Web services, as well as the increasing number of Web APIs and RESTful services which appear to be preferred on the Web.

To this end our research is based on the Minimal Service Model (MSM), originally introduced together with hRESTS [Kopecky´ et al., 2008] and WSMOLite [Vitvar et al., 2008], and slightly modified for the purposes of this work. The MSM, driven by Semantic Web best practices, builds upon existing vocabularies, namely SAWSDL, WSMO-Lite and hRESTS, depicted in Figure 1 with the sawsdl, wl, and rest namespaces respectively. In a nutshell, the MSM is a simple RDF(S) integration ontology based on the principle of minimal ontological commitment; it captures the maximum common denominator between existing conceptual models for services. Thus, the MSM does not aim to be yet another service model to bring further heterogeneity to the SWS landscape; it is instead an integration model at the intersection of existing formalisms, able to capture the core semantics of both Web services and Web APIs in a common model, homogeneously supporting publication and discovery. Still, the MSM is devised in a way such that framework-specific extensions can remain attached, to the benefit of clients able to comprehend and exploit those formalisms.

The MSM, denoted by the msm namespace in Figure 1, defines Services which have a number of Operations. Operations in turn have input, output and fault MessageContent descriptions. MessageContent may be composed of mandatory or optional MessageParts[7]. The intent of the message part mechanism is to support finer-grained input/output discovery, as available in SAWSDL, OWL-S and WSMO, especially including support for optional parts.

Minimum Service Model

Figure 1: Minimal Service Model.

SWS frameworks [Sheth, 2003, Vitvar et al., 2008] thus far have provided support for semantically describing different subsets of the following aspects of services:

  • Functional semantics defines service functionality, that is, the function a service offers to its clients when it is invoked. This information is of particular relevance when finding services and when composing them.
  • Nonfunctional semantics defines any specific details concerning the implementation or running environment of a service, such as its price or quality of service. Nonfunctional semantics provide additional information about services that can help rank and select the most appropriate one.
  • Behavioural semantics specifies the protocol (i.e., ordering of operations) that a client needs to follow when invoking a service.
  • Information model defines the semantics of input, output, and fault messages.

To attach these semantics to the service model, we adopt the RDF mapping of SAWSDL introduces earlier, which defines three kinds of annotations over WSDL and XML Schema, namely model reference, lifting schema mapping, and lowering schema mapping. The schema mapping annotations provide grounding from the service’s Information Model to the concrete on-the-wire messages, whereas the model references can be used for pointing to ontologies covering functional semantics, nonfunctional semantics , behavioural semantics and the information model.

The WSMO-Lite vocabulary [Vitvar et al., 2008] completes the MSM by providing classes for significantly describing the above four aspects of service semantics and by supplying type information to the generic model references. In particular, WSMO-Lite captures nonfunctional semantics through the concept of Nonfunctional Parameter, and functional semantics via the concepts

Condition, Effect, and Functional Classification Root. The reader may note that WSMO does not have direct support for functional classifications; still, the majority of discovery engines for WSMO have indirectly applied the notion of classifications through hierarchies of Web Services or Goals (e.g. in [Stollberg et al., 2007, Domingue et al., 2008]).

Behavioural semantics are likely the biggest source of heterogeneity between SWS frameworks; SAWSDL even omits this aspect altogether. We therefore do not prescribe any particular approach to describing behavioural semantics of services and defer this instead to specific applications and frameworks. Thanks to its simplicity, the MSM captures the essence of services in a way that can support service matchmaking and invocation, while remaining largely compatible with WSMO-based descriptions of Web services, with OWL-S services, and with services annotated according to SAWSDL, WSMO-Lite, and MicroWSMO.

iServe: a Linked Services Publishing and Discovery Platform

iServe uses as its core conceptual model the MSM and it currently includes a number of import mechanisms able to deal with WSDL files including SAWSDL annotations, with descriptions adopting the WSMO-Lite specific extensions, with MicroWSMO annotations of Web APIs as well as with OWL-S service descriptions. These import mechanisms transform the service descriptions into the appropriate terms according to the MSM. Additionally, iServe automatically generates rdfs:definedBy links – pointing to the definition file in case additional information is required – and rdfs:seeAlso links – pointing to documentation.

Once imported, iServe publishes the semantic annotations of services as Linked Data. Thus every service is assigned a resolvable HTTP URI, through which, humans and machines can access the service descriptions in HTML or in RDF using content negotiation. The registry additionally provides a SPARQL endpoint allowing advanced querying over the services annotations, as well as a read and write RESTful API so that services can easily be retrieved and published from remote applications. The RESTful API is completed with a number of semantic discovery methods that provide more refined discovery than that supported directly via SPARQL, by exploiting the semantic descriptions of services, RDFS inferencing, and similarity measures for more accurate results.

iServe architecture

Figure 2: High-level architecture of iServe.

On top of iServe’s RESTful API, the registry is complemented by a crawler. Currently it has only been used for targeted import for there are not many SWS descriptions available on the Web. At the time of this writing, iServe registers about 2000 SWS coming from the OWL-S test collection[8] and the SAWSDL test collection[9], 50 services coming from the Jena Geography Dataset[10] annotated manually for evaluation purposes, a test import of around 30 services indexed by, and around 20 real services annotated in the context of the use cases of the EU projects SOA4All and NoTube. The current implementation already shows how Web services and Web APIs can be described by means of an homogeneous conceptual model – the Minimal Service Model – and how they can be published as Linked Data, therefore better promoting their discovery based on the use of the well established and adopted Linked Data principles.

Services for the Web of Data

The notion of services as well-defined, independent, invokable and distributed pieces of functionality is indeed a very powerful architectural notion for developing distributed systems. Providing functionality in this way independently from the underlying technology provides the capacity for maintaining a loose coupling between integrated components which, when it comes to an environment like the Web, appears as a highly beneficial (if not necessary) feature. Services, may they be traditional Web services or RESTful services, provide therefore a suitable architectural abstraction for the integration of processing capabilities over the Web of Data in a loosely coupled manner. In the remainder of this section we shall cover what services can provide to the Web of Data both as a means for providing new sources of data as well as for processing existing assertions.

Integrating Legacy Systems

Currently a good part of the Web of Data is generated from existing databases by using tools such as D2R [Bizer et al., 2009]. Indeed, this allows exposing large amounts of data which would otherwise remain private or, in the best case, offered through means that are not that convenient for automated processing. In other cases data is already stored in RDF and can be exposed easily[11]. There is, however, a large body of information owned by companies which are either not interested in offering the information publicly on the Web given its commercial value and/or its sensitivity, or because they do not have the technical skills or interest in exposing the information as Linked Data. Similarly, there is a growing number of streams of data provided by sensors through highly heterogeneous formats and interfaces, which exhibits considerable integration and processing limitations [Sheth et al., 2008].

Web 2.0 developers have long realised the value of Web APIs for accessing highly valuable data on demand. Additionally, Semantic Web researchers have acknowledged the benefit that could be brought by adapting or wrapping these additional sources of information like Web APIs and sensors, so that they can be turned into Linked Data producers, see for instance [Sheth et al., 2008, Sequeda and Corcho, 2009] and the Flickr Wrappr[12]. To a certain extent, the work on sensors is more advanced since there already exists proposals for exposing sensors observations as Linked Data [Page et al., 2009]. The work around exposing Web APIs as Linked Data is, however, more an art than a science due to the lack of standard description languages and the extreme heterogeneity characterising Web APIs.

We previously highlighted that Linked Services are such that their inputs and outputs are RDF. As a consequence, they represent a natural means for exposing as Linked Data valuable information previously enclosed within silos, through the annotation of existing Web APIs and WSDL services. Web APIs could in this way be invoked by interpreting their semantic annotations (see Section 4.2), and RDF information could be obtained on demand. In this way, data from legacy systems, state of the art Web 2.0 sites, or sensors, which do not embrace Linked Data principles could be made available as Linked Data easily.

This approach is explored in the context of several use cases from European projects such as SOA4All [Davies et al., 09] and NoTube [Qing et al., 2010]. Our current experience, although preliminary at this stage, shows already the applicability and potential of bringing legacy systems to the Web of Data in this manner. Indeed proper care should be taken in order to ensure that Linked Data principles are followed in these cases (see Section 2.3). We anticipate, however, that at least for services strictly adhering to REST principles this should be relatively straight-forward since they should already define URIs for the resources and offer convenient means for exposing and exploring them.

Processing Linked Data

Integration and fusion of disparate data coming from the Web of Data hardly takes place nowadays and therefore applications do not perform any ulterior processing of this data other than for presenting it to the user [Bizer et al., 2009]. Generating new data based on what has been found or the provisioning of addedvalue services that exploit this data thus remains a pending issue. For instance, something as simple and useful as a unit transformation service is still to be provided for the Web of Data. To a certain extent this is natural since the Web of Data is precisely about data; and storing an RDF triple per possible transformation result would simply be absurd since there are infinite possibilities. There is, however, a clear need for enabling the processing of Linked Data in ways such that application developers could conveniently apply them over data gathered at runtime to carry out computations as simple as unit transformations, more complex as deriving similarities between products or services based on the reviews published by users on, or even more advanced as envisioned for the Semantic Web [Berners-Lee et al., 2001].

The Web of Data provides large amounts of machine-processable data ready to be exploited and, as we saw earlier, services provide a suitable abstraction for encapsulating functionality as platform and language independent reusable software. It therefore seems natural to approach the development of systems that process Linked Data by composing Linked Services. These services should be able to consume RDF data (either natively or via lowering mechanisms), carry out the concrete activity they are responsible for (e.g., unit conversion), and return the result, if any, in RDF as well. The invoking system could then store the result obtained or continue with the activity it is carrying out using these newly obtained RDF triples combined with additional sources of data. In a sense this is quite similar to the notion of service mashups [Benslimane et al., 2008] and RDF mashups [Phuoc et al., 2009] with the important difference that services are, in this case, RDF-aware and their functionality may range from RDF-specific manipulation functionality up to highly complex processing beyond data fusion. The use of services as the core abstraction for constructing Linked Data applications is therefore more generally applicable than that of current data integration oriented mashup solutions.

It is worth noting in this respect the benefit brought by having services annotations available on the cloud as we saw earlier. When developing applications that process Linked Data, discovering useful services could be driven by the data that needs to be manipulated. For instance, developers could easily discover services that manipulate a concrete kind of data or those that produce a certain type by sending SPARQL queries to service registries like iServe [Pedrinaci et al., 2010b], or using advanced semantic discovery mechanisms. And, as opposed to traditional Web services repositories like UDDI-based ones, developers would benefit from the existence of semantic annotations in order to filter them based on the semantics of inputs, outputs, their classification with respect to well-known taxonomies, etc. The reuse of ontologies and vocabularies would in turn contribute towards increasing the compatibility of services. In this way, Linked Data application developers would have access to an ever growing body of reusable components ready to be combined and exploited.

The Services Ecosystem

Integrating services with the Web of Data as depicted before would give birth to a services ecosystem on top of Linked Data, whereby people would be able to collaboratively and incrementally construct complex systems by reusing the results of others, gradually taking us closer to the ambitious vision initially presented for the Semantic Web. In this process, we anticipate that two main families of services will emerge depending on whether they are domain-independent or not.

On the one hand, task-specific yet domain-independent services will allow developers to perform some of the typical tasks involved when processing Linked Data. These activities range from relatively basic activities such as transforming data between different schemas to more complex actions such as determining how trust-worthy a piece of data is or even, eventually, to carry out knowledge intensive tasks, e.g., Parametric Design or Diagnosis [Schreiber et al., 1999]. These domain-independent services which are already starting to appear (for example, [Euzenat, 2004]) can in fact be seen from a Knowledge Engineering perspective as a new generation of Problem-Solving Methods (PSM) adapted to the Web as some researchers already start considering [van Harmelen et al., 2009].

This new family of PSMs for the Web of Data will, however, require adapting prior techniques to the new environment, notably with respect to the location, size, and quality of the data to be manipulated. In fact, traditional PSMs were applied within closed environments often with small amounts of manually curated data, whereas in this new scenario data would be obtained automatically from the Web for automated processing, and it would therefore have to be validated, fused, cleaned, and filtered prior to any execution since this would otherwise yield execution errors or incorrect results. We expect that a good deal of domain-independent services will precisely be devoted to performing these tasks. For instance entity resolution, ontology alignment, data cleansing, data fusion, provenance analysis, and trust analysis are some of the domain-independent services that we anticipate would be necessary to develop for the Web of Data. As a side effect, though, it is likely that data quality in the Web of Data will increase as software matures, and especially as it starts being processed by applications which would indirectly detect inconsistencies and incorrect data.

On the other hand, we refer as domain-dependent services to those abstracted away from the technicalities and specificities of Linked Data and generic tasks. This kind of services will be for example those directly providing access to traditional systems in order to obtain some data and carry out actions like sending an SMS or booking a hotel. These services will only be relevant for a particular domain, e.g., hotel services, and will mostly be populated by services directly addressing end-users and therefore better showcasing the potential of the Semantic Web from an end-user perspective. It is worth noting, however, that a wide proliferation of advanced domain-specific solutions for end-users will only occur when a sufficient set of stable domain-independent services able to solve complex tasks will be available. For instance, cross organisational business integration would most likely have to build upon on advanced ontology alignment support for transforming data between different schemas [Jung, 2009]. The systematic development of these applications in a sustainable, efficient, and robust manner shall only be achieved through reuse, and services are a particularly suitable abstraction to carry this out on a Web scale.

6        Conclusions and Outlook

Despite the appealing characteristics of service-orientation principles and technologies, their uptake on a Web-scale has been significantly less prominent than initially anticipated. This limited adoption is due to a number of issues of both socio-economic and technical nature. From a socio-economic perspective service orientation has for the most part targeted enterprises which, thus far, have been reluctant to publish functionality of the Web. From a technical perspective, service technologies have exhibited a limited level of support for automating activities such as service discovery and composition. SWS have managed to overcome some of the technical limitations of Web services but have in turn introduced additional complexity and overheads. Consequently, SWS have not gained any significant adoption either.

In parallel, the Web is witnessing a dramatic evolution with the advent of Web 2.0 and Linked Data technologies. Web 2.0 has triggered a socialisation of the Web which has placed individuals at the centre of the Web and is widely based on somewhat altruistic contributions of free data and manual labour from users. The Linked Data initiative is in turn devoted to creating what is referred to as the Web of Data, which already provides publicly large amounts of interconnected data concerning a wide range of domains described in terms of light weight ontologies for supporting automated processing.

We have argued that the advent of the Web of Data together with the rise of Web 2.0 technologies and social principles constitute the final necessary ingredients that will give birth to a new wave of services on the Web, which we refer to as Linked Services. We have explored the relationship between services and the Web of Data. In particular we have highlighted that Linked Data represent appropriate principles for publishing services on the Web. We have illustrated how Web services and RESTful services can be brought into the Web of Data by means of simple RDF vocabularies and supporting tools. We have highlighted the fact that the current evolution of the Web of Data is gathering the necessary motivation for the development of advanced applications that process Linked Data. We have outlined that Linked Services are particularly well-suited for supporting developers in creating applications that process Linked Data. We have discussed how the evolution towards more complex Linked Data applications could be supported and we have identified the need for making publicly available domain-independent services that carry out common tasks such Data Cleansing or Trust Analysis [Jung, 2010].

The overall vision outlined herein represents the roadmap for the research we are currently carrying out trying to expand the capabilities of the Linked Data applications as well as trying to promote and support the use of services on the Web through light weight semantic annotations. This research, like the principles it builds upon, will strive to provide data, resources, tools and engines publicly on the Web in order to eventually lead to the wider uptake of services on a Web scale.


[Andrews et al., 2003] Andrews, T., Curbera, F., Dholakia, H., Goland, Y., Klein, J., Leymann, F., Liu, K., Roller, D., Smith, D., Thatte, S., Trickovic, I., and Weerawarana, S.: Business Process Execution Language for Web Services Version 1.1;, May 2003.

[Auer, 2008] Auer, S., Kobilarov, C. B. G., Lehmann, J., Cyganiak, R., and Ives, Z.: DBpedia: A Nucleus for a Web of Open Data; In Proceedings of 6th International Semantic Web Conference, 2nd Asian Semantic Web Conference (ISWC+ASWC 2007), pages 722–735. November 2008.

[Al-Masri and Mahmoud, 2008] Al-Masri, E. and Mahmoud, Q.: Investigating Web Services on the World Wide Web; In 17th International World Wide Web Conference, pages 795–804, 2008.

[Becker and Bizer, 2008] Becker, C. and Bizer, C.: DBpedia Mobile: A LocationEnabled Linked Data Browser; In Linked Data on the Web (LDOW2008), 2008.

[Benslimane et al., 2008] Benslimane, D., Dustdar, S., and Sheth, A.: Services Mashups: The New Generation of Web Applications; IEEE Internet Computing, 12(5):13–15, 2008.

[Brickley and Guha, 2002] Brickley, D. and Guha, R. V.: RDF Vocabulary Description Language 1.0: RDF Schema;, 2002

[Bizer et al., 2009] Bizer, C., Heath, T., and Berners-Lee, T.: Linked Data – The Story So Far; International Journal on Semantic Web and Information Systems (IJSWIS), 2009.

[Broekstra et al., 2002] Broekstra, J., Kampman, A., and van Harmelen, F.: Sesame: A Generic Architecture for Storing and Querying RDF and RDF Schema; International Semantic Web Conference (ISWC 2002), 2002.

[Berners-Lee et al., 2001] Berners-Lee, T., Hendler, J., and Lassila, O.: The Semantic Web; Scientific American, (5):34–43, May 2001.

[Berners-Lee et al., 2007] Berners-Lee, T., Hollenbach, J., Lu, K., Presbrey, J., d’ommeaux, E. P., and m.c. schraefel: Tabulator redux: Writing into the semantic web; 2007.

[Brickley and Miller, 2007] Brickley, D. and Miller, L.: FOAF Vocabulary Specification 0.91;, November 2007.

[Chi, 2008] Chi, E.: The Social Web: Research and Opportunities; Computer, 41(9):88 – 91, Sep 2008.

[Domingue et al., 2008] Domingue, J., Cabral, L., Galizia, S., Tanasescu, V., Gugliotta, A., Norton, B., and Pedrinaci, C.: IRS-III: A Broker-based Approach to Semantic Web Services; Web Semantics: Science, Services and Agents on the World Wide Web, 6(2):109–132, 2008.

[Davies et al., 09] Davies, J., Domingue, J., Pedrinaci, C., Fensel, D., GonzalezCabero, R., Potter, M., Richardson, M., and Stincic, S.: Towards the Open Service Web; BT Technology Journal, 26(2), 2009.

[d’Aquin et al., 08] d’Aquin, M., Motta, E., Sabou, M., Angeletou, S., Gridinoc, L., Lopez, V., and Guidi, D.: Toward a New Generation of Semantic Web Applications; IEEE Intelligent Systems, 23(3):20–28, 2008.

[Erl, 2007] Erl, T.: SOA Principles of Service Design The Prentice Hall ServiceOriented Computing Series. Prentice Hall, July 2007.

[Euzenat, 2004] Euzenat, J.: An API for ontology alignment; In 3rd conference on International Semantic Web Conference (ISWC), 2004.

[Fielding, 2000] Fielding, R. T.: Architectural Styles and the Design of Network-based Software Architectures PhD thesis, University of California, Irvine, 2000.

[Farrell and Lausen, 2007] Farrell, J. and Lausen, H.: Semantic Annotations for WSDL and XML Schema (SAWSDL); Recommendation, W3C, August 2007.

[Fensel et al., 2007] Fensel, D., Lausen, H., Polleres, A., de Bruijn, J., Stollberg, M., Roman, D., and Domingue, J.: Enabling Semantic Web Services: The Web Service Modeling Ontology Springer, 2007.

[Hadley, 2009] Hadley, M.: Web Application Description Language; Member submission, W3C, August 2009.

[Hepp, 2006] Hepp, M.: Products and Services Ontologies: A Methodology for Deriving OWL Ontologies from Industrial Categorization Standards; International Journal on Semantic Web and Information Systems (IJSWIS), 2(1):72–99, 2006.

[Hendler and Golbeck, 2008] Hendler, J. and Golbeck, J.: Metcalfe’s law, Web 2.0, and the Semantic Web; Web Semantics: Science, Services and Agents on the World Wide Web, Jan 2008.

[Heß et al., 2004] Heß, A., Johnston, E., and Kushmerick, N.: ASSAM: A Tool for Semi-automatically Annotating Semantic Web Services; In McIlraith et al. [McIlraith et al., 2004], pages 320–334.

[Haarslev and Möller, 2003] Haarslev, V. and Möller, R.: Racer: A Core Inference Engine for the Semantic Web; In Second International Semantic Web Conference ISWC 2003, Florida, October 2003.

[Heath and Motta, 2008] Heath, T. and Motta, E.: Revyu: Linking reviews and ratings into the Web of Data; Web Semantics: Science, Services and Agents on the World Wide Web, 6(4):266–273, 2008.

[Hately et al., 2004] Hately, L. C. A., von Riegen, C., and Rogers, T.: UDDI Specification Version 3.0.2; Technical report, OASIS, 2004.

[Jung, 2008] Jung, J. J.: Query transformation based on semantic centrality in semantic social network; Journal of Universal Computer Science, 14(7):1031–1047, 2008.

[Jung, 2009] Jung, J. J.: Semantic Business Process Integration based on Ontology Alignment; Expert Syst. Appl., 36(8):11013–11020, 2009.

[Jung, 2010] Jung, J. J.: Reusing Ontology Mappings for Query Segmentation and Routing in Semantic Peer-to-Peer Environment; Information Sciences, 180(17):3248–3257, 2010.

[Kawamura et al., 2004] Kawamura, T., Blasio, J.-A. D., Hasegawa, T., Paolucci, M., and Sycara, K. P.: Public Deployment of Semantic Service Matchmaker with UDDI Business Registry; In McIlraith et al. [McIlraith et al., 2004], pages 752–766.

[Kopecky´ et al., 2008] Kopeck´y, J., Gomadam, K., and Vitvar, T.: hRESTS: an HTML Microformat for Describing RESTful Web Services; In The 2008 IEEE/WIC/ACM International Conference on Web Intelligence (WI2008), Sydney, Australia, November 2008. IEEE CS Press.

[Küster and König-Ries, 2008] Küster, U. and König-Ries, B.: Towards Standard Test Collections for the Empirical Evaluation of Semantic Web Service Approaches; Int. J. Semantic Computing, 2(3):381–402, 2008.

[Martin et al., 2004] Martin, D., Burstein, M., Hobbs, J., Lassila, O., McDermott, D., McIlraith, S., Narayanan, S., Paolucci, M., Parsia, B., Payne, T., Sirin, E., Srinivasan, N., and Sycara, K.: OWL-S: Semantic Markup for Web Services; Member submission, W3C, 2004 W3C Member Submission 22 November 2004.

[Maleshkova et al., 2009a] Maleshkova, M., Kopecky´, J., and Pedrinaci, C.: Adapting SAWSDL for Semantic Annotations of RESTful Services; In Workshop: Beyond SAWSDL at OnTheMove Federated Conferences & Workshops, 2009.

[Maleshkova et al., 2009b] Maleshkova, M., Pedrinaci, C., and Domingue, J.: Supporting the Creation of Semantic RESTful Service Descriptions; In Workshop: Service Matchmaking and Resource Retrieval in the Semantic Web (SMR2) at 8th International Semantic Web Conference, 2009.

[McIlraith et al., 2004] McIlraith, S. A., Plexousakis, D., and van Harmelen, F., editors The Semantic Web – ISWC 2004: Third International Semantic Web Conference,Hiroshima, Japan, November 7-11, 2004. Proceedings, volume 3298 of Lecture Notes in Computer Science. Springer, 2004.

[McIlraith et al., 2001] McIlraith, S., Son, T., and Zeng, H.: Semantic web services; Intelligent Systems, IEEE, 16(2):46 – 53, Jan 2001.

[Noy et al., 2001] Noy, N. F., Sintek, M., Decker, S., Crub´ezy, M., Fergerson, R. W., and Musen, M. A.: Creating Semantic Web Contents with Protégé-2000; IEEE Intelligent Systems, 2(16):60–71, March/April 2001.

[Oren et al., 2008] Oren, E., Delbru, R., Catasta, M., Cyganiak, R., Stenzhorn, H., and Tummarello, G.: a Document-oriented Lookup Index for Open Linked Data; IJMSO, 3(1):37–52, 2008.

[O’Reilly, 2005] O’Reilly, T.: What Is Web 2.0: Design Patterns and Business Models for the Next Generation of Software;, 2005.

[Pedrinaci et al., 2010a] Pedrinaci, C., Domingue, J., and Krummenacher, R.: Services and the Web of Data: An Unexploited Symbiosis; In Linked AI: AAAI Spring Symposium ”Linked Data Meets Artificial Intelligence”, 2010 Accepted for publication.

[Pedrinaci et al., 2010b] Pedrinaci, C., Liu, D., Maleshkova, M., Lambert, D., Kopecky´, J., and Domingue, J.: iServe: a Linked Services Publishing Platform; In Proceedings of Ontology Repositories and Editors for the Semantic Web at 7th ESWC, 2010.

[Phuoc et al., 2009] Phuoc, D. L., Polleres, A., Hauswirth, M., Tummarello, G., and Morbidoni, C.: Rapid Prototyping of Semantic Mash-ups Through Semantic Web Pipes; In Quemada, J., Leo´n, G., Maarek, Y. S., and Nejdl, W., editors, WWW, pages 581–590. ACM, 2009.

[Page et al., 2009] Page, K., Roure, D. D., Martinez, K., Sadler, J., and Kit, O.: Linked Sensor Data: RESTfully serving RDF and GML; In 2nd International Workshop on Semantic Sensor Networks (SSN09), collocated with the 8th International Semantic Web Conference (ISWC-2009), 2009.

[Patel-Schneider et al., 2004] Patel-Schneider, P. F., Hayes, P., and Horrocks, I.: OWL Web Ontology Language Semantics and Abstract Syntax;, February 2004 Last Visited: March 2005.

[Pilioura and Tsalgatidou, 2009] Pilioura, T. and Tsalgatidou, A.: Unified Publication and Discovery of Semantic Web Services; ACM Trans. Web, 3(3):1–44, 2009.

[Qing et al., 2010] Qing, H., Benn, N., Dietze, S., Siebes, R., Pedrinaci, C., Liu, D., Lambert, D., and Domingue, J.: Two-staged Approach for Semantically Annotating and Brokering TV-related Services; In The IEEE International Conference on Web Services (ICWS), 2010.

[Richardson and Ruby, 2007] Richardson, L. and Ruby, S.: RESTful Web Services O’Reilly Media, Inc., May 2007.

[Schreiber et al., 1999] Schreiber, G., Akkermans, H., Anjewierden, A., de Hoog, R., Shadbolt, N., de Velde, W. V., and Wielinga, B.: Knowledge Engineering and Management: The CommonKADS Methodology MIT Press, 1999.

[Sabou, 2006] Sabou, M.: Building Web Service Ontologies PhD thesis, Vrije Universiteit Amsterdam, 2006.

[Sequeda and Corcho, 2009] Sequeda, J. and Corcho, O.: Linked Stream Data: A Position Paper; In 2nd International Workshop on Semantic Sensor Networks (SSN09), collocated with the 8th International Semantic Web Conference (ISWC2009), 2009.

[Sabou et al., 2008] Sabou, M., d’Aquin, M., and Motta, E.: Exploring the Semantic Web as Background Knowledge for Ontology Matching; Journal of Data Semantics, (Accepted for publication) 2008.

[Sheth, 2003] Sheth, A.: Semantic Web Process Lifecycle: Role of Semantics in Annotation, Discovery, Composition and Orchestration; Invited Talk at WWW 2003 Workshop on E-Services and the Semantic Web, May 2003.

[Sheth, 2007] Sheth, A.: Beyond SAWSDL: A Game Plan for Broader Adoption of Semantic Web Services; IEEE Intelligent Systems Trends & Controversies, 22(6), November–December 2007.

[Stollberg et al., 2007] Stollberg, M., Hepp, M., and Hoffmann, J.: A Caching Mechanism for Semantic Web Service Discovery; In 6th International and 2nd Asian Semantic Web Conference (ISWC2007+ASWC2007), pages 477–490, November 2007.

[Sheth et al., 2008] Sheth, A., Henson, C., and Sahoo, S.: Semantic Sensor Web; Internet Computing, IEEE, 12(4):78–83, 2008.

[Srinivasan et al., 2004] Srinivasan, N., Paolucci, M., and Sycara, K.: Adding OWL-S to UDDI: Implementation and throughput; In Proceedings of 1st International Conference on Semantic Web Services and Web Process Composition (SWSWPC 2004), 2004.

[van Harmelen et al., 2009] van Harmelen, F., ten Teije, A., and Wache, H.: Knowledge Engineering Rediscovered: Towards Reasoning Patterns for the Semantic Web; In K-CAP ’09: Proceedings of the fifth international conference on Knowledge capture, pages 81–88, New York, NY, USA, 2009. ACM.

[Vinoski, 2002] Vinoski, S.: Putting the “Web” into Web Services: Interaction Models, Part 2; IEEE Internet Computing, 6(4):90–92, 2002.

[Vitvar et al., 2008] Vitvar, T., Kopecky, J., Viskova, J., and Fensel, D.: WSMO-Lite Annotations for Web Services; In Hauswirth, M., Koubarakis, M., and Bechhofer, S., editors, Proceedings of the 5th European Semantic Web Conference, LNCS, Berlin, Heidelberg, June 2008. Springer Verlag.

[1] See

[2] See

[3] See

[4] See

[5] See Working Group

[6] See

[7] The addition of message parts is a small extension to the original MSM.

[8] See

[9] See

[10] See

[11] See

[12] See

Interoperability Solutions for European Public Administrations (ISA)


Administrative procedures have the reputation of being lengthy, time-consuming and costly. Electronic collaboration between Public administrations can make these procedures quicker, simpler and cheaper for all parties concerned, in particular when transactions need to be done cross-border and/or cross-sector.


The Interoperability Solutions for European Public Administrations (ISA) program of the European Commission facilitates such transactions through more than 40 Actions organized into five categories:

and four clusters:

Our interest in the Semantics of Service immediately engages us with two Actions falling within the Trusted Information Exchange cluster:

Common services – Action 1.1 – Improving semantic interoperability in European eGovernment systems.

Common frameworks – Action 1.3 – Accessing Member State information resources at European level.

ISA Action 1.1 – Improving semantic interoperability in eGovernment systems

Homepage accessed on 20141122.

Four Core Vocabularies 1 have been developed under ISA Action 1.1 so far:

  • Core Person: captures the fundamental characteristics of a person, e.g. the name, the gender, the date of birth, the location.
  • Registered organisation: captures the fundamental characteristics of a legal entity (e.g. its identifier, activities) which is created through a formal registration process, typically in a national or regional register.
  • Core Location: captures the fundamental characteristics of a location, represented as an address, a geographic name or geometry.
  • Core Public service: captures the fundamental characteristics of a service offered by public administration.

Public administrations can use and extend the Core Vocabularies in the following contexts:

  • Development of new systems: the Core Vocabularies can be used as a default starting point for designing the conceptual and logical data models in newly developed information systems.
  • Information exchange between systems: the Core Vocabularies can become the basis of a context-specific data model used to exchange data among existing information systems.
  • Data integration: the Core Vocabularies can be used to integrate data that comes from disparate data sources and create a data mesh-up.
  • Open data publishing: the Core Vocabularies can be used as the foundation of a common export format for data in base registries like cadastres, business registers and service portals.

See also:

  1. Core vocabularies are simplified, re-usable, and extensible data models that capture the fundamental characteristics of an entity in a context-neutral fashion.

ISA Action 1.3 – Improving semantic interoperability in eGovernment systems

Homepage accessed on 20141122.

Study ICatalog of (Web) Services (CoS) – focused on the set-up of a common catalog containing cross-border services and its interoperability aspects. Its scope was limited to online administration-to-administration web services that involve the exchange of information.

Study IIFederated Catalog of Public Services (FCOPS) – focused on the set-up of a catalog of public services offered by all public administrations at all government levels in all EU Member States, which are the members of the ISA Program. Models of catalogs were defined and investigated according to ISA’s LOST (Legal, Operational, Semantic and Technical) principles. The specific goals were to:

  • Analyze the existing public services models and recommend what should be done to set up a common public services model
  • Determine the feasibility of building a European FCOPS
  • Define a roadmap with concrete set of steps on how to implement the catalog

See more about Action 1.3: Catalog of services and a presentation on Action 1.3: Catalog of Services – November 2014.

The study consists of two Work Packages (WPs), each with a Final Report as main deliverable:

WP 1 focused on the current state of affairs in the Member States in the field of catalogs of public services – D1.3 Phase I Final report
WP 1 -Current State of Affairs
– August 2013.

WP 2 examined the requirements and different scenarios for establishing a FCOPS, and proposed an action plan on how to achieve this – D2.2 Phase II Final report
WP 2 -Requirements and Scenarios
– June 2013.