Understanding how services operate as part of large scale global networks, the related risks and gains of di erent network structures and their dynamics is becoming increasingly critical for society. Our vision and research agenda focuses on the particularly challenging task of building, analyzing, and reasoning about global service networks. This paper explains how Service Network Analysis (SNA) can be used to study and optimize the provisioning of complex services modeled as Open Semantic Service Networks (OSSN), a computer-understandable digital structure which represents connected and dependent services.
Services that o er di erent capabilities are also distributed over space and time. These services are related, dependent, connected, and form networks. They are brought together as new services through composition or aggregation. In many situations, the value of services and networks is in uenced by existing suppliers, competitors, value-adders and customers. It is important to be able to understand and reason about these networks to assist, e.g., decision-making involving strategic investments in service innovation. However, this is hard because of their scale (number of services a network may have) and the complexity of technical aspects such as the temporal and spatial distribution, and the business aspects such as the diversity of marketing, operations, business models and nancial aspects.
We resort to Service Network Analysis (SNA) that o ers a systematic and
scienti c analysis of service networks to address the above challenges. SNA views
service systems and service relationships [
The dynamic nature of service networks indicates that their topology might
be shaped according to some intrinsic property, e.g. service cost, availability,
or extrinsic property, e.g. perceived customer preference. This dynamic
behavior has been veri ed in many elds. For example, Web Science looks at World
Wide Web models as a large directed graph with an apparent random character.
Nonetheless, the topology of this graph has evolved to a scale-free network [
Having provided a motivation for the importance of SNA, the contribution of this paper is to aggregate and present four approaches originating from di erent elds that can be applied to analyse service networks: 1. Optimization, 2. Evolutionary analysis, 3. Cooperative analysis and 4. Value analysis.
While these approaches have been developed in isolation and have been often
applied to distinct elds (e.g. complex systems, logistics, economics, and
markets), they all constitute solutions for SNA. In the future, a convergence of these
approaches is needed to build a comprehensive body of scienti c methods for
service network analysis. A second contribution of this paper is to describe how,
from a technical perspective, these service networks can be build. We relied
on the recent developments of the Linked USDL (http://linked-usdl.org)
language to represent services and a service relationship model OSSR [
In this paper, we explore SNA as introduced above and o er some
preliminary concepts, models and insights. Section 2 introduces the main terms and
concepts that will be used throughout the paper. In Section 3 we provide a
motivating scenario, from the eld of cloud computing, to highlight the
importance of constructing complex services by aggregating simpler building blocks
into service networks. Section 4 explains how Open Semantic Service Networks
(OSSN)[
To address the growing importance of connecting service systems, we have
introduced the concepts that constitute an OSSN [
OSSNs are global service networks which relate services with the assumption that rms make the information of their service systems openly available using suitable models. Therefore, service systems, relationships, and networks are said to be open when their models are transparently available and accessible by external entities and follow an open-world assumption. The objective of open services is very similar to the one explored by the linked data initiative (http://linkeddata.org): exposing, sharing, and connecting pieces of data and information on the Semantic Web using URIs and RDF. Networks are said to be semantic since service and relationships can be represented by using shared models, common vocabularies, and semantic Web theories and technologies. Service networks bring together several players (e.g. service creators, aggregators, providers, marketplaces, and consumers ) that work together to deliver value to consumers.
The (re)construction of OSSNs is the result of a peer-to-peer social process.
Firms, groups and individuals (i.e. the community) are equal participants which
freely cooperate to provide information on services and their relationships to
ultimately create a unique global, large-scale service network. The principles
of OSSN (re)construction were presented in our previous work [
Our scenario is from the eld of cloud computing. As cloud applications spread,
such as platform-as-a-service (PaaS) and software-as-a-service (SaaS), the
dependencies between applications increases. For example, Heroku, Instagram,
Pinterest and Net ix all establish a relationships with Amazon EC2 since they
dependent on its services. This has many implications e.g., a change in Amazon EC2
characteristics (e.g. its cost, reliability, and performance) can in uence all the
dependent services.
Let us consider the service network from Fig. 1. Two service aggregators (SA1
and SA2), part of the ACME corporation, decided to construct two new services:
ACME Customer Relationship Management (SC1RM ) and ACME Business
Intelligence (SB2I ). These two services where constructed from the aggregation of an
existing cloud processing service provided by Heroku6 (SH3er) and a storage
service provided by Amazon Elastic Block Store7 (SA4ma). In addition, SC1RM also
relied on the SugarCRM service8(SS5ug), an open-source, web-based customer
relationship management SaaS platform. On the other hand, SB2I relied on the
BIME service9 (SB6im), a SaaS solution for business analytics and data
visualization. The relationships [
R1( SH3er, SA4ma, SS5ug)
R2( SH3er, SA4ma, SB6im)
The two new services, i.e. SC1RM and SB2I , are commercialized by service provider SP5 and SP6. The atomic services SH3er, SA4ma, SS5ug, and SB6im are provided by SP1, SP2, SP3, and SP4, respectively. Service consumers (Ci) can 6 http://www.heroku.com/ 7 http://aws.amazon.com/ebs/ 8 http://www.sugarcrm.com/ 9 http://www.bimeanalytics.com/ purchase and use any of the aggregated services. To operate, aggregated services must purchase computing/processing units from atomic services. In other words, there is a dependency between, e.g., SC1RM and SH3er. Furthermore, SC1RM has a service complementor osCommerce10 CP1 and SB2I has a service competitor SAS Visual Analytics CO1.
SNA can be used to, e.g., minimize the cost of providing the aggregated service SC1RM . The study of the network can suggest the use of other data storage and processing services which are less expansive than SH3er and SA4ma, possibly with a slightly lower reliability. Minimizing the cost of a service network can be mapped to an assignment problem which in turn is mapped to fundamental combinatorial optimization problems. This branch of service network analysis will be explored in section 5.1. The e ect of having a higher number of complementors CPi and a lower number of competitors COi can also be studied since it in uences the perceived value of services SC1RM and SB2I . In this case, ndings from the led of social network analysis can be used to explore the in uence of actors in a network. 4
The service network from the previous section can be constructed and represented using OSSN by accessing, retrieving and combining information from service and relationship models. OSSN are networks which relate services with the assumption that rms make the information of their services openly available using suitable models. In other words, we make the assumption that the description of services SC1RM , SB2I , SH3er, SA4ma, SS5ug, and SB6im is openly and remotely available. 4.1
In OSSN, services are modeled using the family of languages named *-USDL (the
Uni ed Service Description Language)[
As an example of service description modeling, we illustrate how the service SugarCRM SS5ug from Section 3 was modeled using Linked USDL. The information used to model the service was retrieved from its web site. A service and a vocabulary model were created. The vocabulary contained domain dependent 10 http://www.oscommerce.com/ is a vendor specialized in customizable web shops platforms 11 Linked-USDL = http://linked-usdl.org/ 12 -USDL = http://www.genssiz.org/research/service-modeling/alpha-usdl/ 13 USDL = http://www.w3.org/2005/Incubator/usdl/ concepts from the eld of CRM systems (e.g., taxonomies of common installation options). Since Linked USDL only provides a generic service description language, domain speci c knowledge needs to be added to further enrich the description of services. The excerpt from Listing 1.1 illustrates the description of the SugarCRM service (the example was written using the Turtle language14). 1 <#service_SugarCRM> a usdl:Service ; 2 ... 3 dcterms:title "SugarCRM service instance"@en ; 4 usdl:hasProvider :provider_SugarCRM_Inc ; 5 usdl:hasLegalCondition :legal_SugarCRM ; 6 gr:qualitativeProductOrServiceProperty 7 crm:On_premise_or_cloud_deployment , 8 crm:Scheduled_data_backups , 9 crm:Social_media_integration , 10 crm:Mobile_device_accessibility . 11 ... 12 :offering_SugarCRM a usdl:ServiceOffering ; 13 ... 14 usdl:includes <#service_SugarCRM> ; 15 usdl:hasPricePlan 16 :pricing_SugarCRM_Professional , 17 :pricing_SugarCRM_Corporate , 18 :pricing_SugarCRM_Enterprise , 19 :pricing_SugarCRM_Ultimate ; 20 usdl:hasServiceLevelProfile :slp_SugarCRM . 21 ...
Listing 1.1. SugarCRM service modeled with Linked USDL
The description starts with the identi cation of the provider (line 4), the legal usage conditions (line 5), the general properties of the service (e.g., deployment, schedules backups, integration, and mobile accessibility), and its price plans (line 15). 4.2
Service networks rely on a fundamental element which connects service systems:
relationships. The Open Semantic Service Relationship (OSSR) model [
An OSSN network is modelled as a time-varying hypergraph SN , such as SN (t) = fS(t); C(t); R(t)g, where S(t) is the set of services provided and C(t) is the set of service consumers, with both being modeled with USDL. R(t) is the set of relationships modeled with OSSR connecting consumers and services provided.
For example, binary relationships can be depicted as edges in a directed graph (see section 5). Time is represented by the parameter t, whose granularity is set to appropriately model the market (e.g., days). Consumers alter the topology of a service network by di usion when they adopt or abandon a service by adding or deleting an OSSR relationship to it.
To construct a service network SN , USDL and OSSR models are remotely
accessed and retrieved (an overview description of the infrastructure to access
and retrieve USDL and OSSR instances is described in [
A wide spectrum of techniques and algorithms can be developed to study OSSNs.
For example, reasoning techniques can be developed to explore the notion of
relationships as bonds. By discovering strong cliques, we can hypothesize that
the stronger the relationships, the stronger the uni cation and the greater the
commonality of fates. As a result, it would be possible to infer that a tightly
coupled service network will sink or swim together. Other fundamental
algorithms which are valuable to implement come from the eld of network science.
For example, algorithms to detect if an OSSN is a scale-free networks [
We present four methods under the umbrella of SNA: 1) OSSN optimization, 2) evolutionary analysis of OSSN, and 3) cooperative analysis of OSSN, and 4) value analysis. Fig. 2 conceptualizes and contextualizes these methods with service networks. Optimization deals with constructing networks by selecting the best services, with regard to some criteria, from some set of available alternatives. Evolutionary and cooperative analysis deals with the study of networks' structures as a function of time. In other words, how do networks expend, adapt, or collapse overtime based on internal and external factors. Finally, service value network analysis deals with the establishment of rules and regulations which ensure that the construction of networks in society follows a fair and unbiased processes.
These four methods enable to study di erent aspects of a service network and, therefore, they do not compete between them. They provide di erent views on a network. For example, optimization (the rst method described in the next section) can be used to selected the most desirable combination of services to achieved an initial goal. On the other hand, cooperative analysis studies how the behaviour of customers (e.g. a new service subscription or a change of the service provider) and the characteristics of the service o ered (e.g. its price) in uence the ows of material and/or immaterial resources (e.g. raw materials) inside a service network. SN = (N; E; ln; le), is a node and edge labeled directed graph (initially we are modeling acyclic graphs) where edge labeled directed graph (initially we are modeling acyclic graphs) where The nodes N are either services S or consumers C. The consumer nodes are sinks (no outgoing edges). Certain services are atomic (i.e., not formed as compositions of simpler services). In our initial modeling we consider them to be sources (i.e., no incoming edges). There will be edges from services to other services as well as consumers.
The node labels indicate the semantics of services or the demands of customers. The edge labels are abstract representations of type. A service, in general, takes one or more input types (colors) and produces one or more output types (colors). Initially, type matching is simple, but we plan to generalize it to model subsumption hierarchies. The color of an edge will be set to the most general of the two matching colors (i.e. the more general type). If there is no match, there will be no edge. Below, an algorithm will be outlined for maximal service network construction. This network will then be trimmed with its ow quanti ed by using mathematical optimization techniques.
The optimal construction of a service network is divided into two phases: maximal color-compliant service network construction and cost minimization. The algorithm for phase I builds a service network from three sets of nodes, atomic services (sources), composite services (intermediate nodes), and consumers (sinks). Starting with the sources, all intermediate and consumer nodes are connected by edges that are color compliant, e.g., if an intermediate node needs a blue input and green input and there exist sources producing/outputting these colors, then this intermediate node is added to the graph. This process continues through k stages, a parameter indicating the maximum number of stages (i.e., distance from source to sink) desired.
Once the graph has been created, it can be reduced to an optimal form using Linear Programming. Such problems require an objective function in terms of decision variables, and constraints on the values of those decision variables. Our objective function is the cost of the network, and our decision variables represent the ow of material through the network and the amount of production at intermediate nodes. Flow is determined by the edges and intermediate nodes, so each of these will have a decision variable. The ow is constrained by the supply, production, or demand capacity of the nodes in the network, and by the required number of inputs for each non-source node. Once the constraints are gathered, a Linear Programming algorithm such as the Simplex algorithm, can be used to nd the optimal values for the decision variables. These values determine the optimal amount of ow through the network and the value of the objective function estimates the minimum cost. 5.2
In our second example, let us assume that each service system contains a value proposition communicated to customers (i.e, the attractiveness elements or preferential attachment). Service value is judged from the perspective of consumers as they compare services among the alternatives. For simplicity reasons, we assume that the value proposition is similar for all service systems and it is the price of the services calculated from a usdl-price:PricePlan15. This concept if part of the Linked USDL family.
Since our objective is to forecast the evolution of a service network over time, we use the following function to calculate the Market Share of each service provided M S(si) = degree(si)=m; where degree(si) is the number of relationships established by service si with service consumers and m is the total number of relationships established between providers and consumers. Overtime, customers change preferences by changing from one service system to another service system provider.
100% 75% 50% 25%
Market share
Forecast 100% 75% 50% 25%
Market share 4
5
Time (t)
Let us assume that the (re)constructed SN topology shows that overtime
the market share is the one represented in Fig. 3 at t = 3. One question to be
answered is: "what will happen to the market in the future if the conditions are
not changed?" (i.e. the value propositions of si remain the same and m ci).
According to Bass model [
+ Total Services + Si KPI = Net gains
Sk KPI =
Resource Limit + +
+ + KPI Gain per Individual Service + Si + Sj KPI = Net gains a)
Service system Sk
Time b)
Fig. 3 illustrates that from the four services provided, three also rise in market share during the early stages, reach a peak, and then decline as the service leader accelerates because of the increasing returns e ect of preferential attachment. In this case, all but one service provided leaves the market, leaving one monopoly competitor. The SNA eld can, therefore, develop mathematical models which can help to forecast the evolution of service network overtime. 5.3
In our third example, we explored the suitability of OSSNs to model system dynamics. Instead of looking at causes and their e ects in isolation, we analyze service networks as systems made up of interacting parts. Once an OSSN is created from distributed service models, cause-e ect or network e ect diagrams can be derived for the network. For example, Fig. 4 shows service systems Si, Sj , Sk, and directed edges illustrating internal and external relationships.
Causal relationships connect Key Performance Indicators (KPI) from different services' and within services. The pattern represented by this OSSN is commonly known as the 'Tragedy of the Commons' archetype. It hypothesizes that if the two services Si and Sj overuse the common/shared service Sk, it will become overloaded or depleted and all the providers will experience diminishing bene ts. Service Si and Sj provide services to costumers. To increase net gains, both providers increase the availability of service instances. As the number of instances increases, the margin decreases and there is the need to increase even more the number of instances available. As the number of instances increases, the stress on the availability of service Sk is so strong that the service collapses or cannot respond anymore as needed. At that point, service Si and Sj can no longer fully operate and the net gain is dramatically reduced for all the parties involved as shown in Fig. 4.b). 5.4
While the previous three sections considered structural aspects, such as the
analysis, optimization and evolution of the network, the described methods do not
take into account the participants' behavior in a service network. For example,
depending on the market mechanism implemented in the service marketplace,
providers might have an incentive to report their service characteristics (such
as price, non-functional attributes, etc.) untruthfully to the system in order to
increase their chance of being allocated. Hence, the market mechanism used in
the marketplace has to be able to handle such behavior and still yield an e
cient market outcome. This is the focus of the economic concept of Service Value
Networks (SVNs, see [
The abstract model and topology of service aggregation in SVNs is shown
in Fig. 5. Given that the consumer requests an aggregated service consisting
of two service functionalities, we have two candidate pools of atomic services
from di erent service providers. From a Mechanism Design perspective, which
is the focus of most current work on SVNs, the question is how we can select a
combination of these atomic services out of the candidate pools that best satis es
the consumer requirements (in Fig. 5, there are 3 2 = 6 di erent combinations
for the requested aggregated service). In SVNs, this is implemented through a
\complex service auction" [
The mechanism implements two steps. In the rst step, the calculation of the allocation, the mechanism computes the di erent potential combinations of atomic services to the desired aggregated service (the aggregation operation of service attributes depends on the attribute type, e.g. the price for the aggregated service is the sum of prices for the atomic services). The mechanism selects the aggregated service with the highest (positive) di erence between consumer valuation minus the costs of the atomic services. In the next step, the calculation of the payments, the mechanism implements a Vickrey-Clarke-Groves (VCG) payment scheme to determine the actual payments to the providers of the allocated atomic services. In contrast to other payment schemes, the VCG scheme is desirable as service providers have the incentive to report the attributes and prices of their services truthfully to the marketplace, without the incentive to manipulate. This property is achieved by rewarding service providers according to their relative importance (added value) to the SVN, which means they can receive an additional discount on their service provisioning price.
This mechanism, with the described allocation and payment calculations, has
certain desirable properties. For example, it is known to be allocative e cient,
i.e. it selects the best combination of atomic services given the consumer
preferences. Further, as mentioned earlier it is also strategy-proof, which means that
the dominant strategy for service providers is to submit their service attributes
truthfully to the marketplace, as other strategies will not yield better outcomes
for the providers. However, the Impossibility Result by [
For the study of SVNs, Service Network Analysis has been applied in various
research questions. Considering the dynamic behavior of the SVN, we can look at
the incentives the providers have to join the network. As a competitive and vital
SVN has to provide many (functionally) di erent services to accommodate for
diverse consumer requirements, Conte et al. [
While current work on SVNs has mainly focused on economic topics, the augmentation of the SVN concept with semantic capabilities to an OSSN along with the use of Social Network Analysis promise to be fruitful. Such an amalgamation would enable a better description and usability of SVNs as well as an improved understanding of their dynamic behavior. 6
In most work done so far, existing approaches fail to adhere to service-dominant
logic [
For example, e3service [
In [
Allee [
From a business perspective, there is an apparent trend for companies (e.g.
service providers) to specialize by focusing on core competencies and becoming
a member of adaptable networks [
Nonetheless, a challenge yet to be solved involves modeling and optimizing
the functions of service-centric organizations from technological, business and
legal points of view, extending work on optimizing Web process [
We envision a world that is well connected via global OSSNs. Semantics will have a major role in creating a large-scale and integrated service network. Organizations, groups and individuals will have tools and platforms to advertise their know-how, capabilities and skills in the form of services to the world. A huge number of detailed rm-generated (or even user-generated) services will be available worldwide. Some services will be shared, composed, co-created, personalized, others are crowdsourced.
As service networks emerge, their study will enable to understand how a service-based society grows and changes overtime. Service network analysis (SNA) can provide theories, mathematical models, algorithms, techniques, and tools to achieve this goal. This paper presented four applications of SNA: optimization, evolution, network e ect, and service value. Optimization targets to construct networks of services which minimize the overall cost. Network evolution relies on time to study and forecast how a network structure will evolve overtime. Network e ect explores the impact of changing the characteristics of one or more service nodes in the other services and in the network itself. Finally, service value network enable to analyze the in uence providers' strategies in the network. These four methods provide the rst building blocks to demonstrate the practical application of SNA to better understand how service networks function in a global, interconnected service-based societies.