World Wide Value Web: Automated Design of Real-World Multi-party Services on the Web

World Wide Value WebAutomated Design of Real-World Multi-party Services on the Web

Dr. Pieter De Leenheer

Tuesday 25 December 12

The Value Web is there already ...

• Ever wondered what happens when you click these “links” ?

Legend

Value

Transfer

Value

port

Value

interface

Consumer

need

Connect.

element

Actor

Boundary

element

OR

element

AND

element

ActivityMarket

segment

Value

object

[...]

Explosion

element

[MONEY] [MONEY]

[Right to make tracks public]

[MONEY] [Right to clear a track]





[MONEY] [Right to collect fees] [Right to collect fees] [MONEY]

[MONEY] [Track]

[MONEY] [Right to make tracks public] [MONEY] [Right to make tracks public]




Legend

Value

Transfer

Value

port

Value

interface

Consumer

need

Connect.

element

Actor

Boundary

element

OR

element

AND

element

ActivityMarket

segment

Value

object

[...]

Explosion

element

[MONEY] [MONEY]








[MONEY] [Track]


Legend

Value

Transfer

Value

port

Value

interface

Consumer

need

Connect.

element

Actor

Boundary

element

OR

element

AND

element

ActivityMarket

segment

Value

object

[...]

Explosion

element

[MONEY] [MONEY]








[MONEY] [Track]






N+1 Tethered Value Webs• walls slowing down innovation• no matter what’s being purchased: 1 mediator who takes the cream

http://www.statista.com/




Ex. 1: High-speed Train Station in Greater London



• OR

• purchase a water-proof wall to keep water out



• OR


• purchase a pump to dispense ground water from the tunnel



• OR



• new value object (i.e., asset): unlimited water resource

• provided by “pumping service”



• OR





• value integrator: the London Water Authority (e.o.) in need of water resources



• OR





• value integrator: the London Water Authority (e.o.) in need of water resources

• new service value network that turns a problem into opportunity is win-win for both parties


Ex. 2: Learning Languages with DuoLingo



• OR

• offer certified online language courses in return for a subscription fee



• OR


• value objects: fee, certificate



• OR



• offer certified language course for free in return for written assessments via sentence translations

• new value object: language-to-language sentence translations



• OR





• through text translating service



• OR






• value integrator: content providers



• OR






• value integrator: content providers

• articulating tacit value objects hidden in existing service relationships creates new value for both parties.


More Examples of Successful Networked Value Propositions Cleverly Combine Web Relations

• http://www.slideshare.net/boardofinnovation/10-business-models-that-rocked-2010-6434921

• talent for the happy few, but how to automate this design process ?


http://www.slideshare.net/boardofinnovation/10-business-models-that-rocked-2010-6434921

http://www.slideshare.net/boardofinnovation/10-business-models-that-rocked-2010-6434921

Overview of the Claims



1. Our world is a large-scale non-linear network of rich relationships between technologies, people, and organisations, emerging from the Web.




2. Web relationships are a catalyst for innovation, i.e., a Value Web, that organisations should harness to devise new forms of value co-creation. To this end, enterprises must abandon value-chain thinking.





3. One challenge is to articulate the structure and composition of value objects inherent to these relationships that would lead them to gravitate towards unanticipated value propositions.






4. Presuming service-centric thinking, and non-linear patterns of the Web, Service Value Networks (SVNs) lie at the center of this gravitation; forming the hubs of the Value Web.






4. Presuming service-centric thinking, and non-linear patterns of the Web, Service Value Networks (SVNs) lie at the center of this gravitation; forming the hubs of the Value Web.

5. (Service) Value Web technologies should embody generative principles similar to those that lead to the success of the Web itself. In other words, Internet-based SVN technologies should allow for unanticipated contribution of value (through service) to the Web by enabling anyone to share and trade their value objects, just like previous generations of the Web did for knowledge and social sharing.




How come ?


Sources: Nova Spivack, John Breslin, Mills Davis, www.opte.org


http://www.opte.org

http://www.opte.org



http://www.opte.org

http://www.opte.org


“By carefully excluding features that are not universally useful Internet technologies became easily adopted on a massive scale and gave the Web a

generative [i.e. self-reproductive] character” (Zittrain, 2009).


http://www.opte.org

http://www.opte.org

Agent webs That know,

Learn & reason As humans do

Incre

asi

ng

Kn

ow

led

ge C

on

nectivi

ty &

Reaso

nin

g

Increasing Social Connectivity

The Ubiquitous WebConnects Intelligence

The Semantic WebConnects Knowledge

The Social WebConnects People

The WebConnects Information

ArtificialIntelligence

PersonalAssistants

IntelligentAgents

Ontologies

Thesauri &Taxonomies

SemanticSearch

Bots

Blogjects

Semantic Website & UI Semantic

BlogSemantic

Wiki

AutonomicIntellectualProperty

Spime

Semantic AgentEcosystems

SmartMarkets

Multi-userGaming

SemanticSocial networks

SemanticCommunities

Wiki CommunityPortals

Marketplaces& Auctions

BlogsRSS

SocialNetworks

Email

Conferencing

Instant MessagingP2P

File Sharing

PIMS

Web Sites

Search Engines

KnowledgeBases

Content Portals

EnterprisePortals

“Push”Publish & Subscribe

Databases

File Servers

SocialBookmarking

Semantic Desktop

SemanticEnterprise

Desktop

Context-AwareGames

Mash-ups

SemanticEmail

NaturalLanguage





http://www.opte.org

http://www.opte.org





http://www.opte.org

http://www.opte.org




Web Science: The Web’s relational patterns exhibit “long tail”

distributions: “80% of sales goes to 20% of the offerings”


http://www.opte.org

http://www.opte.org



Why is this not happening ?


From Chain to Network

value-in-exchange value-in-use

transaction relationships

operand resource (goods) operant resource (knowledge, consumer)

marketing push consumer pull

technology service and content

customer acquisition customer retention

Service = the applications of competences (knowledge and skills) for the benefit of a party

Service = action; not object


The Problem of SOA: Service-oriented Architecture

Norman & Ramirez (1993): “the key strategic task is the reconfiguration of roles and relationships among this constellation of actors in order to mobilise the creation of value in new forms and by new players.”


The Problem of SOA: Service-oriented Architecture• believed to be core enabling technology, however no large-scale adoption for our service economy

• a componential approach inspired by product innovation: “bill of materials” and “urban architecture”





➡ clever idea but with lack of appreciation of inherent traits of service co-production: variety, intangibility, and coopetition






• biased by the enterprise-centric vision, hence electronic business implementations:

• rely on hierarchy of functional components, i.e.: Web services for exchange of data and functionality

• enforce how to execute a certain business operation in a fixed pre-defined manner: time dependency and control flow









➡ completely ignores aspects related to the exchange of value: e.g., strategy, proposition, roles, resourcing, pricing, quality and regulatory compliance









➡ completely ignores aspects related to the exchange of value: e.g., strategy, proposition, roles, resourcing, pricing, quality and regulatory compliance

• complement SOA with value abstraction level: declare knowledge about what the business domain constitutes in terms of assets and relationships that allows to reactively adapt its role in changing value propositions.

• Service-dominant logic: ontological analysis of “service” as a perdurant (“action”), rather than an endurant (“object”)....



Service Network Approaches: State of the Art

• dotted circles:

• process-based

• planning problem

• solid circles:

• value-based

• design problem

➡ low tendency towards decentralised and automated approaches

➡ contamination of process-thinking in network-centric approaches

➡ lonely at the top?

DesignNone Analysis MatchingBundling Composition Dynamic Composition

Ente

rpris

e-ce

ntric

:hi

erar

chica

l pro

cess

-driv

en o

rgan

isatio

nNe

twor

k-ce

ntric

:de

cent

ralis

ed re

latio

nshi

p-dr

iven

orga

nisa

tion

ICT support:

Value Chain

(Porter, 1985)

BMO (Oster-walder, 2004)

e3value (Gordijn,

2002)

e3service (de

Kinderen, 2009)

Servigu-ration

(Baida, 2006)

Value Networks

(Allee, 2002)

REA (McCarthy,

1982)

GVP (Zlatev, 2007)

O-WSP (Omela-yenko, 2006)

(Razo-Zapata et

al., BUSITAL,

2010)

(Gordijn et al.,

HICCS, 2011)

(Razo-Zapata,

BUSITAL 2011)

Wiki-nomics

(Tapscott, 2008)

Digital Capital

(Tapscott, 2000)

VBC (Nakamura, 2006 )

(Traverso, 2004)

Dynami-CoS (Da

Silva, 2011)

u-Service (Lee, 2011)

CPC (Letia, 2008)

(Kohl-born, 2010)

(Becker, 2009)

Ontomat (Agarwal,

2004)

Service Architectu-res (Booth,

2004)

METEOR-S (2005)

SNN(Bitsaki, 2008)














3.One challenge is to articulate the structure and composition of value objects inherent to these relationships that would lead them to gravitate towards unanticipated value propositions.






4.Presuming service-centric thinking, and non-linear patterns of the Web, Service Value Networks (SVNs) lie at the center of this gravitation; forming the hubs of the Value Web.






4.Presuming service-centric thinking, and non-linear patterns of the Web, Service Value Networks (SVNs) lie at the center of this gravitation; forming the hubs of the Value Web.

5. (Service) Value Web technologies should embody generative principles similar to those that lead to the success of the Web itself. In other words, Internet-based SVN technologies should allow for unanticipated contribution of value (through service) to the Web by enabling anyone to share and trade their value objects, just like previous generations of the Web did for knowledge and social sharing.


Service Value Networks

• service co- production, i.e. ”bundling”

• in function of well-articulated needs.

• reflects an acceptable trade-off between

• value proposition (to maximize short-term profit) and

• market accuracy (to minimize consumer sacrifice)

• fractal system: the Value Web is a SVN in which every peer itself can be an SVN

• thus SVN composition becomes a complex problem










An SVN is a complex system of peers that establish the necessary relationships to collectively produce (hence co-produce) value (in terms of a real-world service) for their environment (Razo-Zapata, De Leenheer, & Gordijn, 2011).











teaching “introduction to databases”












ability to normalise a database












ability to normalise a database

certificate / diploma


SVN Composition Problem: Design vs. Planning• Composition is a Design- rather than

Planning- problem

• “service artifact”: what value is exchanged rather than how and when

• value network analysis

• patterns of exchange ?

• causal effect of value within and on environment?

• value accuracy?

• self-adaptation principles?

4.1. SVN COMPOSITION 63

Figure 4.1: The SVN Composition Framework.

edge but also the inferences that are needed to produce new knowledge as well as Knowledgethe interactions among them. Inferences carry out reasoning processes whereas Inferencesdynamic knowledge roles are the run-time inputs and outputs of inferences [82].Because supply and demand are always evolving, they are described as dynamicknowledge roles. As can be observed in Figure 4.1, due to the application of in-ferences, the rest of the knowledge components are also dynamically produced.Finally, whereas the propose, verify and modify subtasks must be perform bya computer (broker), the critique subtask must be perform by a human (cus- Brokertomer). In this way, the interaction between the customer and a broker provides Customeralso a dialogue in which the customer can influence the composition of SVNsby providing feedback. The elements of this dialogue-based interaction (i.e thepropose, verify, critique and modify subtasks) are explained in the followingparagraphs.

4.1.1 Propose

According to Chandrasekaran, given a design goal, the propose subtask gen-erates a solution [19]. In our case, the goal is to compose a SVN to cover agiven customer need. The next paragraphs provide a detailed explanation of theinferences related to this subtask, i.e. O↵ering, Laddering, Matching, Bundlingand B2B Linking.

4.1.1.1 O↵ering

Before composing any SVN, a description of the o↵erings of services suppliersand enablers is required. Such description must allow the creation of B2C and


Articulating needs: from a long & happy life down to toothpaste

• I.S. requirements engineering meets marketing theory:

• separation of structure and solution (e.g., means-end chaining, quality function deployment, problem frames, i*) reveals different functions of product attributes:

• product (toothpaste): attribute (minty) -> consequence (neat image, increase social inclusion) -> value (sense of beloning) <= need for a happy life

• product (toothpaste): attribute(calcium;teeth strengthener) -> consequence (stay healthy) <= need for a long life [note: attribute(minty) not relevant]

• .... and semantically encode this in consequence ladders, based on a customer perspective ontology:

• Kinds of needs: physical good (house), human resource, monetary resources, information, capability (course), experience (museum visit), state change (hair cut, car wash, a flight)

S. de Kinderen et al. / An ontology for needs-driven service bundling in a multi-supplier setting 3

Problem

recognition

Information

search

Post-

purchaseEvaluation Purchase

Fig. 1. The Customer Buying Behavior Model, cf. Kotler (2000)

Most scholars refer to the above characteristics, but use them in different combinations to provide theirown interpretation of what a service is. Some emphasize one specific aspect, such as “services are deeds,processes or performances” (see Bitner et al. (2008)) while others, most notably Vargo and Lusch (2004)conclude that everything is a service.

For this article, we adopt a business science interpretation of the term service. Of particular importancefor us are the following two aspects. Services have an intangible nature, therefore automated reasoningabout services is challenging, as opposed to physical products. Services produce valuable outcomes, thelatter providing us with matchmaking capabilities with customer needs.

2.2. Customer needs

The e3service ontology is unique in a way that it considers analysis of customer needs key. To ensurea needs-driven service bundling process, we require an understanding of the steps that a customer usuallytakes to arrive from the goals that s/he wants to achieve, to the decision to purchase a service offering.Marketers provide us with several buying behavior theories that help us understand the main steps thatcustomers use in this process. Most prominent amongst these is the Customer Buying Behavior (CBB)model, which we find in amongst others Kotler (2000); Solomon (2003); Loudon and Della-Bitta (1993).

The CBB model consists of the five steps depicted in Fig. 1: (1) problem recognition, in which the cus-tomer becomes aware of a need that is to be satisfied (2) information search, in which the customer seeksout benefits required to satisfy this need (3) evaluation, in which the customer decides upon the productthat maximizes the desired features and minimizes the negative features (4) product buying, in whichthe customer actually buys the product and finally (5) the post-purchase phase, in which the customerevaluates the product in use-situations. 2

Following the steps of the CBB model (Kotler, 2000, p. 177-178), we discuss (1) separation betweenproblems and solutions, (2) how a customer seeks products, and (3) how products are evaluated by bal-ancing positive and negative service features.

2.2.1. Separation of problem and solutionThe CBB model sees problem recognition and information search (hence finding a solution) as two

separate phases, hence emphasizing the explicit separation between problem specification and problemsolving. The emergence of a customer need is caused by a deviation between the customer’s current andideal state. This may happen by an alteration of the current state (e.g. the need for a new car can be initiatedby a breakdown of the current car) or ideal state (e.g. through the new car of the neighbor). Only thereafter,the customer searches for solutions satisfying his need through information search.

2.2.2. Gradual specification of customer need into solutionsThis is related to CBB’s Information search step. Customer needs analysis is often treated as problem

specification, whereby a goal from the customer is translated into something that is specific enough to finda solution for. The amount of levels required for this specification then depends on the level of abstractionof the customer’s goal. Here, one may consider ‘leading a happy life’ as the ultimate abstract goal that

2Fig. 1 suggests that the customer goes through the CBB steps sequentially. This is however not the case. Depending onthe situation in which it used, the CBB model contains feedback loops between steps, and some steps even might be omittedaltogether (Kotler, 2000, p. 179). For example: In the case of habitual buying, such as a pack of salt, the customer can skip thesteps “information search” and “evaluation of alternatives”.


Customer Perspective Ontology & Example (1) in the domain of Assisted Living for Dementia Patients

6 S. de Kinderen et al. / An ontology for needs-driven service bundling in a multi-supplier setting

Specified by

Depends on

Core enhancing

1…*

0…*

0…*

Functional consequence

Contained in

0…*

Scale

0..1Has

0…*Optional bundling

Want

Quality consequence

Need

0..*

Consists of

Has 1…*

Has 0...1

Core enhancing0…*

Optional bundling

Consequence

nominal

ordinal

interval

ratio

0…*

Fig. 2. The e3service customer perspective ontology

original social chart. Since the tool conforms exactly to the reasoning steps from e3service , our assump-tion is that we can also validate the usefulness of e3service through such a demonstration. The tool demoinvolved scenario walkthroughs, where realistic consumer needs - such as a customer that tries to find ameal-preparation service - were taken as starting points to show how the tool interacts with the informalcarer. For each scenario walkthrough, the domain expert then commented to what extent the demonstratedprinciples could constitute a useful addition to the existing social chart.

4. The e3service ontology

This section discusses the concepts and relationships of the e3service ontology, exemplified by therunning dementia-care case study. Section 5 shows how to reason with the ontology.

The e3service ontology takes two perspectives on services: the customer perspective (Sect. 4.1) and thesupplier perspective (Sect. 4.2). Additionally, there is a pricing ontology (see Sect. 4.3). Major parts ofthe e3service ontology have been published earlier, in de Kinderen (2010), de Kinderen et al. (2009) andde Kinderen et al. (2009).

We define the ontology in terms of UML class diagrams. The ontology is also available as Protégéspecification. Moreover, the ontology has been implemented in RDF (static part) and Java (reasoning part)(see Sect. 6).

4.1. The customer perspective

The customer perspective ontology, depicted in Fig. 2, concentrates on modeling concepts that representhow a customer yields value from acquiring a service bundle.

The ontology is based on concepts from established customer needs literature. Most notably, see Arndt(1978); Kotler (2000) for a discussion on needs wants and demands, Gutman and Reynolds (1982);Woodruff (1997); Gutman and Reynolds (1988); Gutman (1997) for a discussion on means-end chainingand Holbrook (1999) for a discussion on the different ways in which a product/service can be valuable.

Note that the instantiation of this ontology for our running dementia care case, a customer perspectivecatalog of dementia care services, can be found in Fig. 3. During our discussion of the concepts of thee3service ontology, we will frequently refer to this catalog to show how an instantiation of the e3serviceontology is created.

Need. A need represents a problem statement independently from a solution direction (see Arndt (1978)and Kotler (2000)).

EXAMPLE: A commonly occurring problem statement from Dutch informal carers of persons withdementia is ‘I cannot cope anymore. What can help?’. This need does not include a notion of a solutionyet, as nothing is stated about the type of support that the informal carer seeks out.

Kinderen, de S.; De Leenheer et al. An ontology for needs-driven service bundling in a multi-supplier setting. In J. of Applied Ontology, 2013 (to appear)

S.deK

inderenetal./An

ontologyfor

needs-drivenservice

bundlingin

am

ulti-suppliersetting

7

Dagsocieteit

Dinnerdelivery

I cannot cope anymore,what can help?

Meal preparation

DietSugar freeKosherFlesh as main courseMeat as main course

Diningtable

Contact type: In personPreparation: Hot

Meal preparation

Social contactsdementia-patient

Social contactsinformal carer

Physical activities forperson with dementia

Social support forperson with dementia

Social supportinformal carer

PreparationFrozenHot

….…..

….

need

Want

Functionalconsequence

Scales ofquality

consequences

Practical supportfor person with dementia

…..

…..

Loaningservice

Possibility toloan eg. an(electrical)wheelchair

OB

Duration: <= 6 Months> 6 Months

-

Handyman

Adjustment:Large, eg.StairliftMinor, eg.Ramps

Adjustments tohome

Meal delivery

…..

OB

OB

-

-

Transportation

Transportation

Contact typeInternetIn person

….

Recreational activities


Casemanagement

Keeping informedabout dementia

patient

C/E -

…..

Fig.3.Partialcustomercatalogue

forthedem

entiacase

Consequence.

Aconsequence

isanything

thatresults-directly

orindirectly-from

consuming

aservice

Gutm

anand

Reynolds

(1982,1988);

Gutm

an(1997).

Thus,a

consequenceis

definedin

abottom

-upm

anner:itresultsfromservice

consumption.W

ediscussin

detailhowa

consequenceresultsfrom

aservice

inourdiscussion

ofthesupply-side

concepts‘Serviceproperty’,‘R

esource’and‘Service

bundle’(seethe

supply-sideontology

inSect.4.2).

–Functional

consequence.A

functionalconsequence

representsthe

functionalgoal

thatcan

beachieved

throughconsum

ptionof

aservice.It

representsthe

primary

functionthat

acustom

eris

interestedin.

EXAMPLE:The

functionalconsequence‘M

ealpreparation’.–

Quality

consequence.Aquality

consequenceexpressesqualitative

propertiesofother,e.g.functional,consequences

incustom

erterm

inology.Because

itexpresses

thequalitative

propertiesof

anotherconsequence,a

qualityconsequence

cannotbeacquired

separately:Italways

dependson

(arelation


Customer Perspective Ontology & Example (1) in the domain of Assisted Living for Dementia Patients

6 S. de Kinderen et al. / An ontology for needs-driven service bundling in a multi-supplier setting

Specified by

Depends on

Core enhancing

1…*

0…*

0…*

Functional consequence

Contained in

0…*

Scale

0..1Has

0…*Optional bundling

Want

Quality consequence

Need

0..*

Consists of

Has 1…*

Has 0...1

Core enhancing0…*

Optional bundling

Consequence

nominal

ordinal

interval

ratio

0…*

Fig. 2. The e3service customer perspective ontology

original social chart. Since the tool conforms exactly to the reasoning steps from e3service , our assump-tion is that we can also validate the usefulness of e3service through such a demonstration. The tool demoinvolved scenario walkthroughs, where realistic consumer needs - such as a customer that tries to find ameal-preparation service - were taken as starting points to show how the tool interacts with the informalcarer. For each scenario walkthrough, the domain expert then commented to what extent the demonstratedprinciples could constitute a useful addition to the existing social chart.

4. The e3service ontology

This section discusses the concepts and relationships of the e3service ontology, exemplified by therunning dementia-care case study. Section 5 shows how to reason with the ontology.

The e3service ontology takes two perspectives on services: the customer perspective (Sect. 4.1) and thesupplier perspective (Sect. 4.2). Additionally, there is a pricing ontology (see Sect. 4.3). Major parts ofthe e3service ontology have been published earlier, in de Kinderen (2010), de Kinderen et al. (2009) andde Kinderen et al. (2009).

We define the ontology in terms of UML class diagrams. The ontology is also available as Protégéspecification. Moreover, the ontology has been implemented in RDF (static part) and Java (reasoning part)(see Sect. 6).

4.1. The customer perspective

The customer perspective ontology, depicted in Fig. 2, concentrates on modeling concepts that representhow a customer yields value from acquiring a service bundle.

The ontology is based on concepts from established customer needs literature. Most notably, see Arndt(1978); Kotler (2000) for a discussion on needs wants and demands, Gutman and Reynolds (1982);Woodruff (1997); Gutman and Reynolds (1988); Gutman (1997) for a discussion on means-end chainingand Holbrook (1999) for a discussion on the different ways in which a product/service can be valuable.

Note that the instantiation of this ontology for our running dementia care case, a customer perspectivecatalog of dementia care services, can be found in Fig. 3. During our discussion of the concepts of thee3service ontology, we will frequently refer to this catalog to show how an instantiation of the e3serviceontology is created.

Need. A need represents a problem statement independently from a solution direction (see Arndt (1978)and Kotler (2000)).

EXAMPLE: A commonly occurring problem statement from Dutch informal carers of persons withdementia is ‘I cannot cope anymore. What can help?’. This need does not include a notion of a solutionyet, as nothing is stated about the type of support that the informal carer seeks out.

Kinderen, de S.; De Leenheer et al. An ontology for needs-driven service bundling in a multi-supplier setting. In J. of Applied Ontology, 2013 (to appear)

want (≠ consequence) is a set of consequences at least one party likes to offer (NAPCS, etc.)

S.deK

inderenetal./An

ontologyfor

needs-drivenservice

bundlingin

am

ulti-suppliersetting

7

Dagsocieteit

Dinnerdelivery

I cannot cope anymore,what can help?

Meal preparation

DietSugar freeKosherFlesh as main courseMeat as main course

Diningtable

Contact type: In personPreparation: Hot

Meal preparation


Social contactsinformal carer

Physical activities forperson with dementia

Social support forperson with dementia

Social supportinformal carer

PreparationFrozenHot

….…..

….

need

Want

Functionalconsequence

Scales ofquality

consequences

Practical supportfor person with dementia

…..

…..

Loaningservice

Possibility toloan eg. an(electrical)wheelchair

OB

Duration: <= 6 Months> 6 Months

-

Handyman

Adjustment:Large, eg.StairliftMinor, eg.Ramps

Adjustments tohome

Meal delivery

…..

OB

OB

-

-

Transportation

Transportation

Contact typeInternetIn person

….

Recreational activities


Casemanagement

Keeping informedabout dementia

patient

C/E -

…..

Fig.3.Partialcustomercatalogue

forthedem

entiacase

Consequence.

Aconsequence

isanything

thatresults-directly

orindirectly-from

consuming

aservice

Gutm

anand

Reynolds

(1982,1988);

Gutm

an(1997).

Thus,a

consequenceis

definedin

abottom

-upm

anner:itresultsfromservice

consumption.W

ediscussin

detailhowa

consequenceresultsfrom

aservice

inourdiscussion

ofthesupply-side

concepts‘Serviceproperty’,‘R

esource’and‘Service

bundle’(seethe

supply-sideontology

inSect.4.2).

–Functional

consequence.A

functionalconsequence

representsthe

functionalgoal

thatcan

beachieved

throughconsum

ptionof

aservice.It

representsthe

primary

functionthat

acustom

eris

interestedin.

EXAMPLE:The

functionalconsequence‘M

ealpreparation’.–

Quality

consequence.Aquality

consequenceexpressesqualitative

propertiesofother,e.g.functional,consequences

incustom

erterm

inology.Because

itexpresses

thequalitative

propertiesof

anotherconsequence,a

qualityconsequence

cannotbeacquired

separately:Italways

dependson

(arelation


Customer Perspective Ontology & Example (2) in the Educational Domain

• consequences generated from open databases: http://www.accreditedqualifications.org.uk


This reasoning step is performed by a human user who is guided by thecustomer catalogue in Fig. 4.3 and the relationships defined in the e3service Customer catalogueontology in Fig. 3.1. For instance, in Fig. 4.3, the customer need “How can Iimprove my programming skills?” can be refined into the FCs : Web Application Customer needDevelopment, Event Driven Programming and Data Analysis and Design, whichcan be refined into more detailed FCs that better describe a customer needin terms of specific requirements [25, 76, 79]. In this case, if the customerchooses Web Application Developments, it can be refined into three specificFCs : Designing and developing a web site (FC1), website management (FC2)and web server scripting (FC3), as depicted in Fig. 4.3.

Figure 4.3: Customer catalogue designed with the ontology in Fig. 3.1.

The set of requested FCs is denoted by F while its cardinality is denotedby M = |F|. In this way, F = {FC1, FC2, . . . , FC

M

} which is reckon by the Defining F and Mcustomer through the laddering step, and M is the the number of specific FCsbeing requested.

4.1.1.3 Matching

Once a service catalogue has been defined and a customer need has beenmapped onto FCs, the next step requires to match the customer requested FCs(F) with all the services in the service catalogue that can completely or partiallyo↵er the FCs as required by the customer [76, 79]. The matched services, i.e.the services that can o↵er all or some FCs, are then grouped into a matchingpool (MP). Matching pool

Fig. 4.4 represents graphically the matching process. The lines matchingfunctional consequences on both sides are a graphical representation of a stringmatching process in which customer requested FCs retrieve supplier o↵ered FCs. String matchingIn the example, we can see that for the requested F = {FC1, FC3, FC5}, the


http://www.accreditedqualifications.org.uk/




Supplier Perspective Ontology



value activity



value activity

value object



value activity

value object

value interface: reciprocity


Service Value Network for Edu Services

64 Interactive Composition of SVNs

B2B relationships, i.e. what can be exchanged not only between customers andsuppliers but also between suppliers and enablers.

By making use of the supplier ontology described in Sect. 3.2.1, service sup-pliers and enablers (actors) can describe their o↵erings in terms of functionalconsequences (FCs), i.e. what functionalities they can o↵er to the customers.This is an o↵-line inference since the service catalogue must be previously de-signed. In this way, when the composition starts, the service o↵erings can beService catalogueretrieved from a service catalogue.

Figure 4.2: A service catalogue designed with the ontology in Fig. 3.3.

The Fig. 4.2 depicts a sample of a service catalogue in which both servicesuppliers and enablers represent what they can o↵er to customers and servicesupplier respectively. The set of service suppliers and enablers is denoted bySC, and its cardinality is given by S = |SC| .Defining SC and S

4.1.1.2 Laddering

As we already explained in Sect. 3.1, since a customer need is usually ex-pressed in abstract terms that might lead to di↵erent solutions, it is necessaryto perform a mapping from those abstract terms onto specific requirements tofacilitate the construction of a possible solution. Within marketing literature,the notion of laddering has been widely used to represent how customers linkspecific product attributes to high-level values [51, 25]. In our case, by makinguse of a customer catalogue (e.g. Fig. 4.3) built with the e3service customerontology in Sect. 3.1, we can apply laddering to refine customer needs into FCs.

3.2. SERVICE SUPPLIERS 51

(in this case an educational service) by means of the generic service supplierontology explained in Sect. 3.2.2 and Sect. 3.2.3. We have harvested a pub-licly available database of educational services and selected only one serviceto exemplify how the supplier ontology is used. The database is available athttp://register.ofqual.gov.uk/, the website of the National Database of Accred-ited Qualifications (NDAQ) containing details of recognized awarding organiza-tions and regulated qualifications in England, Wales and Northern Ireland.

Figure 3.2: Example of a service profile.

Fig. 3.2 depicts how The City and Guilds of London Institute (actor) per-forms a value activity (service element - Teaching Course 500/3474/1) thato↵ers a Diploma in ICT (value object) through the value port P1 in the valueinterface VI1. As can be observed, Value Interface Instances are depicted as A servicerows, Value Port Instances are part of Value Interface Instances and AttributeInstances are depicted as cells being part of Value Port Instances. In this ex-ample we represent that the Diploma in ICT can have specific attributes suchas ID and Assessment (A or B 3). Moreover, depending on the values of the at-tributes, The City and Guilds of London Institute can require di↵erent amountsof money for the Diploma being o↵ered, which is depicted in the instances ofthe value port P2.

Furthermore, as elaborated in Sect. 3.2.3, value interface instances can havedependencies with other interface instances. The idea is that, because valueobjects such as the Diploma in ICT can have very specific attributes, othervalue objects also with specific attributes can be required. In this case, thevalue interface instance with attribute ID 222 depends on the value interfaceinstance with attribute ID 701. This kind of dependencies allows services tofocus on their core business (in this case the course 500/3474/1) and delegate

3A = E-assessment, and B = Practical Demonstration/Assignment


Matching

• generating the candidate service space

• explosive !


Figure 4.4: Matching FCs.

MP contains the services A,B,C,D and E since A o↵ers FC1, B o↵ers FC3,C o↵ers FC1, D o↵ers FC3 and E o↵ers FC5.

The MP is actually a subset of SC, i.e. MP ✓ SC, where SC represents allthe services stored in the service catalogue as defined in Sect. 4.1.1.1. Once theDefining MPMP has been computed, the next step is to find the combinations of services(i.e. service bundles) that can jointly satisfy a customer need by providing therequested FCs.

4.1.1.4 Bundling

As observed in the matching step, because sometimes a single service cannot provide all the required FCs, it is necessary to combine two or more servicesto provide a solution (in terms of FCs) for the customer need. In this step(bundling), therefore, the objective is to find groups of services that jointlyo↵er a solution for the customer. These groups of services working together areknown as service bundles .

Furthermore, since the generation of such bundles is similar to combinato-rial optimization problems where exhaustive search is not feasible, heuristics totackle this issue are required [21, 76], In previous research [25], it was assumedthat the service bundles were already present within the environment. A keyAutomatically bundling

services contribution of this research is that we remove that assumption and present anew algorithm to automatically generate service bundles.

Although a complete solution for a customer need is a service bundle o↵eringall the required FCs, most of the time it is not possible to find such solutionbecause of two reasons: 1) Incomplete Solution Space - there are no candidate


Bundling

• clustering

• push heuristic


services for the required FCs, i.e. customer requested FCs can not be foundon the supplier side, and 2) Overlapping - there is no service bundle in whichthe service suppliers do not overlap in term of FCs . This issue is related to the Complete and incomplete

solutionsassumption that, although two service suppliers can o↵er di↵erent value objects,they can not be part of the same service bundle if the value objects provide atleast one similar FC, i.e. there is one FC overlapping.

For instance, consider two educational services. The first service o↵eringa Diploma in ICT 1 that provides FCs such as Testing ICT Systems (FC1),Object Oriented Programing (FC2) and Database Software Use (FC3). Thesecond one o↵ering a Certificate in ICT Skills 2 that provides the followingFCs : Data analysis and data structure design (FC4), and Database SoftwareUse (FC3). As can be observed, it would be meaningless to o↵er both services Services overlappingin the same bundle since they both provide Database Software Use.

Furthermore, if the customer need requires FC1 and FC4, two incompletesolutions must be presented to the customer, i.e. two bundles composed of oneservice each: the first bundle composed of the service o↵ering the Diploma inICT and the second bundle composed of the service o↵ering the Certificate inICT Skills.

�Service ID FC1 FC2 FC3 FC4

S17 1 1 1 0S16 1 1 0 1S15 1 0 1 1S14 1 0 1 1S13 1 0 0 1S12 1 0 0 0S11 1 0 0 0S10 0 1 1 1S9 0 1 0 1S8 0 1 0 1S7 0 1 0 0S6 0 0 1 1S5 0 0 1 0S4 0 0 1 0S3 0 0 0 1S2 0 0 0 1S1 0 0 0 1

Table 4.1: Matrix representation for a matching pool (MP). S15 and S14 areservices, that might be provided by di↵erent suppliers, o↵ering exactly thesame requested FCs. The same holds for S11�12, S8�9, S4�5, S1�3.

In this step we present a bundling algorithm that takes into account the twoissues described above (Incomplete Search Space and Overlapping) and gener-ates service bundles for the requested FCs (again service bundles can providecomplete or incomplete solutions depending on the FCs being requested). The

1O↵ered by The City and Guilds of London Institute with the Qualification Number500/3474/1

2O↵ered by ABC Awards with the Qualification Number 100/6398/5


services for the required FCs, i.e. customer requested FCs can not be foundon the supplier side, and 2) Overlapping - there is no service bundle in whichthe service suppliers do not overlap in term of FCs . This issue is related to the Complete and incomplete

solutionsassumption that, although two service suppliers can o↵er di↵erent value objects,they can not be part of the same service bundle if the value objects provide atleast one similar FC, i.e. there is one FC overlapping.

For instance, consider two educational services. The first service o↵eringa Diploma in ICT 1 that provides FCs such as Testing ICT Systems (FC1),Object Oriented Programing (FC2) and Database Software Use (FC3). Thesecond one o↵ering a Certificate in ICT Skills 2 that provides the followingFCs : Data analysis and data structure design (FC4), and Database SoftwareUse (FC3). As can be observed, it would be meaningless to o↵er both services Services overlappingin the same bundle since they both provide Database Software Use.

Furthermore, if the customer need requires FC1 and FC4, two incompletesolutions must be presented to the customer, i.e. two bundles composed of oneservice each: the first bundle composed of the service o↵ering the Diploma inICT and the second bundle composed of the service o↵ering the Certificate inICT Skills.

�Service ID FC1 FC2 FC3 FC4

S17 1 1 1 0S16 1 1 0 1S15 1 0 1 1S14 1 0 1 1S13 1 0 0 1S12 1 0 0 0S11 1 0 0 0S10 0 1 1 1S9 0 1 0 1S8 0 1 0 1S7 0 1 0 0S6 0 0 1 1S5 0 0 1 0S4 0 0 1 0S3 0 0 0 1S2 0 0 0 1S1 0 0 0 1

Table 4.1: Matrix representation for a matching pool (MP). S15 and S14 areservices, that might be provided by di↵erent suppliers, o↵ering exactly thesame requested FCs. The same holds for S11�12, S8�9, S4�5, S1�3.

In this step we present a bundling algorithm that takes into account the twoissues described above (Incomplete Search Space and Overlapping) and gener-ates service bundles for the requested FCs (again service bundles can providecomplete or incomplete solutions depending on the FCs being requested). The

1O↵ered by The City and Guilds of London Institute with the Qualification Number500/3474/1

2O↵ered by ABC Awards with the Qualification Number 100/6398/5


Bundling

• clustering

• push heuristic


given algorithm is composed of four steps: 1) Cluster generation, 2) Clustercombination, 3) Cluster solution and 4) Cluster expansion.

1) Cluster generation Since FCs are the matching point between the cus-tomer and supplier ontologies, it is meaningful to represent a service by meansof its FCs. For instance, if the required consequences are: FC1, FC2, FC3

and FC4, a service providing FC1, FC2 and FC3 can be represented througha binary vector � = [1110], where the first three 1s stand for the FCs o↵eredby the service, and the last 0 means that the service does not provide FC4.Moreover, we highlight that this sorting does not imply any preference aboutFCs, i.e. all the FCs might be equally important. Later in this chapter weexplain how customer preferences over FCs are addressed (Sect. 4.1.2).

In this way, a matching pool (MP ) of services can be described as depicted inTable 4.1. Moreover, based on its � vector, services can be clustered, i.e. servicesare assigned to the same cluster if they o↵er exactly the same FCs. The purposeof a cluster is to group services o↵ering exactly the same FCs as well as to focusthe searching of possible bundles on the clusters , i.e. explore combinations ofSearching focus on clustersclusters rather than combinations of a huge number of individual services. Inaddition, since a cluster contains one or more services, it can be seen as a set ofpartial solutions for a customer need.

We identify two types of clusters, upper and lower clusters. The Table 4.2depicts the set of upper clusters. As can be observed, because all the upperclusters provide FC1, i.e. they are overlapped in FC1, they cannot be combinedwith each other. The name upper cluster comes from FC1 being the mostsignificant bit (msb) in the vector �, i.e. they have the highest � values.

Cluster ID Elements Cluster.� Cluster.msb

C14 {S17} [1110] 8C13 {S16} [1101] 8C11 {S14, S15} [1011] 8C9 {S13} [1001] 8C8 {S11, S12} [1000] 8

Table 4.2: Upper Clusters.

2) Cluster combination Contrary to the upper clusters, some of the lowerclusters can be combined with each other. Two lower clusters (see Table 4.3) canbe combined if, and only if, their FCs do not overlap. E.g., since C1.� = [0001]and C2.� = [0010] do not overlap, C1 and C2 can be combined and added toC3’s elements (C3 becoming a merging cluster, see Table 4.6 for the final result).

We use the C1 � C2 expression to denote that the services inside C1 can becombined with the services within C2 . In simple words, it means that three� to denote a combination

of clusters steps are performed. 1) take an element from C1, 2) take an element from C2,and 3) combine the two elements. Consequently, based on Table 4.3, C1 � C2

produces the following combinations: {S1, S4} , {S1, S5}, {S2, S4}, {S2, S5},{S3, S4}, {S3, S5}.

To combine lower and upper clusters we follow a heuristic called push onelevel up. The idea is that clusters can push their elements (services) to a clusterPush one level up



C7 {S10} [0111] 4C5 {S8, S9} [0101] 4C4 {S7} [0100] 4C3 {S6} [0011] 2C2 {S4, S5} [0010] 2C1 {S1, S2, S3} [0001] 1

Table 4.3: Lower Clusters.

with higher msb. Moreover, we apply a bottom-up approach starting at thecluster with the lowest msb and sequentially moving up until reaching the lowercluster with the highest msb. In short, the heuristic is that the elements of acluster (services and other clusters) only need to be pushed one level up (intoclusters with higher msb) since at some point they will be combined with upperclusters that provide the FCs that they are missing, i.e. generating solutionsproviding more FCs, in the end, it would be possible to generate completesolutions. Moreover, the decision of pushing clusters just one level up obeysalso to a save-computation-time strategy.

In this way, a cluster Cx

can only be combined with clusters CY

that meetthe following conditions:

1. CY

.msb is equal to (Cx

.msb) ⇤ 2, and

2. CY

.� does not overlap with Cx

.�, i.e. they do not provide the same FCs.

For instance, C2 can be combined with C4 and C5 since

1. C4.msb = C5.msb = (C2.msb) ⇤ 2 = 4, and

2. C4.� and C5.� do not overlap with C2.�.

The Table 4.4 depicts all the operations that are performed following theheuristic push one level up. As can be observed, when combining clusters, newclusters can be generated . For instance, the operation C2 � C4 generates the Generating new clustersnew cluster C6 (see Table 4.6 for the final result). In addition, the new generatedclusters can also push their elements levels up since the pushing process startswith the lowest bundles, e.g. C6 is later combined with C8 and C9. Moreover,it is also possible not only to generate new clusters but also to generate newclusters that provide all the FCs, e.g. C6 � C9 and C7 � C8 both push theirelements into the new generated cluster C15.

The Tables 4.5 and Table 4.6 show respectively the upper and lower clustersafter applying the heuristic push one level up. To facilitate the reading, we usethe operation C

x

�Cy

to indicate which clusters are combined within a mergingcluster. For example, C14 (a merging cluster) contains the service s17 as well asthe combinations of services produced by the operation C6 � C8.

3) Cluster solution Once the upper and lower clusters have been repro-cessed (using the push one level up heuristic), the next step is to generate thepool of solution clusters. A solution cluster might contain single services or a


given algorithm is composed of four steps: 1) Cluster generation, 2) Clustercombination, 3) Cluster solution and 4) Cluster expansion.

1) Cluster generation Since FCs are the matching point between the cus-tomer and supplier ontologies, it is meaningful to represent a service by meansof its FCs. For instance, if the required consequences are: FC1, FC2, FC3

and FC4, a service providing FC1, FC2 and FC3 can be represented througha binary vector � = [1110], where the first three 1s stand for the FCs o↵eredby the service, and the last 0 means that the service does not provide FC4.Moreover, we highlight that this sorting does not imply any preference aboutFCs, i.e. all the FCs might be equally important. Later in this chapter weexplain how customer preferences over FCs are addressed (Sect. 4.1.2).

In this way, a matching pool (MP ) of services can be described as depicted inTable 4.1. Moreover, based on its � vector, services can be clustered, i.e. servicesare assigned to the same cluster if they o↵er exactly the same FCs. The purposeof a cluster is to group services o↵ering exactly the same FCs as well as to focusthe searching of possible bundles on the clusters , i.e. explore combinations ofSearching focus on clustersclusters rather than combinations of a huge number of individual services. Inaddition, since a cluster contains one or more services, it can be seen as a set ofpartial solutions for a customer need.

We identify two types of clusters, upper and lower clusters. The Table 4.2depicts the set of upper clusters. As can be observed, because all the upperclusters provide FC1, i.e. they are overlapped in FC1, they cannot be combinedwith each other. The name upper cluster comes from FC1 being the mostsignificant bit (msb) in the vector �, i.e. they have the highest � values.


C14 {S17} [1110] 8C13 {S16} [1101] 8C11 {S14, S15} [1011] 8C9 {S13} [1001] 8C8 {S11, S12} [1000] 8

Table 4.2: Upper Clusters.

2) Cluster combination Contrary to the upper clusters, some of the lowerclusters can be combined with each other. Two lower clusters (see Table 4.3) canbe combined if, and only if, their FCs do not overlap. E.g., since C1.� = [0001]and C2.� = [0010] do not overlap, C1 and C2 can be combined and added toC3’s elements (C3 becoming a merging cluster, see Table 4.6 for the final result).

We use the C1 � C2 expression to denote that the services inside C1 can becombined with the services within C2 . In simple words, it means that three� to denote a combination

of clusters steps are performed. 1) take an element from C1, 2) take an element from C2,and 3) combine the two elements. Consequently, based on Table 4.3, C1 � C2

produces the following combinations: {S1, S4} , {S1, S5}, {S2, S4}, {S2, S5},{S3, S4}, {S3, S5}.

To combine lower and upper clusters we follow a heuristic called push onelevel up. The idea is that clusters can push their elements (services) to a clusterPush one level up



C7 {S10} [0111] 4C5 {S8, S9} [0101] 4C4 {S7} [0100] 4C3 {S6} [0011] 2C2 {S4, S5} [0010] 2C1 {S1, S2, S3} [0001] 1

Table 4.3: Lower Clusters.

with higher msb. Moreover, we apply a bottom-up approach starting at thecluster with the lowest msb and sequentially moving up until reaching the lowercluster with the highest msb. In short, the heuristic is that the elements of acluster (services and other clusters) only need to be pushed one level up (intoclusters with higher msb) since at some point they will be combined with upperclusters that provide the FCs that they are missing, i.e. generating solutionsproviding more FCs, in the end, it would be possible to generate completesolutions. Moreover, the decision of pushing clusters just one level up obeysalso to a save-computation-time strategy.

In this way, a cluster Cx

can only be combined with clusters CY

that meetthe following conditions:

1. CY

.msb is equal to (Cx

.msb) ⇤ 2, and

2. CY

.� does not overlap with Cx

.�, i.e. they do not provide the same FCs.

For instance, C2 can be combined with C4 and C5 since

1. C4.msb = C5.msb = (C2.msb) ⇤ 2 = 4, and

2. C4.� and C5.� do not overlap with C2.�.

The Table 4.4 depicts all the operations that are performed following theheuristic push one level up. As can be observed, when combining clusters, newclusters can be generated . For instance, the operation C2 � C4 generates the Generating new clustersnew cluster C6 (see Table 4.6 for the final result). In addition, the new generatedclusters can also push their elements levels up since the pushing process startswith the lowest bundles, e.g. C6 is later combined with C8 and C9. Moreover,it is also possible not only to generate new clusters but also to generate newclusters that provide all the FCs, e.g. C6 � C9 and C7 � C8 both push theirelements into the new generated cluster C15.

The Tables 4.5 and Table 4.6 show respectively the upper and lower clustersafter applying the heuristic push one level up. To facilitate the reading, we usethe operation C

x

�Cy

to indicate which clusters are combined within a mergingcluster. For example, C14 (a merging cluster) contains the service s17 as well asthe combinations of services produced by the operation C6 � C8.

3) Cluster solution Once the upper and lower clusters have been repro-cessed (using the push one level up heuristic), the next step is to generate thepool of solution clusters. A solution cluster might contain single services or a


Bundling

• clustering

• push heuristic70 Interactive Composition of SVNs

Clusters to be combined Merging Cluster

C1 � C2 ! C3

C2 � C4 ! C6

C2 � C5 ! C7

C3 � C4 ! C7

C4 � C8 ! C12

C4 � C9 ! C13

C5 � C8 ! C13

C6 � C8 ! C14

C6 � C9 ! C15

C7 � C8 ! C15

Table 4.4: Operations.

C. ID Elements Cluster.� Cluster.msb

C15 {C6 � C9, C7 � C8} [1111] 8C14 {S17, C6 � C8} [1110] 8C13 {S16, C4 � C9, C5 � C8} [1101] 8C12 {C4 � C8} [1100] 8C11 {S14, S15} [1011] 8C9 {S13} [1001] 8C8 {S11, S12} [1000] 8

Table 4.5: Upper Clusters - after merging.


C7 {S10, C2 � C5, C3 � C4} [0111] 4C6 {C2 � C4} [0110] 4C5 {S8, S9} [0101] 4C4 {S7} [0100] 4C3 {S6, C1 � C2} [0011] 2C2 {S4, S5} [0010] 2C1 {S1, S2, S3} [0001] 1

Table 4.6: Lower Clusters - after merging.

combination of clusters such that their FCs: 1) do not overlap, and 2) matchall the required consequences. E.g., if we combine C14 with C1, we can ob-serve that C14.� does not overlap with C1.�, besides C14 � C1 provide all therequired consequences, i.e. C14.� = [1110] together with C1.� = [0001] provide, FC1, FC2, FC3 and FC4. Therefore, C14 � C1 is a solution cluster.

4) Cluster expansion Finally, we must expand the solution clusters to geta pool of solution bundles. The Table 4.7 depicts the pool of solution clusterstogether with the pool of solution bundles. As can be observed, all the solution



C1 � C2 ! C3

C2 � C4 ! C6

C2 � C5 ! C7

C3 � C4 ! C7

C4 � C8 ! C12

C4 � C9 ! C13

C5 � C8 ! C13

C6 � C8 ! C14

C6 � C9 ! C15

C7 � C8 ! C15











Bundling

• clustering

• push heuristic



C1 � C2 ! C3

C2 � C4 ! C6

C2 � C5 ! C7

C3 � C4 ! C7

C4 � C8 ! C12

C4 � C9 ! C13

C5 � C8 ! C13

C6 � C8 ! C14

C6 � C9 ! C15

C7 � C8 ! C15












C1 � C2 ! C3

C2 � C4 ! C6

C2 � C5 ! C7

C3 � C4 ! C7

C4 � C8 ! C12

C4 � C9 ! C13

C5 � C8 ! C13

C6 � C8 ! C14

C6 � C9 ! C15

C7 � C8 ! C15












C1 � C2 ! C3

C2 � C4 ! C6

C2 � C5 ! C7

C3 � C4 ! C7

C4 � C8 ! C12

C4 � C9 ! C13

C5 � C8 ! C13

C6 � C8 ! C14

C6 � C9 ! C15

C7 � C8 ! C15












C1 � C2 ! C3

C2 � C4 ! C6

C2 � C5 ! C7

C3 � C4 ! C7

C4 � C8 ! C12

C4 � C9 ! C13

C5 � C8 ! C13

C6 � C8 ! C14

C6 � C9 ! C15

C7 � C8 ! C15











Bundling

• clustering

• push heuristic


bundles provide all the required FCs. E.g. the bundle composed of the servicesS4, S7 and S13 is a complete solution since S4 provides FC3, S7 provides FC2

and S13 provides FC1 and FC4.

Solution Clusters Solution Bundles

C6 � C9 {C2 � C4}� C9

C7 � C8 {{S10}, {C3 � C4}, {C2 � C5}}� C8

C6 � C9 {{S4, S7}, {S5, S7}}� S13

C7 � C8 {{S10}, {S1, S4, S7}, {S1, S5, S7},{S2, S4, S7}, {S2, S5, S7},{S3, S4, S7}, {S3, S5, S7},{S4, S8}, {S4, S9},{S5, S8}, {S5, S9}}� {S11, S12}

C6 � C9 {S4, S7, S13}, {S5, S7, S13}C7 � C8 {S10, S11}, {S10, S12},

{S1, S4, S7, S11}, {S1, S5, S7, S11},{S2, S4, S7, S11}, {S2, S5, S7, S11},{S3, S4, S7, S11}, {S3, S5, S7, S11},{S4, S8, S11}, {S4, S8, S12},{S4, S9, S11}, {S4, S9, S12}, {S5, S8, S11},{S5, S8, S12}, {S5, S9, S11}, {S5, S9, S12}

Table 4.7: The pools of solution clusters and bundles.

The Algorithm 1 depicts the pseudo code of our proposed bundling algo-rithm, which takes as inputs the set of required consequences (F) and thematching pool (MP ). As can be observed, there are four main steps, wherethe first three steps are mainly driven by M , which is the cardinality of F , i.e.the number of requested FCs.

Quick review The four steps are performed as follows. First, the servicesare clustered based on a binary vector � which represents the FCs o↵ered byeach one of the services within the cluster. Second, the so-called lower clus-ters are combined with each other. As already mentioned, the clusters can becombined if, and only if, their FCs do not overlap. In this way we assurethat a cluster keeps the property of being a pool of non-overlapping services.Third, it generates the solution clusters by combining the upper clusters withthe lower clusters. Finally, the algorithm expands the pool of solution clustersand generates the pool of solution bundles.

To finalize, since our bundling algorithm is based on heuristic reasoning,neither all the possible service bundles nor the optimal bundles are guaranteedto be generated. Nonetheless, the bundling algorithm o↵ers a tractable way togenerate service bundles matching customer requirements.

4.1.1.5 B2B Linking

The Fig. 4.5 depicts two service bundles composed of three services o↵ering threevalue objects and performed by two actors (educational institutions). Assumingthat the laddering comes up with the FCs : Data structures and algorithms


bundles provide all the required FCs. E.g. the bundle composed of the servicesS4, S7 and S13 is a complete solution since S4 provides FC3, S7 provides FC2

and S13 provides FC1 and FC4.

Solution Clusters Solution Bundles

C6 � C9 {C2 � C4}� C9

C7 � C8 {{S10}, {C3 � C4}, {C2 � C5}}� C8

C6 � C9 {{S4, S7}, {S5, S7}}� S13

C7 � C8 {{S10}, {S1, S4, S7}, {S1, S5, S7},{S2, S4, S7}, {S2, S5, S7},{S3, S4, S7}, {S3, S5, S7},{S4, S8}, {S4, S9},{S5, S8}, {S5, S9}}� {S11, S12}

C6 � C9 {S4, S7, S13}, {S5, S7, S13}C7 � C8 {S10, S11}, {S10, S12},

{S1, S4, S7, S11}, {S1, S5, S7, S11},{S2, S4, S7, S11}, {S2, S5, S7, S11},{S3, S4, S7, S11}, {S3, S5, S7, S11},{S4, S8, S11}, {S4, S8, S12},{S4, S9, S11}, {S4, S9, S12}, {S5, S8, S11},{S5, S8, S12}, {S5, S9, S11}, {S5, S9, S12}

Table 4.7: The pools of solution clusters and bundles.

The Algorithm 1 depicts the pseudo code of our proposed bundling algo-rithm, which takes as inputs the set of required consequences (F) and thematching pool (MP ). As can be observed, there are four main steps, wherethe first three steps are mainly driven by M , which is the cardinality of F , i.e.the number of requested FCs.

Quick review The four steps are performed as follows. First, the servicesare clustered based on a binary vector � which represents the FCs o↵ered byeach one of the services within the cluster. Second, the so-called lower clus-ters are combined with each other. As already mentioned, the clusters can becombined if, and only if, their FCs do not overlap. In this way we assurethat a cluster keeps the property of being a pool of non-overlapping services.Third, it generates the solution clusters by combining the upper clusters withthe lower clusters. Finally, the algorithm expands the pool of solution clustersand generates the pool of solution bundles.

To finalize, since our bundling algorithm is based on heuristic reasoning,neither all the possible service bundles nor the optimal bundles are guaranteedto be generated. Nonetheless, the bundling algorithm o↵ers a tractable way togenerate service bundles matching customer requirements.

4.1.1.5 B2B Linking

The Fig. 4.5 depicts two service bundles composed of three services o↵ering threevalue objects and performed by two actors (educational institutions). Assumingthat the laddering comes up with the FCs : Data structures and algorithms


Animation of the Bundling

• http://www.youtube.com/watch?v=sUnPs53F-cA


http://www.youtube.com/watch?v=sUnPs53F-cA

http://www.youtube.com/watch?v=sUnPs53F-cA

e3value Business Proposition for the Service Bundles


Algorithm 1 Bundling algorithm - General view.1: procedure Bundling(F ,MP ) . F = required consequences, MP = matching pool2: n |MP | . number of services o↵ering at least one required FC

3: M |F| . number of required consequences4: C Generate clusters(n,M)5: Combine lower clusters(C,M)6: SolC Get solution clusters(C,M)7: solutions Expand solution clusters(SolC)8: return solutions

9: end procedure

Figure 4.5: Alternative Service Bundles.

(FC1), Data Analysis and Data Structure Design (FC2), and Data Represen-tation and Manipulation for IT (FC3), these bundles represent two alternativesto solve the customer need depicted in Fig. 4.3.

Moreover, as explained in Sect. 3.2, since service suppliers can delegate non-core-business activities to other entities, the suppliers within a bundle can alsohave dependencies with service enablers, i.e. services that support the function-Service enablersing of a service supplier.

At this step, we solve the B2B dependencies that service suppliers withinB2B dependenciesbundles might have [40, 78, 79]. To this aim, we make use of the service profilesof each service supplier, gathering information about not only what value objectsthey request but also what kind of dependencies they have based on their valueinterface instances (see Fig. 3.2). The process consists of looking for serviceswithin the service catalogue that can provide the resources required by a suppliertaking care of the constraints that might be imposed by the value interfaceinstances, i.e. matching services that o↵er value objects with specific attributes.

For instance, within a bundle, an educational service o↵ering a given courseto the customer side might rely on a service such as a digital library that allowsstudents to access reading material. Although the final customer only cares


B2B Linking with e.g., Software Services4.1. SVN COMPOSITION 73

Figure 4.6: e3value model of a SVN for a specific customer need requiring: Datastructures and algorithms (FC1), Data Analysis and Data Structure Design(FC2), and Data Representation and Manipulation for IT (FC3).

about the B2C relationships with the service bundle, the educational serviceneeds to solve its dependencies with other service enablers to allow the servicebundle being sustainable. The Fig. 4.6 depicts an SVN generated based onsolving the B2B dependencies of the Service Bundle in Fig. 4.5.

The Propose subsection (See 4.1.1) has explained how the first subtask ofthe propose-verify-critique-modify (PVCM) methods is addressed. Briefly, thePropose subtask is composed of five inferences which can be summarized asfollows:

• Laddering: maps a customer need onto specific Functional Consequences(FCs),

• O↵ering: describe the service o↵erings in term of FCs,

• Matching: takes as input a customer need expressed in terms of FCs andfinds the pool of services that can partially or completelly provide the FCsrequired by the customer,

• Bundling: having a pool of services that plausibly can provide the customer-required FCs, this inference generates the combinations of one or moreservices (service bundles) that can jointly o↵er a solution to a customerneed,

• B2B Linking: in case the service bundles might contain dependencies withother services, this inference finds the service enablers that support the


Verification

• checking whether the design satisfies functional and non-functional specifications


Figure 4.11: SVNs o↵ering di↵erent FCs. Customer requirements are given byFC1, FC2 and FC3. Although SV N1 o↵ers the requested FCs, it also o↵ersa non-required FC. Moreover, SV N2 and SV N3 both only o↵er two of therequested FCs.

SV N1 SV N2 SV N3

P Many (1.0) Some (0.90), Many (0.10) Some (0.45), Many (0.45)M Few (1.0) Few (0.10), Some (0.90) Few (0.45), Some (0.45)N Few (0.70), Some (0.25) Few (1.0) Few (0.70), Some (0.25)Applied rules 1,2 1,3,5,7 1-8Aggregation Perfect(0.7),Good(0.25) Perfect (0.1), Good(0.1), Perfect(0.45), Good(0.95)

Average(0.1), Poor(0.9) Average(0.7), Poor(0.7)

Score 8.69 3.66 5.68

Table 4.8: Important measures in the verification process for each SVN.

Fig. 4.13 illustrates how the defuzzification is performed for SV N1, SV N2

and SV N3. Finally, Table 4.8 summarizes some of the important results thatallow computing the final scores for SV N1 , SV N2 and SV N3. As can beobserved, for SV N1 only the rules 1 and 2 are applied whereas for SV N2 therules 1,3,5 and 7 are applied.

To sum up, in this subsection a Fuzzy Inference System (FIS) has been in-troduced to verify how well the SVNs composed by the propose subtask meetthe customer requirements. For each SVN, the FIS determines a score thatreflects whether the composed SVN provides the FCs as required by the cus-tomer. Based on this score the SVNs can be ranked so the customer has a betterunderstanding on which alternative SVN represents a good solution. Once theSVNs have been ranked, the customer can select one SVN (most lilkely the bestranked) and then (s)he has two options: i) to agree on the design of the selectedSVN and then take it as a solution, or ii) to provide scores for the FCs thatare provided and non-required by the given SVN, with this information otherSVNs can be composed to better match the customer requirements. The nextsubsection (4.1.3) elaborates more on these issues.

4.1.3 Critique

According to Chandrasekaran, if the design task has been unsuccessful, th cri-tique step identify the source of failure within a design [19]. Furthermore, theSources of failuremain goal at this stage is mostly about finding ways to improve the design [82].In our case, if a customer is not satisfied by any composed SVN, (s)he will iden-tify the sources of failure for a selected SVN, otherwise (s)he will select a SVNto satisfy her/his need (Fig. 4.1). More specifically, since SVNs might miss


We define two ways to express preferences for a given FC: wc

and wa

whichare defined by:

wc

: fc ! [0, 1], wa

: fc ! {0.5} (4.2)

where wc

is the weight that a customer assigns to a FC and wa

is a predefinedWeightsweight assigned to a given FC. We have chosen a 0.5 value assuming that non-required FCs have a moderate influence in the fitness of a SVN. Nonetheless,this value can be adapted to the composition context or even gathered throughcrowd-sourcing, i.e. through web-based platforms, customer preferences for FCcan be obtained. Later on, we can compute the total weights (preferences) forFC

Customer

and FCSV N

in the following way:

TWC

=X

fc

i

2FC

Customer

wc

(fci

), TWSV N

= WP

+WM

+WN

(4.3)

where TWC

is the total weight for the FCs as requested by the customer andTW

SV N

is the total weight of all FCs in a SVN. The values for WP

, WM

andW

N

are computed as follows:

WP

=X

fc

i

2FC

P

wc

(fci

), WM

=X

fc

i

2FC

M

wc

(fci

), WN

=X

fc

i

2FC

N

wa

(fci

)

(4.4)where W

P

is the sum of the weights wc

(fci

) for the functional consequencesFCs that are provided as required by the customer in a SVN. Similarly, W

M

isthe sum of the weights for FCs that are missing in a SVN, and W

N

is the sumof the weights for the FCs that are non-required in a SVN.

2) Computing Fractions: Although the computed weights (WP

,WM

andW

N

) provide basic information to what extend SVNs match the customer pref-erences for FCs, it is still necessary to have a better description in term of theproportion of FCs that are provided, missing and non-required within SVNs.To this aim, for each SVN we compute three fractions: 1) F

P

the fraction ofFractionsthe provided FCs to the total FCs as requested by the customer. 2) F

M

thefraction of the missing FCs to the total FCs as requested by the customer.3) F

N

the fraction of the non-required FCs to the total FCs in a SVN. Thesefractions are computed by Eq.4.5:

FP

=W

P

TWC

, FM

=W

M

TWC

, FN

=W

N

TWSV N

(4.5)

3) Fuzzification: The computed fractions FP

, FM

and FN

give a numericrepresentation on the proportion of (provided, missing and non-required) FCswithin SVNs. Nonetheless, to allow natural-language based reasoning about thefitness of the SVNs, it is required to map the F

P

, FM

and FN

fractions ontohuman-understandable concepts. The main advantages of such reasoning are:1) concepts are understandable (processable) for both, humans and machines,and 2) the rules to determine the fitness of SVNs can also be represented innatural language, which makes it easy to understand also for a non-technical


Calculating Fractions


We define two ways to express preferences for a given FC: wc

and wa

whichare defined by:

wc

: fc ! [0, 1], wa

: fc ! {0.5} (4.2)

where wc

is the weight that a customer assigns to a FC and wa

is a predefinedWeightsweight assigned to a given FC. We have chosen a 0.5 value assuming that non-required FCs have a moderate influence in the fitness of a SVN. Nonetheless,this value can be adapted to the composition context or even gathered throughcrowd-sourcing, i.e. through web-based platforms, customer preferences for FCcan be obtained. Later on, we can compute the total weights (preferences) forFC

Customer

and FCSV N

in the following way:

TWC

=X

fc

i

2FC

Customer

wc

(fci

), TWSV N

= WP

+WM

+WN

(4.3)

where TWC

is the total weight for the FCs as requested by the customer andTW

SV N

is the total weight of all FCs in a SVN. The values for WP

, WM

andW

N

are computed as follows:

WP

=X

fc

i

2FC

P

wc

(fci

), WM

=X

fc

i

2FC

M

wc

(fci

), WN

=X

fc

i

2FC

N

wa

(fci

)

(4.4)where W

P

is the sum of the weights wc

(fci

) for the functional consequencesFCs that are provided as required by the customer in a SVN. Similarly, W

M

isthe sum of the weights for FCs that are missing in a SVN, and W

N

is the sumof the weights for the FCs that are non-required in a SVN.

2) Computing Fractions: Although the computed weights (WP

,WM

andW

N

) provide basic information to what extend SVNs match the customer pref-erences for FCs, it is still necessary to have a better description in term of theproportion of FCs that are provided, missing and non-required within SVNs.To this aim, for each SVN we compute three fractions: 1) F

P

the fraction ofFractionsthe provided FCs to the total FCs as requested by the customer. 2) F

M

thefraction of the missing FCs to the total FCs as requested by the customer.3) F

N

the fraction of the non-required FCs to the total FCs in a SVN. Thesefractions are computed by Eq.4.5:

FP

=W

P

TWC

, FM

=W

M

TWC

, FN

=W

N

TWSV N

(4.5)

3) Fuzzification: The computed fractions FP

, FM

and FN

give a numericrepresentation on the proportion of (provided, missing and non-required) FCswithin SVNs. Nonetheless, to allow natural-language based reasoning about thefitness of the SVNs, it is required to map the F

P

, FM

and FN

fractions ontohuman-understandable concepts. The main advantages of such reasoning are:1) concepts are understandable (processable) for both, humans and machines,and 2) the rules to determine the fitness of SVNs can also be represented innatural language, which makes it easy to understand also for a non-technical


4.1.2.3 Defuzzification

Based on the fuzzy set generated during the analysis, for each SVN we computea value that is the final score of the SVN. This defuzzification process can beperformed in many di↵erent ways. Due to its simplicity, we have used a discreteversion of the so-called center of gravity (COG) method that easily computes a Center of gravity (COG)score and uses the following equation [95]:

Defuzzification(A) = D(A) =

Py

max

y

min

y ⇤A(y)P

y

max

y

min

A(y)(4.7)

where A is the fuzzy set computed during the analysis and Y is the set of ele-ments for which we want to determine a degree of truth with respect to A. Sincewe want to give scores within the range [0, 10], Y = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10}.To clarify the three steps previously described we present an hypothetical ex-ample generated based on the information gathered from the NDAQ database.

4.1.2.4 Example

Fig. 4.11 depicts a setting in which three SVNs o↵er di↵erent FCs 4. SV N1

can fully o↵er the required FCs whereas SV N2 and SV N3 are networks thatcan partially cover the customer need. For instance, SV N3 not only o↵ers FC2

and FC3 which are required FCs but also misses FC1 and o↵ers FCw

that isnot required by the customer. Assuming the customer weights in Fig. 4.11 and Computing weightsapplying Eqs. 4.2, 4.3, and 4.4, we obtain the following fractions for SV N3:

FP

=0.8 + 1.0

0.6 + 0.8 + 1.0=

1.8

2.4= 0.75 (4.8)

FM

=0.6

0.6 + 0.8 + 1.0=

0.6

2.4= 0.25 (4.9)

FN

=0.5

0.6 + 0.8 + 1.0 + 0.5=

0.5

2.9= 0.17 (4.10)

Afterwards, by making use of the membership functions in Fig. 4.8, we can Fuzzificationcompute the values for P , M and N . Since, P is some (0.45) and many (0.45),M is few (0.45) and some (0.45) and N is few (0.70) and some (0.25), duringthe analysis step the rules 1 to 8 are applied as depicted in Fig. 4.13. As can be Inference rulesobserved, the AND operator is replaced by themin function and the aggregation Aggregationstage simply sums up the rules’ outcomes.

Once the rules’ outcomes have been aggregated, the defuzzification step com- Defuzzificationputes the COG to produce the final score for SV N3 by means of Eq. 4.7, withA as depicted in Fig. 4.13.c, Y = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}. In Eq. 4.11 we de-pict how the computation was carried out which can also be visually verified inFig. 4.13.c.

D(A) =1 ⇤ 0.17 + 2 ⇤ 0.70 + 3 ⇤ 0.70 + 4 ⇤ 0.45 + 5 ⇤ 0.70

0.17 + 0.70 + 0.70 + 0.45 + 0.70 + 0.45 + 0.80 + 0.80 + 0.45 + 0.45+

6 ⇤ 0.45 + 7 ⇤ 0.80 + 8 ⇤ 0.80 + 9 ⇤ 0.45 + 10 ⇤ 0.45

0.17 + 0.70 + 0.70 + 0.45 + 0.70 + 0.45 + 0.80 + 0.80 + 0.45 + 0.45= 5.68 (4.11)

4For simplicity we only depict the FCs o↵ered by each SVN omiting the overall structure



SV N1 SV N2 SV N3



Score 8.69 3.66 5.68





4.1.3 Critique






Py

max

y

min

y ⇤A(y)P

y

max

y

min

A(y)(4.7)


4.1.2.4 Example





FP

=0.8 + 1.0

0.6 + 0.8 + 1.0=

1.8

2.4= 0.75 (4.8)

FM

=0.6

0.6 + 0.8 + 1.0=

0.6

2.4= 0.25 (4.9)

FN

=0.5

0.6 + 0.8 + 1.0 + 0.5=

0.5

2.9= 0.17 (4.10)



D(A) =1 ⇤ 0.17 + 2 ⇤ 0.70 + 3 ⇤ 0.70 + 4 ⇤ 0.45 + 5 ⇤ 0.70

0.17 + 0.70 + 0.70 + 0.45 + 0.70 + 0.45 + 0.80 + 0.80 + 0.45 + 0.45+

6 ⇤ 0.45 + 7 ⇤ 0.80 + 8 ⇤ 0.80 + 9 ⇤ 0.45 + 10 ⇤ 0.45

0.17 + 0.70 + 0.70 + 0.45 + 0.70 + 0.45 + 0.80 + 0.80 + 0.45 + 0.45= 5.68 (4.11)

4For simplicity we only depict the FCs o↵ered by each SVN omiting the overall structure


Memberships: Fuzzify


an Average SVN. Since in the fuzzification step P,M and N may have di↵erentdegrees of truth for each linguistic label (membership function), more than onerule can apply when analyzing a given SVN. E.g., as explained before, P cansimultaneously be some and many, consequently the rules in which P is notonly some but also many have to be applied, the same holds for M and N .

1: IF P is many AND M is few AND NR is few THEN Perfect2: IF P is many AND M is few AND NR is some THEN Good3: IF P is many AND M is some AND NR is few THEN Good4: IF P is many AND M is some AND NR is some THEN Good5: IF P is some AND M is few AND NR is few THEN Average6: IF P is some AND M is few AND NR is some THEN Average7: IF P is some AND M is some AND NR is few THEN Poor8: IF P is some AND M is some AND NR is some THEN Poor... ... ... ... ... ... ... ... ... ... ... ... ... ... ...N: IF P is few AND M is many AND NR is many THEN Bad

Figure 4.9: Inference rules.

To apply the inference rules another observation is important. Since thedegrees of truth are expressed in values within the range [0, 1], the traditionalboolean AND operation does not work here, i.e. after applying the rule (1) asAND operationdepicted in Fig. 4.9, IF P is many AND M is few AND N is few the valuecannot be 0 or 1 but a value that reflects the degrees of truth of P , M and N .The AND operator in fuzzy logic is usually replaced by the min function [97],i.e. the output of the rule is the minimum degree of truth among P , M and N .E.g. for the rule (1), if P = 0.6, M = 0.4 and N = 0.6, the output is Perfectwith a degree of truth 0.4.

Aggregation Once the inference rules has been applied, their outputs must beaggregated into a new fuzzy set. Several aggregation methods can be used at thispoint, in our case we sum up the outcomes for each applied rule and combinethem into a fuzzy set given by the membership functions in Fig. 4.10. Theprocess is analogous to fill a set of silos based on the outcomes of the applied rules(the shape of the silos is given by the functions in Fig. 4.10). E.g. If the rules 2,Defuzzification functions4 and 6 (as depicted in Fig. 4.9) are applied during the analysis and their valueoutcomes are respectively Good = 0.2, Good 0.4, Average 0.3, at the end we havea fuzzy set which is the combination of the fuzzy sets Good and Average but withmaximum degrees of truth of 0.6 and 0.3 respectively. The set of membershipfunctions in Fig. 4.10 is also defined by means of Eq. 4.10, nonetheless this timethe values are ↵ = 1/8, �2 = 0.16 and c = {0.0, 2.5, 5.0, 7.5, 10.0}.

Figure 4.10: Defuzzification Functions.


Figure 4.8: Fuzzification Functions.

audience such as business or marketing experts. For instance, rules can bedefined using basic concepts like: IF the fraction of provided FCs is many andthe fraction of missing FCs is few and the fraction of non-required FCs is alsofew, then the fitness of the SVN is Perfect. The main issue, however, is theimprecision that might arise due to assumptions such as predefined preferencesfor non-required FCs or inaccuracy when mapping numerical values onto words.Fuzzy Logic techniques are good candidates to deal with this kind of issues sincethey o↵er tolerance for imprecision whereas still achieving robustness, low costcomputation and a good rapport with reality [97].

In this way, at this step, we determine the amount of provided (P ), missing(M) and non-required (N) functional consequences for each SVN in terms ofthree linguistic labels: few, some and many. To this aim, we compute themembership degree of the fractions F

P

, FM

and FN

to the functions in Fig. 4.8(also known as Fuzzy Sets) [97]. The membership degree (also known as degreeof truth) of a fraction in a fuzzy set allows to determine how true is the fact thatthe analyzed fraction is few, some or many. The degree of truth of a fraction isdetermined by the following distribution function [97]:

Fuzzification(x) = µ(x) =1p2⇡�2

e�↵

(x�c)2

�

2 (4.6)

where x = {FP

, FM

, FN

}, �2 is the variance, ↵ is a parameter that deter-mines how width the fuzzy set is and c is the location of the fuzzy set, i.e.the highest point of the fuzzy set. For all the fuzzy sets ↵ = 2.0, �2 = 0.16.The values for c are 0.0, 0.5 and 1.0 for the fuzzy sets few, some and manyrespectively. The values for ↵, �2, c were chosen to cover the fraction valuesFP

, FM

and FN

within the range [0, 1] [97]. At the end, for each SVN we havedi↵erent degrees of truth for P , M and N . E.g. a SVN with a fraction valueFP

= 0.8 has the following degrees of truth for P : few (0.0), some (0.32) andmany (0.60), which means that P simultaneously belongs to the fuzzy sets someand many but with di↵erent degrees of truth. i.e. it is not only true (with a0.32 value) that the SVN provides some FCs but also true (with a 0.60 value)that the SVN provides many FCs as required by the customer.

4.1.2.2 Analysis

Inference rules Once we have computed the degrees of truth of P , M andNR for each SVN, we can then analyze how good they are by making use ofa set of inference rules that we designed following common sense criteria anddepicted in Fig. 4.9. E.g., a SVN with P = some, M = few and N = few is





Py

max

y

min

y ⇤A(y)P

y

max

y

min

A(y)(4.7)


4.1.2.4 Example





FP

=0.8 + 1.0

0.6 + 0.8 + 1.0=

1.8

2.4= 0.75 (4.8)

FM

=0.6

0.6 + 0.8 + 1.0=

0.6

2.4= 0.25 (4.9)

FN

=0.5

0.6 + 0.8 + 1.0 + 0.5=

0.5

2.9= 0.17 (4.10)



D(A) =1 ⇤ 0.17 + 2 ⇤ 0.70 + 3 ⇤ 0.70 + 4 ⇤ 0.45 + 5 ⇤ 0.70

0.17 + 0.70 + 0.70 + 0.45 + 0.70 + 0.45 + 0.80 + 0.80 + 0.45 + 0.45+

6 ⇤ 0.45 + 7 ⇤ 0.80 + 8 ⇤ 0.80 + 9 ⇤ 0.45 + 10 ⇤ 0.45

0.17 + 0.70 + 0.70 + 0.45 + 0.70 + 0.45 + 0.80 + 0.80 + 0.45 + 0.45= 5.68 (4.11)

4For simplicity we only depict the FCs o↵ered by each SVN omiting the overall structureTuesday 25 December 12

Defuzzy and Rank



SV N1 SV N2 SV N3



Score 8.69 3.66 5.68





4.1.3 Critique




SV N1 SV N2 SV N3



Score 8.69 3.66 5.68





4.1.3 Critique



an Average SVN. Since in the fuzzification step P,M and N may have di↵erentdegrees of truth for each linguistic label (membership function), more than onerule can apply when analyzing a given SVN. E.g., as explained before, P cansimultaneously be some and many, consequently the rules in which P is notonly some but also many have to be applied, the same holds for M and N .

1: IF P is many AND M is few AND NR is few THEN Perfect2: IF P is many AND M is few AND NR is some THEN Good3: IF P is many AND M is some AND NR is few THEN Good4: IF P is many AND M is some AND NR is some THEN Good5: IF P is some AND M is few AND NR is few THEN Average6: IF P is some AND M is few AND NR is some THEN Average7: IF P is some AND M is some AND NR is few THEN Poor8: IF P is some AND M is some AND NR is some THEN Poor... ... ... ... ... ... ... ... ... ... ... ... ... ... ...N: IF P is few AND M is many AND NR is many THEN Bad

Figure 4.9: Inference rules.

To apply the inference rules another observation is important. Since thedegrees of truth are expressed in values within the range [0, 1], the traditionalboolean AND operation does not work here, i.e. after applying the rule (1) asAND operationdepicted in Fig. 4.9, IF P is many AND M is few AND N is few the valuecannot be 0 or 1 but a value that reflects the degrees of truth of P , M and N .The AND operator in fuzzy logic is usually replaced by the min function [97],i.e. the output of the rule is the minimum degree of truth among P , M and N .E.g. for the rule (1), if P = 0.6, M = 0.4 and N = 0.6, the output is Perfectwith a degree of truth 0.4.

Aggregation Once the inference rules has been applied, their outputs must beaggregated into a new fuzzy set. Several aggregation methods can be used at thispoint, in our case we sum up the outcomes for each applied rule and combinethem into a fuzzy set given by the membership functions in Fig. 4.10. Theprocess is analogous to fill a set of silos based on the outcomes of the applied rules(the shape of the silos is given by the functions in Fig. 4.10). E.g. If the rules 2,Defuzzification functions4 and 6 (as depicted in Fig. 4.9) are applied during the analysis and their valueoutcomes are respectively Good = 0.2, Good 0.4, Average 0.3, at the end we havea fuzzy set which is the combination of the fuzzy sets Good and Average but withmaximum degrees of truth of 0.6 and 0.3 respectively. The set of membershipfunctions in Fig. 4.10 is also defined by means of Eq. 4.10, nonetheless this timethe values are ↵ = 1/8, �2 = 0.16 and c = {0.0, 2.5, 5.0, 7.5, 10.0}.

Figure 4.10: Defuzzification Functions.



Figure 4.1: The SVN Composition Framework.

edge but also the inferences that are needed to produce new knowledge as well as Knowledgethe interactions among them. Inferences carry out reasoning processes whereas Inferencesdynamic knowledge roles are the run-time inputs and outputs of inferences [82].Because supply and demand are always evolving, they are described as dynamicknowledge roles. As can be observed in Figure 4.1, due to the application of in-ferences, the rest of the knowledge components are also dynamically produced.Finally, whereas the propose, verify and modify subtasks must be perform bya computer (broker), the critique subtask must be perform by a human (cus- Brokertomer). In this way, the interaction between the customer and a broker provides Customeralso a dialogue in which the customer can influence the composition of SVNsby providing feedback. The elements of this dialogue-based interaction (i.e thepropose, verify, critique and modify subtasks) are explained in the followingparagraphs.

4.1.1 Propose

According to Chandrasekaran, given a design goal, the propose subtask gen-erates a solution [19]. In our case, the goal is to compose a SVN to cover agiven customer need. The next paragraphs provide a detailed explanation of theinferences related to this subtask, i.e. O↵ering, Laddering, Matching, Bundlingand B2B Linking.

4.1.1.1 O↵ering

Before composing any SVN, a description of the o↵erings of services suppliersand enablers is required. Such description must allow the creation of B2C and


Animation of all Steps in SVN Composition

• http://www.youtube.com/watch?v=tsrbv7-cN0A


http://www.youtube.com/watch?v=tsrbv7-cN0A

http://www.youtube.com/watch?v=tsrbv7-cN0A

5. The (Service) Value Web should be built on technologies that embody the same generative principles that lead to the success of the Web itself.

In other words, Internet-based SVN technologies should allow for unanticipated contribution of value (through services) to the Web by enabling anyone to share and trade their value objects, just like previous generations of the Web did for knowledge and social sharing.


Service Description Languages


Current Assumptions and Biases that impede our goal for on-the-fly adaptation

• we assume that the consumer explicitly articulates a need, and ranks proposed service bundles accordingly;

• we distinguish strongly between consumer and provider perspectives: strength and weakness;

• we should integrate sacrifices (such as our price models) better and investigate their effect on the size of the solution space (during matching), and ranking of bundles (during verification);

• current verification is based on case-to-case base, while it would be also interesting to consider the effect of global market trends on accuracy;

• we should further investigate the effect of business and technical constrains on the size of solution space (during matching)

• measure the value creation using the derived e3value business models

• case studies: global software development, ambient assisted living

• Collibra: changing enterprise governance from data management processes outwards


References

• De Leenheer, P.; Cardoso, J.; Pedrinaci (2013) Ontological Representation and Governance of Business Semantics in Compliant Service Networks. In Proc. of IESS 2013, LNBIP, to appear

• De Leenheer, P., Christiaens, S., Meersman, R. (2010) Business Semantics Management: a Case Study for Competency-centric HRM. In Journal of Computer for Industry 61(8): 760-775, Elsevier

• Limonad, L.; De Leenheer, P.; Linehan, M.; Hull, R.; Vaculín, R. (2012) Ontology of Dynamic Entities. In Proc. of ER 2012, LNCS, Springer, pp. 345-358

• Kinderen, de S.; De Leenheer et al. An ontology for needs-driven service bundling in a multi-supplier setting. In J. of Applied Ontology, 2013 (to appear)

• Normann, R.; Ramírez, R. (1993) From Value Chain to Value Constellation: Designing Interactive Strategy, Harvard Business Review 71:65-77

• Razo-Zapata, I.; De Leenheer, P.; Gordijn, J.; Akkermans, H. (2012) Fuzzy Verification of Service Value Networks. In Proc. of CAiSE 2012, Springer LNCS 7328, pp. 95-110

• Razo-Zapata, I.; Gordijn, J.; De Leenheer, P.; Akkermans, H. (2011) Dynamic Cluster-based Service Bundling: a Value-oriented Framework. In Proc. of IEEE CEC 2011, Luxembourg, Luxembourg, IEEE Press

• Razo-Zapata, I.; De Leenheer, P.; Gordijn, J.; Akkermans, H. (2012) Service Network Approaches. In Barros, A.; Oberle, D. Handbook of Service Description: USDL and its Methods, Springer, pp. 45-74

• Vargo and Lusch (2004) Evolving towards a new dominant logic for marketing. Journal of Marketing 68:1-19• Zittrain (2009) The Future of the Internet and How to Stop it. Yale University Press (http://futureoftheinternet.org/

download)


http://e3value.few.vu.nl/docs/bibtex/pdf/IEEE-CEC-2011-Razo-Zapata.pdf




http://futureoftheinternet.org/download




Acknowledgements

• Ivan Razo-Zapata

• Jaap Gordijn

• Davor Meersman


Date post:	19-Jan-2015
Category:	Technology
Upload:	pieter-de-leenheer
View:	1,497 times
Download:	0 times

World Wide Value Web: Automated Design of Real-World Multi-party Services on the Web

Technology