=Paper=
{{Paper
|id=Vol-447/paper-3
|storemode=property
|title=QUATRO Plus: Quality You Can Trust?
|pdfUrl=https://ceur-ws.org/Vol-447/paper1.pdf
|volume=Vol-447
|dblpUrl=https://dblp.org/rec/conf/esws/ArcherFKKKP09
}}
==QUATRO Plus: Quality You Can Trust?==
https://ceur-ws.org/Vol-447/paper1.pdf
QUATRO Plus: Quality You Can Trust?
Phil Archer1 , Elena Ferrari2, Vangelis Karkaletsis3, Stasinos Konstantopoulos3,
Antonis Koukourikos3, and Andrea Perego2
1
i-sieve technologies, Athens, Greece
phil@i-sieve.com
2
DICOM, Università degli Studi dell’Insubria, Varese, Italy
{elena.ferrari,andrea.perego}@uninsubria.it
3
IIT, NCSR ‘Demokritos’, Athens, Greece
{vangelis,konstant,kukurik}@iit.demokritos.gr
Abstract. The QUATRO Plus project, a follow on from the original
QUATRO Project, aims to balance the wisdom of the crowds with the
knowledge of the experts. It uses a mixture of authenticated data sources
and the opinions of end users expressed through social networking soft-
ware to build a dataset that is authoritative and trustworthy. The dataset
describes online resources using RDF with the upcoming W3C Recom-
mendation, POWDER, as the underlying transport and storage mecha-
nism. Data can be added to or queried through a variety of tools provided
by the project, some of which are described in detail in this paper.
1 Introduction
There is a great deal of opinion online that is an expression of ‘the wisdom of
the crowds,’ the classic example of this being Wikipedia. Much of this material
is recognised as being extremely good. Mistakes and occasional deliberate false-
hoods do occur however—something that Wikipedia itself recognises.4 On the
other hand, there are many experts whose opinions are expressed on the Web
and whose opinions are open to authentication: trustmark operators. Well known
examples of this include TRUSTe, Health on the Net (HON), and VeriSign.
The Quality Content And Description project (QUATRO Plus)5 seeks
to balance the wisdom of the crowds with the wisdom of the experts. QUATRO
Plus follows on directly from an earlier project and has been instrumental in the
development of the new W3C Protocol for Web Description Resources (POW-
DER). This paper presents the QUATRO Plus project with particular emphasis
on its creation of POWDER documents, exposure of their provenance and their
potential trustworthiness based on the reputation of the individual or organisa-
tion that created them.
The rest of this paper is organised as follows. Section 2 briefly introduces
the POWDER specification and especially its trust and authentication features.
Section 3 outlines the architecture of the QUATRO Plus platform and discusses
4
See http://en.wikipedia.org/wiki/Wisdom of the crowds.
5
Project website: http://www.quatro-project.org/
2 Phil Archer et al.
1
2 < a t t r i b u t i o n>
3
4 2007−12−14T 0 0 : 0 0 : 0 0
5
6
7 < i r i s e t>
8 < i n c l u d e h o s t s>exa m p l e . com
9
10 < d e s c r i p t o r s e t>
11 < d i s p l a y t e x t>E v e r y t h i n g on exa m p l e . com i s r e d and s q u a r e
12 < d i s p l a y i c o n s r c=” h t t p : // a u t h o r i t y . exa m p l e . o r g / i c o n s / r ed −s q u a r e . png ” />
13 < e x : c o l o u r r d f : r e s o u r c e=” h t t p : // r g b . o r g / vo ca b#r e d ” />
14 s q u a r e
15
16
17
Fig. 1. A simple POWDER document
how it serves metadata, whereas Sect. 4 focuses on one of its main metadata
creation components, the Quality Social Network. Finally, Sect. 5 discusses re-
lated work, whereas Sect. 6 concludes the paper and outlines future research
directions.
2 The POWDER Specification
The Protocol for Web Description Resources (POWDER) is a general purpose
content and quality labelling protocol based on Semantic Web technologies, de-
veloped by the W3C POWDER Working Group.6
POWDER was designed as a practical means of adding RDF to collections
of resources. To take a simple example, POWDER expresses statements such
as ‘everything on example.com is red and square; this statement was asserted
by the entity described at http://authority.example.org/company.rdf#me on 14th
December 2007.’ This statement is exemplified in Fig. 1.
At the core of POWDER is the Description Resource (DR) [1], an association
between:
– the DR scope, a set of resources [2]. Scope is expressed as an iriset, a series
of restrictions on the IRIs of resources that are in scope (lines 7-9 in Fig. 1).
– a formal description of all resources in scope (lines 13 & 14 in Fig. 1). This
is expressed by the descriptorset, which comprises RDF properties defined in
external vocabularies (as ex:colour and ex:shape at lines 13 & 14 in Fig. 1).
And
6
The POWDER Working Group was chartered in March 2007 and is due to complete
its work in June 2009. More details on the WG and the specification are available
at http://www.w3.org/2007/powder/.
QUATRO Plus: Quality You Can Trust? 3
1
2
3 < r d f : D e s c r i p t i o n r d f : a b o u t=” h t t p : //www . exa m p l e . com/ f o o / ”>
4 E v e r y t h i n g on exa m p l e . com i s r e d and s q u a r e
5
6 < e x : c o l o u r r d f : r e s o u r c e=” h t t p : // r g b . o r g / vo ca b#r e d ” />
7 s q u a r e
8
9
Fig. 2. The output of the describe (u, D) function, where D set includes as element
the POWDER document in Fig. 1 and u = http://www.example.com/foo/. Note that
properties text and logo map from elements displaytext and displayicon, respectively, in
Fig. 1.
– a textual and/or pictorial summary of the formal description. This is ex-
pressed as descriptorset members using the POWDER-defined displaytext and
displayicon elements (lines 11 & 12 in Fig. 1).
Furthermore, POWDER documents associate attribution blocks (lines 2–5 in
Fig. 1) with one or more DR blocks. Attribution blocks provide information
that can be used assess the trustworthiness of the DRs in a document, including
the labelling authority that created it, creation time, validity period, and so on.
Note that POWDER attribution blocks have the only purpose of denot-
ing the authorship of a given POWDER document, thus making users able to
verify its authenticity, and to decide its degree of trustworthiness. Such an ap-
proach is more specific with respet to the one of the Open Provenance Model
(OPM) [3], which aims at providing a detailed description of the creation, usage,
and derivation of a given artifact along its whole lifecycle. However, although
OPM addresses issues that are not requirements according to the POWDER
specification, such technology can be effectively applied to POWDER in order
to document the history of POWDER documents.
Finally, POWDER documents can be explicitly associated with the resources
they apply to by using a variety of methods. For this purpose, the describedby
relationship has been defined to be used in (X)HTML link elements, HTTP
Link headers [4], ATOM entries [5], and RDFa [6]. As illustrated in Sect. 2.1,
POWDER processors allow one to apply arbitrary POWDER documents to a
resource by specifying the IRI of a specific POWDER document, or by access-
ing a repository of POWDER documents. For a detailed description of all the
possible options, we refer the reader to Sect. 4 of [1].
2.1 Processing POWDER Documents
POWDER processors implement a describe(u, D) function that returns an
rdf:Description of resource u given a set D of POWDER documents (see Fig. 2
for an example).
4 Phil Archer et al.
1 @prefix owl : .
2 @prefix ex : .
3 @prefix rdfs : .
4 @prefix wd rs : .
5 @prefix xsd : .
7 <> a owl : O n t o l o g y ;
8 wd rs : i s s u e d ”2007−12−14T00 : 0 0 : 0 0 ” ;
9 wd rs : i s s u e d b y .
11 : i r i s e t 1 a owl : C l a s s ; owl : e q u i v a l e n t C l a s s [
12 a owl : C l a s s ;
13 owl : i n t e r s e c t i o n O f (
14 [ a : R e s t r i c t i o n ; owl : o n P r o p e r t y wd rs : m a t c h e s r e g e x ;
15 owl : h a s V a l u e
16 ” \:\/\/(([ˆ\/\?\#]∗)\@) ?([ˆ\:\/\?\#\@] +\ .) ? ( e x a m p l e \.com ) (:([ 0 −9] +) ) ?\/” ˆˆ x s d : s t r i n g ; ] ) ] .
17 : d e s c r i p t o r s e t 1 a owl : C l a s s ;
18 r d f s : s u b C l a s s O f [ a owl : C l a s s ;
19 owl : i n t e r s e c t i o n O f (
20 [ a owl : R e s t r i c t i o n ; owl : o n P r o p e r t y e x : c o l o r ; owl : h a s V a l u e ]
21 [ a owl : R e s t r i c t i o n ; owl : o n P r o p e r t y e x : s h a p e ; owl : h a s V a l u e ” s q u a r e ” ]
22 ) ];
23 wd rs : l o g o ;
24 wd rs : t e x t ” E v e r y t h i n g on e x a m p l e . com i s r e d and s q u a r e ” .
26 : iriset 1 rd fs : subClassOf : descriptorset 1 .
Fig. 3. The OWL/RDF transform of the POWDER document in Fig. 1 in N3 syntax
[9, 10]. Note how attribution is expressed as owl:Ontology annotations and POWDER-
defined descriptors as descriptorset annotations; POWDER-defined properties are in-
stances of owl:AnnotationProperty.
POWDER documents can, to a large degree, be processed entirely as XML
but are, in fact, transporting OWL/RDF graphs that can be accessed by per-
forming a transformation on the XML (see Fig. 3 for an example). In Semantic
POWDER (POWDER-S), DRs express an rdfs:subClassOf relation between the
class of resources that fall within scope and the class of resources that have the
given RDF description. The POWDER document as a whole is an owl:Ontology
instance containing a number of such assertions, and the attribution block ex-
presses owl:AnnotationProperty triples with the owl:Ontology instance as subject.
This achieves the same goal as Named Graphs [7] as used for example by Bizer
and Cyganiak in [8]: the explicit declaration of provenance information, but
within existing RDF/OWL specifications.
A semantic extension bridges abstract (but named) resources and the string
representations of their IRIs. This extension reflects the basic premise of POW-
DER, absent from RDF semantics, that the structure of a resource’s IRI is a
property of the corresponding resource, upon which inference can be drawn;
this allows POWDER documents to assert propositions about ‘all resources on
example.com.’
POWDER-S offers processors the option of using standard semantic inference
and querying tools to process POWDER/XML documents. The process of trans-
forming POWDER/XML documents into POWDER-S documents is detailed in
the POWDER specification [11] and also implemented as an XSLT script [12]
associated with the POWDER namespace via GRDDL [13]. Together with the
fact that POWDER processors respond with RDF descriptions even when only
processing XML, this transformation makes POWDER tools fully compatible
with other Semantic Web tools and technologies.
QUATRO Plus: Quality You Can Trust? 5
1
2
3 The E x e m p l a r y D e s c r i p t i o n A u t h o r i t y
4 < f o a f : n i c k>EDA
5
6
Fig. 4. A simple FOAF Agent
2.2 Attribution and Authentication
Trust is a human quality and is rarely deterministic in an absolute sense; whether
trust is conferred on any data is always a balance between the likelihood of it
being true and the consequence of it being false. If a DR says that a particular
cake recipe is really good but following it produces an inedible lump, little harm
is done. If a DR states that some medical advice is peer reviewed and can be
used as the basis for diagnosis and treatment, then the consequences of false-
hood are clearly much more serious and a more robust authentication method
is appropriate.
Although the POWDER specification is very detailed on the form and mean-
ing of POWDER documents and the output of conforming processors, it is de-
liberately non-prescriptive about trust and authentication methods; any method
can be defined by a DR publisher that is appropriate for the particular context
(the choice of method made by the publisher may itself be a factor in a user’s
assessment of trust!). What POWDER does offer is a detailed specification of
how to endow DRs with attribution and authentication credentials, the raw in-
formation upon which a variety of trust methodologies can operate.
The issuedby attribution element is mandatory for all POWDER documents
and provides the IRI that identifies a description of the entity that created the
document and that is therefore responsible for the claims made in the DR(s) it
contains. The strong recommendation is that such descriptions are provided as
instances of the Agent class in either of the widely used FOAF [14] or Dublin
Core [15] vocabularies.7
An RDF-aware system can use the IRI to retrieve the properties of the creator
(see Fig. 4 for an example) and use them to decide if the document is trustwor-
thy. Besides whatever properties are defined by the vocabulary used to describe
the creator, POWDER also defines the authenticate property that points to a re-
source that gives information on how to authenticate any POWDER documents
that were created by this Agent. Such a resource may be a machine readable
document, such as a WSDL8 file, that can be read and acted upon automati-
cally. However, it is equally valid for the resource to be a simple text document
that needs to be read by a person who then implements a suitable system to
take advantage of the available authentication mechanism. This approach is in
7
foaf:Agent is defined at http://xmlns.com/foaf/spec/#term Agent. dcterms:Agent is
defined at http://dublincore.org/documents/dcmi-terms/#classes-Agent.
8
Web Services Description Language. See [16].
6 Phil Archer et al.
Fig. 5. The architecture of the QUATRO Plus platform
tune with the POWDER paradigm as the aim is to provide strong assurance
of the identity of the creator of a given description. Such a data source will al-
most always require a human stage in assessing trustworthiness. Furthermore,
as already noted, trust is very context-sensitive so that prescribing a single au-
thentication mechanism would have been inappropriate. Section 3 discusses the
particular authentication system used by QUATRO Plus.
There are other features of POWDER designed to facilitate trust in the
data. These include certification, that is, where a DR has a scope of exactly one
resource and a descriptor set that includes either or both of two descriptors:
certified, which has a range of xsd:boolean, and sha1sum, the value of which is a
SHA-1 checksum of the described resource. Using these, it’s possible for a DR
to certify the veracity of another. Another feature designed to increase the trust
that can be placed in a POWDER document is the supportedby property. This
can link to any form of data published by a third party that agrees with the
assertions in the DR. This data may be in any form, the assumption being that
an application will be built with a specific data source in mind. As an example,
a POWDER document may assert that a website is suitable for children. Sup-
porting evidence might come from a commercial content classification service.
3 Working with Trustmarks: The QUATRO Proxy
The full suite of QUATRO Plus software (whose architecture is depicted in
Fig. 5) includes a search result annotation tool, LADI+ and a browser exten-
sion, ViQ+, both of which recognise the presence of links to DRs and provide
authentication information to end users. A further tool, the Label Management
QUATRO Plus: Quality You Can Trust? 7
Environment (LME), was developed under a related project, MedIEQ.9 In this
paper, we focus on two components of QUATRO Plus suite: The QUATRO
Proxy and the Quality Social Network (QSN).
The QUATRO Proxy is a set of Web services provided by multiple servers
that, among other functionalities, is able to describe resources by locating, pro-
cessing and authenticating POWDER documents. Given the IRI of a Web re-
source, QUATRO Proxy locates relevant POWDER documents, authenticates
them and uses them to compose a description of the resource. The description
is returned to the client along with details of the provenance and authenticity
of the data.
As mentioned earlier in this section, relevant POWDER documents are lo-
cated by implementing a cascade of discovery techniques. In increasing order of
efficiency these are: by checking the website itself and following links to POW-
DER documents (using the ‘describedby’ relationship type); by following a link
to a POWDER document provided in the request to the proxy made by the
client; and by looking the site up directly in the trustmark operators’ databases
to which it has access. Except in the latter case, once the relevant POWDER
documents have been identified, they are authenticated by looking them up in
the relevant trustmark operator’s database—an automated version of the ‘click
to verify’ model operated by all online trustmarks.
The security of the transactions and the authenticity of the server and the
databases to which the proxy connects are guaranteed via message signing and
SSL tunnelling. Equally importantly, trustmark operators can be confident that
their data is secure.
4 Creating Descriptions: The Quality Social Network
Recent years have seen an increasing diffusion of Web-based social networks (WB-
SNs) giving their members the ability to specify and share metadata concerning
online resources. Examples of such WBSNs are del.icio.us (http://del.icio.us),
RawSugar (http://rawsugar.com), Flickr (http://flickr.com), and Last.fm
(http://last.fm), which support the so-called social or collaborative tagging.
Although, so far, users’ tags and content providers’ content labels have evolved
separately, we think it desirable that a convergence is established between them.
The ideas underlying collaborative tagging may contribute in addressing some
of the main open issues which prevented the success of content labelling. In
particular, existing content labels describe a very limited subset of the Web,
thus making impossible to widely exploit their advantages in term of informa-
tion filtering and discovery. Moreover, content labels must be consistent with
the resources they describe. However, since Web resources may change very fre-
quently, this requires an effort which is not acceptable to content providers. In
such a scenario, collaborative tagging can make any end user a possible label
author, thus increasing not only the number of labelled resources, but also the
9
MedIEQ focused on the labelling of medical resources. Please see http://medieq.org/
for more details.
8 Phil Archer et al.
number of labels associated with the same resource. Thanks to this, it would be
simpler to verify whether a label actually describes the associated resource, and
it would be also possible to assess the trustworthiness of a given description by
statistically analysing the whole set of labels associated with a resource.
Based on these considerations, the QUATRO Plus platform includes a so-
cial networking service, referred to as Quality Social Network (QSN), giving its
members the ability to specify user-defined POWDER documents (referred to
as labels) and ratings. As any POWDER document, besides containing a set of
descriptors (the DR descriptorset component), a label states who has created it
(POWDER attribution block), the set of resources it applies to (the DR iriset),
and its validity period (corresponding to the validfrom and validuntil attribution
elements available in POWDER). By contrast, ratings are used by WBSN mem-
bers to endorse part or all the descriptors in an existing label, or to express their
disagreement about the claims it contains. Three types of ratings are supported:
positive (agreement), negative (disagreement), and neutral (‘I don’t know’). The
trustworthiness of the descriptors contained in a label is then statistically de-
termined based on the number of positive/negative/neutral ratings associated
with them. The results of such an evaluation are made available to the other
components of the QUATRO Plus architecture, and to their end users.
QUATRO Plus plans to use these features as a basis to enforce personalised
access to Web resources through the support for user preferences. User prefer-
ences give end users the ability to state which action should be performed by
their user agent upon detection of resources carrying labels with given descrip-
tors and trustworthiness. For instance, an end user may ask to be notified when
a given resource does or does not satisfy a given set of requirements concerning
medical resources, or decide that the user agent must block access to a resource if
more than 20% of the existing labels state that it contains pornographic content.
It is important to note that, although a publicly accessible QSN will be
set up,10 the QSN is a service which can be installed and run by any institu-
tion/organisation that wishes to set up a social network supporting collaborative
labelling and rating. For this reason, the QSN gives system administrators the
ability to configure the service depending on the requirements of the institu-
tion/organisation running it.
In the following section, we provide an overview of the main features sup-
ported by the QSN. The work reported here is a partial implementation of the
multi-layer personalisation framework we have proposed in [17], which provides
the foundations of the approach adopted in QUATRO Plus for supporting Web
access personalisation through the use of Semantic Web technologies and social
networking.
4.1 Labels and Ratings Evaluation and Aggregation
The main purpose of supporting collaborative labelling and rating of Web re-
sources, is to statistically analyse the labels and ratings specified by WBSN
10
A prototype version of the QSN, accessible only by invited members, is currently
available at http://dawsec.dicom.uninsubria.it/qui/
QUATRO Plus: Quality You Can Trust? 9
members and identify the most objective descriptions of a resource which may
then be used for content/service personalisation. In order to achieve this, the
labels associated with a resource are aggregated, and for each descriptor they
contain, a trust value is computed.
There are several different formulae which may be used for trust computation,
each addressing the requirements of given application domains, granting different
degrees of accuracy, and having different computational costs. For these reasons,
we plan to support a set of trust computation algorithms, among which QSN
administrators can choose the one which is most suitable to their purposes, also
taking into account the trade-off between efficiency and accuracy.
Basically, we plan to support two main trust computation solutions. Accord-
ing to the basic option, all QSN members as equally trustworthy. Consequently,
the trustworthiness of a label is determined only by the more or less wide consen-
sus on the claims it contains. Such consensus dependes on the possible existence
of multiple occurrences of the same label, specified by different authors, and
on the positive, negative, and neutral ratings associated it. All this informa-
tion is then used to compute a trust score for a given label. There exist several
algorithms proposed for reputation systems [18], such as Eigentrust [19] and
PeerTrust [20], which might provide a suitable solution.
Such an approach can be enhanced by associating each QSN member with a
reputation score, which can then be used to assign a specific weight to his/her
labels and ratings. This is an issue that has been thoroughly investigated by
recommender systems [21], among which there exist examples of online commu-
nities, such as MovieLens [22] and MyWOT (http://mywot.com).
Reputation can determine the trustworthiness of a given QSN member for
all the users in a community. However, it may be often the case that a given
member A considers more trustworthy the opinions of member B than the ones
of member C, even though the reputation of C is higher than the one of B. Rec-
ommender systems have addressed such issue by computing a similarity measure
between user profiles. More recently, a different approach has been adopted by
a few WBSNs. For instance, FilmTrust [23] and Moleskiing [24] propose al-
gorithms which make use of trust relationships explicitly established between
WBSN members to predict the degree of trust existing even between members
not directly connected. Finally, in an earlier paper [17], we have explored also
the possibility of using trust policies, thanks to which a given QSN member can
denote the set of users he/she considers as trustworthy based on either their
identity or characteristics.
4.2 Privacy Issues
Labels and ratings are not sensitive by themselves, but they may convey sensitive
information about end users’ opinions and tastes, which can be misused. On
the other hand, labels’ and ratings’ accountability is fundamental in order to
make their authors responsible of what they claim. QUATRO Plus raises other
issues surrounding privacy protection and accountability. If user-defined labels
and ratings are public (i.e., accessible by everyone, even by people who are not
10 Phil Archer et al.
members of the QSN), the whole Web community can benefit of them. However,
this may result in an inappropriate disclosure of private information.
Based on this, when specifying a label or a rating, QSN members are given
various protection options:
– the label/rating is private—it can be seen only by the QSN member who
created it;
– the label/rating is visible only to the direct contacts of the QSN member
who created it;
– the label/rating is visible only to the group members of the QSN member
who created it;
– the label/rating is visible to any QSN member;
– the label/rating is public—also non QSN members can access it.
When aggregated, labels and ratings no longer carry information concerning
their authors, so they do not explicitly disclose private information. Nonetheless,
QSN administrators are given the possibility of setting which labels and ratings
must be taken into account (i.e. any label, only public labels, etc.) and whether
the aggregate view is a service available only to QSN members or also to non
QSN members.
4.3 Web Access Personalisation
As mentioned earlier, we plan to use the aggregate view of labels and ratings
as a basis to enforce Web access personalisation. In order to achieve this, QSN
members will be given the ability of specifying user preferences denoting the
action to be performed when detecting a resource associated with a given set
of descriptors, having a given trustworthiness. The supported actions are block
and notify. In case of a block action, access to the requested resource is denied,
and the user agent returns the notification message. In case of a notify action,
access is granted, and the user agent returns the notification message. In both
cases, the end users will be given the possibility of verifying why a given action
has been performed by displaying (a) the content of the labels associated with
the accessed resource, (b) the corresponding aggregate view, and (c) the user
preferences which have been satisfied. Note however that, as far as single labels
are concerned, the end user will be able to display only the ones he/she is au-
thorised to see, based on the disclosure policies specified by their authors (see
Sect. 4.2).
The language and the tools to be used for user preference specification and
enforcement are currently one of the issues we are investigating. A possible option
is to use a rule language, such as N3 Rules [9, 10], and a reasoner (e.g., Cwm11 ).
However, languages such as N3 Rules have an expressivity which is far higher
than the one required by our user preferences, and they are not necessarily the
best solution, in terms of efficiency. For this reason, an alternative option is to
develop a user preference language which could be efficiently processed by a piece
of software designed specifically for this purpose.
11
See http://www.w3.org/2000/10/swap/doc/cwm.html.
QUATRO Plus: Quality You Can Trust? 11
Fig. 6. QSN architecture
4.4 QSN Architecture
The QSN architecture, depicted by Fig. 6, consists of a set of repositories (the
member base, which stores members’ data and relationships, the user preferences
repository, the labels and ratings repositories), storing all QSN data, and of
a set of modules in charge of carrying out the supported services (members’
registration and authentication, user preferences evaluation, labels and ratings
retrieval and aggregation).
In particular, the labels and ratings retrieval module will be in charge of
returning the set of labels and ratings satisfying a query, which may concern all
the labels and ratings associated with a resource, all the labels and ratings spec-
ified by a given member, etc. Such a service will be used by the label aggregator,
which is in charge of aggregating and evaluating labels. In turn, the service pro-
vided by the label aggregator will be used by the user preferences evaluation
module, which will verify whether a given resource satisfies one or more of the
user preferences specified by the end user, and then returns the action to be
performed by the user agent (ViQ+).
All QSN services will be accessed by the other components of the QUATRO
Plus platform through SOAP interfaces. For example, LADI+ will use the re-
sults of the statistical analyses of labels and ratings to associate trust values of
any available descriptors with search results whilst QUAPRO+ can return the
aggregated data regarding the users’ ratings on labels applied to a given resource
when returning descriptions of it.
The QSN user interface (QUI) will allow end users to register/authenticate on
the network, manage their personal data and relationships, specify user-defined
labels and ratings, and browse them as a full list or as an aggregate view. As far
12 Phil Archer et al.
as the specification of user-defined labels and ratings is concerned, this task will
be performed either by an annotation tool, developed specifically for the QSN,
or by the Label Management Environment (LME), which end users will be able
to transparently access, directly from the QUI. Finally, the QUI can be accessed
by QSN members either from ViQ+ or directly through the browser.
5 Related Work
The idea of annotating Web content is not new. It has been proposed as a means
of allowing content consumers to easily identify the content that best fits their
needs, by means of using established vocabularies to describe the latter.
Such annotations might relate a variety of content aspects, such as language
or languages the content is available in, thematic categorisation, suitability for
certain devices, suitability for certain age groups, etc. Web content annotation
has many analogies that precede the Web and Web content, such as book and
film ratings or food and consumer goods labelling for health and safety reasons.
5.1 Visible Trustmarks
Trustmarks of varying kinds have been in existence for as long as people have
communicated. The established online form is a logo on a Web page that, when
clicked, returns a certificate delivered from the trustmark operator’s system con-
firming that the trustmark is genuine. This has a number of problems however:
– The user must manually follow a link, check that the certificate is genuine
and still valid, and so on.
– The website must permit the content provider to add the logo linking back
to trustmark operator, which is not always the case.
– Trustmarks are, typically, only visible on a specific page but refer to the
whole site, so they can be missed when following search results or external
links.
– The trustmarks are invisible to search engines and linked data systems.
These are precisely the problems that the QUATRO Plus project sets out
to address. The ‘click to verify’ action, and all the authentication, can be done
automatically. The result can then be presented to the user through the browser
chrome (or some other method independent of the Web page) while the data is
also made available in a machine-readable format.
5.2 Semantic Web Approaches to Trust
A great deal of work has been done to enable the assessment of the quality, rele-
vance and trustworthiness of online resources using Semantic Web technologies.
Mention has already been made of the WIQA Policy Framework [8]. In that
paper, Bizer and Cyganiak identify three metrics for assessing quality: content-
based (i.e., analyzing the information content itself), context-based (the available
QUATRO Plus: Quality You Can Trust? 13
metadata) and ratings-based. The QUATRO Plus / POWDER approach is a po-
tential source of both context and ratings-based data. A DR may provide any
kind of description of one or more resources, including ratings and details such
as who created the resource(s) and when, whether a document was created by
a company with a commercial interest and so on. Crucially, such descriptions
always carry metadata about themselves so that trustworthiness of the context
and ratings-based metrics can themselves be assessed.
In [25], Heath and Motta found that in their particular field of study (travel
recommendations), experience and affinity were the key terms for building a sys-
tem that personalised search results. Since recommendations about travel des-
tinations is inherently subjective, the wisdom of the crowds (mediated through
semantic relationships) is entirely appropriate. QUATRO Plus has the poten-
tial to contribute additional information to the system such as awards given to
restaurants by professional food critics, clean beach awards given by environ-
mental health experts and so on.
In both these cases, QUATRO Plus offers reliable data of known provenance
that complements the approach taken.
5.3 Platform for Internet Content Selection (PICS)
One of the W3C’s earliest efforts at content labelling was the Platform for In-
ternet Content Selection (PICS) [26], addressing some of the issues with using
HTTP headers and HTML meta elements to annotate Web content.
PICS specified a format for bundling content annotations along with scop-
ing and attribution information. Scoping amounted to coupling annotations
with a URL prefix, including the ability to specify how more specific prefixes
override more generic ones. Although this mechanism allowed for the mass-
annotation of websites, it was far too restrictive as URL prefixes cannot cap-
ture very commonly encountered groupings. Imagine, for example, a site where
all pages under http://www.red.example.com/ share features that pages under
http://www.green.example.com/ do not.
An important feature of PICS was the rating system, a unique identifier for
a vocabulary of rating terms. A client application still had to know how to
interpret the terms in the vocabulary, but the rating-system mechanism avoided
the danger of misinterpreted annotations.
Finally, clients requested PICS documents through a specific HTTP request-
for-labels made to either the content provider’s Web server or third-party services
known as label bureaux. This mechanism allowed for independent annotation
services, so that content consumers can decide which annotation provider to
trust. Client-side trust policies could also refer further attribution information
(such as validity dates) included in PICS documents. The major drawbacks of
the HTTP request-for-labels mechanism, however, was the requirement for an
extension to the HTTP protocol to handle the PICS-specific HTTP request for
labels, and that it did not provide for discovery mechanisms so that user agents
had to know in advance which label bureaux to ask.
14 Phil Archer et al.
6 Conclusions and Future Work
The linear model of content annotation that posits a label at one end of a chain
and a user agent/gateway at the other was developed in the 1990s and promoted
particularly by child protection organisations. Experience shows that such a sys-
tem, for all its political attractions, is very unlikely to gain widespread adoption.
In contrast, social networking and shared tags have been phenomenally success-
ful. The QUATRO Plus project and its forerunner (simply called QUATRO)
have tried to marry these two facts together as the fundamental point remains:
some people really do know more than others about given subjects. In some
situations, experience and knowledge must be allowed to counterbalance pub-
lic perceptions that may or may not be well-informed. By helping to develop
a technical platform to support its aims, one that fits in with much broader
efforts to develop the Semantic Web; by providing a suitable software infrastruc-
ture and by encouraging the publication of interoperable data, complete with an
assessment of trust, we believe that balance can be realised.
Any system needs ongoing technical development and maintenance, together
with an infrastructure to support it. Alongside that is the need for dissemination
and community building, The latter drives the former and it is the building of
that community to which the QUATRO Plus partners are now turning their
attention.
Acknowledgements QUATRO Plus is co-funded by the European Union’s
Safer Internet Programme (SIP-2006-211001) and includes partners from indus-
try, academia and the non-profit sector.
References
1. Archer, P., Smith, K., Perego, A.: Protocol for Web description re-
sources (POWDER): Description Resources. W3C Working Draft (April 2009)
http://www.w3.org/TR/powder-dr/.
2. Archer, P., Perego, A., Smith, K.: Protocol for Web description resources
(POWDER): Grouping of Resources. W3C Working Draft (April 2009)
http://www.w3.org/TR/powder-grouping/.
3. Plale, B., Miles, S., Goble, C., Missier, P., Barga, R., Simmhan, Y., Futrelle, J.,
McGrath, R., Myers, J., Paulson, P., Bowers, S., Ludaescher, B., Kwasnikowska, N.,
den Bussche, J.V., Ellkvist, T., Freire, J., Groth, P.: The Open Provenance Model
(v1.01). Technical Report 16148, University of Southampton, School of Electronics
and Computer Science (July 2008) http://eprints.ecs.soton.ac.uk/16148/.
4. Nottingham, M.: Web linking. IETF Internet-Draft, Internet Engineering
Task Force (April 2009) http://www.ietf.org/internet-drafts/draft-nottingham-
http-link-header-05.txt.
5. Nottingham, M., Sayre, R.: The Atom syndication format. IETF RFC 4287, Inter-
net Engineering Task Force (December 2005) http://www.ietf.org/rfc/rfc4287.txt.
6. Adida, B., Birbeck, M.: RDFa primer: Bridging the human and data Webs. W3C
Working Group Note, W3C (October 2008)
QUATRO Plus: Quality You Can Trust? 15
7. Carroll, J.J., Bizer, C., Hayes, P.J., Stickler, P.: Named graphs. Journal of Web
Semantics 3(4) (December 2005) 247–267
8. Bizer, C., Cyganiak, R.: Quality-driven information filtering using the WIQA
policy framework. Journal of Web Semantics 7(1) (January 2009) 1–10
9. Berners-Lee, T., Connolly, D.: Notation3 (N3): A readable RDF syntax. W3C
Team Submission (January 2008) http://www.w3.org/TeamSubmission/n3/.
10. Berners-Lee, T., Connolly, D., Kagal, L., Scharf, Y., Hendler, J.: N3Logic: A logical
framework for the World Wide Web. Theory and Practice of Logic Programming
8(3) (May 2008) 249–269
11. Konstantopoulos, S., Archer, P.: Protocol for Web description re-
sources (POWDER): Formal semantics. W3C Working Draft (April 2009)
http://www.w3.org/TR/powder-formal/.
12. Clark, J.: XSL Transformations (XSLT) — Version 1.0. W3C Recommendation,
W3C (November 1999)
13. Connolly, D.: Gleaning resource descriptions from dialects of languages (GRDDL).
W3C Recommendation, W3C (September 2007) http://www.w3.org/TR/grddl/.
14. Brickley, D., Miller, L.: FOAF Vocabulary Specification v0.91. Namespace Docu-
ment (November 2007)
15. DCMI Usage Board: DCMI metadata terms. DCMI Recommendation, Dublin
Core Metadata Initiative (January 2008)
16. Booth, D., Liu, C.K.: Web Services Description Language (WSDL) —
Version 2.0 Part 0: Primer. W3C Recommendation, W3C (June 2007)
http://www.w3.org/TR/wsdl20-primer/.
17. Carminati, B., Ferrari, E., Perego, A.: Combining social networks and Semantic
Web technologies for personalizing Web access. In: CollaborateCom 2008. (Novem-
ber 2008) http://www.dicom.uninsubria.it/dawsec/pubs/collaboratecom2008.pdf.
18. Jøsang, A., Ismail, R., Boyd, C.: A survey of trust and reputation systems for
online service provision. Decision Support Systems 43(2) (2007) 618–644
19. Kamvar, S.D., Schlosser, M.T., Garcia-Molina, H.: The Eigentrust algorithm for
reputation management in P2P networks. In: WWW 2003. (2003) 640–651
20. Xiong, L., Liu, L.: PeerTrust: Supporting reputation-based trust for peer-to-peer
electronic communities. IEEE Transactions on Knowledge and Data Engineering
16(7) (2004) 843–857
21. Adomavicius, G., Tuzhilin, A.: Toward the next generation of recommender sys-
tems: A survey of the state-of-the-art and possible extensions. IEEE Transactions
on Knowledge & Data Engineering 17(6) (June 2005) 734–749
22. Sen, S., Lam, S.K., Rashid, A.M., Cosley, D., Frankowski, D., Osterhouse, J.,
Harper, F.M., Riedl, J.: tagging, communities, vocabulary, evolution. In: CSCW
2006. (2006) 181–190
23. Golbeck, J.A.: Generating predictive movie recommendations from trust in social
networks. In: iTrust 2006. (2006) 93–104
24. Avesani, P., Massa, P., Tiella, R.: A trust-enhanced recommender system applica-
tion: Moleskiing. In: ACM SAC 2005. (2005) 1589–1593
25. Heath, T., Motta, E.: Personalizing relevance on the Semantic Web through trusted
recommendations from a social network. In: ESWC’06 SWP Workshop. (2006)
26. Resnick, P., Miller, J.: PICS: Internet access controls without censorship. Com-
munications of the ACM 39(10) (1996) 87–93