PERSONALISED RESOURCE DISCOVERY SEARCHING OVER

MULTIPLE REPOSITORY TYPES

Using user and information provider profiling

Boris Rousseau, Parisch Browne, Paul Malone, Mícheál ÓFoghlú

Telecommunications Software Systems Group (TSSG), Waterford Institute of Technology, Waterford, IRELAND

Paul Foster, Venura Mendis

BTexact Technologies, Orion Building, Adastral Park, Martlesham Heath, Suffolk, IP5 3RE, United Kingdom

Keywords: E-Learning, personalisation, profiling, user profile, information provider profile, resource discovery,

multiple repository types.

Abstract: The success of the Information Society, with the overabundance of online multimedia information, has

become an obstacle for users to discover pertinent resources. For those users, the key is the refinement of

resource discovery as the choice and complexity of available online content continues to grow. The work

presented in this paper will address this issue by representing complex extensible user and information

provider profiles and content metadata using XML and the provision of a middle canonical language to aid

in learner-to-content matching, independent of the underlying metadata format. This approach can provide a

federated search solution leading to personalise resource discovery based on user requirements and

preferences, seamlessly searching over multiple repository types. The novelty of the work includes the

complex extensible user profiles, information provider profiles, the canonical language and the federated

search strategy. Although, the work presented is focused on E-Learning, the general ideas could be applied

to any resource discovery or information retrieval system.

1 INTRODUCTION

Most current E-Learning resource discovery systems

(GESTALT, 1998) employ relatively simple

algorithms when searching for content. The more

complex of these resource discovery services (RDS)

may use a thesaurus to find similar words and build

a set of related pages. This allows the user to be

provided with a ‘Find Similar Pages’ option.

Although early research in the area of Artificial

Intelligence (Spark, 1978) provided some solutions,

a successful Information Retrieval system should

take into account the identity of the end user

performing the search. Such identity incorporates

their needs, what type of content they prefer or

indeed whether they have available resources to

access the content found. Future successful RDSs

must consider these factors when executing search

queries. This can be best achieved through the use of

metadata. Metadata (Day, 2000) is traditionally

defined as information on the resources, not only to

indicate the nature of the content, but also to provide

information about the user as well as the provider of

the content.

In this work, models for describing this metadata

are explored and new ones developed where

necessary. This paper will also demonstrate how this

metadata can be used to provide personalised

resource discovery, thus facilitating the user’s

requirement for pertinent information, and the

provider’s requirement for maintaining multiple,

diverse, repository types, while at the same time

facilitating the business model that can support this.

The Information Society Technologies (IST)

project GUARDIANS (GUARDIANS, 2000) has

undertaken this approach and this paper describes a

large part of the work. This paper will present and

address the issues faced during the project regarding

Rousseau B., Browne P., Malone P., ÓFoghlú M., Foster P. and Mendis V. (2004).

PERSONALISED RESOURCE DISCOVERY SEARCHING OVER MULTIPLE REPOSITORY TYPES - Using user and information provider proﬁling.

In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 35-43

DOI: 10.5220/0002649800350043

 SciTePress

user and information provider profiling as well as

the search strategies to enable customisable resource

discovery. The first section will cover the user and

information provider profiling, in addition to content

metadata extensions for preferences weighting

facilitating the ranking of results found. The main

section will present the work carried out on the

canonical language and the federated search strategy

handling multiple repository types to provide

personalised resource discovery.

2 PROFILING

2.1 User Profile

With the growth of e-commerce and the Internet, the

gathering of information on preferences, activities

and characteristics of clients and customers has

become quite important, as this information can

facilitate personalised content delivery. The idea of

user profiling can be summarised as “one that

renders the user with an experience that is tailored to

his/her current situation” (Suryanarayana, Hjelm

2002). The issue of user profiling is quite complex

as such profile should be as generic as possible with

a compromise between storing too little and too

much information. A starting point is to use XML

(eXtensible Markup Language, 2000), in order to be

exportable to other system components and to

incorporate information about the user’s preferred

language and the technological platform for

communication.

As described in (Rousseau et al., 2003),

investigation of the existing profiling standards

revealed that while each of the standards

investigated had its own merits it was felt that none

fully addressed the profiling needs of the project.

Specifications researched include the IMS Learner

Information Package (IMS LIP, 2001) specification,

the IEEE Public and Private Information (IEEE

PAPI, 2002) Learner standard, and the TV Anytime

(TV Anytime, 1999) and vCard (vCard, 1996)

Specifications. Based on this research, a complex,

extensible, generic user profile model is developed

mainly based on the IMS LIP specification, the

Generic User Profile (GUP). The profile contains

information on the user stored in ‘sections’ as shown

in Table 1.

Table 1: Sections within the GUARDIANS Generic User

Profile

GUP Section Description

Accessibility User platform, applications,

disabilities, languages etc.

Affiliation Organisations with which the

learner has associations or

membership.

Usage History What the learner has previously

searched and/or accessed.

Relationship Relationships between GUP

elements.

Interest Interests of the learner, classified

by domain.

Contact Details The learner’s contact details.

Qualifications Formal qualifications that the

learner has previously acquired.

Goals Learning objectives of the

learner.

Although the IMS LIP specification is quite

complete, it was felt that extensions where needed to

facilitate the search process.

A first extension is the usage history building on

the TV Anytime forum specification. Such extension

allows to record previous searches information and

to reuse it in future ones. A usage history instance

fragment follows:

<InstanceId>history_01</InstanceId>

<defaulttype>Select</defaulttype>

</item>

</action>

</actionhistory>

</history>

The ‘action’ subsection depicts whether the user

has selected some content (itemused=”Content”) or

an INSP (itemused=”INSP”). In the case above, if

the user decides to do his next search using

“astronomy” as a keyword, “NASA” related content

is more likely to come up on his search results.

A second extension allows the user to further

refine his/her preferences indicating greater or lesser

importance with weightings. These weightings help

at the query-building phase to refine further the

search and at the rating of results found stages. A

weighted fragment follows:

ICEIS 2004 - HUMAN-COMPUTER INTERACTION

With this in place, the user gets resources

preferably in Italian. English content is also suitable

but with a lesser importance.

2.2 Information Service Provider

Profile

In the standards development community, much of

the profiling work seems excessively devoted to user

profiles and how they might be used for resource

discovery personalisation and communication with a

single Service Provider. However, in GUARDIANS

this process is seen as only half of the story, the user

should choose which service provider he/she wants

to enrol with or use. This section details the

requirements for storing information on service

providers. It will determine what information must

be stored in order to adequately describe a service

provider and its content and how it may be

represented.

The Information Service Provider (INSP) Profile

data-model can be used to store information on a

service provider on a per-repository basis. It stores

both domain specific metadata and generic metadata

and defines an aggregated view of the information

stored in its repository. As the INSP Profile is

essentially an aggregation or summary of the content

and services hosted or offered by the INSP, one of

its functions must be to act as a collection level

description such as those defined by Research

Support Library Programme (RSLP, 1999) and

UKOLN (UKOLN). In addition it must also function

as a searchable resource and be able to operate

across multiple domains.

IMS Learning Resource Metadata Information

Model Version 1.2 (IMS, 2001) is a specification

derived from Version 3.5 Learning Object Metadata

Scheme working document of the IEEE LTSC LOM

Working Group (IEEE LTSC LOM, 2002). The

differences lie in the redefinition of some of the

elements, some of which have a new type and/or a

new description and/or a new multiplicity. Although

both specifications are suitable for the INSP profile,

IMS seems to have scope according to up-to-date

work coming from IMS and IEEE.

The key information held in the INSP profile is

shown in Table 2.

Table 2: Sections within the GUARDIANS INSP profile

INSP Section Description

Service Name The common name of

the service.

Description Description of the

service.

Contact Details Information service

contact details.

Metametadata Detail about the

metadata contained in

the profile.

Legal Constraints Legal Issues surrounding

the use of the content

held by the INSP.

Payment information Details of payment

methods accepted by the

INSP.

Security Information Level of security

required to access the

various elements in this

profile.

Relationships Relationships between

instance metadata stored

locally or in other INSP

profiles.

Classifications Classification system.

Taxonpaths A taxonomic path in a

specific classification

system. Each succeeding

level is a refinement in

the definition of the

higher level.

Keywords Selected keywords from

multiple keyword lists.

Keywords can be

weighted.

One important element in the INSP profile is the

classification. Similarly to a library it represents the

categories covered by the service provider. An

instance fragment follows:

<choice>Educational content in

Science</choice>

</purpose>

<source>www.cosmos.com</source>

<taxon>

<identifier>Stars</identifier>

</taxon>

…

</taxonpath>

</classification>

PERSONALISED RESOURCE DISCOVERY SEARCHING OVER MULTIPLE REPOSITORY TYPES: USING USER

AND INFORMATION PROVIDER PROFILING

From this extract, the ‘identifier’ section

indicates that the service provider offers some

resources on the subject of Software Engineering.

There is a requirement in the INSP data model to

allow INSPs to offer hosting to content providers in

multiple domains or application areas. In addition,

some information service providers may wish to

extend their profile with other information outside of

the standard structure defined above.

The obvious solution is to use a two-category

structure: one generic category (for contact details,

legal constraints, payment information, etc) and an

unbound number of domain specific categories (for

keywords and classifications) to handle multiple

repository types.

2.3 Storage and Retrieval of profile

information

The overall system is based upon profiles. The

success of the query depends largely on how those

profiles are mapped and on the profile retrieval

performance. For storage and retrieval of user and

service provider profiles information, different

solutions were investigated:

• Local client file system storage.

• Central RDBMS Database storage.

• Central LDAP Directory Service.

Although, storing profile information on a local

drive is convenient, the issue of storing space would

arise, as the number of profiles gets larger. Such

solution would require investigating and

implementing compression mechanisms, which

would ultimately slow down the profile information

retrieval process.

The second solution is to store the data in a

relational database system such as Oracle (Chang et

al., 2000). The general principles are discussed in

(Kanne and Moerkotte, 2000) and (Florescu and

Kossmann, 1999). However, at the time of the

implementation, no mature, reliable XML database

management solution could solve this problem.

Also, such a deployment is heavyweight and

consequently would affect the systems’

performance.

The final and chosen option is to store the data in

a Directory Service such as the Lightweight

Directory Access Protocol, a simple variant of the

X.500 ISO standard (Howes et al., 1995). Some

basic classes for persons in an organisation are

directly derived from the X500 Directory Services

standards. A directory is a specialised database

optimised for storage, manipulation and retrieval of

information in a globally scalable system, supporting

sophisticated filtering capabilities and high volume

search request.

However, unlike database management systems,

directory services do not support complicated

transactions or rollbacks. A profile in a directory

service refers to a person or an organisation having

one or more relationships with the organisation

responsible for the maintenance of the directory. For

instance, an Internet Service Provider that stores

personal details and access rights. For an LDAP

Directory Service to implement a profile, the

information should be stored as classes ranging from

abstract to more specific ones. For instance, abstract

classes could be general information such as contact

details and description.

3 CANONICAL LANGUAGE

An underlying concept of the GUARDIANS is that

it has no knowledge of the metadata formats on

which the repositories are based. Hence, some

mechanism is needed to transform the metadata from

one format to another. For example, the water

molecule (H2O) can be written in HTML as

something like:

Mathematicians will represent this molecule in

MathML (MathML, 2001) as:

Canonical

Language

Format 1

HTML

Format 4

(graphical)

Format 2

(WYSIWYG)

Format 3

(MathML)

Format 5

(MoDL)

Figure 1: Canonical data model solution

ICEIS 2004 - HUMAN-COMPUTER INTERACTION

<math>

<msub>

</msub>

</math>

Whereas chemists might use the MoDL

(Molecular Dynamics Language) (MoDL, 1999)

</DEFINE>

After investigation, the most suitable model for

transforming structured metadata was determined to

be a canonical data model rather than providing the

transformations between all those formats. This

model requires that as a new metadata format be

added it is only necessary to create a mapping

between the new metadata format and the canonical

middle language. This is shown in Figure 1.

The Basic Semantic Registry (ISO/IEC 11179

Part 6, 1997) makes use of this canonical language

(CL) to translate from one metadata format into one

or more other(s). Most significantly, the BSR is used

to map user preferences to content metadata

representing the information provider’s content,

regardless of the content metadata format. It is the

job of the BSR to provide mappings between the

user profile and each of these target metadata types.

As the BSR is responsible for all XML metadata

conversions, it uses XSLT (XSL Transformations,

1999). Of course because XSLT is quite a specific

language, it is sometimes necessary to re-format the

data to enable its successful processing and

translation.

4 FEDERATED SEARCH

STRATEGY

The personalised resource discovery strategy is

performed by a software component set called

Mediation and Search Services (MSS, figure 2),

which comprises a Mediator and a Search Service.

For the purpose of the federated search strategy we

will only concentrate on the components of interest

for the search: the Mediation and Search Services.

The resource discovery strategy works on a three-

phase basis. The initial pass, performed by the

Mediator, is to select INSPs who can provide

content suitable for the learner initiating the search.

The second phase of the search is performed by the

Search Service component of the MSS, which

searches on the user selected INSPs for the relevant

content. The third phase is the results collection and

ranking, delivering the pool of results back to the

user.

4.1 Finding the relevant INSP

This is achieved through the Mediator component

matching the user’s profile (GUP) with the INSP

profiles stored in the Directory and Registry Service

(DRS). The underlying technology used in this

matching is XSLT (XSL Transformations, 1999).

The Mediator serves as the middle tier

interacting with the INSP profile, the graphical user

interface (UAS in figure 2) and the actual Search

Service. From the UAS, in interaction with the user

User

pplication

Services

(UAS)

Directory &

Registry

Services

(DRS)

Mediation

& Search

Services

(MSS)

INformation

Services

(INS)

Network

ccess

Technologies

Services

(NATS)

Figure 2: GUARDIANS Functional Architecture

PERSONALISED RESOURCE DISCOVERY SEARCHING OVER MULTIPLE REPOSITORY TYPES: USING USER

AND INFORMATION PROVIDER PROFILING

profile, the user defines his current preferences and

keywords to formulate a search request. Those user

preferences are then mapped to the canonical

language. A sample extract is as follows:

Table 3: GUP preferences mapping to the canonical

language

GUP CL

//accessibility/language=

’en’

//general/description/

language=’en’

//accessibility/preference

/refinement=’video/quic

ktime’

//general/technical/prefer

ence=’video/quicktime’

//accessibility/preference

s/preference/type=’Oper

ating System’

//general/technical/title=

’Operating System

Those types of mappings are expressed in XSLT

(Bradley, 2000) as simple ‘select-match’

expressions, which are used to select the current

element to match it to a particular target element.

Once this mapping is achieved, an XPath query

(Clark and DeRose, 1999) can be formulated in the

canonical language with the relevant user

information. This query formulation process is

similar to the second phase’s, hence will be

described in the next section.

4.2 Finding the relevant resources

For each of the INSPs selected by the Mediator, a

subset of the INSP profile is retrieved from the DRS.

Through this mechanism the location, type (query

type) and format (metadata schema) of each INSP’s

metadata repository is discovered.

In the next step, preferences are processed using

an XSLT document similar to the one defined in

Table 3 but this time with the INSP information.

Once the previous step is performed a general query

is formulated, based on the canonical language (CL).

However, this query cannot be executed against any

repository since it is schema “unaware”. To obtain a

schema specific query, the Search Service has to call

the BSR. A query is built according to the type and

format of the repository. An example of such query

for an English-speaking user requesting a search

with ‘universe’ as a keyword and ‘quicktime’ as his

preferred format could be:

<querystring>[!CDATA

/lom/general/language[text()=”en”] |

/lom/general//lom [(/lom/general/title/

langstring[contains(text(),"universe") or

/lom/general/description/langstring[

contains(text(),"universe")]]

/lom/technical/format[text()=”video/

quicktime”]

</querystring>

<querylanguage>XPATH</querylanguage>

</search>

In this example, ‘universe’ is the only input from

the user, language and format are derived from the

user’s profile. This query is then delivered to the

location specified and results added to the overall

pool of results.

4.3 Results handling

A result manager component of the MSS is in charge

of collecting, ranking and re-formatting the results

before they are sent back to the user. The idea is to

create a result collection, which is parsed and used

for ranking. This result collection is ranked

according to the preferences, keywords, etc. This

was developed in such a way to allow extension of

the rating algorithm, for instance to use other fields

from the User Profile. Once accomplished, the

results are stored into the following structure:

<description>The movie describe some of the

different satellites studying the Sun: The

hubble telescope, Solar max, and the very

large array in new Mexico; and information

about the solar system.</description>

<location>http://www.ist-guardians.org/

astronomy/Sun.mov</location>

<format>application/quicktime movie</format>

</resourceResult>

This example represents the XML structure of the

results, which are returned by the Search Service.

Basically, results are represented the same way as a

normal web search engine with the following

features describing the XML document above:

• A title to identify the result found.

• A description, to briefly identify what the

resource is about.

• A location, to provide a link to the resource

found.

• A ranking, as a degree of relevance of this

result that will be displayed to the user later

on.

ICEIS 2004 - HUMAN-COMPUTER INTERACTION

• A format representing the resource format for

display purposes. For instance, the type of

movie, audio, text to know the platform and

requirement for displaying it to the user.

• An INSP name, in case the user wants to

come back to this particular INSP later on.

(Part of the action history in the user profile)

The results discovered are rated using a simplistic

mechanism as follows:

1: an arbitrary base score (30%) is given for all

results (since it has been found, matching a

minimum of one keyword).

2: a score is given for each keyword found in the

title (20%).

3: a score is given for each keyword found in the

description (6%).

4: a bonus score is given if the title or description

contains more than one keyword (10 and 5%).

5: a score is given for the preference level at which

the result was found (3%).

A greater number of preferences matched, means

a better, more personalised result for the user.

5 CONCLUSION AND FUTURE

WORK

As the number of resources on the Internet continues

to grow, user faces the issue of information

overload. In this situation, reliable search tools that

enable personalisation of resource discovery and

information retrieval becomes crucial. This is

specifically true in the E-Learning domain, where

users (students or researchers) have a clear need to

retrieve information relevant to their preferences

(language, interest, platforms, etc). Simple text-

based search engines are not sufficient anymore.

Several improvements have been made in the past

few years showing a better search accuracy. For

instance, Google (Page, 1998) now provides a new

improved HTML based search engine with regard to

quicker delivery and better search accuracy.

Although such tools are successful, it is only a

partial solution, it only delays the problem as more

and more information (educational, recreational, etc)

becomes available and as the number users still

increases.

A complete solution to this problem is the

combination of current and emerging technologies

and standards to offer better interaction between the

user and federated search strategies. This solution,

presented in this paper, has been researched and

implemented by the author in the E-Learning context

of the GUARDIANS European project as part of its

Mediation Service component. It offers a very

efficient search facility that enhances the user’s

learning experience through the usage of smart

functionalities (components and resources) for

discovery and retrieval of educational resources.

Although this work is based on E-Learning

resources, one could easily fit it into the overall

context of information retrieval on the Internet.

Although the search strategy using XPath

performed successfully, it was felt that the queries

were limited with regards to performance and

flexibility. Indeed, to incorporate most of the user

preferences, the XPath was complicated and major

results reformatting had to be performed. For this

reason this initial work was extended and an XML

querying solution using XQuery (Boag et al., 2002)

was proposed in (Rousseau, Leray, 2003).

Another important point to consider is that the

matching between user profile and metadata was

limited but it would be possible to include more

information about the user. For example, the search

could use the affiliation to only search for INSPs

with which the user is affiliated. It could also use

qualification details to only retrieve information

relevant to the user’s previous qualification,

diplomas and other licenses or certifications.

However, the user profile might still be extended

with new features according to future development

in the E-Learning community.

The ranking algorithm implementation was an

interesting part of the work. However, with time

constraints and the lack of information about how to

actually develop a precise ranking mechanism. It

was first developed as a basic algorithm, only taking

care of the keywords. For a possible

commercialisation purpose, it should have taken into

account a few more feature from the user profile,

specifically provide a complex algorithm using

weighting attributes. It is expected that the authors

will be involved into future research in the area of

stopwords, tf.ifd, coordination level, etc (Baeza-

Yates, 1999).

Nowadays, there is a lot of hype around agents.

In the context of this work, an agent would enable

automatic discovery of the user’s platform

capabilities. This issue was encountered while

populating the user profile with some device

information. Hence, a typical user would be unaware

of its device(s) capabilities and will require an

automatic detection mechanism. Also, a good

balance between static and dynamic profile

information should be researched further to present

the user with a reliable profile that requires the

minimum efforts to fill in. A form of interaction

between the profiling mechanism and the user is

therefore crucial to get some feedbacks to the

system.

The abstract representation of data on the World

Wide Web, also known as the Semantic Web

PERSONALISED RESOURCE DISCOVERY SEARCHING OVER MULTIPLE REPOSITORY TYPES: USING USER

AND INFORMATION PROVIDER PROFILING

(Berners et al., 2001), is gradually taking over the

current HTML based Web. In the near future, the

development in the area will lead in significant new

possibilities to enhance and process the data that

they merely display at present. Hence, programs will

be able to exchange information with others and

automate services. Such functionalities together with

the flexibility of XML will be promoted using

agents, that will help users tracking down resources

and match them against their criteria and

preferences.

REFERENCES

Baeza-Yates, R. and Ribeiro-Neto, B. Modern Information

Retrieval, Addison-Wesley, ACM Press, 1st edition

(May 1999). ISBN: 020139829X

Berners-Lee T., Hendler J. and Lassila O. The Semantic

Web. A new form of Web content that is meaningful

to computers will unleash a revolution of new

possibilities. Scientific American, May 17, 2001.

Boag, S., Chamberlin, D., Fernandez, M.F., Florescu, D.,

Robie, J., Simeon, J., and Stefanescu, M., eds 2002.

XQuery 1.0: An XML Query Language. W3C Working

Draft, http://www.w3.org/TR/2002/WD-xquery-

20020430/

Bradley, N., September 7, 2000. The XSL Companion.

Addison-Wesley Pub Co; 1st edition. ISBN:

0201674874

Chang, B., Scardina, M., Karun, K., Kiritzov, S., Macky,

I., Novoselsky, A. and Ramakrishnan, N., 2000.

ORACLE XML Handbook. Oracle Press. ASIN:

007212489X

Clark, J., and DeRose, S. eds 1999. XML Path Language

(XPath) version 1.0. W3C Recommendation.

http://www.w3.org/TR/xpath

Day, M., Summer 2001. Metadata in a nutshell. Article

published in Information Europe, p. 11. Information

Europe quarterly magazine of EBLIDA.

Florescu, D., and Kossmann, D., 1999. A Performance

Evaluation of Alternative Mapping Schemes for

Storing XML Data in a Relational Database. Technical

report no. 3680, INRIA, Le Chesnay Cedex, France.

GESTALT, Getting Educational Systems Talking Across

Leading-Edge Technologies, EU project (started in

October 1998) Web Site:

http://www.fdgroup.com/gestalt/

GUARDIANS project, 2000. Gateway for User Access to

Remote Distributed Information and Network

Services, started in October 2000. Web site:

http://www.ist-guardians.tv/

Howes, T., Yeong, W. and Kille, S, 1995. Lightweight

Directory Access Protocol (LDAP). The Internet

Society, RFC 1770.

IEEE Learning Technology Standards Committee WG12:

Learning Object Model, 6 March 2002.

http://ltsc.ieee.org/wg12/

IEEE PAPI, 01 February 2002. IEEE Public and Private

Information draft 8 specifications. Available at:

http://www.edutool.com/papi/

IMS LIP version 1.00, 18 March 2001. Learner

Information Package Specification. Available at:

http://www.imsglobal.org/profiles/index.cfm

IMS Learning Resource Metadata Specification, 01

October 2001.

http://www.imsglobal.org/metadata/index.cfm

ISO/IEC 11179 Part 6, 1997. Information technology -

Specification and standardization of data elements -

Part 6: Registration of data elements. Available online

at: http://www.iso.ch/

Kanne, C.-C., and Moerkotte, G., 2000. Efficient Storage

of XML Data. Proceeding of the International

Conference on Data Engineering. San Diego,

California, p. 198.

MathML, W3C math specification, 21 February 2001,

http://www.w3.org/Math

MoDL - Molecular Dynamics Language project, 29 July

1999, http://xml.coverpages.org/modl.html

Page, L., Brin, S. The anatomy of a large-scale

hypertextual Web search engine, Computer Networks

and ISDN Systems, Volume 30, Issues 1-7, April 1998,

Pages 107-117.

RDS. Distributed Systems Technology Centre: Research

Data Unit. Resource Discovery – A Definition.

Available from

http://www.dstc.edu.au/RDU/rd_define.html

Rousseau, B., Browne, P., Malone, P., ÓFoghlú, M., 2003.

User Profiling for Content Personalisation in

Information Retrieval. In the ACM Symposium on

Applied Computing, special track on Information

Access and Retrieval. ACM Press

Rousseau, B., Leray, E., ÓFoghlú, M., 2003. Metadata &

Information Management issues in XML-based

Mediation. In ICEIS’03 Fifth International

Conference on Enterprise Information Systems,

Angers, FRANCE, pages 375-379. ICEIS Press.

RSLP. Research Support Libraries Programme British

initiative, September 1999, http://www.rslp.ac.uk/

Spark, K. Jones, Artificial Intelligence: What can it offer

to Information Retrieval. In Proceedings of the

Informatics 3, Aslib, ed., London, 1978.

Suryanarayana, L., Hjelm, J., May 7-11, 2002. Profiles for

Situated Web. Johan. In WWW 2002, Honolulu,

Hawaii, USA. ACM Press 1-58113-449-5/02/0005.

TV Anytime forum, formed in 1999. Web site, available

at: http://www.tv-anytime.org/

UKOLN, centre of expertise in digital information

management, University of Bath, England.

http://www.ukoln.ac.uk/

ICEIS 2004 - HUMAN-COMPUTER INTERACTION

vCard, December 1996. The Internet Mail Consortium

(IMC), available at:

http://www.imc.org/pdi/vcardoverview.html

XML, the eXtensible Markup Language 1.0 second

Edition. W3C Recommendation 6 October 2000,

http://www.w3.org/TR/REC-xml

XSL Transformations (XSLT), version 1, W3C

Recommendations, 16 November 1999,

http://www.w3.org/TR/xslt

PERSONALISED RESOURCE DISCOVERY SEARCHING OVER MULTIPLE REPOSITORY TYPES: USING USER

AND INFORMATION PROVIDER PROFILING