CONTENT ENRICHMENT THROUGH DYNAMIC

ANNOTATION

George R S Weir

Department of Computer and Information Sciences, University of Strathclyde, Glasgow, UK

George Lepouras, Costas Vassilakis

Department of Informatics and Telecommunications, University of Athens, Athens, Greece

Keywords: Dynamic annotation, Web content enrichment

Abstract: This paper describes a technique for interceding between users and the information that they browse. This

facility, that we term ‘dynamic annotation’, affords a means of editing Web page content ‘on-the-fly’

between the source Web server and the requesting client. Thereby, we have a generic way of modifying the

content displayed to local users by addition, removal or reorganising any information sourced from the

World-Wide Web, whether this derives from local or remote pages. For some time, we have been exploring

the scope for this device and we believe that it affords many potential worthwhile applications. Here, we

describe two varieties of use. The first variety focuses on support for individual users in two contexts

(second-language support and second language learning). The second variety of use focuses on support for

groups of users. These differing applications have a common goal which is content enrichment of the

materials placed before the user. Dynamic annotation provides a potent and flexible means to this end.

1 INTRODUCTION

Our research over the past five years suggests that a

desirable characteristic in many application areas is

the ability to intercept information requested by a

Web user and modify its content on-the-fly before it

arrives on the user’s display. Such activity has no

sinister intent, but affords an effective means of

supplementing or filtering Web content in a ‘just in

time’ fashion so that the user’s information exposure

is better honed to the needs or wishes of that user.

The technology required to affect such a system

builds upon a common component in network

architectures, viz., a caching Web proxy server.

In a Web context, many organisations require

that users browse the Web in a controlled and

monitored fashion. Often, this entails the use of a

Web proxy facility such as ‘squid’ (http://

www.squid-cache.org/). These systems sit between a

user’s client machine and remote Web servers. The

local Web client software is configured to relay all

Web requests to the proxy server. In turn, this proxy

forwards the request to the appropriate remote site.

Since the remote Web server receives the request

from the proxy machine, its response is directed to

the address of the proxy server. Upon receipt of the

requested document, the proxy server determines

which internal client should receive this response

and relays the document to the appropriate host.

Such behaviour provides a number of security

advantages to the local network. In the first place,

local users cannot access external Web sites except

via the proxy server. Thereby, the proxy provides a

central point at which a record of all Web browsing

can be maintained. This proxy log is a useful device

for monitoring the behaviour of Web users by

recording who is browsing and what information

they are viewing. A second benefit of Web proxies

is that they often allow administrators to impose

restrictions on the local user population by curtailing

access to specific sites. A third proxy benefit is the

fact that such servers often provide a caching

facility. This reduces duplicate requests for remote

documents, in line with the expiry times specified on

retrieved Web pages, and thereby reduces data

152

R S Weir G., Lepouras G. and Vassilakis C. (2004).

CONTENT ENRICHMENT THROUGH DYNAMIC ANNOTATION.

In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 152-157

DOI: 10.5220/0002643301520157

 SciTePress

ure 1: We

rox

in networ

confi

uration.

traffic to and from the network. Finally, use of a

Web proxy affords a measure of protection for local

client machines. Since external servers only see the

proxy address, the addressing scheme and number of

machines behind the proxy can be concealed from

external view. The usual network position of the

Web proxy is illustrated in Figure 1.

2 DYNAMIC ANNOTATION

To this context of proxy-based Web browsing we

bring the possibility of intercepting incoming Web

pages, scanning, and optionally modifying their

content. This facility we term ‘dynamic annotation’.

‘Dynamic’ reflects the fact that content can be

modified on-the-fly between the user’s request and

delivery of the selected Web page. ‘Annotation’

denotes a principal use that we have found for this

mechanism.

Note that our use of annotation is more generic

than other Web annotation approaches (e.g., Kahan

et al, 2001 and Ovsiamnikov et al, 1999).

Commonly, such systems apply annotation as a

commenting facility available to those viewing

Web-based information. In such contexts, users are

able to add their comments by annotating the

information that they view. Contributed comments

are then visible to other users. For example, Annotea

(op. cit.) supports ‘shared annotations (comments,

notes, explanations, or other types of external

remarks that can be attached to any Web document

or a selected part of the document)’. Significantly,

Annotea does not modify the original Web

document but stored comments externally in an

annotation server. A metadata facility for describing

annotations and a pointer mechanism for locating the

annotations in the annotated document are the basis

for unifying the document and annotation display for

the end user.

While a manual facility is included in the third of

our annotation applications (described below), the

feature that differentiates our approach to annotation

is that the system itself adds ‘extra’ information to

the Web pages requested by the user. This is closer

to the Magpie approach described by Dzbor et al

(2003). In the following, we outline the operation of

our annotation facility and describe its use in several

example applications.

2.1 Annotation mechanism

In order to achieve the interception and modification

of Web content we have employed a Web server that

supports proxying. The Apache server (cf.

http://www.apache.org) fits this bill and has the

added benefit that it allows the use of customised

dynamically loadable modules. This convenience

permitted the development of an add-on module that

filters the content of retrieved Web pages. The

filtering process relies upon a parsing mechanism

that, in turn, applies a dictionary of terms. These are

the terms upon which some modification is required.

Our initial versions of the annotation facility

used a static dictionary with pre-defined terms. The

later version supports a dynamic dictionary,

allowing the terms of interest to be changed during

the operation of the annotation system. This feature

provides a basis for adapting the Web content

modification strategy during its operation. Work on

this adaptive mechanism, which employs feedback

from the user’s interaction, is currently in progress.

The architecture for the dynamic annotation facility

is shown in Figure 2.

CONTENT ENRICHMENT THROUGH DYNAMIC ANNOTATION

153

ure 2: D

namic annotation mechanism.

2.2 Interaction support

We have developed applications using dynamic

annotation for two different contexts in which the

user’s native language is a significant factor. The

first case is the context of second language support

the second is the context of second language

learning.

Second language support is an approach that we

have developed to the ‘second language problem’ in

human-computer interaction (Weir et al., 1996;

Lepouras & Weir, 1999). This is the setting in which

non-native speakers of English are obliged to

operate with English-based computerised

information systems. A common consequence in this

scenario is difficulty in user interaction. For such

individuals, language introduces particular

difficulties and misunderstanding in the use of such

interactive systems.

One might expect that fully language-localised

software products or information systems would

solve all problems of language misunderstanding for

the end user. Experience suggests that this is not the

case. Instead of full remedy, the localisation strategy

adds or replaces existing problems with new ones.

We have characterised several sources of such

problems (Lepouras & Weir, 1999), including:

1. the absence of standardised terminology;

2. mixed language environments, and

3. cross-language confusion.

At first sight, Web information systems seem an

unlikely setting for the second language problem.

Web pages deliver information, so language

comprehension will be an issue but the relevance of

this to user interaction is not obvious. The difficulty

is that reading information from Web pages is a

major component of the interaction context. Web

users do more than read delivered information. The

Web is an interactive system in which the user must

negotiate a route through selected information

resources. This means that the user’s ability to

comprehend the visible labels on Web links will

affect their choice of successive resources.

The user selects links based upon presumptions

of their relevance and interest value. Where

hyperlinks rely upon textual description, the user’s

interpretation of the text plays a dominant role in

determining interaction. From this we conclude that

language comprehension significantly affects the

success of user interaction in the Web context. In

consequence, user’s whose first language differs

from that used in the Web pages are more likely to

experience degraded interaction. This is the feature

of the second language problem.

Our work on second language support has

focused on examples in Greek and Chinese.

Although these languages require font sets that

complicate their use in conjunction with English

language displays, we are able to combine local

language support (e.g., Chinese text) with the

original language (i.e., English).

In aiming to support the second-language user

we could aim for full translation of Web page

content, e.g., English to Chinese, but translation is

not our principal concern. In addition, experience

suggests that targeted use of local language support

can enhance user interaction by improving

comprehension whilst minimising additional risk of

language-based dissonance. We have explored a

variety of mechanisms for supplementing Web

information with native language annotations. One

approach adds selective translations in the form of

pop-ups, to pre-selected English terms. An example

of this technique is shown in Figure 3.

ICEIS 2004 - HUMAN-COMPUTER INTERACTION

154

Figure 3: Chinese annotation on English Web content.

In this example, English terms that may not be

readily understood by the native Chinese speaker are

annotated with Chinese translations or explanations.

This is achieved by searching incoming Web page

content, locating terms that require elucidation and

modifying the Web page content to include the

Chinese addendum. In this instance, we have

changed the HTML code to include an icon and

generate pop-ups with suitable Chinese language

content. The target English expressions are also

altered so that they appear in red. This enables the

Chinese user to identify readily those terms for

which supplements are available.

Other annotation techniques include the use of

frames to add a second source of commentary on the

English original. This technique appeals to those

users who prefer that the original content is left

unaltered. Whatever the specific annotation facility,

each has the aim of content enrichment.

2.3 Language learning support

A related but distinct area of application for dynamic

annotation is a setting in which users wishing to

learn English may use Web browsing as a basis for

enriching their familiarity with English language.

We have explored the use of Web-based content

enrichment as a means of adding explanations in the

native language of the user (Weir & Lepouras,

2001). This provides a native language support that

does not seek to smooth user interaction but rather

aims to facilitate better understanding of English

language and texts expressed therein.

A particular strength in this use of dynamic

annotation is that language learners may freely

explore Web sites without constraint. The support

facility is tuned to add annotation in a variety of

forms to specific sets of English terminology. This

context is considerably richer in varieties of support

since the variability of linguistic needs may be

extensive.

An area that merits greater exploration is the

application of criteria to determine the deployment

of native language annotation. While our work in

this aspect is still at an exploratory stage, we have

identified several plausible factors to serve as drivers

for such dynamic support:

1. Learner selected experience level

2. Frequency table

3. Fog index

4. Meta information

The simplest approach is learner driven. This

provides a range of support levels and allows

learners to vary the degree of annotation applied.

Other techniques, such as frequency analysis of

English terms and expressions, are potential

supplements to user selection. This information

would enable the selective application of native

language support, based upon the likely exposure of

the learner to such terms. In addition, a Fog Index

may be determined for each paragraph, in order to

gauge the likely need for clarification. Such metrics

as the Gunning fog index (Gunning & Kallan, 1994)

provide a heuristic measure of text readability based

upon the average number of words in each sentence

and the number of polysyllabic words in the given

sample of t Although our purpose is not to modify

original English content, the fog metric is a viable

technique for guiding annotations. Furthermore,

such insight may focus attention on paragraphs to

which automated summarisation techniques may be

CONTENT ENRICHMENT THROUGH DYNAMIC ANNOTATION

155

appropriate. Finally, a rating system may be

introduced to signify the language complexity for

specific Web pages. Either this could be provided by

enlightened authors as 'meta' content to their pages

or generated dynamically and added to cached

versions of incoming pages. Through dynamic

annotation techniques driven by metrics for English

language complexity, there is scope to provide a

virtual 'English Assistant' that will ensure that

English content is appropriately enriched to support

user learning.

2.4 Group support

Our application of dynamic annotation has been

extended to a third context: group support (cf.

Lepouras et al, 2003). Here, we have a Web-based

conferencing facility that supports the management

and dissemination of domain experience in an

organisation's operational environment. In this

context, workers often share valuable experience and

knowledge but this is usually exchanged in an

informal rather than formal manner. This tacit or

implicit knowledge is often lost as it resides in email

and memos that afford no easy method of retrieval,

classification, or cross reference.

The content enriched system that we have

prototyped allows for dynamic classification of local

knowledge in a systematic manner that eases its

retrieval from the knowledge database. In operation,

the system requires that individuals login and

configure a profile of user preferences. Thereafter,

the user can register for topic-specific discussions,

exchange messages with other system users, create

direct links to favourite or interesting messages,

annotate messages and rate them according to the

validity or importance of their content (Figure 4).

Dynamic annotation makes an appearance

through automatic recognition of key terms and

automatic enhancement of messages via auto-

generated links to other messages with content of

relevance to topics in which the user has expressed

an interest (Figure 5). In concept, this shares some

aspects of the ‘ComMentor’ system developed at

Stanford University (Roscheisen et al., 1994). One

major difference is that our system is active in

automatically cross-referencing content and in its

use of term equivalence in support of user searches.

At the heart of the system’s classification

component is an explicit taxonomy of the

organisation. This taxonomy includes lists of

equivalent terms whose nodes can be

interconnected; thereby the user may search for a

term in messages and retrieve a message even

though the exact term does not appear in the search

pattern.

ure 4: Link creation dialo

ue.

ICEIS 2004 - HUMAN-COMPUTER INTERACTION

156

Figure 5: Message with annotated keywords.

The taxonomy and the general discussion

categories are maintained through the system

administration screens. Through these screens the

administrator can define new general discussion

categories, connect them to the taxonomy and add

key-terms to describe the topics. In its present form,

this application has extensible content enrichment

but is not yet adaptive.

3 GENERAL APPLICATION

Our purpose in this paper is to detail the benefits of

dynamic annotation as a means of content

enrichment and to indicate the general applicability

of this approach wherever the delivery of Web-

sourced information can be enhanced to better suit

the needs or interests of the individual user.

The variety of application areas extends from

supporting the interaction of individuals for whom

English is a second language, through to annotation

in the context of group interaction. Between these

extremes, dynamic annotation affords support for

those seeking to use Web resources as a basis for

learning English, and as a means of building upon

otherwise implicit institutional knowledge.

Next on our research agenda is the addition of

feedback from user interaction with content and

annotations as a basis for building adaptivity into the

mechanism for dynamic annotation. This will allow

us to explore ways of allowing user behaviour to

direct the mode and content of annotation.

REFERENCES

Dzbor, M., Domingue, J. & Motta, E., 2003. Magpie –

towards a semantic web browser. In Proceedings of

the 2nd International Semantic Web Conference 2003

(ISWC 2003), Florida, USA

Gunning R. and Kallan R. A., 1994. How to take the fog

out of business writing, Dartnell.

Kahan, J., Koivunen, M. R., Prud'Hommeaux, E. &

Swick, R. R., 2001. Annotea: An Open RDF

Infrastructure for Shared Web Annotations. In Proc. of

the 10

World Wide Web Conference (WWW10),

Hong Kong.

Lepouras, G., Vassilakis, C, Weir, G. R. S., 2003. A

System to Support Dissemination of Knowledge and

Sharing of Experiences in the Working Environment,

International Journal of Continuing Engineering

Education and Life-Long Learning, Volume 13, Nos.

3/4, 2003, 248-257.

Lepouras, G. and Weir, G.R.S., 1999. It's not Greek to me:

terminology and the second language problem', ACM

SIGCHI Bulletin, vol.31 No.2, (April), 17-24, ACM

Press.

Lepouras, G. and Weir, G.R.S., 2003. Subtitled

interaction: Complementary support as an alternative

to localization, International Journal of Human

Computer Studies, Volume 59, Issue 6, December,

941-957.

Ovsiannikov, I.A., Arbib, M.A., and Mcneill, T.H.,

Annotation Technology. International Journal of

Human-Computer Studies, 1999. 50(4): p. 329-362.

Roscheisen, M., Mogensen, C. & Winograd, T., 1994.

Shared Web annotations as a platform for third-party

value added information providers: Architecture,

protocols, and usage examples, Technical Report

CSDTR/DLTR, Stanford University.

Weir G.R.S. and Lepouras G, 2001. English Assistant: A

Support Strategy for On-Line Second Language

Learning, IEEE International Conference on

Advanced Learning Technologies, IEEE Press.

Weir, G.R.S., Lepouras G. & Sakelleridis, L., 1996.

Second language help for Windows applications, in M.

A. Sasse, R. J. Cunningham and R. L. Winder (Eds.)

People and Computers XI, Proceedings of HCI'96,

Springer, London, 129-138.

CONTENT ENRICHMENT THROUGH DYNAMIC ANNOTATION

157