AN EXTENDED ARCHITECTURE FOR ADAPTATION
OF SOCIAL NAVIGATION
Manel Mezghani, Corinne Amel Zayani, Ikram Amous and Faiez Gargouri
MIRACL Laboratory, ISIMS, El Ons City, Street of Tunis Km 10, Sakiet Ezziet 3021, Sfax, Tunisia
Keywords: Adaptation, Social Navigation, Architecture, Recommendation, Tagging Behaviour.
Abstract: Social navigation is a way for users to navigate through social information such as resources and social
annotation (tags). However, due to the growth of social networks, the user could be lost. To avoid this
problem of disorientation, we try to adapt the social information through a recommendation technique by
providing useful information according to the user’s needs. In this paper, we present an extended
architecture of social recommender system. The originality of this architecture rely on the way to combine
the collective intelligence of the social network with the user’s behaviour especially his tagging behaviour
.
1 INTRODUCTION
In general, social navigation is a way for users to
navigate through social information such as
resources and social annotation (tags). Social
navigation is classified as direct or indirect (Farzan,
2009). The direct interaction of users with each other
in the form of recommendation or guiding is defined
as direct social navigation. Tracing activities of
users to guide new users in the system is defined as
indirect social navigation. Indirect social navigation
can be classified as: collaborative filtering and
history-enriched information spaces (Brusilovsky, et
al., 2010). Collaborative filtering aims to help users
to navigate according to information of all users.
History-enriched information spaces provide support
for navigating by making individual action of others
visible.
Even if, indirect social navigation explores
different techniques (collaborative filtering and
history-enriched information spaces) and due to the
growth of social networks, the user could be lost. To
avoid this problem of disorientation, some
researchers have been done to adapt the social
information through a recommendation technique by
providing useful information according to the user’s
needs. Some researchers study either how to
recommend pertinent resources based on user profile
(Zheng, et al., 2011) (De Meo. et al., 2010), or how
to recommend relevant tags based on user tagging
behaviour (Musto, et al., 2009), or how to
recommend both resources and tags (Carmagnola, et
al., 2011) (Nauerz, et al., 2008).
There are several architectures which adapt
social information based on recommendation
technique. In general, these architectures use the
same modules that exist in the classic architecture of
adaptation systems like AHA! (De Bra, et al., 2003):
the user modelling module and the adaptation
module which could be devised in three sub modules
(Brusilovsky, 1996): adaptation of presentation,
adaptation of contents and adaptation of navigation.
In social context, adaptation architectures add social
modules such as: social networking module
(Carmagnola, et al., 2011) which define social
objects (i.e.: users, resources, etc.) and social
interactions (i.e.: assigning an annotation, rating a
resource, etc). From this social networking module,
some architectures define a sub module that specify
the tagging behaviour of user (which resource is
been tagged by which user) (Nauerz, et al., 2008)
(Kim, et al., 2010).
Architectures which adapt social information
based on recommendation technique and using these
modules like (Nauerz, et al., 2008) (Carmagnola, et
al., 2008), don’t take into consideration the
ambiguity associated to tags and so the quality of
recommendation could decrease. However,
(Carmagnola, et al., 2011) architecture takes into
consideration the semantic of tags. But, tag
ambiguity is not treated efficiently which affect the
recommendation quality. Another limit is that the
architecture depends on the TV partner and uses the
log file to build a user profile.
540
Mezghani M., Amel Zayani C., Amous I. and Gargouri F..
AN EXTENDED ARCHITECTURE FOR ADAPTATION OF SOCIAL NAVIGATION.
DOI: 10.5220/0003907005400545
In Proceedings of the 8th International Conference on Web Information Systems and Technologies (WEBIST-2012), pages 540-545
ISBN: 978-989-8565-08-2
Copyright
c
 2012 SCITEPRESS (Science and Technology Publications, Lda.)

In order to overcome limits related to these
architectures and to avoid the disorientation of the
user, we introduce an extended architecture of social
adaptation systems. The extended architecture
recommends information suitable to users, based on
analyzing tags, tagging behaviour and user profiles.
The originality of this architecture rely on the way to
combine the collective intelligence of the social
network with the user’s behaviour especially his
tagging behaviour.
We try through this architecture to adapt indirect
social navigation. Although the majority of projects,
dealing with social navigation are exploring
collaborative filtering, we try to employ both
collaborative filtering and history-enriched
information spaces. From this latter, individual
action of other users would be extracted from social
annotation (tags) assigned by users.
This paper is organized as follow: Section 2
presents an overview of the related works. Section 3
is dedicated to the presentation of our extended
architecture. Section 4 presents the detailed modules
used in our extended architecture. In Section 5 we
will conclude the paper by presenting some future
works.
2 RELATED WORKS
In social networks, the adaptation of social
information (resources and tags) could be based on a
recommendation technique. The latter, is classified
in (De Meo, et al., 2010): i) Content Based approach
(CB), which aim to recommend objects that are
relevant to the user; ii) Collaborative Filtering
approach (CF), which aim to use the collective
intelligence of the social network to recommend
social information. As the CF becomes widely used,
tag-based CF becomes more present in literature
(Kim, et al., 2011).
There are many techniques using Tag-based CF:
(Wang, et al., 2010) employ tag-based CF for
integrating the individual user’s tagging history in
the recommendation of tags and content of
resources, in order to adapt social navigation.
(Wang, et al., 2010) don’t update for the user profile
through time. (Zheng, et al., 2011) use the
importance and usefulness of tag and time
information in a CF context, when predicting user’s
preferences. From this prediction, they examine how
to exploit such information to build an effective
resource-recommendation model, but tags used are
not filtered and ambiguous. Researches of (Wang, et
al., 2010) and (Zheng, et al., 2011) and don’t
consider the semantic ambiguity associated to tags.
Contrary to (Zhao, et al., 2008), who suggest a tag-
based collaborative filtering, based on the semantic
distance among tags (from WorldNet dictionary) for
calculating user similarity. However, the similarity
measure is not very accurate since it doesn’t treat tag
ambiguity.
In this work, we are interested in architectures
which use the tag-based CF for recommendation. In
portals, (Nauerz, et al., 2008) analysis user’s tagging
behaviour to learn interests and preferences of users,
groups or communities, for better adaptation and
recommendation of tags and resources.
(Carmagnola, et al., 2008) presented “iCITY” as an
adaptive, social, multi-device recommender guide
which deals with cultural events taking place in
Torino city. These cultural events considered as
resources, are recommended based on the tagging
behaviour of users.
All these works don’t take into consideration the
ambiguity associated to tags and so the quality of
recommendation could decrease. However,
(Carmagnola, et al., 2011) suggest an architecture
derived from the “iCITY” architecture named
“iDynamicTv”, to recommend TV content (video)
and navigate through resources, tags and users. The
“iDynamicTv” architecture is a good way to
discover and organize TV content; it takes into
consideration the semantic of tags. However, tag
ambiguity is treated with a semantic similarity using
WordNet dictionary only and doesn’t consider spam
and personal tags which affect the recommendation
quality. This architecture depends on the TV partner
for TV content. Another limit is the use the log file
to build a user profile especially the no tracking
event in a log file.
We try to overcome the limitations in the
existent tag-based CF architecture, by extending the
“iDynamic” architecture.
From techniques listed above, we try to
overcome their limits. We try to integrate tagging
history in our architecture system in a different way
by analyzing tagging behaviour from the whole
users and update each user’s profile to adapt social
navigation. Analyzed tags are filtered to guaranty
accuracy. User’s similarity is calculated from their
similar behaviour, similar annotations (extracted
from WordNet dictionary) and from similar users
sharing common interests.
In table 1, we compare these architectures
according to specific criteria. Criteria are devised
into two main categories. The user
category, which
compare how a user is represented through: a static
way (by gathering information that rarely changes
ANEXTENDEDARCHITECTUREFORADAPTATIONOFSOCIALNAVIGATION
541

like name, age, etc.,) or dynamic way (by gathering
information that frequently changes like the tagging
behaviour), user’s interest update as they change
over time and user similarity. The tag category
specifies if the architecture takes into consideration
the semantic aspect of the folksonomy (tags), filters
inappropriate tags and if it considers tags’ weight
(the weight reflect the degree of importance of the
tag).
Table 1: Comparison of architectures which use the tag-
based CF for recommendation.
Characteristic
Reference
User Tag
Static
Dynamic
Update
Similarity
Semantic
Filtering
Weight
N
auerz et al., 2008
9 9
Carmagnola et al., 08
9 9 9
Carmagonla et al., 11
9 9 9 9 9
Our architecture
9 9 9 9 9 9 9
3 ARCHITECTURE
From architectures discussed above and from the
comparison between these architectures and our
architecture (in table 1), we try to explain deeply our
approach of adaptation. We present first a
motivating scenario which explains the purpose of
our architecture. Then, we present an overview of
the architecture across its different components and
interactions.
3.1 Motivating Scenario
Let’s take an example of a user who uses his social
network to navigate through its different resources
and tags.
In social networks, the number of resources is
regularly growing. The user has a limited view of
what exist in his social network. In consequence,
pertinent information may exist but the user doesn’t
see it. In spite of the fact that the user is connected to
other users (friends) and is a member of groups
(users sharing common interests), he could not have
pertinent information and could be lost or influenced
by other bad users (i.e.: spammers) while navigating.
In order to avoid these limits, we try to adapt the
indirect social navigation. This adaptation ensures
that the user will have the pertinent information he
needs by recommending relevant resources and tags.
To adapt indirect social navigation, we need to
analyse different social elements especially the user
through his profile and his social behaviour. We
need also to analyse resources and tags and how
these elements are relevant to the user and could
affect his social navigation.
3.2 Architecture Overview
In this paper, we present an extended architecture to
adapt social navigation by recommending tags and
resources. The architecture is inspired by
(Carmagnola, et al., 2011), which combines web2.0,
social networking and user-model personalization.
Architecture of (Carmagnola, et al., 2011), is
proposed to get a powerful tool for discovering
organizing of content in interactive television. We
try to overcome some limitations in this system,
which cause the dependence of the content of the TV
partners, no filtering of inappropriate tags and limits
of using log file to build a user profile.
In figure 1, we present the main components of
our system and relationships between different
modules of the architecture. This work aims to
prevent the disorientation of user in social networks.
It offers the possibility to navigate through resources
and tags. The system can detect similar users
according to their similar behaviour, similar
annotations (extracted from WordNet dictionary)
and from similar users sharing common interests. It
analyzes the tagging behaviour and filters no
appropriate tags to improve tag-based
recommendation.
The databases (DB) presented in this architecture
are:
DB social network:
The data exploited in this approach are
extracted from a specific social network (delicious,
movieLens, etc). Social information recommended is
depending on the social network (i.e.: bookmarks in
Delicious, scientific articles in CiteUlike, music in
Last.fm).
DB user model:
From the DB social network, this module
specifies information about users and networks of
users (interest, preferences, friends, professional
relationships, etc.).
DB Contents:
From the DB social network, this module stores
information about the resources of the social
network (type of resource, tags associated by each
users, metadata, etc.)
We present a simplified scenario for the
communication between modules. Architecture’s
modules will be detailed in section 4. After
WEBIST2012-8thInternationalConferenceonWebInformationSystemsandTechnologies
542

Figure 1: Extended architecture.
defining databases, interactions will be as follow: (1)
The module user modelling gets information from
the BD user model to create a user profile and
update it as his information change through time. (2)
From DB user model and BD contents, the tagging
behaviour module defines the relation between user
and resource through a common annotation (tag). (3)
The social networking module takes information
from both DB user model and BD content to
construct a social network. The social network is
defined by social elements (i.e.: users, resources)
and social interactions (i.e.: friends, annotations). (4)
The social networking module tries to define
similarity between users by calculating it through the
WordNet dictionary. (5) The filtering module detects
tag’s synonyms and homonyms, etc., from
functionality present in the WordNet dictionary. (6)
The tagging behaviour module contains
inappropriate tags which are detected due to the
filtering module. (7) In order to increase the quality
of adaptation, the filtering module provides
appropriate tag already filtered. (8) The adaptation
module needs social connections as friends, active
users, etc., from the social networking module, to
recommend social information. (9) The adaptation
module needs social elements such as resources to
recommend them. (10) Interaction between users
and the system. This interaction is the only input-
output in the architecture. The input is the user’s
information (especially static information) and
user’s request. The output is the result of the
adaptation process (the recommended resources and
tags).
4 MAIN AND SECONDARY
MODULES
Modules are divided in two categories: i) The main
modules which are presented in the most social
adaptation architecture and already presented in
section 1 (the user modelling module, social
networking module and adaptation module) ii) the
secondary modules (the tagging behaviour module,
the filtering module and the dialog manager
module).
4.1 Main Modules
User modelling module:
The user is an important entity in the adaptation
process. This module aims to represent each single
user in the social network. It defines information
needed to represent a user profile, extracted from the
DB user model.
In the literature, a user profile is constructed
either in a static way, by gathering information that
rarely changes like name, age, etc., or in a dynamic
way, by gathering information that frequently
changes. In this module, we consider both static and
dynamic information.
In the most classic adaptation system, a user
profile contains personal information like name, age,
etc. In a social context, a user has social connections
or relationships and interests which are represented
in his profile through a FAOF (Friend-Of-A-Friend)
vocabulary. FOAF is based on the RDF/XML
vocabulary. Usually, interest in FOAF file, specifies
a resource. An extension has been made with “e-
FOAF:interest” vocabulary which provide more
detailed vocabularies related to user interests.
In a social environment, many approaches
suggest to define a tag-based user profile. It can be
constructed in an explicit way, by analyzing the tag
defined by the user (Firan, et al., 2007); or in an
implicit way, by observing his tagging behaviour
(Carmagnola, et al., 2011) (Carmagnola, et al.,
2008) (Nauerz, et al., 2008) and enrich user profile
by neighbour tags (Kim, et al., 2011). But tags are
ambiguous and need to be filtered (by the filtering
module) for a better profile construction.
Although tags are important elements reflecting
user’s interests on a resource, they are not
represented in his FOAF profile. Associating tags in
a FOAF file is not existent in the literature as far as
we are considered. So we try to extend the FOAF
file, by adding a new attribute which specifies the
tag assigned by the user, in order to extract from the
FOAF file information especially tags so we will
ANEXTENDEDARCHITECTUREFORADAPTATIONOFSOCIALNAVIGATION
543

analyze user profile once in order to decrease the
execution time.
Social networking module:
This module exploits the user modelling by
analyzing the similarity between users to build
networks of similar users using same tags (similarity
between tags is deduced from WordNet) and access
the user’s profiles to build networks of friends
(Carmagnola et al., 2011). This module is able to
identify similar users with a similar tagging
behaviour (Nauerz, et al., 2008). Based on social
relation, it is able to send information such as most
popular users, friends, etc. for the adaptation
module.
Adaptation module:
This module usually treats three dimensions of
adaptation: content, presentation and/or navigation.
We are interested in the adaptation of navigation
because it’s a way to avoid the disorientation of user
(Farzan, R., 2009).
In a social recommendation context, adaptation
layer performs various recommendations and
adaptations such as navigation adaptation model,
content adaptation model, etc. (Nauerz, et al., 2008).
Adaptation module may just adapt the content to the
user and personalizes the presentation (Carmagnola,
et al., 2008).
This module takes in entry the filtered
folksonomy, the social network elements and the
content to achieve the task of adaptation by
recommending social information which include: i)
Resource recommendation: a recommending
technique which recommends resources according to
data present in the DB contents, tag and users needs;
and ii) Tag recommendation: a recommending
technique which recommends tags according to the
user’s tagging behaviour and needs.
Adapting more than an information leads to offer
the user more than a possibility to navigate through
social information.
4.2 Secondary Modules
Tagging behaviour module:
Contains information about users who annotate,
by means of plain keywords known as tags,
resources of various types (i.e.: photos, videos,
scientific papers, etc). The result of the collaborative
tagging practice is also known as folksonomy (De
Meo, et al., 2010). This tagging behaviour is usually
presented as a 3D matrix (Wang, et al., 2010) (Kim,
et al., 2010) which link tags (t), users (u) and
resources (r). This matrix is usually very hard to
analyze, due to the fact that tagging data are
generally sparse. To deal with this problem, we
create three matrixes, analogously to (Wang, et al.,
2010), each one represent a simplified view:
 User–Tag (UT): Element (u, t) equals the
number of resources that user u tagged with
tag t.
 Resource–Tag (RT): Element (r, t) equals the
number of users that tagged resources r with
tag t.
 User–resource (UR): Element (u, r) equals the
number of tags that user u assigned to
resources r.
Tagging behaviour module is able to extract the
user’s preference and interest (based on the
assumption that tagging expresses interest in a
resource), update then the user profile according to
the evolution of his behaviour.
Filtering module:
The Filtering module aims to decrease tag
ambiguity present in the tagging behaviour module
in order to provide a correct tag to the adaptation
module.
From the tagging behaviour module, analyzing
the folksonomy presents a challenge. In
(Carmagnola et al., 2008) tags are analyzed with the
support of WordNet to classify them in various
categories (i.e.: subjective tags, which reflect the
user’s point of view, or free tags, which are not
derived from the textual description of the cultural
event). (De Meo, et al., 2010) introduce the
“authoritative” tag (i.e. tags having a high
PageRank) to enhance recommendation.
“Authoritative” tags are exploited to refine user’s
query and so improve recommendation. These
works, do not consider the semantic ambiguity in
folksonomy.
In our architecture, this module tries to filter
noisy tags by means of different techniques. From
folksonomy, this filtering module firstly detects
personal tags which don’t reflect the content of the
resource tagged but a personal opinion (i.e.: like,
awesome, etc.) (Gupta, et al., 2010). Then, it detects
spam (fake tags that are generated in order to
enhance the visibility of some resources or to
confuse users) (Liu, et al., 2009) by means of
specific algorithms. Finally, this module performs a
semantic analyze from WordNet
dictionary to detect
synonyms, tags which belong to the same topic, etc.
The purpose of this last step is to generate tags
(already filtered) which are similar, so in a
recommendation context, we can recommend
resources to the user who is interested in a topic
described by means of different tags.
WEBIST2012-8thInternationalConferenceonWebInformationSystemsandTechnologies
544

Dialog Manager:
This module manages interactions between
users and the system.
5 CONCLUSIONS
The history-enriched information spaces considered
as tags assigned by a user are combined with the
collaborative filtering to adapt social navigation. The
originality of our architecture relies on analyzing
information like the tagging behaviour, social
environment (i.e. relationships, friends) and the
merging of different techniques and methods to
recommend useful information according to the
user’s behaviour and the user’s environment. We use
the social information and user’s needs to overcome
the problem of disorientation. This architecture tries
to overcome limits such no updating for the user
profile, as his preference changes; the semantic
analysis of tags and the detection of the
inappropriate ones.
Our work is in its first step. In perspectives we
try to develop the resource and tag recommendation
techniques and extend the FOAF profile. We try to
combine the wisdom of the administrator (metadata)
and the user expression (tags) to recommend
resources and evaluate our method in large
databases. For the user profile, we try to figure out
pertinent user’s interests by analysing his
<foaf:interest> and his tags which reflect an
interest.
REFERENCES
Brusilovsky, P., 1996. Methods and techniques of adaptive
hypermedia. In User Modeling and UserAdapted
Interaction, 6(2-3), p.87-129. Springer Press.
Brusilovsky, P., Casselc L. N., Delcambreb L., M., L.,
Foxd, E., A., Furutae, R., Garciaf, D., D., Shipman F.,
M., Yudelsona, M., 2010. Social navigation for
educational digital libraries. In RecSysTEL’10 1st
Workshop on Recommender Systems for Technology
Enhanced Learning. Procedia Computer Science.
Carmagnola, F., Cena, F., Console, L., Cortassa, O., Gena,
C., Goy, A., Torre, I., 2008. Tag-based user modeling
for social multi-device adaptive guides. In User
Modeling and User-Adapted Interaction. Kluwer
Academic Publishers.
Carmagnola, F., Cena, F., Console, L., Grillo, P., Perrero,
M., Simeoni, R., Vernero, F., 2011. Supporting
content discovery and organization in networks of
contents and users. In Multimedia Systems. Springer
Press
De Bra, P., Aerts, A., Berden, B., De Lange, B., Rousseau,
B., Santic, T., Smits, D., and Stash, N., 2003. AHA!
The adaptive hypermedia architecture. In
HYPERTEXT’03: Proceedings of the fourteenth ACM
conference on Hypertext and hypermedia, pages 81–
84. ACM Press.
Farzan, R., 2009: A Study of social navigation support
under different situational and personal factors.
Doctoral dissertation, Pittsburgh University.
Firan, C. S., Nejdl, W., Paiu, R., 2007. The benefit of
using tag-based profiles. In LA-Web’07, Proceedings
of the Latin American Web Congress pp. 32–41. IEEE
Computer Society.
Gupta, M., Li, R., Yin, Z., Han, J., 2010. Survey on social
tagging techniques. In SIGKDD Explorations 12(1):
58-72, ACM Press.
Kim, H.-N., Roczniak, A., Lévy, P., & Saddik, A., 2010.
Social media filtering based on collaborative tagging
in semantic space. In Multimedia Tools and
Applications, vol. 50, no. 1. Springer Press.
Kim, H.-N., Alkhaldi, A., El Saddik, A., and Jo, G.-S.,
2011. Collaborative user modeling with user-
generated tags for social recommender systems. In
Expert Systems with Applications 38, 8488–8496.
Pergamon Press
Liu, K., Fang, B. and Zhang, Y., 2009. Detecting Tag
Spam in Social Tagging Systems with Collaborative
Knowledge. In FSKD'09 Proceedings of the 6th
international conference on Fuzzy systems and
knowledge discovery. 427-431. Publisher IEEE Press
Piscataway.
Meo, P. D., Quattrone, G., and Ursino, D., 2010. A query
expansion and user profile enrichment approach to
improve the performance of recommender systems
operating on a folksonomy. In Proceedings of User
Model. User-Adapt. Interact., 41-86. Kluwer
Academic Publishers
Musto, C., Narducci, F., De Gemmis, M., Lops, P. &
Semeraro, G. 2009. STaR: a Social Tag Recommender
System. In ECML PKDD Discovery Challenge 2009,
215-227. CEUR Workshop Proceedings Publishers.
Nauerz, A., Pietschmann, S., & Pietzsch, R. 2008. Social
Recommendation and Adaptation in Web Portals. In
Proceedings of the Workshop on “Adaptation for the
Social Web” ASW. Springer Press.
Zhao, S., Du, N., Nauerz, A., Zhang, X., Yuan, Q. and R.
Fu, 2008. Improved recommendation based on
collaborative tagging behaviors. In Proceedings of the
13th international conference on Intelligent user
interfaces, pp. 413–416, ACM Press.
Zheng, N., Li, Q., 2011. A recommender system based on
tags and time information for social tagging systems.
In Expert systems with Applications 4575-4587.
Pergamon Press.
Wang, J., Clements, M., Yang, J., De Vries, A. P., &
Reinders, M. J. T., 2010. Personalization of tagging
systems. In Information Processing & Management,
pp. 58-70. Elsevier Ltd.
ANEXTENDEDARCHITECTUREFORADAPTATIONOFSOCIALNAVIGATION
545