F: A P2P ITS FOR RECOMMENDING ADDITIONAL LEARNING

CONTENTS BY MEANS OF FOLKSONOMIES

Marta Rey-L

opez

Conseller

ıa de Educaci

on e O.U., Xunta de Galicia, Spain

Fernando A. Mikic-Fonte, Ana M. Peleteiro, Juan C. Burguillo, Ana Bel

en Barrag

ans-Mart

ınez

Departamento de Enxe

ner

ıa Telem

atica, Universidade de Vigo, Vigo, Spain

Keywords:

Collaborative tagging, Folksonomies, Recommender, ITS, P2P, e-Learning, Additional contents.

Abstract:

Intelligent Tutoring Systems (ITSs) aim at providing personalized and adaptive tutoring to students by the in-

corporation of a student modeling component. Besides, Web 2.0 technologies have achieved great acceptance,

bringing new applications which empower multiuser collaboration, such as collaborative tagging systems. In

this paper, we present L

F, a peer-to-peer system which provides students with new additional learning ele-

ments related to the course they are following. This is achieved through the collaboration of several ITSs and

using folksonomies.

1 INTRODUCTION

In the last few years, Web 2.0 technologies have

achieved great acceptance among the users. In the

ﬁeld of Web 2.0 applications, collaborative tagging

is becoming a popular practice to annotate resources

on the Web (which has even reached e-learning ini-

tiatives). In such a scenario, both the users and the

contents have their own tag clouds. In the case of

users, the tag cloud is composed of the tags the user

has ever assigned, whereas for the contents, it is com-

posed of the tags users have ever assigned to it. In

both cases the weight of the tags, i.e., its importance,

is proportional to the number of times they have been

assigned (by the user or to the content, respectively).

From annotations provided by users, a new structure,

called folksonomy, arises. It shows the relationships

between the different tags in the system. Applying

folksonomies in the ﬁeld of e-learning —and extend-

ing collaboration from users to systems—, some in-

teresting scenarios will appear when multiple systems

exchange information in order to learn from their ex-

periences. This is particularly interesting in learning

and tutoring systems, to offer speciﬁc contents to each

student taking into account previous experiences.

Intelligent Tutoring Systems (ITSs) (Brusilovsky

et al., 1997) were designed as intelligent tutors based

on knowledge, to serve as a guide in the student learn-

ing process. They are programs that aim at improving

the learning process through personalization of con-

tents depending on the student’s skills and knowledge

about the topics that they are learning about. They

try to emulate the way in which a human tutor guides

his/her students throughout the learning process.

Concerning the use of folksonomies in ITSs, it

makes that one resource is described based on the tag-

ger’s experiences and beliefs (John and Seligmann,

2006). The cited proposal uses user’s tags and book-

marks to ﬁnd an expert user on a particular topic when

needed. In ours, the ITSs, based on the past experi-

ences, are the ones which act as experts when they

recommend additional learning content to users who

are having trouble on the topic.

In order for ITSs to collaborate with each other,

Peer-to-Peer (P2P) seems to be an appropriate tech-

nology, since it has been very popular and suc-

cessful for managing distributed databases. Peer-to-

peer (P2P) architectures scale and self-organize them-

selves in the presence of a highly variable population

of nodes, with network and computer failures, with-

out the need of a central server and the overhead of its

administration. The inherent characteristics of such

architectures are typically scalability, and increased

access to resources (Theotokis and Spinellis, 2004).

This paper presents L

F (Light Long-Life

Learning with Folksonomies), a P2P framework

135

Rey-López M., A. Mikic-Fonte F., M. Peleteiro A., C. Burguillo J. and Belén Barragáns-Martínez A. (2010).

L4F: A P2P ITS FOR RECOMMENDING ADDITIONAL LEARNING CONTENTS BY MEANS OF FOLKSONOMIES.

In Proceedings of the 12th International Conference on Enterprise Information Systems - Software Agents and Internet Computing, pages 135-138

DOI: 10.5220/0002871101350138

 SciTePress

model made of ITSs which collaborate with each

other to provide users with additional learning mate-

rial related to the course they are following. Those

ITSs propose this additional material taking into ac-

count the users’ interests and knowledge, as well as

the previous success or failure of other similar users

with the additional content. In order for the ITSs to

recommend this new material, L

F provides a reason-

ing algorithm based on folksonomies which are cre-

ated from the tags the users annotate the learning ele-

ments with.

In this paper, we describe the system (Sect. 2),

according to its P2P structure, its content and user

modeling and its decision capabilities. Finally, we

present the conclusions and expose our future lines

of research (Sect. 3).

2 SYSTEM DESCRIPTION

F is a framework model, where users can annotate

the learning elements of the system using tags. In

F, each user is associated with an ITS which stores

his/her proﬁle and provides the user with learning ma-

terial.

If the ITS detects that the user is having trou-

ble with a particular learning element (see Fig. 1),

it looks for additional related content in its own

database, and communicates with other ITSs using a

P2P model (Sect. 2.1) to obtain related material from

their databases. The recommendation of new learning

elements is based on two different criteria. On the one

hand, the system looks for contents which are similar

to the one the user is following (by comparing con-

tents’ tag clouds, see Sect. 2.3). On the other hand, it

searches for a learning element which has been suc-

cessful for similar users. For this to be possible, con-

tents have a new type of tag cloud: the target user tag

cloud, which is obtained from the tag clouds of the

users who have followed the learning element, taking

into account their success or failure. In this manner, to

compare users with target users, the method explained

in Sect. 2.3 is used.

The last step in the process of ﬁnding new ad-

ditional content is selecting the most appropriate

one from those the consulted ITSs have proposed

(Sect. 2.4).

2.1 Peer-to-Peer System Description

We consider our system, composed by multiple ITSs,

as a P2P system. We organize the whole system

in a similar way to Kazaa (Liang et al., 2004), i.e,

the peers are hierarchically distributed in two levels.

There are some ITSs whose agents behave as peer

leaders (a peer leader is dynamically selected taking

into account the number of times its learning elements

have been followed) who have a set of children (nor-

mal peers) linked to them. The peer leader has a sum-

mary of the tag clouds of the learning elements avail-

able in the repositories of its children, determining its

proﬁle.

Now we will describe the behavior of the system

considering a new peer ITS (IT S

new

) joining the P2P

system, to publish information, to search for a com-

mittee of peers, and ﬁnally to fetch the appropriate

learning elements to solve its problems:

• Join: A new ITS (IT S

new

) is set up and wants to

join the ITS P2P network. IT S

new

selects the peer

leader IT S

closer to its contents’ tag clouds and

connects with it as a normal peer (child).

• Publish: IT S

new

sends to IT S

a summary of

contents’ tag clouds for indexing purposes.

• Search: When IT S

new

has problems to ﬁnd an ad-

ditional learning element for a particular user, it

connects with its leader IT S

asking for a solu-

tion to its problem. Then, IT S

checks its own

repository and contacts with its children return-

ing to IT S

new

a committee of peers with similar

learning elements. If none of its children has any

tag clouds.

The simplest way to measure this value would be

to consider the number of coincident tags in both tag

clouds, i.e., the higher the number of coincident tags

and its weight, the higher the degree of relationship

between the two tag clouds.

But our algorithm does not only consider the num-

ber of coincident tags and their weights (direct rela-

tionship, R

) (Eq. (1)), but also the degree of

relationship between the tags of both tag clouds (one-

hop relationship, R

) (Eq. 2)).

) =

∑

∀i/t

∈|T

∩T

(−1)

|w(t

)| |w(t

(1)

) =

∑

∀i/t

∈T

∀ j/t

∈T

(−1)

|w(t

)| |w(t

)|·r

i j

(2)

where D

is the set of tags in the tag cloud of item

i, D

is the set of tags in the tag cloud of user u, w(t, D)

is the weight of tag t in the tag cloud D, and r

i j

is the

relationship in the folksonomy between the tags t

and

. In equations above, n = 1 i f w(t

)·w(t

) < 0

and n = 2 otherwise

In this manner, the total relationship takes into ac-

count both Eq. (1) and (2), and it is calculated

R(D

) = α R

) + (1 − α) R

) (3)

being α the parameter used to indicate the impor-

tance of each type of relationship (α ∈ [0,1]).

These equations can be used to measure three dif-

ferent degrees of relationships: i) R

is the one be-

tween the original content and the additional one; ii)

L4F: A P2P ITS FOR RECOMMENDING ADDITIONAL LEARNING CONTENTS BY MEANS OF FOLKSONOMIES

137

DOI

is the relationship between the user and the tar-

get user proﬁle, concerning his/her interest; and iii)

DOK

is the relationship between the user and the tar-

get user proﬁle, concerning his/her knowledge.

2.4 Final Decision

Each ITS of the committee of peers needs to deter-

mine which is the best learning element to fulﬁll the

learning needs of the user studying the original con-

tent. For this to be possible, it follows this algorithm:

1. First, it ﬁlters the contents according to R

, sort-

ing them according to this value and discarding

those ones which differ more than a magnitude or-

der with respect to the previous one.

2. Next, it measures R

DOI

and R

DOK

for the remain-

ing candidates and calculates R

(the relationship

regarding the user) as follows:

= |U

| · (βR

DOI

+ (1 − β)R

DOK

) (4)

being β a constant provided by the original ITS

which reﬂects the degree of importance the user

gives to his/her interests and knowledge, and |U

the number of users who have followed the learn-

ing element c

3. It selects that content with the highest R

and

sends R

and R

to the original ITS.

4. Finally, the original ITS selects the most appropri-

ate element from those received following steps

1–3, asks for it to the correspondent ITS and of-

fers it to the student.

3 CONCLUSIONS AND FUTURE

WORK

In this paper, we have presented L

F, a P2P system

composed of multiple ITSs which collaborate with

each other in order to provide users with additional

material for the courses they are following. While we

centered our description on the ITS domain, we con-

sider that the model described here can be extended

and useful, for recommending learning contents, to

the whole spectrum of learning systems.

For the recommendation of the aforementioned

material, the ITSs do not only measure the similar-

ity of the additional content with the original one, but

also the appropriateness of the material for the user.

This is accomplished by maintaining for each piece

of content a target user proﬁle, created from the in-

formation of success or failure of all the users which

have studied this element.

For the user and content model, we have selected

a solution which is being very successful in the ﬁeld

of Web 2.0: collaborative tagging and folksonomies.

Based on this solution, this paper presents as well a

reasoning algorithm which aims to help ITSs deter-

mine which is the most appropriate additional learn-

ing element for a given user and original learning ma-

terial. The main advantage of this solution is its ﬂexi-

bility and adaptability, since it changes in time reﬂect-

ing the opinion of current users.

In the ﬁeld of reasoning, future work will address

the improvement of the system ﬁnding a solution for

those users who are passive and do not collaborate

in tagging contents. This fact prevents the system to

obtain a user proﬁle for these users. A possible solu-

tion would be obtaining the user proﬁle from the tag

clouds of the contents he/she has followed.

REFERENCES

Brusilovsky, P., Nakabayashi, K., and Ritter, S. (1997). In-

telligent Educational Systems on the World Wide Web.

Workshop at the 8th World Conference on Artiﬁcial

Intelligence in Education.

John, A. and Seligmann, D. (2006). Collaborative tagging

and expertise in the enterprise. In Collaborative Web

Tagging Workshop held in Conjunction with WWW

2006, Boulder, Colorado (USA).

Liang, J., Kumar, R., and Ross, K. W. (2004). Understand-

ing KaZaA. http://cis.poly.edu/∼ross/papers/Unders-

tandingKaZaA.pdf.

Michlmayr, E., Cayzer, S., and Shabajee, P. (2007). Add-

A-Tag: Learning Adaptive User Proﬁles from Book-

mark Collections. In 1st International Conference on

Weblogs and Social Media (ICWSM’2007), Boulder,

Colorado (USA).

Rey-L

opez, M., D

ıaz-Redondo, R., Fern

andez-Vilas, A.,

and Pazos-Arias, J. (2010). Computational Intelli-

gence For Technology Enhanced Learning, chapter

Use of Folksonomies in the Creation of Learning Ex-

periences for Television. Springer-Verlag.

Theotokis, S. A. and Spinellis, D. (2004). A survey of peer-

to-peer content distribution technologies. ACM Com-

put. Surv., 36(4):335–371.

ICEIS 2010 - 12th International Conference on Enterprise Information Systems

138