Recommendation of Learning Resources based on Social Relations
Mohammed Tadlaoui
1,2
, Karim Sehaba
3
and Sébastien George
4
1
Université de Tlemcen, LRIT, Tlemcen, Algeria
2
INSA de Lyon, LIRIS, UMR5205, F-69676, Lyon, France
3
Université de Lyon, CNRS, Université Lyon 2, LIRIS, UMR5205, F-69676, Lyon, France
4
LUNAM Université, Université du Maine, EA 4023, LIUM, 72085, Le Mans, France
Keywords: Recommender Systems, Educational Resources, Social Networks, Social Influence.
Abstract: Recommender systems are able to estimate the interest for a user of a given resource from some information
about similar users and resources properties. In our work, we focus on the recommendations of educational
resources in the field of Technology Enhanced Learning (TEL) and more specifically the recommendations
which are based on social information. Based on the results of research in recommender systems and TEL,
we define an approach to recommend learning resources using social information present in social networks.
We have developed a formal model for the calculation of similarity between users and the generation of
three types of recommendation. We also developed a platform that implements our approach.
1 INTRODUCTION
Social media are increasingly used in education.
They are either integrated into an LMS or used in
standalone mode (Popescu, 2014). This type of
media focuses on the interactions and mutual
support between learners.
In social networks (Guy and Carmel, 2011),
users suffer from information overload due to the
multitude of resources and interactions related to the
multitude of social relationships. To address this
problem, we propose to recommend the users
relevant resources based on existing social
relationships.
Thus, the objective of our work is to propose an
approach that allows customizing educational
resources based on connections in social networks.
Our work is based on two principles identified by
research in social science:
the co-citation regularity (Bhagat et al., 2011)
that stipulates that similar individuals tend to
refer or to connect to the same resources;
social influence (Sun and Tang, 2011) indicates
that individuals tend to follow the behaviour of
their friends.
The first principle is used in classical
recommender systems. These systems are mainly
based on the evaluation of similar users for a given
user to predict her/his preferences. However, this
kind of recommendation systems ignores the social
influence connections between users. This type of
connections can be used to increase the accuracy and
relevance of recommendations.
According to the second principle, people who
are socially connected can share the same interests
or similar interests. So the users of a system can be
easily influenced by their friends and be interested in
their activities. This principle is used in social
recommender systems.
In recent years, a particular research area of
recommender systems has emerged. It concerns
recommender systems for Technology Enhanced
Learning (TEL). Drachsler (Drachsler, 2012)
explains that this type of systems uses the
experiences of a community of learners to help
learners of this community to more effectively
identify learning content or peers students from a set
of potentially very wide choices.
Several recommendation systems dedicated to
TEL were developed during the last decade. One of
the first systems is Altered Vista (Recker and
Walker, 2003). It collects assessments that users
attribute to educational resources and propagates
them in the form of "word-of-mouth"
recommendations on the qualities of resources.
RACOFI (Anderson et al., 2003) is a similar system;
it incorporates an inference engine based on rules.
LSRS (Huang et al., 2009) is a recommendation
system which is based on sequencing rules and the
analysis of learning groups. ReMashed (Drachsler et
425
Tadlaoui M., Sehaba K. and George S..
Recommendation of Learning Resources based on Social Relations.
DOI: 10.5220/0005452304250432
In Proceedings of the 7th International Conference on Computer Supported Education (CSEDU-2015), pages 425-432
ISBN: 978-989-758-108-3
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
al., 2009) asks learners to evaluate information from
an informal learning network. This system uses
these evaluations and tags associated with resources
to make the recommendation.
The existing recommendation systems mainly
use evaluations from users and their similarities to
propose them resources adapted to their needs.
However, they do not exploit the information
contained in the profiles of similar users. Thus, in
order to customize and make relevant the
recommendation, we propose to use, in addition to
the profile of the target user, the information present
in the friends’ profiles. It is mainly about the
characteristics of friends’ profiles, information on
resources visualization, their utilities relative to
learning fields and the results of exercises performed
by a learner. Based on the co-citation regularity and
the social influence theories, we hypothesize that the
use of this information can help to make the
recommendation richer, comprehensive and tailored
to the needs of the user. The research question that
we try to answer in this paper is “How to generate
more specific recommendations, and therefore more
relevant ones, basing on social relations and user
evaluations? ”.
Our approach is adapted to the traditional
acquisition which is based on the sharing of
resources. Nevertheless, being based on educational
social network, the proposed approach can provide
more interactions between users and so encourage
collaborative learning.
The characteristics of a user’s profile can be
incomplete, outdated or inappropriate. In our
approach, as the recommendation is based on user’s
characteristics, the quality of this recommendation is
relative to the quality of these characteristics. In
addition to the recommendation of learning
resources, we think that social relationships can also
help to enrich or correct the information stored in the
learner profile based on the information present in
the friends’ profiles.
The paper is organized as follows: the first part
presents the state of the art on recommendations in a
social context and in the context of TEL. The second
part presents the approach we propose to address the
problem of information overload. Our approach
identifies three types of recommendations of
educational resources (recently viewed resources,
popular resources and useful resources) and also a
type of recommendation to help users to complete
their profiles. The third part contains an illustration
of our approach and the last part presents a
conclusion and perspectives to our work.
2 RELATED WORK
Recommender systems have existed since the 90s
(Resnick and Varian, 1997). The most commonly
used methods in these systems are based on
collaborative filtering (Goldberg et al., 1992) or are
content-based systems (Pazzani and Billsus, 2007).
The first method recommends resources from the
similarity between user’s preferences. The second
method is based on the recommendation of resources
that are similar to resources for which the user has
expressed an interest in the past.
The algorithm of collaborative filtering was
extended to be more scalable for large user bases.
This extended method is called “item to item
collaborative filtering” and it is one of the most
widely deployed recommender methods today. For
instance, this method is used by Amazon (Linden et
al., 2003) for recommending products and LinkedIn
(Wu et al., 2014) for recommending peoples, jobs,
companies, groups, and other entities
recommendations.
Since the last decade, recommender systems
have increasingly used social information to
improve the quality and relevance of the
recommendation. Bellogína et al. (Bellogína et al.,
2013a), (Bellogína et al., 2013b) divide the social
recommender systems into four types:
1. Friend Based Recommender: this type of
system uses collaborative filtering method but just
by taking into account users that are explicitly
declared as friends by a user.
2. Social Popularity Recommender: in this type
of system, it is the most popular resources for the
friends of a user that are recommended to her/him.
3. Personal Social Recommender: systems that
are part of this type use the distance between users
in the social graph to make the recommendation.
The more users are far from a given user, the less the
weight of evaluations of their resources is important
in the formula for calculating the recommendation.
4. Hybrid Recommender: this type of system
can use several methods of recommendation to take
advantage of the benefits of each one.
Drachsler et al. (Drachsler et al., 2008) explain
that recommender systems used in the educational
field are different from those of other fields such as
e-commerce. This difference is due to the fact that
the objectives and user models in both types of
systems are not the same. In (Drachsler et al., 2013),
authors provide an analysis and comparison
framework between recommender systems for TEL.
These systems are classified according to several
categories: supported tasks, user model, domain
CSEDU2015-7thInternationalConferenceonComputerSupportedEducation
426
model, personalization, architecture, location and
recommendation mode. In our work, we consider the
first category that handles user tasks supported in a
recommender system for TEL. The tasks that can be
performed by the users of such systems are:
1. Find novel resources: recommendation of
new resources in particular resources never viewed
by the user.
2. Find interesting users: recommendation of
other users for which a particular user may be
interested, for example, offer an expert user in a
domain or provide a user with similar interests.
3. Find good pathways: recommendation of
learning path through educational resources, for
example, propose a list of possible paths for the
same resources to achieve a learning objective.
The approach that we propose fits into the first
type of social recommender systems (friend-based
recommender) and also in the first category of
recommender systems for TEL (find novel
resources).
3 SOCIAL RECOMMENDATION
APPROACH
Since the context of our work concerns the social
networks for learning, our approach (Tadlaoui,
2014) illustrated in Figure 1 is based on 1) data that
describes users and stored in their profiles, 2) data
on different types of links between users and
between the groups to which they belong and 3)
feedbacks on results of exercises made by learners.
All these data will be used to make
recommendations to learners.
Figure 2 illustrates the overall principle of our
approach. Each user of the system is described with
some information that characterizes her/his profile
and is linked with friendships links with other
system users. Users can view educational resources
and they can also evaluate the quality and usefulness
of the resources they have already seen.
In our approach, we define friendship as a link
explicitly declared by a user of the system. Two
types of friendship are supported, friends that the
user knows in real life and has declared as friends on
the social network and those she/he knows only on
the social network. In both cases, following the
principle of social influence, these friends may have
the same interests, and following the principle of co-
citation regularity, if these friends have some
similarities in common, so they will be more likely
to have the same interests.
Figure 1: Overall architecture of the proposed approach.
Figure 2: Overall principle of the proposed approach.
From the information that describes the users,
resources, and links between them, the system
generates lists of recommended resources for a given
user. These lists are customized for each user of the
system and are divided into three types, namely
recently viewed resources, popular resources and
useful resources.
The overall process of the recommendation
system involves the following steps:
1. Select a type of recommendation;
Visualization
Evaluation of the
quality and the utility
Friendship
link
Popular resources
Useful resources
Recently viewed resources
Recommend
ed resources list
Profile
data
Tests
data
Social
data
Recommender s
y
stem
Recommend
ed resources list
Recommended
resources lists
Profile update
RecommendationofLearningResourcesbasedonSocialRelations
427
2. Select users related to the current user by a
friendship link;
3. Calculate the degree of similarity between the
active user and her/his friends, following several
criteria (explained in section 5);
4. Calculate the score of resources based on actions
(visualization, evaluation and utility) performed
by the active user’s friends on resources;
5. Present the user with a list of resources ordered
by score according to the selected type of
recommendation.
4 FORMALIZATION
In this section we define the basic concepts for
calculating scores for the recommendation of
learning resources.
U is the set of all users of the system and B[u] =
{v ∈ U: v friend with u} is the set of friends of the
user u. IV
u
, IE
u
, IU
u
represent respectively the sets of
visualized resources, evaluated resources according
to their qualities and evaluated resources in terms of
utility by the user u.
D is the set of all teaching domains represented in
the system and E[u] = {d ∈ D: d is a domain of u}
represents areas of interest of the user u. For
example, if the system is used in a research
laboratory, domains can then be topics of research.
If it is a university, domains can be specialties or
course modules.
The user profile information are represented by
the set C. Formally, C[u] is the set consisting of
characteristic /value pairs that define the profile of
the user u. C[u] = { (c, val) | c ∈ C , val is the value
of the characteristic c of the user u} . The values of
continuous type may be replaced by discrete values.
For example, the value of age can be transformed
into child, adolescent, adult …
Visu(u, i) is used to know the resources that are
visualized by a given user. This function is equal to
1 if the user u viewed the resource i and 0 otherwise.
t(u, i) represents the number of days since the date of
the last visualization of the resource i by the user u.
Eval(u, i) is used to know the evaluation of the
quality of a resource by a user. For example, the user
u can evaluate the resource i from 1 to 5. If the
resource has not been evaluated then this function is
equal to 0. Eval(u, .) is the average rating of the user
u for all the resources that she/he has evaluated .
In addition to the evaluation of the quality of a
resource, a user can evaluate its utility according to a
specific domain. Util(u, i, d) represents the
evaluation of the usefulness of the resource i in the
context of work d(domain) of the user u. Util(u, .) is
the average evaluation of the user u of all the
resources that she/he has evaluated .
Seval is the set of co-evaluated resources in
terms of quality by users u and v: Seval = IEu
IEv.
Sutil represents the co-evaluated resources in terms
of usefulness by users u and v: Sutil = IUu
IUv.
5 SOCIAL SIMILARITY
The majority of works on recommendation use the
Pearson correlation coefficient for calculating the
similarity between two users of a system. These
works are mainly interested in resource evaluation to
calculate the similarity between users. Because our
work takes place in the context of social networks
for learning, we propose a new method for
calculating the similarity which is based on 1) the
similarity of the choice of users’ visualizations and
evaluations, 2) strength of the link between these
users and 3) the similarity between users’ profiles.
This new formula, denoted SocialSim(u, v),
represents the social similarity between users u and
v:
We used the Pearson correlation coefficient to
calculate the similarity in terms of evaluation
EvalSim(u, v) and we adapted this coefficient to
SocialSim(u,v) = (EvalSim(u,v) + UtilSim(u,v) + VisuSim(u,v) + LinkS(u,v) + ProfilSim(u,v)) / 5
(1)
(
,
)
=
∑
(
(
,
)
−
∈
(
,.
)
).(
(
,
)
−
(
,.
)
)
∑
(
(
,
)
−
∈
(
,.
)
)
.
∑
(
(
,
)
−
∈
(
,.
)
)
(2)
(
,
)
=
∑
∑
(
(
,,
)
−
∈
(
,.,
)
).(
(
,,
)
−
(
,.,
)
)
∑
(
(
,,
)
−
∈
(
,.,
)
)
.
∑
(
(
,,
)
−
∈
(
,.,
)
)
∈
[
]
∩
[
]
([]∩[])
(3)
CSEDU2015-7thInternationalConferenceonComputerSupportedEducation
428
calculate the similarity in terms of utility UtilSim
(u, v).
The formula for the similarity in terms of
visualization is defined as follows:
(
,
)
=
(
∩
)
(
∪
)
(4)
If both users have previewed no resources then
the union is null. So in this case, we do not apply the
rule and the visualization similarity is equal to 0.
In social networks (Sun and Tang, 2011), the link
between two users is stronger if they have in
common multiple neighbors. In our work, the
strength of the relationship between two users is
defined using the number of common friends and
their total number of friends. The strength of the link
between users u and v is defined by the following
formula:
(
,
)
=
(
[
]
∩
[
])
+2
(
[
]
∪
[
])
(5)
The last element that we have integrated into the
formula of social similarity is the similarity related
to characteristics present in users profiles. It will
take into account the similarities between users in
terms of preferences, knowledge, goals… This
similarity is related to the number of common
characteristics between the two users and the total
number of characteristics. The formula that
calculates such similarity between users u and v is:
(
,
)
=
(
[]∩[]
)
(
)
(6)
6 RECOMMENDATION TYPES
6.1 Recommendation of Recently
Viewed Resources
The system can provide to a user a list of resources
that have been recently viewed by similar users. This
type of recommendation is useful for collaborative
learning. Indeed, it allows users to follow courses at
the same time as their friends to be able to
collaborate and help each other on these different
courses. Recommendation score of the resource i for
the user u is determined by the following formula:
(
,
)
=
(
,
)
.
(
,
)
.
(
,
)
∈
[
]
(7)
α is a decay factor. More a user has viewed a
resource recently, the higher the score of this
resource increases. We were inspired by the work of
Guy et al. (Guy et al., 2009) and (Guy et al., 2010)
to take into account the time in this formula.
6.2 Recommendation of Popular
Resources
A user can also see a list of recommended resources
highly rated by her/his friends. Recommendation
score of resource i for user u is determined by the
formula:
(
,
)
=k
(
,
)
.
(
,
)
∈
[
]
(8)
Only friends of user u who have evaluated item i
can be used in the calculation of this score. This
subset of B[u] is denoted by B
i
[u]. k represents the
normalizing factor. It is usually given as k=
1
∑
|SocialSim
(
u,v
)
∈
[
]
|
⁄
The Seval formula uses the weighted sum
approach (Adomavicius andTuzhilin, 2005).
6.3 Recommendation of Useful
Resources
A list of resources can be recommended to a user
based on their utilities according to learning domains
of this user. Recommendation score of resource i for
user u is determined by the formula:
(
,
)
=k
(
,
)
.
∈
[
]
∑
(
,,
)
∈
[
]
(
[
]
)
(9)
Only domains of interest of the user u which
have been used to evaluate resource i by user v can
be integrated in the calculation of this score. This
subset of E[u] is denoted by E
vi
[u].
7 PRINCIPLE OF THE PROFILE
UPDATE
The characteristics of a user can be partially filled,
even outdated or inappropriate. As our approach is
based in part on the characteristics of the user, it is
necessary to update the user profile to be able to
make a high quality recommendation.
The profile data can be filled manually by the
RecommendationofLearningResourcesbasedonSocialRelations
429
user or/and set by the system based on her/his
behavior. In addition to the two methods previously
explained, the principle of social influence (Sun and
Tang, 2011) leads us to enrich the user's profile with
the information that characterizes the profiles of
her/his influential friends, this by using one or the
both of the following methods:
Recommend to the user the most common
characteristics present in the profiles of her/his
friends. For example, the studies level the most
common among her/his friends. The system
will update the user profile if she/he validates
this recommendation;
Update directly the user's profile with the
values of the most common friends’
characteristics. This information may be exact
or may be uncertain, so the system associates
to them probabilities of accuracy (degree of
reliability) and stores them in the user's profile.
8 EVALUATION
To evaluate our approach, we have wanted first to
use datasets extracted from existing educational
recommender systems. Among these datasets, we
can mention Mendeley (Jack et al. 2010), MACE
(Wolpers and Niemann 2010) and APOSDLE
(Beham et al. 2010). After studying this type of
datasets, we found it impossible to use them to
evaluate our model. The data they provide do not
contain all the data we need to conduct the
evaluation, such as social relationships between
users and resource evaluations in terms of utility.
We have considered the possibility to complete
these datasets with missing information but the
whole new modified dataset can be incoherent and it
can make our simulation wrong.
To address this problem, we established a process
for studying the feasibility of our approach, evaluate
that it proposes relevant resources and compare it
with other existing approaches. 1) The first step was
to create a reduced dataset consisting of 10 users and
9 resources and evaluate the approach with it, 2) The
second step was to develop a learning platform that
implements our approach and test it with true users,
3) In parallel with this step we are trying to have a
dataset from an existing platform named ACCEL
(Delache et al., 2007) to run on it the evaluation.
8.1 Simulation on a Created Dataset
For the first step of the evaluation we have made a
simulation on a dataset that we have created. This
evaluation helped us, to test the algorithms related to
our formulas to evaluate their effectiveness and to
refine them. We have developed a prototype in Java
that calculates the similarities between users and
calculates and displays the three lists of
recommendation that we propose in our approach.
The dataset that we have created contains all the
information that our model needs. It uses the
information about 10 users and 9 resources. This
dataset contains mainly 1) user characteristics such
as age, preferences ... 2) social relationships between
users and 3) evaluation values that users attribute to
educational resources.
8.2 Design and Test of the Platform of
Icraa
To evaluate our approach with real users, we have
developed a learning platform, named icraa (Icraa is
a soCial leaRning And Authoring environment),
which implements our formal models to recommend
educational resources.
The effectiveness of our approach is measured by
the users’ evaluations on the recommendations that
are proposed by the system.
The learning platform is currently used by 10
teachers from the University of Tlemcen (Algeria).
The evaluation is conducted on 3 classes of 25, 28
and 40 students. The teachers are asked to upload
educational resources related to their courses and we
estimate that we will have more than 300 learning
resources by the end of April 2015.
8.2.1 Platform Functionalities
Resources Upload: teachers who have rights can
upload the resources of their courses into the
platform and describe them by some metadata.
Resources Access: All resources are accessible
to all system users (students and teachers).
Resource Download: all users can download all
resources that have been added by their teachers.
Resource Evaluation: a user of the platform can
evaluate the quality of a resource and its utility
according to the user’s domains. This functionality
is illustrated in figure 3.
Resource Recommendation: the system
provides the three types of recommendation of our
approach namely recently viewed resources, popular
resources and utile resources. When a user chooses
one of these three types, the system displays a list of
the 3 best rated recommended resources.
Social features: Icraa platform provides multiple
social features that can be found in social networks
as Facebook.
CSEDU2015-7thInternationalConferenceonComputerSupportedEducation
430
8.2.2 First Results
The platform was installed at the end of November
2014 and after 4 weeks of its usage 4 teachers
started to upload their courses on the platform. They
uploaded about 50 resources.
We are noticing that users have created about 340
friend relationships and there were 400 resources
visualizations. Users were connecting frequently
into the platform; there was an average of 33 users’
connections by day. The resources of the system
were visualized 875 times and there were 305 of
resources evaluations by users.
As first results of this experimentation we noticed
that 80% of the users found that the recommended
resources were relevant. Also 82% of the users
found the given recommendation useful.
The evaluation will be continued in the second
semester of this academic year and it will finish by
the end of April 2015.
8.2.3 Ongoing Evaluation
The first results that we have are interesting but not
sufficient. Currently, we are working on the next
step of the evaluation with icraa platform. We
divided the set of the users of this system into 3
groups:
G1: students of this group have recommend-
dations that respect our approach;
G2: students of this group have recommend-
dation that respect the algorithm of Friend
Based Recommender System (explained in
section 2);
G3: students of this control group have recom-
mendation in a random order.
Once the results collected, it is necessary that the
group G1 will be the most satisfied with the
recommendations provided by the system and G2
will be more satisfied than G3.
Figure 3: Interface for evaluating the quality and the utility
of a resource.
8.3 Simulation on ACCEL Dataset
ACCEL is a distance learning platform developed
and used by the University of Lille (France). It is an
acronym of « Apprentissage Collaboratif et
Communauté En Ligne » which means
Collaborative Learning and Online Community.
ACCEL is used in life learning context with
students which have different backgrounds when
icraa is used with young students in university. All
courses that are followed by students who use
ACCEL are 100 % in distance learning but students
who use icraa in the University of Tlemcen follow
courses in blended learning mode.
The dataset extracted from the ACCEL platform
can help us to evaluate our approach. It can give us a
complementary evaluation compared to the icraa
platform since the context of use and the type of
users are not the same between these two systems.
The problem with this platform is that it does not
contain the functionality of evaluating resources and
the functionality of declaring friend relationships.
We are working with the ACCEL team to
incorporate these functionalities on their platform.
All their users will use these new functionalities and
after some weeks of use we will extract a dataset
which contains information that we need for our
evaluation.
9 CONCLUSIONS
In this paper we have presented an approach to
recommend educational resources based on social
relationships. We have developed a formal model
for the calculation of similarity between users and
the generation of three types of recommendation of
educational resources. We also presented an
illustration and evaluation that we have followed to
test, refine and validate our approach.
The platform icraa that we developed allowed us
to have positive first results on the evaluation of our
approach. More than 80% of users are satisfied with
the recommendations made by the system. With the
help of this platform, we continue in the next months
to do a comparative evaluation between our
approach and other recommendation approaches.
Our recommendation approach is based on
collaborative filtering that uses evaluations of users.
This approach can be enriched using a hybrid
recommendation method that also uses a
recommendation based on the content. Another
perspective of our work can be the use of social
information (profiles, relationships, affiliations ...)
RecommendationofLearningResourcesbasedonSocialRelations
431
present on public social networks such as Facebook
or LinkedIn. This will help us to improve and enrich
social information in the recommendation system.
REFERENCES
Adomavicius, G., Tuzhilin, A., 2005. Toward the next
generation of recommender systems: A survey of the
state-of-the-art and possible extensions. Knowledge
and Data Engineering, 17(6): 734-749.
Anderson, M., Ball, M., Boley, H., Greene, S., Howse, N.,
McGrath, S., and Lemire, D., 2003. Racofi: A rule-
applying collaborative filtering system. In IEEE/WIC
COLA 2003.
Beham, G., Stern, H., and Lindstaedt, S. N., 2010.
APOSDLE-DS A Dataset from the APOSDLE Work
integrated Learning System. In 1st Workshop on
Recommender Systems for TEL.
Bellogína, A., Cantadora, I., Castellsa, P., and Díeza, F.,
2013. Exploiting Social Networks in
Recommendation: Multi-Domain Comparison. In DIR
2013.
Bellogína, A., Cantador, I., Díez, F., Castells, P., and
Chavarriaga, E., 2013. An empirical comparison of
social, collaborative filtering, and hybrid
recommenders. ACM TIST, 4(1): 14.
Bhagat, S., Cormode, G., and Muthukrishnan, S., 2011.
Node classification in social networks. In Social
Network Data Analytics Springer US. 115-148.
Delache, D., D’Halluin C., Fichez, E., et al., 2007.
Environnements numériques et pratiques
collaboratives d’apprentissage.
Drachsler, H., 2012. Recommender systems for learning.
http://fr.slideshare.net/Drachsler/recsystel-lecture-at-
advanced-siks-course-nl. Visited 28/11/2014.
Drachsler, H., Hummel, H., and Koper, R., 2008.
Identifying the Goal, User model and Conditions of
Recommender Systems for Formal and Informal
Learning. Journal of Digital Information, 10(2): 4-24.
Drachsler, H., Pecceu, D., Arts, T., Hutten, E., Rutledge,
L., Van Rosmalen, P., Hummel, H., and Koper, R.,
2009. ReMashed–recommendations for mash-up
personal learning environments. In Learning in the
synergy of multiple disciplines. Springer 788-793.
Drachsler, H., Verbert, K., and Duval, E., 2013.
Recommender systems for learning. Springer.
Goldberg, D., Nichols, D., Oki, B. M., and Terry, D.,
1992. Using Collaborative Filtering to Weave an
Information Tapestry. Communications of the ACM.
Guy, I., and Carmel, D., 2011. Social recommender
systems. In 20th international conference companion
on World wide web. ACM. 283-284.
Guy, I., Zwerdling, N., Carmel, D., Ronen, I., Uziel, E.,
Yogev, S., and Ofek-Koifman, S., 2009. Personalized
recommendation of social software items based on
social relations. In Proceedings of the third ACM
conference on Recommender systems. ACM. 53-60.
Guy, I., Zwerdling, N., Ronen, I., Carmel, D., and Uziel,
E., 2010. Social media recommendation based on
people and tags. In Proceedings of the 33rd
international ACM SIGIR conference on Research and
development in information retrieval. ACM. 194-201.
Huang, Y. M., Huang, T. C., Wang, K. T., Hwang, W. Y.,
2009. A Markov-based Recommendation Model for
Exploring the Transfer of Learning on the Web.
Educational Technology & Society. 12(2): 144-162.
Jack, K., Hammerton, J., Harvey, D., Hoyt, J., Reichelt, J.,
and Henning, V., 2010. Mendeleys reply to the datatel
challenge. Procedia Computer Science, 1(2): 1-3.
Linden, G., Smith, B. and York, J., 2003. Amazon. com
recommendations: Item-to-item collaborative filtering.
Internet Computing, IEEE, 7(1), 76-80.
Pazzani, M.J., and Billsus D., 2007. Content-based
recommendation systems. In The adaptive web.
Springer Berlin Heidelberg. 325-341.
Popescu, E., 2014. Providing collaborative learning
support with social media in an integrated
environment. World Wide Web. 17(2): 199-212.
Recker, M. M., and Walker, A., 2003. Supporting “Word-
of-Mouth” Social Networks through Collaborative
Information Filtering. Journal of Interactive Learning
Research. 14(1): 79–99.
Resnick, P., and Varian, H. R., 1997. Recommender
systems. Communications of the ACM. 40(3): 56-58.
Sun, J., and Tang, J., 2011. A survey of models and
algorithms for social influence analysis. In Social
network data analytics. Springer US. 177-214.
Tadlaoui, M., 2014. Recommandation de ressources
pédagogiques basée sur les relations sociales.
RJCEIAH 2014.
Wolpers, M. and Niemann, K., 2010. dataTEL challenge:
CAM for MACE. In 1st Workshop on Recommender
Systems for Technology Enhanced Learning.
Wu, L., Shah, S., Choi, S., Tiwari, M. and Posse, C., 2014,
The Browsemaps: Collaborative Filtering at LinkedIn.
In 6th Workshop on Recommender Systems and the
Social Web, collocated with ACM RecSys 2014.
CSEDU2015-7thInternationalConferenceonComputerSupportedEducation
432
