MOOCs Recommender System using Ontology and Memory-based
Collaborative Filtering
Kahina Rabahallah
1
, Latifa Mahdaoui
1
and Faiçal Azouaou
2
1
Department of Computer Science, USTHB University, Algiers, Algeria
2
Higher National School of Computer Science, ESI, Algiers, Algeria
Keywords: MOOCs, Personalized Recommender System, Ontology, Item-based Approach, User-based Approach,
Cold-Start Problem.
Abstract: With Massive Open Online Courses (MOOCs) proliferation, online learners are exposed to various
challenges. Therefore, the lack of personalized recommendation of MOOCs can drive learners to choose
irrelevant MOOCs and then lose their motivation and surrender the learning process. Recommender System
(RS) plays an important role in assisting learners to find appropriate MOOCs to improve learners’
engagements and their satisfaction/completion rates. In this paper, we propose a MOOCs recommender
system combining memory-based Collaborative Filtering (CF) techniques and ontology to recommend
personalized MOOCs to online learners. In our recommendation approach, Ontology is used to provide a
semantic description of learner and MOOC which will be incorporated into the recommendation process to
improve the personalization of learner recommendations whereas CF computes predictions and generates
recommendation. Furthermore, our hybrid approach can relieve the cold-start problem by making use of
ontological knowledge before the initial data to work on are available in the recommender system.
1 INTRODUCTION
Massive Open Online Courses (MOOCs) are a
current trend in the field of e-learning and it attracts
millions of learners, to be engaged to enjoy massive
free open education courses. Prior studies indicate
that MOOCs are to date suffering from low
completion rate and Dropout problem (Goldberg et
al, 2015; Murphy et al, 2016; Xing et al, 2016;
Dhorne et al, 2017). With proliferation of MOOC
development in recent years, MOOC platforms (e.g.
Coursera
1
, Udacity
2
, edX
3
, etc) have more than
millions of learners and online courses. How to find
the most suitable MOOC among all proposed within
the platform? how can MOOC be efficient to
address the needs of its learners? are critical
challenges. In this context, we assume that more
personalized MOOCs’ recommendation can answer
these questions (Pang et al, 2017; Piao et al, 2016).
Previous studies have shown that E-learning
Recommender System (ERS) is considered an
1
https://www.coursera.org
2
http://www.udacity.com/
3
https://www.edx.org
effective key solution to overcome the information
overload. The Course Recommendation System is a
system that recommends the best combination of
subjects wherein the learners are interested (Aher et
al, 2013). In conventional recommender systems like
CF, Content-Based (CB), the recommendation
process are based merely on ratings. However,
literature reviews have shown that these
recommenders suffer from cold-start problem
(Bobadilla et al, 2012; Sun et al, 2017) and do not
consider the additional information about user and
items in making recommendations(Adomavicius et
al, 2011).
The initial insufficiency of ratings or preferences
leads to the occurrence of the cold start problem,
hence it becomes difficult to provide reliable
recommendations. Generally, the cold start problem
is triggered by three factors: new community, new
item and new users (Schafer et al, 2007; Sun et al,
2017). Moreover, in the context of e-learning,
learners have different characteristics like
knowledge level, which influence personalization of
recommendations. These additional learner
characteristics need to be incorporated into the
recommendation process (Verbert et al, 2012; Tarus
Rabahallah, K., Mahdaoui, L. and Azouaou, F.
MOOCs Recommender System using Ontology and Memory-based Collaborative Filtering.
DOI: 10.5220/0006786006350641
In Proceedings of the 20th International Conference on Enterprise Information Systems (ICEIS 2018), pages 635-641
ISBN: 978-989-758-298-1
Copyright
c
2019 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
635
et al, 2017). Our approach uses ontological
knowledge to address these major problems in
MOOCs recommendation process.
Traditional course recommendation systems are
integrated in closed e-learning environments (Aher
et al, 2013), in this paper, we propose a MOOCs
recommender system combining CF and ontology to
recommend personalized MOOCs to online learners
which can be better applied to a course
recommendation in MOOC platform. Our
contributions in this paper are summarized as
follow:
Combining item-based and user-based
approaches, and use ontological knowledge to
address the cold-start problem;
We propose a hybrid approach which uses the
ontological knowledge to integrate the
characteristics of learner and MOOC in
computing similarities and generating
recommendations for the learners.
The remainder of this paper is organized as
follows. Section 2 gives a brief review on
recommendation techniques relevant to this work,
and discusses related work about MOOCs
recommender systems. Section 3 presents our hybrid
recommendation technique. Finally, Section 4
concludes the paper.
2 RELATED WORK
This section gives a brief review on recommendation
techniques relevant to this work, and discusses
related work about MOOCs recommender systems.
2.1 Recommendation Technique
Several techniques have been proposed and used for
recommendation generation.
2.1.1 Collaborative Filtering (CF)
The most successful and widely used
recommendation technique is Collaborative Filtering
(CF), which is founded on the basic assumption that
users who have shown similar interests in the past
will share common interests in the future (Goldberg
et al, 1992). Collaborative filtering algorithms are
located under two different categories, namely
Memory-Based and Model-Based approaches.
Memory-Based provides recommendations based on
the similarities between users (or items) and ratings,
to calculate predicted values. Two subcategories of
memory-based CF are user-based and item-based CF
approaches (Lü et al, 2012; Ghazarian et al, 2015)
which we briefly explain below.
User-based: As its name suggests, it focuses on
the user. It consists in Looking for users most
similar to the target user, known as neighbour
users, through resorting to similarity measures.
After that, the unknown ratings are predicted
based on the users’ similarities values and the
ratings given to items by the similar users
(Koohi et al, 2017; Wang et al, 2006);
Item-based: Focuses on item instead of user.
It uses to find k-most similar items to items that
the target user has rated and then items’
similarities values and the ratings of target user
in the similar items are used to predict
unknown items’ rating (Sarwar et al, 2001).
Traditional similarity measures such as Pearson
Correlation Coefficient (PCC), COsine Similarity
(COS) and their variants, have been widely used in
CFs to calculate similarity (Desrosiers et al, 2011).
2.1.2 Ontology-based (OB)
Ontology is originally defined by Gruber (1992) as
an “explicit specification of a conceptualization”.
Modeling the information at the semantic level is
one of the main goals of using ontologies (Guarino
et al, 2009). We use ontology instead database to
provide a semantically rich description, which will
allow automatic processing. Different ontology
representation languages like Web Ontology
Language (OWL) are used to create ontology.
Moreover, the logical model allows the use of a
reasoner which can infer facts about given
situations.
In context of e-learning recommender systems,
ontology is used to model knowledge about the
learner (user context), knowledge about the learning
resource, and the domain knowledge (Yu et al,
2007) to take them into consideration in
recommendation process. Ontology-based are
knowledge-based which is developed to deal with
the cold start problem (Sun et al, 2017) since their
recommendations make use of the ontological
knowledge. We use also ontological knowledge to
allow deducting additional information about the
current context of learners and therefore,
personalizing the recommendation of annotated
MOOCs.
2.1.3 Hybrid Approaches
Combine two or more approaches, e.g. ontology-
based and collaborative Filtering, to gain better
ICEIS 2018 - 20th International Conference on Enterprise Information Systems
636
performance with less of the drawbacks of any
single solution (Burke, 2007).
2.2 Recommender Systems for MOOCs
Commonly, the personalized Recommender System
is usually divided into three main basic components:
The first one is about the Recommendation
Technique, the second one is about the
recommending Item and the last one is about
Personalization.
Figure 1: Basic components of personalized RS.
Recommendation system is widely becoming
popular on online learning. For instance, Pang et al.
(2017) propose an improved CF named as Multi-
Layer Bucketing Recommendation (MLBR) to
recommend courses on MOOC platform. At the
same time, MLBR fixes data sparsity and cutting
down the time cost on recommendation including
offline similarity calculation, online similarity
research, and update of similarity. Furthermore,
they extend MLBR with map-reduce technique to
improve the efficiency.
Bousbahi et al. (2015) propose a Recommender
System (MOOC-Rec) using the Case Based
Reasoning (CBR) approach and a special retrieval
information technique to recommend the most
appropriate MOOCs fitting her/his request based on
learner profile, needs and knowledge.
Piao et al. (2016) investigated three different
users modelling strategies based on the collected
LinkedIn dataset, for personalized MOOC
recommendations in a cold start situation. Results
showed that the Skill-based user modelling strategy
performs better than the Job and Edu-based ones.
Tang et al. (2017) proposed a personalized
behaviour recommendation in a MOOC to predict
the behaviour. They stipulate that this approach
touches on factors more aligned with
personalization, since the prediction of behaviour is
an aggregation of the student’s cognitive abilities,
affective state, and preferences. They investigated
the suitability of this behavioural prediction
approach by applying it to an expanded set of 13 UC
Berkeley MOOCs run on the edX platform.
Zhang et al. (2017) propose the course
recommendation model-oriented MOOC platform,
MCRS, which greatly improves the data storage
level and efficiency of calculation. The experiments
are carried out on Hadoop and Spark and the results
shows that MCRS is more efficient than traditional
Apriori algorithm and Apriori algorithm based on
Hadoop.
2.3 Discussion
Previous related works on personalized MOOCs RS
show that in the personalization process, the authors
focus on learner behaviors, on collected details
about learners by relying on their LinkedIn profiles,
rather than learner’s cognitive abilities. Unlike
previously mentioned approaches, our method
incorporates learner’s cognitive abilities in the
recommendation process to make personalization,
and it attempts to solve the cold-start problem by
using ontological knowledge in computing
similarities. Therefore, we use these similarities
values alongside ratings in computing predictions to
improve the recommendation accuracy.
The novelty of our approach focuses specifically
on ontology-based recommender system within
MOOC platforms of which to the best of our
knowledge, no study has been conducted to describe
the learner and MOOC using ontology, and integrate
these semantic descriptions into the recommendation
process.
3 PROPOSED METHOD
By inspiring from (Tarus et al, 2017), our method’s
focus is on prediction process, solving the cold start
problem and to improve the personalization of
learner recommendations. Indeed, we hypothesis
that ontological knowledge may be the bridge that
overcomes the limitations of cold-start problem and
the absence of integration the learner characteristics
(learner’s cognitive abilities) into the
recommendation process. Our approach involves
three steps, which are shown in Figure 2 and are
introduced in following subsections: (1) creating
ontologies to represent knowledge about the learner
and MOOC, (2) computing similarities based on
ontological knowledge and ratings as well as
predictions for the target learner, (3) generation
of top N MOOCs by the collaborative filtering
recommendation engine .
MOOCs Recommender System using Ontology and Memory-based Collaborative Filtering
637
Figure 2: Hybrid recommendation model.
3.1 Semantic Representation of MOOC
and Learner
To provide personalization in MOOCs
recommendations, several approaches have been
proposed: Based on LinkedIn profiles (Piao et
al,2016), base on learner behaviors (Tang et al,
2017; pang et al, 2017), etc.
In this paper, we focus on learner’s cognitive
abilities, we propose a semantic representation of the
learner profile inspired by (Rabahallah et al, 2016;
Labib et al, 2017). It is based on two type of
information which is acquired both explicitly and
implicitly (see Figure 3).
Static Information, including personal data
(name, gender, age, username and password,
type of learner which may be student;
professionals; etc.), language, Format;
Dynamic information, including different
dimension such as “Interest”, “prerequisite”
and “knowledge level” about a specific
domain, “specialty”.
On the other hand, with the light of the MOOC
platforms (e.g. Coursera, Udacity, edX, Udemy,
FUN, etc.) we used all the information collected
about MOOC for semantic description (see Figure
4). MOOC ontology contain more specific
information such as MOOC’s name, specialty,
domain, language, the start and the end of the
MOOC’s Session, the prerequisites required to
access, knowledge level which may be: beginner,
intermediate and advanced, university who created it
and the learning activities will propose by the
MOOC.
Figure 3: The learner profile conceptual model.
Figure 4: the MOOC profile conceptual model.
3.2 Computing Similarities and
Predictions
In order to predict unknown MOOCs’ ratings for the
target learner and solving the cold start problem, our
proposed method does the following steps:
ICEIS 2018 - 20th International Conference on Enterprise Information Systems
638
Step 1: It computes similarities between pairs
of MOOCs Sim(
) looking for the k
most similar MOOCs and between the pairs of
Learners Sim (
) to find the k most
similar learners, using both ratings and
ontological knowledge;
Step 2: It predicts learner’ unknown ratings on
MOOCs based on similarities values and the
ratings given to k most similar MOOCs by the
target learner.
3.2.1 Computing MOOCs’ Similarity
To calculate the similarity and make the predictions
for the target learner, the recommendation engine
will use both the ontological knowledge and the
ratings given by the target learner and other learners
on MOOCs. In computing the similarities, some
traditional similarity measures have been widely
used in CFs (Patra et al, 2015; Ahn et al, 2008).
However, cosine similarity (COS) among items does
not consider the dierence in each user’s use of the
rating scale. To address this problem, an adjusted
cosine similarity (ACOS) is presented (Sarwar et al,
2001).
Figure 5 indicates Learner-MOOC rating matrix.
Let L denote the set of all learners L = {l
1
, l
2
, l
3
, ...,
l
s
}, let M be the set of all possible MOOCs M= {m
1
,
m
2
, m
3
,......m
n
}, and Let R be the rating function that
measures the usefulness of MOOC m
i
to learner l
v
.
The possible rating values are defined on a
numerical scale from 1 (very irrelevant) to 5 (very
relevant).
Figure 5: Learner-MOOC rating Matrix.
In this paper, we use an extension of Adjusted
Cosine Similarity proposed in (Tarus et al, 2017) to
compute the similarities of MOOCs. In the proposed
extension, ontological information is utilized in
computing the mean rating
. Formally, the
similarity Sim(
) between two MOOCs m
i
and m
j
is calculated as follows (eq. 1):
Sim(
)=
(1)
Where
denotes the rating given to MOOC
m
i
by learner l
v,
, i and j integers {1, 2,...n}, i ≠ j,
is the mean rating of all the ratings provided by
learner l
v
based on ontological knowledge. Unlike in
pure CF, ontological information is utilized in
computing the mean rating
(Tarus et al, 2017).
3.2.2 Learners’ Similarity Computation
Experimental analyses show that Pearson
Correlation Coefficient (PCC) has outperformed
other similarity measures in user-based CF
(Aggarwal, 2016; Jannach et al, 2011). For this, we
use PCC to calculate ontological similarity
Sim(
) between two Learners l
v
and l
u
. The
PCC between users v and u can be measured through
(eq. 2):
Sim(
)=
Where
denote the rating given to m
i
by the
learner l
v
,
is the mean rating of all the ratings
provided by learner l
v
based on ontological
knowledge, v and u integers {1,2,...s}, u ≠ v.
3.2.3 Making Prediction
Once we obtain the set of k most similar MOOCs (k
nearest neighbors) and the k most similar learners
using respectively the Adjusted Cosine Similarity
and PCC similarity, the next step is to compute
predictions of unknown ratings for the target learner.
In the case of insufficient ratings, the principle is to
predict the rating
providing by target learner l
t
for a MOOC m
j
(k nearest neighbors) using the
rating given to m
j
by other similar learners (nearest
neighbors) obtained by eq. (2).
The predicted ratings are computed from the k
most similar MOOCs (k nearest neighbors)
obtained by eq. (1) and the ratings given on it by
the target learner. To compute the predictions of
ratings, we use the following prediction algorithm
(eq. 3).
(3)
Where
is the predicted rating for unrated
MOOC m
i
by the target learner l
t
, K denotes the
MOOC m
i
’s similar MOOC set (K nearest
neighbors ), Sim(m
i
, m
j
) is the similarity between
(2)
MOOCs Recommender System using Ontology and Memory-based Collaborative Filtering
639
two MOOCs m
i
and m
j
, and
is the rating of
MOOC m
j
(the neighbors ) by the target learner l
t
.
3.3 Generating Individual
Recommendation List
After predicting all unknown MOOCs rating in a
Learner-MOOC matrix, it is necessary to generate
recommendations list (top N) for the target learner.
For this step, the CF recommendation engine based
on the predicted ratings for the target learner and
ontological knowledge to generating
recommendation. After that, the filtering process
consists to eliminate the MOOCs that do not adapt
for the target learner’s profile (i.e. remove the
MOOC that don’t correspond to the preference and
need of learner, like language, knowledge level).
The list (top N) generated is ranked according to
their similarities with MOOC m
i
. Algorithm 1 shows
how the top N MOOCs recommendation list is
generated.
Algorithm 1: Generate Recommendation List.
Input :
Set of MOOCs :
M = {m
1
,m
2
,m
3
,
m
4
,....., m
n
}
set of learners:
L = {l
1
, l
2
,l
3,
..., l
s
}
Ontological Knowledge :
O = {Learner , MOOC }
Learner ratings on MOOCs :
R= {
,
,..,
} where
Rϵ {1,2,3,4,5}
Output :Top N recommendations list
Method
1: for each m
i
є M, o є O, do
2: Compute ontological similarity
Sim(
)using eq. (1)
end for
3: for each l
v
є L, o є O, do
4: Compute ontological similarity
Sim(
) using eq. (2)
end for
5: for each unknown ratings of
the target learner
6: Compute predicted ratings
using eq. (3)
end for
7: Filter the MOOCs according to
the learner profile
8: return Top N recommendations list
for the target learner l
t
.
4 CONCLUSIONS
In this paper, we propose a hybrid recommendation
approach based on ontology and collaborative
filtering to recommend MOOCs to learners within
MOOC platforms. The proposed hybrid
recommendation algorithm incorporates ontological
knowledge about the learner and MOOC into
recommendation process, while Collaborate filtering
predicts ratings and generates recommendations.
The novelty of our approach focuses specifically
on ontology-based recommender system within
MOOC platforms of which to the best of our
knowledge, no study has been conducted to describe
the learner and MOOC using ontology.
Directions for future research include the
experimental of the proposed approach in real
situation to show that the proposed hybrid algorithm
can obtain better performance and accuracy than
other related algorithms. We plan also to integrate
other techniques such as machine learning like
support vector machine (SVM).
REFERENCES
Adomavicius, G., Tuzhilin, A., 2011. Context-aware
recommender systems. In Recommender systems
handbook (pp. 217-253). Springer US.
Aggarwal, C.C., 2016. Recommender systems. Springer
International Publishing.
Aher, S.B., Lobo, L.M.R.J., 2013. Combination of
machine learning algorithms for recommendation of
courses in E-Learning System based on historical
data. Knowledge-Based Systems, 51, 1-14.
Ahn, H.J., 2008. A new similarity measure for
collaborative filtering to alleviate the new user cold-
starting problem. Information Sciences, 178(1), 37-51.
Bobadilla, J., Ortega, F., Hernando, A., Bernal, J., 2012.
A collaborative filtering approach to mitigate the new
user cold start problem. Knowledge-Based
Systems, 26, 225-238.
Bousbahi, F., Chorfi, H., 2015. MOOC-Rec: a case based
recommender system for MOOCs. Procedia-Social
and Behavioral Sciences, 195, 1813-1822.
Burke, R., 2007. Hybrid web r ecommender systems. In:
The adaptive web, pp. 377–408.
Dhorne, L., Deflandre, J.P., Bernaert, O., Bianchi, S.,
Thirouard, M., 2017. Mentoring Learners in MOOCs:
A New Way to Improve Completion Rates?.
In European Conference on MOOC (pp. 29-37).
Desrosiers, C., Karypis, G., 2011. A comprehensive
survey of neighborhood-based recommendation
methods. Recommender systems handbook, 107-144.
Goldberg, L.R., Bell, E., King, C., O’Mara, C.,
McInerney, F., Robinson, A., Vickers, J., 2015.
Relationship between participants’ level of education
ICEIS 2018 - 20th International Conference on Enterprise Information Systems
640
and engagement in their completion of the
Understanding Dementia Massive Open Online
Course. BMC medical education, 15(1), 60.
Goldberg, D., Nichols, D., Oki, B.M., Terry, D., 1992.
Using collaborative filtering to weave an information
tapestry. Communications of the ACM, 35(12), 61-70.
Ghazarian, S., Nematbakhsh, M.A., 2015. Enhancing
memory-based collaborative filtering for group
recommendersystems. Expert systems with
applications, 42(7), 3801-3812.
Guarino, N., Oberle, D., Staab, S., 2009. What is an
Ontology? In Handbook onontologies (pp. 1–17).
Berlin Heidelberg: Springer.
Gruber, T.R., 1992. Ontolingua: a mechanism to support
portable ontologies.
Jannach, D., Zanker, M., Felfernig, A., Friedrich, G.,
2010. An introduction to recommender systems.
In 25th ACM Symposium on Applied Computing.
Koohi, H., Kiani, K., 2017. A new method to find
neighbor users that improves the performance of
Collaborative Filtering. Expert Systems with
Applications, 83, 30-39.
Lü, L., Medo, M., Yeung, C. H., Zhang, Y.C., Zhang,
Z.K., Zhou, T., 2012. Recommender systems. Physics
Reports, 519 (1), 1–49.
Labib, A.E., Canós, J.H., Penadés, M.C., 2017. On the
way to learning style models integration: a Learner's
Characteristics Ontology. Computers in Human
Behavior, 73, 433-445.
Murphy, J., Tracey, J.B., Horton-Tognazzini, L., 2016.
MOOC camp: A flipped classroom and blended
learning model. In Information and Communication
Technologies in Tourism 2016(pp. 653-665). Springer.
Pang, Y., Jin, Y., Zhang, Y., Zhu, T., 2017. Collaborative
filtering recommendation for MOOC
application. Computer Applications in Engineering
Education, 25(1), 120-128.
Piao, G., Breslin, J.G., 2016. Analyzing MOOC Entries of
Professionals on LinkedIn for User Modeling and
Personalized MOOC Recommendations.
In Proceedings of the 2016 Conference on User
Modeling Adaptation and Personalization.
Patra, B. K., Launonen, R., Ollikainen, V., Nandi, S.,
2015. A new similarity measure using Bhattacharyya
coefficient for collaborative filtering in sparse
data. Knowledge-Based Systems, 82, 163-177.
Rabahallah, K., Azouaou, F., Laskri, M. T., 2016.
Ontology-Based Approach for semantic description
and the discovery of e-learning web services.
In Intelligent Networking and Collaborative Systems
(INCoS),(pp. 117-124). IEEE.
Sun, G., Cui, T., Shen, J., Xu, D., Beydoun, G., Chen, S.,
2017. Ontological Learner Profile Identification for
Cold Start Problem in Micro Learning Resources
Delivery. In Advanced Learning Technologies
(ICALT), 17th International Conference on (pp. 16-
20).
Schafer, J. B., Frankowski, D., Herlocker, J., Sen, S.,
2007. Collaborative filtering recommender systems.
In The adaptive web (pp. 291-324). Springer Berlin
Heidelberg.
Sarwar, B., Karypis, G., Konstan, J., Riedl, J., 2001. Item-
based collaborative filtering recommendation
algorithms. In Proceedings of the 10th international
conference on World Wide Web (pp. 285-295). ACM.
Tarus, J.K., Niu, Z., Yousif, A., 2017. A hybrid
knowledge-based recommender system for e-learning
based on ontology and sequential pattern
mining. Future Generation Computer Systems, 72, 37-
48.
Tang, S., Pardos, Z. A., 2017. Personalized Behavior
Recommendation: A Case Study of Applicability to 13
Courses on edX. In Adjunct Publication of the 25th
Conference on User Modeling, Adaptation and
Personalization (pp. 165-170). ACM.
Verbert, K., Manouselis, N., Ochoa, X., Wolpers, M.,
Drachsler, H., Bosnic, I., Duval, E., 2012. Context-
aware recommender systems for learning: a survey
and future challenges. IEEE Trans Learn Technol
5(4):318–335.
Wang, J., De Vries, A.P., Reinders, M.J., 2006. Unifying
user-based and item-1077 based collaborative filtering
approaches by similarity fusion. In Paper presented at
the proceedings of the 29th annual international ACM
SIGIR conference on research and development in
information retrieval.
Xing, W., Chen, X., Stein, J., Marcinkowski, M., 2016.
Temporal predication of dropouts in MOOCs:
Reaching the low hanging fruit through stacking
generalization. Computers in Human Behavior.
Yu, Z., Nakamura, Y., Jang, S., Kajita, S., Mase, K., 2007.
Ontology-based semantic recommendation for
context-aware e-learning. Ubiquitous Intelligence and
Computing, 898-907.
Zhang, H., Huang, T., Lv, Z., Liu, S., Zhou, Z., 2017.
MCRS: A course recommendation system for
MOOCs. Multimedia Tools and Applications, 1-19.
MOOCs Recommender System using Ontology and Memory-based Collaborative Filtering
641
