International Journal of Computer Applications (0975 – 888)

Volume 47– No.9, June 2012

A Survey Paper on E-Learning Recommender System

Reema Sikka Amita Dhankhar Chaavi Rana

M.Tech (SE)-II Associate Professor Associate Professor
University Institute of University Institute of University Institute of
Engineering and Technology Engineering and Technology Engineering and Technology
MDUHaryana India MDU Haryana India MDU Haryana India

ABSTRACT represent some information about their preferences. If not, the

A recommender system in an e-learning context is a software recommender requests preference information from the user
agent that tries to ”intelligently” recommend actions to a explicitly. Based on the known preferences of the seeker and
learner based on the actions of previous learners. This those of other people, the recommender recommends items
recommendation could be an on-line activity such as doing an the seeker probably will like. The seeker may use the
exercise, reading posted messages on a conferencing system, recommendation to select items from the universe of
or running an on-line simulation, or could be simply a web alternatives or to communicate with like-minded other users
resource. These recommendation systems have been tried in e- as the recommender also identifies people with similar
commerce to entice purchasing of goods, but haven’t been interests.
tried in e-learning. This paper suggests the use of web mining
techniques to build such an agent that could recommend on-
1.1Recommendation Systemsfor E-
line learning activities or shortcuts in a course web site based Learning
on learners’ access history to improve course material An ”e-learning task recommender” is a recommendation
navigation as well as assist the online learning process. These system that would recommend a learning task to a learner
techniques are considered integrated web mining as opposed based on the tasks already done by the learner and their
to off-line web mining used by expert users to discover on- successes, and based on tasks The similarity of the learners
line access patterns could be established using user profiles, or could be based on
common previous access patterns made by other “similar”
Keywords learners. The similarity of the learners could be established
E-learning, Recommender Systems, Collaborative Filtering, using user profiles, or could be based on common previous
Content-based filtering. access patterns. In principle, there are two major parts in the
design of such an agent: a ”learning” module that learns from
1. INTRODUCTION past access patterns and infers an individual or common
E-learning is indeed a revolutionary way to provide education access model; and an ”advising” module that applies the
in life long term, comparing with the traditional face-to-face learned model at given times to recommend actions. There are
style teaching and learning. Nowadays more and more people many ways to implement this process, such as data clustering,
have benefited from various e-learning programs. However, association rule mining, or collaborative filtering [11] etc.
high diversity of the learners on the Internet poses new
challenges to the traditional “one-size-fit-all” learning model, Software agents that recommend actions, products, or other
in which a single set of learning resource is provided to all items have been used in some applications. However, to the
learners. In fact, the learners could have various interests; best of our knowledge there is no distance learning system to
even sharing with the common interests, they may have date that provides such automated facilities to automatically
different levels of expertise, and hence they cannot be treated suggest learning activities or resources. In the field of
in a uniform way. It is of great importance to provide a electronic commerce, however, given the lucrative prospects,
personalized system which can automatically adapt to the a significant research effort has been made to devise elaborate
interests and levels of learners. methods to take advantage of customers’ accesses and
purchase behaviours in order to enhance the purchasing
Recommender Systems are implemented in commercial and experience and customer satisfaction by user profiling and
non-profit websites to predict the user preferences. The main smart recommendations, and thus increase profit.
functions of recommendation systems include analyzing user Recommenders are used to boost sales by displaying products
data and extracting useful information for further predictions or services a consumer is likely to be interested in. Systems
A recommender system is a piece of software that helps users for recommendation such as Amazon.com that suggests books
to identify the most interesting and relevant learning items or other products to purchase related to a current purchase
from a large number of items. Recommender systems may be based on preference information and other users purchases.
based on collaborative filtering (by user ratings), content- The techniques used are, however, very simple and not always
based filtering (by keywords), and hybrid filtering (by both accurate or even effective. Basically, the program compares
collaborative and content-based filtering). The following the set of items purchased by the current customer with the set
figure is a general model of recommendation which of items purchased by other customers, selects the customers
summarizes the recommendation process by identifying the with the bigger item overlap with the current customer’s item
communication among its components. A recommendation set, then finally picks some items not yet bought by the
seeker can be considered as a user of the system. The seeker customer but present in the baskets of customers with high
may directly ask for a recommendation, or the recommender overlap and presents them as a recommendation list to the
may produce recommendations without the seeker’s demand. current customer. More sophisticated methods take into
Seekers may get into contact with preference provider and account ratings of products given by customers and select

27
 International Journal of Computer Applications (0975 – 888)
Volume 47– No.9, June 2012

products for recommendation from customers that rate items of failure. Thus, it has never been taken seriously
the same way or level as the current customer. This technique by any researcher or vendor and only serves as
also used in information retrieval for retrieving text reference point1, helping to compare the quality of
documents that are similar is called collaborative filtering the results obtained by the utilization of a more
[11]. Recommendation of movies or music compact discs, sophisticated algorithm [15].
such as moviefinder.com, uses collaborative filtering by
predicting a person’s preferences as a linear weighted  Frequent sequences can help build recommender
combination of other people’s preferences. Since we do not systems. For example, if a customer frequently
have ratings for course material, collaborative filtering is not rates items we can use the frequent pattern to
applicable for devising an accurate recommender agent for e- recommend other items to him. The only problem
learning activities. Moreover, we are interested in is that this method will only be efficient after the
recommending beneficial learning activities to enhance on- customer makes minimum purchases.
line learning, as well as recommending shortcuts or jumps to
some resources to help users better navigate the course  Collaborative filtering algorithms (CF) are
materials. algorithms that require the recommendation seekers
to express their preferences by rating items. In this
1.1.1 What Makes Recommendations Different in algorithm, the roles of recommendation seeker (a
E-learning from Other Domains user) and preference provider are merged; the more
An e-learning recommender system includes: users rate items (or categories), the more accurate
 Items liked by learners might not be pedagogically the recommendation becomes. In most CF
appropriate for them. For example, a learner without approaches, there is a list of users U = u1,u2,,,, u m
prior background on the techniques of web mining and a list of items I = i1,i2,,,,,in. Each user ui has a
may only be interested in knowing the state-of-the- list of item Iui on which he has expressed his
art of web mining techniques in ecommerce. Then, opinion [16].
it should be recommended that he/she read some  Content based algorithms are algorithms that
review papers. For example, an editorial article by attempt to recommend items that are similar to
two of the leading researchers in this area [5], items the user liked in the past. They treat the
although there are many high quality technical recommendation's problem as a search for related
papers related to his/her interest. On the other hand, items. Information about each item is stored and
for the learner coming from industry with some used for the recommendations. Items selected for
prior knowledge who wants to know how web recommendation are items that content correlates
mining can be utilized to solve e-commerce the most with the user's preferences [17]. For
problems, should be recommended, because the example, whenever a user rated an item, the
paper is the KDD-Cup 20001 organizers' report on algorithm constructs a search query to and other
how web mining can support business decision popular items by the same author, artist, or director,
making for a real-life e-commerce vendor, and or with similar keywords or subjects [14]. Content
points out challenges, as well as lessons learned based algorithms analyse item descriptions to
from the competition, which can benefit both identify items that are of particular interest to the
researchers and industry practitioners. By contrast in user.
other domains, recommendations are made based 2. LITERATURE SURVEY
purely on users' interests.
Recommender system in an e-learning context is a software
agent that tries to ”intelligently” recommend actions to a
 Customization should not only be made about the learner based on the actions of previous learners. This
choice of learning items, but alsoabout their recommendation could be an on-line activity such as doing an
delivery.For example, some instructors will exercise, reading posted messages on a conferencing system,
recommend learners to read an interesting or running an on-line simulation, or could be simply a web
magazine article, such as a related article in resource. These recommendation systems have been tried in e-
Communications of ACM, before a technical paper, commerce to entice purchasing of goods, but haven’t been
because they believe it will help learners tried in e-learning. This paper suggests the use of web mining
understand the technical paper and make them less techniques to build such an agent that could recommend on-
intimidated. However, this is not the case in e- line learning activities or shortcuts in a course web site based
commerce recommendations, where site managers on learners’ access history to improve course material
prefer to leave the list of recommended items navigation as well as assist the online learning process. These
unordered to avoid leaving an impression that a techniques are considered integrated web mining as opposed
specific recommendation is the best choice. to off-line web mining used by expert users to discover on-
1.1.2 Recommender System Approaches line access patterns. [1]
Recommender system can be built with many approaches. The focus in eLearning has recently been shifted from the
Below are some of them: supporting tools towards the learning person. It has turned out
that the learning success can significantly be improved if the
 Random prediction algorithm is an algorithm that learning content is specifically adapted to individual learners’
randomly chooses items from the set of available preferences, learning progress and needs. Personalization has
items and recommends them to the user. Since the evolved as a promising concept to take into consideration
item's selection is done randomly, the accuracy of individual needs. Due to the dynamic nature of eLearning, this
the algorithm is based on luck; the greater the paper focuses not only on static but also particularly on
number of items is the chance of good selection dynamic aspects of personalization. [2]
lowers. Random prediction has a great probability

28
 International Journal of Computer Applications (0975 – 888)
Volume 47– No.9, June 2012

A personalized e-learning system which can automatically the process to select the appropriate recommendations, and d)
adapt to the interests and levels of learners is the system the management of the users’ interactions. We are defining a
which is designed based on the IEEE Learning Technology user-centred evaluation approach that copes with those issues
Systems Architecture (IEEE LTSA) to achieve high and drives the recommender building process in three
scalability and reusability. A feedback extractor with fusion consecutive steps: 1) elicitation of pedagogically sound
capability is proposed to combine multiple feedback measures recommendations validated by users with a collaborative
to infer user preferences. User profile, which stores user review, 2) acquisition and validation of the user features to
preferences and levels of expertise, is collected by user select the appropriate recommendations for the current
profiler to deliver personalized information using the context, and 3) analysis of the recommendations provided and
collaborative filtering algorithm. [3] evaluation of their impact on the us. [8]
The majority of current web-based learning systems are closed Recommender systems have been a useful tool to recommend
learning environments where courses and learning materials items in many online systems, including e-learning. However,
are fixed and the only dynamic aspect is the organization of not much research has been done to measure the learning
the material that can be adapted to allow a relatively outcomes of the learners when they use e-learning with
individualized learning environment. In this paper, we recommender system. Instead, most of the researchers were
propose an evolving web-based learning system which can focusing on the accuracy of the recommender system in
adapt itself not only to its users, but also to the open Web. predicting the recommendation rather than the knowledge
More specifically, the novelty with respect to the system lies gain by the learners. This research aims to compare the
in its ability to find relevant content on the web, and its ability learning outcomes of the learners when they use several types
to personalize and adapt this content based on the system's of e-learning recommender systems. Based on the comparison
observation of its learners and the accumulated ratings given made, we propose a new e-learning recommender system
by the learners. Hence, although learners do not have direct framework that uses content-based filtering and good learners’
interaction with the open Web, the system can retrieve ratings to recommend learning materials, and in turn is able to
relevant information related to them and their situated learning increase the student’s performance. The results show that
characteristics. [4] students who used the proposed e-learning recommender
system produced a significantly better result in the post-test.
Recommendation process in distributed information systems The results also show that the proposed e-learning
extracts similarities and differences between recommender system has the highest percentage of score gain
recommendations in stores and the recommendations applied from pre-test to post-test. [9]
to an e-learning environment. It also explains the phenomena
of self-organization and cooperative emergence in complex 3. CONCLUSION
systems coupled with bio-inspired algorithms to improve A recommender system is a program that sees what a user is
knowledge discovery and association rules. Finally, the doing and tries to recommend courses of action it thinks
present recommendation is applied to e-learning by proposing would be beneficial to the user. This is the idea behind some
recommendation by emergence in Multi-Agent System systems used in electronic commerce sites to recommend
architecture. [5] products to customers they might wish to purchase based on
Recommended learning resources are computed based on the their previous purchasing history as well as the purchasing
current learner’s recent navigation history, as well as history of those who bought similar goods. To date, this hasn’t
exploiting similarities and dissimilarities among learners’ been proposed for on-line learning environments and no
preferences and educational content. The proposed framework known e-learning system uses such a software agent to
for building automatic recommendations in e-learning enhance the on-line learning experience as described in this
platforms is composed of two modules: an off-line module paper.
which pre-processes data to build learner and content models, We have proposed an approach to build a software agent that
and an online module which uses these models on-the-fly to uses data mining techniques in order to build a model that
recognize the students’ needs and goals, and predict a represents on-line user behaviours, and uses this model to
recommendation list. Recommended learning objects are suggest activities or shortcuts. These suggestions can help
obtained by using a range of recommendation strategies based learners better navigate the on-line materials by finding
mainly on content based filtering and collaborative filtering relevant resources faster using the recommended shortcuts and
approaches, each applied separately or in combination. [6] assist the learner choose pertinent learning activities that
An enormous amount of learning materials in e-learning has should Improve their performance based on on-line behaviour
led to the difficulty on locating suitable learning materials for of successful learners.
a particular learning topic, creating the need for content
recommendation tools within learning context. In this paper, 4. REFERENCES
we aim to address this need by proposing a novel framework [1] Zaiane, O. R.,“Building a Recommender Agent for e-
for an e-learning recommender system. Our proposed Learning Systems”,University of Alberta, Edmonton,
framework works on the idea of recommending learning Alberta, Canada, 2002.
materials based on the similarity of content items (using [2] Armin, U., and Kandpal, D., and Tochtermann,K.,
Vector Space Model) and good learners’ average rating ”First Steps towards Personalization Concepts in
strategy. This paper presents the overall architecture of the eLearning” Know-Centre Graz, 2003.
proposed system and its potential implementation via a
prototype design. [7] [3] Li, X., and Chang, S. K., “A Personalized E-Learning
System Based on User Profile Constructed Using
When building a knowledge-based recommender along the Information Fusion” University of Pittsburgh, USA,
eLearning life cycle, the following issues have to be 2006
considered: a) the user interface design of the tools required,
b) the process to design/generate the recommendations, c)

29
 International Journal of Computer Applications (0975 – 888)
Volume 47– No.9, June 2012

[4] Tang, Y. T. ; and McCalla, G.,” Smart [11] S. Chee, J. Han, and K. Wang. Rectree: An efficient
Recommendation for an Evolving E-Learning System” collaborative filtering method. In 3rd Int. Conf. On Data
Dept. of Computer Science, University of Saskatchewan, Warehousing and Knowledge Discovery (DAWAK
2007 2001), LNCS 2114, pages 141–151, Munich, Germany,
September 2001. Springer Verlag.
[5] Gil, A. B., and García-Peñalvo, F., J., ”Learner Course
Recommendation in E-Learning Based on Swarm [12]O. R. Za¨ýane and J. Luo. Towards evaluating learners’
Intelligence”, Department of Computer Science, Sciences behaviour in a web-based distance learning environment.
Faculty University of Salamanca, Salamanca, Spain, In Proc. of IEEE International Conference on Advanced
2008. Learning Technologies (ICALT01), pages 357–360,
Madison, WI, August 2001.
[6] Khribi, M. K., Jemni, M., & Nasraoui, O. (2009).
Automatic Recommendations for E-Learning [13] O. R. Za¨ýane, M. Xin, and J. Han.Discovering web
Personalization Based on Web Usage Mining Techniques access patterns and trends by applying OLAP and data
& Information Retrieval. Educational Technology & mining technology on web logs. In Proc. Advances in
Society, 12 (4), 30–42. Digital Libraries ADL’98, pages 19–29, Santa Barbara,
CA, USA, April 1998
[7] Ghauth , K., I., B., and Abdullah, N., A., ” Building an
E-Learning Recommender System using Vector Space [14] Linden, G., Smith, B., and York, J. Amazon.com
Model and Good Learners Average Rating” Multimedia recommendations: Item-to-item collaborative filtering.
University , and University of Malaya, 2009 Ninth IEEE IEEE Internet Computing 7 (2003), 76-80.
International Conference on Advanced Learning
Technologies. [15] Papagelis, M., and Plexousakis, D. Qualitative analysis
of user-based and item-based prediction algorithms for
[8] Santos, O.C., and Boticario, J.G., ” Building a recommendation agents. Engineering Applications of
knowledge-based recommender for inclusive eLearning Artificial Intelligence 18, 7 (2005), 781-789.
scenarios” Artificial Intelligence Department, Spain,
2009 [16] Sarwar, B., Karypis, G., Konstan, J., and Reidl,J. Item-
based collaborative filtering recommendation algorithms.
[9] Ghauth, K., I., and Abdullah, N., A., ” Measuring In WWW '01: Proceedings of the 10th international
learner’s performance in e-learning recommender conference on World Wide Web (NewYork, NY, USA,
systems”. Multimedia University, 2010 2001), ACM, pp. 285-295.
[10] R. Agrawal, T. Imielinski, and A. Swami. Mining [22] VAN Meteren, R., and van Someren, M. Using content-
association rules between sets of items in large databases. based filtering for recommendation.
In Proc. 1993 ACM-SIGMOD Int. Conf. Management of
Data, pages207–216, Washington, D.C., May 1993.

30