International Journal on Recent Researches In ISSN (Print) : 2347-6729

Science, Engineering & Technology ISSN (Online) : 2348-3105

A Journal Established in early 2000 and upgraded to International journal in 2013 and is in
existence for the last 10 years. It is run by Retired Professors from NIT, Trichy. Volume 2, Issue 9,
It is an absolutely free (No processing charge No publishing charge etc) Journal Indexed in September 2014
DIIF and SJIF.
Research Paper DIIF IF :1.46
Available online at: www.jrrset.com SJIF IF: 1.329
Chief Editor : Dr. M.Narayana Rao, Rtd. Professor, NIT, Trichy.

Developing a Recommender system for web based learning

Baseera. A1
Assistant Professor, Department of Computer Science
Tirupattur Arts and Science College,
Tirupattur. Vellore District.

Dr.M.V.Srinath2,
Professor, Dept of Computer Science and Engineering,
Mahendra Engineering College,
Namakkal, Tamilnadu.

Abstract - A Recommender System (RS) in an E-learning context is a software agent that tries to intelligently
recommend actions to a learner, based on the actions of previous learners. This activity could be through online or
from a simple web resource. An attempt is made in this paper to make a detailed review of the RS used for various
applications with their merits and demerits .The review includes special reference to be made to data-mining
techniques used for building the RS for online learning activities, which are based on learners access history to
improve course materials navigation as well as assist the online learning process. The RS that are reviewed include
I) Personal Recommender Systems II) Collaborative or Social Filtering Recommender Systems III) Context Based
Recommender Systems IV) Personal Logic Recommender Systems V) News Dude Systems etc. In general a RS
provides the users details about items they might wish to acquire. Recommendations made by such systems can help
the users, navigate through large information available. The major contribution of present work is that the detailed
review presented in this paper paves way for building knowledge based RS which overcomes the limitations of
existing RS and helps the learners for effective use of E-learning modules.

Keywords- Recommender Systems, Classifier Systems, Decision Support Tools, Navigation, Customer Purchasing
Habits.

1.0 Introduction knowledge accessible to those who need it, when

they need it-anytime, anywhere." In recent times,
"E-learning is not just e-training. E-learning is the
educational institutions have started re-examining the
overarching umbrella that encompasses education,
way training and education is imparted. Elearning has
information, communication, training, knowledge
now become a crucial part of their strategy to deliver
management, and performance management. It is the
knowledge. E-learning is the effective learning and
web-enabled system that makes information and
International Journal on Recent Researches in Science, Engineering and Technology, Vol.2,Issue97,Sept. 2014.
ISSN (Print) 2347-6729 ; ISSN (Online) 2348-3105
teaching process that uses a wide spectrum of
technologies to create E-digital content for more
engaging and interactive learning experiences,
mainly Internet or computer-based, to reach learners
at a distance. E-learning is not a replacement killer
for traditional classroom-teaching; it is an enhancer.
Organizations choose E-learning to complement their
traditional instructor-led training programs to
increase the efficiency and productivity of their
employees and overall effectiveness of the business.
In today’s hyper-competitive market, more and more
academic institutions, enterprises and government
organizations across the globe are turning to cutting-
edge technologies to enhance and complement their
offerings and achieve competitive advantage in their
industry. Anyone who has ever taught in a traditional
classroom setting knows how difficult it is to give
individual instruction while at the same time meeting
the needs of the class as a whole. Instructors must
pace their presentations for the majority of learners in
the class. If some students are having a difficult time
mastering a new concept, it is virtually impossible for
the instructor to know about this unless they speak
up. Without knowing this, the instructor cannot adjust
his/her pace. In addition, even if the instructor is
aware of these students having difficulty mastering a
concept, he/she still needs to pace the instruction for
the majority of the class. This inability to pace
instruction to each student’s needs is a major
drawback to traditional instruction. It penalizes both
bright and slower students and is inefficient. This is
not the case with E-learning. Using E-learning
students can learn at the pace that suits them best
while getting the information that they need. A brief
review of the existing recommender systems is
presented. In general the various step involved in
developing the recommender system are i. Analyze
user data, ii. Identify the requirements of users, iii.
Compiling user data verses user requirements, and iv.
Finally zero down to the data required from the data
available and present it to the user. The various
algorithms such as i. Random prediction algorithm
[7], ii. Use of frequent sequence pattern for
developing algorithms [8], iii. Collaborative filtering
algorithms [9], and iv. Contents based algorithms [5],
are normally used to develop the recommender
software for E-learning modules. In Personal RS, it
caters to the requirements of individuals. This
enables users of all learning capacities at the pace
that suits them the best [6]. The most well known
type of RS is the Collaborative or Social Filtering
type. These systems aggregate data about customers
purchasing habits or preferences, and make RS to
other users based on similarity in overall purchasing
patterns. Content-based RS are classifier systems
derived from machine learning research. These
systems use supervised machine learning to induce a
classifier that can discriminate between items likely
to be uninteresting [2]. A third type of RS is Content
Based RS. This system provides advice to users about
items they might wish to purchase or examine.
Recommendations made by such system can help
users navigate through large information spaces of
product description, news articles or other items [1].
The fourth type of RS is one that uses Knowledge-
Based approach to generating a recommendation,
reasoning about what products meet the user’s
requirements. The Personal Logic recommender
system offers a dialog that effectively walks the user
down a discrimination tree of product features.
International Journal on Recent Researches in Science, Engineering and Technology, Vol.2,Issue97,Sept. 2014.
ISSN (Print) 2347-6729 ; ISSN (Online) 2348-3105

Others have adapted quantitative decision support 2.0 Literature Review

tools for this task. The system allows users to
The latest trend is focus on learner than supporting
navigate by stating their preference with respect to a
tools. The FindMe technique is one of knowledge-
given restaurant, there by refining their search
based similarity retrieval. There are two fundamental
criteria. Each of these approaches has its strengths
retrieval modes: similarity-finding and tweak
and weakness. As a collaborative filtering system
application. In the similarity case, the user has
collects more rating from more users, the probability
selected a given item from the catalog (called the
increases that someone in the system will be a good
source) and requested other items similar to it. To
match for any given new user. However, a
perform this retrieval, a large set of candidate entities
collaborative filtering system must be initialized with
is initially retrieved from the database. This set is
a large amount of data, because a system with a small
sorted based on similarity to the source and the top
base of ratings is unlikely to be very sensitive to the
few candidates returned to the user. Web-based
number of rated items that can be associated with a
learning systems are also available. The main
given user [8]. The fifth type of RS is NewsDude RS
limitations of web based systems are that they are
this system avoids the above problem by using a
closed learning environments. The main drawback,
nearest-neighbor classifier that works with few
namely locating the required learning material from
examples, but the system can only base its
among the vast data variable is addressed through
recommendations on ratings it has, and cannot
designing an E-learning recommender based on
recommend stories unless they are similar to ones the
learner’s average rating. All above aspects are
user has previously rated. A knowledge-based RS
considered, while designing the current recommender
avoids some of these drawbacks. It does not have a
system.
ramp-up problem since its recommendation does not
depend on the basis of user rating. It does not have to
3.0 Methodology of Developing The Proposed
gather information about a particular user because its Recommender System
judgments are independent of individual tastes. These
To start with it is essential to build a knowledge
characteristics make knowledge based
based system for the complete E-learning life cycle.
recommender's not only are valuable systems on their
The knowledge based system should address the
own, but also highly complementary to other type of
following issues namely, user interface tools
RS [3]. Above review reveals that each type of RS
required, the methodology of generating the
has its own merits and drawbacks. This review
developed recommender, the criteria for selecting the
analysis paves way for the design and development
suitable recommender. The recommender software is
of knowledge–based RS [10].
designed considering all above aspects. The basic
source of material for this approach is the vast input
data available. Knowledge engineering of the type
described above is necessary for building a
International Journal on Recent Researches in Science, Engineering and Technology, Vol.2,Issue97,Sept. 2014.
ISSN (Print) 2347-6729 ; ISSN (Online) 2348-3105
knowledge-based RS. This is the inevitable “price of
admission” for a knowledge based approach, a price
that is not incurred by knowledge tweak method such
as collaborative filtering or machine learning.
However, these tweak methods suffer from the ramp-
up problem. The differing strengths of these
approaches suggest that they may be complementary
rather than computing approaches for the generation
of recommendations. A particular benefit of FindMe
system is that they gather preference information
without requiring that user make their ratings
explicit. Rather than requiring the user to input his or
her preferences as a starting point, Find Me systems
let the user browse through a catalog using
qualitative ratings as navigation aids. Each navigation
step informs the system about the user’s preferences
at a finer grain of detail then a single “buy” decision,
and a user is likely to make several such navigation
steps. Within FindMe systems, it is often one case
that a retrieval strategy fails to discriminate the
returned items completely. The system might be
required to arbitrarily select 10 items to return, for
example, out of a topmost bucket of size 20. In such a
case, we consider the result “under-discriminated”.
Eliminating under-discriminated results in FindMe
system can be difficult because it requires the
addition of one or more new similarity metrics, with
attendant knowledge-engineering tasks, or it may
require a more in-depth representation of the
products. Collaborative
filtering, however, can add additional discrimination
without requiring knowledge engineering. Consider a
system
that, in addition to the FindMe recommender
component, has also a collaborative filtering engine,
where ratings are obtained by recording each
example the user has seen and the user’s reaction to
it. If a user names an item as a starting point, we can
consider that very high rating. Since the user is
seeking something similar to what he or she has seen
and like before. The exit point of the system could
also be considered a high rating, since the user stops
searching. But this is less reliable since the user may
have given up without finding anything satisfactory.
Each tweak along the way can consider a navigation
rating, since the user has found something to dislike.
With this technique, we can accumulate ratings for
users with only the overhead of logging their FindMe
activity. These ratings can be used to compute
correlations between users. The operation of this
recommender is likely to be tweak if it starts with a
small amount of data, so we would not want to
present its suggestions directly to users. However, we
will not make a bad suggestion if we only examine
the items in the topmost under-discriminated bucket.
After we have performed the normal FindMe
ranking, we look at the user’s profile to rate and rank
each items in the topmost bucket. There is little risk
in applying collaborative filter are the random, we
will still be selecting items that are equally similar as
far as our knowledge –based system is concerned.
4.0 Design and Development of the
Recommender Software
i) The software is essentially a knowledge data base
capable of handling vast data relating to all areas of
interest.
ii) The data is stored under different headings and is
available to the users at the click of the mouse.
International Journal on Recent Researches in Science, Engineering and Technology, Vol.2,Issue97,Sept. 2014.
ISSN (Print) 2347-6729 ; ISSN (Online) 2348-3105

iii) The software also provides methodology of IEEE 3rd International Conference on Computing &
analyzing the content, structure the same and stores Processing (Hardware/Software) Beijing, August
it. (2012), pages 513-518.
iv) The data may be retrieved as and when required
for a specific application. [2] Bahram Amini, Roliana Ibrahim, Mohd Shahizan
v) The software should also guide the users as to the Othman, Discovering the Impact of Knowledge in
history of the previous users both online/ offline with Recommender Systems: A Comparative Study,
their opinions on the RS used. International Journal of Computer Science and
vi) Summing up the software include a) given the Engineering Survey (IJCSES) Vol 2, No. 3, August
content topic, the software should provide the RS (2011).
best suited for the current application b) list out the
available structured data relating to the current [3] Bobadilla, J., Serradilla. F., Dernal. J. A new
application c) methodology of using the listed data d) collaborative filtering metric that improves the
should provide related information or the source for behavior of recommender systems. Elsevier Journal.
the same e) list out the advantages of the RS used Vol. 23, Issue 6, August 2010, pages 520-528.
over other available RS f) should provide possible
suggestions for the best use of the RS. [4] Frank McSherry and Iiya Mironov, Differentially
Private Recommender Systems, KDD’09 June 2009,
5.0 Conclusions Paris, France, ACM 978-60558-459-9/09/06. pages
1. Literature review revealed that, the tools required 150-161.
for developing the E-learning modules are arranged,
based on the users and no systematic generalized [5] Gabor Takacs, Istvan Pilaszy, Scalable
tools are available for a general application. Collaborative Filtering Approaches for Large
2. The present work proposes a generalized design of Recommender Systems. Journal of Machine Learning
tools for effectively developing varied groups of Research (2009), pages 623-656.
Elearning modules.
3. The review reveals that a knowledge based RS can [6] Linden, G., Smith, B., and York, J. Amazon.com
be designed which addresses the limitation of recommendations: Item-to-item collaborative
existing RS. filtering. IEEE Internet Computing 7 (2003), pages
76-80.
References

[1] A. Costanzo, A fuzzy mobile recommender [7] Papagelis, M., and Plexousakis, D. Qualitative
system. Software Engineering and Service Science analysis of user-based and item-based prediction
(ICSESS), 2012 algorithms for recommendation agents. Engineering
International Journal on Recent Researches in Science, Engineering and Technology, Vol.2,Issue97,Sept. 2014.
ISSN (Print) 2347-6729 ; ISSN (Online) 2348-3105

Applications of Artificial Intelligence 18, 7 (2005),

pages 781-789.

[8] R. Agrawal, T. Imielinski, and A. Swami. Mining

association rules between sets of items in large
databases. In Proc. 1993 ACM-SIGMOD Int. Conf.
Management of Data, pages 207–216, Washington,
D.C., May 1993.

[9] Sarwar, B., Karypis, G., Konstan, J., and Reidl,J.

Itembased collaborative filtering recommendation
algorithms. In WWW '01: Proceedings of the 10th
international conference on World Wide Web
(NewYork, NY, USA, 2001),
ACM, pages 285-295.

[10] Shishehchi, S.Review of Personalized

recommendation techniques for learners in e-learning
systems, Semantic Technology and Information
Retrieval (STAIR), International Conference on
Knowledge Technologies, Malaysia,
June (2011) pages 277-281.