The Annual University Journal on Research and Applications, 2019, Vol-1, Issue 1
Online Recommendation System Using
Collaborative Filtering-Based Round Robin
First Author Second Author
Affiliation Affiliation
firstauthor@email.com secondauthor@email.com
Abstract— Recommender Systems have been widely applied in
many commercial e-commerce sites to help users deal with the
information overload problem. This paper aims for
recommending infrequently purchased products by exploiting
user generated content such as user product click streams data.
From product click stream data, user profiles are generated
using the user profiling approach. User profiles are used to
find similar-minded users accurately. Collaborative Filtering-
based Round Robin (CFRRobin) is used to personalize
recommendation based on the user profiles.
Keywords—: Round Robin, User Profiling, Collaborative –
Filtering
I. INTRODUCTION
Recommender systems help users to decide which
products they would like to purchase on e-commerce
sites. Many websites use collaborative filtering (CF)
technique which generates personalized recommendations
to users based on user’s preferences and ratings data.
Collaborative Filtering creates suggestions for users based
on their neighbor’s preferences. Collaborative Filtering
approach requires a large amount of rating data. This
paper is not directly applicable for recommending
products that are not frequently purchased by users such
as cars and house as it is difficult to collect rating data for
such products from the users. The system is aimed to use
the collaborative filtering and search-based techniques to
generate personalized recommendations for infrequently
purchased products. Collaborative Filtering-based using
round robin (CFRRobin) generates queries based on the
neighbor user’s profile and then a search to the product
collection by using the generated queries is conducted to
find the products that may be interested by the target user.
II. RELATED WORKS
There are two kinds of products offered to users on e-
commerce sites –low involvement products (LIP), such as
books, videos, soap, and high involvement products
(HIP), and such as electronic devices, cars, and houses
[3]. Collaborative Filtering (CF) approach has been
widely applied for recommending LIP because a rich
source of data is available for learning user’s preferences
and for generating personalized recommendations
according to the preference of similar users [4].
Twitter is a social information network where short
messages or tweets are shared among a large number of
users through a very simple messaging mechanism. With
a population of more than 100M users overwhelmed by
the massive amount of information available and the huge
number of people they can interact with. To overcome the
above information overload problem, recommender
systems can be introduced to help users make the
Page 1
Third Author
Affiliation
thirdauthor@email.com
appropriate selection. Researchers have begun to study
recommendation problems in Twitter but their works
usually address individual recommendation tasks. There is
so far no comprehensive survey for the realm of
recommendation in Twitter to categorize the existing
works as well as to identify areas that need to be further
studied. Su Mon Kywe, Ee-Peng Lim aimed to fill this
gap by introducing taxonomy of recommendation tasks in
Twitter, and to use the taxonomy to describe the relevant
works in recent years and it proposes a few research
directions for recommendation tasks in Twitter [5].
Online news reading has become very popular as the
web provides access to news articles from millions of
sources around the world. A key challenge of news
websites is to help users find the articles that are
interesting to read. Jiahui Liu, Peter Dolan presented
personalized news recommendation system in Google
News [6]. For users who are logged in and have explicitly
enabled web history, the recommendation system builds
profiles of users’ news interests based on their past click
behavior. To understand how users’ news interest changes
over time, they first conducted a large-scale analysis of
anonym zed Google News users click logs. Based on the
log analysis, they developed a Bayesian framework for
predicting users’ current news interests from the activities
of that particular user and the news trends demonstrated in
the activity of all users. They combined the information
filtering mechanism using learned user profiles with an
existing collaborative filtering mechanism to generate
personalized news recommendations. The combined
method was deployed in Google News. Experiments on
the live traffic of Google News website demonstrated that
the combined method improves the quality of news
recommendation and increases traffic to the site.
III. THETORETICAL BACKGROUND
A Recommender System (RS) as a computer system
that provides advice to users about items they might wish
to purchase. A recommender system provides individual
personalization to each user by customizing its
recommendations and presenting different items for each
user according to her/his tastes.
To provide a personalized set of recommendations, a
recommender system incorporates a user’s wishes into a
user model and exploits suitable recommendation
algorithms to map the user model into targeted product
suggestions. There are three steps involved in a
recommender system: acquiring preferences from a user’s
input data; computing the recommendation using proper
 The Annual University Journal on Research and Applications, 2019, Vol-1, Issue 1

techniques; and finally presenting the recommendation where denotes the user’s interest/preference the
results to user. value of attribute user preference for each
attribute value is calculated based on the number of
A. Recommendation Techniques
products with the attribute value that have been
There are many recommendation techniques such as
the following viewed by a user.

 Collaborative Filtering Recommender System E. Neighborhood Formation

Neighborhood formation is a key component of the
 Content-based Recommender System collaborative filtering approach in which a set of similar
 Knowledge-based Recommender System users or neighbors for a target user is generated [2].
Let tu =< , ,…, > be a target user’s profile
 Hybrid Recommender System
and =< , ,…, > be a previous user’s profile
created, based on the user click stream data.
B. Collaborative Filtering-Based Round Robin Fusion The equation to calculate the similarity between the target
Method(CFRRobin) user and each of the previous user is
Hybrid Recommender System is to combine sim (tu, )=
Collaborative filtering and search-based techniques.
Collaborative filtering and search-based techniques are
used to generate personalized recommendations for
infrequently purchased products. F. Round Robin Method
Round Robin method is simple data fusion
Collaborative Filtering-based round robin fusion
(CFRRobin) method generates queries based on the technique. Round Robin method contains two processes:
Product Retrieval and merging and Product Ranking[1].
neighbor user’s profile and then a search to the product
collection by using the generated queries is conducted to G. Product Retrieval and Merging
find the products that may be interested by the target user. CFRRobin approach uses each product of the user’s
neighbors as a query to retrieve other relevant products.
C. Architecture of the System Using neighbor products as a query, a recommender
In figure 1 shows the overview of the system. There system may retrieve more products that satisfy use’s
are three portions in the system. preferences than by directly recommends the products.
 User Profiling Approach Then, the retrieved products from all the queries are
 Neighborhood Formation using Cosine similarity merged by applying Round Robin method to generate a
 Round Robin Method set of candidate products.
Let B1 { , , , } is a set of target user’s
neighbor and { } be a set of products
viewed by the neighbor which is generated using user
profiling method.
Instead of using the products in as the
candidates for recommendation, the attribute values of
each products =< > are used as a query to
retrieve products
=< ,…, >
Let ={ = , = ,…, = }be a set of
Fig 1. Architecture of the System product attributes in the database DB. To measure the
similarity between the products and , the following
D. User Profiling
method can be used
User profiling is defined as to generate a user profile
sim( , )=
that represents the user’s preference or interest in products
or product attributes. It based on the user’s online click
=
stream data when the user browses products on an e-
commerce site. A user profile is represented by a set of Based on similarity, a list of ranked products can be
attribute values. A user profile (up) is formally
generated, =< ,…, > where sim( , ) >
represented as:
up=< sim( , ) >…> sim( , ) . Round Robin method is
adopt to merging and selecting final products from
=1 different products lists , , ,…, . Round

Page 2
 The Annual University Journal on Research and Applications, 2019, Vol-1, Issue 1

Robin method selects a product from the top of each B. Experimental Results
for each round and then starts again from the top of the
list for the remaining products in each . From the The objective of this experiment is to verify that the
user profiles based on user click stream data that can
ranked products in U …U , the top N products are improve recommendation accuracy by combining the
chosen from neighbor Bi ,as . By combing the collaborative filtering and search-based approach.
products in for all neighbors, a set of candidate
products can be obtained, Table 1: Precision and Recall results with
 .
Threshold value = 50%, (50 users in 150
Products)

H. Product Ranking Users Total Number Number of Precision Recall

Product Ranking is to rank the products in the number of Match
of Retrieve product(NM)
candidate list and to select the top N products to product (NR)
recommend. (NT)
Let =< ,…, > ,
= (u ), i=1,….,n be the target user’s highest User1 150 94 
86 91.48% 57.33%
preference attribute values . Let be the set of candidate User2 150 110 90 81.81% 60%
products generated by CFRRobin , . And
=< ,…, > is the product with its User3 150 62 55 88.7% 36.67%
attributes and attribute’s value. The similarity between
and , denoted as sim( , ),is used to rank the User4 150 110 94 85% 62%
products in .The top N products are selected as the
final products to be recommended from the ranked User5 150 110 94 85% 62%
products.

IV. EXPERIMENTAL RESULT

The results in Table 1 show the precision and recall
for five users using 150 products and 50 user’s click
The objective of this experiment is to verify that the
stream data. Based on the number of products that match
user profiles based on user click stream data that can
the returned products from the system, the result for each
improve recommendation accuracy by combining the
user is calculated. So, the precision and recall for 150
collaborative filtering and search-based approach.
. products in 50 users on five of ten attributes is 86.4% and
55.6%.
A. Evaluation Method

Precision measures the ability of the system to Table 2: Precision and Recall results with
Threshold value = 80%, (50 users in 150
present only items that are relevant, and it can be seen as Products)
the measure of exactness. Precision is defined as the ratio
of the retrieved items that are relevant (NM) and the Users Total Number Number of Precision Recall
number of all retrieved items (NR) number of Match
of Retrieve product(NM)
Precision= product (NR)
(NT)
Recall measures the ability of the system to present
User1 150 94 92 97.87% 61.33%
all the relevant items and it can be seen as the measure of
98.18% 72%
completeness. Recall is defined as the ratio of the
User2 150 110 108
retrieved items that are relevant (NM) and the number of
items that should be returned (NT) User3 150 62 60 96.7% 40%
Recall=
User4 150 110 108 98.18% 72%
To evaluate this system, NM is the number of
User5 150 110 108 98.18% 72%
retrieved cars that are relevant from the retrieved cars, NR
is the number of retrieved cars, and NT is the total number The results in Table 2 show the precision and recall
of cars in the system. These cars have some attributes that for five users using 150 products and 50 user’s click stream
of interest to the user. The precision and recall are data. Based on the number of products that match the
calculated for each user
returned products from the system, the result for each user

Page 2
 The Annual University Journal on Research and Applications, 2019, Vol-1, Issue 1
is calculated. So, the precision and recall for 150 products
in 50 users on eight of ten attributes is 97.8% and 63%.
The performance of the system will increase and
generate accurate results when the system has numerous
user click stream data. The system’s performance will
decrease when the system has few user click stream data.
The oval with vertical straight-line represents all product
lists in the system; the oval with dot is the set of relevant
items (Precision); the circle with black dot represents the
item which relevant by the review process (Recall).
Fig 2. Precision and Recall results
C. Conclusion
Recommender system (RS) have been widely
applied for recommending products on e-commerce sites
to overcome information overload issues and help users in
selecting final products to purchase. The popularly used
collaborative filtering (CF) recommendation approach
requires a large amount of explicit ratings data for making
meaningful recommendations. However, this data is not
always available as it requires high involvement from the
users to provide explicit ratings of the products they
already know. Consequently, recommender systems are
not currently popular for recommending luxury products
because the users only purchase few such items in their
Page 2
lifetime and thus, it is impossible for them to provide
sufficient ratings for the recommender system to make
meaningful recommendation.
For recommending infrequently purchased products,
this system uses collaborative filtering based round robin
(CFRRobin) technique. This paper intend to improve the
recommendations using CFRRobin. This paper intent to
provide an effective personalized recommendation system
for online user
REFERENCES
[1] Si,L. & Callan J. (2003). A semisupervised learining method to
merge search engin result, ACM Transactions on Information
Systems, 21(4), 457-491.
[2] Adomavicius, G. & Tuzhilin, A. (2005). Toward the next
generation of recommender systems: a survey of the state-of-the-art
and possible extensions. IEEE Transactions on Knowledge and
Data Enginnering, 17(6), 734-749.
[3] K. Srikumar and B. Bhasker, Personalized product selection in
Internet business, Journal of Electronic Commerce Research, vol.5,
no.4 pp.216-227 , 2004 .
[4] X. Su and T.M.Khoshgoftaar. A survey of collaborative filtering
techniques, Advances in Artificial Intelligence, 2009.
[5] Su Mon Kywe, Ee-Peng Lim and Feida Zhu. A Survey of
Recommender Systems in Twitter, Singapore Management
University, Singapore.
[6] Jiahui Liu, Peter Dolan, Elin Ronby Pedersen. Personalized News
Recommendation Based on Click Behavior, 1600 Amphitheatre
Parkway, Mountain View, CA 94043, USA
[7]