A still remains high because of its applications, which

A Survey On Recommendation System

Atul Mishra

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!

order now

of Computer Engineering                                              
Deptt of Computer Engineering

University of Science And Technology                       YMCA University of
Science And Technology 

[email protected]                                                       
[email protected]




In the past years ,the way of communication have been changed between
the web and the users. With the ample amount of data being present on the
web,it became very important to use filtering techniques to find our interest
of information. Recommendation systems has its applications in almost every
field including e-commerce, e-bussiness, e-government, e-library, e-learning
etc. In this paper, we introduce various filtering techniques to ease our work
of finding relevant information.Content based and collaborative techniques are
the most widely used and are discussed in the paper with their pros and cons.
However, the newest research in this area- Hybrid recommendations are also
included in this paper.


1.      Introduction

Since the appearance of the first papers on
collaborative filtering since the mid-1990s 1,2,3, Recommender systems became
an important research area.The interest in this area still remains high because
of its applications, which help users deal with the information overload and
provide personalized recommendations to them.

With the immense increase in the amount of
information available, retrieving information of our interest has became very
difficult. With the development of Information Retrieval systems such as
Google, Alta Vista, the task of retrieving information has become quite easy
but upto some extent.

The Recommendation systems have fully solves
this problem. It has now become the most powerful tool in electronic commerce. In recent years, Recommendations systems have become
immensely popular and are used in variety of areas such as in movies, in dating
websites, in restaurants, in social tags and in news.


systems are softwares that provide suggestions for items to be of use to the
user.The suggestions can be related to various decision-making processes such
as which item to buy, which news to read , which movies to watch , which music
to listen. Over the years, various approaches of recommender systems have been
created – collaborative ,content-based
filtering.However  the newest technology
“hybrid recommendations-mixture of collaborative and content’ also comes into
picture now a days.

Recommender system are good
for both the users and the service providers4.They actually reduce cost of finding
and then selecting items4.In e-commerce recommender system is an effective
way to increase profits.They are very useful in libraries.

Item is the term which is
used to denote what the system recommends to users7. In the immortal words
of Steve Jobs: “A lot of times, people
don’t know what they want until you show it to them.” Customers may love your
movie, your product, your job opening- but they may not know it exists. The job
of the recommender system is to open the customer/user up to a whole new
products and possibilities, which they would not think to directly search for


2.      Recommendation filtering techniques

A recommendation system must
predict an item that is worth recommending.In order to do this, a system must
able to predict the utility of some of items, and then decide what items to
recommend based on their comparison 5.


2.1  Content based filtering

This is the most famous
filtering and is used widely .Content based recommendation is targeted at
personal level and considers individual preferences and contents of the
products for generating recommendations.

Basically these content based
recommendations system elaborate a specific profile of each content and
then  perform some correlation matrices
on data.Various items are compared with items previously rated by the user and
best matching items are recommended. Collaborative algorithm does two main
tasks :

a user profile that describes the types of items the user likes/dislikes

the user profile to some reference characteristics (with the aim to predict
whether the user is interested in an unseen item)


Various advantages
of content based filtering are :

based systems are user dependent.

These systems are highly transparent.



Lack of diversity : No suitable suggestions if the analyzed content does not contain enough
information to discriminate items the user likes from items the user does not

(b)   Scalability : Content based
filtering is not scalable.


2.2  Collaborative filtering Recommendations

Another filtering technology
that is widely used in recommender systems is Collaborative Filtering.As
compared to content based filtering ,collaborative recommender system can
automatically filter the information that the content based system could not
represent and gives up to date recommendations and informations 6.

Collaborative filtering was
developed in late 1990’s and it is the most famous filtering technique till
date .Various online web services such as Amazon and Netfix are utilizing this
filtering techniques.

Collaborative filtering algorithms usually separated into two




2.2.1 Memory-based
Collaborative Filtering

Memory based collaborative filtering
store  the entire customer ratings into
their memory.

They are also called lazy recommendation
algorithms. They do not immediately attempt to calculate customer precedence of
an object. Typical examples of this approach are
neighbourhood-based CF and item-based/user-based top-N recommendations.

Two types of memory based CF are there:




Item/Object-based filtering :  Calculate similarity
between items and make recommendations. Items usually don’t change much, so
this often can be computed off line. Item/Object-based
filtering was recommended by Sarwar et al at 2001 7.
Items/objects are compared for similarity. The neighborhood of most likely
objects is recognized, for every object that belongs to the customer who is
active .

Item/object Based Algorithm8

For every object “o” that “c” has preference for yet

For every object “p” that “c” has a preference for

Compute a similarity s between “o” and “p”

Add “c”‘s preference for “p”, weighted by s, to a running

Return the top objects , ranked by weighted average



User/Customer-based filtering: Recommend items by finding
similar users. This is often harder to scale because of the dynamic nature of

User/customer based Algorithm8

For every object “o” that “c” has no preference for yet

For every other customer “d” that has a preference for “o”

Compute a similarity s between “c” and “d”

Add “d”‘s preference for “o”, weighted by s, to a running

Return the top objects, ranked by weighted average


2.2.2 Model-based Collaborative Filtering: memory
based  tackle the task of “guessing” how
much a user will like an item that they did not encounter before. For that they
utilize several machine learning algorithms to train on the vector of items for
a specific user, then they can build a model that can predict the user’s rating
for a new item that has just been added to the system.

Popular model-based
techniques are Bayesian Networks, Singular Value Decomposition, and
Probabilistic Latent Semantic Analysis9.



Scalability6: There are
millions of users and items present.Thus a large amount of computation power is
important to calculate recommendation. For example, with millions of customers
(C) and millions of distinct items (O), a CF algorithm with the complexity of
O(n) is already too large. Also, many systems need to react immediately to
online requirements and make recommendations for all users regardless of their
purchases and ratings history, which demands a high scalability of a CF system.

Data sparsity10 : If a
customer or user  has evaluated very few
items then its quite difficult to know his taste and his preferences and in
this case he could be related to wrong neighbourhood8.So, this lack of
information is quite harmful.

Cold start problem4 :  This is a situation where a recommendor
system do not have adequate information about a customer or object in order to
make relevant predictions.


2.3 Hybrid recommendations

Hybrid systems are the
newest recommendor system and combines the best feature of both content based
and collaborative based filtering .

Types of hybrid systems are

Weighted hybridisation4 : This technique combine the
results of collaborative and content based filtering to generate a prediction
by integrating the techniques of both the recommendations. It increases the
performance of the whole system.The main idea of using this technique is to
conquer  the disadvantages of any
individual technique. It computes the weighted sum of the scores and then
compare them.

(b)   Switching hybridisation11
: This technique switches to other recommendations techniques based on the
needs of the moment.It solves the new user problem .This method solves the
problem which is specific to any one method.One example of switching hybrid is “Daily
Learner” that switches between content and collaborative filtering according to
the need.

(c)    Mixed hybridisation12
:Instead of having only one recommendation per object, this technique combine
results of different recommendations techniques at the same time.Each object
have many recommendations associated from different recommendations techniques.In
this technique the result of an individual does not affect the performance in

Cascade Hybridisation 13: Refinement is done in cascade
hybridisation. Results from one recommendations acts as input to other
recommendation techniques. Due to this, this process is tolerant to
noise.Example of Cascade hybridisation is EntreeC that used a collaborative recommender
and knowledge based recommender.


3.  Phases of
Recommendation process

3.1 Loading and formatting

Dataset is a collection of
interests and likes of various users by using which we can recommends products
and items to them..Dataset can be downloaded from various websites such as
movielens,frappe, CoMoDa

3.1.1 Data
collections are done by two methods-explicitly and implicitly. 

data  include the following:

a customer to rate an object.

a customer to rank a collection of object from most favourite to least

two object to a customer and asking to choose the better one of them.

a customer to create a list of object that they likes.


data collection includes the following:

observing the object that a customer views in an online store.

its viewing time.

a track of the object that a customer purchases online.


3.1.2 Datasets are normally in the form 14

{ critic,title,rating}

where critic is the user,

title is the item it rated,

rating is the rating given by the user.


The next step is to arrange the data in a
format that is useful to build the recommendation engine .The current data
contains a row containing critic,title and rating.This has to be converted to
matrix format containing critics as rows,title as columns, and ratings as the
cell values.

The data can be viewed as :























3.2 Calculating similarity between the users

This is the very important
step as we have to find the similarity between two users to recommend items to
them. Various similarity measures are available 15.

1.      Distances

We often want to compare two feature vectors, to measure how similar
they are. We hope that similar patterns will behave in a similar way.


using correlation

The distance is not well normalised .So, we use correlations.


Correlation Coefficient

Recommending items to users

The final step is to recommend items to the users.





Conclusion and Future Work

This paper presented the
various techniques and algorithm to build the recommender system. We study
various research paper and realised that Collaborative filtering is the mostly
used filtering technique but there are various problems related to this such as
data sparsity,cold start problem ,scalability. We have also introduced various
advantages and disadvantages related to these techniques. Various areas where
still much research is to be done in coming years has also been discussed (hybrid


5.  References

Hill, W., L. Stead, M.
Rosenstein, and G. Furnas. Recommending and evaluating choices in a virtual
community of use. In Proceedings of CHI’95.


Resnick, P., N. Iakovou, M.
Sushak, P. Bergstrom, and J. Riedl. GroupLens: An open

architecture for
collaborative filtering of netnews. In Proceedings of the 1994 Computer

Supported Cooperative Work
Conference, 1994.


Shardanand, U. and P. Maes.
Social information filtering: Algorithms for automating

‘word of mouth’. In Proc. of
the Conf. on Human Factors in Computing Systems, 1995.


4.Recommendation systems :
principles,methods and evaluations  by



5.Introduction to
Recommender Systems Handbook by Francesco Ricci, Lior Rokach and

Bracha Shapira


6.A Survey on Recommender
Systems based on Collaborative Filtering Technique  by Atisha sachan.



7.Sarwar.B, Karypis.G,
Konstan.J, and Riedl.J. (2001). “Item-Based Collaborative Filtering
Recommendation Algorithms”. In WWW ’01: Proceedings of the 10th
International Conference on World Wide Web, New York, NY, USA.11.


8.Item based and user based
recommendation difference in Mahout by data science



9.Memory-Based vs.
Model-Based Recommendation Systems by yasserebrahim


10.Survey Paper on  Recommendation System  by 
Mukta kohar,Chhavi Rana


11. An Improved Switching
Hybrid Recommender System Using Naive Bayes Classi?er and Collaborative
Filtering by Mustansar Ali Ghazanfar and Adam Pru¨gel-Bennett ?


12.A mixed hybrid
recommender system for given names by Rafael Glauber1, Angelo Loula1, and Jo˜ao
B. Rocha-Junior2


13.A Hybrid Recommender
System Using Link Analysis and Genetic Tuning in the Bipartite Network of
BoardGameGeek.com by Brett Boge.


14.Building recommendation
engines by By Suresh Kumar Gorakala


15.Similarity and
recommender systems by Hiroshi Shimodaira?



I'm Angelica!

Would you like to get a custom essay? How about receiving a customized one?

Check it out