A LITERATURE SURVEY ON RECOMMENDATION ENGINE

Rajesh Kumar
Department of Computer Science and Engg.
M.tech, GITM, Bilaspur

Abstract

In the last sixteen years, more than 200 research articles were published about research-paper
recommender systems. We reviewed these articles and present some descriptive statistics in this
paper, as well as a discussion about the major advancements and an overview of the most
common recommendation concepts and approaches. Recommendation engine are software tool
and techniques providing suggestion to the users. Now a day’s a wide impact in online
marketing. Recommendation system are subclass of information filtering system that predict
rating or preference that a user would give to an items (such as book, music or movies) or social
items. The suggestion provided are aimed at supported their users in various decision making
processes such as what items to buy, what music to listen, what policy to buy depends upon its
benefits. However the recommender systems are highly successfully and advisable for people
with online presence. Recommendation systems apply data mining techniques and prediction
algorithms to predict user’s interest on information, products and service among the large
number of available items. The vast growth of information on the internet as well as number of
visitors to websites have some key challenges to recommender system. To address these
challenges we have explored several collaborative filtering techniques such as the item based
approach, which identify relationship between items and users. In this paper we introduced the
topic of recommender system. It provides ways to evaluate efficiency, scalability and accuracy of
recommendation system. The paper also analyzes different techniques of user based and item
based. Moreover a simple experiment was done to achieve the recommendation search.

Keywords
Recommender engine, Recommender system, Collaborative Filtering, Content–Based Filtering,
Item Based recommendations.

1. Introduction
We are currently living in an era of information. We are surrounding by a huge amount of data in
the form of reviews, blogs and comments on various websites. The increasing importance of the
web as a medium for electronic and business transactions has served as a driving force for the
development of recommender system technology. The explosive growth of the world-wide web
and the emerging popularity of e-commerce has huge collection of data. Recommendation
systems were developed to help to close the gap between information collection and analysis by
filtering all of the available information to present what is valuable to the user. For e.g., Consider
a user bought a product from e commerce website and he want to rate this product is good or bad
so that other users can see the feedback and they should decide whether this product should buy
or not. In such cases, uses are able to easily provide feedback with simple with a click of a
mouse. A methodology to provide feedback is in the form of rating, in which user select
numerical values from specific evaluation system (e.g. five-star rating system) that specify their
likes and dislikes of various items. One of the most successful such technologies is the
Recommender system; as defined by. Deshpande and G. Karypis “a personalized information
filtering technology used to either predict whether a particular user will like a particular item
(prediction problem) or to identify a set of N items that will be of interest to a certain user (top-N
recommendation problem)". Over the years, various approaches for building recommender
systems have been created; collaborative Filtering has been a very successful approach in both
research and practice, and in information filtering and e-commerce applications. Collaborative
filtering works by creating a matrix of all items and users' preferences. In order to recommend
items for the target user, similarities between him and other users are computed based on their
common taste. This approach is called user-based approach. A different way to recommend
items is by computing the similarities between items in the matrix. This approach is called item
based approach.

2. Background
Recommender system can be built with many approaches.
Below are some of them.
 Random prediction algorithm is an algorithm that randomly chooses items from the set of
available items and recommends them to the user. Since the item’s selection done
randomly, the accuracy of the algorithm is based on luck; the greater the number of items
is the chance of good selection lowers. Random prediction has a great probability of
failure. Thus it has never been taken seriously by any researchers or vendor.

 Frequent sequence can help build recommender systems. For example, If a customer
frequently rates items we can use the frequent pattern to recommend other items to him.
The only problem is that this method will only be efficient after the customer makes
minimum purchases.

 Collaborative Filtering algorithm are algorithms that requires the recommendation user to
express their preferences by rating items. In this algorithm the role of recommendation
users and preference provider are merged; the more users rate items, the more accurate
the recommendation becomes.

 Content based algorithms are algorithms that attempt to recommend items that are similar
to items the user liked in past. They treat the recommendation’s problem as a search for
related items. Information about each item is stored and used for the recommendations.
Items selected for recommendation are items that content correlates the most with the
user’s preferences. For example whenever a user rated an items, the algorithm constructs
a search query to find other popular items by the same author, artist, songs etc. Content
 based algorithm analyze item descriptions to identify items that are of particular interest
to the user.

Many others approaches for recommender system exist. However, collaborative filtering
algorithm have come to be the best of recommendation algorithms. As per researches
collaborative filtering algorithms have been extensively adopted by both research and e-
commerce recommendation systems in order to provide an intelligent mechanism to filter out the
access of information available and to provide customers with the prospect to find out items that
they will probably like according to their logged history of prior transactions.

3. Contribution
This paper has three primary research contribution:
1) Analysis of the user based and item based prediction algorithms.
2) Formation of a hybrid model that uses both item based and user algorithm for more
accurate prediction.
3) An experiment to show how from unstructured data converted into structured and make a
judgement of best product.

4. Organization

My work will be primarily based on collaborative filtering algorithms. First we introduce and
describe collaborative filtering. Afterwards I discuss about user based and item based algorithms.
Following that we discuss about of privacy and security issues related to recommender systems
and also describe the metrics used to evaluate recommender.

4.1 Techniques used in Recommendation System:-

Recommendation systems are categorized as of three types namely –
1) Content Based Recommender System
2) Collaborative Based Recommender System
3) Hybrid Recommender System

4.1.1 Content Based Recommender System:-

Content based recommendation systems recommend an item to a user based upon a description
of the item and a profile of the user’s interests. Such systems are used in recommending web
pages, TV programs and news articles etc.
 Item Selection
Features of items

Users
Recommender Recommended
Rating
System Items

User modeling data

Figure 1: Content based Approach

All content based recommender systems has few things in common like means for description of
items, user profiles and techniques to compare profile to items to identify what is the most
suitable recommendation for a particular user. Content-based recommendation systems analyze
item descriptions to identify items that are of particular interest to the user. Because the details of
recommendation systems differ based on the representation of items.

4.1.2 Collaborative Based Recommender System:-

Collaborative Filtering (CF) systems work by collecting user feedback in the form of ratings for
items in a given domain and exploiting similarities in rating behavior amongst several users in
determining how to recommend an item.
The importance of Collaborative Filtering methods is: guessing the usefulness of a particular
item for a particular user depending upon the previous history of other users. Consider Table1.1,
where 'U' and 'S' represent user and item respectively. U1, U2, U3, U4, U5 represent users and
S1, S2, S3, S4, S5 represent items. Assume that we want to suggest a particular item for
'U4'(user 4), S5(item 5) is recommended for the U4(user 4) who is “similar” to U3 (user 3) as
both of them have given high rating for the item S3 User 4 is likely to like item S3. The
recommendations are made by comparing the likes and dislikes of users who have taste that is
similar to that of the current user.
 S1 S2 S3 S4 S5

U1 3 4 3 1

U2 1 2

U3 1 2 5 3

U4 2 1 4 ?

Table 1.1 user-item rating

4.1.3 Hybrid Recommender:-

Hybrid recommender systems are systems that combine multiple recommendations techniques
together to achieve a synergy between them. Several researches have attempted to combine
collaborative filtering and content based approaches in order to smoothen their disadvantages
and gain better performance while recommendation. Depending on domain and data
characteristic, several hybridization techniques are possible to combine CF and CB technique.

5. Challenges of User-based Collaborative Filtering Algorithms

Building a recommender presents two main challenges which are:-
1) Making the right design decisions for the system to be built
2) Finding the best way to represent the data so that it can be efficiently used by the
recommender to provide higher prediction accuracy than single recommender systems.
Another major challenge that we have face and which regards the integration strategy is
finding the best strategy possible so that after integration both the requirements of the
host system and those of the recommender are still met.
In order to take up the above mentioned challenges we are going to study the guidelines that
drive the design of recommender system. Doing so will enable us to make the right choices for
the design of our system. In addition to studying the guidelines and to analyze the advantages
and disadvantages of the integrations strategies for the application.
6. Security and Privacy Issues
Collaborative filtering recommender requires personal information from a user to give personal
information from a user to personalized recommendations. The more users express their
preference items, the more accurate the recommendation they receive become. As with any data
mining systems, users must trust the recommender to protect their information appropriately.
More over since the user does not know how the recommendation is performed. The
recommender system should not violate the trust of the users.

6.1 Privacy Risks

In most of the systems users need to register before they rate or comment about the products the
registration process often requires them to provide some personal information like their names,
birth dates, postal code and email. Since not every users want their information to be disclosed or
misused, the recommender should then protect itself against exposition of users’ information or
misuse of those information. Recommender systems are also confronted with other type of
problems such as security's related problem. Since being recommended is often promise of good
selling, recommender are often target of manipulation from producers or malicious users For
example, a book writer may try to alter the recommendation so that his book get recommended.
Recent research by Dellarocas and others have shown that even popular systems such as Amazon
and eBay have (and are) being manipulated.

6.2 Shilling Attacks

A shilling attack is an attack in which the system's recommendations for a particular item is
manipulated by submitting misrepresented opinions to the system (S. Lam, D.Frankowski, and J.
Riedl, 2006). The attack can have two objectives: decrease the ratings of all the items outside its
target item-set (push attack) to make them more recommended. He may also increase the ratings
(nuke attack) of other items to make its target item-set less recommended.

7. Related Search
In practice, research paper recommender system do not exist. However concepts have been
published and partly implemented that could be used for their realization.

 Prem Melville, Raymon J. Mooney, Ramadass Nagarajan: - It gives an approach to

combine content and collaboration to enhance existing user data and to give better
performance than a pure content based predictor.
 Marko Balabanovic: - An adaptive recommendation service for collection and selection
of web pages. It makes the system more personalized and combines the benefits of
content analysis with the shared user interests.
  Yi Zhang: - Proposes a faster technique to gather a huge number of individual user
profiles even if feedbacks available are less. It uses various parameters for optimization
of joint data
 Toshio Okamoto and Yan Feng: - It captures the relations between conditions that the
learner has been exposed to, which are used to represent the knowledge of the learner.
The relevance and complement of learning resources recommendation are increased by
keyword map. It is better suited for e-learning settings and achieves higher accuracy than
common recommender methods.
 Sarabjot Singh Anand and BamshadMobasher: - A recommender system is proposed
which retrieves relevant preference information from long term memory and uses it in
conjunction with the information stored in short term memory.

8. Conclusions

This thesis has explored the feasibility of the design and the implementation of a recommender
system using Spring 3.0.4 framework and JSON. It has also explored the feasibility of a seamless
integration of the recommender engine in Policy Portal (Application Developed). The approach
proposed for the integration of the recommender was a full integration strategy to the portal
using a custom JSTL. It consisted in an implementation following the JSON paradigm. In other
words, it consisted in implementing the recommender system under the shape of JSON data and
spring controller were used to display the recommended data by different customers view and
comments.

 Modularity and maintainability: the fact that the recommender system and the models
have can be used with other services to makes it modular and very easy to maintain

 Low coupling and high cohesion: the fact that the system is modular implies low
coupling between its components thus providing high cohesion

 Extensibility: given that internally the recommender engine is composed by local services
which manage each a part of its data model, the number of item classes the recommender
manipulates can be extended easily by just adding a web service that manages the underlying
business logic and interfaces with the database

The approach proposed for the implementation of the recommender system was a hybrid method
that combines content based information and collaborative based information. The evaluation the
Hybrid approach has proven that it performs better than single collaborative based
recommenders and single content based recommenders. In that regards, I believe that my
approach is practical for real life applications.

9. References

[1]Francesco Ricci, Lior Rokach, Bracha Shapira, Paul B. Kantor: Recommender systems handbook,
Springer, 2011, pp 332 – 340.
[2]Toby Segaran: Programming collective intelligence, OReilly, 2007.

[3] Ted Husted, Cedric Dumoulin, George Franciscus, David Winterfeldt: Struts in action, Manning,
2003

[4]Christian Bauer, Gavin King: Java persistence with hibernate revised edition of hibernate in
action, Manning, 2005.

[5] Riccardo Bambini, Paolo Cremonesi, Roberto Turrin: A recommender system for an IPTV
Service Provider: a real large scale production environment, Whitepaper March 2010.

[6]Toine Bogers, Antal van den Bosch: Collaborative and content-based filtering for item
recommendation on social bookmarking websites, ACM RecSys ’09 Workshop on Recommender
systems and social web, October 25 2009, New York, NY, USA.

[7] Pasquale Lops, Marco de Gemmis, Giovanni Semeraro: Content-based Recommender Systems:
State of the Art and Trends, Springer Science + Business Media, LLC 2011

[8]Jerome Picault, Myriam Ribiere, David Bonnefoy, Kevin Mercer: How to Get the Recommender
Out of the Lab?, Springer Science + Business Media, LLC 2011

[9]Luo Si, Rong Jin: Flexible Mixture Model for Collaborative Filtering, Proceedings of the
Twentieth International Conference on Machine Learning (ICML-2003), Washington DC, 2003.

[10] Emilio F Zegarra, Tony Effremenko: Using a recommendation engine to personalize your web
application enhancing the user experience with Apache Mahout and WebSphere Application Server,
IBM WebSphere Developer Technical Journal, September 21, 2011.