AD ALTA
JOURNAL OF INTERDISCIPLINARY RESEARCH
A REVIEW OF HYBRID RECOMMENDER SYSTEMS
a
BABAK JOZE ABBASCHIAN,
b
SAMIRA KHORSHIDI
a
University of Louisville, 2301 S 3rd St, Louisville, KY 40208,
USA
b
Indiana University Purdue University, 420 University Blvd.
Indianapolis, IN 46202, USA
Email:
a
b0joze01@louisville.edu,
b
sakhors@iu.edu
Abstract: Due to the huge amount of information available online, the need of
personalization and filtering systems is growing permanently. In today’s world
recommender systems are increasingly used to make suggestions and provide
information about items for users. Also recommendation systems constitute a specific
type of information filtering technique that attempt to present items according to the
interest expressed by a user. There are many techniques that can be applied for
personalization in recommender systems. All these techniques have complementary
strengths and weaknesses. A hybrid recommender system combines two or more
recommendation techniques to gain better system performance and mitigate the
weaknesses of individual ones. In this review paper, we prepared a brief introduction
on hybrid recommendation systems, components of recommendation systems, various
approaches of recommendation systems such as collaborative approach, content-based
approach, hybrid approach and demographic approach.
Keywords: hybrid approach, collaborative approach, content-based approach,
demographic approach.
1 Introduction
It is typically necessary to possess a certain sufficient amount of
information to make good decisions in any situation.
Technologies enable us to easily obtain more information,
especially on the Internet. For example, if an individual want to
rent a movie online, there are numerous choices available.
However, too much information can make decision-making
inefficient, leading to information overload. Personalization
technologies and recommender systems help to overcome this
problem by providing personalized suggestions regarding which
information is most relevant to users. [1]
Recommender systems are used in various online applications
from e-commerce to search engines. There are a number of
techniques used to implement recommender systems, each with
its advantages and disadvantages. Hybrid systems intend to
combine two or more of these techniques in order to obtain
better results. [2]
Recommender systems [3] reached a broad acceptance and
attracted public interest during the last decade, also expanding
the field for new sales opportunities in e-commerce [4, 5].
Recommender systems are divided into two categories in term of
their approach to rating estimation: content-based and
collaborative recommender systems. Content-based
recommendations [6] based on item similarity of the user
preferred to objects in the past. Moreover, collaborative
recommendation systems [7] depend on the ratings given by
individuals with similar taste and preference. However, both
techniques exhibit specific strong and weak points.
Collaborative filtering recommender systems are the most
commonly used systems [8]. They involve the use of the
information provided by other users to make suggestions to a
particular user. This can be compared to what happens in real
life when an item is purchased based on the recommendation.
Collaborative filtering systems differ in the way they use the
information provided by other users to link it to the information
available about the user that it needs to make a prediction for. A
type of collaborative filtering is the use of association rules.
Development of recommender systems depends on e-commerce
but there are also other applications for them such as search
results and news portals customization. Different techniques
have been used, including the nearest neighbor algorithm [9],
association rule mining [10] and neural networks [11].
Hybrid techniques were implemented to overcome some
challenges in the above-mentioned techniques. The challenges
include some aspects of performance, trust security and privacy
issues.
Hybrid approaches unifying collaborative and content-based
filtering less than one single framework, reducing synergetic
effects and mitigating inherent challenges of either paradigm.
Finally, hybrid recommenders operate on both product rating
information and descriptive features. In fact, numerous ways for
combining collaborative and content-based aspects are
conceivable; [8] lists an entire plethora of hybridization methods.
However most widely adopted among these, is the so-called
“collaboration via content” paradigm [12], where content-based
are built to detect similarities among users.
2 A review on the recommendation systems: approaches and
limitations
This section is a review on the basic approaches of
recommendation systems. The approaches include content-
based, collaborative filtering, demographic, and hybrid
approaches. Also there are limitations for the recommendation
approaches are described in details.
Today there are different approaches to recommendation systems
that are used to serve in different contexts based on system
needs. The content-based approach deals with item profiles and
user profiles, and it is designed to recommend text-based items.
The collaborative filtering approach is widely used in
commercial areas. Amazon uses the collaborative filtering
approach to recommend books and other products to its
customers [13]. Recommendation systems based on
collaborative filtering recommend items to a particular user
based on the similar items that have been rated by some other
users, and the target user and the other users share the same
preferences of items or products. The demographic approach
recommender systems use demographic information such as the
gender, age, and date of birth of respective users in order to
recommend items [13].
3 Content-based approach
Content-based approach is one of the most widely used
recommendation approaches. One main component of content-
based is the user modeling process, because the interests of users
are inferred from the items that users interacted with. Items are
usually textual, for instance emails or webpages [14]. There are
actions that are typically established interaction through
downloading, buying, authoring, or tagging an item. Items are
showed containing the items’ features. Features are typically
word-based. Some recommender systems use non-textual
features, such as writing style, layout information, and XML tags
[15].
In content-based approach, the user rates the items, that mean the
recommender system should understand the common
characteristics among the items that the user has rated in the
past. The system then recommends the items that have a high
degree of similarity to the user’s preferences and tastes. For
example, in a movie recommendation system, a content-based
approach tries to understand the common characteristics such as
actors, directors, genres, etc. among the movies that the user has
given high ratings in the past. Then, the system recommends the
movies that have a high degree of similarity to the user’s
preferences [13].
In a content-based recommendation system, a user profile
contains the user’s preferences of items. A user profile can be
obtained by analyzing the content of all rated items [5].
Specifically, this profile is constructed by using the content
(keyword) that has been analyzed using the methods that are
mentioned in the item profile section. Each item in the user’s
profile has a weight that denotes the importance of keyword Ki
to the user [5]. This weight can be computed using average
- 259 -