AD ALTA
JOURNAL OF INTERDISCIPLINARY RESEARCH
Figure 1: Several models of hybrid filtering, using both
collaborative filtering approach and content-based approach
[40].
Case B: incorporating some of the characteristics of the
content-based approach into the collaborative filtering
approach, e.g. the incorporation of personal information in
collaborative filtering approach, to alleviate the problem of
introducing a new user to the RS;
Case C: building a general model that takes into account
characteristics from both collaborative filtering approach
and content-based approach, therefore combining their
results with a machine-learning algorithm (e.g. using
Bayes-Networks);
Case D: incorporating some of the characteristics of the
collaborative filtering approach into the content-based
approach. For example, using the determined
recommendations from the collaborative filtering approach
used as an input to the content-based approach algorithm;
7 Discussion and Conclusion
In the 16 years from 1998 to 2013 more than 200 research
articles were published in the field of recommender systems. The
articles consisted primarily of peer-reviewed conference papers
(59%), journal articles (16%), pre-prints (5%), and other
documents such as presentations and web pages (15%). The few
existing literature surveys in this field cover only a fraction of
these articles, which is why we conducted a comprehensive
survey of recommender systems. The review revealed the
following information [35].
Content-based approach is the predominant recommendation
approaches. From 62 reviewed approaches, 34 used content-
based approach (55%). From these content-based approach
approaches, the majority utilized plain terms contained in the
documents. A few approaches also utilized non-textual features,
such as citations or authors. [8].
The most popular model to store item representations was the
vector space model (VSM). Other approaches modeled their
users as graphs, as lists with topics that were assigned through
machine learning, or as ACM classification hierarchies. The
reviewed approaches extracted text from the title, abstract,
header, introduction, foreword, author-provided keywords,
bibliography, body text, social tags, and citation context.
According to Yang et al result, it concluded that only eleven
approaches applied collaborative filtering approach, and none of
them successfully used explicit ratings. Hence, implicit instead
of explicit ratings were used. Implicit ratings were inferred from
the number of pages the users read, users’ interaction with the
papers and citations. The main problem of collaborative filtering
approach for research papers seems to be scarcity. Vellino
compared implicit ratings on Mendeley (research papers) and
Netflix (movies), and found that scarcity on Mendeley differs
from Netflix by a magnitude of three [30].
As mentioned above, a demographic approach recommends
items to the user based on the user’s demographic information
such as gender, age, and date of birth. Demographic approach
puts the users into groups based on their demographic
characteristics. Also, the users of ages ranging from 18 to 25
years-old will be in one group. The demographic approaches
assume that the users in the same group or category share the
same interests and preferences. The demographic approach
tracks the buying or rating behavior of the users within the same
group or category. The demographic approach first will place the
user into a particular group based on the user’s demographic
information. Then, the system will recommend products or items
to the user based on the buying or rating behavior of the other
users in the group.
In this study a novel hybrid approach we proposed to prediction
of rating, so collaborative filtering approach and content-based
approach were used and finally combined. Although there are
several hybrid recommendation systems, so in order to combine
collaborative filtering approach and content-based approach,
rating and content information are integrated to build a hybrid
model. The main advantages of this hybrid model are less
parameters and more reasonable prediction.
Since content features have a characteristic such as multiplicity,
so this hybrid model has flexibility in size, by what the
computational effort increased substantially. In order to reduce
the runtime in the system, some dimension of hybrid model by
means of singular value decomposition decreased. The hybrid
model compared with basic collaborative filtering approach, this
hybrid approach performed better in prediction accuracy and
runtime. So according to result we can conclude that novel
hybrid model is practical to real-life applications.
Since, over the last years, recommender systems have made
significant progress, accordingly hybrid recommendation model
have been proposed and implemented. This proposed hybrid
model mainly focuses on providing justifications for the
recommendations. This proposed hybrid model is the integration
of content and context data with rating data, and also provides
accurate justifications for recommendations.
Moreover, this proposed hybrid model provide an explanation
interface that shows the recommendations in a group or a
category rather than duplicating the same information which
reduces the time for decision making of customers. This
proposed hybrid model allows the customer to interact with it to
provide feedback on the recommendations and justifications.
The results have clearly shown that interact with the customer
more effectively and boosts the customer’s satisfaction on the
recommender system. Interacting with the recommender systems
allows the customer to achieve their desired product quicker.
This proposed hybrid model is implemented by a prototype web-
based application in the JAVA platform. This proposed hybrid
prototype is implemented for movies; however, it can be easily
implemented for other products. However, all of these advances
with accordance with the current generation of recommender
systems still require further improvements to make
recommendation methods more effective in a broader range of
applications. Specifically, there is lot of work needed in the area
of providing effective explanations that will increase the
customer’s trust on the recommender systems and also boosts
the business of the organizations.
References
1. Gediminas Adomavicius and YoungOk Kwon (2013). New
Recommendation Techniques for Multi-Criteria Rating Systems.
Department of Information and Decision Sciences Carlson
School of Management University of Minnesota.
2. Alex Cristache (2009). Hybrid recommender system using
association rules. School of Computer and Mathematical
Sciences. A thesis submitted to Auckland University of
Technology in partial fulfillment of the requirements for the
degree of Master of Computer and Information Sciences (MCIS)
3. Resnick, P. & Varian, H. R. (1997). Recommender Systems.
Communications of the ACM. 40 (3). 56-58
4. Schafer, B., Konstan, J., and Riedl, J. 1999. Recommender
systems in e-commerce. In Proceedings of the First ACM
Conference on Electronic Commerce. ACM Press, Denver, CO,
USA, 158–166.
- 262 -