See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/300412849

Online book recommendation system

Conference Paper · September 2015

DOI: 10.1109/ICECCO.2015.7416895

CITATIONS READS

27 12,665

3 authors, including:

Konstantin Nikolayevich Latuta Abay Nussipbekov

Suleyman Demirel University Suleyman Demirel University
5 PUBLICATIONS   48 CITATIONS    4 PUBLICATIONS   39 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Abay Nussipbekov on 27 May 2021.

The user has requested enhancement of the downloaded file.

 Online Book Recommendation System

Nursultan Kurmashov Konstantin Latuta

Faculty of Engineering and Natural Sciences Faculty of Engineering and Natural Sciences
Suleyman Demirel University Suleyman Demirel University
Kaskelen, Kazakhstan Kaskelen, Kazakhstan
nursultan.kurmashov@gmail.com konstantin.latuta@sdu.edu.kz

Abay Nussipbekov
Faculty of Engineering and Natural Sciences
Suleyman Demirel University
Kaskelen, Kazakhstan
abai.nusipbekov@sdu.edu.kz

Abstract — Today the amount of information in the internet II. RELATED WORK
growth very rapidly and people need some instruments to find
and access appropriate information. One of such tools is called Over the last 20 years there has been a considerable growth
recommendation system. Recommendation systems help to in the field of recommender systems. The research works like
navigate quickly and receive necessary information. Generally [1], [2] demonstrate a variety of state of the art methods and
they are used in Internet shops to increase the profit. This paper techniques applied today.
proposes a quick and intuitive book recommendation system that
helps readers to find appropriate book to read next. The overall
Recommendation systems use different kind of approaches
architecture is presented with it’s detailed description. We used a to provide relevant recommendation. Traditionally
collaborative filtering method based on Pearson correlation collaborative filtering and content based filtering are used. The
coefficient. Finally the experimental results based on the online content based filtering approach learns the content of the item
survey are provided with some discussions. i.e. product to categorize it to appropriate user based on his
preferences learned from his profile. The collaborative
Keywords—recommendation system, collaborative filtering, filtering, in contrast, doesn’t rely on content and matches items
book. with users based on the idea that those users who agreed in the
past will also agree in the future. The data about their
I. INTRODUCTION preferences can be collected upon ratings they give on the
Nowadays the amount of information especially in Internet items. One of the successful implementations and use of
growth very rapidly. Finding necessary information becomes collaborative filtering was done by Amazon company that
more difficult. Recommendation systems aim to solve this kind recommends their wide range of products in a very efficient
of problems. With the help of them one can quickly access way. Also there is hybrid recommender system, which combine
relevant information without searching the web manually. As both of previously explained approaches.
such many web sites today benefit from recommendation There are many book recommendation systems for
systems to promote and sell their products. There is a wide today [3], [4], [5], [6]. The authors of [7], [8] use social-media
range of products like music, movies, articles and etc. that can information to learn the preferences of users in order to make
be recommended to the customer based on their profiles in right recommendations. Chang et. al. [9] use such an
internet shops or even social networks, browsing history such information like interest fields, number of clicks to suggest
as visited links, browsing activity like number and time of appropriate book.
visits and other online behavior. Online shops are increasing
their sales using such technologies. In this paper we propose a book recommendation web
service that using collaborative filtering provides users with
In this paper we propose using recommendation recommendations on different genres based on the information
systems for recommending books. We developed a system, of their preferences which they provide while making
which learns user preferences by asking to rate books and registration. The advantage of this system is in its speed and
choosing favorite categories and then generate the list of books simplicity. Most of the existing services need a profile history
user most probably would like to read. information and other information that need some time to
In Section 2 literature review on related researches is provide users with recommendations while our aim was to
provided. Section 3 explains the detailed implementation of the generate recommendations for users in a very quick way.
proposed system. In Section 4 the evaluation procedure with
results is presented. Finally in Section 5 we make some
conclusions.
 III. PROPOSED SYSTEM to quickly find beloved book an intuitive search box with a
Existing recommendation services despite their wide range of search and category listing is provided (see Fig.
powerfulness need a strong user profile information and 3).
history. User register to such systems, browse books, rate them, Fig. 2. Choosing genres
write their feedbacks, recommend to others, share, read
appropriate information and etc. Based on such an information
a system makes its recommendations. The examples of such
services are whichbook.net, whatshouldireadnext.com,
lazylibrary.com and etc. Instead our recommender system
focuses on simplicity and speed. The user makes a registration
and is asked to select 10 favorite books from at least 3
categories (genres). Based on this information the system
makes recommendations. Further the user can continue to rate
the books, buy them and add them to read list and thus allow to
improve the quality of recommendations. The system overview
is demonstrated in Fig. 1. A user, using an intuitive search and
filtering interface updates a database by rating the books and
then gets appropriate recommendations. The recommendations
in turn are calculated based on collaborative filtering method.

Ranking books
User using search /
filtering Database

Fig. 3. Searching and rating a book

C. Providing Recommendations
Recommendations After learning user preferences the system provides
using personalized Collaborative recommendations (see Fig. 4). Recommendations can be seen
interface filtering as a simple list or divided by genres.

A. Collecting Data
Fig. 1. System overview
In recommendation system the amount of data plays an
important role. In order to provide accurate recommendations
one have to provide enough data. For that purpose we decided
to collect data from existing resources. Our service is focused
on Russian-speaking users so we decided to collect data from
www.readly.ru - a web service providing tens of thousands of
books from a variety of different genres. To collect the data we
used Scrappy technology. Scrappy is a special framework for
crawling web sites, which are not provided with API and
extracting data. The principle of this library is simple. Firstly
one needs to define the data he wants to scrape, write a spider Fig. 4. Recommended books
to parse the data and then run the spider. We collected data and In our work we decided to use collaborative filtering for
stored it into MySQL database. The amount of books is about generating the recommendations. Collaborative filtering is an
25.000 popular books. approach to make recommendations on different items for
B. Learning User Preferences users by collecting a bunch of information about their
preferences [10]. As it was mentioned before the idea of a
There are two types of users in the system: guest and collaborative filtering is that if two users have same
registered users. Guest can search for appropriate book, read preferences on a particular item (see Fig. 5 user1 and 2) then
descriptions and see the ratings. Whereas registered users can most likely that they will have same opinions on other items
rate items and see the recommendations of the system. rather than with some other random user. A collaborative
In the very beginning, after making registration a user filtering has different types of realizations. One of them are
asked to choose three or more favorite genres (see Fig. 2) and build on a neighborhood-based algorithms that calculate
then rate minimum 10 books to learn his preferences. In order
similarity between two items. In our prooject we calculate Experiment results are demmonstrated in Fig. 6. They show
similarity by using Pearson correlation functiion: the average of user opinions. Itt can be seen that users liked the
easy of use of the system m i.e. the speed of getting
¦ XY − ¦ N ¦
X× Y recommendations (89%) while the quality of recommendations
was estimated at 77%.
r= (1)
(¦ X ) 2
(¦ Y ) 2

(¦ X −2
) − (¦ Y − 2
) ͻͲ
N N
Similarity r between two users X and Y can have values ͺͷ
between 1 and -1 where 1 means two useers are absolutely
ͺͲ
similar and -1 vice versa.
͹ͷ

͹Ͳ
—ƒŽ‹–›‘ˆ ƒ•›‘ˆ—•‡Ȁ•’‡‡†‘ˆ
”‡…‘‡†ƒ–‹‘ ‘• ”‡…‡‹˜‹‰
”‡…‘‡†ƒ–‹‘•

Fig. 6. Expperiment results

Although the promising results
r of online survey users
indicated several problems relaated to system. One of them was
that sometimes the same bookk appears in different categories
which makes recommendationns less useful. This is because
some of the genres are very cllosely related to each other and
therefore the items of that genrees can appear in both of genres.
V. CONC
CLUSION
In this paper we present a recommendation system that is
Fig. 5. User preferences taable
based on collaborative filteringg method. The main goal was the
We used such technologies like Ajax, JRating, Bootstrap in speed of recommendations i.e. to create such a system, which
order to make our system simpler, user frienddly and intuitive to can give qualitative recommenndations to their users without
use. After learning recommendations a userr is given a list of need to be registered for a loong time and have big profile
links where he can download and buy theses books.
b information, browsing history and
a etc. Experiment results show
that the proposed method provides relevant recommendations.
IV. EXPERIMENT RESULTS
The proposed work can be applied for other domains to
As Shani and Gunawardana say [11], experiments on suggest such items like movies,, music and other products.
recommendation systems are divided into thhree categories: a)
offline experiments: when user behavior is simulated i.e. the REFER
RENCES
process of system making predictions and recommendations
and the user corrects them or uses these reecommendation is [1] G. Adomavicius and A. Tuzhhilin, “Towards the Next Generation of
simulated, b) user studies: analyze the quantitative and Recommender Systems : A Survey of the State-of-the-Art and
qualitative measurements and observe the beehavior of a group Possible Extensions,” IEEE Trans.
T Knowl. Data Eng., vol. 17, no. 6,
of subjects that will interact with the system,
s c) online pp. 734–749, 2005.
evaluation: the system is used by real userss on real tasks and
then analyzed. In our research we decidedd to use an online [2] X. Su and T. M. Khoshgoftaaar, “A Survey of Collaborative Filtering
survey because of to the best of our know wledge there is no Techniques,” Adv. Artif. Inteell., 2009.
available database suitable for our task annd because online
evaluation can not be applied yet. Therefoore we decided to [3] G. Linden, B. Smith, and J. York,
Y “Amazon.com
analyze the performance of our system by creating an online Recommendations: Item-to-IItem Collaborative Filtering,” IEEE
Internet Comput., vol. 7, no. 1, pp. 76–80, 2003.
survey.
We selected a group of independent readeers and asked them [4] S. Kumar Jain and C. Rana, “Building
“ a Book Recommender
to give score (0-10) on such parameters ofo the system like system using time based conntent filtering,” WSEAS Trans. Comput.,
quality of recommendations and conveniencce and ease of use vol. 11, no. 2, pp. 27–33, 20112.
of the system. The higher score indicates the relevance of
recommendations. Also additional field foor user notes was [5] “Whichbook.net.” [Online]. Available:
A http://whichbook.net.
added to the questionnaire. [Accessed: 20-May-2015].
[6] “Lazylibrary.com.” [Online]. Available: [9] Y. Park and K. Chang, “Individual and Group Behavior-based
http://www.lazylibrary.com. [Accessed: 24-May-2015]. Customer Profile Model for Personalized Product
Recommendation.,” Expert Syst. with Appli- cations, vol. 36, no. 2,
pp. 1932–1939, 2009.
[7] M. S. Pera, Y. Ng, and N. Conde, “Personalized Book
Recommendations Created by Using Social Media Data,” in Web
Information Systems Engineering - WISE 2010 Workshops, 2010, [10] M. D. Ekstrand, “Collaborative Filtering Recommender Systems,”
pp. 1–14. Found. Trends® Human–Computer Interact., vol. 4, no. 2, pp. 81–
173, 2010.
[8] I. Guy, N. Zwerdling, I. Ronen, D. Carmel, and E. Uziel, “Social
Media Recommendation Based on People and Tags,” in Proc. of [11] G. Shani and A. Gunawardana, “Evaluating Recommendation
SIGIR, 2010, pp. 194–201. Systems,” Springer, 2011.

View publication stats