Content uploaded by Raghavendra C K
Author content
All content in this area was uploaded by Raghavendra C K on Jun 27, 2021
Content may be subject to copyright.
BOOK RECOMMENDATION SYSTEM: A SYSTEMATIC REVIEW AND RESEARCH
ISSUES
Pooja M Kerudi
UG Student
Computer Science and Engineering
BNM Institute of Technology
Bangalore-560070,India
Email -pooja17cs064@bnmit.in
Pratheeksha M
UG Student
Computer Science and Engineering
BNM Institute of Technology
Bangalore-560070,India
Email -pratheeksha17cs069@bnmit.in
Mr. Raghavendra C K,
Assistant Professor
Computer Science & Engineering,
BNM Institute of Technology,
Bengaluru, 560070, India
Email -raghavendrack@bnmit.in
T Sai Srujan
UG Student
Computer Science and Engineering
BNM Institute of Technology
Bangalore-560070, India
Email -sai17cs105@bnmit.in
Abstract— Recommendation systems are intelligent systems
that employ a large database of information to offer the most
accurate and appropriate products to consumers depending on
their interests. Electronic- commerce (E-commerce) replicates
the behavior of the human being and help to positively change
the behavior of the sellers and consumers. Book
Recommendation System (BRS) recommends a set of books to
users based on their previous ratings. The current work
focuses on content-based filtering, collaborative filtering, and
hybrid filtering, along with many other ways for making
recommendations. This study also highlights restrictions such
as sparsity, cold start, and so on.
Keywords— Book Recommendation system, Hybrid Filtering
technique, Cold start, Sparsity, Content Based Filtering,
Demographic Based Recommender system, Collaborative
Filtering technique.
I. INTRODUCTION
Technology's use in everyday life has increased
due to the impact of its fast advances. Technology is seen in
almost every sector. Technology usage in banking, finance,
media, education, etc. has boomed up in the last two
decades. Likewise, online shopping i.e. Electronic
Commerce (E-Commerce) has made shopping more
convenient. E-commerce deals with the exchanging of
various types of commodities and services through an
electronic network. In the past decade, the usage of E-
Commerce websites has increased substantially. The
contribution of E-Commerce in total retail sales increased
from 5.1% in 2007 to 16.0% in 2019.
With the increase in online shopping, the variety of
items sold online has increased exponentially. Also, the
number of people shopping online is rising day-by-day. Due
to this, the users find it difficult to search the products of
their interest. The tools such as Recommendation systems
come into picture. Recommendation systems can be used to
guide the users with items that they might prefer. They filter
out suitable items from a large collection of items. The
usage of Recommendation systems make it easier for the
users to spot relevant items and improve their experience in
online shopping. The online shopping websites can improve
their revenue with the usage of Recommendation systems.
These recommender system uses users interest, applies some
filtering techniques and recommends the items based on the
data. In some of the E-commerce websites, the system uses
demographic data such as age, gender, location to
recommend the items to the customer that he/she might
prefer.
II. TYPES OF RECOMMENDATION SYSTEM
Recommendation system behaves like human
experts which suggest the desired items to the users based
on decision making ability. Nowadays, these
Recommendation Systems (RS) are used in many sectors
such as financial, medical, agricultural, educational,
entertainment sectors etc. Recommendation Systems (RS)
takes huge amount of data from the users, filters the data
according to the algorithms specified by the user and then
recommends the data based on given data. These
Recommender systems can also be used for personal interest
such as in YouTube, where each user gets a specific set of
personalized recommendations based on the past behavior.
A variety of recommendation system techniques work on
various sources of data. The methodologies used to filter the
data are Hybrid Filtering, Content Based Filtering
Demographic based filtering and Collaborative Filtering .
Figure 1: Types of Recommendation Techniques
A. Collaborative Filtering: Collaborative filtering is the
popularly used filtering technique which is used to find the
like-minded users. Based on the user's recent behavior, the
system predicts the desired item for the user in this filtering.
[1].This filtering evaluates user similarities based on their
ratings and recommends new items based on inter-user
comparisons.
Collaborative filtering techniques are further divided into
two main categories:
Journal of Huazhong University of Science and Technology
ISSN-1671-4512
Vol 50
1
Issue 6
i. Memory based filtering techniques: This is a type of
collaborative filtering which approaches to solve the
problem by considering the entire dataset This algorithm
looks for people who are similar to the current user. The
user to whom the recommendation is being made is referred
to as active user. Users who have comparable behavior and
transaction history to the active user are referred to as
similar users [2]. To find the similarity between the users,
the system uses a similarity measure such as Cosine
similarity, Pearson similarity, Jaccard similarity,
Correlation, etc. [3]. The similarity based on Correlation
ranges in [-1, 1]. Correlation value 1 denotes the highest
similarity and correlation value -1 denotes the least
similarity between a pair of users. This technique is also
called user-based collaborative filtering.
ii. Model based techniques: In order to build a system that is
much faster and scalable the model-based technique can be
used. In this technique “model” is the main keyword which
refers to build the model based on the ratings present in the
dataset. The main advantage of this technique is that the
model doesn’t use the complete dataset but predicts the
scalable output. It is implemented using either explicit
information such as ratings or implicit ratings such as user
interactions and behavior. Modeling can be done using a
variety of machine learning methods such as classification,
cluster analysis, rule-based approaches, and so on [2].
B. Content Based Recommendation:
These recommender systems suggest items which are
similar to the items bought by the user. In this technique,
similarity scores between two items are calculated, based on
which similar books are suggested to the active user. Any
attribute or feature of the item can be considered while
calculating similarity [4].
These recommender systems relate various items based on
their features. User can use two methods in order to
implement content-based similarity:
i. Cosine Similarity- It is a metrics used to find the
similarity between the two items without depending upon
the size of dataset. The two vectors have to be plotted on the
multi-dimensional array, and measure the cosine angle
between two vectors which can be helpful for the user to
find the similarity between two items. Since, multi-
dimensional space is used large-sized dataset can also
support this technique [3].
ii. Decision tree classification- In this approach, the whole
dataset is divided into many small numbers of sets which
helps in predicting the output faster. The tree is constructed
based on the past ratings which are rated by the present or
previous user and based on the contents which are present in
the dataset. Since, it uses the previous data of the users this
technique produces the most appropriate output [5].
C. Demographic Based Recommender System: It is a
kind of Recommender which divides the users into
demographic classes using demographic attributes such as
age, profession, gender, education etc. It works same like
collaborative and content based filtering except it doesn’t
require any historical data. Various personal data taken
during user registration can be used in this approach. For
example, if the demographic attribute considered is
profession of the users, then all researchers are
recommended with a particular set of research related
books, all the teachers are recommended with a particular
set of academic books. One more parameter which can be
used in this technique is the gender, based the gender
specified by the user while registering the system tries to
predict the book/items. Hence, the demographic based
filtering helps the users to predict the efficient product
without the usage of historical data [1].
D. Hybrid Recommendation System: To make more
accurate recommendations, hybrid recommender systems
integrate the features of two, three, or more filtering
algorithms. These recommender systems integrate the good
features of each technique to provide recommendations
more effectively [6]. Hybrid filtering is further divided into
i. Weighted hybridization-In this hybridization
technique, weights are allotted to each technique based on
their performance. Initially, equal weights are allotted to
each technique. The technique which proves to be more
efficient than the other(s) techniques, gradually get greater
weights than the other(s).
ii. Switching hybridization- Works based on the
swapping of two or more recommendation technique, and
uses the efficient technique to recommend the item.
iii. Mixed hybridization- This technique collects all the
results from different recommendation technique at the
same time and combines them to provide desired output.
This technique is used when the data set is very large. In this
type of hybridization each element of the hybrid should be
able to produce desired output and then the system assigns
some weights to the output. The output which has highest
value is considered to recommend the predicted output.
III. LITERATURE SURVEY ON VARIOUS
RECOMMENDATION SYSTEM
Recommendation system is a trending application
in recent times. Recommendation systems are very
important both from the user perspective and the company
perspective. Users can conveniently shop using the
applications with the recommendations being received from
the recommendation systems. On the other hand, an
application’s usage, reachability and revenue can be
improved tremendously with the usage of recommendation
systems. Many experts have tried to improve upon the
current Book Recommendation Systems. They have
proposed various new, innovative ideas for the creation of
efficient recommendation systems. This survey paper
discusses some of such unique ideas.
Madhuri Kommineni, P.Alekhya, T.MohanaVyshnavi,
V.Aparna, K Swetha, V Mounika [3] discussed that User
Based Collaborative Filtering technique along with the
cosine rule is more effective to predict the desired books to
the user. In User based filtering the system finds the similar
preferences of several users and recommends the next book
which like-minded user may like to read. This system is
very helpful for the administration purpose as it collects the
feedback from all the users, report them and analyze the
items and recommends most desired output. User profile as
well as item profile is maintained to find the “User
Journal of Huazhong University of Science and Technology
ISSN-1671-4512
Vol 50
2
Issue 6
Behavior” which is very effective in finding the desired
output. In order to build collaborative filtering
recommendation they have used Singular Value
Decomposition (SVD) model which helps to predict more
efficiently and effectively. The quick sort algorithm is used
to sort the dataset based on the keywords provided by the
users after registering. Historical data should be maintained
properly in this system.
Jayanti Rathnavel and Kavita Kelkar [1] proposed a
personalized recommender for recommending books to the
users. In this experiment, they combined the two popularly,
extensively used recommendation techniques i.e.
collaborative and content based techniques to build a hybrid
recommender. They personalized the system by trying to
understand the interests of the users such as favorite author,
favorite genre, etc. They have addressed the
overspecialization problem. Overspecialization is a
limitation in which the recommended books are similar to
those that the current user has already read. Using lightfm
model, overspecialization is overcome, due to which the
recommended set of books also contains the type of books
not explored by the active user. It gives the opportunity to
the active user to explore new kinds of books. The
recommeder can learn the new interests of the active user.
Anand Shanker Tewari and Kumari Priyanka [7] in their
paper proposed a The Book Recommendation System Based
on Collaborative Filtering and Association Rule Mining
(ARM) for College Students employs the User Based
Collaborative Filtering technique to forecast the top n-rated
books for students and academics. This system aims to help
the students to find books based on the price ranges and
publisher’s name. The system employs categorization
approaches, collaborative filtering based on user input, and
association rule mining. Classification techniques are used
to extract a set of rules and patterns in the data and classify
the data to predefined classes, each class is processed
independently while recommending. Similar people are
detected using Pearson's similarity algorithm in user-based
collaborative filtering. ARM determines the correlation of
each users in the given dataset and associate the relation
between users and finds the best suited items. ARM can also
be used to discover interesting associations and relationships
in the data, which can be used for user behaviour analysis.
Based on these techniques, the system recommends books to
the readers.
Praveena Mathew, Bincy Kuriakose and Vinayak Hegde [4]
According to the author, combining content-based filtering
with collaborative filtering produces more effective and
efficient results. Along with these two techniques,
associative rule is used to predict the desired items from a
large collection of items. This method aims to tackle the
problem of sparsity by combining the techniques of Content
Based Filtering, Collaborative Filtering, and Associative
Rule Mining. The system also implements keyword based
recommendation in which, the users enter keywords related
to their interests and the system compares these words in the
datasets to recommend the books. Equivalence class
Clustering and bottom up Lattice Traversal are discussed in
this paper (ECLAT), which aims to find frequently read sets
of books in an efficient way. ECLAT performs using Depth
First Search (DFS), thus scanning the dataset only once and
consuming less time compared to other algorithms.
Kitti Puritat and Kannikar Intawong [8] have proposed a
model for book recommendation system that uses Support
Vector Machine (SVM). They took into account a variety of
factors, including title similarity and book bibliographic
information like author, year, category, number of books,
etc. This model was specifically designed for usage in small
libraries. SVM is a supervised machine learning model that
can be used to solve classification and regression
applications. The SVM is trained using three sources of data
i.e. title similarity, Dewey Decimal Classification (DDC) for
classification and bibliographic features. The model was
found to perform considerably well.
Huayong Liu and Nianlai Jiao [9] have proposed a hybrid
recommendation system with the usage of context
awareness and social network. Various contextual factors
that affect the user choice on books are obtained through the
context aware layer such as gender of the reader, time of
borrowing the book, etc. A user-book-context matrix is
established to represent the contextual theme suitable for
book recommendation. The contextual factors are associated
with the book type in the matrix and then context aware
computing is performed to obtain entropy and the weight of
each contextual factor. On the other hand, user-to-user
similarity is calculated based on Pearson similarity, based
on which nearest k users are considered. The books to which
the active user has not rated are scored using the
corresponding scores of these k nearest users. The obtained
scores are combined with context weights to obtain final
scores, upon which recommendations are made. It is opined
in the paper that with the usage multiple other context
factors and multi-dimensional context factors, the system
can be further improved.
JiabeiLi, TianweiXu, Juxiang Zhou [10] demonstrates how
to use the hybridization method to effectively use content-
based filtering and collaborative filtering techniques. The
combination of popularity, inverse popularity with similarity
and duration of borrowing of book is considered to measure
the user’s interest and likeability on those books. Inverse
popularity highlights that the users who like unpopular
books have similar interest. While applying for inverse
popularity Borrowing Time has been calculated. Borrowing
Time refers to the time between the borrowal and return of
the book. It reflects the reader’s interest on the book. If a
person’s borrowal duration on a book is short, he/she might
be more interested on it. On the other hand, if it is longer,
he/she might be less interested on it.is the time. Including all
the above mentioned parameters, the recommendation is
done. The scalability issues are overcome in this system
with the usage of cloud.
Dharna Patel, Harish Patidar [6] have proposed a
Recommendation Solution for Online Book Portal and have
explained the need of cloud computing while recommending
the books. The value-added feature in this paper is that the
system gets the profession of the user while registering into
the system. To recommend the book collaborative filtering
technique and content-based filtering techniques are being
utilized. Cloud computing has been used in this paper as
Journal of Huazhong University of Science and Technology
ISSN-1671-4512
Vol 50
3
Issue 6
dataset is very large and it is not always possible to store it
in local disk, which is very difficult to recover when in case
of any loss. In order to secure data, one can adopt cloud
computing, which is also known as a storing centre because
it maintains enormous datasets. This system of
recommendation is more suitable to the readers who require
the best book for general purpose rather than specific
purpose. Raghavendra et al, [18,19] provided an study of
existing techniques, similarity metrics and research
opportunities in this area.
IV. TYPICAL DATASETS FOR BOOK
RECOMMENDATION SYSTEM
Table 1: Description of dataset along with their login
credentials
V. RESEARCH ISSUES
A. Cold start – It is one of the most serious
problems that Recommendation Systems encounter. When a
person joins the system for the first time, he or she will have
no history data. Thus, it is difficult to recommend books to
him/her. This issue is called cold start problem [1,4]. The
same problem arises when there is a new book in the
system, where it is difficult to recommend that book to the
existing users. Many researchers have tried to solve cold
start problem in various approaches. Using the demographic
data of the users, this issue can be overcome [1]. Rather than
using historical data, users' demographic data might be
utilized to recommend books to them. This problem can also
be solved with the help of a Knowledge Graph
Convolutional Network. [11]. Many other solutions to this
problem is necessary to provide significant results.
B. Sparsity – This issue arises when there is a lack of ratings
and reviews by the users. With lesser ratings and reviews it
is difficult to understand the user’s taste and give
recommendations. Thus, reducing the effectiveness of the
recommendation system [1,4,7,10]. A possible solution is to
use Knowledge Graphs to alleviate this problem [12].
Another solution would be to use matrix factorization. This
issue opens up new opportunities for improvement.
C. Trust issues – This issue arises when certain users have
lesser history and it is difficult to recommend books to
them. Also, it occurs when it is difficult to decide the
amount of weightage to be given to reviews and ratings. The
users might have varying tastes, thus making it difficult to
recommend books to them. That is, we do not know how
much we can trust the existing ratings of users with respect
to books. Social network data can be used to reduce the
impact of this issue [9]. There is a scope of research in this
issue.
D. Scalability -With the increase in the number of books and
users in the system, the scalability issue arises. The system
requires more and more resources for recommendation [8].
Also, the performance of the system might not be significant
with the increase in number of books and users. Thus, it is
necessary to develop recommendation models that can face
the scaling up of data.
VI. CONCLUSION
With the adoption of shopping and reading online, the
recommendation systems have become a necessity. There
are multiple types of recommendation approaches that can
be used to build a recommender. These types of
recommenders can be combined together in the hybrid
approach to get better results. Also, there are many issues
faced by that arise with respect to recommendation systems
such as sparsity, cold start, scalability, etc. Thus, more
extensive research and development of models in this field
is necessary to alleviate these issues.
REFERENCES
[1] Jayanti Rathnavel, Kavita Kelkar. “Personalized Book
Recommendation System”. In International Journal Of Engineering
And Computer Science ISSN: 2319-7242, Volume 6 Issue 4 April
2017, Page No. 21149-21153.
[2] P. H. Aditya, I. Budi, Q. Munajat “A Comparative Analysis of
Memory-based and Model-based Collaborative Filtering on the
Implementation of Recommender System for Ecommerce in
Indonesia : A Case Study PT X”
[3] MadhuriKomminen,P.Alekhya,T.MohanaVyshnavi,V.Aparna,KSwet
ha,VMounika. "Machine Learning based Efficient Recommendation
System for Book Selection using User based Collaborative Filtering
Algorithm". International Journal of Advanced Research in Computer
and Communication Engineering, 2018..
[4] Ms.PraveenaMathew, Ms. BincyKuriakose, Mr.VinayakHegde."Book
Recommendation System through Content Based and Collaborative
Filtering Method". In Department of Computer Science And
Engineering, international journal of innovations in engineering and
technology (IJIET), ISSN:2319-1058,vol.2, issue 2, pp1-7, April
2017.
[5] Snehalata B. Shirude.Et.al worked on “Classification of Library
Resources in Recommender System Using Machine Learning
Techniques: 52nd Annual Convention of the Computer Society of
India, CSI 2017, Kolkata, India”(January 2018)
[6] Dharna Patel, Dr. Harish Patidar. "Hybrid Recommendation Solution
for Online Book Portal". In International Journal for Research in
Applied Science & Engineering Technology (IJRASET) ISSN: 2321-
9653; IC Value: 45.98; SJ Impact Factor: 6.887 Volume 6 Issue V,
May 2018.
[7] AnandShanker Tewari, Kumari Priyanka. “Book Recommendation
System Based on Collaborative Filtering and Association Rule
Journal of Huazhong University of Science and Technology
ISSN-1671-4512
Vol 50
4
Issue 6
Mining for College Students”. In 2014 International Conference on
Contemporary Computing and Informatics (IC3I).
[8] Kitti Puritat, Kannikar Intawong. "Development of an Open Source
Automated Library System with Book Recommendation System for
Small Libraries". In 2020 Joint International Conference on Digital
Arts, Media and Technology with ECTI Northern Section Conference
on Electrical, Electronics, Computer and Telecommunications
Engineering (ECTI DAMT & NCON).
[9] Huayong Liu, Nianlai Jiao. "A Hybrid Book Recommendation
Algorithm Based on Context awareness and Social Network". In 2020
3rd International Conference on Advanced Electronic Materials,
Computers and Software Engineering (AEMCSE).
[10] JiabeiLi, TianweiXu, Juxiang Zhou. "A personalized
Recommendation Algorithm for College Books based on User
Interest". In 2020 IEEE 2nd International Conference on Computer
Science and Educational Informatization (CSEI).
[11] MadhuriKomminen,P.Alekhya,T.MohanaVyshnavi,V.Aparna,KSwet
ha,VMounika. "Machine Learning based Efficient Recommendation
System for Book Selection using User based Collaborative Filtering
Algorithm". International Journal of Advanced Research in Computer
and Communication Engineering, 2018.
[12] Anoop A, N Ayush Ubale. "Cloud Based Collaborative Filtering
Algorithm For Library Book Recommendation System". Proceedings
of the Third International Conference on Smart Systems and Inventive
Technology (ICSSIT 2020).
[13] E. Uko Okon, B. O. Eke, P. O. Asagba. "An Improved Online Book
Recommender System using Collaborative Filtering Algorithm". In
International Journal of Computer Applications (0975 – 8887)
Volume 179 – No.46, June 2018.
[14] Dharna Patel, Dr. Harish Patidar. "Hybrid Recommendation Solution
for Online Book Portal". In International Journal for Research in
Applied Science & Engineering Technology (IJRASET) ISSN: 2321-
9653; IC Value: 45.98; SJ Impact Factor: 6.887 Volume 6 Issue V,
May 2018.
[15] Hongwei Wang, Fuzheng Zhang, Jialin Wang, Miao Zhao, Wenjie Li,
Xing Xie, Minyi Guo1."Ripple Net: Propagating User Preferences on
the Knowledge Graph for Recommender Systems1". In arXiv:
1803.03467v4 [cs.IR] 25 Aug 2018.
[16] Cheng Zhao, Chenliang Li, Rong Xiao, Hongbo Deng, Aixin Sun.
"CATN: Cross-Domain Recommendation for Cold-Start Users via
Aspect Transfer Network". In arXiv: 2005.10549v2 [cs.IR] 23 May
2020.
[17] HUAXUAN ZHAO, HONGCHEN WU, JINGZHI LI, HUAXIANG
ZHANG AND XINJUN WANG. "Dtree2vec : A High-Accuracy and
Dynamic Scheme for Real-Time Book Recommendation by
Serialized Chapters and Local Fine-Grained Partitioning". SPECIAL
SECTION ON ADVANCED DATA MINING METHODS FOR
SOCIAL COMPUTING, January 2020.
[18] Raghavendra C K, Srikantaiah K.C, Venugopal K. R, " Personalized
Recommendation Systems (PRES): A Comprehensive Study and
Research Issues", International Journal of Modern Education and
Computer Science(IJMECS), Vol.10, No.10, pp. 11-21, 2018. DOI:
10.5815/ijmecs.2018.10.02
[19] R. C K and K. C. Srikantaiah, "Similarity Based Collaborative
Filtering Model for Movie Recommendation Systems," 2021 5th
International Conference on Intelligent Computing and Control
Systems (ICICCS), 2021, pp. 1143-1147, doi:
10.1109/ICICCS51141.2021.9432354.
Journal of Huazhong University of Science and Technology
ISSN-1671-4512
Vol 50
5
Issue 6
