BOOK RECOMMENDATION USING HYBRID

ALGORITHM
Yarlagadda Pranay(19BCE0766), Joshita R(19BCE2405),
Thaneeshkha B S(19BCE2398), V. Sriram(19BCE0934)
October 2021

Abstract
In recent times, E-commerce has spread all over the world. The e-shops are
not similar to the physical shops. The e-shops can have abundant of items
independent of physical boundaries. The information about all these products
is available on the Internet. So now customer is overloaded with information.
The use of recommendation system helps users explore their interested options
with getting overloaded by information. Some of the existing recommendation
systems like collaborative filtering and content-based filtering have problems
like cold start , data sparsity and other performance issues. In this paper we
have proposed a hybrid recommendation systems which aims to solve some of
the existing problems and provide the users with quality recommendations and
improve the accuracy of the system.

Keywords— Collaborative Filtering, Content Based Filtering, Hybrid Filtering,

Recommendation System

Abstract- Done by Joshita(19BCE2405) and Thaneeshkha(19BCE2398)

1 Introduction
The expansion of the Internet has provided the platform for the e-commerce business
to spread all across the world. Day by day new products are coming in the world
market. It is very hard for a customer to use each and every product available.
So the recommendation system helps users in finding the right products for them
by figuring out their interest and using the experiences of other similar users . The
users gives their judgments or suggestions about the products and services generally
in the form of ratings . The Recommendation System uses these user’s judgments to
form recommendations for them and other users. The RS not only rely on existing
customers but also supports in attracting new customers and increases website sales
over their competitors.

1
2 Objective
The main objective of a recommender system from the customer point of view is to
provide suggestions to online users to make better decisions from a variety of options
available. The main goal from the business point is to increase the sales and profit
Recommending popular items involves lower risk in terms of accuracy. However, it is
not always advisable to recommend only the popular ones. Increasing the diversity of
recommendations is essential in businesses. Increasing the diversity and accuracy is
not in the same direction, thus making it important to make ideal decisions between
the two parameters.

3 Existing Systems
Recommender systems are broadly classified as content-based filtering, collaborative
filtering (CF), hybrid approach, and knowledge-based recommender systems.

3.1 Content Based Filtering

Content-based filtering is one popular technique of recommendation or recommender
systems. The content or attributes of the items the user like are referred to as ”content”
.Here, the system uses the user’s features and interests in order to recommend user
with things that user might like or expect. It uses the information provided by the
user over the internet and the information they are able to gather and then they curate
recommendations according to that. The purpose behind content-based filtering is to
classify products with specific keywords, learn what the user likes, look up those terms
in the database, and then recommend similar things.

3.1.1 Content Based Recommendation

Content-based recommendation is an important approach in recommender systems.
The basic idea is to recommend items that are similar with what user liked before.
The core mission of content-based recommender system is to calculate the similarity
between items. There are a lot of methods to model item and the most famous one
is Vector Space Model. The model extracts keywords of the item and calculate the
weight by TF-IDF. For example, set ki as the ith keyword of item dj , wij is the weight
of ki for dj , then the content of dj can be defined as:
Content(dj ) = {w1j , w2j , ...} (1)
As we talked before, content-based recommender system recommends items that are
similar with what user liked before. So the tastes of a user can be modeled according to
the history of what the user liked. Consider Content Based Profile(u) as the preference
vector of user u, the definition is:
1
ContentBasedP rof ile(u) = sumdN (u) Content(d) (2)
|N (u)|
N(u)is what the user u liked before. After calculating content vector Content(.)
and content preference vector ContentBasedP rof ile(.) of all users, given any user
u and an item d, how the user like the item is defined as the similarity between
ContentBasedProfile(u) and Content(d):
p(u, d) = sim(ContentBasedP rof ile(u), Content(d)) (3)

2
3.2 Collaborative Filtering
Collaborative filtering it filters the information by interactions and the data collected
by the system from the users. It’s main idea is based on the users(people) who agreed
in their evaluation of certain items are also can agree again in the future if needed.
This systems main focus is on the relationship between users and items. Similarity of
items is determined by the similarity of ratings for items by the users who have rated
both items.

Figure 1: Collaborative and content-based filtering

3.2.1 Collaborative-filtering Recommendation

The Collaborative-filtering recommendation is considered as one of the most famous
algorithm in recommender systems. This algorithm models works depending on the
user’s taste according to their history of user behavior. GroupLens was the first one
to publish the first paper about collaborative filtering and the paper also raised user-
based collaborative filtering.

User-based collaborative-filtering In this method, it is considered that a user

will like the items which are liked by the other users with whom they might have the
similar taste. So in user-based collaborative-filtering the first step is to sort out or find
users with similar taste. In the case of collaborative filtering,the users are considered
similar when the users like similar items. So lets consider the users u and v, N(u) and
N(v) are items set liked by u and v users respectively. So the similarity of u and v can
be :
|N (u)| ∩ |N (v)|
suv = (1)
|N (u) ∪ |N (v)|

Item-based collaborative-filtering In this method , it assumes that users will

like items that are similar with items that are liked by the user before. In item-based
collaborative-filtering the first step is to find out the items that are similar with what
the other user liked before. The main point of item-based collaborative-filtering is to
calculate the similarity of two items. Lets assume N(i) and N(j) are user sets who like
i and j items respectively. So the similarity of i and j can be defined as:
|N (i)| ∩ |N (j)|
sij = (2)
|N (i) ∪ |N (j)|

3
 User u’s likeability of item i can be calculated by:
X
pu i = sij pu j (3)
jS(i,k)∩N (u)

3.3 Hybrid Filtering

The hybrid system uses information from the previous user–item interactions and con-
tents of their purchased items. Some of the hybrid methods are Weighted, switching,
mixed, cascade,feature combination and feature augmentation.

Figure 2: Hybrid Filtering

3.4 Knowledge Based Filtering

There might arise some conditions when collaborative and content-based filtering does
not work. The knowledge-based filtering uses the explicit knowledge about the users
and the products . Such systems require knowledge about the group of items and
knowledge about the users .

4 Literature Survey

Sno Title Methodology Gap Identified Future work

1. Generating items This paper has Complex system The approach can
recommendations Combined user- might result in be improved by in-
by fusing content item based collab- scalability prob- corporating users’
and user-item orative filtering lems, and Also context related
based collabora- with content cause overfitting. information in the
tive filtering. based filtering recommendation
and has Assigned system.
Tewari,Anand weights for better
Shanker. recommendations.
Year-2020

4
Sno Title Methodology Gap Identified Future work
2. An Improved Introduced hybrid Training of Improve the ac-
Dynamic Collab- dynamic collab- model Can be curacy by using
orative Filtering orative filtering difficult If the Clustering on The
Algorithm Based (HDCF), which is user item Matrix matrix.
on LDA. based on the LDA is sparse.
model. Added a
Di-Fei, Meng, et time-decay func-
al. tion to the subject
Year-2021 model and gave
its variational
inference model.

3. DNNRec: A novel Hybrid rec- It addresses the White-boxing

deep learning ommendation ratings predic- neural network
based hybrid System (user-item tion case of the predictions is
recommender based matrix), recommender tough and ex-
system. Deep learning systems problem plaining the “why
(using leaky ReLu and not the of predictions of
Kiran, R., activation layer. ranking case or deep neural net-
Pradeep Ku- the streaming works is tougher .
mar, and Bharat case. This Area can be
Bhasker. of Use in future
Year-2020 Study.
4. College library Combined collab- This paper Can merge this
personalized rec- orative filtering, doesn’t solve as Module into
ommendation Content Based the cold start other bigger
system based on filtering and ap- problem which is applications.
hybrid recommen- plied K-means a very common
dation algorithm. clustering on the p Inrecommen-
user-item matrix. dation Systems.
Tian, Yonghong,
et al.
Year-2019

5. A hybrid rec- Combined collab- Again use of The model uses

ommendation orative filtering, too many sys- Explicit feed-
system for QA Content Based tems might lead back.In the future
documents. filtering and to scalability implicit feedback
complementary problems. Can also be used.
Li, Ming, et al. based recommen-
Year-2020 dation system
and trained it on
the modelled QA
documents.

5
Sno Title Methodology Gap Identified Future work
6. A hybrid rec- Content based The‘Recommendation Can work on The
ommendation recommendation dimensons’ mod- user-item matrix
algorithm–based algorithm, item ule is called Fre- to improve effe-
intelligent busi- based collabo- quently which ciency.
ness recommenda- rative filtering might lead to
tion system. recommenda- performance
tion algorithm issues.
Yang, Fan. and demography
Year-2018 based.

7. Artist recommen- User based collab- The model di- The system could
dation system orative filtering versifies the be made dynamic
using hybrid combined with recommenda- by including user
method: A novel CBF. tions too much feedback for the
approach. reducing person- recommendations
alization. provided and
Dhruv, Ajay, et al. replacing those
Year- 2019 recommendations
which the user did
not like with new
ones.
8. Automatic rec- User based Col- Cold start prob- CBF faces over
ommendation laborative filtering lem, Sparsity specialization and
system based on (Nearest neigh- and Scalability. content extraction
hybrid filtering bour approach problems.It is to
algorithm. is also used) , work with online
Content based recommender sys-
Sunny Sharma filtering (CBF) tem to check the
,Vijay Rana, and Association user level accep-
Manisha Malho- rule mining. tance , since the
tra. current research
Year-2021 work is based on
offline dataset.

9. Online Book Jaccard Similarity Overspecialization, To solve the trust

Recommendation , Root Mean Data sparsity, issues in the case
System using Square Error Cold start, of rating for gen-
Collaborative (RMSE) used Scalability and uine recommenda-
Filtering (With for evaluation ranking of rec- tion of books for
Jaccard Similar- metrics. ommendation. the user.
ity).

Avi
Rana,K.Deeba.
Year-2019

6
Sno Title Methodology Gap Identified Future work
10. A Collaborative User based Col- Sparsity of rat- It is to enhance
Filtering Based laborative filtering ing matrix. future works in
Library Book and Association user’s satisfaction
Recommendation rule mining. and accuracy by
System. adopting book
bibliography like
Chaloemphon publisher , cate-
Sirikayon, Panita gory , author and
Thusaranon, other features.
Piyalak Pongtaw-
evirat.
Year-2018

11. A Novel Approach Frequent pat- More storage Having concept

for Book Rec- tern intersect and expen- generation ap-
ommendation algorithm(FP sive, Apri- proaches instead
Systems. Intersect Algo- ori(traditional of dealing with
rithm). algorithm). keywords in
P.Devika,R.C.Jisha recommendations.
and G.P.Sajeev.
Year-2016
12. Introducing Hy- Collaborative Sparsity in rat- Increasing the
brid Technique Filtering (Slope ings. mapping of user-
for Optimization one algorithm) item rating matrix
of Book Recom- , Content , due to decline
mender System. based filtering fall in precision.
(LSH/MinHash
Manisha Chan- algorithm) along
daka ,Sheetal Gi- with Demographic
raseb , Debajyoti attributes . On-
Mukhopadhyay. tology approach
Year-December also been used.
2015

13. LIBRS-Library Collaborative fil- It depends on Enhance the

Recommendation tering techniques- rating and sug- library system
System Using Slope one(Rating gest books which user-friendly to
Hybrid Filtering. prediction), are not preferred the user.
Pearson’s Corre- by the user.
Pooja Rani , lation(User and
Pratima Solanki user relation-
, Vedashree ship) and Cosine
Puntambekar, method(Item and
Vivek Borse ,Mrs item relationship)
A.S.Vaidya. .
Year- 2014

7
Sno Title Methodology Gap Identified Future work
14. Product Rec- Item similar- Cold start prob- Recommendation
ommendation ity,Bipartite pro- lem,Stability of products in
Systems Based on jection,Spanning and plasticity more precisely
Hybrid Approach tree. Content problem(tough and accurately.
Technology. based and Collab- to change user’s
orative filtering. profile),Eliciting
Ndengabaganizi the user’s feed-
Tonny James, back.
Dr.K.Rajkumar.
Year-2017

15. E-Learning Rec- User-based collab- Extracting To pay more

ommendation orative filtering negative associ- attention to
System. method combined ation rules from recommend ac-
with online edu- frequent item- cording to tags of
Huiyi Tan , Junfei cation,Association sets;Extraction each course.
GuoYong Li rule. of positive as-
Year-December sociation rules
2008 from infrequent
itemsets.
16. Improving a Hy- Collaborative Cold Start Assess if the book
brid Literary filtering Algo- problem,Hard choice is related
Book Recommen- rithm, Author to include side to content, user
dation System recommenda- features for location, and the
through Author tion (to improve query/item. month in which
Ranking. the book top-n the book was read.
recommendation. The use of fea-
Paula Cristina ture augmentation
Vaz ,David Mar- and dimensional-
tins de Matos ity reduction tech-
Year-June 2012 niques will also be
considered.

17. Book Recommen- Association rule Many parame- Will include other
dation System for Mining,content- ters are needed. factors to analyse
Digital Library based book Low comprehen- relations, apply
Based on User recommending sibility. Huge other data mining
Profiles by Using technique for text size of the result techniques and
Association Rule. categorization. set. compare with the
developed model.
Pijitra Jomsri
Year-October
2014

8
Sno Title Methodology Gap Identified Future work
18. Web-based Per- Collaborative Recommendation Enhance the
sonalized Hybrid filtering, Content- based on exist- accuracy of the
Book Recommen- based filtering and ing features recommender
dation System Knowledge-based only. Lack of systems over time.
filtering. novelty and
Salil Kanetkar Ak- diversity. Scala-
shay ,Nayak Srid- bility.Attributes
har Swamy , Gre- may be inconsis-
sha Bhatia. tent.
Year-August 2014

19. Book Recommen- User based collab- With large data, To expand its use
dation System orative filtering, prediction is to wider users not
Based on Collab- Association rule, slow. Require only for students,
orative Filtering Nearest neigh- high memory to Classify the books
and Association bour. store data. Ac- with more con-
Rule Mining for curacy depends strains.
College Students on the quality of
data.
Anand Shanker
Tewari, Kumari
Priyanka
Year-January
2015
20. Enhancing the Collaborative Sparsity in To study further
Performance of Filtering(CF) metrics. Rep- the potential of
Library Book approach, resentation of building the at-
Recommenda- Probabilistic- Joint Probability tribute model by
tion System by keyword ap- Set of random combining the at-
Employing the proach. variables is in tributes data of
Probabilistic- exponential both the users and
Keyword Model which is very items to solve the
on a Collabo- large. sparsity problem.
rative Filtering
Approach

Noor Ifadaa,Irvan
Syachrudina
,Mochammad
Kautsar Sophana
Year-September
2019

9
Sno Title Methodology Gap Identified Future work
21 A hybrid recom- User based col- Data Sparsity To overcome the
mender system laborative filtering The model has lack of user prefer-
for patron driven and content based limited to ex- ences and reviews
library acquisition filtering pand on the that are internal
and weeding. user’s existing in the library, we
interests. aim to include
Maryem Rhanoui active learning in
,Mounia Mikram the recommender
,Siham Yousf system ,so that it
,Ayoub Kasmi helps actively in-
Year-October volving the users
2020 and acquiring
their reviews.
22 Book Recommen- Demographic It takes lot of The use ofdata
dation System technique, content time to identify cleaning, which
through Content based filtering, the needs of the include removing
Based and Collab- collaborative fil- customer. Trust the irrelevant data
orative Filtering tering and hybrid issues. and keeping the
Method. method, time relevant data for
Praveena based content mining.
Mathew; Bincy filtering, ECLAT
Kuriakose; Algorithm.
Vinayak Hegde.
Year-16thMarch
2016

23 Content-Based Collaborative fil- We cannot plan We can present

Book Recom- tering, Rank cor- to explore richer a quantitative
mending Using relation, LIBRA. representations picture of per-
Learning for Text and their abil- formance on a
Categorization. ity to improve realistic sample;
Raymond J, recommendation books to be rated
Mooney, Loriene performance. where selected at
Roy. randomly.
Year-01 June 2000

24 An Improved Collaborative- Accuracy is not We should target

Online Book Rec- filtering, k-NN determined by on securing rec-
ommender System approach, Python high number ommender system
using Collabo- algorithm for of ratings,Gray data against at-
rative Filtering sorting rating, sheep, shilling tacks and also im-
Algorithm. root mean square attacks, privacy proving on the al-
E. UkoOkon, B. error (RMSE) protections. gorithms used.
O. Eke.
Year-June 2018

10
 Sno Title Methodology Gap Identified Future work
25 Online Book Collaborative- Sparsity, Cold we can rec-
Recommendation based filtering start. ommend the
System by using and association recommendation
Collaborative rule mining. list appearing
filtering and Asso- according to au-
ciation Mining. thor depends on
database.
SuhasiniParvatikar,
Dr. Bharti Joshi.
Year-21 March
2016

26 Online Book Collaborative Large data pre- To generate rec-

Recommendation filtering, Rank- diction is slow. ommendations for
System. ing books using Qualitative rec- users in a very
search / filter- ommendation is quick way.
Nursultan Kur- ing, Scrappy low.
mashov, Kon- technology, Pear-
stantin Latuta. son correlation
function.
Year-25 February
2016

5 Proposed System
In our proposed method , we will be using a hybrid recommendation system which
comprises of two systems, namely content based recommendation and collaborative
filtering.
To solve the sparsity problem of the user-item matrix we will be using the clustering
algorithm . Overall our system consists of three parts :
1. Preprocessing the data and clustering .
2. Training the CBF and CF algorithms individually.
3. Combining the results and presenting them on the website.
This proposed method aims to remove the sparsity problems which leads to per-
formance problems. The combination of 2 recommendation systems also results in a
well balanced list of recommendations which was a problem individually.

5.1 Dataset Used

The dataset being used is the “Goodreads review dataset”. We will be using the Flask
framework to develop the book recommender website.

5.2 Architecture Diagram

The Proposed Architecture Diagram:

11
 Figure 3: Architecture Diagram

References
[1] Manisha Chandak, Sheetal Girase, and Debajyoti Mukhopadhyay. Introducing
hybrid technique for optimization of book recommender system. Procedia Com-
puter Science, 45:23–31, 2015.
[2] P Devika, RC Jisha, and GP Sajeev. A novel approach for book recommendation
systems. In 2016 IEEE international conference on computational intelligence
and computing research (ICCIC), pages 1–6. IEEE, 2016.
[3] Ajay Dhruv, Aastha Kamath, Anuja Powar, and Karan Gaikwad. Artist recom-
mendation system using hybrid method: A novel approach. In Emerging Research
in Computing, Information, Communication and Applications, pages 527–542.
Springer, 2019.
[4] Meng Di-Fei, Liu Na, Li Ming-Xia, and Su Hao-Long. An improved dynamic
collaborative filtering algorithm based on lda. IEEE Access, 9:122568–122577,
2021.
[5] Noor Ifada, Irvan Syachrudin, Mochammad Kautsar Sophan, and Sri Wahyuni.
Enhancing the performance of library book recommendation system by employing
the probabilistic-keyword model on a collaborative filtering approach. Procedia
Computer Science, 157:345–352, 2019.
[6] Ndengabaganizi Tonny James and K Rajkumar. Product recommendation sys-
tems based on hybrid approach technology. International Research Journal of
Engineering and Technology, 4(8), 2017.
[7] Pijitra Jomsri. Book recommendation system for digital library based on user
profiles by using association rule. In Fourth edition of the International Conference

12
 on the Innovative Computing Technology (INTECH 2014), pages 130–134. IEEE,
2014.
[8] Salil Kanetkar, Akshay Nayak, Sridhar Swamy, and Gresha Bhatia. Web-based
personalized hybrid book recommendation system. In 2014 International Confer-
ence on Advances in Engineering & Technology Research (ICAETR-2014), pages
1–5. IEEE, 2014.
[9] R Kiran, Pradeep Kumar, and Bharat Bhasker. Dnnrec: A novel deep learning
based hybrid recommender system. Expert Systems with Applications, 144:113054,
2020.
[10] Nursultan Kurmashov, Konstantin Latuta, and Abay Nussipbekov. Online book
recommendation system. In 2015 Twelve International Conference on Electronics
Computer and Computation (ICECCO), pages 1–4. IEEE, 2015.
[11] Ming Li, Ying Li, Wangqin Lou, and Lisheng Chen. A hybrid recommendation
system for q&a documents. Expert Systems with Applications, 144:113088, 2020.
[12] Praveena Mathew, Bincy Kuriakose, and Vinayak Hegde. Book recommendation
system through content based and collaborative filtering method. In 2016 Inter-
national conference on data mining and advanced computing (SAPIENCE), pages
47–52. IEEE, 2016.
[13] Raymond J Mooney and Loriene Roy. Content-based book recommending using
learning for text categorization. In Proceedings of the fifth ACM conference on
Digital libraries, pages 195–204, 2000.
[14] E Uko Okon, BO Eke, and Prince Oghenekaro Asagba. An improved online
book recommender system using collaborative filtering algorithm. International
Journal of Computer Applications, 179(46):41–48, 2018.
[15] Suhasini Parvatikar and Bharti Joshi. Online book recommendation system by
using collaborative filtering and association mining. In 2015 IEEE International
Conference on Computational Intelligence and Computing Research (ICCIC),
pages 1–4. IEEE, 2015.
[16] Avi Rana and K Deeba. Online book recommendation system using collaborative
filtering (with jaccard similarity). In Journal of Physics: Conference Series,
volume 1362, page 012130. IOP Publishing, 2019.
[17] Pooja Rani, Pratima Solanki, Vedashree Puntambekar, Vivek Borse, and
AS Vaidya. Librs: Library recommendation system using hybrid filtering. Int. J.
Technical Research and Applications, 2(2):78–81, 2014.
[18] Maryem Rhanoui, Mounia Mikram, Siham Yousfi, Ayoub Kasmi, and Naoufel
Zoubeidi. A hybrid recommender system for patron driven library acquisition and
weeding. Journal of King Saud University-Computer and Information Sciences,
2020.
[19] Sunny Sharma, Vijay Rana, and Manisha Malhotra. Automatic recommendation
system based on hybrid filtering algorithm. Education and Information Technolo-
gies, pages 1–16, 2021.
[20] Chaloemphon Sirikayon, Panita Thusaranon, and Piyalak Pongtawevirat. A col-
laborative filtering based library book recommendation system. In 2018 5th In-
ternational Conference on Business and Industrial Research (ICBIR), pages 106–
109. IEEE, 2018.

13
[21] Huiyi Tan, Junfei Guo, and Yong Li. E-learning recommendation system. In 2008
International conference on computer science and software engineering, volume 5,
pages 430–433. IEEE, 2008.
[22] Anand Shanker Tewari. Generating items recommendations by fusing content
and user-item based collaborative filtering. Procedia Computer Science, 167:1934–
1940, 2020.
[23] Anand Shanker Tewari and Kumari Priyanka. Book recommendation system
based on collaborative filtering and association rule mining for college students.
In 2014 International Conference on Contemporary Computing and Informatics
(IC3I), pages 135–138. IEEE, 2014.
[24] Yonghong Tian, Bing Zheng, Yanfang Wang, Yue Zhang, and Qi Wu. College
library personalized recommendation system based on hybrid recommendation
algorithm. Procedia CIRP, 83:490–494, 2019.
[25] Paula Cristina Vaz, David Martins de Matos, Bruno Martins, and Pavel Cal-
ado. Improving a hybrid literary book recommendation system through author
ranking. In Proceedings of the 12th ACM/IEEE-CS joint conference on Digital
Libraries, pages 387–388, 2012.
[26] Fan Yang. A hybrid recommendation algorithm–based intelligent business recom-
mendation system. Journal of Discrete Mathematical Sciences and Cryptography,
21(6):1317–1322, 2018.
[22] [4] [9] [24] [11] [26] [3] [19] [16] [20] [2] [1] [17] [6] [21] [25] [7] [8] [23] [5] [14] [15]
[10] [13] [12] [18]

14