Proceedings of the Third International Conference on Computing Methodologies and Communication (ICCMC 2019)

IEEE Xplore Part Number: CFP19K25-ART; ISBN: 978-1-5386-7808-4

A Survey on Recommendation System for

Bigdata using MapReduce Technology
Maulik Dhamecha Krupa Dobaria Dr. Tejas Patalia
Research Scholar Research Scholar Professor & HOD
Computer Engg. Department, Computer Engg. Department, Computer Engg. Department,
Gujarat Technological V.V.P. Engineering V.V.P. Engineering
University, College, College,
Ahmedabad, Gujarat, India Rajkot, Gujarat, India Rajkot, Gujarat, India
[email protected] [email protected] [email protected]

Abstract— Recently recommender system is emerging as a clients items coordinating their interests and ideally
promising way of communication between user and transform them into conferred purchasers.
massive amount of data. Recommendation system
organizes the data in large amount to determine the Every recommendation system consists of two part,
concentration of analyst and make the information first is a user and the second is an item. We can define
retrieval process easier for the users. For this purpose and user who consume various products, items, services who
criteria, large scale recommendation systems are available has a speculation of that item. We can supply database
such as shopping website, movie, music etc. Many as an input to recommendation algorithm and find the
researchers have developed several methods or algorithms like different items as output for the recommendation.
like collaborative filtering, keyword based filtering etc. Here, inputs includes a dataset of consumer and database
Here we proposed University Recommendation System from universities and output generates the opinion of
with keyword based sentiment analysis as an underlying university.
algorithm. The proposed work introduces customized and
personalized item recommendation and efficiently There are basically three approaches for
formulates the most suitable item to the user. Here, we recommendation system. The first one is “Collaborative
have designed and implemented our system based on the Filtering (CF) algorithm” and the second is “Content-
Hadoop framework using MapReduce technology. based Filtering algorithm”. In CF approach, it constructs
a model from user’s browsing history as well as similar
Keywords— Big data, Hadoop, MapReduce, decision made by other user to predict the products that
Recommendation algoritm, Collaborative Fitering algorithm, the client may have an enthusiasm for. In content-based
Content-based filtering algorithm filtering approach, the characteristics of an item are
I. INTRODUCTION analysed to recommend items to user. Late research has
shown that a composite approach, integrating content-
In previous approaches people rely on reviews of based and collaborative filtering which could be more
their friends or relatives for purchasing any product or viable at few times. Hybrid approaches can be
visiting any place. People prefer recommendation from performed in different ways: first, by solving both
friends rather any strangers. Nowadays in the era of methods forecasts independently and after that joining
information abundance, the number of users who are them. Second, by including content-based abilities to a
used the internet, are increase in very fast manner and CF technique and vice-versa.
information relevant to web is also growing epidemic.
Then also it has number of issues like on the very first With the use of recommendation system, human
site, its complex for users to search any key value from interaction with internet, create new milestone. They can
the higher cumulative of huge density of data. Then expand the era of interactivity and with this they can
again, there is a lot of data overload for the users and elaborate the users’ expectation. The main focus of using
they spend too much time in selection the product. recommendation framework is to enhance the products
Hence, this recommendation system is recognize to sell and increase profits criteria and for that following
summarize these queries and fix on choices by targeting are the technical goals: [1][2][4]
the accuracy of a user [1][13].  Relevance - Clients will probably expend the
Recommender system is playing a supreme role in items they find a fascinating.
our daily life. Relevant to the users’ etiquette, to build a  Novelty – those kind of things which user has
binary model with the user and relevant items, not seen any time before, recommender system
recommendation system is used and also enables clients should recommend that result to target user.
to rapidly find alluring things without being Rehashed proposal of well-known things can
overpowered by unessential data [9]. It is of awesome likewise prompt diminishment in deals assorted
enthusiasm for sellers to prescribe to their potential variety.

978-1-5386-7808-4/19/$31.00 ©2019 IEEE 54

 Proceedings of the Third International Conference on Computing Methodologies and Communication (ICCMC 2019)
IEEE Xplore Part Number: CFP19K25-ART; ISBN: 978-1-5386-7808-4

 Serendipity – as we discuss the  Changing User Preferences – It happens often

Recommendation should be unexpected and the changing of users’ likes and dislikes. So it
surprising to user and that’s why this goal is is difficult to recommend anything to the user.
different from novelty. Let’s take an example In such case user will lose trust in
of this. If we found that user likes action kind recommendation system and stop using it. So it
of movie but every time we cannot recommend is necessary to maintain the quality of
action kind of movie, because these kind of recommendation [18].
movie will be highly similar and for this reason
user may lost their interest. Recommender  Privacy – Privacy has been the most imperative
system should elaborate user to find and query. Deciding the final goal in mind to get
explore new range of interest. the final exact and perfect view, the framework
must obtain the most measure of data
 Diversity – whenever any recommendation conceivable about the client, including the
include the fundamentally same as elements, it statistic information and information about the
increase the threat that the objective client may location of a specific client. Normally the
like on one of them. If the recommendation subject of dependability, security and
contains different types of elements then there confidentiality of the given data emerges [14].
is a more noteworthy possibility that the client
may like minimum one of these products.
II. RELATED WORK
Different strategies utilized as a part of a
recommender framework encounters a portion of the Whenever we are talking about intelligent element
obstacles that might be portrayed as far as essential retrieval and accurate prediction we have to take
issues as [2]: consideration of recommendation theory. Main phase of
the recommendation system is a recommendation
 Cold-Start Problem – It alludes to the engine. It will directly connect with performance of
circumstance when another client or product recommendation. Here we are given literature survey on
just occur the entry into the system, generally the existing recommendation system for other
there are three kinds of cold start problems applications:
generated: if found new user then create issue,
if found new item then create issue and if found Shunmei Meng et al [5] used to demonstrate “clients'
new system then also create issue [15][19]. For inclinations and a client based Collaborative Filtering
these kind of queries, it is hardly expectation to calculation to create proper proposals”. The dynamic
give suggestion as new client event, if there are client gives his/her inclinations by choosing the
few data relevant to client that they can catchphrases from the watchword hopeful rundown, and
accessible and further new product availability, the inclinations of the past clients can be extricated from
no typically reviews are available for users. their audits for administrations as indicated by the
watchword competitor rundown and area thesaurus.
 Data Sparsity – Sparsity issue is one of the
G.Karthika et al [6] has suggested system which
significant issues experienced by recommender
manages the execution to give administrations for hotel
framework and information sparsity has
reservation system. Supervised learning algorithms are
awesome effect on the nature of suggestion.
used to identify sentiment orientation. It determines
The primary purpose for information sparsity is
positive or negative opinion along with negation word
that most clients don't rate the vast majority of
near to each opinion word and then identifies the number
the products and the accessible reviews are
of positive and negative opinions of reviews. Keywords
normally scanty. Collaborative filtering
with positive opinion are considered and similarity is
experiences this issue because it is resting on
ascertained between client inclinations with surveys of
the rating matrix in most cases [3].
the past client by Jaccard and Cosine measures.
 Scalability - System’s capacity to deal with Sanya Sharma et al [7] has proposed that “composite
developing measure of data in an effortless way search algorithm that depends on Cosine similarity
is known as scalability and it’s a most function and rating given by other users”. According to
important property. With colossal development analysis, proposed algorithm refines information on the
in data over web, clearly the recommender bases of traits and also client given appraisals. History of
frameworks are having a blast of information client assumes an imperative part in perceiving client's
and subsequently it is an incredible test to deal interests.
with constantly developing interest [18]. Many
of them recommendation algorithms compute Khushboo R. S. et al [8] utilized client inclinations
incrementally with exploring number of users and competitor benefit quality. Comparable clients are
and products. then arranged utilizing client based synergistic

978-1-5386-7808-4/19/$31.00 ©2019 IEEE 55

 Proceedings of the Third International Conference on Computing Methodologies and Communication (ICCMC 2019)
IEEE Xplore Part Number: CFP19K25-ART; ISBN: 978-1-5386-7808-4

separating calculation. These comparative clients' Collaborative filtering is the technology that focuses on
certain, negative surveys and conclusions in the content the relationship between users and between items to
are separated. Sentiment analysis is used for score make a prediction.” They focus on the reviewing two
calculations. strategies of collaborative filtering: “user-based and
item-based recommendations.”
Feiyue Ye et al [9] has proposed “a CF
recommendation algorithm based on users’ choice and Lun-Chi Chen et al [10] similar client area issues
the association of items. Traditional collaborative were examined and the MapReduce beast constrain
filtering recommendation algorithms present calculation was presented, to improve collaborative
shortcomings such as sparsity of rating information and filtering technology for libraries utilizing a parallel
single appraisals closeness.” To take care of this issue, subtask of the MapReduce algorithm. This kind of
they exhaustively think about the impact of clients' MapReduce algorithms which has parallel subtask,
inclinations for clear cut properties of things and the comprising of keyword likeness calculation and
connection of things on the counts of likenesses. auxiliary similitude calculation, depend on the
transformed list approach utilizing MapReduce. In view
Elena Shakirova [1] used “collaborative filtering of the recreation, the proposed framework enhances the
techniques for a music recommender system. execution of recommender frameworks for libraries.

Sr. Techniques and

Author Name & Title Approach Used Advantages Disadvantages
No. algorithm used
1 “KASR: A Keyword-Aware User-based Jaccard Scalable, Less accuracy of
Service Recommendation Collaborative Coefficient and More productive Jaccard
Method on MapReduce for Filtering approach Cosine Similarity than customary Coefficient
Big Data Applications” [5] strategies method.
Positive marking
of client and
negative marking
of client are not
separated.
Basic content of
sentiments are not
considered for
valuation.

2 “Collaborative Filtering With Collaborative Keyword Rating of each Small set of

Semantic Based Service Filtering approach extraction, keyword is keyword is used.
Recommendation Using On Jaccard calculated. Take more time
MapReduce” [17] Coefficient and for execution.
Cosine Similarity
3 “Recommender System using Composite Composite Refines User’s search
Hybrid Approach” [8] approach Search algorithm, information on history not
(Collaborative Nearest the bases of considered.
Filtering and Neighbour ascribed and
Content-based algorithm, additionally client
Filtering) Cosine Similarity given appraisals.
4 “Hotel Recommendation Collaborative Sentiment Hadoop is utilized There is a cold
System using Hadoop and Filtering approach analysis to expand start and ratings
MapReduce for Big Data” [8] versatility and sparseness
information problem.
security.
5 “A Collaborative Filtering Collaborative Cosine Similarity Contemplated Need to improve
Recommendation Based on Filtering approach Method and interest of in massive
Users’ Interest and Modified Cosine customers and websites
Correlation of Items” [13] Similarity relational analysis recommendations.
Method, its relevant items
Also create set of

978-1-5386-7808-4/19/$31.00 ©2019 IEEE 56

 Proceedings of the Third International Conference on Computing Methodologies and Communication (ICCMC 2019)
IEEE Xplore Part Number: CFP19K25-ART; ISBN: 978-1-5386-7808-4

nearest neighbour
users
6 “Collaborative Filtering for Collaborative User-based and Prepared only Need to analyse
Music Recommender Filtering approach Item-based theoretical basis different
System” [1] collaborative for recommender constraints such as
approach system similarity ration
and functional
ability to improve
system efficiently.

7 “Ontology-based Collaborative MapReduce Similar client area Need to explore

Recommender System using Filtering approach Brute-Force issues were pairwise similarity
MapReduce” [10] algorithm explored, between
Ascertain ontologies.
Keyword
closeness and
Structural
Similarity

III. APPROACHES FOR

RECOMMENDATION From all available methods of recommendation, this
collaborative filtering method is one of the best
Recommendation techniques depend on the data they recommender method [21]. With compare to other
used to ratify products tactics will be separated into filtering methods, we can says that Collaborative
many categories. From that, most of the studies suggest filtering technique is far better to full fill all the
the mainly below four technique. [11] requirements. It’s a methodology for generating
 Collaborative Filtering technique anticipations about customer’s interests by consuming a
 Content based Filtering technique datasets of proclivities of items which may another
 Hybrid Approach client interested [11]. Customer who want the same
result in the previous have to prone with same kind of
proclivities. Improvement into the quality of the
A. Content-based Filtering technique
collaborative filtering method based on the Customer’s
valuation; it will reject the value that can be isolated
On the basic criterion we can say that, whenever we are
from semantic database. Hence, the process of filtering
talking about contents of items in a database, it was
technique depend on the customer’s semantic database.
known as content based filtering technique. It relies on
the contour of users’ analysis perceptron [12]. After
C. Hybrid approach
evaluated the content of customer suggested items, we
can summarise the recommendation of the university.
Hybrid approach founded to combine different
This kind of approach summarised separate customer
techniques of recommendation to reduce the
rated items and uses the datasets of universities to
disadvantages of their common. To find the relevant
generate a history to disparity with other items which are
interest customer, many times hybrid technique adequate
not rated to recommend, like vies any search through its
customer’s data and detail of their relevant items and
engine, recommendation of particular pages is depend
then applied collaborative filtering technique to generate
on the substances of the user requirements [13][16].
perception. For this kind of systems, the most important
Main demerits of this approach is that it requires
thing is to cautious acceptance of parameters. Different
particular datasets to observe, however, as the datasets
filtering techniques will be combined into many
may be limited or weak and therefor datasets availability
different traditions. Hybrid techniques can be
is unusually. Content based technique shows the harsh
implemented into several ways such as – Cascade,
result over specifications, as its produce recommend
Switching, Feature combination, Feature Augmentation,
only which item was relevant to early suggested by the
Meta level, Weighted and Mixed [8].
customer [20].

B. Collaborative filtering approach  Cascade: this is very basic level method. In this
process, higher level option cannot be overruled

978-1-5386-7808-4/19/$31.00 ©2019 IEEE 57

 Proceedings of the Third International Conference on Computing Methodologies and Communication (ICCMC 2019)
IEEE Xplore Part Number: CFP19K25-ART; ISBN: 978-1-5386-7808-4

by lower level. Only possibility is they can [3] Sharma, L., & Gera, A. (2013). A survey of recommendation
system: Research challenges. International Journal of
allow to polish that criterion. Engineering Trends and Technology (IJETT), 4(5), 1989-1992.
 Switching: at the time of particular situation, [4] Nayana, Vaidya; & Khachane, A, R; (2017, June). Keyword
system will decide that which recommender based recommender system for electronic products using weight
will applied based on the user contour and the based recpmmendation algorithm implemented on Hadoop. In
environment of recommender output. There is International Journal of Advances in Electronic and Computer
Science (IJAECS), 2017 International Conference, vol. 4.
option to take any recommendation
[5] Maulik Dhamecha, Dr. Tejas Patalia : MapReduce Foundation of
mechanisms and utilise it. Big Data With Hadoop Environment, ELSEVIER(SSRN) –
 Feature Combination: to generate any specific 2018, ISSN: 1556-5068
recommendation algorithm, use more than one [6] Krishna Parmar, Maulik Dhamecha : Feature based Composite
feature from which they can get the require Approach for Sarcasm Detection using MapReduce, IEEE-2018,
ISBN:978-1-5386-3452-3
output.
[7] Rathnavel, J., & Kelkar, K. Personalized Book Recommendation
 Feature Augmentation: to create any new System.
technique, use one already given recommended [8] Meng, S., Dou, W., Zhang, X., & Chen, J. (2014). KASR: A
technique and after generating that new keyword-aware service recommendation method on mapreduce
technique use for the next level processing. for big data applications. IEEE Transactions on Parallel and
Distributed Systems, 25(12), 3221-3231.
 Meta level: generate one basic model to acquire
[9] K. Jeyaganesh Kumar, & G.Karthika. (2017, March).
knowledge of recommender and use that model Collaborative Filtering With Semantic Based Service
to take response for new method. Recommendation Using On Mapreduce. In SSRG International
 Weighted: at initial level applies collaborative Journal of Computer Science and Engineering- (ICET’17) -
Special Issue, 2348 – 8387.
filtering method and content based filtering
[10] Goral Godhani, Maulik Dhamecha : Simulation of Genre based
method separately. And after that combine the Movie Recommendation system using Hadoop MapReduce
predictive result of individual methods. Technique, IEEE-2017, ISBN - 978-1-5386-1887-5
 Mixed: at a single point of process time use [11] Maulik Dhamecha, Dr. Tejas Patalia : Study of Basic
variation of recommender methods. Functionality for Information Retrieval in Large Scale Data – Big
Data, IJMCS-2018, ISSN: 2320-7868
[12] Sharma, S., Sharma, A., Sharma, Y., & Bhatia, M. (2016, April).
IV. CONCLUSION Recommender system using hybrid approach. In Computing,
Communication and Automation (ICCCA), 2016 International
Nowadays, everything act as a service, so creating and Conference on (pp. 219-223). IEEE.
recommending the service using big data analytics in the [13] Shrote, K. R., & Deorankar, A. V. Hotel Recommendation
social networking will be more efficient and accurate. System using Hadoop and MapReduce for Big Data.
Large number of recommendation techniques are [14] Ye, F., & Zhang, H. (2016, July). A collaborative filtering
available in market like content dependent filtering recommendation based on users' interest and correlation of items.
In Audio, Language and Image Processing (ICALIP), 2016
method, collaborative based filtering method or we can International Conference on (pp. 515-520). IEEE.
combine both and created one hybrid recommendation [15] Chen, L. C., Kuo, P. J., & Liao, I. E. (2015). Ontology-based
method. From this paper we can says that Collaborative library recommender system using MapReduce. Cluster
based filtering method is one of best suggestion Computing, 18(1), 113-121.
algorithm for scalability related problem and also [16] Dwivedi, S. K., & Arya, C. (2016, November). A survey of news
provide better recommendation system. So, the proposed recommendation approaches. In ICT in Business Industry &
system gives the more accurate recommendation of Government (ICTBIG), International Conference on (pp. 1-6).
IEEE.
university for better future of students. In this proposed
system, we will focus on hybrid approach which is [17] G. Adovamavicius and A. Tuzhilin, "Toward the next generation
of recommender systems: A survey of the state-of-the-art and
combination of Content based filtering technique and possible extensions," Knowledge and Data Engineering, IEEE
Collaborative based filtering technique to give more Transactions on 17.6, pp.734-749, 2005.
accurate recommendation of university. This algorithm [18] G. Salton, and M. J. McGill, “Introduction to modern information
will be implementing based on Hadoop framework using retrieval,” New York: McGraw-Hill, 1983.
MapReduce. [19] P. Lops, M. D. Gemmis, and G. Semeraro, "Content-based
recommender systems: State of the art and trends,"
REFERENCES Recommender systems handbook. Springer US, pp. 73-105,2
[1] Shakirova, E. (2017, February). Collaborative filtering for music 011.
recommender system. In Young Researchers in Electrical and [20] X. Su, and T. M. Khoshgoftaar, "A survey of collaborative
Electronic Engineering (EIConRus), 2017 IEEE Conference of filtering techniques," Advances in artificial intelligence, pp. 4
Russian (pp. 548-550). IEEE. 2009.
[2] Maulik Dhamecha, Dr. Tejas Patalia : Fundamental Survey of [21] A. S. Das, M. Datar, A. Garg, S. Rajaram, "Google news
Map Reduce in Bigdata with Hadoop Environment, SPRINGER personalization: scalable online collaborative filtering,"
– 2018 , ISBN 978-981-13-2372-0 Proceedings of the 16th international conference on World Wide
Web, ACM , pp. 271- 280, 2007.

978-1-5386-7808-4/19/$31.00 ©2019 IEEE 58