Int J Syst Assur Eng Manag (June 2021) 12(3):426–436

https://doi.org/10.1007/s13198-021-01087-x

ORIGINAL ARTICLE

Comparative study of recommender system approaches and movie

recommendation using collaborative filtering
Taushif Anwar1 • V. Uma1

Received: 24 March 2018 / Revised: 23 May 2020 / Accepted: 10 March 2021 / Published online: 19 April 2021
Ó The Society for Reliability Engineering, Quality and Operations Management (SREQOM), India and The Division of Operation and
Maintenance, Lulea University of Technology, Sweden 2021

Abstract The increasing demand for personalized infor- 1 Introduction

mation has resulted in the development of the Recom-
mender System (RS). RS has been widely utilized and Recommender systems are information filtering tools and
broadly studied to suggest the interests of users and make whose main aim is to provide personalized recommenda-
an appropriate recommendation. This paper gives an tions. Nowadays exponential growth of information sys-
overview of several types of recommendation approaches tems, increase in number of users and data have become
based on user preferences, ratings, domain knowledge, challenging problems in obtaining essential information.
users demographic data, users context and also lists the RS helps users to overcome the information overload
advantages and disadvantages of each RS approach. In this problem and easily find products and services. RS gener-
paper, we also proposed the movie recommendation based ally helps users to find the products and services that fit
on collaborative filtering and singular value decomposition their personal needs and preferences. Millions of users
plus-plus (SVD??). The proposed approach is compared throughout the globe interact with a wide variety of RS on
with well-known machine learning approaches namely k a daily basis Anwar and Uma (2019c, 2020a).
nearest neighbor (K-NN), singular value decomposition Increasing trends in internet usage have affected almost
(SVD) and Co-clustering. The proposed approach is every aspect of life. Dependency on the internet in per-
experimentally verified using MovieLens 100 K datasets forming online shopping, online booking and using social
and error of the RS is measured using Root Mean Square networking websites etc. is increasing day by day. Through
Error (RMSE) and Mean Absolute Error (MAE). The result RS, users effortlessly get required information from the
shows that the proposed approach gives a lesser error rate enormous amount of data in lesser time. RS predicts the
with RMSE (0.9201) and MAE (0.7219). This approach interest and preferences of users. RS is widely accepted
also overcomes cold-start, data sparsity problems and approach which is earning continuous attention in recent
provides them relevant items and services. years. RS is mostly used in various fields such as e-gov-
ernment, e-learning, e-commerce, social media and tourism
Keywords Recommender system  Knowledge based  Premalatha et al. (2018). A large number of researches are
Context based  Hybrid  Content based  Collaborative being carried out on RS using various methods with the
filtering  Intelligent  Demographic aim of selecting out relevant information from the huge
amount of data. In the current scenario many approaches
for recommendation have been proposed in several
domains such as information retrieval, artificial intelli-
gence and data mining. In Fig. 1, shows the simple model
& Taushif Anwar of recommendation system.
[email protected] The highlights of this paper are as follows:
1
Department of Computer Science, Pondicherry University,
Puducherry 605014, India

123
Int J Syst Assur Eng Manag (June 2021) 12(3):426–436 427

Fig. 1 Simple model of

recommendation system

(a) Overview of several recommendations approaches 2 Types of recommender system

based on user preferences, user ratings, item ratings,
domain knowledge, users demographic data, user In Fig. 2, various types of Recommender systems are
context, user reviews, knowledge representation, and presented.
reasoning methods are provided extensively.
(b) The advantages and disadvantages of each RS 2.1 Content-based recommender system (CBRS)
approach is provided.
(c) The advantage of the proposed approach that com- CBRS mainly works by analyzing the features of an item
bines collaborative filtering with SVD?? is com- preferred by a user and tries to identify the similarity
pared with well-known machine learning approaches between them. The preference of user is calculated based
namely K-NN (k nearest neighbor), SVD (singular on his/her choice in the past. Similar items are identified
value decomposition) and Co-clustering. based on the characteristics of compared items Pazzani and
The remainder of the article is structured as follows: Billsus (2007). CBRS system can start recommending
Sect. 2 explains various types of RS approaches. In Sect. 3, items as soon as information about items are available. For
comparison of RS approaches is presented and the con- appropriate recommendations in CBRS, a large amount of
ceptual goals and various inputs to be given to RS are information about the items is required Kumar and Thakur
explained. Section 4 provides details about the experi- (2018).
mental analysis and Sect. 5 concludes the paper. CBRS is categorized into attribute based approach
Felfernig and Burke (2008) and case based reasoning
approach Adomavicius and Tuzhilin (2005). Attribute
based approach suggests items on the basis of matching
their features to the related learner profile Klašnja-Mili-
ćević et al. (2015) and case based reasoning approach

Fig. 2 Types of recommender

systems

123
428 Int J Syst Assur Eng Manag (June 2021) 12(3):426–436

suggests items which are correlated to user interest. In attribute values directly. On the basis of similarity, a new
general CBRS approach mostly uses simple retrieval approach has been proposed for predicting unknown val-
models like Term Frequency-Inverse Document Frequency ues. A large data set of real web services is used for
(TF- IDF) weighting, keyword matching and Vector Space checking the performance of proposed method. The results
Model (VSM). show that computation time is less, mean absolute error is
Deldjoo et al. (2016) proposed an innovative CBRS that less and prediction precision is higher.
automatically checks the video contents and collects a set
of unique characteristics (motion, color and lighting). The 2.2.1 Trust aware recommender system (TARS)
result shows that the approaches gave more appropriate
recommendations. The proposed approach performs better This is an extension of conventional Collaborative based
not only when visual characteristics are extracted but also RS.Trust is a commitment to a person and an action on the
with small-length videos (movie trailers). basis of faith such that the future hopes for good outcome.
This approach only examines the trust association between
2.2 Collaborative recommender system (CRS) users in order to make the recommendation Hassan (2019).
TARS are more robust and can easily overcome cold start
CRS is a prevalent RS approach, which uses past rating and data sparsity problem because trust can be generated in
data, comments, and review to make appropriate recom- the network. Generally, trust is of two types implicit and
mendations. This technique works based on the users’ explicit. Implicit trust is the information of the trust that is
similarity and without asking for exogenous information implicitly gathered from user behavior and explicit trust is
Kant and Mahara (2018); Anwar and Uma (2019a). For the value of the trust that is explicitly given by users.
finding similarity between items, widely used matrices are Golbeck and handler Golbeck and Hendler (2006) said
cosine and correlation based similarity. The standard CRS that trust is a commitment related to an activity on the basis
approaches are model-based and memory based, which are of hope such that the future activity starts with a good
mostly used in various fields such as social networks (e.g., result. The main aim of this approach is to produce a
Instagram, Facebook), review sites (e.g., Google news, personalized recommendation from trust relation and
Movielens) and e-commerce (e.g., Flipkart, Amazon). One known opinions. TARS are extensively used in e-com-
of the most important benefits of memory based CRS merce, e-learning and social network where trust plays
approach is that their natural idea makes it convenient to essential aspect to improve the relationship between users.
understand and yields convenient results. Besides, the main
strength and backbone of pure CRS approaches is that 2.3 Knowledge based recommender system (KBRS)
frequent data addition can be done without any problem.
This is because, no tagging of items is required when KBRS works on the basis of domain knowledge about user
compared to CBRS recommendations Anwar and Uma and items. This recommending approach is based on rea-
(2019b). The process involved in CRS is shown in Fig. 3. soning as to what items are related to the user’s interest
CRS has some limitations such as cold start Lillegraven Tarus et al. (2017b). Mainly three types of information are
and Wolden (2010) Yan et al. (2020),data sparsity Ahma- required; knowledge about the items, users and the simi-
dian et al. (2019) and scalability Koshti et al. (2019). Cold larity between users and items. KBRS approaches are
start problem exists when a new user and new item enters mainly used for complex domains where items are not
the system and there is no rating available. CRS can’t bought regularly. KBRS approaches support hybridization
recommend in the absence of initial ratings and therefore, it with some other recommendation approach. An important
is very tough to make accurate recommendations (Anwar advantage of KBRS is the nonexistence of cold start (start-
and Uma 2020b). up or ramp up) problems. The main limitation of KBRS is
Aciar et al. (2016) proposed a recommendation based on the necessity of knowledge and engineering skills Ricci
novel similarity computation (i.e., ratio-based method) for et al. (2011).
computing similarity. This system can easily calculate Aciar et al. (2016) proposed a recommender system on
similarity between items or between users using the the basis of the degree of domain knowledge that provided

Fig. 3 Process involved in

collaborative recommendation
system

123
Int J Syst Assur Eng Manag (June 2021) 12(3):426–436 429

a convenient question-answering facility. In this paper, 2.4 Demographics based recommender system
new method has been added to find the reputation of the (DBRS)
candidate users in the group. Result shows that the pro-
posed approach gave high degree of effectiveness and DBRS works on the basis of user demographic data. The
proper recommendations. main goal of this RS approach is to classify the user on the
basis of attributes and user’s demographic data stored in
2.3.1 Ontology based recommender system (OBRS) their profiles (i.e. gender, age, location etc.) for suggesting
the item. Unlike CRS and CBRS demographic approach
OBRS works on the basis of ideal knowledge about the doesn’t need previous user ratings.
users, items, and domain knowledge. In OBRS, ontology is DBRS works on three steps: Collecting input data,
mainly used for knowledge representation George and Lal similarity computation and recommendation prediction.
(2019). OBRS is mainly related to KBRS that uses ontol- Collecting input data is the first step that includes new
ogy for knowledge representation. OBRS also alleviates target user’s demographic information (the user who wants
some limitations related to traditional RS, such as Data recommendations) and also demographic information and
sparsity Zhao et al. (2015) Ramp Up and overspecializa- rating of other users. Similarity Computation step uses user
tion. Meanwhile, OBRS relies more on domain knowledge demographic information to retrieve users having related
rather than ratings. This striking characteristic makes demographic interest thus creating a neighborhood. Ulti-
OBRS most useful and suitable for e-commerce, e-learning mately, recommendation computation stage recommends
and tourism-related field. On the other side, development items that are generally positively rated by surrounding
of ontologies is complex and costly process. users. The process involved in DBRS is shown in Fig. 4.
John Tarus et.al Tarus et al. (2017a) presented a rec- Zhao et al. (2015) proposed a DBRS for product rec-
ommendation approach for e-learning using CBF and ommendations which identify users buying plans from
ontology. Ontology is used to combine recommendation microblogs and make recommendations on the basis of
process into the learner features. Result shows that OBRS matching user demographic data. Conventional Product RS
outperforms than CRS. The proposed approach also over- are generally designed for some particular e-commerce
come start-up limitations in the initial steps of recom- websites, which provides suggestions when users are car-
mendation through ontological knowledge. rying out some e-commerce activities. For experimental
testing data crawled from Sina Weibo microblog is used.
2.3.2 Intelligent recommender system (IRS) The result shows that effectiveness and feasibility are
better in terms of recommendations.
The extension of knowledge-based RS is the IRS. The IRS
adventures knowledge, learns, explores new information 2.5 Context aware recommender system (CARS)
from reviews and identifies preferences. IRS is described
by the components: learning methods, knowledge repre- CARS make better suggestions using particular contextual
sentation model and reasoning mechanisms. Furthermore, condition of the user Aghdam (2019). Preferences of the
IRS has five Knowledge models in the distinct context such user may change according to context (mood, season, time
as items, users, context, criticisms and domain, that can be of day, location, companion, occasion etc.). For example,
considered during the recommendation.It also alleviates while watching a movie (location, time, companion etc.),
cold-start problem because it does not depend upon user traveling (location, time, weather, transport condition etc.)
rating and knowledge engineering processes is not and listening to music (time, location, emotions, occasions
required. IRS also utilizes the information about users etc.) Thaduri et al. (2017). CARS collects the context of
through learning mechanism to increase its performance. the user at the time of making recommendations. There are
Aguilar et al. (2016) proposed IRS which enhances the generally three types of architecture for building CARS:
degree of the recommendations through its knowledge Contextual Modeling, Contextual Pre-filtering, and Con-
representation, learning and reasoning techniques. It also textual Post-filtering.
overcomes some conventional problems of the recom- Mazloom et al. (2017) proposed a technique to analyze
mender system, i.e., cold start problem as recommenda- the collaboration between item and user for suggesting the
tions do not depend on the previous user ratings. prevalence of posts associated with a particular user and
Implementation of IRS has been done using the fuzzy item in social media. The result shows that relational
cognitive map. The main features of IRS are representation improvement in user specific and item-specific post popu-
of diverse knowledge, reasoning and learning capabilities. larity recommendation increased.

123
430 Int J Syst Assur Eng Manag (June 2021) 12(3):426–436

Fig. 4 Process involved in

demographic based
recommendation system

2.5.1 Time aware recommender systems (TiARS) on a case study considering digital cameras. In this tech-
nique domain ontology is used to interpret the information.
TiARS is a specialized form of CARS, and it mainly
focuses on utilizing time as contextual information. The 2.7 Hybrid recommender system (HRS)-approaches
interest of users might change time to time, and this is
mainly considered in this RS. Managing the temporally The HRS approach merges characteristics of multiple fil-
changing preference of the user in RS leads to new hurdles tering approaches e.g. CBRS and CRS approach, by adding
because each particular user interests are distinct due to the advantages of each approach there by achieving better
concept drift problem Tsymbal (2004). In the social net- performance Ghauth and Abdullah (2010). This also alle-
work, numerous items and users with multiple attributes viate various issues of RS such as ramp-up, data sparsity
are changing at the same time and move towards each and overspecialization. HF approach is widely accepted
other. and very helpful as it reduces the limitations of conven-
Rezaeimehr et al. (2018) presented a unique TiARS on tional recommender system (Anwar et al. 2021). Firstly in
the basis of finding overlapping community structure 2002, Burke suggested a taxonomy of seven hybridization
within users. The suggested algorithm finds overlapping techniques Mobasher (2007). He divided hybridization
community structures within the users and assists in technique into seven types. This is shown in Fig. 6 and the
reducing the effect of sparsity. The suggested method uses explanation of each technique is given in Table 1 (Fig. 7).
two real-world datasets and achieves better precision than Gordillo et al. (2017) proposed an idea about HRS for
various state-of-the-art approaches. Learning Object Repositories (LORs) that merges demo-
graphic based RS, content based RS and context aware RS
2.6 Review based recommender system (RBRS) approaches, including the use of popularity and quality
metrics. This paper also explains how to use the model for
This RS has the capability to reduce and remove cold start implementing two RSs for two real Learning Object
and data sparsity problems. Basically, RBRS works on the Repositories: Europeana and ViSH. The experimental
basis of product profile and user profile which are result shows that the presented technique had high user
reviewed. This approach, analyzes and classifies the acceptance in terms of satisfaction, usability and utility.
reviews using different methods like opinion mining and Aslanian et al. (2016) proposed a HRS algorithm by
text analysis. It is then divided based on user and product analyzing the similarity among content characteristics.
profile, and then recommend of items is done similar to This relationship is inserted in HRS to increase their
CBRS and Rating-based CRS. The process involved in accuracy. At first, a novel approach is used to obtain the
RBRS is shown in Fig. 5. content characteristic relation matrix. Then the CRS is
Aciar et al. (2007) proposed the RBRS based on con- revised so that relation matrix can be efficiently merged
sumer product reviews. This proposed approach was based with the algorithm. This algorithm alleviates the ramp-up

123
Int J Syst Assur Eng Manag (June 2021) 12(3):426–436 431

Fig. 5 Review based

recommender system process

Fig. 6 Hybridization technique

Table 1 Taxonomy of hybridization techniques

Sr. No. Technique name Function

1 Weighted Merge several RS component numerically

2 Switching System switches among RS depending upon the current situation
3 Mixed Several recommenders are presented together
4 Feature Combination Combine the different features and given to a single method
5 Feature Augmentation Output from one approach is used as an input characteristic to another approach
6 Cascade Applies a repetitive refinement.One recommendation refines the recommendation given by other
7 Meta-level One approach is used for producing some small model, which is given to input as another approach

problem. In this paper, content based hybrid recommender introduced. After that, the diversity and true positive rate
system showed better results in terms of novelty and on the basis of the new situation is redefined to evaluate the
accuracy. RS. The experiment is done using the MovieLens dataset.
Zhang et al. (2015) presented a HRS on the basis of user The result shows that hybrid algorithm makes more effi-
recommender interaction and calculated its performance cient and appropriate recommendation than non-hybrid
with diversity metrics and true positive rate (recall). At one.
first, user recommender interaction is set. Second, merging
of HRS with K- nearest neighbor and random forest is

123
432 Int J Syst Assur Eng Manag (June 2021) 12(3):426–436

Fig. 7 Recommendation
System model for movie
recommendation

3 Comparison of RS approaches 4.2 Proposed system

3.1 Table 2 provides details about the advantages In the basic collaborative recommender system, the rec-
and disadvantages of various Recommendation ommendation is done by (1) Finding similarity between
system approaches. items and users (2) Neighborhood formation (3) Predicting
the rating matrix and (4) Recommendation of top-N items.
But, in the proposed movie recommendation model, the
recommendation is done using collaborative filtering with
3.2 Table 3 provides details about conceptual goals Singular Value Decomposition (SVD)??. The proposed
and various input to be given to various model is divided into five modules which are explained
Recommendation system approaches. below.
(1) Data pre-processing.
(2) Calculating similarity between movies and users
using the Cosine similarity using Eq. 1.
(3) Finding the Prediction Matrix using the formula in
4 Experiments and result Eq. 2.
(4) Applying the SVD?? approach
4.1 Dataset description
(5) Recommending Top-N movies.
We have used the openly available MovieLens 100 K At first, user-movie ratings are used for generating the
dataset related to Movie domain which contains 100,000 rating matrix. Then cosine similarity is found for generat-
ratings (1-5) of 1682 movies given by 943 users. Each user ing similarity between movie and user. The prediction
has rated at least twenty movies. The details such as User matrix is evaluated using Eq. 2. Then the SVD??
id, Movie id, Movie name, rating, genre are obtained1. approach is applied and finally, Top-N items are
recommended.
Pn
AB Ai Bi
Cosine ¼ ¼ Pni¼1 2 2 ð1Þ
k Ak k Bk i¼1 i Bi
A
1
https://grouplens.org/datasets/movielens/100k/. A and B are features of the movies or items. A dot (.)

123
Int J Syst Assur Eng Manag (June 2021) 12(3):426–436 433

Table 2 Advantages and disadvantages of Recommendation method

Recommendation Advantages Disadvantages
methods

Collaborative It does not depend on the analysis of the information given Cold start or ramp-up problem (can’t handle new items or
filtering by the user. The information which is being used for new users). Unpopular items are not strongly
recommendation is domain independent. Does not require recommended.Bootstrapping. Ratings required. Critical
analysis of the items (features). Better at qualitative mass required. Not as transparent in their
judgments. Adaptability(Requirement of a user may recommendations. Need substantially more user data to
change over time) better at qualitative judgments work wel
Content-based Useful in solving cold start problem when a user in the new The item which is highly correlated with user interest or
community. Characteristics of the user profile are matched profile is only recommended (Overspecialization problem
with characteristics of the items. Domain knowledge is not occurs). Learning item/content must be in the category.
required. Independent of other users. Provide transparency Need category maintenance and modeling.New user
( easily understandable,explanations). Easy to explain and problem. Limited content analysis(insufficient keywords,
understand content may not be automatically extractable)
Knowledge based Grey sheep and Cold start problems are avoided because the Necessity of knowledge and engineering skills. Construction
domain knowledge is used for recommendation. Provides of the knowledge base is a difficult task that demands
noise-free and more reliable recommendation. Great expertise in knowledge representation and considerable
precision is possible. Easily mapping is possible from user domain knowledge
need to products
Ontology based It alleviates some limitations related to traditional RS, such Ontologies creation is a difficult process. Construction of
as Data sparsity, ramp up and overspecialization. The main ontology is expensive and time-consuming
advantage of applying ontologies are pointed
recommendation
Demographics Feedback is not needed. No cold start problem. Metadata Cannot provide personalization. Low accuracy. Too general
based engineering is not needed
Context aware Context-aware is more of a real-time nature. More advanced Contextual factor identification. May lead too much sparsity.
than content based RS because this system continuously in Categorical (Time Morning. Evening) Vs. (Time
synchronization with user movement and generate 6:30,2PM) Numeric context information.Serendipity and
recommendation as per user current context surprise
Trust-aware The decision-making process can become easy. Reduces the In explicit trust user have an extra burden for giving trust
sparsity problem. More relevant recommendation than the information apart from the rating. Sparsity problem in
traditional approach Explicit Trust Information because no trust information is
less than no of ratings
Intelligent RS An inference is used for inferring interesting information. Large knowledge about context and domain are required
Alleviates cold-start problem because do not depend upon
user rating. Knowledge engineering process is not required
Utility based No cold start problem. Non-product features can include. Utility function must be added by the user. The ability of
Preferences changes by sensitively possible suggestion is static (doesn’t learn)
Time aware Minimizing the sparsity effects. Better precision also collecting implicit feedback
possible
Hybrid Combining CF with other RS in attempt to overcome cold- External information required that easily not available.
start, scalability, and data-sparsity problem. Prediction Compare between various recommendation paradigms
performance is improved (rating, domain knowledge, item features, critiques,
requirements etc.) not allowed because all are rarely
available in a dataset. Have increased expense of and
complexity

represents vector dot product ||A|| and ||B|| shows the length In this section, Collaborative filtering is performed with
of the vector. four well-known approaches namely K-NN (k nearest
P   neighbor), SVD (singular value decomposition), SVD??
teN simði; tÞ  Ru;t
Pu;i ¼ P ð2Þ (singular value decomposition plusplus), Co-clustering and
teN ðsimði; tÞÞ the results are analyzed. The comparative analysis shows
In this equation, n is the neighborhood of most related that SVD?? gives a lesser error rate.
movies rated by user u and sim (i,t) is the similarity K-NN is a machine learning approach to find clusters of
between movies i and t. similar users on the basis of common item ratings, and is

123
434 Int J Syst Assur Eng Manag (June 2021) 12(3):426–436

Table 3 The conceptual goals of various recommender systems

Methodology Objective Input

Content-based Recommendations based on the content of item User ratings ? item attributes
Collaborative Recommendations based on past behavior of user User ratings ? item attributes
Knowledge- Based on domain knowledge of item and user knowledge of items ? knowledge of user ?
based Similarity between users and items
Ontology OBRS more trust on domain knowledge instead of ratings Model knowledge of items ? Model
based knowledge of user ? Domain knowledge
Hybrid HRS reduces the limitations of conventional recommendation system User ratings ? item attributes ? domain
Knowledge etc
Demographics DBRS approach is to classify the user on the basis of attributes doesn’t need User demographic data (gender, age, location
previous user ratings like CRS and CBRS etc)
Context aware Works based on CARS preferences of the user (mood, season, time of day, Context of User (location ? time ? mood ?
location, companion, occasion etc) companion etc)
Trust-aware TARS use social media and trust information to make a recommendation. Trust value ?* Trust metric
Main goal is to improve the accuracy of recommendation
Intelligent RS The IRS adventures knowledge, learns, explores new information also User profile ? item profile
concludes reviews and preferences among other things

Table 4 RMSE and MAE comparison using MovieLens 100K

dataset
RMSE MAE

K-NN 0.979 0.7732

SVD 0.9358 0.737
SVD?? 0.9201 0.7219
Co-Clustering 0.9675 0.7582

Fig. 9 MAE Comparison graph

each item. But, the SVD?? model tries to mix the

strengths of the latent model and the neighborhood model.
Co-clustering is a data mining approach that allows
simultaneous clustering of the rows and columns of a
matrix.
For implementing the proposed approach python lan-
guage has been used. For implementing various machine
learning approaches scikit-learn and surprise module were
used. For performing comparative analysis on dataset,
Fig. 8 RMSE Comparison graph packages such as numpy, pandas and matplotlib were used
(Table 4).
used to make predictions applying the average rating of
top-k nearest neighbors. SVD finds the latent factors 4.3 Result analysis
associated with matrices. It is a matrix decomposition
approach that reduces the user-item rating matrix as the
product of two lower-dimensional matrices. The first one Our experiment shows that collaborative filtering based
has a row for each user, while the second has a column for K-NN gives highest Root Mean Square Error (RMSE) of

123
Int J Syst Assur Eng Manag (June 2021) 12(3):426–436 435

0.978 and highest Mean Absolute Error (MAE) of 0.772. In Anwar T, Uma V (2020a) A study and analysis of issues and attacks
terms of the lesser error rate, collaborative filtering based related to recommender system. In: Convergence of ICT and
smart devices for emerging applications, Springer, pp 137–157
SVD?? gives lowest RMSE of 0.9201 and MAE of Anwar T, Uma V (2020b) Book recommendation for eLearning using
0.7219 (Figs. 8, 9). collaborative filtering and sequential pattern mining. In: 2020
international conference on data analytics for business and
industry: way towards a sustainable economy (ICDABI). IEEE,
pp 1–6
5 Conclusion Anwar T, Uma V, Hussain MI (2021) Challenges and applications of
recommender systems in e-commerce. In: Challenges and
This paper discusses the different RS approaches and applications of data analytics in social perspectives, pp 175–188
provides a comparison between them. Various recom- Aslanian E, Radmanesh M, Jalili M (2016) Hybrid recommender
systems based on content feature relationship. IEEE Trans Ind
mender approaches have been developed for several rea- Inf 28:666
sons but they have some limitations. So, it is essential to Deldjoo Y, Elahi M, Cremonesi P, Garzotto F, Piazzolla P, Quadrana
develop RS that is both precise and efficient. This paper M (2016) Content-based video recommendation system based on
proposed a movie recommendation system using collabo- stylistic visual features. J Data Sem 5(2):99–113
Felfernig A, Burke R (2008) Constraint-based recommender systems:
rative filtering and SVD?? (singular value decomposition technologies and research issues. In: Proceedings of the 10th
plus-plus) approaches. The proposed approach is compared international conference on Electronic commerce, ACM, p 3
with well-known machine learning approaches namely George G, Lal AM (2019) Review of ontology-based recommender
K-NN (k nearest neighbor), SVD (singular value decom- systems in e-learning. Comput Educ 142:103642
Ghauth KI, Abdullah NA (2010) Measuring learner’s performance in
position) and Co-clustering. The recommender system e-learning recommender systems. Australasian J Educ Technol
error is evaluated considering RMSE (Root Mean Square 26(6):764–774
Error) and MAE (Mean Absolute Error). The result shows Golbeck J, Hendler J (2006) Inferring binary trust relationships in
that the Collaborative filtering with SVD?? gives a lesser web-based social networks. ACM Trans Internet Technol
6(4):497–529
error rate with RMSE (0.9201) and MAE (0.7219). This Gordillo A, Barra E, Quemada J (2017) A hybrid recommendation
approach also overcome with cold-start and data sparsity model for learning object repositories. IEEE Latin Am Trans
problems to some extent. 15(3):462–473
Hassan T (2019) Trust and trustworthiness in social recommender
systems. In: Companion proceedings of the 2019 world wide
web conference, ACM, pp 529–532
Kant S, Mahara T (2018) Merging user and item based collaborative
References filtering to alleviate data sparsity. Int J Syst Assur Eng Manag
9(1):173–179
Aciar S, Zhang D, Simoff S, Debenham J (2007) Informed Klašnja-Milićević A, Ivanović M, Nanopoulos A (2015) Recom-
recommender: basing recommendations on consumer product mender systems in e-learning environments: a survey of the
reviews. IEEE Intell Syst 22(3):156 state-of-the-art and possible extensions. Artif Intell Rev
Aciar SV, Aciar GI, Collazos CA, González CS (2016) User 44(4):571–604
recommender system based on knowledge, availability, and Koshti V, Abhilash N, Gill KS, Nair N, Christian MB, Gupta P (2019)
reputation from interactions in forums. IEEE Revista Iberoamer- Online partitioning of large graphs for improving scalability in
icana de Tecnologias del Aprendizaje 11(1):18–22 recommender systems. In: Computational intelligence: theories.
Adomavicius G, Tuzhilin A (2005) Toward the next generation of Springer, Applications and Future Directions-Volume II,
recommender systems: A survey of the state-of-the-art and pp 121–135
possible extensions. IEEE Trans Knowl Data Eng 17(6):734–749 Kumar P, Thakur RS (2018) Recommendation system techniques and
Aghdam MH (2019) Context-aware recommender systems using related issues: a survey. Int J Inf Technol 10(4):495–501
hierarchical hidden markov model. Physica A 518:89–98 Lillegraven TN, Wolden AC (2010) Design of a bayesian recom-
Aguilar J, Valdiviezo-Dı́az P, Riofrio G (2016) A general framework mender system for tourists presenting a solution to the cold-start
for intelligent recommender systems. Appl Comput Inf 56:288 user problem. MS. thesis, Institutt for datateknikk og
Ahmadian S, Afsharchi M, Meghdadi M (2019) A novel approach informasjonsvitenskap
based on multi-view reliability measures to alleviate data Mazloom M, Hendriks B, Worring M (2017) Multimodal context-
sparsity in recommender systems. In: Multimedia tools and aware recommender for post popularity prediction in social
applications, pp 1–36 media
Anwar T, Uma V (2019a) Cd-spm: cross-domain book recommen- Mobasher B (2007) Data mining for web personalization. In: The
dation using sequential pattern mining and rule mining. J King adaptive web, Springer, pp 90–135
Saud Univ Comput Inf Sci 56:972 Pazzani MJ, Billsus D (2007) Content-based recommendation
Anwar T, Uma V (2019b) Mrec-crm: Movie recommendation based systems. In: The adaptive web, Springer, pp 325–341
on collaborative filtering and rule mining approach. In: 2019 Premalatha M, Viswanathan V, Suganya G, Kaviya M, Vijaya A
International conference on smart structures and systems (2018) Educational data mining and recommender systems
(ICSSS), IEEE, pp 1–5 survey. Int J Web Portals 10(1):39–53
Anwar T, Uma V (2019c) A review of recommender system and Rezaeimehr F, Moradi P, Ahmadian S, Qader NN, Jalili M (2018)
related dimensions. Springer, Berlin, pp 3–10 Tcars: time-and community-aware recommendation system.
Future Gener Comput Syst 78:419–429

123
436 Int J Syst Assur Eng Manag (June 2021) 12(3):426–436

Ricci F, Rokach L, Shapira B (2011) Introduction to recommender Yan Y, Huang C, Wang Q, Hu B (2020) Data mining of customer
systems handbook. In: Recommender systems handbook, choice behavior in internet of things within relationship network.
Springer, pp 1–35 Int J Inf Manag 50:566–574
Tarus J, Niu Z, Khadidja B (2017a) E-learning recommender system Zhang HR, Min F, He X, Xu YY (2015) A hybrid recommender
based on collaborative filtering and ontology. World Acad Sci system based on user-recommender interaction. Mathematical
Eng Technol Int J Comput Electr Autom Control Inf Eng Problems in Engineering 2015
11(2):225–230 Zhao X, Niu Z, Wang K, Niu K, Liu Z (2015) Improving top-n
Tarus JK, Niu Z, Mustafa G (2017b) Knowledge-based recommen- recommendation performance using missing data. Mathematical
dation: a review of ontology-based recommender systems for Problems in Engineering 2015
e-learning. In: Artificial intelligence review, pp 1–28
Thaduri A, Kumar U, Verma AK (2017) Computational intelligence
Publisher’s Note Springer Nature remains neutral with regard to
framework for context-aware decision making. Int J Syst Assur
jurisdictional claims in published maps and institutional affiliations.
Eng Manag 8(4):2146–2157
Tsymbal A (2004) The problem of concept drift: definitions and
related work. Computer Science Department, Trinity College
Dublin 106(2)

123