Neural Comput & Applic

DOI 10.1007/s00521-016-2817-3

ORIGINAL ARTICLE

Recommender system with grey wolf optimizer and FCM

Rahul Katarya1 • Om Prakash Verma1

Received: 16 September 2016 / Accepted: 19 December 2016

 The Natural Computing Applications Forum 2016

Abstract Recommender systems are contributing a sig- Keywords Recommender systems  Collaborative
nificant aspect in information filtering and knowledge filtering  Gray wolf optimizer  Fuzzy c-mean  Movie
management systems. They provide explicit and reliable
recommendations to the users so that user can get infor-
mation about all products in e-commerce domain. In the 1 Introduction
era of big data and large complex information delivery
system, it is impossible to get the right information in the Recommender systems are now mostly used in all e-com-
online environment. In this research work, we offered a merce applications and knowledge management systems.
novel movie-based collaborative recommender system Recommender systems are responsible for delivering
which utilizes the bio-inspired gray wolf optimizer algo- accurate and reliable information to the specific users
rithm and fuzzy c-mean (FCM) clustering technique and [1–5]. Recommender systems collect the relevant infor-
predicts rating of a movie for a particular user based on his mation and preferences for the users or group of users. For
historical data and similarity of users. Gray wolf optimizer example, if a new user is registered on the Amazon Web
algorithm was applied on the Movielens dataset to obtain site, then recommendation engine will first try to investi-
the initial clusters, and also the initial positions of clusters gate the behavior and pattern of that user, and if recom-
are obtained. FCM is used to classify the users in the mender system is unable to provide sufficient
dataset by similarity of user ratings. Our proposed collab- recommendations, then this type of problem is known as
orative recommender system performed extremely well cold-start problem [6–8]. In this situation, the system is
with respect to accuracy and precision. We analyzed our unaware of the new user’s choices and likes, but as soon as
proposed recommender system over Movielens dataset the user spends the time on the Web site and browses the
which is available publically. Various evaluation metrics web pages of Amazon Web site, then he will get the rec-
were utilized such as mean absolute error, standard devi- ommendation of the products which he likes. Various
ation, precision and recall. We also compared the perfor- applications and business applications such as tourism,
mance of projected system with already established entertainment, web intelligence and online shopping have
systems. The experiment results delivered by proposed seen the power of recommender system, and still it is
recommender system demonstrated that efficiency and moving toward its mature phase. RSs work on the various
performance are enhanced and also offered better recom- filtering procedures such as collaborative filtering (CF),
mendations when compared with our previous work [1]. context based, content based (CB), hybrid and social based.
CF approach is the most widely used tactic which is
adopted by the recommender systems [9–13]. In the CF-
based system, the recommendation is suggested on the
& Rahul Katarya basis rating provided by the users. For example, if users
[email protected]
have given the ratings to an e-commerce Web site of var-
1
Department of Computer Science and Engineering, Delhi ious items such as camera, washing machine and juicer
Technological University, Delhi, India mixer, then recommender system will suggest the products

123
 Neural Comput & Applic

to consumers. In the CB filtering, the decision relies on the • The clustering process was boosted with the utilization
choices which were done in the past [14–16]. If a person of fuzzy c-means clustering technique.
purchased a computer, then RS will probably suggest the • The proposed recommender system was evaluated on
products which are related to computer only, because RS Movielens dataset.
will learn the behavior of this individual. In the demo- • We employed MAE, standard deviation, precision and
graphic-based recommender system, the recommendations recall for analyzing the behavior of the system.
are provided by the various factors related to the human • The proposed recommender system performed superi-
such as country, age, sex and location [17–19]. Nowadays orly as it delivered MAE as 0.68 and performed
hybrid-based recommender systems are working well in the efficiently with respect to time.
e-commerce and web intelligence applications as they used • We compared our proposed RS with the already
the combination of the demographic filtering, CB filtering established systems in which proposed recommender
and CF with various computational intelligence, machine system shows superior outcomes.
learning technologies and algorithms [20–22]. To optimize
This research work is structured into several segments,
the recommended results is always a challenging task,
for example, Sect. 2 discusses the relevant work; Sect. 3
especially in which different domains have been applied
demonstrates the projected system. Section 4 is related to
such as data mining, computational intelligence and
experiments and results, and finally, Sect. 5 shows the
machine learning. The bio-inspired algorithms are most
summary and possible gaps in the proposed work.
respectable and efficient in data optimization. There are
also other proficient bio-inspired algorithms such as
genetic algorithm (GA), artificial immune system (AIS),
2 Related work
ant colony optimization (ACO), artificial bee colony
(ABC), fish swarm algorithm (FSA), bacteria foraging
Recommender system has demonstrated a significant impact
optimization (BFO) and differential evolution. These
in e-commerce business since last two decades. Intelligent
algorithms have shown better outcomes in other domains
systems are recommending the products to the users based on
such as bioinformatics, image processing, operation man-
their past behavior and by analyzing performance in the
agement and data mining. These algorithms can be applied
online environment. A method was offered for guessing the
in recommender systems for expert recommendations. We
likes of the users by the help of collaborative filtering and
have used GWO, a bio-inspired algorithm which is based
matrix factorization [13]. Nowadays data are available in
on the gray wolves [23–26]. The meta-heuristics algorithm
tremendous form, and it is now uncontrollable and hard to
describes the behavior of wolves by analyzing their hunting
analyze from it because now it has turned into big data.
and leadership tactics, and these actions are performed by
Various business intelligence and web-based companies
the numerous types of wolves (carnivores animal) breeds
required the relevant information from these big data so that
such as b, d, a and X. A clustering approach, FCM, is
they can understand the customers need according to the
employed with the meta-heuristic GWO. GWO has been
market demand. A research study was performed for pre-
applied to various engineering domains such as optical
diction ratings jobs in group recommender systems [31]. In
engineering, physics (pressure, heat, tension), fluid
the group recommendation systems, behavior and pattern of
dynamics, medical data (breast cancer, lymphography) and
similar users are clustered in a collective group and then
voting. Fuzzy c-means clustering technique is based on the
collectively analyzed the prediction of a particular group.
features that it delivers diverse membership degree of each
Researchers offered four different types of group recom-
and every individual which concerned distinct clusters
mender systems out of which three were built according to
[27–30]. We first applied the GWO on the Movielens
the existing methodologies, and the last one was applied in a
dataset from which we get the initial location of clusters.
different environment that avoided the data sparsity prob-
By taking these initial positions of clusters, FCM is
lem. In the similar area of research of group recommender
employed which classifies the user in the dataset by uti-
system, a recommender system was produced which took the
lizing similarity of user ratings. We divided the dataset as
involvement of each member presented in particular group
training and test data into the ratio of 70:30. The major
and weighted them accordingly [32]. For the justification of
contribution of this research work is:
their concepts, authors proposed a model named as MCS
• We proposed a novel efficient collaborative movie model with the help of matrix factorization technique and
recommender system by adopting a new meta-heuristic then detected the importance of the contribution of each user
algorithm. presented in the group. Tagging is an essential trait of an
• We employed a bio-inspired meta-heuristic algorithm, online social RS in which the performance of the system is
such as gray wolf optimizer (GWO). enhanced, but there are several issues which are responsible

123
Neural Comput & Applic

for delaying the performance of the tag-based recommender domains and categorized them as electronic (government,
system such as ambiguity, sparsity and redundancy. These resource services commerce, shopping, library, business,
such issues were taken into consideration and an algorithm learning, tourism and group activities) [38]. In the similar
was presented which relied on the deep neural networks and type of study, the performance and behavior of recommender
formed a system which had an abstract form of a tag-based system were analyzed in the e-learning environments [39].
recommender system [33]. A literature survey based on They suggested and examined the techniques for recom-
interactive recommender system was accomplished by mender systems as association rule mining, content based,
researchers in which they focused on serious problems of matrix factorization, tensor factorization and collaborative
recommender systems such as collaborative filtering, con- filtering. Another literature survey was conducted by the
textual information, controllability, diversity and cold start authors in which they deeply studied the recommender sys-
[34]. In the recommender systems, sometimes the informa- tems [40]. They studied the recommender systems on the
tion of the users is not complete and has some missing and various important factors such as evaluation procedures,
faulty information because these impact recommender sys- offline and online evaluation with different types of accuracy
tems failed to understand what products should be recom- issues. We analyzed the behavior of users who have rated the
mended. Based on these errors and mistakes, some authors movie on the scale 1 to 5 and demonstrated that our proposed
termed it as natural noise in the recommender system [35]. system is better than the other existing systems. A collabo-
This natural noise was reduced and managed by presenting a rative recommender system was presented with the utiliza-
system which employed the fuzzy-based methods and pro- tion of genetic algorithm in which researchers focused on the
vided an efficient recommender system. Evolutionary com- various similarity measures [41]. A literature review of the
putation is an emerging research area which is applied to the bio-inspired algorithm was conducted, in which author sur-
different interdisciplinary areas of computer science, web veyed various important swarm algorithms with their
technologies and electronics. A survey was conducted for the applications [42]. A new meta-heuristic algorithm was
involvement of evolutionary computation in the recom- named as kidney-inspired and considered four components
mender system and tracked down the areas of information [43]. Recommender systems for groups are productive and
retrieval, knowledge management and web personalization effectual in nature because information can be shared very
[36]. Authors considered the papers which were based on easily and fastly. A group recommender system was
recent technologies and themes of RS. They furthermore designed to analyze the sentiments of human in online
examined the issues of recommender systems such as nov- microblogs in which authors focused on the TV and movies
elty, diversity and serendipity. Authors categorized and applications [44]. A similar work related to the mining of
explained the research work done so far, with evolutionary news was presented by the researchers in which they adopted
computation, genetic algorithm, evolutionary programming, fuzzy systems [45]. A multi-objective system was introduced
genetic programming, feature weighting approaches, clus- in which authors proposed a new evolutionary algorithm for
tering-based methodologies and latent models. As soon as long tail recommendations [46]. A tag-based recommender
recommender system influence is increasing on big data, system based on the deep neural network was offered by
then online social networks are also growing tremendously authors in which they utilized the two Last.Fm and Delicious
in which the products reviews and ratings are shared by [47]. An evolutionary algorithm was proposed which was
online users who can be friends or strangers. Various Web based on the discretization and cut points. This algorithm
sites such as Flickr, Amazon and Flipkart have millions of followed the multivariate tactic and chromosome deduction
users, and they exchange their experiences, lifestyle methodology with the utilization of various benchmark
involvements, check-ins or point of interests, product ratings datasets [48].
and reviews. In this regard for web social networks, a study
was performed which categorized the social information into
three major parts such as social web search, social search and 3 Proposed recommender system
social recommender systems [37]. They also sub-divided all
these components majorly into the indexing, result ranking, In this segment, we will present the projected RS which is
query recommendations, social content search, collaborative assimilated by the FCM and gray wolf optimizer. The
search, user recommendation, item recommendation and fuzzy c-mean is well-known clustering algorithm which is
topics recommendations. Recommender systems apart from applied to the Movielens dataset. The gray wolf optimizer
the recommendation to online recommendation have a sig- was used in our work as it performed efficiently and
nificant impact on other domains also which are essential for delivered superior result in the optimization process when
activities such as web intelligence and information retrieval compared to other meta-heuristic methodologies [24].
from big data. The researchers performed a study related to In Fig. 1, we demonstrated the workflow of our
the application of recommender systems in other substantial anticipated RS in which first the Movielens dataset is

123
 Neural Comput & Applic

Fig. 1 Workflow of proposed

collaborative movie
Features reduction
recommender system Movielens Dataset Apply fuzzy c means
with grey wolf
clustring
optimizer

Recommendations for Rating prediction

Clustered Data
the users model

engaged and then gray wolf optimizer is applied to find Movielens dataset by applying similarity of the user
the initial locations of the clusters. When we received ratings. The useful recommendations are delivered to
the initial positions with the help of gray wolf opti- the users who received the effects of FCM and gray
mizer, then FCM is used to classify the users in wolf optimizer.

Fig. 2 Pseudocode of proposed

collaborative movie
recommender system

123
Neural Comput & Applic

Table 1 Assessment of state of the art methods with different metrics

Method/system MAE SD Precision Recall

K-means 0.7 0.62 0.53 0.43

PCA-K-means 0.64 0.73 0.57 0.48
K-means-improved 0.69 0.65 0.52 0.42
SOM-cluster 0.79 0.08 0.31 0.23
FCM 0.77 0.53 0.49 0.38
UPCC 0.83 0.67 0.17 0.28
KM-PSO-FCM 0.74 0.77 0.54 0.41
PCA-SOM 1.96 1.15 0.16 0.15
PCA-GAKM 0.92 0.18 0.49 0.33
GAKM-cluster 0.78 0.13 0.37 0.45
PCA 1.99 1.48 0.19 0.22
Proposed 0.68 0.54 0.55 0.49

applied finally. Whereas in Fig. 3, we have demonstrated the

detailed working flow process with all three coefficient
points such as r, R and Q which are valid for clusters. Fig-
ures 2 and 3 demonstrate the proposed recommender system
in which first Movielens data are initialized. With the help of
gray wolf optimizer, features of movies in which users have
rated them are initialized. Then, we estimated the appropri-
ateness of respective explorer negotiator and designated as
finest according to their positions. Then, we refreshed the
spots of recent explorer negotiator. We will check whether
all the explorer negotiators are refreshed or not; if they are
not updated, then again we will refresh the spot of existing
explorer negotiator. After this process, all the coefficients
points r, R and Q are updated, and appropriateness value of
all the explorer negotiator is calculated which is present in
the form of users, who have rated the movies. After the
implementation of gray wolf optimization in Movielens
dataset, then users’ centers of a particular cluster are selected
randomly. The results generated by grey wolf optimization
technique were later on adopted by FCM. A membership
matrix is initialized, and then, center vectors are calculated.
Membership matrices are updated with the help of user rating
and movie ratings which are retrieved by the vectors. Finally,
all the clusters and clusters centers are returned with effec-
tive ratings which are estimated by the help of a mean
absolute error, precision and recall.

4 Experiment outcomes and investigation

Fig. 3 Detailed flowchart of proposed collaborative movie recom- In this segment, we will discuss the analysis of experiments
mender system
of the projected RS which was performed on the Movielens
dataset (http://grouplens.org/datasets/movielens/). We per-
In Fig. 2, we demonstrated the pseudocode of our pro- formed all these experiments on the system which has a
posed recommender system, where initial data are loaded in configuration as Intel I3 processor, 8 GB RAM and Python
GWO culture with three explorer negotiators and FCM is 2.7.10 environment.

123
 Neural Comput & Applic

Fig. 4 Comparison of various

methods and systems for MAE 2.50
2.00
1.50

MAE
1.00
0.50
0.00

VARIOUS METHODS AND SYSTEMS

Fig. 5 Comparison of various

methods and systems for 0.6
precision 0.5
PRECISION

0.4
0.3
0.2
0.1
0

VARIOUS METHODS AND SYSTEMS

Fig. 6 Comparison of various

methods and systems for recall 0.6

0.5

0.4
RECALL

0.3

0.2

0.1

VARIOUS METHODS AND SYSTEMS

4.1 Dataset and evaluation procedures publically. The University of Minnesota collected the
dataset for the GroupLens research project. The Movielens
For the evaluation of our proposed recommender system, dataset contains the 100, 000 ratings (1–5) and 943 users
we considered the Movielens dataset which is available who have rated 1682 movies. Every user from 943 has

123
Neural Comput & Applic

Fig. 7 Comparison of the state 2.50

of art methods and systems with
various metrics
2.00

1.50

1.00

0.50

0.00

MAE Standard deviaon Precision Recall

rated at least 20 movies. To understand the behavior of the of the methods did not respond well, for example, SOM-
user’s recommendation quality, we adopted some evalua- Cluster, UPCC, PCA-GAKM, PCA-SOM, and PCA has
tion matrices. We calculated means absolute error (MAE) 0.79, 0.83, 0.92, 1.96 and 1.99, respectively. As far as other
which is the most popular metric to understand the conduct evaluation metrics is concerned such as standard deviation,
of a recommender system. We also calculated the precision precision and recall, our proposed system also delivered
and recall for the proposed system, so that we can analyze enhanced results.
the interest of users in a particular movie or collection of In Figs. 4, 5 and 6, comparisons of different systems are
movies. compared by MAE, precision and recall, respectively.
P In Fig. 7, all evaluation metrics such as mean absolute
k ogh  sgh k
MAE ¼ ð1Þ error, standard deviation, precision and recall are com-
G
pared for the different existing systems and the proposed
G = entire amount of expected movies, ogh = expected system.
value for user g on item h, sgh = real rating. So we concluded from Fig. 7 and Table 2 that our
 
Interesting \ TopN  proposed system performed well with valuable recom-
Precision ¼  
 ð2Þ mendations and performance for the Movielens dataset.
N
 
Interesting \ TopN 
Recall ¼   ð3Þ
jInterestingj  Table 2 Comparison of speed of different methods and systems for
Movielens dataset
If the MAE value is small, then prediction accuracy is
superior. And if precision and recall values are high, then Method Time (in seconds)
recommender system’s prediction accuracy is better. K-means 24.75
PCA-K-means 67.56
4.2 Results K-means-improved 23.48
SOM-cluster 89.11
In this segment, we mentioned the outcomes and compar- FCM 42.67
isons which were performed on various already existing
UPCC 181.23
systems by adopting Movielens dataset of 100 K.
KM-PSO-FCM 140.43
In Table 1, we have compared proposed system with
PCA-SOM 152.89
other existing systems. The proposed system has a mean
PCA-GAKM 31.25
absolute error as 0.68, which is much better than PCA and
GAKM-cluster 332.31
PCA-SOM methods. K-means, PCA-K-means, K-means-
PCA 19.65
improved, SOM-Cluster, FCM, KM-PSO-FCM and
Proposed 66.45
GAKM-Cluster have MAE ranges from 0.64 to 0.78. Some

123
 Neural Comput & Applic

5 Conclusion and future work 12. Liang X, Xia Z, Pang L, et al. (2016) Measure prediction capa-
bility of data for collaborative filtering. Knowl Inf Syst
49(3):975–1004. doi:10.1007/s10115-016-0920-5
Recommender systems are helpful for the organizations 13. Hernando A, Bobadilla J, Ortega F (2016) A non negative matrix
and e-commerce business in such manner that they help the factorization for collaborative filtering recommender systems
users to choose relevant products from billions of products. based on a bayesian probabilistic model. Knowl-Based Syst
97:188–202
Similarly, companies will get to know their customer more
14. Xu Y, Yin J (2015) Engineering Applications of artificial intel-
in depth such as their behaviors, likes and patterns. We ligence collaborative recommendation with user generated con-
have presented a new and efficient collaborative movie- tent. Eng Appl Artif Intell 45:281–294
based recommender system. Our proposed recommender 15. Puglisi S, Parra-Arnau J, Forné J, Rebollo-Monedero D (2015)
On content-based recommendation and user privacy in social-
system performed superiorly when compared with already
tagging systems. Comput Stand Interfaces 41:17–27
existing systems. As suggested, recommender system 16. Puglisi S, Parra-Arnau J, Forné J, Rebollo-Monedero D (2015)
delivered the mean absolute error, standard deviation, On content-based recommendation and user privacy in social-
precision and recall as 0.68, 0.54, 0.55 and 0.49, respec- tagging systems. Comput Stand Interfaces 41:17–27
17. Zhao WX, Li S, He Y, et al. (2015) Exploring demographic
tively. The performance of proposed system with respect to
information in social media for product recommendation. Knowl
time is also superior as compared to existing systems. As Inf Syst 49(1):61–89. doi: 10.1007/s10115-015-0897-5
far as future work is a concern, we will apply this system 18. Katarya R, Verma OP (2015) Restaurant recommender system
by including demographic features, sentiments, machine based on psychographic and demographic factors in mobile
environment. In: 2015 IEEE international conference on green
learning and big data environment. We can also enhance
computing internet things. pp 907–912
the quality of our proposed recommender system by 19. Al-Shamri MYH (2016) User profiling approaches for demo-
including some important factors such as privacy, trust graphic recommender systems. Knowl-Based Syst 100:1–13
within the group, contextual information and application of 20. Moradi P, Gholampour M (2016) A hybrid particle swarm opti-
mization for feature subset selection by integrating a novel local
hybrid systems.
search strategy. Appl Soft Comput 43:1–14
21. Capdevila J, Arias M, Arratia A (2016) GeoSRS: a hybrid
social recommender system for geolocated data. Inf Syst
57:111–128
References 22. Pessemier T De, Dhondt J, Martens L (2016) Hybrid group rec-
ommendations for a travel service. Multimed Tools Appl 1–25.
1. Katarya R, Verma OP (2016) A collaborative recommender doi10.1007/s11042-016-3265-x
system enhanced with particle swarm optimization technique. 23. Mirjalili S, Saremi S, Mirjalili SM, Coelho LDS (2016) Multi-
Multimed Tools Appl 75:1–15 objective grey wolf optimizer: a novel algorithm for multi-cri-
2. Ortega F, Hernando A, Bobadilla J, Kang JH (2016) Recom- terion optimization. Expert Syst Appl 47:106–119
mending items to group of users using Matrix Factorization based 24. Mirjalili S, Mirjalili SM, Lewis A (2014) Grey wolf optimizer.
Collaborative Filtering. Inf Sci (Ny) 345:313–324 Adv Eng Softw 69:46–61
3. Bobadilla J, Ortega F, Hernando A, Gutiérrez A (2013) Recom- 25. Emary E, Yamany W, Hassanien AE, Snasel V (2015) Multi-
mender systems survey. Knowl-Based Syst 46:109–132 objective gray-wolf optimization for attribute reduction. Procedia
4. Katarya R, Verma OP (2016) Recent developments in affective Comput Sci 65:623–632
recommender systems. Phys A Stat Mech Appl 461:182–190 26. Emary E, Zawbaa HM, Hassanien AE (2016) Binary grey wolf
5. Katarya R, Jain I, Hasija H (2014) An Interactive Interface for optimization approaches for feature selection. Neurocomputing
Instilling Trust and providing Diverse Recommendations. In: 172:371–381
2014 IEEE international conference on computer and communi- 27. Koohi H, Kiani K (2016) User based collaborative filtering using
cation technology (ICCCT). pp 17–22 fuzzy c-means. Measurement 91:134–139
6. Ji K, Shen H (2016) Jointly modeling content, social network and 28. Cannon RL, Dave JV, Bezdek JC (1986) Efficient implementa-
ratings for explainable and cold-start recommendation. Neuro- tion of the fuzzy c-means clustering algorithms. IEEE Trans
computing 218. doi:10.1016/j.neucom.2016.03.070 Pattern Anal Mach Intell 8:248–255
7. Zhao WX, Li S, He Y et al (2016) Connecting social media to 29. Havens TC, Bezdek JC, Leckie C et al (2012) Fuzzy c-means
e-commerce: cold-start product recommendation using algorithms for very large data. IEEE Trans Fuzzy Syst
microblogging information. IEEE Trans Knowl Data Eng 20:1130–1146
28:1147–1159 30. Bezdek JC, Ehrlich R, Full W (1984) FCM: the fuzzy c-means
8. Son LH (2016) Dealing with the new user cold-start problem in clustering algorithm. Comput Geosci 10:191–203
recommender systems: a comparative review. Inf Syst 58:87–104 31. Boratto L, Carta S, Fenu G (2016) Investigating the role of the
9. Da Silva EQ, Camilo-Junior CG, Pascoal LML, Rosa TC (2016) rating prediction task in granularity-based group recommender
An evolutionary approach for combining results of recommender systems and big data scenarios. Inf Sci (Ny) 378:1–20
systems techniques based on collaborative filtering. Expert Syst 32. Wang W, Zhang G, Lu J (2016) Member contribution-based
Appl 53:204–218 group recommender system. Decis Support Syst 87:80–93
10. Zhou Q (2016) Supervised approach for detecting average over 33. Zuo Y, Zeng J, Gong M, Jiao L (2016) Tag-aware recommender
popular items attack in collaborative recommender systems. IET systems based on deep neural networks. Neurocomputing
Inf Secur 10:134–141 204:1–10
11. Yang Z, Xu L, Cai Z (2015) Re-scale adaboost for attack 34. He C, Parra D, Verbert K (2016) Interactive recommender sys-
detection in collaborative filtering recommender systems. Knowl- tems: a survey of the state of the art and future research chal-
Based Syst 100. doi: 10.1016/j.knosys.2016.02.008 lenges and opportunities. Expert Syst Appl 56:9–27

123
Neural Comput & Applic

35. Yera R, Castro J, Martı́nez L (2016) A fuzzy model for managing 42. Kar AK (2016) Bio inspired computing—a review of algorithms
natural noise in recommender systems. Appl Soft Comput and scope of applications. Expert Syst Appl 59:20–32
40:187–198 43. Jaddi NS, Alvankarian J, Abdullah S (2017) Kidney-inspired
36. Horváth T, de Carvalho ACPLF (2016) Evolutionary computing algorithm for optimization problems. Commun Nonlinear Sci
in recommender systems: a review of recent research. Nat Numer Simul 42:358–369
Comput 1–22. doi:10.1007/s11047-016-9540-y 44. Li H, Cui J, Shen B, Ma J (2016) An intelligent movie recom-
37. Bouadjenek MR, Hacid H, Bouzeghoub M (2016) Social net- mendation system through group-level sentiment analysis in
works and information retrieval, how are they converging? A microblogs. Neurocomputing 210:1–10
survey, a taxonomy and an analysis of social information retrieval 45. Iglesias JA, Tiemblo A, Ledezma A, Sanchis A (2016) Web news
approaches and platforms. Inf Syst 56:1–18 mining in an evolving framework. Inf Fusion 28:90–98
38. Lu J, Wu D, Mao M et al (2015) Recommender system appli- 46. Wang S, Gong M, Li H, Yang J (2016) Multi-objective opti-
cation developments: a survey. Decis Support Syst 74:12–32 mization for long tail recommendation. Knowl-Based Syst
39. Klašnja-Milićević A, Ivanović M, Nanopoulos A (2015) Rec- 104:145–155
ommender systems in e-learning environments: a survey of the 47. Zuo Y, Zeng J, Gong M, Jiao L (2016) Tag-aware recommender
state-of-the-art and possible extensions. Artif Intell Rev systems based on deep neural networks. Neurocomputing
44:571–604 204:1–10
40. Beel J, Gipp B, Langer S, Breitinger C (2015) Research-paper 48. Ramı́rez-Gallego S, Garcı́a S, Benı́tez JM, Herrera F (2016)
recommender systems: a literature survey. Int J Digit Libr Multivariate Discretization Based on Evolutionary Cut Points
17(4):305–338. doi:10.1007/s00799-015-0156-0 Selection for Classification. IEEE Trans Cybern 46:595–608
41. Gang L, Chun-ling H, Sheng-bing C (2015) Research on rec-
ommender system based on ontology and genetic algorithm.
Neurocomputing 187:1–6

123