Information Systems 58 (2016) 87104

Contents lists available at ScienceDirect

Information Systems
journal homepage: www.elsevier.com/locate/infosys

Dealing with the new user cold-start problem

in recommender systems: A comparative review
Le Hoang Son n,1
VNU University of Science, Vietnam National University, Vietnam

a r t i c l e i n f o

abstract

Article history:
Received 15 September 2014
Received in revised form
6 October 2014
Accepted 7 October 2014
Recommended by D. Shasha
Available online 24 December 2014

The Recommender System (RS) is an efficient tool for decision makers that assists in the
selection of appropriate items according to their preferences and interests. This system has
been applied to various domains to personalize applications by recommending items such as
books, movies, songs, restaurants, news articles and jokes, among others. An important issue
for the RS that has greatly captured the attention of researchers is the new user cold-start
problem, which occurs when there is a new user that has been registered to the system and no
prior rating of this user is found in the rating table. In this paper, we first present a classification
that divides the relevant studies addressing the new user cold-start problem into three major
groups and summarize their advantages and disadvantages in a tabular format. Next, some
typical algorithms of these groups, such as MIPFGWC-CS, NHSM, FARAMS and HUFCF, are
described. Finally, these algorithms are implemented and validated on some benchmark RS
datasets under various settings of the new user cold start. The experimental results indicate
that NHSM achieves better accuracy and computational time than the relevant methods.
& 2014 Elsevier Ltd. All rights reserved.

Keywords:
Collaborative filtering
NHSM
New user cold start
Recommender systems

Contents
1.
2.
3.

4.

5.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
The analysis of existing methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.1.
MIPFGWC-CS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.2.
NHSM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.3.
FARAMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.4.
HUFCF. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Experiments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.1.
Environment setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2.
Results and discussion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

Tel.: 84 904171284; fax: 84 438623938.

E-mail addresses: sonlh@vnu.edu.vn, chinhson2002@gmail.com
1
Official address: 334 Nguyen Trai, Thanh Xuan, Hanoi, Vietnam.

http://dx.doi.org/10.1016/j.is.2014.10.001
0306-4379/& 2014 Elsevier Ltd. All rights reserved.

88

L.H. Son / Information Systems 58 (2016) 87104

Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Appendix A Supporting information. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

1. Introduction
The growing development of content-based systems
that provide a large amount of data, such as videos,
images, blogs, multimedia, and wikis, brings great challenges for analysts attempting to extract useful knowledge
and capture meaningful events from the massive data.
Machine learning tools should indeed be oriented to what
users intend to do and how they want the results to be
returned in a given format. An efficient tool that assists
decision makers to choose appropriate items according to
their preferences and interests and that is currently widely
used is the Recommender System (RS). Ricci et al. [31]
defined the RS as a special type of information system that
(i) helps to make choices without sufficient personal
experience of the alternatives, (ii) suggests products to
customers, and (iii) provides consumers with information
to help them decide which products to purchase. The RS is
based on a number of technologies, such as information
filtering, classification learning, user modeling and adaptive hypermedia, and it is applied to various domains to
personalize applications by recommending items such as
books, movies, songs, restaurants, news articles and
jokes, among others. It has been applied to e-commerce
to learn from a customer and recommend products
that he or she will find most valuable from among the
available products, thus helping the customer find suitable
products to purchase. Some e-commerce RSs are named as
follows [37,22]. For example, Amazon.com is the most
famous e-commerce RS, structured with an information
page for each book while providing details of the text and
purchase information. Two recommendations are found
herein, including books frequently purchased by customers who purchased the selected book and authors whose
books are frequently purchased. eBay.com is another
example that provides the Feedback Profile feature, which
allows both buyers and sellers to contribute to the feedback profiles of other customers with whom they have
done business. The feedback consists of a satisfaction
rating and a specific comment about the other customer.
On Moviefinder.com, customers can locate movies with a
similar mood, theme, genre or cast through Match Maker
or by their previously indicated interests through WePredict. We clearly recognize that RSs are becoming important and with increasing influence on various practical
applications.
An important issue for RSs that has greatly captured the
attention of researchers is the cold-start problem. This problem has two variants: the new user cold-start problem and
the new item cold-start problem. The new item cold-start
problem occurs when there is a new item that has been
transferred to the system. Because it is a new product, it has
no user ratings (or the number of ratings is less than a
threshold as defined in some equivalent papers) and is
therefore ranked at the bottom of the recommended items
list. Moreover, this problem can be partially handled by staff

members of the system providing prior ratings to the new

item. Thus, the concentration of the cold-start problem is
dedicated to the new user cold-start problem when no prior
rating could be made due to the privacy and security of the
system. It is difficult to give the prediction to a specific item
for the new user cold-start problem because the basic
filtering methods in RSs, such as collaborative filtering and
content-based filtering, require the historic rating of this user
to calculate the similarities for the determination of the
neighborhood. For this reason, the new user cold-start
problem can negatively affect the recommender performance
due to the inability of the system to produce meaningful
recommendations [33]. Addressing this problem has been
the primary focus of various studies in recent years.
The aim of this paper is to provide a comparative review
of those studies that could answer our research question
which (group of) algorithm is the most effective among
all?. For this purpose, we first provide a classification that
divides the relevant studies into three groups: (i) makes
use of additional data sources; (ii) selects the most
prominent groups of analogous users; and (iii) enhances
the prediction using hybrid methods. A table that summarizes the advantages and disadvantages of all groups of
methods is presented. Second, some typical algorithms of
the groups of methods, such as MIPFGWC-CS [46] (the first
group), NHSM [20] (the second group), FARAMS [17] and
HUFCF [42] (the third group), are described in detail.
Finally, these algorithms are implemented and validated
on some benchmark RS datasets, such as MovieLens [23]
and Jester [12], under various settings of the new user cold
start. The experimental results could reveal the answer for
our research question stated above.
The remainder of the paper is organized as follows. In
Section 2, we present a literature review of the relevant
studies according to the three aforementioned groups.
Section 3 elaborates on the four typical methods, namely,
MIPFGWC-CS [46], NHSM [20], FARAMS [17] and HUFCF
[42]. Section 4 presents the comparative experiments of
these algorithms involving benchmark RS datasets. Finally,
Section 5 draws conclusions and delineates the future
research directions.

2. Literature review
The beginning of this section starts with an example
that clearly demonstrates the new user cold-start problem.
Example 1. We have a RS that includes three tables: the
users' demographic data (Table 1), the movies' information
(Table 2) and the rating (Table 3). This type of system is able
to predict the user rating of a movie, which is expressed in
Table 3. Nonetheless, the new user cold-start problem
occurs with a new user, e.g., Kim (User ID: 6) in Table 1,
who has no prior rating such that it is difficult to provide a
prediction for the first movie, e.g., Titanic (ID: 1).

L.H. Son / Information Systems 58 (2016) 87104

Table 1
Users' demographic data.
ID

Name

Age

Gender

Occupation

1
2
3
4
5
6

John
David
Jenny
Marry
Tom
Kim

23
30
29
20
30
25

Male
Male
Male
Female
Male
Female

Student
Doctor
Student
Engineer
Engineer
Doctor

Table 2
Movies' information.
ID

Name

Genre

Date

Sales

1
2
3

Titanic
Hulk
Scallet

Romantic
Horror
Romantic

9/2004
10/2005
6/2009

150
300
200

Table 3
Rating data.
User ID

Movie ID

Rating

1
1
2
2
2
3
3
4
5
5
6

2
3
1
2
3
2
1
2
2
3
1

4
2
4
3
1
2
1
3
3
2
?

In the following, we briefly summarize the relevant

works in regards to the new user cold-start problem. At the
milestone of 2014, there are various works aiming to
handle this problem. Those studies could be divided into
three categories: (i) makes use of additional data sources,
(ii) chooses the most prominent groups of analogous users,
and (iii) enhances the prediction using hybrid methods.
a) The principal idea of the first group is the use of some
additional sources, such as the demographic data (a.k.a.
the users' profile), the users' opinions, and social tags,
for a better selection of the neighbors of the new user.
Vozalis and Margaritis [49] demonstrated a modified
version of k-nearest neighborhood by adding a user
demographic vector to the user profile and embedding
it in the collaborative filtering algorithm for the calculation of similarity. Poirier et al. [27] proposed a
method that exploits blog textual data to reduce the
cold-start problem by labeling subjective texts according to their expressed opinions to construct a user
item-rating matrix and establishing recommendations
through collaborative filtering. Zhang et al. [53] presented a recommendation algorithm that makes use
of social tags, particularly user-tag-object tripartite
graphs, to provide more personalized recommendations when the assigned tags belong to diverse topics.

89

Almazro et al. [3] introduced a hybrid demographicbased and collaborative filtering approach on the movie
domain using demographic data to enhance the recommendation suggestion process. Their method classified
the genres of movies based on demographic attributes,
e.g., user age (child, teenager or adult), student (yes or
no), have children (yes or no) and gender (female or
male). Preisach et al. [28] argued that many user
profiles contain untagged resources that could provide valuable information, especially for the cold-start
problem, and proposed a purely graph-based semisupervised relational approach that uses untagged
posts. Said et al. [34,36] modified the user similarity
calculation method to employ the hybridization of
demographic and collaborative approaches. A modification to the k-nearest neighborhood that calculates
the similarity scores between the target user and other
users was introduced. Wang et al. [50] introduced
Credible and co-clustering filterBot for cold-stArt recommendations (COBA), which uses the rating confidence
level to reduce the dimensionality of the itemuser
matrix. The items and users were co-clustered, and the
ratings within every user cluster were smoothed to
overcome data sparsity. The recommendations were
fused from item and user clusters to predict user
preference. Zhang et al. [52] proposed the Cold-start
Recommendations Using Collaborative Filtering (CRUC)
scheme, which involves formulation, filtering and prediction steps. They assumed that users are tracked by
sensors such that each user has their own location,
which is currently regarded as the item. The itemuser
matrix was normalized and clustered to identify users
who have a significant influence on the recommendation. The prediction steps were performed by taking the
hybrid between the item-based and user-based filtering
methods. Chen et al. [8] employed additional information, such as the social sub-community and an ontology
decision model, to assist the recommendation in the
cold-start problem. The social sub-community was
divided according to the exiting users' history data
and the mining relationship between each other. An
ontology decision model was then constructed on the
basis of sub-community and users' static information,
which makes recommendations for the new user based
on his static ontology information. Guo [11] proposed
three different approaches from the perspective of
preference modeling. First, the ratings of trusted neighbors were merged to form a new rating profile for the
active users based on which better recommendations
can be generated. Second, a novel Bayesian similarity
measure was introduced by taking both the direction
and length of rating vectors into account. Third, a new
information source called prior ratings, based on virtual
product experience in virtual reality environments, was
proposed to inherently resolve the concerned problems. Chen et al. [7] proposed a cold start recommendation method for the new user that integrates a user
model with trust and distrust networks to identify trustworthy users, which are then aggregated to provide
useful recommendations for new users. Demographic

90

L.H. Son / Information Systems 58 (2016) 87104

data or users' profiles are the most common additional

source for solving the cold-start problem. Safoury and
Salah [33] presented a framework for evaluating the
influence of demographic attributes on the user ratings.
This framework was examined using a movie dataset to
evaluate the accuracy and precision of the generated
recommendations. Nazir and Yadav [24] introduced a
profile-based approach that consisted of three main
phases: Fetch, Process and Truncate. The Fetch phase is
concerned with obtaining the required parameters,
such as Global Student Profile or Shared Profile (Profile
Token), for the Process phase. Process is the main
engine where the actual recommendations are generated based on inputs from the Fetch phase. Truncate is
involved with the discarding of the profile-tokens used
during the Fetch phase. Formoso et al. [9] proposed a
novel profile-expansion approach that includes three
types of techniques, namely, item-global, item-local
and user-local, based on the query expansion techniques in information retrieval. The experimental evaluation showed that both item-global and user-local offer
outstanding improvements in precision. Son et al. [46]
presented a novel filtering method based on fuzzy
geographically clustering [45,3844], the so-called
MIPFGWC-CS, that can handle the issues of selected
demographic attributes, the similarities between items
and missing ratings that existed in relevant demographic-based algorithms. Rosli et al. [32] designed a new
measure by combining similarity values obtained from
a movie Facebook Page. First, the users' similarities
were computed according to the rating cast on the
Movie Rating System. Then, the similarity values obtained from a user's genre interest in Like information
extracted from Facebook Pages were combined.
Finally, all of the similarity values were integrated to
produce a new user's similarity value. Lika et al. [18]
proposed a model that incorporates classification methods with demographic data for the identification of
other users with similar behaviors.
" Limitations of the first group: although the additional
data sources are necessary, we sometimes do not have
these types of data for the selection, e.g., in some
e-shopping systems when users do not record their
profiles and associated Facebook/Twitter accounts.
b. The idea of the second group is to improve the methods
that determine the analogous users without the aid of
additional data sources. Ahn [2] addressed the limitations
of the existing methods for the new user cold-start
problem by primarily focusing on the similarity measures,
such as the Pearson coefficient and the cosine measure,
and proposed a heuristic similarity measure, i.e., the socalled PIP (ProximityImpactPopularity) measure. The
Proximity factor is based on the arithmetic difference
between two ratings, the Impact factor considers how
strongly an item is preferred or disliked by buyers, and
the Popularity factor provides greater value to a similarity
for ratings that are further from the average rating of a
co-rated item. Lam et al. [15] discussed a hybrid model
based on the analysis of two probabilistic aspect models
using pure collaborative filtering to combine with users'
information. Sun et al. [47] clustered users based on the

useritem rating matrix and then utilized the clustering

results and users' demographic information to construct a
decision tree to achieve the associations between the
existing users and the new users. The predictions for new
users were made by combining the decision tree with the
collaborative filtering algorithm. Zhou et al. [54] presented functional matrix factorization (fMF), a novel coldstart recommendation method that constructs a decision
tree with each node being a question. fMF enables the
recommender to query a user adaptively according to her
prior responses and associates latent profiles for each
node of the tree to gradually refine the profiles. It also
consists of an iterative optimization scheme that alternates between decision tree construction and latent
profile extraction. Qiu et al. [29] introduced an itemoriented function and incorporated it with a hybrid
algorithm between heat conduction and the probability
spreading process so that the proposed algorithm does
not require any additional information, such as tag. Liu
et al. [21] noted that the existing recommendation
methods lacked a principled model for guiding how to
select the most useful ratings and that ratings on the
selected representatives are considerably more useful for
making recommendations; thus, they proposed a principle approach to identify representative users and items
using representative-based matrix factorization. Bobadilla
et al. [5] presented a new similarity measure using
optimization based on neural learning, which exceeds
the best results obtained with current metrics, and
described the mathematical formalization that shows
how to obtain the main quality measures of a recommender system using leave-one-out cross validation. Said
et al. [35] performed a set of tests to identify whether the
weighting schemes on three common similarity measures using two different movie datasets can be beneficial
for the purpose of overcoming problems related to coldstart, as well as profiling users to generate more accurate
profiles not based on the most popular items. They
claimed that the weighting schemes appear to have little
effect on datasets with a wide rating scale and high
concentration of ratings on popular items. Moreover, the
cosine measure is very insignificantly affected by any
weighting measure and produces identical results regardless of whether weighting is applied. Sun et al. [48]
proposed a novel algorithm that learns to conduct the
interview process guided by a decision tree with multiple
questions at each split. The splits, represented as sparse
weight vectors, are learned through an L_1-constrained
optimization framework. The users are directed to child
nodes according to the inner product of their responses
and the corresponding weight vector. A linear regressor is
learned within each node, using all previously obtained
answers as inputs to predict item ratings. Liu et al. [20]
presented a new user similarity model NHSM that
takes into account the global preference of user behaviors
in addition to the local context information of user ratings
to improve the recommendation performance in the
cold-start situation.
" Limitations of the second group: how to choose the
optimal number of groups and the splitting criteria
is worth considering.

91

L.H. Son / Information Systems 58 (2016) 87104

c. After determining the most analogous users to the new

one, some authors used hybrid methods for the calculation of similarity and/or the prediction of rating. This is
the basic idea of the third group. Leung et al. [16,17]
introduced a collaborative filtering framework based on
Fuzzy Association Rules and Multiple-level Similarity
(FARAMS), which extends existing techniques by using
fuzzy association rule mining and takes advantage of
product similarities in taxonomies to address data sparseness and non-transitive associations. Basiri et al. [4]
proposed a hybrid recommender system using the optimistic exponential type of an ordered weighted averaging
operator to fuse the output of recommender system strategies for the new user cold-start problem. Kim et al. [13]
presented a method for the cold-start problem that
includes the prediction of actual ratings, the identification
of prediction errors for each user, and the construction of
an error-reflected model for the prediction of new users
or items. Ge and Ge [10] claimed that a lower-rank approximation could remove data noise resulting from unstable user behaviors and thus lead to better recommendation quality; based upon this idea, they proposed
Singular Value Decomposition-based Collaborative Filtering. Kim et al. [14] proposed three hybrid recommenders
based on user similarity and two content-boosted recommenders used in conjunction with interaction-based collaborative filtering; they experimentally showed that the
best hybrid and content-boosted recommenders improve
on the interaction-based collaborative filtering method.
Quijano-Snchez et al. [30] extended a group recommender system with a case based on previous group recommendation events. Carrer-Neto et al. [6] presented a hybrid recommender system based on knowledge and social
networks. Negre et al. [25] introduced a process for solving the cold-start problem in cases of data warehouses
composed of four steps: patternizing OLAP queries, predicting candidate operations, computing candidate recommendations and ranking these recommendations. Xie

et al. [51] proposed Elver, which employs an iterative

matrix completion technology and a nonnegative factorization procedure to work with meager content inklings
to recommend and optimize page-interest targeting on
Facebook. Aharon et al. [1] introduced a recommendation
algorithm based on Latent Factor analysis called One-pass
Factorization of Feature Sets (OFF-Set), which is able to
model non-linear interactions between pairs of features
and updates its model per each recommendation-reward
observation in a pure online fashion. Lin et al. [19] described a method that considers the nascent information
culled from Twitter to provide relevant recommendations
in cold-start situations. Nilashi et al. [26] proposed new
recommendation methods using ANFIS and SOM clustering. A hybrid user-based fuzzy collaborative filtering
method has been proposed by Son [42].
" Limitations of the third group: irrelevant users are
still included in the computation of similarities.
d. Table 4 summarizes the relevant works by groups along
with their advantages and disadvantages.

3. The analysis of existing methods

In this section, we provide more details about the
typical algorithms of the groups of methods in Table 4 to
address the new user cold-start problem. These algorithms
are MIPFGWC-CS [46], NHSM [20], FARAMS [17] and HU
FCF [42], and these methods are presented in the subsequent sections.
3.1. MIPFGWC-CS
The basic concept of the MIPFGWC-CS algorithm [46] is
to use fuzzy geographically clustering [45,3844], particularly MIPFGWC, for the determination of similar users with
respect to all attributes in the demographic data. Because

Table 4
Groups of methods for the new user cold-start problem.
Group

Ideas

Typical
algorithms

Advantages

Disadvantages

Makes use of
additional data
sources

Use some data (users' profile, opinions, social

tags) to support the selection of the
neighbors of the new user

# MIPFGWCCS [46]

# Determination of
analogous users is
more accurate

# Additional data are sometimes not

available

Chooses the most

prominent groups
of analogous users

Improve the methods determining the

analogous users by clustering algorithms,
decision trees

# NHSM
[20]

# The similarity
degrees between
users are enhanced
# Additional data are
not required

# How to choose the optimal number

of groups and the splitting criteria
is worth considering

Enhances the
prediction using
hybrid methods

Use hybrid methods for the calculation of

similarity and/or the prediction of ratings

# FARAMS
[17]
# HUFCF
[42]

# Utilizes the results

of existing methods
for prediction
# Controls the final
results by
parameters

# Specification of values of
parameters is hard
# Irrelevant users are still included in
the computation of similarities

92

L.H. Son / Information Systems 58 (2016) 87104

the new user has no prior rating, the demographic data are
the only medium to calculate the similarities between
users. After finding users similar to the new one,
MIPFGWC-CS checks whether they rated the considered
item or not. If ratings are found, then they are considered
to be the representative ratings of users. Otherwise, a
similar item to the considered one is found by the Pearson
coefficient, and the rating on the similar item is assumed
to be the representative rating. Finally, the rating of the
new user to the considered item is approximated by the
weighted average operator of the representative ratings.
Fig. 1 and Table 5 illustrate the idea in detail.
As described above, the MIPFGWC-CS algorithm contains several disadvantages, as follows:
a) Determining the optimal number of clusters for
MIPFGWC is required before running the clustering
algorithm. Although other parameters of MIPFGWC
were suggested by Son et al. [39], how to determine
the optimal number of clusters is still an on-going topic
of research. The exact number of clusters would lead to
more accurate results for finding the similar users to a
new user and thus enhance the prediction accuracy.
b) In Fig. 1, finding a similar item to the considered one by
the Pearson coefficient could somehow not achieve
good results because the Pearson metric has some
limitations where there is a poor signal-to-noise ratio
and negative spikes. In other words, if the relationship
between two variables is non-linear, the Pearson coefficient cannot accurately measure the correlation.
c) The MIPFGWC-CS relies solely on the demographic
data (Fig. 1). If this type of data is not available, the
algorithm cannot be performed.

3.2. NHSM
Liu et al. [20] introduced a new similarity metric called
NHSM to replace the traditional Pearson coefficient or the
cosine similarity measure. This heuristic similarity measure is composed of three factors of similarity, which are
Proximity, Significance and Singularity. Proximity considers the distance between two ratings. Significance shows
that the ratings are more significant if the two ratings are
more distant from the median rating. Singularity represents how the two ratings are different from other ratings.
Furthermore, NHSM integrates the modified Jaccard and
the user rating preference in the design. The definition of
NHSM is stated below.
simu; vNHSM simu; vJPSS simu; vURP ;

1
!
";
simu; vURP 1 #
1 exp # ju #v jj u # v j

simu; vJPSS simu; vPSS simu; vJaccard ;

simu; vJaccard
simu; vPSS

jI u \ I v j
;
jI u j ' jI v j

!
"
Proximity r u;p ; r v;p
pAI

10
11

Fig. 1. The MIPFGWC-CS algorithm.

!
"
!
"
'Signif icance r u;p ; r v;p Singularity r u;p ; r v;p ;

12

!
"
Proximity r u;p ; r v;p 1 #
!
"
Signif icance r u;p ; r v;p

1
!
";
1 exp #jr u;p #r v;p j

13

1
!
";
1 exp # jr u;p # r med jjr v;p #r med j

14

L.H. Son / Information Systems 58 (2016) 87104

93

Table 5
The pseudo-code of the MIPFGWC procedure.
Input

Geo-demographic data X. The number of elements (clusters) NC. The dimension of dataset r. Threshold and other parameters m; ; , ai

i 1; 3, j j 1; C . Geographic parameters ; ; ; a; b; c; d.
Output Final membership values u0 and centers V t 1
k
MIPFGWC
1:
Set the number of clusters C, threshold 4 0 and other parameters such as m; ; 41, ai 4 0 i 1; 3, j j 1; C as in [39]
2:
Initialize centers of clusters V j , j 1; C at t 0
3:

4:

Set geographic parameters ; ; ; a; b; c; d satisfying condition (1)

1:
Use the formulas to calculate the membership values, the hesitation level and the typicality values, respectively
1
ukj
!m 2# 1 ; k 1; N ; j 1; C ;
C
P
X k # V j
i1

hkj

5:

6:
7:

8:

! 2
& #1
C %
P
X k # V j

i1

t kj

X k # V i

k 1; N ;

& 1 ;
2 #1

a2 X k # V j
j

j 1; C ;

X k # V i

3
k 1; N ;

j 1; C :

Perform geographic modifications through Eqs. (56)

kX
#1
C
1 X
wkj ' u0j ' '
w ' uj ;
u0k ' uk '
A j k kj
5
j1
8
b
< popk 'popj 'pckj 'IMdkj
kaj
a
dkj
:
wkj
:
6
0
else
' (
If u0k is a completely monotone increasing sequence or uk Z u0k for most k 1; C , then conclude that there is no suitable solution for the given
geographic parameters. Otherwise, go to Step 7.
Calculate the centers of clusters at t 1 by Eq. (7)
&
N %
P

a1 um
kj a2 t kj a3 hkj ' X k
k1
Vj
& ; j 1; C :
N %
P

a1 um
a2 t kj a3 hkj
kj
7
k1
If the difference jjV t 1 # V t jj r , then stop the algorithm. Otherwise, assign V t V t 1 and return to Step 4.

!
"
Singularity r u;p ; r v;p 1 #

1
#":
! #r r
1 exp # # u;p 2 v;p #p #

15

where u and u are the mean rating and the standard

variance of user u, respectively. I u represents the set of
ratings of user u. The operator ' means the common
ratings between two users. r u;p is the rating of user u on
item p. r med is the median value in the rating scale. Fig. 2
describes the filtering method using the NHSM metric. The
limitations of the NHSM-based filtering algorithm as
follows:
a) The algorithm is based solely on the rating data and
makes no use of additional data, such as demographic
data; thus, it somehow leads inaccurate calculations of
the similarity.
b) The algorithm must assume that the new user has rated
some prior rating in the rating data.

3.3. FARAMS
Leung et al. [17] integrated fuzzy sets theory into
association rule mining techniques and applied the proposed work to the collaborative filtering of recommender
systems. First, the rating data are converted to the transactional database of association rule mining, fuzzified by

fuzzy memberships of linguistic variables and transformed

into the type of transaction ID (TID) items where each
TID is in the form of {Item, linguistic variable}, and each
item is a list of users with equivalent fuzzy memberships
that opted for the {Item, linguistic variable}. Then, an
Apriori-like algorithm is used to define candidate item
sets and possible rules with the support of MinSupp and
MinConf thresholds. The difference between this algorithm and the original Apriori algorithm is the use of
Fuzzy Support FC hhA;X i;hB;Y ii and Fuzzy Confidence
FC hhA;X i;hB;Y ii between two items A and B equipped by their
memberships X and Y respectively (Eqs. (1621)). After
defining the fuzzy rules, the predicting score of a recommendable item is calculated and used to provide the final
rating of the new user. Fig. 3 highlights the concept in
detail.
P
t i A T aj A A t i aj )
FShA;X i
;
16
jTj
FC hA;X i;hB;Y i

FShA [ B;X [ Y i
;
FShA;X i

CovhA;X i;hB;Y i
CORRhA;X i;hB;Y i q;
Var hA;X i nVar hB;Y i

17

18

94

L.H. Son / Information Systems 58 (2016) 87104

Fig. 2. The NHSM-based filtering algorithm.

CovhA;X i;hB;Y i FShA [ B;X [ Y i # FShA;X i nFShB;Y i ;

19

Var hA;X i FShA;X i2 # FShA;X i 2 ;

20

FShA;X i2

ti A T

o2
aj A A t i aj )
jT j

21

)
*
In Eq. (16), A; X represents an Itemset, FuzzySet. t i aj ) is
the value of aj in the ith record of the transactional
database T. t i aj ) is the membership value of t i aj ).
Eqs. (1618) provide the formulas of Fuzzy Supports, Fuzzy
Confidence and Correlation, respectively. The limitations of
the FARAMS algorithm are as follows:

a) The fuzzification in FARAMS could lead to inaccurate

prediction results. The FARAMS algorithm was designed
for movie applications, e.g., MovieLens, Jester and EachMovie, where the linguistic variables are Like, Dislike
and Neutral with pre-defined membership functions.
When applied to other applications, knowing how to set
up the membership functions is a matter of concern.
Wrong membership values would result in the activities
of the algorithm. In fact, not all recommender system
applications require fuzzy parameters; thus, for the sake of

Fig. 3. The FARAMS algorithm.

stability and processing time, the fuzzification step should

be reduced.
b) The limitation of rating data in the NHSM-based filtering algorithm is available.
c) The FARAMS algorithm could be regarded as an efficient method for calculating the similarity between items.

3.4. HUFCF
The basic concept of the HUFCF method [42] is to
integrate the fuzzy similarity degrees between users based
on the demographic data, with the hard user-based degrees
calculated from the rating histories integrated into the final

95

L.H. Son / Information Systems 58 (2016) 87104

Table 6
The experimental results using the Hold-out cross validation method.
Dataset
MAE values
MovieLens
Jester
RMSE values
MovieLens
Jester
Computational
MovieLens
Jester
a

"
Fig. 4. The HUFCF algorithm.

similarity degrees. As such, those degrees would reflect more

exactly the correlation between users in terms of the internal
(attributes of users) and external information (interactions
between users). Each similarity degree (fuzzy/hard) is accompanied by weights automatically calculated according to the
numbers of analogous users. After the final similarity degrees
are calculated, the final rating will be constructed based on
the rating values of neighbors of the considered user. Depending on the domain of a specific problem, the final rating will
be approximated to its nearest value in that domain accompanied by an error threshold, which is normally less than 5%.
A list of nearest values with equivalent error thresholds is also
given as the prediction ratings of a user for an item. Fig. 4
illustrates the concept in detail.
The limitations of the HUFCF algorithm are as follows:
a) If the demographic data are not provided in the data list,
the HUFCF algorithm does not work because the rating
data have no prior ratings of the new user. Thus, the
similarities between the new user and others cannot be
calculated, and the final rating cannot be found.
b) Similar to the deficiencies of the MIPFGWC-CS algorithm, the Pearson coefficient cannot accurately measure the correlation. Thus, a better similarity metric
should be used instead of the Pearson coefficient.
c) In the branch of demographic data, the GFD matrix is
calculated from all users in the system. Indeed, irrelevant
users may be included in the computation of similarities,
thus degrading the performance of the prediction.

"

"

NHSM

FARAMS

HUFCF

0.701

0.641
0.821a

0.636a
0.899

0.697
0.895

0.818a
1.040a

0.878
1.091

0.903
1.098

5.36a
12.56a

31.88
44.94

436.58
362.47

0.866

time (s)
1335.49

Smallest value for a given dataset.

programming language and executed them on a PC

with an Intel Pentium 4 CPU 2.66 GHz, 1 GB RAM, and
80 GB HDD.
Experimental datasets: we use the following benchmark
RS datasets.
" MovieLens 1 M [23]: contains 1,000,209 anonymous
ratings of approximately 3900 movies provided by
6040 MovieLens users. Ratings are discrete values
from 1 to 5. Demographic data are provided in the
following form: Gender: Age: Occupation: Zipcode.
" Jester [12]: contains ratings of 100 jokes from 73,421
users. Ratings are real values ranging from #10 to
10. The value 99 corresponds to null not
rated. Demographic data are no longer supported
for this dataset.
Generating cold-start users: we adopt the Hold-out and
the k-fold cross validation methods, where the users in
the testing set are the cold-start users. For each coldstart user, we use those algorithms to predict the
ratings for items that have been rated, except three
rated items selected to be the basis for the calculation
of similarity. Each trial is measured by the evaluation
indices. The final results are computed as the average
value of those according to users and trials.
Evaluation indices: we use the Mean Absolute Error
(MAE) and Root Mean Square Error (RMSE) for the
validation of accuracy.
#
1 X##
MAE
p # r u;i #;
22
N u;i u;i
RMSE

"

MIPFGWC-CS

s
"2
1 X!
p #r u;i :
N u;i u;i

23

where pu;i r u;i is the predicted (real) rating of user u for

item i.
Experimental objectives: we compare the accuracy and
the computational time of the algorithms to determine
the most effective algorithm.

4. Experiments
4.1. Environment setup

" Experimental tools: we have implemented MIPFGWC-CS

[46], NHSM [20], FARAMS [17] and HUFCF [42] in the C

4.2. Results and discussion

First, we present the comparative results of the algorithms using the Hold-out cross validation method described

96

L.H. Son / Information Systems 58 (2016) 87104

Fig. 5. The MAE and RMSE values of algorithms on the MovieLens dataset.

Fig. 6. The MAE and RMSE values of algorithms on the Jester dataset.

in Table 6. The experimental results are evaluated by the

evaluation indices. In this table, the results of MIPFGWC-CS
on the Jester dataset are null because this dataset does not
support the demographic data, which are essential for the
calculation in MIPFGWC-CS. The experimental results have
clearly shown that the MAE, RMSE values and the computational time of NHSM are mostly smaller than those of the
other algorithms. To visualize the experimental results, the
MAE and RMSE values of the algorithms on the MovieLens
and Jester datasets are presented in Fig. 5 and 6, respectively.
It is clear that the MAE and RMSE values of NHSM are
the smallest among all of the algorithms. For instance, the

MAE value of NHSM on the Jester dataset is 0.821, which

approximates to 91.3% and 91.7% of those of FARAMS and
HUFCF, respectively. Similarly, the RMSE value of NHSM
on the Jester dataset is 1.04, which approximates to 95.3%
and 94.7% of those of FARAMS and HUFCF, respectively.
The only case in which the MAE value of NHSM is larger
than those of other algorithms occurs on the MovieLens
dataset with MAE values of NHSM, MIPFGWC-CS, FARAMS
and HUFCF being 0.641, 0.701, 0.636 and 0.697, respectively. Despite this fact, the MAE value of NHSM is only
larger than that of FARAMS and is smaller than those
of other algorithms. Thus, the experimental results have

97

L.H. Son / Information Systems 58 (2016) 87104

shown that NHSM obtains better accuracy than the other

algorithms in terms of the MAE and RMSE evaluation
indices.
Nonetheless, the MAE and RMSE values of the algorithms vary according to the datasets. Specifically, in the
case of the dataset having both the demographic and
rating sets, such as MovieLens, the accuracies of algorithms are considerably better than those in the case of the
dataset having the rating set only, such as Jester. The mean
accuracies of all algorithms in terms of MAE and RMSE on
the MovieLens dataset are 0.67 and 0.87, respectively.
Those values on the Jester dataset are 0.87 and 1.08,
respectively. This result clearly demonstrates that all algorithms would work more efficiently on the data having
both the demographic and rating sets than on the data
having the rating set only. Fig. 5 and 6 demonstrate this
fact with the maximal value of bars calculated on the xaxis in the case of the MovieLens dataset being smaller
than that in the case of Jester.
The computational time is another advantage of NHSM.
This algorithm takes only 5.36 and 12.56 s to produce the
results on the MovieLens and Jester datasets, respectively,
while the other algorithms spend more time than NHSM.
The computational time of FARAMS on the MovieLens and
Jester datasets is 31.88 and 44.94 s, respectively. Similarly,
the computational time of HUFCF is 436 and 362 s,
respectively. The computational time of MIPFGWC-CS on
Table 7
The results of k-fold cross validation on the MovieLens dataset.
Fold

MAE
MIPFGWC-CS

2
3
4
5
6
7
8
9
10
Fold
2
3
4
5
6
7
8
9
10
Fold
2
3
4
5
6
7
8
9
10
a

NHSM

0.804
0.758a
0.782
0.679a
0.798
0.659a
0.705
0.687
0.692
0.684
0.708
0.665
0.723
0.632a
0.693
0.615
0.672
0.591a
RMSE
MIPFGWC-CS
NHSM
1.203
1.072a
1.045
0.945a
0.998
0.909a
0.990
0.858a
0.972
0.963
0.806a
0.982
0.882
0.822a
0.856
0.885
0.824
0.801a
Computational time (s)
MIPFGWC-CS
NHSM
963.26
4.3a
1123.2
5.6a
1254.2
6.3a
1321.2
6.2a
1345.4
6.8a
1235.2
7.7a
1543.6
8.0a
1537.7
10.3a
1843.4
10.9a

Smallest value for a given dataset.

FARAMS

HUFCF

0.790
0.744
0.765
0.658a
0.707
0.641a
0.648
0.605a
0.603

0.806
0.793
0.731
0.698
0.672a
0.652
0.641
0.625
0.608

FARAMS
1.138
1.025
0.986
0.972
0.840a
0.880
0.941
0.859
0.845

HUFCF
1.163
1.032
0.969
0.912
0.893
0.886
0.842
0.843a
0.840

FARAMS
48.3
49.6
51.6
54.3
58.5
60.8
65.2
76.8
80.3

HUFCF
345.32
356.70
489.63
478.34
498.43
552.18
603.22
643.43
668.98

Table 8
The results of k-fold cross validation on the Jester dataset.
Fold

MAE
MIPFGWC-CS

2
3
4
5
6
7
8
9
10
Fold
2
3
4
5
6
7
8
9
10
Fold
2
3
4
5
6
7
8
9
10
a

NHSM
a

0.898

0.844a

0.825a

0.887

0.814a

0.795

0.747a

0.745

0.703a
RMSE
MIPFGWC-CS
NHSM

1.203a

1.102a

1.003a

0.993a

0.987a

1.002

0.982

0.972

0.897a
Computational time (s)
MIPFGWC-CS
NHSM

12.4a

13.4a

15.2a

15.6a

17.2a

18.4a

18.9a

18.5a

18.4a

FARAMS

HUFCF

0.903
0.878
0.853
0.835a
0.819
0.784a
0.793
0.742a
0.723

1.123
1.102
1.097
1.002
0.992
0.947
0.909
0.832
0.808

FARAMS
1.304
1.134
1.145
1.091
0.989
0.963a
0.942a
0.923a
0.912

HUFCF
1.534
1.432
1.269
1.101
0.994
0.982
0.985
0.934
0.915

FARAMS
75.2
76.0
79.4
85.6
88.2
91.8
93.8
96.9
100.3

HUFCF
304.5
334.2
365.6
398.0
415.6
405.9
425.3
420.6
489.5

Smallest value for a given dataset.

the MovieLens dataset is 1335 s. The computational time

reflects the mechanism of an algorithm and its efficiency
in terms of processing. To this extent, NHSM is the most
effective algorithm because it takes little processing time
while keeping the best accuracy. Through the results in
Table 6 and Figs. 5 and 6, we have extracted the following
remarks of the efficiencies of algorithms with the Hold-out
cross validation method.

" The accuracy of NHSM is mostly better than those of

"

the relevant algorithms, such as MIPFGWC-CS, FARAMS

and HUFCF, especially on the RS data, which have both
the demographic and rating data, e.g., MovieLens.
The computational time of NHSM is also better than
that of the other algorithms.

Second, we performed another test of the algorithms

using the k-fold cross validation method to generate the
new user cold-start. The experimental results on the
MovieLens and the Jester datasets are presented in
Table 7 and 8. Each table presents the MAE, RMSE and
the computational time of all algorithms according to the
number of folds. For instance, in the 5-fold setting, the
dataset is randomly divided into 5 parts, in which 4 parts
are used for the training set and the remaining 1 part is

98

L.H. Son / Information Systems 58 (2016) 87104

Fig. 7. The MAE values of algorithms by the number of folds on the MovieLens dataset.

Fig. 8. The RMSE values of algorithms by the number of folds on the MovieLens dataset.

L.H. Son / Information Systems 58 (2016) 87104

99

Fig. 9. The average MAE values of algorithms by the number of folds on MovieLens.

Fig. 10. The average RMSE values of algorithms by the number of folds on MovieLens.

reserved for the testing set. All users in the testing set are
the new user cold-starts, and all their rated items in the
testing set are cleared. Applying the experimental algorithms to the dataset, we calculate the MAE, RMSE and the
computational time of the algorithms. We continue to
randomly divide the original dataset into 5 parts and
perform similar tasks until exceeding 5 repetitions of
division. The final MAE, RMSE and the computational time
of algorithms are calculated to find the average results of
the 5 isolated division times. Through this cross validation
method, we can assume that the generated new user cold-

start is not dependent on the division of the original

dataset, such as in the Hold-out method. In addition, this
makes various settings of the new user cold-start for the
validation of all algorithms.
In Table 7, we illustrate the MAE and RMSE values of
algorithms by the number of folds on the MovieLens
dataset in Figs. 7 and 8, respectively. By taking the average
MAE (resp. RMSE) values of the algorithms by the number
of folds, we create the bar charts in Fig. 9 (resp. Fig. 10).
These figures demonstrate the average accuracy of algorithms regardless of the cross validation method used to

100

L.H. Son / Information Systems 58 (2016) 87104

Fig. 11. The MAE values of algorithms by the number of folds on the Jester dataset.

Fig. 12. The RMSE values of algorithms by the number of folds on the Jester dataset.

generate the new user cold-start. These results clearly

show that the MAE value of NHSM is better than those of
other algorithms, with the numbers being approximate to
90.7%, 96.9% and 95.8% of those of MIPFGWC, FARAMS and
HUFCF, respectively. These numbers in terms of RMSE in
Fig. 10 are 96%, 97.1% and 98.3%, respectively. The descending order of the algorithms in terms of accuracy is NHSM,
FARAMS, HUFCF and MIPFGWC-CS.
From Figs. 1114, we illustrate the experimental results
on the Jester dataset. In Fig. 11 (resp. Fig. 12), the MAE

(resp. RMSE) values of the algorithms by the number of

folds on the Jester dataset is depicted. From these figures,
we calculate the average MAE (resp. RMSE) values of the
algorithms by the number of folds on Jester in Fig. 13 (resp.
Fig. 14).
We clearly recognize that the MAE and RMSE values of
the NHSM algorithm are the smallest among all of the
algorithms. According to Fig. 13, the average MAE value of
NHSM is equal to 99% and 82.4% of those of FARAMS and
HUFCF, respectively. Analogously, the average RMSE value

L.H. Son / Information Systems 58 (2016) 87104

101

Fig. 13. The average MAE values of algorithms by the number of folds on Jester.

Fig. 14. The average RMSE values of algorithms by the number of folds on Jester.

of NHSM in Fig. 14 is equal to 97.2% and 90.1% of those of

FARAMS and HUFCF, respectively. These results clearly
show that NHSM achieves better accuracy than other
relevant algorithms, such as FARAMS and HUFCF, even
on the dataset that has only the rating set, e.g., Jester.
Finally, we present the average computational time of
algorithms on both datasets in Fig. 15. The results have
demonstrated that the processing time of NHSM is still the
smallest among all of the algorithms. Through the results
in Tables 7 and 8 and Figs. 715, we have extracted the

following remarks regarding the efficiencies of algorithms

with the k-fold cross validation method.

" The descending order of the algorithms in terms of

accuracy is NHSM, FARAMS, HUFCF and MIPFGWC-CS,

with the average MAE values being 0.66370.143,
0.68470.130, 0.691 70.288 and 0.731, respectively.
The average RMSE values of these algorithms are
0.91570.1, 0.94370.101, 0.93170.196 and 0.953,
respectively.

102

L.H. Son / Information Systems 58 (2016) 87104

Fig. 15. The average computational time of algorithms on both datasets (s).

" The average computational time of NHSM is approxi-

mately 11 s, which is smaller than that of FARAMS

(74 s), HUFCF (455 s) and MIPFGWC-CS (1351 s).

5. Conclusions
In this paper, we concentrated on the new user coldstart problem that negatively affects the recommender
performance due to the inability of the recommender
systems to produce meaningful recommendations. A comparative review of the relevant studies addressing the new
user cold-start problem was performed according to three
groups, namely, (i) makes use of additional data sources,
(ii) chooses the most prominent groups of analogous users,
and (iii) enhances the prediction using hybrid methods. We
also discussed the advantages and disadvantages of these
groups and noted the typical algorithm(s) of each group.
Details of the typical algorithms along with their theoretical
analyses, such as MIPFGWC-CS [46], NHSM [20], FARAMS
[17] and HUFCF [42], were examined. An experimental
validation on the benchmark recommender systems datasets, namely, MovieLens and Jester, under various settings
of the new user cold-start was performed. The experimental results, which were presented in tables and bar chart
figures, revealed the efficiencies of the algorithms in terms
of accuracy and the computational time.
Through the above results and discussion, our findings
in this article can be summarized as follows. First, NHSM is
the most effective algorithm among all of the investigated
methods in terms of both accuracy and computational
time. The MAE and RMSE values of the NHSM algorithm
are approximately 0.663 70.143 and 0.915 70.1, respectively. The computational time of NHSM is approximately
11 s. Second, the descending order of the algorithms in
terms of accuracy is NHSM, FARAMS, HUFCF and
MIPFGWC-CS. Third, all algorithms, especially NHSM, are
stable according to various cross-validation methods, such

as the Hold-out and k-fold used to generate the new user

cold-start. These concluding remarks have clearly answered
our research question stated in Section 1.
The explanation for those remarks demonstrating the
superiority of the NHSM metric over other algorithms, such
as FARAMS, HUFCF and MIPFGWC-CS, could be observed
from the mechanism of NHSM. In cases of the RS data
having the rating only, such as Jester, HUFCF made use of
the Pearson coefficient, which has some limitations, such as
a poor signal-to-noise ratio and negative spikes, to calculate
the similarity. FARAMS utilized the fuzzy association rules
with the support of a fuzzification method to work with the
rating dataset. As stated in limitation #a of Section 3.3, the
fuzzification could lead to inaccurate prediction results if an
unsuitable method, e.g., the center of gravity or the max/
min method, etc., is applied. The NHSM metric, on the other
hand, utilized a better similarity metric than the Pearson
coefficient and did not employ the fuzzification, thus avoiding the ambiguousness of dealing with the fuzzy parameters and enhancing the prediction accuracy. Therefore,
the accuracy of NHSM is better than those of FARAMS and
HUFCF in this case, as illustrated in Tables 6 and 8. In cases
of the RS data having both the demographic and rating,
such as MovieLens, some algorithms relied on either the
demographic or the rating datasets, such as MIPFGWC-CS
and FARAMS, respectively, such that their accuracies were
not high. A hybrid algorithm using both the demographic
and the rating datasets, such as HUFCF, could be a good
choice. However, the problem of the fuzzification of the
demographic dataset that existed in MIPFGWC-CS, FARAMS
and HUFCF impedes the high accuracy of algorithms in the
context of the new user cold-start situation. An interesting
observation is that if a specific cross validation method,
such as the Hold-out method, is chosen to generate the
cold-start users, the accuracy of NHSM could not be the best
among all, as proven in Table 6. Nevertheless, if taking large
samples and diverse cross validation methods, such as the

L.H. Son / Information Systems 58 (2016) 87104

k-fold in Table 7, the advantages of NHSM are more obvious,

with the number of times that the accuracy of NHSM is
better than those of other algorithms being large. Additionally, NHSM has low computational complexity because it
requires less computational time than other algorithms.
From these results, we could clearly recognize the efficiency
of NHSM.
Referring back to the Introduction section, we recognize the practical implication and insightfulness of the
new user cold-start problem. Therefore, our further
research directions could be (i) proposing a hybrid method
to enhance NHSM in terms of accuracy, (ii) improving the
association rules to large orders in the FARAMS method,
(iii) proposing a general similarity measure that is better
than the NHSM metric, and (iv) investigating applications
of NHSM and its variants to the forecasting problems.
Those directions will enrich the knowledge of developing
techniques in the fields of recommender systems and
applied intelligence in the future.

Acknowledgments
The authors are greatly indebted to the anonymous
reviewers for their comments and their valuable suggestions, which improved the quality and clarity of this paper.
Other thanks are sent to Ms. Hoang Thi Thu Huong, FPT
and Mr. Donald B. Samuel, WHO for the language editing.
This work is sponsored by the NAFOSTED under Contract
no. 102.05-2014.01.
Appendix A. Supporting information
Supplementary data associated with this article can be
found in the online version at http://dx.doi.org/10.1016/j.is.
2014.10.001.

[9]

[10]

[11]

[12]
[13]

[14]

[15]

[16]

[17]

[18]

[19]

[20]

[21]

[22]

References
[1] M. Aharon, et al., OFF-set: one-pass factorization of feature sets for
online recommendation in persistent cold start settings, in: Proceedings of the 7th ACM Conference on Recommender systems,
2013, pp. 375378.
[2] H.J. Ahn, A new similarity measure for collaborative filtering to
alleviate the new user cold-starting problem, Inf. Sci. 178 (1) (2008)
3751.
[3] D. Almazro, G. Shahatah, L. Albdulkarim, M. Kherees, R. Martinez,
W. Nzoukou, A Survey Paper on Recommender Systems, 2010,
arXiv:1006.5278.
[4] J. Basiri, A. Shakery, B. Moshiri, M.Z. Hayat, Alleviating the cold-start
problem of recommender systems using a new hybrid approach, in:
Proceedings of the 5th IEEE International Symposium on Telecommunications (IST 2010), 2010, pp. 962967.
[5] J. Bobadilla, F. Ortega, A. Hernando, J. Bernal, A collaborative filtering
approach to mitigate the new user cold start problem, Knowl.-Based
Syst. 26 (2012) 225238.
[6] W. Carrer-Neto, M.L. Hernndez-Alcaraz, R. Valencia-Garca,
F. Garca-Snchez, Social knowledge-based recommender system.
Application to the movies domain, Expert Syst. Appl. 39 (12) (2012)
1099011000.
[7] C.C. Chen, Y.H. Wan, M.C. Chung, Y.C. Sun, An effective recommendation method for cold start new users using trust and distrust
networks, Inf. Sci. 224 (2013) 1936.
[8] M. Chen, C. Yang, J. Chen, P. Yi, A method to solve cold-start problem in
recommendation system based on social network sub-community and

[23]
[24]

[25]

[26]

[27]

[28]

[29]

[30]

103

ontology decision model, in: Proceedings of the 3rd International

Conference on Multimedia Technology (ICMT 2013), 2013, pp. 159166.
V. Formoso, D. Fernndez, F. Cacheda, V. Carneiro, Using profile
expansion techniques to alleviate the new user problem, Inf.
Process. Manag. 49 (3) (2013) 659672.
S. Ge, X. Ge, An SVD-based collaborative filtering approach to
alleviate cold-start problems, in: Proceedings of the 9th IEEE
International Conference on Fuzzy Systems and Knowledge Discovery (FSKD 2012), 2012, pp. 14741477.
G. Guo, Integrating trust and similarity to ameliorate the data
sparsity and cold start for recommender systems. in: Proceedings
of the 7th ACM Conference on Recommender systems, 2013,
pp. 451454.
Jester, Jester Online Joke Recommender Dataset, http://www.ieor.
berkeley.edu/ * goldberg/jester-data/ (accessed September 2013).
H.N. Kim, A. El-Saddik, G.S. Jo, Collaborative error-reflected models
for cold-start recommender systems, Decis. Support Syst. 51 (3)
(2011) 519531.
Y.S. Kim, et al., Hybrid techniques to address cold start problems for
people to people recommendation in social networks, PRICAI 2012:
Trends in Artificial Intelligence, Springer, Berlin, Heidelberg, 2012,
206217.
X.N. Lam, T. Vu, T.D. Le, A.D. Duong, Addressing cold-start problem in
recommendation systems, in: Proceedings of the 2nd ACM International Conference on Ubiquitous Information Management and
Communication, 2008, pp. 208211.
C.W.K. Leung, S.C.F. Chan, F.L. Chung, A collaborative filtering framework based on fuzzy association rules and multiple-level similarity,
Knowl. Inf. Syst. 10 (3) (2006) 357381.
C.W.K. Leung, S.C.F. Chan, F.L. Chung, An empirical study of a crosslevel association rule mining approach to cold-start recommendations, Knowl.-Based Syst. 21 (7) (2008) 515529.
B. Lika, K. Kolomvatsos, S. Hadjiefthymiades, Facing the cold start
problem in recommender systems, Expert Syst. Appl. 41 (4) (2014)
20652073.
J. Lin, K. Sugiyama, M.Y. Kan, T.S. Chua, Addressing cold-start in app
recommendation: latent user models constructed from twitter
followers, in: Proceedings of the 36th International ACM SIGIR
Conference on Research and Development in Information Retrieval,
2013, pp. 283292.
H. Liu, Z. Hu, A. Mian, H. Tian, X. Zhu, A new user similarity model to
improve the accuracy of collaborative filtering, Knowl.-Based Systems 56 (2014) 156166.
N.N. Liu, X. Meng, C. Liu, Q. Yang, Wisdom of the better few: cold
start recommendation via representative based rating elicitation, in:
Proceedings of the 5th ACM conference on Recommender systems,
2011, pp. 3744.
N. Manouselis, et al., Recommender systems challenge 2012, in:
Proceedings of the 6th ACM Conference on Recommender Systems,
2012, pp. 353354.
MovieLens, Movie Lens dataset, http://grouplens.org/datasets/
movielens/ (accessed September 2013).
U.B.M.M. Nazir, A. Yadav, A mechanism for handling cold start in
book recommender system by sharing student profile, Int. J. Appl.
Innov. Eng. Manag. 2 (12) (2013) 318322.
E. Negre, F. Ravat, O. Teste, R. Tournier, Cold-start recommender
system problem within a multidimensional data warehouse, in:
Proceedings of the IEEE 7th International Conference on Research
Challenges in Information Science (RCIS 2013), 2013, pp. 18.
M. Nilashi, O.B. Ibrahim, N. Ithnin, Hybrid recommendation
approaches for multi-criteria collaborative filtering, Expert Syst.
Appl. 41 (8) (2014) 38793900.
D. Poirier, F. Fessant, I. Tellier, Reducing the cold-start problem in
content recommendation through opinion classification, in: Proceedings of the IEEE/WIC/ACM International Conference on Web
Intelligence and Intelligent Agent Technology (WI-IAT 2010), 1, 2010,
pp. 204207.
C. Preisach, L.B. Marinho, L. Schmidt-Thieme, Semi-supervised tag
recommendation-using untagged resources to mitigate cold-start
problems, Advances in Knowledge Discovery and Data Mining,
Springer, Berlin, Heidelberg, 2010, 348357.
T. Qiu, G. Chen, Z.K. Zhang, T. Zhou, An item-oriented recommendation algorithm on cold-start problem, Europhys. Lett. 95 (5) (2011)
58003.
L. Quijano-Snchez, D. Bridge, B. Daz-Agudo, J.A. Recio-Garca, A
case-based solution to the cold-start problem in group recommenders, Case-Based Reasoning Research and Development, Springer,
Berlin, Heidelberg, 2012, 342356.

104

L.H. Son / Information Systems 58 (2016) 87104

[31] F. Ricci, L. Rokach, B. Shapira, Introduction to Recommender Systems

Handbook, Springer, US, 2011.
[32] A.N. Rosli, T. You, I. Ha, K.Y. Chung, G.S. Jo, Alleviating the cold-start
problem by incorporating movies facebook pages, Clust. Comput.
(2014) 111.
[33] L. Safoury, A. Salah, Exploiting user demographic attributes for
solving cold-start problem in recommender system, Lect. Notes
Softw. Eng. 1 (3) (2013) 303307.
[34] A. Said, E.W. De Luca, B. Kille, B. Jain, I. Micus, S. Albayrak, KMulE: a
framework for user-based comparison of recommender algorithms,
in: Proceedings of the 2012 ACM International Conference on
Intelligent User Interfaces, 2012, pp. 323324.
[35] A. Said, B.J. Jain, S. Albayrak, Analyzing weighting schemes in
collaborative filtering: cold start, post cold start and power users,
in: Proceedings of the 27th Annual ACM Symposium on Applied
Computing, 2012, pp. 20352040.
[36] A. Said, T. Plumbaum, E.W. De Luca, S. Albayrak, A comparison of
how demographic data affects recommendation, in: Adjoint Proceedings of the 19th International Conference on User modeling,
Adaption, and Personalization, 2011.
[37] B. Shapira, Recommender Systems Handbook, Springer, US, 2011.
[38] L.H. Son, B.C. Cuong, P.L. Lanzi, N.T. Thong, A novel intuitionistic
fuzzy clustering method for geo-demographic analysis, Expert Syst.
Appl. 39 (10) (2012) 98489859.
[39] L.H. Son, B.C. Cuong, H.V. Long, Spatial interactionmodification
model and applications to geo-demographic analysis, Knowl.-Based
Syst. 49 (2013) 152170.
[40] L.H. Son, N.D. Linh, H.V. Long, A lossless DEM compression for fast
retrieval method using fuzzy clustering and MANFIS neural network, Eng. Appl. Artif. Intell. 29 (2014) 3342.
[41] L.H. Son, Enhancing clustering quality of geo-demographic analysis
using context fuzzy clustering Type-2 and particle swarm optimization, Appl. Soft Comput. 22 (2014) 566584.
[42] L.H. Son, HUFCF: a hybrid user-based fuzzy collaborative filtering
method in recommender systems, Expert Syst. Appl. 41 (15) (2014)
68616870.
[43] L.H. Son, Optimizing municipal solid waste collection using chaotic
particle swarm optimization in GIS based environments: a case
study at Danang City, Vietnam, Expert Syst. Appl. 41 (18) (2014)
80628074.

[44] L.H. Son, DPFCM: a novel distributed picture fuzzy clustering

method on picture fuzzy sets, Expert Syst. Appl. 42 (1) (2014)
5166.
[45] L.H. Son, P.L. Lanzi, B.C. Cuong, H.A. Hung, Data mining in GIS: a
novel context-based fuzzy geographically weighted clustering algorithm, Int. J. Mach. Learn. Comput. 2 (3) (2012) 235238.
[46] L.H. Son, N.T.H. Minh, K.M. Cuong, N.V. Canh, An application of fuzzy
geographically clustering for solving the cold-start problem in
recommender systems, in: Proceeding of 5th IEEE International
Conference of Soft Computing and Pattern Recognition (SoCPaR
2013), 2013, pp. 4449.
[47] D. Sun, C. Li, Z. Luo, A content-enhanced approach for cold-start
problem in collaborative filtering, in: Proceedings of the 2nd IEEE
International Conference on Artificial Intelligence, Management
Science and Electronic Commerce (AIMSEC 2011), 2011, pp. 4501
4504.
[48] M. Sun, F. Li, J. Lee, K. Zhou, G. Lebanon, H. Zha, Learning multiplequestion decision trees for cold-start recommendation, in: Proceedings of the Sixth ACM International Conference on Web Search and
Data Mining, 2013, pp. 445454.
[49] M. Vozalis, K.G. Margaritis, Collaborative filtering enhanced by
demographic correlation, in: Proceedings of the AIAI Symposium
on Professional Practice in AI of the 18th World Computer Congress,
2004.
[50] W. Wang, D. Zhang, J. Zhou, COBA: a credible and co-clustering
filterbot for cold-start recommendations, Practical Applications of
Intelligent Systems, Springer, Berlin, Heidelberg, 2012, 467476.
[51] Y. Xie, Z. Chen, K. Zhang, C. Jin, Y. Cheng, A. Agrawal, A. Choudhary,
Elver: recommending facebook pages in cold start situation without
content features, in: Proceedings of the IEEE International Conference on Big Data 2013, 2013, pp. 475479.
[52] D. Zhang, Q. Zou, H. Xiong, CRUC: Cold-start Recommendations
Using Collaborative Filtering in Internet of Things, 2013, arXiv:1306.
0165.
[53] Z.K. Zhang, C. Liu, Y.C. Zhang, T. Zhou, Solving the cold-start problem
in recommender systems with social tags, Europhys. Lett. 92 (2)
(2010) 28002.
[54] K. Zhou, S.H. Yang, H. Zha, Functional matrix factorizations for coldstart recommendation, in: Proceedings of the 34th ACM International Conference on Research and Development in Information
Retrieval, 2011, pp. 315324.