International Core Journal of Engineering 2020-26 | Page 127

recommendation system and discuss the requirements and
objectives for cellphone recommendation.
structural user choice model (SUCM) to cope with apps in a
hierarchical taxonomy [1]. They modeled the user choices as
two phases, in the first phase, a user decides which type of
apps to choose. In the second phase, the user chooses apps in
the selected category [1]. However, for some complicated
classifications that cannot be split into only two phases, the
method becomes complex and difficult to implement [1].
A. Importance of Cellphone Recommendation
We find out that with all kinds of cellphones in the
cellphone market, consumers are perplexed by the markets
and do not know how to choose a cellphone. For instance,
they are unlikely to know about the functionalities and
performance of a cellphone, and whether they are
comfortable with the operating systems. Also, they do not
know how to satisfy their requirements on cellphone without
wasting much money. Besides, consumers need a visualized
evaluation of specific aspects like photography, business use
to choose from the large scope of market cellphones.
IV. O UR P ROPOSED M ETHOD
Large-scale technology companies like Huawei and
Xiaomi have investigation groups to decide what features
and characteristics should be equipped in the next generation
of their cellphones [4][12]. For example, several corporations
conduct requirement analysis and make modifications based
on the last generation of their products [13]. They may
modify rear cameras to achieve better photographing or hide
the front camera under the screen [14]. Our proposed method
is designed to satisfy consumers in assisting them to choose
from these techniques and facilities, and offer them a suitable
cellphone. However, there are some challenges: 1) the
characteristics of every consumer are highly different, and it
is difficult to model the unique traits using a universal
method; 2) some consumers are unfamiliar with cellphones,
and thus unable to give any information about cellphone
preference; 3) whether a consumer will make a purchase
depends on a complex set of factors. As such, we define
three research questions: 1) what are the factors and features
that impact a cellphone purchase? 2) how to build a
connection between target users’ information and specific
cellphones? 3) how to model the set of factors and features to
represent portraits of every consumer? 4) how to develop the
model to correctly recommend the cellphone with the highest
purchase rate and customer satisfaction rate.
Consequently, we propose a cellphone recommendation
system so as to help consumers choose the most appropriate
cellphones and create the most satisfactory cellphone using
experience. In consideration of the popularity of cellphones,
there are a huge amount of target consumers. As such, the
cellphone recommendation system should be researched on.
B. Business Requirement and Objectives for Cellphone
Recommendation
It is surveyed that there are few functionally-similar
products in the market. Hence, the system possesses a high
market share, as well as indirectly increase the overall profits
from cellphone sales, either by increasing the number of
consumers or by increasing the willingess for each consumer
to buy the suitable cellphone recommended by the system.
Moreover, the requirement also aims to give those cellphone
businesses a larger exposure to the market [7]. In this paper,
we mainly aim at the objectives of increasing the cellphone
purchase rate and the customer satisfaction rate.
Our proposed method, CIMR, answers all the four
research questions defined above. When applied within the
surveyed consumers in an offline way, CIMR is able to fulfill
all the requirements and objectives answered by consumers.
III. R ELATED W ORK
Machine learning techniques are widely applied on
recommender systems and on predicting users’ preferences
[8][9][10][11], and we discuss a few of them in this section.
The cellphone recommendation issue stated in this paper
highly resembles several prediction and decision-making
problems.
A. An Overview of CIMR
The overview of CIMR is shown in the follow as Fig. 1.
A. Personalized Trip Recommendation Based on Users’
information
Researchers from Melbourne University proposed a point
of interest (POI) recommendation mechanism for tourists
based on user interests, visit duration and visit recency in
2017 [8]. It addressed the problems for tourists to choose
POIs of an unfamiliar city considering their own interests [8].
The proposed method formulated the tour recommendation
problems based on the Orienteering problem, which
considered a user’s trip constraints such as time limitations
and the needs for the tour to start and end at specific POIs [8].
There are several drawbacks with this mechanism: first, it
didn’t separate similar POIs to each other, which would lead
to random allocations of those similar POIs; second, user
interests can be conflicted with POI popularity, and may not
come up with a more precise result. In addition, the POIs
lacked enough features to build connections with users.
Fig. 1. Overview of CIMR
The method consists of seven procedures: 1) we obtain
surveyed data about users’ personal situation(i.e. hobbies and
cellphone usage) and their cellphones; 2) we obtain target
consumers who are willing to buy a new cellphone; 3) we
then carry out data preprocessing, splitting users’ data based
one different cellphones; 4) we use the survey data to train
machine learning models to predict the purchase rate
(probability) of a particular consumer; 5) we calculate the
purchase rate for various cellphones; 6) targeted at different
cellphone, we compare the corresponding predicted purchase
B. Mobile App Recommendation
The researchers from IBM research center proposed a
105