See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/257704736 Mobile application service networks: Apple’s App Store Article in Service Business · March 2014 DOI: 10.1007/s11628-013-0184-z CITATIONS 32 READS 837 4 authors, including: Jieun Kim Massachusetts Institute of Technology 35 PUBLICATIONS 111 CITATIONS SEE PROFILE Chulhyun Kim Induk University 31 PUBLICATIONS 453 CITATIONS SEE PROFILE Hakyeon Lee Seoul National University of Science and Technology 59 PUBLICATIONS 895 CITATIONS SEE PROFILE All content following this page was uploaded by Jieun Kim on 27 May 2014. The user has requested enhancement of the downloaded file.
28
Embed
Jieun Kim Yongtae Park Chulhyun Kim - ResearchGate · Jieun Kim • Yongtae Park • Chulhyun Kim ... e-mail: [email protected] 123 Serv Bus DOI 10.1007/s11628-013-0184-z. 1 Introduction
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/257704736
Mobile application service networks: Apple’s App Store
Article in Service Business · March 2014
DOI: 10.1007/s11628-013-0184-z
CITATIONS
32
READS
837
4 authors, including:
Jieun Kim
Massachusetts Institute of Technology
35 PUBLICATIONS 111 CITATIONS
SEE PROFILE
Chulhyun Kim
Induk University
31 PUBLICATIONS 453 CITATIONS
SEE PROFILE
Hakyeon Lee
Seoul National University of Science and Technology
59 PUBLICATIONS 895 CITATIONS
SEE PROFILE
All content following this page was uploaded by Jieun Kim on 27 May 2014.
The user has requested enhancement of the downloaded file.
Next, to identify the growth pattern of the mobile service sector, the numbers of
apps accumulated over time was examined for each category (see Fig. 2). Game has
increased exponentially since the App Store opened; Entertainment and Books
follow, with a roughly one-month time lag. Education, lifestyle, and utilities have
been rising since early 2009 and the others since mid-2009. Unlike other categories,
Travel began to grow suddenly and rapidly after May 2009. Thus, game and
entertainment services have played trigger roles in the expansion of mobile apps,
but books and education and leisure and life-relevant services are also becoming
important mobile app services. The other sectors (except weather) are in similar
positions—growing and reaching around 2,000.
In terms of pricing, 30 % of all apps are free, with an average price of $2.56.
More than 50 % of paid apps are fixed at low prices, under $0.99 in all categories.
However, the distributions of app prices are not identical among categories.
Services that provide simple information, such as news and weather, or have free
business models, such as social networking, entertainment, and games, have a
greater proportion (about 80 %) of free or cheap apps. On the contrary, services
dealing with professional information, such as medical, navigation, and reference,
or charged contents such as books have a relatively large proportion (about 10 %) of
paid apps costing more than $10. Medical, navigation, and business have expensive
apps, costing over $50. There are some extremely expensive apps, some costing
$999.99; MobiGage NDI in business, for instance, is a metrology iPhone app used in
Fig. 1 Total number and proportion of mobile app services
J. Kim et al.
123
the inspection of manufactured parts and assemblies and has very specialized
functions such as measurement methodologies and industry-standard fitting
algorithms. The distribution of expensive and cheaper apps is also reflected in
their average prices: the category with the highest average price is medical ($7.9),
and that with the lowest is news ($0.8).
With regard to ratings, out of a possible 10 points (5 stars), all apps have a low
average rating (2.78 points), and most are rated as 0 stars. User ratings are often
higher than three stars. This could mean that users are seldom satisfied with their
apps or tend to evaluate them after downloading them (producing a low response
rate). The distributions of the app ratings are somewhat different among categories.
Games has a significantly higher rating than the others (about 70 % of its scores are
higher than 3 stars, and the average is 5.41 points), whereas productivity (with a
1.13 average), books, and travel have a greater proportion of 0 stars (about 80 %).
Both price and rating affect developers’ revenues and users’ satisfaction, but they
contribute differently across service sectors. It is thus worth mapping the categories in
a price-rating matrix, as in Fig. 3. The price-rating matrix indicates a clear negative
relationship between rating and price. When price is higher, the rating lowers almost
linearly (as in games, finance, education, and books). However, several categories
have comparatively high ratings despite their high price (as in medical, navigation,
business, and reference), low ratings despite their medium price (as in travel and
productivity), or medium ratings despite their low price (as in the rest).
So far, the descriptive statistics have shown the market trends in the Apple App
Store, which can be helpful in identifying the App Store’s general phenomena.
However, they are not enough for understanding the characteristics of the store’s
‘‘mobile app services’’. This paper focuses on the relational characteristics among
the mobile service sectors and mobile app services; a network analysis identifies
Fig. 2 Service growths in category
Mobile application service networks: Apple’s App Store
123
structural characteristics such as the cohesion of services in a certain mobile service
sector and positional characteristics such as the App Store’s most influential service
sector.
4 Research framework
4.1 Method
This paper develops the mobile app service networks through text-mining-based
network analysis approach. Much of the literature on network analysis investigates
links typically through citation or co-citation. However, citation analysis has
fundamentally two main limitations: new patents tend to be less cited than old ones
and may miss citations to contemporary patents; citation-based analysis cannot be
used for patents in databases which do not require citations (Yoon and Kim 2012).
Responding these problems, several literatures have attempted to construct network
through other bibliometric techniques which uses the sharing of keywords or the
similarities of properties between documents. The examples of keyword-based
approaches are text-mining-based patent network (Yoon and Park 2004), property-
function-based patent network (Yoon and Kim 2011, 2012), a semantic networks of
keywords (Kim 2008), and co-word network (Callon et al. 1991; Engelsman and
van Raan 1994; Su and Lee 2010). These papers are based on an assumption that
sharing the same keyword implies these two documents (patent or research)
partially overlap each other. In the same manner, this paper assumes that the sharing
of same keyword corresponds to the overlap between app services, and the
similarity based on this overlap can be interpreted as the relationship. To produce a
Fig. 3 Price-rating matrix
J. Kim et al.
123
database (including an app-by-app description), a text-mining-based network
analysis was applied in the following four stages:
1. Setting of keyword vector: since the detailed descriptions of each app are
expressed in natural language, a text mining that extracts keywords from
documents was used to transform the unstructured data into analyzable
structured data.
2. Construction of association matrix: based on the keyword vector, an association
matrix was constructed using the relationship among services, quantified in
terms of distance or similarity.
3. Development of service network: by applying the association matrix as input, a
service network was generated with nodes (categories or apps) and links
rendered visually.
4. Analysis of interrelationships: the interrelationships among categories or
services were analyzed based on quantitative indexes.
4.1.1 Setting of keyword vector
Text mining, the process of finding interesting patterns, models, directions, trends,
or rules from unstructured text, is an automated discovery of knowledge from texts
(Berry 2003). Structuring the input text usually involves parsing, along with the
addition and removal of derived linguistic features, and subsequent insertion into a
database. In text mining, a keyword vector is the general method of handling large
amounts of unstructured text to extract information from structured data (Yoon and
Park 2004).
This study’s text mining extracted 2,357 keywords from the documents, but only
563 keywords were selected after the elimination of redundant words and the
consideration of total occurrence and semantic meaning. Using the selected
keywords, documents with no occurrences were eliminated from the 100,830 apps;
some apps were not relevant to the keywords because they had very few keywords
by which they analyze the characteristics. Then, the frequencies of the keywords’
appearance in individual documents were entered into the keyword vector, resulting
in the keyword vectors of 1,919 apps, with 563 keywords ultimately constituted.
Table 3 shows the frequencies of the keyword vectors resulting from the text
mining. The keyword vectors were used to construct an association matrix and to
conduct the network analysis.
4.1.2 Construction of association matrix
The association matrix was constructed by quantifying the degree of similarity
between keyword vectors. This study used cosine similarity, a representative
measurement for the similarity between two vectors of n dimension, and calculated
the cosine of the angle between them. The cosine similarity is represented using a
dot product and magnitude, as below:
Mobile application service networks: Apple’s App Store
123
Vi � Vj
Vij j Vj
����
where Vi and Vj are the keyword vectors of mobile apps or categories.
The resulting cosine similarity ranges from -1 (exactly opposite) to 1 (exactly
the same), with 0 usually indicating independence and the in-between values
indicating intermediate similarity or dissimilarity. In the case of information
retrieval, the cosine similarity of two mobile app services ranges from 0 to 1
because the frequency of keywords is positive.
4.1.3 Development of service network
The degree of connectivity was decided based on the threshold value that the analyzer
is supposed to determine. The connectivity between Vi and Vj was set to 1 in the
association matrix if the cosine similarity was larger than the selected threshold value.
Otherwise, the connectivity was set to 0 and considered a weak relationship.
Determining the threshold value was subjective, and the results may be strongly
dependent on the threshold value. A few representative services can be selected if we
set a higher threshold values, while many representative services can be identified if
we set a lower threshold value. There were two alternatives. At an intermediate level,
the decision maker could select a reasonable threshold value so that the number of
representative services would become clearly relevant, or multiple threshold values
could have been applied in a sensitivity analysis. In this paper, the threshold values
were selected to yield the relevant number of representative services. After deciding
on the threshold value, service networks were developed using visualization software.
The networking software package UCINET 6, a popular network analysis program,
was used to depict the network and compute the quantitative indexes.
4.1.4 Analysis of interrelationships
Structure and interrelationship characteristics were identified through the quantitative
indexes drawn from network theory as complements to visualization for effective
description. This paper considers two network characteristic types, network structure
property, and node centrality, measured using metrics such as density, centralization,
degree centrality, closeness centrality, and betweenness centrality, among the primary
methods of understanding networks and their participants.
Regarding network structure property, we examined the shape of the network by
density and centralization. First, network density represents the degree of interaction
Table 3 Example of keyword
vectorCity Video Story Online TV Upload
L152 1 0 1 0 0 1
L54 1 0 1 0 0 1
L263 1 1 0 1 1 0
L109 0 1 1 0 0 1
L205 0 1 1 0 0 1
J. Kim et al.
123
in a network (Meagher and Rogers 2004), calculated as the proportion of the
number of actual relations between categories divided by the maximum possible
number of ties that would be present if the network were complete (Scott 2000).
This is based on the idea that the more the actors are connected to one another, the
more cohesive their network is. Second, network centralization indicates whether
the interaction is equally distributed or centralized on a few nodes. While
centralization provides information about the compactness of the overall structure of
the network (like its density), while density indicates the overall level of network
cohesion, centralization measures the degree to which an entire network is focused
around a few central nodes (Wasserman and Faust 1994). Thus, density corresponds
to the concept of ‘‘mean’’ inferred from the number of relationships between actors,
while centralization corresponds to the ‘‘variance’’ in the interrelationships.
Consequently, density and centralization can provide important clues to the nature
of network structures. For instance, if a network is a complete type in which all
nodes are linked, its density will be 1 and its centralization 0. If a network is a circle
type in which the actors form a big circle, its density will be 0.5 and its
centralization 0. If a network is a star type, in which the actors are related to only
one central hub, its density will be 0.5 but its centralization 1.
Node centrality is a primary measure used to evaluate the position of actors in a
network. It is divided into three types of centrality: degree, betweenness, and
closeness (Freeman 1979). It has been argued that centrally located services occupy
strategic positions that allow them greater access to information, knowledge, and
resources. This applies particularly to the context of the mobile App Store, where
potentially complementary technologies, information, and services are dispersed
among numerous firms. First, degree centrality refers to the number of ties the actors
have. This paper does not distinguish between in-degree or out-degree centrality
(Scott 2000) because we assume that the relationship between two actors does not
have a direction. Actors with a high relationship degree are generally connected or
adjacent to many actors and should be considered as being in a prominent location
where ‘‘value’’ flows. A low degree tends to characterize actors at the periphery.
However, degree centrality may be criticized because it takes into account only the
immediate ties an actor has and ignores the indirect ties to others. In many instances,
a firm may be tied to a large number of other firms that are rather disconnected from
the network as a whole. Second, in response to this deficiency, closeness centralityhas been used to emphasize the ‘‘nearness’’ of an actor to all others in the network
using the reciprocal of the geodesic distances (Scott 2000). Lastly, betweennesscentrality measures the extent to which actor k lies on the path ‘‘between’’ the other
actors in the network: an actor of relatively low degree may play an important
intermediary role and so be very central in the network. The existence of such a
structural hole allows the relevant actor to act as a broker (Freeman 1979). Table 4
summarizes the indexes used to analyze network interrelationships.
4.2 Overall procedure
The objective of this paper is to gain a deeper understanding of the structure and
complexity of the relationships among the mobile app services of the App Store. Its
Mobile application service networks: Apple’s App Store
123
overall procedure is organized as shown in Fig. 4. Network visualizations were
generated in a top-down manner by creating first a macro view of the mobile app
categories and then a micro view of the mobile app services. Thus, the networks
were developed in two levels—a category network for macro-level analysis and an
app network for micro-level analysis. In the category network, the centrality
measures (the degree, closeness, and betweenness) of each category, representing
the position and influence of the category in the macro-level network, were derived.
We divided the mobile app categories into utilitarian and hedonic segments based
on their values; thus, whether the centrality measures are different between the two
segments can be assessed. This was analyzed by a Mann–Whitney U test, a non-
parametric statistical test for two unpaired groups. We selected a non-parametric
test because there are only small centrality data samples from non-Gaussian
populations for the two segments. An app network was constructed for each
category, and the network structure property measures (of density and centraliza-
tion), indicating the structural shape and compactness of each network, were
computed. Then, a Mann–Whitney test was conducted on these measures. Finally, a
cluster analysis on the mobile app categories was performed to identify the new
taxonomy based on the relationships among the mobile service sectors.
5 Networks of mobile app services
This section presents the results of the visualization and analysis of the mobile app
service networks of the two sub sections—the macro-level category and micro-level
app networks—while also incorporating the Mann–Whitney test for comparing the
network indexes of the utilitarian and hedonic segments.
5.1 Macro-level analysis: category network
In the macro view, the categories the apps belong to are represented as a single
vertex. Figure 5 shows the global relations among all categories. In the macro-level
analysis, 1,919 apps’ keyword vectors were merged according to their categories,
and the average frequency was filled with the keyword vectors of 20 categories.
Some categories may have consisted of more apps than others, and a summation of
Table 4 Definition of network indexes
Class Index Definition
Network structure
property
Density The degree of the overall level of network cohesion and
interaction
Centralization The degree to which an entire network is focused
around a few central nodes
Node centrality Degree The number of direct edges nodes has
Closeness The nearness of an node to all (direct and indirect) other nodes
Betweenness The extent to which node lies on the path between the various
other nodes
J. Kim et al.
123
the links could have reflected tie strengths; thus, we normalized the cosine similarity
with the overall number of apps in the categories. As shown in Fig. 4, the thickness
of their edges is proportional to the degree of linkage between nodes. To visually
differentiate between the segments’ categories, colors were used for the nodes.
Utilitarian segments were depicted with black spheres, whereas hedonic segments
were depicted with white spheres.
The visualization prompts several key observations. First, several central
categories appear in the utilitarian and hedonic segments of the App Store
structure. Utilities in the utilitarian segment and entertainment in the hedonic
segment seem to be the most central categories. Meanwhile, specialized categories
from the utilitarian segment such as navigation, medical, finance, and healthcare and
fitness appear to be relatively peripheral to the rest of the App Store. Second, there
seems to be a strong relationship among education and reference, utilities and
finance, and utilities and business. The reference category in the App Store is a
portal to new offerings that not only fit under books and guides but also include
interactive graphics and audio for both children and adults. It is thus natural for
reference to be associated with educational services. The contributing keywords are
Mobile application service networks: Apple’s App Store
123
are utilities, lifestyle, games, education, and business; however, their index values
differ slightly. For example, education, games, and lifestyle have the same degree
but occur as game, education, and lifestyle of closeness; hence, the indirect
interaction is active in game but inactive in lifestyle. The betweenness of business
and travel is comparatively high; thus, they seem to have an intermediary role in the
whole network. The results also show that relative importance does not depend on
the scale of the category.
Lastly, to assess whether the node centralities differ between utilitarian and
hedonic segments, a Mann–Whitney U test was applied to the data in Table 5 using
SPSS 12.0K software. The p values yielded were 0.220 in degree, 0.046 in
closeness, and 0.422 in betweenness. At the 5 % significance level, the null
hypothesis was rejected only in closeness centrality. Consequently, the closeness
centralities are likely to differ but the degree and betweenness centralities do not
significantly differ between the utilitarian and hedonic segments. The closeness
centrality of the hedonic segment is larger than that of the utilitarian segment. Thus,
the hedonic segment appears to be more central than the utilitarian segment in terms
of the direct and indirect nearness. A plausible explanation for this is that the
categories of the hedonic segment have more general content and functionalities
than do those of the utilitarian segment and thus have more services in common
with other categories. In other words, the categories of the utilitarian segment are
constituted by more specialized services.
Fig. 5 Category network (cutoff value = 0.17)
J. Kim et al.
123
5.2 Micro-level analysis: app network
At the micro level, we constructed app networks for each category to examine the
characteristics of each mobile service sector in more detail. For example, to develop
the utility app network, 103 apps’ keyword vectors were used to yield an association
matrix; the resulting visualization is shown in Fig. 6. Network metrics were also
computed: the network density is 0.0305 and the centralization 0.0725; the top five
services with high centrality are listed in Table 6.
First, regarding network structure property, both density and centralization are
lower than those of the previous networks. The degree of interaction between apps
in Utility and the degree to which a few apps dominate the relationship seem not
much greater than in other categories. Second, regarding node centrality, the service
listed in Table 6 turns out to be the most influential service as intuitively identified
by the visual network (and depicted by red spheres). The service with the highest
average rank of three index values is HodgePodge (S601), which is a utility
collection with a clean, user-friendly interface. The nine utilities include location,
battery, alarm clock, tip calc, converter, flashlight, ruler, level, and random number.
These utilities appear to be representative of the basic functions that help us in our
daily lives and are thus influenced by many apps, such as Battery God Lite (S56),
Battery–Flashlight (S58), and Utilitybox (S248). Thus, they naturally exhibit an
active flow of knowledge across other services.
Table 5 Node centrality of category network
Segment Category Degree Closeness Betweenness
Utilitarian Business 0.3158 0.3725 0.0684
Education 0.3684 0.3800 0.1268
Finance 0.1579 0.3220 0.0058
Healthcare and fitness 0.1579 0.3167 0.0049
Medical 0.0526 0.2836 0
Navigation 0.1053 0.2969 0.0039
News 0.1053 0.3167 0
Productivity 0.2632 0.3257 0.0194
Reference 0.2105 0.3276 0.0093
Utilities 0.5263 0.4043 0.1615
Weather 0 0.0500 0
Hedonic Books 0.2632 0.3585 0.0251
Entertainment 0.6316 0.4318 0.3018
Games 0.3684 0.3878 0.0476
Lifestyle 0.3684 0.3725 0.0505
Music 0.1579 0.3333 0
Photography 0.2632 0.3585 0.0085
Social networking 0.2105 0.3519 0.0468
Sports 0.0526 0.3245 0
Travel 0.2632 0.3585 0.0611
Mobile application service networks: Apple’s App Store
123
The same micro-level analysis was conducted on the remaining 19 categories. Thus,
20 individual app networks were developed and their network metrics computed.
Table 7 shows the results of the network structure properties (Note that the top five
values are represented in bold strokes for each index), and Table 8 represents the most
central apps selected by the average rank of the three node centrality measures. Density
and centralization show similar patterns: if the density is high, centralization is high,
meaning that most app networks are star-shaped, not circular. However, the overall level
of density and centralization is quite low, and absolute connectivity is not frequent. The
categories with relatively high density and centralization are photography, music,
reference, and navigation; thus, they have influential leader services that dominate the
others, such as Felaur PDF Reader (Photography), Guitar Jam (Music), EnglishDictionary and Thesaurus by Ultralingua (Reference), and Sygic Mobile Maps SEAsia—Turn-by-Turn Voice Guided GPS Navigation (Navigation). Since most of the
services in all categories flock to these central services, they develop into similar classes
in terms of content and functionality. Thus, central apps in highly centralized networks
will be representative of most of those service sectors.
In the Mann–Whitney U test for the data in Table 7 assessing whether the
network structure properties differ between utilitarian and hedonic segments, the
p values yielded 0.766 in density and 0.970 in centralization. Consequently, all of
the null hypotheses are accepted, and density and centralization are not likely to
differ between the segments.
Fig. 6 App network of Utilities (cutoff value = 0.2)
J. Kim et al.
123
6 Clusters of mobile app services
As shown in the previous networks, the network structure and importance properties
vary across the 20 mobile App Store categories. This study has attempted to
compare between the utilitarian and hedonic segments’ network indexes using a
value-based typology proactively defined based on mobile service value, but a
significantly different measure, closeness centrality, appeared. Mobile app service
categories can thus be re-categorized according to a relationship-based taxonomy.
In order to group the mobile app service categories based on the pattern of network
characteristics, a cluster analysis was implemented using five network indexes.
Network structure property measures were gathered from app networks (see
Table 7) to include the internal cohesion of app services in categories, and node
Table 6 Top five node centrality of app network of utilities
Rank Degree Closeness Betweenness
1 S324 0.9141 S601 0.1686 S601 0.1935
2 S128 0.7408 S385 0.1667 S385 0.1315
3 S601 0.7354 S617 0.1643 S168 0.0763
4 S22 0.7293 S260 0.1637 S260 0.0755
5 S172 0.7262 S324 0.1635 S98 0.0655
Table 7 Network structure
property of app networksCategory Density Centralization
Books 0.0575 0.0862
Business 0.0477 0.0707
Education 0.0591 0.1011
Entertainment 0.0246 0.0516
Finance 0.1251 0.1439
Games 0.0373 0.0746
Healthcare and Fitness 0.0440 0.0770
Lifestyle 0.0210 0.0414
Medical 0.0619 0.0621
Music 0.1708 0.1954
Navigation 0.1222 0.1712
News 0.1016 0.1015
Photography 0.1809 0.1956
Productivity 0.0440 0.0821
Reference 0.1494 0.1758
Social networking 0.0988 0.1531
Sports 0.1058 0.1574
Travel 0.0586 0.0918
Utilities 0.0305 0.0725
Weather 0.1050 0.2100
Mobile application service networks: Apple’s App Store
123
Ta
ble
8T
he
most
centr
alap
ps
inap
pnet
work
sof
each
cate
gory
Cat
egory
Cen
tral
app
Des
crip
tion
Books
Jane
Aust
enC
oll
ecti
on
The
coll
ecti
on
of
Jane
Aust
in,
anE
ngli
shnovel
ist
of
real
ism
,in
cludin
gS
ense
and
Sen
sibil
ity,
Pri
de
and
Pre
judic
e,M
ansfi
eld
Par
k,
Em
ma,
Nort
han
ger
Abbey
,
Per
suas
ion
Busi
nes
su
Foll
ow
itA
do
cum
ent
del
iver
yto
bac
ko
ffice
imag
ing
syst
eman
dtr
ack
ing
reg
ard
less
of
ind
ust
ry,
incl
ud
ing
pro
du
cin
gre
al-t
ime
loca
tion
info
rmat
ion
,ca
ptu
rin
gre
al-t
ime
pro
of
of
del
iver
y/s
ervic
eco
mple
tion
wit
hsi
gnat
ure
captu
re,
and
forw
ardin
gal
l
rela
ted
do
cum
enta
tio
nfo
rim
med
iate
refe
ren
ce
Ed
uca
tio
nO
xfo
rd–
Hac
het
teF
ren
chD
icti
on
ary
Aco
mp
lete
gu
ide
toth
eF
ren
chan
dE
ng
lish
lan
gu
ages
,fr
om
gen
eral
tosp
ecia
list
and
tech
nic
alte
rmin
olo
gy
,fr
om
fore
ign
lan
gu
age
tosl
ang
Ente
rtai
nm
ent
Gir
mit
An
appli
cati
on
for
Vid
eoG
irm
itan
dR
adio
Gir
mit
along
wit
hm
ovie
gal
leri
esan
d
Kar
nak
ata
New
sR
SS
feed
,in
clu
din
gK
ann
ada
mu
sic
vid
eos,
mo
vie
trai
lers
,co
med
y
sho
ws,
reci
pes
and
oth
ery
ou
tub
ev
ideo
sre
late
dto
Kan
nad
a
Fin
ance
RE
Cal
c–R
eal
Est
ate
Mo
rtg
age
Lo
an
Cal
cula
tor
Aca
lcula
tor
support
ing
the
month
lypay
men
t,te
rm,
inte
rest
rate
or
loan
amount
for
a
loan
,an
nu
alp
rop
erty
tax
,h
om
eow
ner
’sin
sura
nce
and
mo
rtg
age
insu
ran
ce,
asw
ell
asa
do
wn
pay
men
tam
ou
nt/
per
cen
tag
e
Gam
esT
he
Sim
pso
ns
Arc
ade
Th
eS
impso
nA
rcad
eg
ame
tog
etd
ou
gh
nu
t-ch
asin
gd
ared
evil
try
Hea
lthca
rean
dfi
tnes
sC
ontr
acti
on
Tim
eran
dF
etal
Kic
k
Co
un
ter
Th
eto
ols
aid
ing
for
cert
ain
key
po
ints
asth
ep
rog
ress
of
use
r’s
pre
gn
ancy
,in
clu
din
g
feta
lk
ick
/mo
vem
ent
cou
nte
r,co
ntr
acti
on
inte
rval
/du
rati
on
tim
er,
con
trac
tion
log
,
and
ho
spit
al‘T
oD
o’
chec
kli
sts
Lif
esty
leM
art’
sE
ver
yday
Fo
od
:F
resh
and
Eas
yR
ecip
es
Are
cip
ein
form
atio
nap
pin
clu
din
gre
cip
efi
nder
,sh
uffl
e,w
atch
and
lear
nv
ideo
,
excl
usi
ve
dai
lyar
ticl
esby
food
edit
ors
,din
ner
tonig
ht,
shoppin
gli
stto
ols
,fa
vori
te
list
,an
dsh
are
fun
ctio
nal
ity
Med
ical
Dai
lyD
ose
of
Med
ical
Kn
ow
led
ge
Th
ed
eliv
ery
of
am
edic
ald
ose
ever
yd
ayto
hel
pm
edic
alp
rofe
ssio
nal
sst
aycu
rren
tin
thei
rm
edic
alk
no
wle
dge
Mu
sic
Gu
itar
Jam
Alo
op
-bas
edan
dre
alti
me
musi
csy
stem
that
allo
ws
any
one
tocr
eate
jam
so
nth
efl
y
Nav
igat
ion
Sygic
Mobil
eM
aps
SE
Asi
a–T
urn
-by-
Tu
rnV
oic
eG
uid
edG
PS
Nav
igat
ion
Are
al-t
ime,
turn
-by
-tu
rnv
oic
eg
uid
edn
avig
atio
nsy
stem
wit
h7
-dig
itp
ost
alco
des
and
safe
tyca
mer
as
J. Kim et al.
123
Ta
ble
8co
nti
nu
ed
Cat
egory
Cen
tral
app
Des
crip
tion
New
sL
oca
l1
2M
obil
eL
oca
lN
ews
Lo
cal
new
s,n
atio
nal
new
s,in
tern
atio
nal
new
s,w
orl
dw
ide
bu
sin
ess
new
s,sp
ort
sn
ews
and
sco
res,
wea
ther
,p
oli
tics
,en
tert
ain
men
tn
ews,
wei
rdn
ews
wit
hm
ult
i-co
nn
ecti
on
vid
eosu
pport
,ea
synew
ssh
arin
g,
and
Use
rG
ener
ated
Conte
nt
Ph
oto
gra
ph
yA
ctio
nC
amer
aA
cam
era
app
lica
tio
nd
esig
ned
for
sim
ula
tin
gth
ew
orl
dw
ide
fam
ous
clas
sic
film
cam
era,
wh
ich
can
tak
ea
fou
r-p
ictu
rese
ries
on
on
ep
ho
to
Pro
du
ctiv
ity
Fel
aur
PD
FR
ead
erA
PD
Fre
ader
sup
po
rtin
gp
assw
ord
pro
tect
edP
DF
file
,n
ewP
DF
encr
yp
tio
nfr
om
Imag
es,
Google
PD
Fse
arch
and
dow
nlo
ad
Ref
eren
ceE
ng
lish
Dic
tio
nar
yan
dT
hes
auru
sb
y
Ult
rali
ngua
Aco
lleg
iate
-lev
eldic
tionar
yfo
rst
uden
tsan
dtr
avel
ers,
off
erin
gth
ousa
nds
of
entr
ies,
plu
ssl
ang,
tech
nic
alte
rms,
idio
mat
icex
pre
ssio
ns,
etc.
So
cial
net
work
ing
Fac
ebo
ok
Qu
ick
Sta
tus
Pro
Ato
ol
allo
win
gto
qu
ick
lyp
ost
stat
us
mes
sag
esto
Fac
eboo
kb
ycu
stom
key
boar
ds
to
po
stF
aceb
oo
ksy
mb
ols
,co
mm
on
tex
tte
rms,
and
com
mo
np
hra
ses
Sport
sS
port
sBook:
Pro
Footb
all
Edit
ion
The
late
stpro
fess
ional
footb
all
new
s,up
todat
esc
hed
ule
and
gam
esp
read
s
Tra
vel
Lo
nel
yP
lan
etT
rav
elG
uid
esO
ver
75
gu
ides
and
aud
iop
hra
seb
ook
san
dh
un
dre
ds
of
han
dy
revie
ws,
stri
kin
gim
ages
and
inte
ract
ive
map
s—al
lse
arch
able
,p
ort
able
and
easi
lyp
erso
nal
ized
by
loca
tio
n-
awar
ete
chno
log
y
Uti
liti
esH
od
geP
odg
eA
uti
lity
coll
ecti
on
wit
ha
clea
n,
use
r-fr
ien
dly
inte
rfac
e.T
he
nin
eu
tili
ties
incl
ud
e
loca
tion
,b
atte
ry,
alar
mcl
ock
,ti
pca
lc,
con
ver
ter,
flas
hli
gh
t,ru
ler,
lev
el,
and
rand
om
nu
mb
er
Wea
ther
NA
IPS
Pil
ot
wea
ther
and
NO
TA
Min
form
atio
nd
irec
tfr
om
Air
serv
ices
Au
stra
lia
Mobile application service networks: Apple’s App Store
123
centrality measures were collected from category network (see Table 5) to
incorporate the roles of categories in the overall App Store. A two-step cluster
analysis was performed using the SPSS 12.0K program. First, through hierarchical
clustering, the dendrogram was used to identify the appropriate number of clusters,
determined to be three. Then, a K-means clustering was executed to classify the
categories into their most homogeneous groups. The center of cluster and p value
from the ANOVA are shown in Table 9. The null hypotheses were all rejected at the
10 % significance level; thus, the clusters are different in all five indexes. The
names of the clusters and the classification of the categories are suggested in
Table 10.
The results show explicit differences. First, cluster 1 shows a high level of
density and centralization but a low node centrality. Thus, the categories in this
cluster are specialized and dominated by a few leader services within their service
boundary, due to a lack of content variety, and have little information, content,
function, or knowledge interaction with other service areas. Although they may not
help to mediate the convergence among mobile service areas in the network
category, the overall network is fertilized by the insuring of the internal stability of
the service sector. We named the cluster solitary specialist, and the utilitarian
categories peripheral in the category network correspond mainly to it, as Table 10
shows. Music and sports in the hedonic segment also seem to have specialized
features in relational patterns. Second, cluster 2 is the opposite of cluster 1,
including its low level of network structure property but high node centrality. They
are very comprehensive in their coverage of various relationships between other
service areas, but their internal concentration is inactive. Even though the three
measures of node centrality are high, their degree is higher than their closeness;
thus, their interrelationship is more direct than indirect. Due to their particularly