A Community Detection Approach for Smart-Phone Addiction

Recognition

Fabio Cozzolino

, Vincenzo Moscato

, Antonio Picariello

and Giancarlo Sperli

DIETI, University of Naples Federico II, Italy

ITEM National Lab, CINI, Naples, Italy

Keywords:

Smart-Phone Addiction, Knowledge Discovery, Social Network Analysis, Community Detection.

Abstract:

In this paper, we present a novel approach for Smart-Phone Addiction recognition that leverages community

detection algorithms from the Social Network Analysis (SNA) theory. Our basic idea is to model data con-

cerning users’ behavior while they are using mobile devices as a particular social graph, discovering by means

of SNA facilities patterns that better identify users with a high predisposition to smart phone addiction. Even-

tually, several experiments on a sample of users monitored for several weeks have been carried out to verify

effectiveness of the proposed approach in correctly recognizing the related addiction degree.

1 INTRODUCTION

During the last years Psycho-Informatics has attracted

more and more the interest of researchers in order to

better understand human behavior within the mod-

ern “data-rich” world: it consists of the application

of novel methodologies for acquisition, management

and analysis of vast quantities of psychological data,

combining behavioral psychology and computer sci-

ence techniques (Markowetz et al., 2014).

Indeed, one of the most natural way for eliciting

nowadays persons’ habit is the analysis of their smart

phones’ data. Unfortunately, the variety of function-

alities that such devices offer – including the use of

the Internet for web browsing, on-line games, digi-

tal cameras, GPS navigation, and a lot of interactive

and social applications – can deeply capture attention

of users, who could be dangerously distracted from

real events ((Hooper and Zhou, 2007)). Several recent

studies (Leung, 2008) have shown how their excessive

use, namely Smart-Phone Addiction, can generate dif-

ferent complications for users’ health, especially psy-

chological pathology: lack of self-control, abstinence,

insomnia, social isolation, depression, difﬁculty of

concentration, as well as signs of irritability, restless-

ness, stress and mood changes (Ch

oliz et al., 2016;

Ha et al., 2008; Ben-Yehuda et al., 2016).

In according to the most recent vision, Smart-

Phone Addiction can be deﬁned as “an unstoppable

and uncontrollable desire of using a smart phone de-

spite its negative and harmful effects” (De-Sola et al.,

2017). The smart-phone addiction does not easily

ﬁt the standard classiﬁcation of psycho pathologi-

cal disorders provided by the Diagnostic and Statis-

tical Manual of Mental Disorders (DSM). Thus, try-

ing to recognize persons’ predisposition with respect

to such new pathology, by means of the application

of novel methodologies and techniques from Psycho-

Informatics, turns out to be very important.

The main idea behind our work is to propose a

novel methodology based on the application of com-

munity detection algorithms from the Social Network

Analysis theory on a particular “social graph” that

considers users’ behaviors during the usage of their

mobile phone. In particular, we provide answers to

questions such as: How much smart-phone addicted

is a particular user? What is the app preferred by a

smart-phone addicted user?

To these aims, we designed and realized a frame-

work for monitoring and evaluating users’ behavior

with respect to the use of mobile devices for support-

ing smart-phone addiction diagnosis and assessment.

Successively, we have analyzed the collected data for

identifying possible social patterns that characterize

users with a high predisposition to smart phone addic-

tion, in conjunction with the analysis of well-known

self-assessment tests that are currently used by psy-

chologists to determine the onset of such pathology.

Eventually, an experimental evaluation was car-

ried out on a signiﬁcant sample of users (aged be-

tween 19 and 50 years) monitored for several week

in order to verify the reliability and effectiveness of

Cozzolino, F., Moscato, V., Picariello, A. and Sperli, G.

A Community Detection Approach for Smart-Phone Addiction Recognition.

DOI: 10.5220/0007839100530064

In Proceedings of the 8th International Conference on Data Science, Technology and Applications (DATA 2019), pages 53-64

ISBN: 978-989-758-377-3

the proposed approach in correctly recognizing and

calssifying the related addiction degree.

The paper is organized as in the following. Sec-

tion 2 provides a review of the Related Work on

smartphone addiction problem. In Section 3, we

describe the proposed methodology for evaluating

users’ smartphone addiction with respect to their mo-

bile devices. System architecture and experimental

results are then presented and discussed in Sections 4

and 5. Finally, conclusions are reported in Section 6.

2 RELATED WORK

The most diffused clinical methods – based on psy-

chometric questionnaires and interviews – for smart

phone-addiction prediction presents several limita-

tions that can be synthesized in the following points

(Markowetz et al., 2014): i) coarse temporal granular-

ity, ii) considerable costs, iii) distortion and poor ob-

jectivity of the available data, iv) impossibility of spe-

cialists to perform an ongoing patient’s assessments

and interventions, v) subjectivity of evaluations.

In more details, researchers actually rely on spe-

ciﬁc clinical experiments and self-assessed psycho-

metric tests to perform diagnosis of user’s mental ill-

ness related to the smart phone addiction pathology.

Even though these methods have found a widespread

application in research, they are not used yet in clin-

ical practice due to difﬁculty in managing the related

data and to the cost of obtaining and storing them.

Several works have recently offered solutions to

the smart phone addiction problem to overcome the

discussed limitations.

First of all, several correlations between the ex-

cessive usage of smart-phones and the Internet Ad-

diction (which literature is quite consolidated) have

been commonly observed in many studies (Ben-

Yehuda et al., 2016), even if a recent review (De-

Sola Guti

errez et al., 2016) discusses some peculiar

characteristics that clearly distinguish the two phe-

nomena. The authors have shown that smart-phone

addicted users are mostly young and female that seek

social gratiﬁcations, while Internet-dependent indi-

viduals are more likely to be males and socially in-

troverted.

The majority of works focusing on smart-phone

addiction proposed statistical approaches to correlate

smart-phone addiction to different mental problems.

(Bian and Leung, 2015) deﬁned a statistical model

that underlines how some smart-phone usage pat-

terns within a social context can be considered spe-

ciﬁc symptoms of smart-phone addiction. A different

perspective has been then analyzed in (Van Deursen

et al., 2015), where authors demonstrate as social

stress can inﬂuence a smart-phone addiction behav-

ior. In addition, (Samaha and Hawi, 2016) and (Sano

and Picard, 2013) discovered interesting relationships

among smart-phone addiction, level of stress and

school performances.

Concerning frameworks to support the smart-

phone addiction analysis, (Lee et al., 2014) realized a

system, namely SAMS (Smart-phone Addiction Man-

agement System and Veriﬁcation) able to perform

a statistical analysis of relationships between smart

phone apps and the possible levels of dependency.

Furthermore, the Smart-phone Overdependence Man-

agement System (SOMS) (Lee et al., 2016) has been

implemented to analyze user behavioral models that

can directly cause excessive dependence on smart

phones and also to prevent and to monitor excessive

smart-phone usage managing the assessment of pa-

tients. Finally, in (Lawanont and Inoue, 2017) it has

been designed an architecture for the recognition of

smart-phone addiction based on a classiﬁcation model

that analyses only some particular psychometric vari-

ables (such as average/minimum/maximum duration

of smart phone usage per unlock, number of apps’

context-switches, average duration of smart phone,

number of unlocks, number of reebots, etc.) acquired

by mobile devices.

Summing up, the results derived by the adoption

of SAMS and SOMS, an other very recent studies

(Lawanont and Inoue, 2017) showed strong correla-

tions between dependency on smart-phones and eval-

uation of the daily usage of these devices, both by

means acquisition and analysis of some speciﬁc psy-

chometric variables and statistics relating to the inter-

action between users and mobile applications. Table

1 summarizes the main prominent approaches.

3 METHODOLOGY

Here we propose a Multi-source Smart phone Addic-

tion Analysis (MSAA), a novel approach for recogniz-

ing users affected by the smart-phone addiction syn-

drome. We model the interactions between users and

mobile devices as a particular social graphs leverag-

ing community detection algorithms from Social Net-

work Analysis theory to infer useful social patterns

that describe different categories of social addiction

degree.

The MSAA approach is formed by four main

phases:

• the ﬁrst stage concerns acquisition and clean-

ing of data generated by the different actors

(users/participants and psychologists/supervisor);

DATA 2019 - 8th International Conference on Data Science, Technology and Applications

Table 1: Smart-phone addiction approaches.

Authors Outcome

(Bian and Leung, 2015)

Identiﬁcation of some smart phone usage patterns

within a social context as symptoms of smart phone

addiction symptoms.

(Van Deursen et al., 2015)

Analysis of social stress’s inﬂuence on

smart phone addiction behavior.

(Samaha and Hawi, 2016)

Identiﬁed relationships among smart phone

addiction, level of stress and school performances.

(Sano and Picard, 2013)

Use of wearable sensors and mobile phones

for stress recognition.

(Lee et al., 2014)

Identiﬁcation of relationships between smart phone

apps and the possible levels of dependency.

(Lee et al., 2016)

Analysis of user behavioral models that can

directly cause excessive dependence on smart phones.

(Lawanont and Inoue, 2017)

Analysis of psychometric variables for

smartphone addiction classiﬁcation.

• in the second phase, we model the gathered in-

formation through a graph data structure, namely

Initial Global Graph (IGG);

• the third phase produces an enriched version of

IGG, namely Final Global Graph (FGG), by com-

puting new edges based on the analysis of three

different types of relationships (between users,

between users and apps and between apps);

• the fourth and last stage performs a community

detection algorithm on FGG for classifying users

into four communities related to different levels of

smart-phone addiction combining self-assessment

test scores and the data obtained by the monitor-

ing phase.

3.1 Knowledge Base Building

Formally, we model users’ behavior while they are us-

ing their smart-phones as a directed a-cyclical graph,

namely Initial Global Graph (IGG).

Deﬁnition 3.1 (Initial Global Graph). An Initial

Global Graph is the pair IGG = (V, E), V being a

set of Vertices, composed by four entities: users,

supervisors, tracking days, apps; E being a set of

Edges, formed by three types of relationships: user-

to-supervisor, user-to-tracking day, tracking day-to-

app.

Table 2 describes in more details both entities and

relationships in the IGG graph.

The IGG has been implemented by means of a par-

ticular property graph, in which both nodes and edges

are particular Abstract Data Type (ADT). Supervi-

sor/clinical and user have both personal attributes

but on one hand a supervisor can also choose test’s

type, psychometric variables and weights for the Re-

cency Frequency Duration (RFD) analysis (Lee et al.,

2014); on other hand, user node has the obtained score

to the assessment test. Furthermore, several features

related to the number of locks, reboots, average, max-

imum and minimum usage for unlock and pedometer

values have been chosen for app nodes.

In Figure 1 an example of IGG is shown.

3.2 Knowledge Discovery Process

In this step, we perform an enrichment of IGG by

computing new edges between the existing nodes. We

leverage three different types of relationships: be-

tween users, between users and apps and between

apps. These relationships are generated by means

of a knowledge discovery process that aims to infer

several useful correlations between involved entities

from different points of view.

The user-to-user edges are focused on the dif-

ference between smart-phone addiction levels of two

users. More in details, we compute the Smart-

phone Addiction User Level (SAUL) for each user,

which corresponds to the sum of two terms: the self-

assessment test score (T S) and the average weighted

sum of the different psychometric variables related to

the smart-phone usage. The SAUL value is deﬁned as

in the following:

SAUL = T S +

∑

j=1

(

∑

T D

i=1

· PV

T D

) (1)

and N

T D

being respectively the number of psy-

chometric variables related to the smart-phone use

and the tracking days for each user, PV

representing

the j−psychometric variable related to i−th monitor-

ing day and w

corresponds to the weight assigned

A Community Detection Approach for Smart-Phone Addiction Recognition

Table 2: Initial Global Graph entities and their relationships.

Label Meaning

Nodes

Supervisor

The clinician/supervisor that monitors the various users.

She/he registers to the platform the self-assessment tests for users.

She/he selects weights to be attributed to the psychometric variables.

User

User/participant who executes the self-assessment test.

She/he is then monitored for a given period.

TrackingDay

Single day during which a user is monitored.

It contains the daily values of the psychometric variables for a given smartphone.

App

Single app used by a user during a tracking day.

It contains different daily usage data of the app.

Edges

Monitored

Relationship between supervisor and monitored user.

Produced

Relationship between the psychometric variables of a tracking day and users.

Relationship between data about used apps and tracking days.

Figure 1: Example of IGG graph at the end of the monitoring of 3 users. It is possible to note that the presence of a Supervisor

node (in green), 3 User nodes (in red), 7 Tracking Day nodes (in purple) for each user and several App nodes (in blue) for

each tracking day.

to of j−th psychometric variable. Successively, the

edge direction is assigned by comparing the SAUL co-

efﬁcients of user pairs, since it indicates a greater de-

pendence of the source user node with respect to the

destination one. Finally, we compute the relationship

weight as difference between the SAUL coefﬁcients

of the analyzed user pairs.

The second family of relations is composed by

user-to-app relationships that connect each user to the

related most used apps according to a RFD analysis

for each tracking day. In particular, the RFD analy-

sis is based on the following three parameters: i) Re-

cency (R) corresponds to the elapsed time since the

last use of the application by a user u within a certain

period T ; ii) Frequency (F) is the number of times a

user u has interacted with the application a within a

certain period T; iii) Duration (D) represents the total

duration of effective interaction with the application

a by a user u during the period T . This analysis aims

to provide an estimation about user’s preferences of a

given application. The RFD score is deﬁned as fol-

lows:

RFD = w

· R + w

· F +w

· D (2)

, w

being the assigned weights to each com-

ponent of RFD analysis based on its importance and

according to the application goals. Analyzing the

RFD value it is easy to understand how the applica-

tions have been used more recently, more frequently

and for longer times will probably be preferred by

users.

Finally, the app-to-app edges represent the usage

relationships between pairs of user apps in the same

DATA 2019 - 8th International Conference on Data Science, Technology and Applications

Figure 2: Example of FGG graph.

monitoring day. In particular, these relationships con-

sider several parameters, such as the usage chronolog-

ical order of apps in a single monitoring day, the dif-

ferences in terms of duration/frequency of usage be-

tween apps, the differences in terms of the quantity of

bytes transmitted/received via Internet connection be-

tween apps. The weight of each edge is computed as

the difference between the pairs of homologous fea-

tures related to two examined nodes.

In conclusion, we have an enriched version of

IGG, namely Final Global Graph (FGG), inferring

the discussed new edges for increasing information

necessary to smart-phone addiction analysis. We de-

ﬁne the FGG as in the following.

Deﬁnition 3.2 (Final Global Graph). The Final

Global Graph is the pair FGG = (V, E), V being a

set of Vertices composed by users, supervisors/clinics,

tracking days and apps; E being a set of Edges

composed by user-to-supervisor, user-to-tracking day,

tracking day-to-app, user-to-user, user-to-app and

app-to-app relationships.

Figure 2 shows an example of FGG, derived from

the IGG graph of Figure 1.

3.3 Smart-Phone Addiction Community

Detection Algorithm

The FGG can be seen as a sort of knowledge base for

supporting several applications. Here, we describe the

proposed approach for community detection over the

extracted FGG.

In our vision, the inherent semantic of user-to-

user relationships plays a key role for identifying user

nodes’ groups according to their smart-phone addic-

tion level. In addition, we also exploit the RFD anal-

ysis values between users and apps belonging to the

“communication” and “social” categories (i.e. What-

sapp, Facebook, Messenger, etc), because, as shown

in (Salehan and Negahban, 2013), they represent the

most useful applications for smart-phone addiction.

In particular, we deﬁne a Weighted Users Ma-

trix that jointly considers the two described contribu-

tions to identify groups of users suffering of the same

pathology.

Deﬁnition 3.3 (Weighted Users Matrix). A Weighted

User Matrix is the matrix:

Θ = {θ

i j

} =

(

(1 − ∆

SAUL

i j

) + (1 −

∑

a∈A

∆

RFD

i j|a

) i f i 6= j

0 i f i = j

A being a subset of apps in FGG, ∆

SAUL

i j

is the

difference of SAUL values between user i and user j

and ∆

RFD

i j|a

represents the sum of difference of RFD

analysis between two users i and j w.r.t. apps in A.

Following the idea discussed in (Gupta and Ku-

mar, 2016), we propose as community detection ap-

proach a vertex selection strategy that guarantees

high coverage and good conductance on expansion

of communities. However, we enhance the meth-

ods in (Gupta and Kumar, 2016) in according to: (i)

the data-model, that integrates both information about

users and their behaviors with respect to used smart

phone’s apps modeled by a property graph data struc-

ture; (ii) the comparison of users’ behavior with re-

spect to the apps relevant for smart-phone addictions;

(iii) a new way to build the user-to-user matrix com-

bining topological features and nodes’ attributes.

In the following, we report the algorithm exploited

for community detection.

More in details:

A Community Detection Approach for Smart-Phone Addiction Recognition

Algorithm 1: Community detection algorithm.

1: procedure SA Community Detection(FGG)

2: C ←

3: Compute Matrix θ

4: while more visited nodes do

5: C

6: u ← argmax

u∈U

{

∑

v∈U

}

7: C

← C

∪ {u}

8: while (φ(C

) − φ(

) ≥ 0) do

9: u ← argmax

u∈U

{

∑

v∈U

}

10: C

← C

∪ {u}

11: end while

12: C ← C ∪C

13: i ← i + 1

14: end while

15: return C

16: end procedure

• (lines 5-7) – the algorithm identiﬁes the nodes

showing the highest weight degrees as seed nodes.

The weighted degree of node u is computed as the

sum of column related to user u of the Weighted

User Matrix.

• (lines 9-11) – successively, the conductance mea-

sure has been used to evaluate the quality of

community during the expansion phase: in fact,

the increase of users in the examined commu-

nity corresponds to a decrease of conductance

value. The conductance is deﬁned as φ(C

) =

cut(C

)

min{deg(C

),deg(

)}

, where cut(C

) denotes the size

of a cut induced by C

is the complement set of

and deg(C

) is the sum of degrees of vertices in

• (lines 8-11) – once a seed node is identiﬁed, we

perform an incremental expansion of the commu-

nity for including the user that maximize the de-

creasing value of conductance. This is an iterative

process until the conductance difference related to

communities computed in successive steps does

not assume a negative value.

4 SYSTEM ARCHITECTURE

The system consists of a client-side application, to

be installed on the users’ mobile devices, and a web

server-side application, responsible for the data ac-

quisition and analysis through the methodological ap-

proach previously illustrated.

The entire system architecture together with the

adopted technologies are shown in Figure 3, that has

been deployed – at the moment – only for Android

Platforms..

More in details, the client-side consists of an An-

droid app: after the sign-up procedure, users can exe-

cute a self-assessment test whose typology is chosen

by clinicians. In addition, once the mobile device has

been set to start the smart-phone addiction recogni-

tion process, the user can access to a personal web

page on the server, containing the summary data con-

cerning personal test score and the current daily mon-

itoring statistics (both in terms of apps’ usage and of

interactions with the smart phone).

The Android app continuously monitors the run-

ning applications on the mobile device (Apps Usage

Statistics module), and also, the different user/smart-

phone interactions (Smart-phone Usage Statistics

module), locally storing the usage records (mainly

on a SQLite database but also through Shared Prefer-

ences mechanisms) by means of the Data Persistence

module. Furthermore, users can perform a real-time

self-monitoring of their smart phone usage level (Self-

Usage Check module) by viewing along the monitor-

ing period both the usage data related to the individual

apps and the entire interaction with the mobile device.

The data acquired by the mobile device using Android

apps are locally and daily stored and sent to the web

server at the end of the observation week of monitor-

ing (Statistic Server Upload module).

From the server side, the clinician performs the

registration operations to the system (Sign up), setting

also the type of test to be submitted to the users and all

the weights related to the psychometric variables. The

clinician can access to a personal web page contain-

ing the summary data (at various levels of detail) of

the users involved in the monitoring process. In par-

ticular, it is possible to view the information related

to the daily statistics and those resulting from several

analytics (Graph Analysis, RFD Analysis, Link Anal-

ysis and Graph Community Detection). Daily usage

records from the Android app at the end of the moni-

toring period are then stored stored into Neo4j graph

database. All the data processing and analytics fa-

cilities have been implemented using Scala functions

within the Apache Spark framework.

5 EXPERIMENTS

5.1 Dataset and Experimental

Environment

To test the reliability and the effectiveness of the de-

veloped system, several experiments were carried out

on a sample of about 50 users (aged between 12 and

77) monitored for a period of 7 days (from Monday

DATA 2019 - 8th International Conference on Data Science, Technology and Applications

Figure 3: System Architecture and Data Flow.

to Sunday).

Figure 4 shows the characterization of the users.

Note that we selected users in order to have the

same number of samples with respect to different age

groups (12-17, 18-27, 28-40, 40-60, 60-77). People

were randomly selected from a urban area, related

(relatives, friends) to students of our Lab, and no a-

priori knowledge was used except that they used ex-

tensively their smart phones for work or fun

. The

dataset is also composed by information about users’

behavior with respect to the use of mobile devices (i.e.

bytes transmitted/received via Internet or data con-

The participants provided informed consent (by parents

in case of minors) and all the assessed data were preserved

in a suitably protected database.

A Community Detection Approach for Smart-Phone Addiction Recognition

Figure 4: Dataset characterization.

nection between apps and so on). In this evaluation

we show how the use of our system offers interesting

perspectives, automatically detecting and classifying

behaviours and life-style of users.

The main characteristics of the used client and

server side hardware/software infrastructure for ex-

periments are summarized in the Table 3.

5.2 Experimental Protocol

The experimental protocol is composed by 4 main

stages.

The ﬁrst stage consists of the registration of the

clinician/supervisor to the system by creating a proper

account, together with the setting of the particular

type of test to be administered to the participants

(choosing between IAT, UADI, NMP-Q, MPACS and

SAS-SV) and the setting of the psychometric vari-

ables weights related to the RFD analysis and to the

calculation of the SAUL coefﬁcient of each user.

The second stage consists of explaining to the po-

tential participants the aims of the experiments: if one

chooses to participate, the Android application will be

installed on her/his smart phone, also registering the

necessary information within the system and the sub-

sequent execution on the app of the speciﬁc assess-

ment test of the level of smart-phone addiction. The

application will compute the test score and send it to

the server.

In the third stage, all the subjects – that have com-

pleted the test – can start the weekly monitoring of

their devices. Monitoring acts as a background ser-

vice allowing subjects to close the application and use

their smart phone normally.

In the fourth stage, at the end of the monitoring

period, the clinician/supervisor uses the data related

to all users who have completed the monitoring and,

through appropriate interface, obtains different types

of statistics (at various levels of granularity) as well

as the result of the community detection algorithm.

The aim of the provided evaluation is to:

• detect the kind of applications that exerts more in-

ﬂuence on users and may represent a possible fea-

ture for predicting smart-phone addiction;

• detect the users’ categories that shows a possible

correlation with smart-phone dependence;

• analyze the usage patterns for better discriminate

addicted vs not addicted users;

• compare the outcome of our proposed technique

with respect to surveys methods.

5.3 Popularity and Category

Applications Analysis

First of all, we have conducted a popularity analysis

of the applications w.r.t. the weekly usage of users’

smart-phones . The top 10 ranked applications (out

of a total of 124) on the basis of the duration and fre-

quency of their average daily use can be seen in Table

The contrast between duration and frequency of

use for each application is due to the related category.

WhatsApp, Messenger, Instagram and Snapchat, be-

longing to the Social/Communication category, have a

greater tendency to be used more frequently, with less

usage time. Other categories such as Game and Me-

dia & Video, to which applications such as YouTube

or the FarmVille game belong, for example, are used

less frequently: however, once a user runs these ap-

plications, the usage time is extended.

According to the study of (Salehan and Negahban,

2013; Lee et al., 2014), which shows a correlation be-

tween the use of applications belonging to the social

DATA 2019 - 8th International Conference on Data Science, Technology and Applications

Table 3: Hardware/software infrastructure.

Client-side

Hardware

Category Smartphones

Manufacturer and/or Samsung, Huawei, Honor, Oppo, Xiaomi,

model Motorola, Vivo, HTC, Meizu, Cubot

CPU da 1.9 GHz Quad Core

a Quad core 2.3 GHz + Quad core 1.7 GHz

RAM 2-4 GB

Storage da 16 GB (only internal storage)

a 128 GB (with micro SD extention)

Software

O.S. Android 6.0 (Marshmallow) +

Database SQLite

Framework Android Volley

Libraries Android MPAndroidChart

Server-side

Hardware

Category Ultrabook

Manufacturer and/or Dell XPS 14

model

CPU Intel i7-3667U Dual-core 2.00 GHz, 2.50 GHz

RAM 8 GB

Storage 500 GB (SSD) + 32 GB (SSD)

Software

O.S. Windows 7 64 bit

Database Neo4j

Framework Apache Spark, Apache Tomcat

Libraries GraphX

category and the smartphone dependence, in our case

there is a high percentage of daily use (both in terms

of frequency and duration) of this category highlight-

ing a potential presence of a smart-phone addiction.

5.4 Smart-Phone Addiction Detection

via Community Detection

The goal of such experiments is to compare the results

proposed community detection algorithm (based on

the combination of SAUL values from different tests

and psychometric variables) in distinguishing smart-

phone addicted (S.A.) users from not smart-phone ad-

dicted (No S.A.) with respect t the outcomes provided

by the MPAC test.

Figure 5 summarizes the obtained results.

The achievement of a lower percentage of S.A.

users compared to that produced only by the MPAC

test follows the results of previous studies (Mon-

tag et al., 2015a; Lin et al., 2015; Boase and Ling,

2013) which showed how the total weekly use of

the smart phone is overestimated by the participants

which are not very reliable in providing an effec-

tive estimate of their interaction with the mobile de-

vice (both in emotional terms through the answers

to the items/questions of the self-assessment test and

in terms of quantitative estimation of the number of

weekly hours used with the device as answer to fur-

ther questions that are integral to those of the test).

We want to note as the users classiﬁed as S.A. by

our algorithm have shown, as a result of the RFD anal-

ysis, a clear preference in the use of applications ac-

cording to the ordering: 1) Whatsapp, 2) Facebook

and 3) Youtube.

In addition, a further survey concerning these

users showed that this preference was also found in

the top 10 used apps in terms of the usage frequency

only. In turn, for what concerns only the duration, the

resulting ranking showed how users show a prefer-

ence in the order: 1) Youtube, 2) Whatsapp, 3) Face-

book.

These results conﬁrmed the ﬁndings of previ-

ous studies (Montag et al., 2015b; Olivencia-Carri

et al., 2016) which identify the WhatsApp application

as one of the driving forces behind the use of smart

phones, attributing to its overuse a high potential cor-

relation with smart phone dependency.

Eventually, it should be noted that the majority of

S.A. users belong to the group 18-27, conﬁrming the

trend that sees the phenomenon of smart- phone ad-

diction is growing among the youth population.

6 CONCLUSIONS

In this paper we introduced a novel methodology for

smart-phone addiction classiﬁcation based on the ap-

plication of community detection algorithms from the

SNA theory.

In particular, we:

A Community Detection Approach for Smart-Phone Addiction Recognition

Table 4: Ranking of the top 10 applications based on the duration and frequency of their average daily use.

Rank Sorting by usage duration Sorting by usage frequency

App name Frequency (%) Duration(%) App name Frequency (%) Duration(%)

1 Whatsapp 18.86 17.48 Whatsapp 18.86 17.48

2 YouTube 10.02 8.69 YouTube 10.02 8.69

Android

0.28 5.52 Facebook 7.06 4.97

browser

4 Facebook 7.06 4.97 Messenger 5.22 1.19

Amazon

0.48 4.48 Instagram 5.07 1.27

shopping

6 Messenger 3.28 4.34 Snapchat 3.45 2.09

Amazon

1.03 3.92 AliExpress 3.28 4.34

Kindle

Zynga

0.76 3.71 Fifa calcio 3.12 1.48

Poker

Google

1.27 3.00

Amazon

3.04 0.37

Maps shopping

10 AliExpress 3.45 2.09

Google

2.89 0.87

Maps

Figure 5: Results of the comparison between the score of the MPACS test and the outcome of the proposed community

detection algorithm.

• modeled data related to users’ behavior with re-

spect to the use of mobile devices as a particular

social graph;

• discovered by means of the SNA algorithms pat-

terns that better characterize users with a high pre-

disposition to smart phone addiction;

• designed and realized a system supporting smart-

phone addiction analysis;

DATA 2019 - 8th International Conference on Data Science, Technology and Applications

• performed several experiments on a sample of

users to verify the reliability and effectiveness of

the proposed approach in correctly recognizing

the related addiction degree.

We think that the empirical study reported in this

paper represents an important starting point to illus-

trate the advantages of the inclusion of Social Net-

work Analysis tools and methodologies in the psy-

chological/psychiatric ﬁeld.

The combination of self-report data and actual be-

havioral monitoring provides a clearer picture of a pa-

tient, as well as a more in-depth view of his potential

dependency status, useful to psychiatric doctors. Fu-

ture work will be devoted to extend experimentation

increasing the number of human subjects and com-

paring our approaches with different and more recent

ones.

ACKNOWLEDGEMENT

This work is part of the Synergy-net: Research and

Digital Solutions against Cancer project (funded in

the framework of the POR Campania FESR 2014-

2020).

REFERENCES

Ben-Yehuda, L., Greenberg, L., and Weinstein, A.

(2016). Internet addiction by using the smartphone-

relationships between internet addiction, frequency of

smartphone use and the state of mind of male and

female students. Journal of Reward Deﬁciency Syn-

drome & Addiction Science, 2(1):22–27.

Bian, M. and Leung, L. (2015). Linking loneliness, shy-

ness, smartphone addiction symptoms, and patterns of

smartphone use to social capital. Social Science Com-

puter Review, 33(1):61–79.

Boase, J. and Ling, R. (2013). Measuring mobile phone

use: Self-report versus log data. Journal of Computer-

Mediated Communication, 18(4):508–519.

oliz, M., Pinto, L., Phansalkar, S. S., Corr, E., Mujjahid,

A., Flores, C., and Barrientos, P. E. (2016). Devel-

opment of a brief multicultural version of the test of

mobile phone dependence (tmdbrief) questionnaire.

Frontiers in psychology, 7:650.

De-Sola, J., Talledo, H., Rubio, G., and de Fonseca, F. R.

(2017). Development of a mobile phone addiction

craving scale and its validation in a spanish adult pop-

ulation. Frontiers in psychiatry, 8:90.

De-Sola Guti

errez, J., Rodr

ıguez de Fonseca, F., and Rubio,

G. (2016). Cell-phone addiction: a review. Frontiers

in psychiatry, 7:175.

Gupta, S. and Kumar, P. (2016). Community detection in

heterogenous networks using incremental seed expan-

sion. In Data Science and Engineering (ICDSE), 2016

International Conference on, pages 1–5. IEEE.

Ha, J. H., Chin, B., Park, D.-H., Ryu, S.-H., and Yu, J.

(2008). Characteristics of excessive cellular phone use

in korean adolescents. CyberPsychology & Behavior,

11(6):783–784.

Hooper, V. and Zhou, Y. (2007). Addictive, dependent,

compulsive? a study of mobile phone usage. BLED

2007 Proceedings, page 38.

Lawanont, W. and Inoue, M. (2017). A development of clas-

siﬁcation model for smartphone addiction recognition

system based on smartphone usage data. In Interna-

tional Conference on Intelligent Decision Technolo-

gies, pages 3–12. Springer.

Lee, H., Ahn, H., Choi, S., and Choi, W. (2014). The sams:

Smartphone addiction management system and veriﬁ-

cation. Journal of medical systems, 38(1):1.

Lee, S.-J., Rho, M. J., Yook, I. H., Park, S.-H., Jang, K.-S.,

Park, B.-J., Lee, O., Lee, D. K., Kim, D.-J., and Choi,

I. Y. (2016). Design, development and implementation

of a smartphone overdependence management system

for the self-control of smart devices. Applied Sciences,

6(12):440.

Leung, L. (2008). Leisure boredom, sensation seeking, self-

esteem, and addiction. Mediated interpersonal com-

munication, 1:359–381.

Lin, Y.-H., Lin, Y.-C., Lee, Y.-H., Lin, P.-H., Lin, S.-H.,

Chang, L.-R., Tseng, H.-W., Yen, L.-Y., Yang, C. C.,

and Kuo, T. B. (2015). Time distortion associated with

smartphone addiction: Identifying smartphone addic-

tion via a mobile application (app). Journal of psychi-

atric research, 65:139–145.

Markowetz, A., Błaszkiewicz, K., Montag, C., Switala, C.,

and Schlaepfer, T. E. (2014). Psycho-informatics:

big data shaping modern psychometrics. Medical hy-

potheses, 82(4):405–411.

Montag, C., Błaszkiewicz, K., Lachmann, B., Sariyska,

R., Andone, I., Trendaﬁlov, B., and Markowetz, A.

(2015a). Recorded behavior as a valuable resource for

diagnostics in mobile phone addiction: evidence from

psychoinformatics. Behavioral Sciences, 5(4):434–

442.

Montag, C., Błaszkiewicz, K., Sariyska, R., Lachmann,

B., Andone, I., Trendaﬁlov, B., Eibes, M., and

Markowetz, A. (2015b). Smartphone usage in the 21st

century: who is active on whatsapp? BMC research

notes, 8(1):331.

Olivencia-Carri

on, M., P

erez-Marﬁl, M., Ramos-Revelles,

M., and L

opez-Torrecillas, F. (2016). Personalidad y

su relaci

on con el uso versus abuso del tel

efono m

ovil.

Acci

on Psicol

ogica, 13(1):109–118.

Salehan, M. and Negahban, A. (2013). Social networking

on smartphones: When mobile phones become ad-

dictive. Computers in Human Behavior, 29(6):2632–

2639.

Samaha, M. and Hawi, N. S. (2016). Relationships among

smartphone addiction, stress, academic performance,

A Community Detection Approach for Smart-Phone Addiction Recognition

and satisfaction with life. Computers in Human Be-

havior, 57:321–325.

Sano, A. and Picard, R. W. (2013). Stress recognition us-

ing wearable sensors and mobile phones. In Affective

Computing and Intelligent Interaction (ACII), 2013

Humaine Association Conference on, pages 671–676.

IEEE.

Van Deursen, A. J., Bolle, C. L., Hegner, S. M., and Kom-

mers, P. A. (2015). Modeling habitual and addic-

tive smartphone behavior: The role of smartphone us-

age types, emotional intelligence, social stress, self-

regulation, age, and gender. Computers in human be-

havior, 45:411–420.

DATA 2019 - 8th International Conference on Data Science, Technology and Applications