SocialCount

Detecting Social Interactions on Mobile Devices

Isadora Vasconcellos e Souza

, Jo

ao Carlos Damasceno Lima

1,2

, Benhur de Oliveira Stein

1,2

and Cristiano Cortez da Rocha

Programa de P

os-Graduac¸

ao em Inform

atica, Universidade Federal de Santa Maria (UFSM), Santa Maria, RS, Brazil

Departamento de Linguagens e Sistemas de Computac¸

ao, Universidade Federal de Santa Maria (UFSM),

Santa Maria, Brazil

Centro de Inform

atica e Automac¸

ao do Estado de Santa Catarina (CIASC), Florian

opolis, SC, Brazil

Keywords:

Social Interactions, Context-aware, Mobile Applications.

Abstract:

With mobile devices increasingly powerful and accessible to the majority of the population, applications have

begun to become increasingly intelligent, customizable and adaptable to users’ needs. To do this, context-

aware applications are developed. In this work, we create an approach to infer social interactions through the

identiﬁcation of the user’s voice and to recognize their social context. Data from the social context of the user

has been useful in many real-life situations, such as identifying and controlling infectious disease epidemics.

1 INTRODUCTION

In recent years, mobile devices have started to be part

of everyday life for most people. In the beginning

users acquired the devices just to make calls, but over

the years it has become a tool to optimize and help

users in various types of tasks. In this way, the appli-

cations had the need to obtain data of human factors

such as activity, social interactions, health, etc. to cu-

stomize processes. Thus, mobile computing began to

develop context-sensitive applications.

Context-aware systems offer entirely new oppor-

tunities for application developers and end-users, gat-

hering context data and adapting the behavior of sy-

stems accordingly. Especially in combination with

mobile devices, these mechanisms are of high value

and are used to greatly increase usability. (Baldauf

et al., 2007)

Mobile devices are used to interact with other

users through calls, messages or social networks. In

addition to being mostly close to the user. As such,

they are a great tool for detecting social interactions,

both face-to-face and virtual. The SocialCount ap-

plication proposed in this paper makes the inference

of social interactions face-to-face. These data can

be used in several areas, such as: marketing (word-

of-mouth), business (enterprise community detection)

and health (infectious disease control).

In section 2 we present the concepts of context-

aware computing, social-aware computing, ubiqui-

tous computing, social context and social interaction

and how these concepts relate. Section 3 discusses re-

lated works. Section 4 presents the SocialCount met-

hodology. Section 5 presents a case of use of infecti-

ous disease control with the use of social interaction

data. Finally, in section 6 is the conclusion.

2 CONTEXT-AWARE

COMPUTING

The Context-aware Computing emerged to address

the challenges of mobile computing, where applicati-

ons have begun to explore the changing environment

in which they are run (Schilit et al., 1994). It is desira-

ble that mobile device applications and services react

to their current location, time, and other attributes of

the environment, and adapt their behavior according

to changing circumstances as context data can change

rapidly (Baldauf et al., 2007). Such context-aware

software adapts their functions, contents, and interfa-

ces according to the user’s current situation with less

distraction of the users. Thus, aim at increasing usabi-

lity and effectiveness by taking environmental context

into account (Temdee and Prasad, 2017).

The word ”context” is very comprehensive, it is

possible to ﬁnd different deﬁnitions proposed by dif-

Vasconcellos e Souza, I., Damasceno Lima, J., de Oliveira Stein, B. and Cortez da Rocha, C.

SocialCount.

DOI: 10.5220/0006696605110518

In Proceedings of the 20th International Conference on Enterprise Information Systems (ICEIS 2018), pages 511-518

ISBN: 978-989-758-298-1

511

ferent authors. ”Context is any information that can

be used to characterize the situation of an entity. An

entity is a person, place, or object that is considered

relevant to the interaction between an user and an ap-

plication, including the user and applications themsel-

ves” (Dey, 2001). Three important aspects of context

are: where you are, who you are with, and what re-

sources are nearby. This includes lighting, noise level,

network connectivity, communication costs, commu-

nication bandwidth and even social situation (Schilit

et al., 1994).

The context can be divided into ﬁve sub-

categories: Environmental context, Personal context,

Social context, Task context and Spatio-temporal con-

text (Kofod-Petersen and Mikalsen, 2005). These

sub-categories are deﬁned as (Cassens and Kofod-

Petersen, 2006):

• Environmental context: This part captures the

user’s surrounding, such as things, services, pe-

ople, and information accessed by the user.

• Personal context: This part describes the mental

and physical information about the user, such as

mood, expertise and disabilities.

• Social context: This describes the social aspects

of the user, such as information about the different

roles an user can assume.

• Task context: the task context describes what the

user is doing, it can describe the user’s goals, tasks

and activities.

• Spatio-temporal context: This type of context is

concerned with attributes like: time, location and

the community present.

One of the key features of a context-aware appli-

cation is not to be intrusive. In other words, the ap-

plication needs to detect user context information wit-

hout any kind of intervention that can change the cur-

rent context. If the user is working and the application

alerts him to make a decision or provides some infor-

mation, the user’s status can be changed from ”wor-

king” to ”using mobile”, in this way the application

changes the user’s current task context. Therefore,

context-aware computing uses the notions of ubiqui-

tous computing.

Ubiquitous computing can also be called syno-

nymously the generalized computation (Temdee and

Prasad, 2017). However, there are some differences.

The concept of pervasive computing implies that the

computer is embedded in the environment invisibly

to the user. The computer has the ability to obtain in-

formation from the environment in which it is ship-

ped and use it to dynamically build computer mo-

dels, such as controlling, conﬁguring and tuning the

application to better meet the needs of the device or

user. Ubiquitous computing comes from the need to

integrate mobility with the functionality of pervasive

computing. In other words, any moving computing

device can dynamically construct computational mo-

dels of the environments in which we move and conﬁ-

gure their services depending on the need (de Araujo,

2003).

Since the beginning of social networks, the rese-

archers are using the sensing data to understand hu-

man behavior, mobility, and activity, and ultimately

helping to solve social problems (Yu et al., 2012).

For this reason, context-aware computing has ge-

nerated a new emerging research topic in computer

science called social-aware. ”Humans, however, are

social beings. Hence, the notion of social context-

awareness (in short social awareness) extends the vi-

sion of context-aware computing” (Kabir et al., 2014).

Socially aware applications are services that ex-

ploiting any information that describes the social con-

text of the user, like social relations, social interacti-

ons, or social situations and embody the ability to

trace and model ongoing social processes, structures,

and behavioral patterns (Ferscha, 2012).

Social context is deﬁned as a set of information

derived from direct or indirect interactions among pe-

ople in both virtual and physical world. Direct in-

teraction contains face to face conversation, video

conferencing, etc.. Indirect interaction includes co-

locating for a period of time, joining the same event,

etc. (Kabir et al., 2014). The SocialCount recognizes

the user’s social interactions to get information about

the user’s social context.

According to the sociological approach social

context is the way that people can relate easily, in-

cluding the culture in which the individual lives and

has been educated and the people and institutions with

whom he interacts (Kolvenbach et al., 2004) (Carter,

2013). Some authors add places and activities to the

concept (Adams et al., 2008). Therefore, according

to the deﬁnition of Schilit et al., the social context ad-

dresses two of the main aspects of the context: where

are you and who you are with.

Biamino proposed a more speciﬁc interpretation

that would meet the point of view of pervasive and

ubiquitous computing. They suggest that the social

context can be represented through networks. ”In our

vision social contexts are more similar to social ag-

gregations or social groups, identiﬁed as a number

of nodes in a given location, linked by some kind of

ties (relations) that determine their nature” (Biamino,

2011). The authors deﬁne a social context as a 3-tuple

that, describe the network:

ICEIS 2018 - 20th International Conference on Enterprise Information Systems

512

cxt =< Size, Density, Typeo f Ties > (1)

In this paper the deﬁnitions of Size, Density and

Type of Ties were adapted from the work of Biamino.

The Size represents the number of nodes in a deﬁned

location:

• Small (n ≤ 5): a network with a small number of

nodes.

• Private (5 < n ≤ 20): a network with a few nodes;

• Open (20 < n ≤ 50): a relatively large network;

• Wide (n > 50): a network with a very large num-

ber of nodes;

Density represents the number of connections be-

tween the nodes:

• Clique: a fully connected graph;

• Easy: a graph with easy to close triangles;

• Hard: a graph with many isolated nodes and hard

to close triangles;

Type of Ties is deﬁned by the main type of relati-

ons between the nodes of the network:

• Unknown: no relation exists between two nodes;

• Acquaintance: two nodes are not close friends,

but they interact with each other;

• Friends: two nodes with a friendship-kind of rela-

tion;

SocialCount uses data collected from social in-

teractions to generate social graphs and classify the

context criteria according to 3-tuple. In the next

section are presented and discussed related works that

also collect social interactions.

3 RELATED WORK AND

DISCUSSION

The related works were selected with the purpose of

presenting the state of the art and the most used met-

hodologies of detection of social interactions in mo-

bile devices. The main characteristics raised for de-

tecting interactions are: interpersonal distance, user

location, relative position and conversation activity.

The most common approach used by researchers

to recognize social interactions between two indivi-

duals is the Bluetooth ID search (BTID) or Wi-Fi ser-

vice ID (SSID) of nearby devices. All devices/people

found are classiﬁed as social interactions. This met-

hod was used in CenceMe (Miluzzo et al., 2007),

SoundSense (Lu et al., 2009), E-Shadow (Teng et al.,

2014), PMSN (Zhang et al., 2012), among others.

This approach is simple and does not require speciali-

zed hardware and sensors, but the accuracy is limited

by the range of Bluetooth (about 10 meters) and Wi-Fi

(approximately 35 meters for indoor environments).

DARSIS (Palaghias et al., 2015) was developed

to quantify social interactions in real time. The rela-

tive orientation of the users was used to obtain face

direction and interpersonal distance. The proximity

between the participants of the interaction is calcu-

lated through the RSSI samples of the user device’s

Bluetooth that are trained by learning machine (Multi-

BoostAB with J48). Samples are taken from three de-

vice position combinations: screen to screen, screen

to back and back to back. The proximity is classi-

ﬁed as public area, social zone, personal zone and

intimate zone. The relative orientation of the user

is known through uDirect (Hoseinitabatabaei et al.,

2014) which identiﬁes the relative orientation bet-

ween the Earth’s coordinates and the user’s locomo-

tion and predicts the direction of the face without re-

quiring a ﬁxed position of the device.

In Multi-modal Mobile Sensing of Social Inte-

ractions (Matic et al., 2012) they used a set of met-

hods for the sensing of the interactions: the interper-

sonal distance, the relative position of the user, the

direction of the face and the veriﬁcation of speech

activity. The interpersonal distance between the de-

vices is captured in RSSI, where one device works as

a Wi-Fi access point (Hot Spot) and another as a Wi-

Fi client. The relative position is calculated by the

position of the torso in relation to the coordinates of

the Earth, always considering the same position of the

mobile device. The speech activity is detected by an

accelerometer installed on the user’s chest. The de-

vice conﬁgured as a Wi-Fi access point is characteri-

zed as an intrusive process because the user generally

does not use his mobile device for this purpose. Like-

wise the accelerometer on the chest because is not a

commonly used device.

SCAN (Social Contex-Aware smartphone Notiﬁ-

cation system) (Kim and Lee, 2017) detects the user’s

social context and blocks smartphone notiﬁcations so

as not to distract the user while he or she is inte-

racting. The system sets breakpoints to release no-

tiﬁcations according to the following criteria: silence

(when there is no conversation for 5 seconds or more),

movement (when a person in the group leaves the ta-

ble), user alone (when the person is alone waiting for

friends) and use (when the other person participating

in the interaction is using the smartphone). Social in-

teractions are known through identiﬁcation of close

people and conversation. SCAN periodically searches

for BLE beacons to detect the presence of other pe-

ople and announce their own presence, the BLE bea-

SocialCount

513

Table 1: Comparative table of relation works and SocialCount.

Intrusive

Detection from

speech sound

Detection from

who is speaking

Approach of interaction detection

CenceMe,

SoundSense,

E-Shadow,

PMSN

No No No

Interpersonal distance

(ID Bluetooth e Wi-Fi)

DARSIS No No No

Interpersonal distance (RSSI Bluetooth),

user relative orientation and face direction

Multi-model Yes Yes No

Interpersonal distance (RSSI Bluetooth),

user relative orientation and

detection from speech sound

SCAN No Yes No

Interpersonal distance (BLE beacons) and

detection from speech sound

SocioGlass Yes No No Image detection with Google Glass

SocialCount No Yes Yes

Interpersonal distance (ID Bluetooth),

detection from speech sound and

detection from who is speaking

cons have been chosen for deployment because they

do not require pairing and connection actions, and

have a low power consumption. For the detection of

conversation was used the algorithm YIN (De Che-

veign

e and Kawahara, 2002) that estimates the fun-

damental speech frequency and identiﬁes the human

voice.

SocioGlass (Xu et al., 2016) to promote additi-

onal information about the people who the user is

interacting. There are 28 biographical information

items that are classiﬁed into 6 groups: work, perso-

nal, education, social, leisure and family. The system

uses Google Glass and an Android application that

communicate via Bluetooth. Interactions are detected

through facial recognition, Google Glass is responsi-

ble for providing the image of the individual who is

participating in the interaction, the application recei-

ves the image, performs the processing and searches

for a combination in the local database. When you do

the recognition, the information related to the person

in question is displayed on the Google Glass screen.

The authors also implemented a smartphone-only ver-

sion, where the face image is captured by the device’s

camera and the information is displayed on the mobile

screen.

To analyze the related works, we insert them into a

bus stop scenario. In this case, there are many people

waiting for buses, some people are talking, others are

quiet looking toward the cars. The Table 1 presents a

comparative board between related works and Social-

Count.

The works that only approach the interpersonal

distance to consider an interaction are submitted to

have a low accuracy, mainly in situations of the real

life. According to the scenario, many people are

physically close and do not interact with each other.

In this case several interactions would be considered

wrongly. The good thing about this methodology is

that Bluetooth and Wi-Fi technology are compatible

with most smartphones available in the market.

DARSIS obtained a good accuracy in verifying in-

terpersonal distance. In addition, they used the di-

rection of the face to identify the interactions, which

is a good approach considering that a person tends to

direct the face to the other person when communica-

ting. However, they do not consider the conversation.

Considering the scenario, this can easily lead to mis-

takes, where two people may be standing next to each

other and looking in opposite directions.

Multi-modal and SCAN consider distance and

conversation. The Multi-modal uses an intrusive ap-

proach to the veriﬁcation of the conversation, which

damages the naturalness of the user’s daily actions.

SCAN uses the YIN algorithm for the same purpose,

which identiﬁes the human voice without the need for

external resources to the smartphone. There may still

be errors in the scene, as people around the user may

be talking.

SocioGlass used images captured with Google

Glass. This can be considered an intrusive procedure,

since few users own the device and use it regularly.

The authors have made a version that works only on

the smartphone, but users need to focus the camera on

the person’s face, which detracts from the usability of

the application.

To solve the problems mentioned in the scenario,

SocialCount uses a set of approaches. Interpersonal

distance is implemented through Bluetooth, the de-

tection of the conversation by the YIN algorithm and

the detection from who is speaking. This last que-

ICEIS 2018 - 20th International Conference on Enterprise Information Systems

514

stion is fundamental for the correct consideration of

interactions. In the next section the SocialCount met-

hodology is described in detail.

4 SocialCount

SocialCount is a mobile application developed for the

Android platform. Its main purpose is to detect the

user’s social interactions and provide data that descri-

bes the social context. The interactions considered by

the application are only face-to-face, that is, interacti-

ons mediated by a means of communication are not

considered.

For inference of social interactions, SocialCount

detects: human voice, who is speaking, location, and

nearby devices. Based on related work, the differen-

tial of SocialCount is the use of the recognition of

who is speaking for the inference of social interacti-

ons. During the conversation, the application checks

whether the speaker is the user or someone near him.

The recognition was developed with widely known

methodologies, and the focus of this work is not to

elaborate a new method.

Figure 1 shows how SocialCount works. The ap-

plication is responsible for recognizing human voice

in the environment, recording audio, locating nearby

people and identifying the current location. The ser-

ver recognizes who is speaking and stores the data in

the database.

The application remains listening for the presence

of human voice in the environment. When Social-

Count detects voice, it records about 8 seconds and

sends the audio to the server. The YIN (De Chev-

eign

e and Kawahara, 2002) algorithm is used to iden-

tify human voice in the environment. The algorithm is

developed by the TarsosDSP (Six et al., 2014) library,

which performs real-time audio processing.

The server veriﬁes that the user is participating

in the conversation through voice prints previously

stored in the database. Voice prints are a set of au-

dios that contain speech frequency. To record the au-

dios were elaborated phrases that contained the con-

sonant phonemes of the native language in several vo-

wel contexts.

The veriﬁcation is developed with the Recognito

(Crickx, 2014), which is a library that performs text-

independent recognition of speakers in Java. The li-

brary generates an universal template with all sto-

red voice prints. Each audio input for checking

who is speaking Recognito computes the relative dis-

tance using the variables: identiﬁed voice print (VPI),

unknown voice print (VPU), and universal (UM) mo-

del. VPI represents the voice prints that have already

been identiﬁed, VPU is the audio input and UM is

the universal model that represents the average of all

stored voice prints. ”If you put them on a line, you

can calculate the distance between IVP/UVP and the

distance between UVP/UM. Based on those numbers,

you can tell how relatively close the unknown voice

print is to the identiﬁed one. The UM acts as a max

distance value” (Crickx, 2014).

SocialCount can classify an interaction in two

ways: participation and monitoring. In participation,

the server identiﬁes who is speaking is the user, so he

is participating in the interaction. In monitoring, the

server identiﬁes who is talking is not the user, it is so-

meone who was close to the user’s device. Therefore,

the user may not be participating in the interaction.

Data from interactions stored as monitoring are only

used to increase the accuracy of the inference of the

interactions stored as participation.

After ﬁnd who is talking, the server sends a re-

sponse to the application. The application then sear-

ches for nearby devices. The Bluetooth ID of nearby

devices is used as additional information to increase

the inference accuracy of users who are participating

in the interaction. If the user chooses not to turn Blu-

etooth on to save battery, SocialCount continues to

function normally.

Finally, SocialCount detects the location of the in-

teraction and sends all the data to the server to store.

The stored data are used to generate social graphs. A

social graph can represent the interactions between

users of SocialCount at a particular location over a

period of time or the interactions performed by a par-

ticular user. The Figure 2 presents social graphs in

two different environments: a research lab and a re-

staurant.

To classify the Type of Ties (ToT), we use the

equation proposed by Palaghias (2016) to calculate

the conﬁdence of each social relation between a pair

of users and the average of interactions performed:

P(r) =

Q(r)

(2)

where Q(r) is the number of inferences of inte-

ractions that are related to social relation r and N is

the total number of social interactions inferences. In

order to adapt the classiﬁcation according to the social

characteristics of each user, we calculated the average

number of interactions performed per node according

to the equation:

M(n) =

T · N

(3)

where T is the total number of nodes that have

interacted with the current node n. Then we classify

ToT as follows:

SocialCount

515

Figure 1: SocialCount ﬂowchart.

Figure 2: Examples of social graphs: (a) Social graph generated in a research lab, (b) Social graph generated in a university

restaurant.

ToT =







P(r) = 0, ToT = Unknown

P(r) < M(n), ToT = Acquaintance

P(r) ≥ M(n), ToT = Friends

(4)

The social graph of the research laboratory pre-

sented in the Figure 2 is deﬁned by:

Researchlab =< Private, Easy, Friends >.

It is private because it has 9 nodes, Easy because

it is possible to close triangles easily and Friends be-

cause most social relations have P(r) ≥ M(n). And

the graph of University restaurant is deﬁned by:

Restaurant =< Private, Hard, Acquaintance >.

It is private because it has 14 nodes, Hard because

it is difﬁcult to close triangles and Acquaintance be-

cause most social relations have P(r) < M(n).

5 USE CASE

The purpose of this section is to better address the be-

neﬁts of bringing social awareness to mobile devices.

We present a brief use case in which we describe how

the inference of social interactions can help in real life

events such as in combating the spread of infectious

diseases.

Human contact is the most important factor in

the transmission of infectious diseases (Clayton and

Hills, 1993). Many diseases spread to human popula-

tions through contact between infectious individuals

(people carrying the disease) and susceptible indivi-

duals (people who do not yet have the disease, but can

get it) (Newman, 2002). These contacts generate net-

works called contact networks, which are networks of

interaction through which diseases spread and deter-

mine whether and when individuals become infected

ICEIS 2018 - 20th International Conference on Enterprise Information Systems

516

and thus who can serve as an early and accurate sur-

veillance sensor (Herrera et al., 2016).

Epidemiological models are based on the SIR mo-

del. A SIR model computes the theoretical number

of people infected with a contagious illness in a clo-

sed population over time. The name of this class of

models derives from the fact that they involve cou-

pled equations relating the number of susceptible pe-

ople S(t), number of people infected I(t), and number

of people who have recovered R(t) (Weisstein, 2017).

Based on this model, it is possible to determine if an

epidemic is increasing or decreasing.

S + I + R = N (5)

Traditionally, public health is monitored through

research and the aggregation of statistics obtained

from health care providers. These methods are expen-

sive, slow, and can be biased. An infected person is

only recognized after a doctor sends the necessary in-

formation to the appropriate health agency. Affected

people who do not seek treatment, or do not respond

to research, are virtually invisible to traditional met-

hods.

Social graphs can act as networks of contacts to

determine who are susceptible people (S) through the

people with whom a contaminated user interacted. Fi-

gure 3 presents the social graph of ”user-01” based on

the contexts of Figure 2.

Figure 3: Social graph of ”user-01”.

If user-01 is the only one infected, 6 people may

be susceptible to illness. The network is recursi-

vely extended for each infected user. In this way,

users can be quickly found, informed and kept awake.

Early identiﬁcation of infected individuals is crucial

in the prevention and containment of outbreaks of de-

vastating diseases. The most effective way to ﬁght

an epidemic in urban areas is to quickly conﬁne in-

fected individuals to their homes. The agility of vac-

cination ranks second in effectiveness (Eubank et al.,

2004). Information about people’s social interactions

can signiﬁcantly reduce latency and improve the over-

all effectiveness of public health monitoring.

6 CONCLUSION

This work proposed a new approach for the detection

of social interactions carried out in the daily life of

the user. The approach consists of joining previously

used methods such as: Bluetooth search to ﬁnd pe-

ople close to the user, localization and conversation

identiﬁcation. And new methods such as detection of

the user who is talking. The collected data are used

to generate social graphs that clearly demonstrate the

relationships among users of a group.

The application collected satisfactory data for the

development of social graphs capable of identifying

people susceptible to infectious diseases. Providing

the possibility for health professionals to intervene

with agility in the control of epidemics. As future

work, we intend to insert and adapt SocialCount in ot-

her areas, such as: sociology (identiﬁcation of inclu-

sion and social exclusion), business (employee relati-

onship mapping), marketing (sales mapping by word-

of-mouth), etc.

ACKNOWLEDGMENTS

The authors would like to thank CAPES for par-

tial funding of this research and the UFSM/FATEC

through project number 041250 - 9.07.0025 (100548).

REFERENCES

Adams, B., Phung, D., and Venkatesh, S. (2008). Sensing

and using social context. ACM Transactions on Mul-

timedia Computing, Communications, and Applicati-

ons (TOMM), 5(2):11.

Baldauf, M., Dustdar, S., and Rosenberg, F. (2007). A sur-

vey on context-aware systems. International Journal

of Ad Hoc and Ubiquitous Computing, 2(4):263–277.

Biamino, G. (2011). Modeling social contexts for pervasive

computing environments. In Pervasive Computing

and Communications Workshops (PERCOM Works-

hops), 2011 IEEE International Conference on, pages

415–420. IEEE.

Carter, I. (2013). Human behavior in the social environ-

ment. AldineTransaction.

Cassens, J. and Kofod-Petersen, A. (2006). Using activity

theory to model context awareness: A qualitative case

study. In FLAIRS Conference, pages 619–624.

SocialCount

517

Clayton, D. and Hills, M. (1993). Statistical methods in

epidemiology. Oxford University Press.

Crickx, A. (2014). Recognito: Text independent speaker

recognition in java. https://github.com/amaurycrickx/

recognito.

de Araujo, R. B. (2003). Computac¸

ao ub

ıqua: Princ

ıpios,

tecnologias e desaﬁos. In XXI Simp

osio Brasileiro de

Redes de Computadores, volume 8, pages 11–13.

De Cheveign

e, A. and Kawahara, H. (2002). Yin, a fun-

damental frequency estimator for speech and music.

The Journal of the Acoustical Society of America,

111(4):1917–1930.

Dey, A. K. (2001). Understanding and using context. Per-

sonal and ubiquitous computing, 5(1):4–7.

Eubank, S., Guclu, H., Kumar, V. A., Marathe, M. V., et al.

(2004). Modelling disease outbreaks in realistic urban

social networks. Nature, 429(6988):180.

Ferscha, A. (2012). 20 years past weiser: What’s next?

IEEE Pervasive Computing, 11(1):52–61.

Herrera, J. L., Srinivasan, R., Brownstein, J. S., Galvani,

A. P., and Meyers, L. A. (2016). Disease surveillance

on complex social networks. PLoS computational bi-

ology, 12(7):e1004928.

Hoseinitabatabaei, S. A., Gluhak, A., Tafazolli, R., and He-

adley, W. (2014). Design, realization, and evaluation

of udirect-an approach for pervasive observation of

user facing direction on mobile phones. IEEE Tran-

sactions on Mobile Computing, 13(9):1981–1994.

Kabir, M. A., Colman, A., and Han, J. (2014). Socioplat-

form: a platform for social context-aware applicati-

ons. In Context in Computing, pages 291–308. Sprin-

ger.

Kim, C. P. J. L. J. and Lee, S.-J. L. D. (2017). “don’t bot-

her me. i’m socializing!”: A breakpoint-based smartp-

hone notiﬁcation system.

Kofod-Petersen, A. and Mikalsen, M. (2005). Context: Re-

presentation and reasoning. Special issue of the Re-

vue d’Intelligence Artiﬁcielle on” Applying Context-

Management.

Kolvenbach, S., Grather, W., and Klockner, K. (2004). Ma-

king community work aware. In Parallel, Distributed

and Network-Based Processing, 2004. Proceedings.

12th Euromicro Conference on, pages 358–363. IEEE.

Lu, H., Pan, W., Lane, N. D., Choudhury, T., and Campbell,

A. T. (2009). Soundsense: scalable sound sensing for

people-centric applications on mobile phones. In Pro-

ceedings of the 7th international conference on Mo-

bile systems, applications, and services, pages 165–

178. ACM.

Matic, A., Osmani, V., Maxhuni, A., and Mayora, O.

(2012). Multi-modal mobile sensing of social in-

teractions. In Pervasive computing technologies for

healthcare (PervasiveHealth), 2012 6th international

conference on, pages 105–114. IEEE.

Miluzzo, E., Lane, N., Eisenman, S., and Campbell, A.

(2007). Cenceme–injecting sensing presence into so-

cial networking applications. Smart sensing and con-

text, pages 1–28.

Newman, M. E. (2002). Spread of epidemic disease on net-

works. Physical review E, 66(1):016128.

Palaghias, N., Hoseinitabatabaei, S. A., Nati, M., Gluhak,

A., and Moessner, K. (2015). Accurate detection

of real-world social interactions with smartphones.

In Communications (ICC), 2015 IEEE International

Conference on, pages 579–585. IEEE.

Palaghias, N., Loumis, N., Georgoulas, S., and Moessner,

K. (2016). Quantifying trust relationships based on

real-world social interactions. In Communications

(ICC), 2016 IEEE International Conference on, pages

1–7. IEEE.

Schilit, B., Adams, N., and Want, R. (1994). Context-

aware computing applications. In Mobile Computing

Systems and Applications, 1994. WMCSA 1994. First

Workshop on, pages 85–90. IEEE.

Six, J., Cornelis, O., and Leman, M. (2014). TarsosDSP, a

Real-Time Audio Processing Framework in Java. In

Proceedings of the 53rd AES Conference (AES 53rd).

Temdee, P. and Prasad, R. (2017). Context-aware commu-

nication and computing: Applications for smart envi-

ronment.

Teng, J., Zhang, B., Li, X., Bai, X., and Xuan, D.

(2014). E-shadow: Lubricating social interaction

using mobile phones. IEEE Transactions on Compu-

ters, 63(6):1422–1433.

Weisstein, E. W. (2017). Sir model. http://mathworld.

wolfram.com/SIRModel.html.

Xu, Q., Chia, S. C., Mandal, B., Li, L., Lim, J.-H., Mu-

kawa, M. A., and Tan, C. (2016). Socioglass: social

interaction assistance with face recognition on goo-

gle glass. Scientiﬁc Phone Apps and Mobile Devices,

2(1):7.

Yu, Z., Yu, Z., and Zhou, X. (2012). Socially aware compu-

ting. Chinese Journal of Computers, 35(1):16–26.

Zhang, R., Zhang, Y., Sun, J., and Yan, G. (2012). Fine-

grained private matching for proximity-based mobile

social networking. In INFOCOM, 2012 Proceedings

IEEE, pages 1969–1977. IEEE.

ICEIS 2018 - 20th International Conference on Enterprise Information Systems

518