Discovering Good Links Between Objects in the Internet of Things

Francesco Buccafurri

, Gianluca Lax

, Serena Nicolazzo

, Antonino Nocera

, Luca Console

and Assunta Matassa

DIIES, University Mediterranea of Reggio Calabria, Via Graziella, Localit

a Feo di Vito, 89122 Reggio Calabria, Italy

Dipartimento di Informatica, University of Torino Corso Svizzera, 185 10149 Torino, Italy

Keywords:

Internet of Things, Network Efﬁciency, Assortativity, Twitter.

Abstract:

The Internet of Things is an emerging paradigm allowing the control of the physical world via the Internet

protocol and both human-to-machine and machine-to-machine communication. In this scenario, one of the

most challenging issues is how to choose links among objects in order to guarantee an effective access to

services and data. In this paper, we present a new selection criterion that improves the classical approach. To

reach this goal, we extract knowledge coming from the social network of humans, as owners of objects, and we

exploit a recently proven property called interest assortativity. The preliminary experimental results reported

in this paper give a ﬁrst evidence of the effectiveness of our approach, which performs better than classical

strategies. This is achieved by choosing only not redundant links in such a way that network connectivity is

preserved and power consumption is reduced.

1 INTRODUCTION

The Internet of Things refers to a new paradigm com-

posed of networked interconnection of everyday ob-

jects, which are often smart, e.g., equipped with ubiq-

uitous intelligence. This innovative scenario will

increase the ubiquity of the Internet by integrating

every object for interaction via embedded systems.

Moreover, it will lead to a highly distributed network

of devices where machine-to-machine and human-to-

machine communication will be possible.

One of the basic problems to face is how to build

the Internet of Things. Indeed, the choice of a strat-

egy to drive the formation of communities of objects

has a direct impact on different aspects relevant from

the application point of view. A ﬁrst desiderata is

that the network of objects has a sufﬁcient connec-

tivity degree, to guarantee that the potential beneﬁts

arising from the communication among objects are

not inhibited. According to this principle, one could

think of a highly connected network, ideally a com-

plete graph. However, the trade-off to solve regards

the limited computational and power capabilities of

smart objects, for which the number of connections

should be minimized.

The typically adopted approach to establishing

a direct connection between two objects is mainly

based on proximity (Union, 2005; Zhang et al., 2011;

Evangelos A et al., 2011). Instead, we deﬁne a new

strategy leveraging the properties of the objects, and

estimating how much similar properties should en-

force a direct link between two objects. This is done

by matching object properties to human interests, and

by measuring the assortativity degree of such inter-

ests in the human social network of owners supposed

to be Twitter (Buccafurri et al., 2015a). The claim

is that the higher such an interest assortativity (Buc-

cafurri et al., 2015b), the higher the potential bene-

ﬁt of directly connecting the corresponding objects.

This process allows us to discover good links between

objects, which guarantee good network connectivity

among similar objects by limiting the node degree and

then the related inefﬁciencies. We tested the above

strategy experimentally and obtained very promising

results. According to a number of social-network-

analysis measures (Buccafurri et al., 2015c), we con-

clude that the network of objects created by means of

our approach is much better than the network obtained

by using the classical one.

1.1 Motivating Example

To better explain our goal, we present the following

real-life situation.

Francis

is a runner. He likes measure his per-

formance and is especially interested in knowing his

speed during his activity. He is used to measure his

advancement by a smart bracelet in companion to his

102

Buccafurri, F., Lax, G., Nicolazzo, S., Nocera, A., Console, L. and Matassa, A.

Discovering Good Links Between Objects in the Internet of Things.

DOI: 10.5220/0006475601020107

In Proceedings of the 14th International Joint Conference on e-Business and Telecommunications (ICETE 2017) - Volume 6: WINSYS, pages 102-107

ISBN: 978-989-758-261-5

smartphone. Both ensure him a full tracking of his

personal activity data. He has a Twitter account and

he is follower of many sportive man, personal coach,

instructor because he wants to stay well informed

about all news in the ﬁeld.

Lucy

is always watching her weight. She likes to

be in ﬁt and to eat biologic food, she likes read about

nutrient information on food, herbs’ properties, cos-

metics’ use and so on. To accomplish her goal, she is

used to adopt a series of mobile apps to track about

food, sports and to stay informed about these top-

ics. On Twitter, for instance, she follows famous ac-

tresses, personal trainers and nutritionists to get well

informed about these topics.

Steven

is a student. He spends a lot of time stay-

ing sit and working on the laptop. He is worried about

assuming a right posture then he had bought a pin able

to give a vibration whenever his posture is wrong or

he is staying to long. He has a Twitter account and he

is follower of healthy products in general: from pos-

ture, to food, to sports and so on. He does not like to

practice sports.

By following the interest assortativity approach,

we found out ’health’ like a common macro-interest,

pushing on this topic of interest, we are able to sug-

gest friendship between these components supporting

their owners in achieving their goals, however they

look like truly different. In fact,

Francis

and

Lucy

have never considered the importance of a right pos-

ture for the wellbeing; conversely,

Steven

does not

take care enough about sports and food, while he is a

very sedentary person. What the new approach allows

is the establishment of friendship between this series

of devices to allow the full accomplishment of per-

sonal goals of everybody, suggesting them new ones

and novel ways to accomplish the same results.

This is discussed in the next section, in which we

show that, in the cases like those described above,

the knowledge typically used to establish a connec-

tion between objects would produce unsatisfactory re-

sults. For example, by considering proximity (i.e., the

fact that the objects meet each other a given number

of times with a sufﬁcient frequency), we should con-

clude that there is no reason to connect the above ob-

jects. Indeed, probably, the above objects never meet.

1.2 Structure of the Paper

The plan of this paper is as follows. In Section 2,

we present our approach to chose links in a IoT envi-

ronment. Section 3 describes the preliminary experi-

mentation carried out to study the effectiveness of our

technique. Section 4 deals with literature related to

our work. Finally, in Section 5, we draw our conclu-

sions.

2 DISCOVERING GOOD LINKS

According to the Internet-of-Thing paradigm, an en-

tity on a network has to be notiﬁed of the availabil-

ity of desirable services or devices on the network in

order to form a link. Typically, the fact that two ob-

jects get in touch somewhere and sometimes (maybe

because the corresponding owners meet in a certain

location) is enough to trigger (with a given thresh-

old) the establishment of a link between the two ob-

jects (Union, 2005; Zhang et al., 2011; Evangelos A

et al., 2011). This property is called proximity. The

aim of this section is to identify possible enhanced

ways to discover potentially beneﬁcial links. To do

this, preliminarily, we consider which are the candi-

date properties (existing in the literature) we can use

to build a more complex model. They are: (i) prox-

imity, (ii) homogeneity, i.e., they are the same kind of

object created by the same manufacturer; (iii) owner-

ship, i.e., they belong to the same user; (iv) friendship,

i.e., owners are mutual friends in a social network.

Arguing that the decision regarding the insertion

of a link between two objects could rely on a mix of

the above properties, we deﬁne a decision function to

decide whether a link between two objects ⟨x,y⟩ has

to be inserted or not.

Observe that all the above properties give us some

information about the direct relationship between two

objects. Our proposal aims to use also some indirect

knowledge coming from the social network of own-

ers to support the computation of the above decision

function (Buccafurri et al., 2016b). In addition, de-

spite the classical selection criteria that are based on

the sole proximity, we use all the above direct prop-

erties. Therefore, we introduce two measures, which

we combine to compute the aimed decision function.

These are:

1. T

dir

x,y

, which derives from the direct knowledge

about objects and owners, and

2. T

ind

x,y

, which encodes some indirect knowledge.

As indirect knowledge we exploit a recently

proven property occurring in online social networks,

called interest assortativity (Buccafurri et al., 2016a).

According to this result, it is possible to have a mea-

sure of the correlation between a given human interest

and the presence of links between humans.

To exploit the above indirect property, we need to

match objects to humans, in such a way that interests

are someway preserved. To do this, we deﬁne a taxon-

omy on top of the properties and/or aims of objects.

Discovering Good Links Between Objects in the Internet of Things

103

This taxonomy allows us to associate each object x

with a set of (human) interests I

(belonging to a given

domain I) derived from the owners’ social network.

Clearly, the following reasoning is valid only if all the

involved humans have a social network account. To

build this taxonomy, we consider how many times a

user with a certain interest, say i, owns a given object

x. This way, we can deﬁne an occurrence degree O

of i w.r.t. x as the ratio between the number of users

with interest i owing x and the total number of occur-

rences of the interest i in the network.

Therefore, given two objects ⟨x,y⟩, we compute

the overlapping set of associated interests as I

x,y

∩I

, and, for each common interest i ∈I

x,y

, we com-

pute the assortativity level IA

of i in the considered

social network, and the common occurrence degree

x,y

deﬁned as the mean between O

and O

At this point, we are ready to deﬁne how T

ind

x,y

computed. In particular:

ind

x,y

∑

i∈I

x,y

·IA

x,y

In words, T

ind

x,y

is obtained as a mean between as-

sortativity degrees of common interests weighted by

the common occurrence degrees. Since objects ⟨x,y⟩

are both related to I

x,y

, we expect that the higher the

value T

ind

x,y

, the higher the linking power of interests in

x,y

should be also for objects.

Finally, we combine the two values T

dir

x,y

and T

ind

x,y

to obtain our boolean function to decide whether to

add a link between the two objects. Speciﬁcally, given

two objects ⟨x,y⟩, a new link is established if:

F(T

dir

x,y

ind

x,y

) ≥th

where th is a suitable threshold value, and F is a pa-

rameter of our model to set by experiments.

3 EXPERIMENTS

In this Section, we describe our experimental cam-

paign carried out in order to validate our approach. In

particular, we started from a set of humans and ob-

jects, and we built two networks:

1. a network of objects obtained by adding links ac-

cording to our approach based on interest assorta-

tivity;

2. a network of objects formed through the classical

proximity-based criterion.

The obtained results showed that the quality of the

ﬁrst network is better in terms of efﬁciency than the

second one. In the following, we will explain which

is the measures adopted to evaluate the network efﬁ-

ciency.

3.1 Testbed and Dataset

Our experiments were carried out on a machine

equipped with a 2 Quad-Core E5440 processor and 16

GB of RAM. The operating system was Linux Ubuntu

Server 14.04.4 LTS, with kernel version 4.2.0-35,

Java Virtual Machine version 1.8.0 45 (64-Bit). We

wrote our code in Java by also exploiting some fea-

tures of Neo4J (neo, 2016), a graph database manage-

ment system.

For our experiments, we used a Neo4j graph data

set, called GraphofThings, consisting of nodes cat-

egorized by one or more labels and connected by

instances of directed relationships. We obtained it

from a github repository maintained by the GraphAl-

chemist group (Gra, 2016).

Figure 1 shows an instance of the graph model

representing interactions (i.e., arcs) between entities

(i.e., nodes) used in our experiments.

For our experimental campaign, we needed only

some of the entities showed in the above schema. In

particular, the nodes we took into consideration are:

• Human. A user equipped with a device.

• User. A node representing a social network proﬁle

(i.e., Facebook , Linkedin, Twitter, etc.). Observe

that, not all the users are humans.

• Machine. A node that indicates any wearable or

mobile device. It posses a tag type that indicates

the family which it belongs to.

• Interest. A node that holds a single interest cate-

gory.

• Location. A node that indicates a physical place.

It can be equipped with a number of attributes (in-

dicating for example an event, an activity, a store,

a park, etc.).

The main relationships involved in our dataset are:

• Uses. A directed relationship between a human

and any number of devices he wears.

• Located. An action taken by a device implying

that a user were in a speciﬁc location.

• Friend. An implicitly bi-directional relationship

implying a connection in a given social network.

• Has. A relationship implying that a user has a

speciﬁc interest.

WINSYS 2017 - 14th International Conference on Wireless Networks and Mobile Systems

104

Figure 1: Graph representation of the GraphofThings nodes and relationships.

3.2 Results and Analysis

As ﬁrst step, we generated the network of humans and

objects according to the classical proximity criterion.

Speciﬁcally, we added a link between two machines

if they had got in touch at least once in a certain loca-

tion.

Secondly, we generated the network of objects by

following our criterion. This task required some more

complex steps. Indeed, for two objects ⟨x,y⟩, we

added a link between them only if T

dir

x,y

·T

ind

x,y

≥ th,

where T

dir

x,y

is a boolean function returning 1 if x and

y had got in touch in a given location at least once

or their owners have a friendship relation in the corre-

sponding social network, 0 otherwise. In other words,

for this ﬁrst investigation, we consider only proxim-

ity and friendship as direct properties and we set the

function F by the simple product between T

dir

x,y

and

ind

x,y

Once both the networks were created, we mea-

sured some quality parameters to compare them. Ta-

ble 1 shows the results of our analysis carried out by

exploiting UCINET (uci, 2016), a well-known soft-

ware package for the analysis of social network data.

From the results reported above, we can observe

that the network built by using the assortativity-based

approach shows a lower average degree and thus a

lower density level. Although, in general, high val-

ues of density imply a higher probability of reaching

target nodes, this has also a negative impact on the

network efﬁciency. Indeed, the higher the value of

density, the higher the trafﬁc level (and duplication)

in broadcast communications.

We can make a similar reasoning for all parame-

ters that have a direct relationship with the number of

point-to-point communications that nodes have to

maintain, such as average degree. Observe that

network efﬁciency is a very important aspect in an

Internet-of-Things scenario, in which smart objects

have to preserve the battery consumption. Obviously,

in this context, what we have to minimize the number

of contacts per object, provided that the efﬁciency of

the network (also in terms of cohesion) is preserved.

By considering degree centralization, our network

achieves slightly better results, allowing us to con-

clude that it is more incline to have information ac-

cumulation points that can be used as seeds to start

the information propagation (Mislove et al., 2007).

Moreover, better results are achieved for component

ratio and connectedness. These parameters measure

the cohesion of the network and show us that our

network has a single connected component (so all

the nodes are reachable) and that the connectivity is

higher than the classical network. The latter prop-

erty means that the expected number of hops to reach

a given target is reduced in our network w.r.t. the

classical one. Hence, also for the compactness and

fragmentation point of view, we can conclude that the

strategy based on assortativity chooses only not re-

dundant links that allow a full connectivity. We ob-

serve that this has not a cost in terms of network re-

silience. Indeed, according to the deﬁnition of frag-

mentation, as it is 0.676 vs 0.776 of the classical net-

work, we obtain a network that is more resilient.

In summary, the obtained results allow us to state

that the approach based on interest assortativity has

been able to build a network showing better efﬁciency

in terms of both node reachability and cohesion level.

An important achievement is that nodes exhibit a re-

duced average degree w.r.t. the classical network.

This aspect has a deep impact on the trafﬁc level (and

duplication) in broadcast communications which is

Discovering Good Links Between Objects in the Internet of Things

105

Table 1: Statistics of our dataset.

Proximity-based Our Approach

Avg Degree 12.017 10.075

Density 0.101 0.085

Deg Centralization 0.265 0.273

Component Ratio 0.529 1

Connectedness 0.224 0.324

Compactness 0.162 0.190

Fragmentation 0.776 0.676

clearly reduced in our network.

4 RELATED WORK

In recent years, IoT has gained much attention from

researchers and practitioners, because this new sce-

nario is opening new opportunities for a large number

of novel applications (Xia et al., 2012; Leavell and

Cooper, 2016; Kopetz, 2011; Atzori et al., 2010). One

of the most challenging issue is how to build the net-

work of objects to access services and data in this new

scenario (Zhang et al., 2011). Some proposals are

presented in (Edwards, 2006; Kortuem et al., 2010;

Wang et al., 2010; Yap et al., 2008; Jara et al., 2014).

In particular, in (Kortuem et al., 2010), smart items

perceive and instruct their environment by capturing

and interpreting user actions, to analyze their obser-

vations and to communicate with other objects and

the Internet. (Wang et al., 2010) presents a search

engine useful to ﬁnd suitable smart devices. The re-

trieval is performed thanks to the description con-

tained within the objects themselves, or other user

deﬁned information. A user can query the engine

to ﬁnd a particular mobile object, or a list of objects

that ﬁt the description. Also the system presented in

(Yap et al., 2008) allows humans to search for and

locate smart things as needed, although it is based

on a hierarchical architecture consisting of tags, sub-

stations and basestations. The architecture presented

in (Jara et al., 2014) exploits social tools to perform

node discovery. It allows users to register their own

sensors into a common infrastructure and then dis-

cover the available resources through mobile. Our

proposal strongly leverages on interest assortativity

(Buccafurri et al., 2016a). The concept of assortativ-

ity was ﬁrstly proposed by (Newman, 2002) in terms

of degree-degree assortativity. Therein the authors

demonstrated that social networks are often assorta-

tively mixed, in the sense that the nodes in the net-

work having many relationships tend to be connected

to other nodes highly connected themselves. Starting

from (Newman, 2002), further studies concerning so-

cial network assortativity have been proposed, such

as (Goh et al., 2003; Newman and Park, 2003; Catan-

zaro et al., 2004; Wilson et al., 2009). In particular,

(Goh et al., 2003) studies the relationship between as-

sortativity and betweenness centrality correlation for

scale-free networks. (Newman and Park, 2003) ana-

lyzes the relation between assortativity and clustering

in social communities discovering that these commu-

nities are characterized by both high levels of cluster-

ing and assortative mixing. By contrast, (Catanzaro

et al., 2004) compares different type of networks (e.g.,

technological, biological and social networks), show-

ing that only social networks are typically assortative

with respect to the degree whereas the others appear,

in general, to be disassortative. (Wilson et al., 2009)

models interaction relationships among users through

an interaction graph.

5 CONCLUSION

Internet of Things paradigm is an extremely promis-

ing scenario that opens new challenging perspec-

tives in terms of interaction between humans and ma-

chines. In this context, a lot of improvements can be

done. For instance, one of the most critical issues

is how to choose the links among objects in order to

build an efﬁcient network.

In this paper, we present a new selection criterion

that improves the classical proximity approaches. To

do this, we relied on the results presented by (Buc-

cafurri et al., 2016a). This work shows that Twitter is

highly assortative in users’ interests (i.e., users behave

uniformly w.r.t. different topics) and it presents a new

social network metric, called interest assortativity.

This position paper shows a ﬁrst experimental ev-

idence of this result by illustrating that our approach

performs better than the classical strategies. Indeed,

our selection criterion allowed us to pick only not re-

dundant links. This result is advantageous in terms of

network connectivity and battery consumption, which

are two crucial aspects for smart devices. This en-

courages us to more deeply analyze this issue in the

next future also by testing some different ways to

combine direct and indirect knowledge.

REFERENCES

(2016). GraphofThings. http://github.com/GraphAlchemist/

GraphofThings.

(2016). Neo4J: World’s leading graph database.

http://neo4j.com/.

(2016). UCINET Software. http://sites.google.com/site/

ucinetsoftware.

WINSYS 2017 - 14th International Conference on Wireless Networks and Mobile Systems

106

Atzori, L., Iera, A., and Morabito, G. (2010). The internet of

things: A survey. Computer networks, 54(15):2787–

2805.

Buccafurri, F., Lax, G., Nicolazzo, S., and Nocera, A.

(2015a). Comparing twitter and facebook user behav-

ior: privacy and other aspects. Computers in Human

Behavior, 52:87–95.

Buccafurri, F., Lax, G., Nicolazzo, S., and Nocera, A.

(2016a). Interest assortativity in twitter. In Proceed-

ings of the 13th International Conference on Web In-

formation Systems and Technologies (WEBIST). WE-

BIST.

Buccafurri, F., Lax, G., Nicolazzo, S., and Nocera, A.

(2016b). A model to support design and development

of multiple-social-network applications. Information

Sciences, 331:99–119.

Buccafurri, F., Lax, G., and Nocera, A. (2015b). A New

Form of Assortativity in Online Social Networks.

International Journal of Human-Computer Studies,

80:56–65.

Buccafurri, F., Lax, G., Nocera, A., and Ursino, D. (2015c).

Discovering Missing Me Edges across Social Net-

works. Information Sciences, 319:18–37.

Catanzaro, M., Caldarelli, G., and Pietronero, L. (2004).

Social network growth with assortative mixing. Phys-

ica A: Statistical Mechanics and its Applications,

338(1):119–124.

Edwards, W. K. (2006). Discovery systems in ubiquitous

computing. IEEE Pervasive Computing, 5(2):70–77.

Evangelos A, K., Nikolaos D, T., and Anthony C, B. (2011).

Integrating rﬁds and smart objects into a uniﬁedin-

ternet of things architecture. Advances in Internet of

Things, 2011.

Goh, K., Oh, E., Kahng, B., and Kim, D. (2003). Between-

ness centrality correlation in social networks. Physical

Review E, 67(1):017101.

Jara, A. J., Lopez, P., Fernandez, D., Castillo, J. F., Zamora,

M. A., and Skarmeta, A. F. (2014). Mobile digcovery:

discovering and interacting with the world through the

internet of things. Personal and ubiquitous computing,

18(2):323–338.

Kopetz, H. (2011). Internet of things. In Real-time systems,

pages 307–323. Springer.

Kortuem, G., Kawsar, F., Sundramoorthy, V., and Fitton, D.

(2010). Smart objects as building blocks for the inter-

net of things. IEEE Internet Computing, 14(1):44–51.

Leavell, P. and Cooper, D. (2016). The internet of things:

Strategic considerations for bankers. Journal of Digi-

tal Banking, 1(1):22–32.

Mislove, A., Marcon, M., Gummadi, K. P., Druschel, P.,

and Bhattacharjee, B. (2007). Measurement and anal-

ysis of online social networks. In Proceedings of the

7th ACM SIGCOMM conference on Internet measure-

ment, pages 29–42. ACM.

Newman, M. and Park, J. (2003). Why social networks are

different from other types of networks. Physical Re-

view E , 68(3):036122.

Newman, M. E. (2002). Assortative mixing in networks.

Physical review letters, 89(20):208701.

Union, I. T. (2005). The Internet of Things. 7th Edition,

International Telecommunication Union Internet Re-

ports.

Wang, H., Tan, C. C., and Li, Q. (2010). Snoogle:

A search engine for pervasive environments. IEEE

Transactions on Parallel and Distributed Systems,

21(8):1188–1202.

Wilson, C., Boe, B., Sala, A., Puttaswamy, K., and Zhao, B.

(2009). User interactions in social networks and their

implications. In Proc. of the ACM European Confer-

ence on Computer systems (EuroSys’09), pages 205–

218, Nuremberg, Germany. ACM.

Xia, F., Yang, L. T., Wang, L., and Vinel, A. (2012). Internet

of things. International Journal of Communication

Systems, 25(9):1101.

Yap, K.-K., Srinivasan, V., and Motani, M. (2008).

Max: Wide area human-centric search of the phys-

ical world. ACM Transactions on Sensor Networks

(TOSN), 4(4):26.

Zhang, D., Yang, L. T., and Huang, H. (2011). Searching

in internet of things: Vision and challenges. In 2011

IEEE Ninth International Symposium on Parallel and

Distributed Processing with Applications, pages 201–

206. IEEE.

Discovering Good Links Between Objects in the Internet of Things

107