=Paper=
{{Paper
|id=Vol-1959/paper-10
|storemode=property
|title=The Brain is a Social Network
|pdfUrl=https://ceur-ws.org/Vol-1959/paper-10.pdf
|volume=Vol-1959
|authors=Claudio Tomazzoli,Silvia Francesca Storti,Ilaria Boscolo Galazzo,Matteo Cristani,Gloria Menegaz
|dblpUrl=https://dblp.org/rec/conf/kdweb/TomazzoliSGCM17
}}
==The Brain is a Social Network==
https://ceur-ws.org/Vol-1959/paper-10.pdf
The Brain is a Social Network
Claudio Tomazzoli, Silvia Francesca Storti, Ilaria Boscolo Galazzo, Matteo
Cristani, and Gloria Menegaz
University of Verona
{claudio.tomazzoli, silviafrancesca.storti, ilaria.boscologalazzo,
matteo.cristani, gloria.menegaz}@univr.it
Abstract. Social Network Analysis is employed widely as a means to
compute the probability that a given message flows through a social net-
work. This approach is mainly grounded upon the correct usage of three
basic graph-theoretic measures: degree centrality, closeness centrality and
betweeness centrality. We developed a model, using Semantic Social Net-
work Analysis, that overcomes the drawbacks of general indices and we
found that this model can be applied, after appropriate adaptations, to
a very di↵erent domain such as brain connectivity.
1 Introduction
Social Networks are considered, on the current panorama of web applications, as
the principal virtual space for online communication. Therefore, it is of strong
relevance for practical applications to understand how strong a member of the
network is with respect to the others.
Traditionally, sociological investigations have dealt with problems of defining
properties of the users that can value their relevance (sometimes their impor-
tance, that can be considered di↵erent, the first denoting the ability to emerge,
and the second the relevance perceived by the others). Scholars have developed
several measures and studied how to compute them in di↵erent types of graphs,
used as models for social networks. This field of research has been named Social
Network Analysis. Sometimes the same name is attributed to a wider context,
where we also mean to include analysis of the ways in which such values arise
(for instance, processes able to change importance of members), or to provide
methods for employing these measures in applications.
Majorly, scholars dealt with the Social Network Analysis from the viewpoint
of information flow, namely they provide models of importance (and other as-
pects as well) to understand how probable would be that a piece of information
passed through a given node. Mainly, the information flow has been studied
for propagation of viruses (both in medical and in computer security contexts),
news spread-out (and hence, studies about viral marketing as well), and message
passing in certain application contexts.
The analysis model we propose in this paper can be extended to the study of
brain, with potentials for for a deeper understanding of neurological disorders.
In particular, in the last two decades, the study of brain functional connectiv-
ity (FC) from multivariate neuroimaging data (eletroencephalography [EEG],
functional magnetic resonance imaging [fMRI] etc.) has become an increasingly
active field of research providing novel and crucial insights into normal brain
functions in resting state as well as during tasks and their disruption in brain
pathologies. Network modeling and analysis applied in this field of research en-
able the study of functional brain connections by extracting significant aspects
of network organization. In this paper we provide a model, the social semantic
analysis [6] that associates vertices with regions of the brain, signals are weights
on the edges connecting regions of the brain, and activities are mapped onto
topics of the semantic analysis. On this basis we can rapidly evidentiate the
regions that are involved in a given activity, since they exhibit a greater cen-
trality measure related to that topic. On the contrary, when a region is involved
in more than one single topic, it exhibits a great transtopic centrality measure.
When we assume that a specific topic represents a pathological condition, as for
instance epilepsy, a centrality measure would allow detecting the source of the
seizures. These “hub” nodes would play a leading role in the seizure generation
and propagation[16]. Transtopic concept could further validate this hypothe-
sis: we could distinguish between a source that a↵ects other higher functions
(high transtopic centrality measure) and an isolated source of such activity (low
transtopic centrality measure).
The purpose of this paper is to give account to the aspects showed above. We
provide a model of Social Network Analysis that takes into account topics, and
show that it can foresee information flow for message treating those topics in a
more accurate way than classical topic-free social network analysis. We also name
Semantic Social Network Analysis the techniques we studied in this investigation
to cover a part of research that some previous studies did not cover satisfactorily.
The rest of the paper is organised as follows: in Section 4 we discuss related
work on the subject. Further we employ Section 2 to provide the actual technical
part of the paper and in Section 3 . Finally Section 5 takes some conclusions
and sketches further work.
2 Semantic Social Network Analysis
In this section we introduce the theme of graph theory, we apply to Social Net-
work Analysis and then we show how the model may be applied to the study
of human brain connections. The basis of both is the very general notion of a
labelled graph, that we assume to be known to the reader, and specify in terms
of form of the labels in Subsection 2.1. Social Network Analysis is extended
in semantic terms in Subsection 2.2 and the model for brain connectomics is
explained in Subsection 2.3.
2.1 Graph theory preliminaries and Social Network Analysis
A graph is a pair G = hV, Ei, where V is a finite set of vertices, and E is a set of
edges. A graph G is labelled when to each vertex or to each edge is associated a
label, determined by ⇤, a function that associates vertices and edges to the label
sets (that are thus denoted by (V) and (E), or simply by , meant to be the
union of the above). We use the term node and the term vertex indi↵erently.
In this paper we concentrate ourselves onto indirected graphs, and delay
the investigation on directed graphs to further work. We also assume that the
graphs we deal with have no circular edge (although we do not assume them to
be acyclic).
To treat the notion of distance we employ notions derived from classic al-
gorithmic graph theory, as widely discussed in [10]. The distance between two
vertices v1 and v2 , denoted by (v1 , v2 ), in a graph, is the length of the shortest
path connecting v1 and v2 .
W e now introduce three basic measures of Social Network Analysis, and
discuss several flaws they exhibit.
Definition 1. A node v of a graph is said to have absolute degree centrality k
when the number of edges incident to v is k.
From absolute degree centrality we can easily derive relative degree centrality, as
the absolute degree centrality weighted by the size of the graph. In other terms,
if a graph has n vertices, and a node v has absolute degree centrality k, the
relative degree centrality of v is nk .
We now introduce the second measure of social network analysis.
Definition 2. A node v of a graph is said to have farness f such that:
N
X
f (vi ) = (vi , vk )
k=1,k6=i
The closeness centrality of a vertex is the reciproce of the farness of v:
1
c(v) =
f (v)
Closeness centrality only works for connected graphs. Conceptually, however,
two vertices that are not connected ar as far as infinite, thus being closeness
centrality of these equivalent to 0. This is clearly faulty. There have been several
proposals to solve this aspect, mainly by means of techniques that are based
upon weights. For the purpose of this research we assume that networks are
connected.
The third measure we consider here is betweeness centrality. It is obtained
as the number of pairs of vertices that are traversed by a path containing the
measured vertex, or such that the vertex is between the elements of the pair.
Definition 3. The betweeness centrality of a vertex v is the number K of pairs
in the graph for which v is between.
Analogously to the previous analysis about degree centrality we can note that
large networks may exhibit wider spectrum of betweeness centrality than smaller
ones, and conversely, less wide.
2.2 Semantic Social Network Analysis
Social Network Analysis starts from unlabelled and, for the settings of this in-
vestigation, indirected graphs.
This, as discussed in Section 1 is unrealistic. An individual can be con-
nected to another over a certain topic, but definitively disconnected over another
topic [17, 5]. Moreover, the link between individuals can be indipendent from the
shared topics. These concepts are known as homophily and have been discussed
in Section 4.
If we measure centralities by means of unlabelled graphs, we may be rather
misleading in defining the relevance (or importance, in some sense) of a vertex
in the graph, since an individual can exhibit strong connections on one specific
topic and weak ones on other topics, providing therefore a di↵erentiated degree
centrality, in particular, and analogously for closeness and betweeness measures.
Consider two individuals John and Alice belonging to the same school but not
sharing a hobby like music. A message regarding a class has a good probability
of being forwarded from John to Alice while one regarding the hobby hasn’t.
So we can say that John and Alice are connected over a topic school but are
disconnected over topic music.
A method to provide this is to add a label to vertices in the graph with label
corresponding to a measure of depths relative to a set of topics T = {t1 , . . . , tk }.
The label will be vector D = {d1t1 , . . . dntk } whose component ditj is the depth over
the topic tj of the individual represented by vertex i as can be seen in Figure 2a.
The goal of centrality measures is to provide a tool for foreseeing information
d2t1 , d2t2 , d2t3 , (0.3, 0.5, 0.5, 0.0,
d2t4 , d2t5 0.0)
N2 Alice
d1t1 , d1t2 , d1t3 , (0.9, 0.7, 0.0, 0.5,
d1t4 , d1t5 0.0)
d5t1 , d5t2 , d5t3 , (0.9, 0.5, 0.7, 0.3,
N1 John
d5t4 , d5t5 0.0)
N5 Charlie
N3 Bob
d3t1 , d3t2 , d3t3 , (0.5, 0.7, 0.9, 0.3,
N4 Annie
d3t4 , d3t5 0.0)
d4t1 , d4t2 , d4t3 , (0.3, 0.0, 0.7, 0.0,
d4t4 , d4t5 0.0)
a)Introducing labels for depth in topics. b)Example of s.n. with labels on topics
Fig. 2: Social networks (s.n) as labelled graphs
flow. The basic assumption we make here is that when someone is not involved
in a specific topic, it is rather unlikely that she promotes the flow of a piece of
information through the vertex she occupies.
Considering the aforementioned two individuals John and Alice, their in-
terests in the set of topics T = {gossip, music, sport, cooking, politics} can be
expressed as Da = (0.3, 0.5, 0.5, 0.0, 0.0) for Alice and Dj = (0.9, 0.7, 0.0, 0.5, 0.0)
for John, meaning that while both are interested in gossip and music and not
interested in politics, Alice is keen to sport and John is not but he likes cooking
while Alice does not, as is expressed in Figure 2b in which topic labels are about
gossip, music, sport, cooking, politics.
An individual has depth on a certain topic measuring the degree of involve-
ment on it; an individual has also an activation threshold which describes the
inverse of the likelihood of that individual of becoming active when “hit” by a
message. The notion introduced here is inspired by that used in scientific evalu-
ation as proposed in [18].
2.3 The model for the Human Brain
This model can in our opinion be extended to the field of brain connectomics, in
which brain regions are studied along with their connections one to each other.
Each brain region can be represented as a vertex in our graph, while the arcs
represent the connections. Several studies have represented data of brain con-
nectivity as matrices of correlation values between pairs of signals from di↵erent
brain regions. These experiments are relative to several brain activities, such as
memory, language, motor tasks, or even to resting-state conditions (rest). We
make the assumption that these correlation matrices represent semantic graphs
in which the specific brain activity can be regarded as a topic. Our labeled graph
model can be adopted seamlessly to this assumption so that brain regions and
their connections, both physical and functional, can be represented with our la-
beled graph. The weights in the arcs are the values of the correlation matrix for
the specific brain activity which we call equivalent to topic.
From the Semantic Closeness centrality of a node on various topics is possible
to derive a new measure, that we named Transtopic Centrality (TC), in order
to quantify the actuation of a region considering all topics. We can define it as a
cross-topic measure. To explain, the TC can be defined as the good chance that
an region becomes engaged when a signal regarding di↵erentPr topics arrives to it.
For a node v, we can express this as: T C(v) = i=1 ↵i · SC(vi ) where r
is the number of topics and SC(v) is the Semantic Centrality of node v. The
Transtopic Centrality of a region is defined as the weighted sum of the Semantic
Closeness centralities in all topics. Intuitively, the formula finds the actors that
are closer to all other actors in the network for all topics.
In the human brain, the weights associated with the links can also have
negative values, while our model for the social network assumes only values
between zero and one. There are two possible alternative solutions: considering
the data as a transform from positive values only and considering two di↵erent
possible origin for the data: the positive and the negative network. The first
case is named “softening” while the latter “reaction” and will both be explained
in the following paragraph. A third approach can be to consider and study the
positive and the negative network separately.
Softening Here we consider the hypothesis of the negative value between two
nodes to have a meaning of ”attenuation” of the semantic value of the message
flowing between two nodes, meaning the node softens the message.
A negative value is therefore indicating that the meaning of the message is
somehow “softened” or attenuated. A positive message means that the meaning
of the message is made “stronger” or sharper , while a zero means that the
message is passed through as it has been seen in input.
This interpretation leads to a model with threshold at ‘zero’ which might as
well be mapped in a similar one in which values are all above zero.
As our data are all in the range [ 1, 1] we can define a mapping function
from this range to [0, 1] without loss of generality
1
0.5
1 0.5 0.5 1
0.5
1
Fig. 3: A possible mapping function for the softening hypothesis
Reaction Here we consider the hypothesis of the negative value between two
nodes to have a meaning of ”reaction” or inversion of the polarity of the semantic
value of the message flowing between two nodes, meaning the node reacts to the
message.
A negative value is therefore indicating that the meaning of the message is
somehow “inverted”. A positive message means that the meaning of the message
is kept with bigger values pointing to a stronger meaning, while a zero means
that the message is passed through as it has been seen in input.
This interpretation leads to a model in which the network is the result of the
superimposition of two di↵erent networks, the positive one and the negative one.
In this model our positive network is obtained from the original neutralizing
all connections with negative values, while the negative one is obtained from
the original neutralizing all connections with positive values and keeping the
absolute values of the others. With neutralizing we mean set the value to zero.
This leads to two di↵erent Centrality measures, the positive centrality (Cp )
and the negative centrality (Cn ) so that our centrality can be obtained as
C = f (Cp , Cn )
where f can be any reasonable function that considers weights of positive and
negative values, for instance f = Cp 2Cn
-.75 .75
2 1 2 1 2 1
.75 .75
.5 .5 .5 .5
3 6 = 3 6 + 3 6
-.4 -.25 .4 .25
4 5 4 5 4 5
Fig. 4: The resulting graph can be considered the composition of the positive and the
negative graphs.
3 Experiments
To test the expressivity of the methods introduced above in section 2, we ap-
plied our model to a magnetic resonance imaging (MRI) dataset1 composed of
structural and functional exams of several conditions of a single subject.
We assess how eight major domains (“topics”), that sample the diversity of
neural systems, modulate a connectivity network of 139 brain regions. The topics
include: 1) Emotion processing; 2) Gambling: designed to assess reward process-
ing and decision making; 3) Language processing (semantic and phonological
processing); 4) Motor: visual, motion, somatosensory, and motor systems; 5)
Relational processing; 6) Resting state condition; 7) Social cognition (Theory of
Mind) and 8) Working Memory (WM): working memory/cognitive control sys-
tems. The image acquisition, protocols and data preprocessing are described in
detail in [3]. Stimuli were projected onto a computer screen behind the subject’s
head within the imaging chamber. Brain parcellation, based on the Harvard-
Oxford Probabilistic MRI Atlas (HOA) and the cerebellar Atlas2 as included in
FSL, was performed. From these atlases, 139 regions were extracted. In details:
cortical and subcortical regions, considering separately the right and left hemi-
spheres, along with cerebellar ROIs divided into left, vermis and right regions.
For each of these regions, a representative mean time series was extracted by
1
HCP dataset (http://www.humanconnectome.org)
2
https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSL
Table 1: Most Central brain regions in di↵erent topics
Rank Emotion Gambling Language Motor Relational Rest Social WM
01 OLs.R PRG.R Crus II.L AG.R OLs.R OLi.L OP.R OLs.R
02 PRG.L FP.R OLi.R Crus I.R F2.L Crus II.L PRG.R PRG.R
03 VI.R Crus II.L LG.R PCN.L POG.R LG.L SGa.R FP.R
04 POG.L OLs.R OP.R OLs.L Crus II.L PRG.L PRG.L Crus I.R
05 CO.L OLs.L OLs.R PCN.R FP.R CO.L VI.L PCN.L
06 TO2.R FP.L PRG.L LG.R F1.R Crus I.L SGp.L Crus II.R
07 SGa.L F2.R PRG.R PT.R Crus I.R TOF.L OF.L F2.R
08 AG.R Crus II.R F2.R OF.L CALC.R Crus II.R SGp.R VI.R
09 F2.R Crus I.R LG.L F1.L VI.R AG.L Crus II.L F1.R
10 SGp.R CN.R F1.R TOF.L OLi.R VI.L PT.R TO2.R
11 VI.L OLi.R F1.L SGp.R PRG.L SMC.R VIIb.L AG.R
12 SPL.L VI.R OF.L Crus II.R F2.R VI.v SGa.L LG.R
13 F2.L VI.L PT.R POG.L CALC.L T1p.L VI.R SGp.R
14 IX.L F2.L CGa.L CGp.R Crus I.L OF.L POG.L OLi.R
15 CGa.R AG.R Crus II.R SPL.R AG.R INS.L Crus I.R OLs.L
averaging the time series of all voxels within the area and used as reference. A
139⇥139 symmetric connectivity matrix was then derived for each topic by calcu-
lating the Pearson correlation coefficient between pairs of nodes. These matrices
can be considered as the matrices defining our graph, in which the activities are
regarded as “topics”.
Generally, Emotion processing task involves activation of amygdala extending
into the hippocampus, as well as bilateral activation in medial and lateral orbital
frontal cortices, and some activation of ventral temporal cortex. Gambling task
is designed to assess reward processing and decision making and involves stria-
tum, insula, ventral medial prefrontal and orbitofrontal. Language processing
elicits robust activation in ventral lateral prefrontal cortex and in both superior
and inferior temporal cortices, including the anterior temporal poles bilaterally.
Activation is expected to be stronger on the left than on the right. Motor task
comprises right and left hand, foot and tongue movements and involves primary
motor, premotor, striatum, retinotopic visual areas, and cerebellum. Relational
elicits consistent activation in bilateral anterior prefrontal cortex, including tem-
poral parietal junction and superior temporal cortex regions. Social task involves
medial prefrontal cortex, temporal parietal junction and inferior and superior
temporal sulcus. Finally, working memory should activated dorsolateral and an-
terior prefrontal, inferior frontal, precentral gyrus, anterior cingulate, and dorsal
parietal regions (Barch). Finally the Resting state elites the so-called resting
state networks. Moving from simple regional activations toward a deeper un-
derstanding of the communication patterns between brain regions and network
organization, our results showed that there is extensive activation of visual re-
gions (OL) in most of the tasks, which is not surprising given the use of visual
stimuli (Fig.11). In addition to these visual areas, Transtopic highlights the com-
mon regions between tasks involving the frontal and parietal lateral cortex.
In Figure 14 the transtopic value of 30 regions is projected into a cortical
surface (L, left side; R, right side): an increase in node size (sphere radius)
represents the increase in transtopic value.
In table 1 regions with the highest degree of semantic centrality are exposed:
for the sake of readability we highlighted a few example regions that appear in
more than one topic. It can be easily noted that some regions appear in this top
fifteen but at di↵erent position while there are regions that selectively appear
only on certain topics but not in the others.
300
300
200
200
100
100
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Emotion Gambling
Fig. 6: Semantic Centrality of brain regions for topic Emotion (left) and Gambling
(right)
In figures 6, 8, 10, 12 the centrality value for each region is shown: on the
x axis the region number and on the y one the centrality value. For the sake of
space we show only the images with centrality values of the positive matrix.
As can be easily seen di↵erent regions have di↵erent centrality values de-
pending on the topic.
The values of the Transtopic Centrality for the test subject are shown in
Figure 13; It is worthy of note that the calculation has been done with equal
weights for each topic.
The picture shows how the transtopic centrality can be an efficient instrument
for quick confrontation with single topics: if a node has a high centrality in a
certain topic and also in transtopic it is likely to have high centrality also on
other topics whereas otherwise in other topics the odds are it has a low centrality.
4 Related Work
The reference literature can be considered as articulated in three themes:
300
250
250
200
200
150
150
100
100
50
50
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Language Motor
Fig. 8: Semantic Centrality of brain regions for topic Language (left) and Motor (right)
300 250
200
200
150
100
100
50
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Relational Social
Fig. 10: Semantic Centrality of brain regions for topic Relational (left) and Social (right)
– Studies about implicit social links that exist among users of the internet (or
of an internet application), or about enrichment of social web;
– Investigations of the semantics of social networks;
– Research about Social Network Analysis and relationships to semantic issues.
Regarding the first topic, we can look at methods for social link extraction,
as discussed below, as one of the best structured investigations on the theme.
This specific method for extracting social networks from the web using similarity
between collective contexts is proposed in [2]. The authors construct three social
networks on the same set of named entities. They use Jaccard, overlap and
Normalized Google Distance (NGD) [4] coefficients to retrieve degree of closeness
between entities. They show how actors may be assigned di↵erent relevance
degrees and that actors having higher ranking results may be assigned lower
ranks and inversely by choosing another measure to perform the ranking. In our
500
300
400
200
300
200
100
100
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Rest Working Memory
Fig. 12: Semantic Centrality of brain regions for topic Rest (left) and Working Memory
(right)
300
250
200
150
100
50
0 20 40 60 80 100 120 140
Fig. 13: Transtopic Centrality
perspective their work is solid, but lacks in one important aspect, the authors
build homophily on the based of the contents.
This is a technique to build a network, and not an analysis of the network
itself, as we do in this work. Su↵ering the same issue is the work of [11], where
the authors present a new framework for applying Social Netork Analysis to
RDF representations of social data. In particular, the use of graph models un-
derlying RDF and SPARQL extensions enables us to extract efficiently and to
parameterize the classic Social Network Analysis features directly from these
representations.
The main criticisms to the proposed approach lie on the fact that, as already
shown in many practical cases, it makes a lot of di↵erence, in terms of under-
stading of the structure of similarity between nodes, to know the relevance of
the two nodes. In fact, similarity can be used, as done, for instance in [7], for
community detection, where members are related to each other based on their
L! R!
Fig. 14: Node cortical transtopic of 30 regions projected into a cortical surface (L, left
side; R, right side): an increase in node size (sphere radius) represents the increase in
transtopic value.
similarity in semantic terms. This is di↵erent in terms of relationship, with re-
spect to measuring the relevance and study attactivity. Clearly, being interested
in Football lies on liking it, but the community is formed around authoritative
persons, for instance journalists. A more practical research has been documented
in [19] where an application of semantic social networks and attraction theory to
web based services is carried out. The relation between trust and Social Network
Analysis has been investigate in [20] and specified as a means for understanding
deeply the meaning of centrality and other measures as related to authority. The
same concept is employed to provide a framework for the general interpretation
of the logic bases of recommendation systems in [8].
The studies cited above all aim at discovering network links by means of
mining techniques. On the other hand, the introduction of notions derived from
semantic web into social networks is the core quest of many recent studies, in-
cluding [21]. As a complete reference to the current literature about meaning
of social links, and relationships between social web and semantics, readers can
look at [13]. More deeply, in [14] a direct and explicit comparison between social
networks and the semantic web is carried out. This paper proposes a parallel
between networked knowledge of members in a network and the basic notions
of semantic web. The same issue is dealt with, with the specificity of a known
technique, the semantic networks, in [9]. More generally, the semantic web meth-
ods are employed for understanding the meaning of social networks as sharing
platforms for common knowledge, in [15].
The idea of using Social Network Analysis as a means for forecasting the
probability of a message to pass through a given member of the network itself
is not novel at all. Base of our analysis is the criticisms to the roughness of the
employed measures, criticisms that are not novel anyhow. This has been dealt
in two distinct ways: by using semantic methods for habilitating the forecast
processes: in particular in [22], authors use semantic networks for foreseeing the
behaviour in facebook.
On the other hand, many criticisms are applied to centrality measures ([12]).
The main criticisms, that are met by the above mentioned investigations as
well as by researches tending to correct the flaws of the general methods for
centrality measures, and the measures themselves, lie on the weakness of the
notion of similarity derived from the notion of centrality. The above mentioned
notion of similarity as derived from centrality measures, and its applications to
the notion of reciprocity, a concept that has a crucial importance, for instance, in
asymmetric social networks (Instagram, Twitter) are dealt with in [1]. Authors
show that centrality measures as used so far are unsuccessful in forecasting the
information flows.
5 Conclusions
In this paper we investigated the application of a model of Semantic Social Net-
work analysis, based on topic centrality, to the description of the brain functional
connectivity evoked by di↵erent conditions of the spectrum of human behaviours.
We presented preliminary experimental data, that, by some examples of these
connections, show that the proposed model can be e↵ective in enhancing the
understanding of the shared brain functioning during di↵erent tasks or resting
state condition.
In further work, we shall investigate also ways in which this diagnostic process
will be expressed by means of large experiments, and verify its usefulness with
medical teams.
References
1. Suraj Bandyopadhyay A. Ramachandra Rao. Measures of reciprocity in a so-
cial network. Sankhy: The Indian Journal of Statistics, Series A (1961-2002),
49(2):141–188, 1987.
2. Rasim Alguliev, Ramiz Aliguliyev, and Fadai Ganjaliyev. Investigation of the role
of similarity measure and ranking algorithm in mining social networks. Journal of
Information Science, 37(3):229–234, 2011.
3. D. M. Barch, G. C. Burgess, M. P. Harms, S. E. Petersen, B. L. Schlaggar, M. Cor-
betta, M. F. Glasser, S. Curtiss, S. Dixit, C. Feldt, D. Nolan, E. Bryant, T. Hartley,
O. Footer, J. M. Bjork, R. Poldrack, S. Smith, H. Johansen-Berg, A. Z. Snyder, and
D. C. Van Essen. Function in the human connectome: task-fMRI and individual
di↵erences in behavior. Neuroimage, 80:169–189, Oct 2013.
4. Rudi Cilibrasi and Paul M. B. Vitányi. The google similarity distance. IEEE
Trans. Knowl. Data Eng., 19(3):370–383, 2007.
5. Matteo Cristani, Elisa Burato, Claudio Tomazzoli, and Katia Santacà†. The spider-
man behavior protocol: exploring both public and dark social networks for fake
identity detection in terrorism informatics. In CEUR Workshop Proceedings, vol-
ume 1, pages 1–12, 2015.
6. Matteo Cristani, Claudio Tomazzoli, and Francesco Olivieri. Semantic social net-
work analysis foresees message flows. In Proceedings of the 8th International Con-
ference on Agents and Artificial Intelligence, pages 296–303, 2016.
7. Juan David Cruz, Cécile Bothorel, and François Poulet. Community detection and
visualization in social networks: Integrating structural and semantic information.
ACM Trans. Intell. Syst. Technol., 5(1):11:1–11:26, January 2014.
8. Elnaz Davoodi, Keivan Kianmehr, and Mohsen Afsharchi. A semantic social
network-based expert recommender system. Applied Intelligence, 39(1):1–13, 2013.
9. Stephen Downes. Semantic networks and social networks. The Learning Organi-
zation, 12(5):411–417, 2005.
10. David Eppstein. Finding the k shortest paths. SIAM J. Comput., 28(2):652–673,
February 1999.
11. Guillaume Erétéo, Fabien L. Gandon, Olivier Corby, and Michel Bu↵a. Semantic
social network analysis. CoRR, abs/0904.3701, 2009.
12. Andrea Landherr, Bettina Friedl, and Julia Heidemann. A critical review of central-
ity measures in social networks. Business and Information Systems Engineering,
2(6):371–385, 2010.
13. Loet Leydesdor↵. Advances in science visualization: Social networks, semantic
maps, and discursive knowledge. CoRR, abs/1206.3746, 2012.
14. P. Mika. Social networks and the semantic web. In Web Intelligence, 2004. WI
2004. Proceedings. IEEE/WIC/ACM International Conference on, pages 285–291,
Sept 2004.
15. Peter Mika. Flink: Semantic web technology for the extraction and analysis of
social networks. Web Semantics: Science, Services and Agents on the World Wide
Web, 3(2):211 – 223, 2005. Selcted Papers from the International Semantic Web
Conference, 2004 ISWC, 20043rd. International Semantic Web Conference, 2004.
16. S. Storti, I. Boscolo Galazzo, S. Khan, P. Manganotti, and G. Menegaz. Exploring
the Epileptic Brain Network using Time-Variant E↵ective Connectivity and Graph
Theory. IEEE J Biomed Health Inform, page In press, Sep 2016.
17. Claudio Tomazzoli, Matteo Cristani, and Diana Fogoroasi. Measuring homophily.
In CEUR Workshop Proceedings, pages 1–12, 2016.
18. Yasar Tonta and Hamid R. Darvish. Di↵usion of latent semantic analysis as a
research tool: A social network analysis approach. Journal of Informetrics, 4(2):166
– 174, 2010.
19. Soe-Tysr Yuan and Yan-Lin Fei. A synthesis of semantic social network and at-
traction theory for innovating community-based e-service. Expert Systems with
Applications, 37(5):3588 – 3597, 2010.
20. Yu Zhang, Huajun Chen, and Zhaohui Wu. A social network-based trust model
for the semantic web. In LaurenceT. Yang, Hai Jin, Jianhua Ma, and Theo Un-
gerer, editors, Autonomic and Trusted Computing, volume 4158 of Lecture Notes
in Computer Science, pages 183–192. Springer Berlin Heidelberg, 2006.
21. Lina Zhou, Li Ding, and Tim Finin. How is the semantic web evolving? a dynamic
social network perspective. Computers in Human Behavior, 27(4):1294 – 1302,
2011. Social and Humanistic Computing for the Knowledge Society.
22. Jolene Zywica and James Danowski. The faces of facebookers: Investigating social
enhancement and social compensation hypotheses; predicting facebook and o✏ine
popularity from sociability and self-esteem, and mapping the meanings of pop-
ularity with semantic networks. Journal of Computer-Mediated Communication,
14(1):1–34, 2008.