Credibility-based Model for News Spreading on Online Social Networks
Vincenza Carchiolo
1 a
, Alessandro Longheu
2 b
, Michele Malgeri
2 c
, Giuseppe Mangioni
2 d
and Marialaura Previti
2
1
DMI, University of Catania, Italy
2
DIEEI, University of Catania, Italy
Keywords:
Credibility, Trust, Social Network, Information Diffusion.
Abstract:
Trustworthiness in Online Social Networks has become essential to discriminate the goodness of both different
information as well as the users it originates. Here a model for news spreading in directed online social
networks (OSNs) that takes into account trustworthiness-related issues is introduced. In particular we add
a credibility network on top of the acquaintance network naturally present in OSNs to model the changing
of each node’s opinion about his/her neighbors every time a piece of news comes from them over the OSN.
We examine three different scenarios of news spreading over OSNs and propose a model suitable for each
scenario, evaluating its applicability using a real world weighted directed network.
1 INTRODUCTION
Thanks to the rapid growth of Online Social Networks
(OSNs) (Persia and D’Auria, 2017) and the enormous
amount of information available every day, users of
these platforms tend to acquire and disseminate news
over them. This process, called social contagion (Ho-
das and Lerman, 2014), can amplify the spread of in-
formation in a social network. The decision to propa-
gate or not news can depend on the news contents or
the trust in people that publish news.
The goal of the work presented in this paper is to
provide a model for information propagation in OSNs
that takes into account trust-related issues. In partic-
ular, we consider the directed acquaintance network
naturally presents in OSNs (e.g., Twitter) as the ba-
sis through which news spreads over the network and
we introduce an overlay weighted directed network
(the credibility network) that have edges in the oppo-
site direction respect to edges in the acquaintance net-
work, because for each piece of news spreads from an
individual the receivers form or modify their opinion
about the spreader’s credibility. The credibility values
are led by two metrics related to the topology of the
a
https://orcid.org/0000-0002-1671-840X
b
https://orcid.org/0000-0002-9898-8808
c
https://orcid.org/0000-0002-9279-3129
d
https://orcid.org/0000-0001-6910-0112
network: the trust and the reliability. After provid-
ing a quantitative definition for each metric, we apply
the model with such credibility network in three dif-
ferent simulation scenarios of information spreading
exploiting as starting point a real network. This paper
follows our previous works on the same topic (Car-
chiolo et al., 2018a)(Carchiolo et al., 2018b).
The paper is organized as follows: in section 2
an overview of existing related works is presented,
while in sec. 3 metrics and credibility network are
introduced and the different way in which a piece of
news can be propagated over the social networks are
detailed through a progressively improved model re-
adapted for each kind of propagation. In section 4 the
simulations are illustrated and discussed, and finally
in section 5 we outline our concluding remarks and
future works.
2 RELATED WORKS
Several attempts to model the news spreading over
OSNs through implicit and explicit use of trust has
been performed. The concept of trust as a way
to assess the quality of information, people, goods,
and/or virtual entities spans different research areas,
from recommendation systems (Massa and Avesani,
2007) (Carchiolo et al., 2015b) to e-commerce (Fung
and Lee, 1999), distributed on-line services (Wang
34
Carchiolo, V., Longheu, A., Malgeri, M., Mangioni, G. and Previti, M.
Credibility-based Model for News Spreading on Online Social Networks.
DOI: 10.5220/0009340500340042
In Proceedings of the 5th International Conference on Complexity, Future Information Systems and Risk (COMPLEXIS 2020), pages 34-42
ISBN: 978-989-758-427-5
Copyright
c
2020 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
and Emurian, 2005), security and privacy (Vasude-
van et al., 2012) and many others. In some cases a
trust network is created for each user and it contains
his friends as nodes and an associated trust value for
each of them as edges weights, provided that usually
trust is not for free, rather some effort must be paid
to earn trustworthiness from others (Carchiolo et al.,
2015a).
Goldbeck (Golbeck et al., 2003) proposed a
method for creating a trust network on the seman-
tic web allowing users to express the level of trust
for each person they know about a specific topic.
This weighted network was used to infer trust values
between individuals not directly connected to each
other. In another work (Golbeck, 2005) authors pro-
posed TidalTrust to derive the trust relationship based
on the premise that neighbors with higher trust ratings
are likely to agree with each other about the trust-
worthiness of a third party, so for a fixed trust rat-
ing shorter paths have a lower average difference and
higher trust ratings have a lower average difference.
This work was extended by Zhang et al. (Zhang et al.,
2006) including pairwise trust ratings and reliability
factors of the entities in the network and using an
edge-weighted network for trust assessment. It cal-
culates the similarity of two raters by comparing their
ratings about the same provider and then it is adopted
to decide which neighbor recommendation should be
followed. Comparing two recommendations, the rec-
ommendation from a rater that is more similar to the
trustor will be chosen. Dubois et al. (DuBois et al.,
2011) presented a method to compute trust and dis-
trust combining an inference algorithm that relies on
a probabilistic interpretation of trust based on random
graphs with a modified spring-embedding algorithm
in order to classify hidden trust edges as positive or
negative.
Other works that exploit the OSN structure has
been proposed by Hang and Singh (Hang and Singh,
2010) that employed a graph-based approach based
on similarity between each node’s friends trust net-
work for measuring trust with the aim to recommend
a node in a social network using the trust network,
Kuter et al. (Kuter and Golbeck, 2007) that pro-
posed a Bayesian trust inference model for estimat-
ing the confidence on the trust information obtained
from specific sources, Caverlee et al. (Caverlee et al.,
2008) that proposed a social trust model that exploits
both social relationships and feedback by users after
each interaction to evaluate trust where users’ feed-
back have a different weight based on their link qual-
ity (higher weights belong to users with a lot of links
with user having high trust ratings) and Zuo et al.
(Zuo et al., 2009) that proposed a model that uses a
trust certificate graph and calculates trust along a trust
chain, then, it exploits the composability of trust in the
form of fusion of relevant trust chains to form a base
trust chain set.
In contrast to most of the previous exposed works
that derive data from users feedback, Kim (Kim et al.,
2008) built a Web of trust without using explicit user
ratings. His approach consists on calculating users
expertise in a certain topic, which involves calcu-
lating the quality of reviews using the reputation of
raters and then the reputation of writers, calculating
the users affinity to the category, where the user affin-
ity to the ratings is derived from the average number
ratings and reviews provided by them in each cate-
gory, finally deriving degree of trust from the user’s
affinity to the topic and another users expertise on the
topic.
Another set of OSNs trust models that exists in lit-
erature only use interactions among users within the
network to calculate trust. Liu et al. (Liu et al., 2008)
proposed an approach for predicting trust in online
communities using the interaction behaviors of OSNs
users. This model considered the temporal factor, i.e.,
the time difference between two connected users re-
spective actions and described a supervised learning
approach that automatically predicts trust between a
pair of users using the user factors, representing ev-
idence derived from actions of individual users, and
the interaction factors, representing the evidence de-
rived from interactions between pairs of users.
Nepal et al. (Nepal et al., 2010) proposed STrust,
a social trust model based only on interactions within
the social network. The model consists of two types
of trust: the popularity trust refers to the acceptance
and approval of a member in the community and rep-
resenting the trustworthiness of the member from the
perspective of other members in the community, and
the engagement trust refers to the involvement of the
member in the community and representing the trust
the member has towards the community. This model
aims to increase the social capital of the community
by encouraging positive interactions within the com-
munity and, so, increase the social trust in the com-
munity.
Adali et al. (Adali et al., 2010) evaluated trust
based on communication behaviors of OSNs users.
Behavioral trust is calculated based on two types of
trust: conversation trust, that specifies how long and
how frequently two users communicate with each
other, and propagation trust that is obtained from one
user to other users in the network and indicates the de-
gree of trust placed on the information and implicitly
on the user that created the information.
In the last decade some hybrid trust models that
Credibility-based Model for News Spreading on Online Social Networks
35
use both interactions among OSNs users and OSNs
structure has been created, even if the literature on
these promising models is limited. Trifunovic et al.
(Trifunovic et al., 2010) proposed a social trust model
for opportunistic networks. This model uses two com-
plementary approaches for social trust establishment:
explicit social trust that is based on consciously es-
tablished social ties and produces a general decrease
of trust with the growth of the number of links be-
tween pairs of users, and the implicit social trust that
is based on frequency and duration of contact between
two users, but take into account not only the length
of interaction, but also the similarity in order to avoid
that a set of negative long interactions produces a high
trust between pairs of users.
Zinoviev et al. (Zinoviev et al., 2010) proposed
a game theoretical model of the information forward-
ing and feedback mechanisms in a social network that
take into account the personalities of the sender and
the receiver, including their perceived knowledge-
ability, reputation, and desire for popularity, and the
global characteristics of the network.
Wu et al. (Wu et al., 2017) investigated the dy-
namics of competitive information diffusion over a
connected social network, proposing a modified SIR
model for two competitive information, where each
individual may turn to either of the two information
after interacting with a spreader, while the spreader
associated with one information may change into the
other information. The population is divided into
three subgroups: innovators, ordinary and laggard
subgroups, and they observed that innovators and
larger network degree can help to increase the cover-
age of the information among the population but they
cannot help one information to compete with the other
one. Moreover, innovators cannot always accelerate
the convergence speed, which depends more on the
network topology.
3 THE CREDIBILITY-BASED
MODEL
As discussed previously, the simple acquaintance net-
work naturally presents in each OSNs is not sufficient
to model the propagation of news because such net-
works do not take into account several factors that
come into play when someone decides to propagate
the news. To this purpose we introduce a duplex net-
work composed by a directed acquaintance network
A = hN, Ei and a directed weighted credibility net-
work C = hN, E
0
, ci with edges in opposite direction
respect to those in the acquaintance network because
if a node spreads a piece of news the receiving node
forms an opinion about the spreader modeled through
the weight c [1, 1] (-1 indicates the highest credi-
bility whereas 1 indicates the lowest).
In directed OSNs the basic assumption is that be-
tween pairs of individuals there is not necessarily a
mutual interest in published posts, so incoming and
outgoing edges of acquaintance network hold differ-
ent roles: outgoing edges represent the link with peo-
ple interested in what we publish, while the incoming
ones are a source of inspiration for arguments of our
interest and that in some cases we want to repost. Ac-
cording to this reason we introduce two amounts in
the credibility network structure: trust and reliability.
The trust (T) in an individual indicates how much
he/she is considered trustworthy by its neighbors;
high trust values indicate that who is in contact with
him/her appreciates the contents he/she posted and
considers him/her a person who verifies the news be-
fore reposting it (it is related to incoming edges of
credibility network);
The reliability (R) of an individual shows its abil-
ity to select which neighbors he/she will accept news
from to repost, hence this parameter indirectly influ-
ences his/her ability to post true news, i.e. it is related
to outgoing edges of credibility network.
In OSNs, many users tend to link with others who
share the same news, while malicious users create
multiple accounts to repost the news. The members
of these two groups of accounts increase each other’s
trust so the network remainder that is in contact with
them can assign a low level of reliability to offset the
high level of trust; this is performed in order to re-size
their weight in the credibility network and therefore to
attenuate the echo chambers phenomenon (Baumann
et al., 2019).
3.1 Trust and Reliability Metrics
The trust and reliability parameters described above
are closely linked each other and influence the credi-
bility of an individual in his neighborhood. In partic-
ular, trust of v N is defined as:
T
t+1
(v) =
1
|v|
in
uU
in
(v)
R
t
(u)c(u, v) (1)
U
in
(v) is the set of neighbors pointing the node v
and R
t
(u) is the reliability at time t. We define relia-
bility of v N:
R
t+1
(v) = 1
1
|v|
out
uU
out
(v)
|T
t+1
(u) c(v, u)|
2
(2)
U
out
(v) is the set of neighbors who are pointed
by node v and T
t+1
(u) is the trust calculated with the
COMPLEXIS 2020 - 5th International Conference on Complexity, Future Information Systems and Risk
36
eq. 1.
Since credibility c [1,1], T [1, 1] as well,
where -1 indicates full distrust and 1 indicates full
trust; each neighbor has actually a different weight
in determining trust depending on his reliability.
To establish the reliability of a node, the closer
this value to the values of other users that contributed
to generate credibility (i.e. the smaller the difference
|T
t+1
(u)c(v, u)|), the higher its reliability. The value
2 in the denominator is used as normalization factor to
balance the discrepancy between c(v, u) and T(u) that
can be at most 2. From equation 2 R [0, 1], where 0
indicates that he/she emphasizes news of little interest
to his neighborhood and conversely 1 indicates that
he/she agrees with his neighbors opinion.
In OSNs the state when observation starts can-
not be usually considered as neutral in which nobody
knows others neither opinions exist, so it is likely to
set initial values of c(u, v) u, v N. To set in par-
ticular values for T and R we use the equations (1)
and (2) by setting R = 1 and iterating the two equa-
tions recursively until convergence occurs, i.e. un-
til |T
t+1
(v) T
t
(v)|< ε and |R
t+1
(v) R
t
(v)|< ε with
proper values for ε.
If a node does not have incoming or outgoing
edges, it is not possible to calculate the trust or re-
liability values respectively using previous equations
hence we suppose that nodes without incoming edges
are hypothetically pointed by all the other N 1 nodes
of the network and similarly nodes without outgo-
ing edges are supposed to point at each other node
in the network. For all these edges we set weight to
1, meaning that maximum credibility is assigned; the
approach resembles that used in defining the telepor-
tation in PageRank algorithm (Gleich, 2014).
Equations 1 and 2 in this scenario become:
T
t+1
(v) =
1
N 1
uN
u6=v
R
t
(u)c(u, v) =
1
N 1
uN
u6=v
R
t
(u)
(3)
(4)
R
t+1
(v) = 1
1
N 1
uN
u6=v
|T
t+1
(u) c(v, u)|
2
= 1
1
2(N 1)
uN
u6=v
|T
t+1
(u) 1|
3.2 News Spreading Models
The process of news diffusion can occur in different
ways: sometimes the news propagates over the net-
work without opposition, while in other cases con-
flicting opinions arise. To address this, we consider
three kinds of news spreading models:
No-competitive news spreading model when
news can propagate over the network without op-
position
Competitive news spreading model in which
there are two different thought factions about the
same topic that propagate the news at the same
time and a group of target OSN users reached by
both factions is called to decide which side they
are on
Competitive news spreading model with delay
same as previous but one line of thought is dis-
seminated after the other and a group of targert
OSN users is called to decide if publish the re-
traction after the propagation of first news
In the following we describe these models in a fur-
ther refinement.
3.2.1 No-competitive Model
After the setting of the initial values on credibility
network, to consider OSN users previous activities
we suppose that a set of nodes S becomes spreader,
activating themselves to propagate a piece of news.
The ignorant neighbors I exposed to that piece of
news must decide whether to repost it or not therefore
the sum of the credibility of the edges linking with
spreader nodes must exceed the activation threshold
of the inactive nodes.
The activation threshold g
t
(v) of a node v must
take into account the contribution of incoming and
outgoing edges that in the previous interactions con-
tributed to help that node to create an opinion about
its neighborhood. This two contributions are embed-
ded in trust and reliability values calculated before the
node is requested to evaluate a new piece of news
posted by a spreader, hence the threshold value will
be:
g
t
(v) = R
t1
(v)T
t1
(v) (5)
Therefore, an inactive node v will become active
if:
1
|U
S
out
(v)|
uU
S
out
(v)
c(v, u) > g
t
(v) (6)
U
S
out
(v) = U
out
(v) S is the set of neighbors
spreading the news. If there are new spreaders at time
t, the procedure is repeated for their inactive neigh-
bors otherwise the propagation ends.
3.2.2 Competitive Model
As described above in this scenario two conflicting
piece of news are spread simultaneously, hence at a
given instant t
0
two groups of spreader S
1
and S
2
are activated. For each subsequent instant for both
Credibility-based Model for News Spreading on Online Social Networks
37
pieces of news the activation of the neighboring in-
active nodes is attempted according to the following
equations:
1
|U
S
1
out
(v)|
uU
S
1
out
(v)
c(v, u) > g
t
(v) (7)
1
|U
S
2
out
(v)|
uU
S
2
out
(v)
c(v, u) > g
t
(v) (8)
If a node is exposed to both ideas at the same time
and both sums of credibilities exceed the threshold
g
t
(v) a further comparison is necessary to evaluate
which group of nodes mostly influences the inactive
node:
1
|U
S
1
out
(v)|
uU
S
1
out
(v)
c(v, u) >
1
|U
S
2
out
(v)|
uU
S
2
out
(v)
c(v, u)
(9)
If the first term is higher than the second the dis-
puted node is activated for the first piece of news, for
the second one elsewhere. At the end of each step if
there are new spreaders the attempts of propagation
are carried out on the nodes directly reached by the
new spreader, otherwise they end.
3.2.3 Competitive Model with Delay
In this type of propagation a piece of news coming
from untrustworthy nodes is spread and after a certain
number of iterations the retraction comes by trustwor-
thy nodes; the viceversa can also occur, i.e. after the
spread of a true piece of news a group of malicious
spreaders propagate a piece of news with conflicting
content.
In this case the diffusion of the malicious piece of
news and the further diffusion of the retraction at the
end of the propagation must be measured, also allow-
ing nodes that posted the malicious news to publish
the retraction if the influence of benevolent nodes is
greater than the one of malicious nodes. Therefore,
the equations are the same of previous case but the
execution times are different.
4 EXPERIMENTS
4.1 Dataset
In this work we used the Wikipedia edit war net-
work topology to carry out our simulations (e.g.
(Sumi et al., 2011)). It is composed by 116,836 nodes
and 2,027,871 directed edges related to the changes
performed by users on pages previously modified by
others. Each edge weight falls into the range [-9,12]
that depends on the number of words modified in fa-
vor (if it is an attempt to expand the information) or
against another user (if it is an attempt to reverse pre-
vious changes).
We considered only the first contact between each
pair of nodes, filtering the edge list and normalizing
the values so that they fall within the range [-1,1] in
order to use them as c(u, v). This is important be-
cause it avoids that simulations begin from random
trust and reliability values that would not be repre-
sentative of the previous interaction history between
pairs of nodes.
4.2 Simulator Workflow
To evaluate the model we implemented a simulator
that carries out the following steps in order to emulate
the spreading processes on OSN:
It reads the edgelist and generates the correspond-
ing weighted directed network
exploiting weights, it calculates trust, reliability
and the threshold value (respectively with eqs. 1,
2 and 5) for each node
It creates the trust ranking and selects the percent-
age of seeds, i.e. the initial spreaders of a piece
news (0.1%, 0.2%, 0.5%, 1%, 2% and 5% of to-
tal network nodes) from the top or the bottom of
ranking and actives them depending on which cat-
egory of nodes it wants to use as seeds of news
propagation, high trust seeds (HTS) or low trust
seeds (LTS))
For each inactive node that has at least a spreader
as neighbors it checks the eqs. 6 or 7 and 8 (in
competitive cases also 9 if the threshold is over-
come by both factions) and activates the inactive
nodes whose threshold are exceeded
At the end of each loop, checks the list of new
spreaders and if it is not empty restarts the loop;
When the list of new spreaders does not contain
new nodes, it saves relevant information about
propagation.
4.3 Simulations
In the case of a no-competitive model we have two
cases of initial spreader nodes: trustworthy (HTS) and
not trustworthy (LTS) and simulator calculates how
many nodes decide to propagate the piece of news
from each group of initial spreader nodes.
In the case of a competitive model it calculates
how many nodes decide to propagate the piece of
COMPLEXIS 2020 - 5th International Conference on Complexity, Future Information Systems and Risk
38
news from two conflicting factions, high and low trust
nodes. In the case of disputed nodes, i.e. nodes that
come into contact with both groups of spreaders, it is
interesting to see how many people follow each of the
two sides.
In the case of a competitive model with delay in
addition to the number of nodes that decide to prop-
agate the piece of news of each faction, it calculates
the number of nodes changing their opinion after the
propagation of retraction.
Therefore, the following five variants of the afore-
mentioned evolution of the model will be examined:
High trust seeds in no-competitive model
Low trust seeds in no-competitive model
High trust seeds vs low trust seeds in competitive
model
High trust seeds vs low trust seeds in delayed
competitive model
Low trust seeds vs high trust seeds in delayed
competitive model
4.4 Results
The distribution of c, T , R and g calculated in the sec-
ond step of simulator workflow on the Wikipedia edit
war network are shown in figs. 1, 2, 3 and 4 respec-
tively.
Figure 1: Distributions of credibility initial values for
Wikipedia edit war dataset.
Most interactions in this network are to reverse
text changes hence most c have a small negative value.
Accordingly to assign negative credibility values to
their neighbors, the reliability values are very high in
the initial phase while the g threshold histogram have
the same form of the trust histogram, it just appears
more contracted along the X axis due to the reliability
resizing. The simulator outputs are shown in tables 1,
2 and 3.
In no-competitive model (table 1) we note that us-
ing the same number of seeds HTS can always propa-
gate a piece of news more effectively than LTS. When
Figure 2: Distributions of trust initial values for Wikipedia
edit war dataset.
Figure 3: Distributions of reliability initial values for
Wikipedia edit war dataset.
Figure 4: Distributions of g threshold initial values for
Wikipedia edit war dataset.
the seeds are few (0.1%) compared to the total size
of the network, LTS cannot convince anyone to re-
post the piece of news while in the last case (5%) the
propagation of the LTS is equal to 1/4 of HTS one,
probably thanks to the large number of promoters.
In competitive model (table 2) it can be observed
that for the same number of seeds the propagations of
both factions are smaller than in no-competitive case.
Particularly interesting is the cases of the propaga-
tion of LTS with few seeds (0.1%, 0.2% and 0.5%)
in which only nodes with few or no contacts with the
opposite faction are convinced to propagate the piece
of news while almost all nodes reached by HTS has
Credibility-based Model for News Spreading on Online Social Networks
39
Table 1: No-Competitive model. Number of nodes that decide to propagate the piece of news deriving from LTS and HTS
respectively.
% of seeds LTS HTS
0.1 116 13681
0.2 234 14904
0.5 591 16551
1 7657 17763
2 4696 20655
5 6253 24830
Table 2: Competitive model. Number of nodes that decide to propagate a piece of news and, in this set, number nodes in
touch with opposite faction for news deriving from LTS and HTS respectively.
% of seeds LTS LTS (HTS) HTS HTS (LTS)
0.1 116 0 13662 10905
0.2 234 1 14863 11989
0.5 590 6 16443 13218
1 1215 47 17512 15042
2 2564 128 20019 15042
5 6153 312 23245 14763
Table 3: Competitive model with delay. Number of nodes that decide to propagate the piece of news at the end of each
propagation and number of reversed nodes that change their mind after the second propagation for news deriving from LTS
and HTS respectively and vice versa.
% of seeds LTS HTS Reversed HTS LTS Reversed
0.1 116 13681 19 13681 116 0
0.2 234 14903 40 14904 234 1
0.5 591 16551 110 16551 588 0
1 7657 17681 5421 17763 5738 3721
2 4696 20582 2413 20655 3458 852
5 6253 24778 1720 24830 6117 83
contacts with the opposite faction and decided any-
way to propagate the HTS piece of news. This means
that when there are many seeds with low trust (i.e. the
piece of news is perceived as false) it is not the content
of the news that plays a fundamental role in its propa-
gation rather the influence of the numerous neighbors
that propagate it.
In the competitive model with delay (table 3) the
HTS propagate the news more effectively than the
LTS as in the previous case but when the denial comes
from the HTS for all the percentage of seeds there
are some nodes that change ideafor a small number
of seeds (0.1%, 0.2% and 0.5%) there are no publi-
cations of the retraction. In each case the number of
retractions is higher for HTS respect to LTS.
5 CONCLUSIONS AND FUTURE
WORK
In this work, we proposed a model that exploits trust
and reliability with 3 variations in order to describe
the different way a piece of news can be propagated
through OSNs: in the no-competitive model each
piece of news is free to propagate over the network
without opposition, in competitive model two differ-
ent factions propagate two pieces of news with op-
posite contents at the same time in order to convince
the undecided social network users to align with their
own thought group and finally in competitive model
with delay the second piece of news propagation hap-
pens after the propagation of the first one in order to
convince other user to change their minds and publish
a retraction.
We implemented a simulator and used a real
weighted directed network as starting point for the
simulation. This simulator exploits different number
COMPLEXIS 2020 - 5th International Conference on Complexity, Future Information Systems and Risk
40
of seeds belonging to two different groups of users,
high trust seeds and low trust seeds. The purpose is
twofold, we evaluated the influence in news propaga-
tion of large groups of seeds respect to smaller ones
and the importance of a high trust respect to a low
one.
We discovered that as in real OSNs occurs, if a
piece of news is propagated by a small group of user
with low trust it does not propagate, indeed in such
cases only the initial seeds remain involved in news
spreading while if the group of seeds has a high num-
ber of members it propagates news over the network
but with a minor impact respect to the case of the
same number of seeds with high trust. This means
that trust is an important metric in this scenarios but
also the influence of neighborhood plays a key role in
the decision to further spread or not.
Currently we are working to an improved model
where dynamics is introduced by inserting a credibil-
ity update mechanism to check whether and to what
extent the changing of user behaviors (e.g. a HTS
suddenly start to spread fake news or viceversa) af-
fects the credibility and the related spread.
REFERENCES
Adali, S., Escriva, R., Goldberg, M. K., Hayvanovych,
M., Magdon-Ismail, M., Szymanski, B. K., Wallace,
W. A., and Williams, G. (2010). Measuring behavioral
trust in social networks. In 2010 IEEE International
Conference on Intelligence and Security Informatics,
pages 150–152. IEEE.
Baumann, F., Lorenz-Spreen, P., Sokolov, I. M., and
Starnini, M. (2019). Modeling echo chambers and po-
larization dynamics in social networks.
Carchiolo, V., Longheu, A., Malgeri, M., and Mangioni,
G. (2015a). The cost of trust in the dynamics of best
attachment. Computing and Informatics, 34(1):167–
184.
Carchiolo, V., Longheu, A., Malgeri, M., and Mangioni,
G. (2015b). Searching for experts in a context-aware
recommendation network. Comput. Hum. Behav.,
51(PB):1086–1091.
Carchiolo, V., Longheu, A., Malgeri, M., Mangioni, G., and
Previti, M. (2018a). Introducing credibility to model
news spreading. In Cherifi, C., Cherifi, H., Karsai, M.,
and Musolesi, M., editors, Complex Networks & Their
Applications VI, pages 980–988, Cham. Springer In-
ternational Publishing.
Carchiolo, V., Longheu, A., Malgeri, M., Mangioni, G., and
Previti, M. (2018b). A trust-based news spreading
model. In Cornelius, S., Coronges, K., Gonc¸alves,
B., Sinatra, R., and Vespignani, A., editors, Complex
Networks IX, pages 303–310, Cham. Springer Interna-
tional Publishing.
Caverlee, J., Liu, L., and Webb, S. (2008). Socialtrust:
tamper-resilient trust establishment in online com-
munities. In Proceedings of the 8th ACM/IEEE-CS
joint conference on Digital libraries, pages 104–114.
ACM.
DuBois, T., Golbeck, J., and Srinivasan, A. (2011). Pre-
dicting trust and distrust in social networks. In 2011
IEEE third international conference on privacy, se-
curity, risk and trust and 2011 IEEE third interna-
tional conference on social computing, pages 418–
424. IEEE.
Fung, R. and Lee, M. (1999). Ec-trust (trust in electronic
commerce): Exploring the antecedent factors. In Pro-
ceedings of the 5th Americas Conference on Informa-
tion Systems, pages 517–519.
Gleich, D. F. (2014). Pagerank beyond the web.
Golbeck, J. (2005). Personalizing applications through inte-
gration of inferred trust values in semantic web-based
social networks. In Semantic Network Analysis Work-
shop at the 4th International Semantic Web Confer-
ence, volume 16, page 30. Publishing.
Golbeck, J., Parsia, B., and Hendler, J. (2003). Trust net-
works on the semantic web. In International work-
shop on cooperative information agents, pages 238–
249. Springer.
Hang, W. and Singh, M. (2010). Trust based recommenda-
tion based on graph similarities.
Hodas, N. O. and Lerman, K. (2014). The simple rules of
social contagion. Scientific reports, 4:4343.
Kim, Y. A., Le, M.-T., Lauw, H. W., Lim, E.-P., Liu, H., and
Srivastava, J. (2008). Building a web of trust with-
out explicit trust ratings. In 2008 IEEE 24th Inter-
national Conference on Data Engineering Workshop,
pages 531–536. IEEE.
Kuter, U. and Golbeck, J. (2007). Sunny: A new algo-
rithm for trust inference in social networks using prob-
abilistic confidence models. In AAAI, volume 7, pages
1377–1382.
Liu, H., Lim, E.-P., Lauw, H. W., Le, M.-T., Sun, A., Srivas-
tava, J., and Kim, Y. (2008). Predicting trusts among
users of online communities: an epinions case study.
In Proceedings of the 9th ACM conference on Elec-
tronic commerce, pages 310–319. ACM.
Massa, P. and Avesani, P. (2007). Trust-aware recommender
systems. In RecSys ’07: Proceedings of the 2007 ACM
conference on Recommender systems, pages 17–24,
New York, NY, USA. ACM.
Nepal, S., Sherchan, W., and Bouguettaya, A. (2010). A
behaviour-based trust model for service web. In 2010
IEEE International Conference on Service-Oriented
Computing and Applications (SOCA), pages 1–4.
IEEE.
Persia, F. and D’Auria, D. (2017). A survey of online so-
cial networks: Challenges and opportunities. In 2017
IEEE International Conference on Information Reuse
and Integration (IRI), pages 614–620.
Sumi, R., Yasseri, T., Rung, A., Kornai, A., and Kertesz, J.
(2011). Edit wars in wikipedia. 2011 IEEE Third Intl
Conference on Privacy, Security, Risk and Trust and
Credibility-based Model for News Spreading on Online Social Networks
41
2011 IEEE Third Intl Conference on Social Comput-
ing.
Trifunovic, S., Legendre, F., and Anastasiades, C. (2010).
Social trust in opportunistic networks. In 2010 IN-
FOCOM IEEE Conference on Computer Communi-
cations Workshops, pages 1–6. IEEE.
Vasudevan, A., Owusu, E., Zhou, Z., Newsome, J., and Mc-
Cune, J. M. (2012). Trustworthy execution on mo-
bile devices: what security properties can my mobile
platform give me? In Proceedings of the 5th interna-
tional conference on Trust and Trustworthy Comput-
ing, TRUST’12, pages 159–178, Berlin, Heidelberg.
Springer-Verlag.
Wang, Y. D. and Emurian, H. H. (2005). An overview of on-
line trust: Concepts, elements, and implications. Com-
puters in Human Behavior, 21(1):105 – 125.
Wu, J., Chiclana, F., Fujita, H., and Herrera-Viedma, E.
(2017). A visual interaction consensus model for so-
cial network group decision making with trust propa-
gation. Knowledge-Based Systems, 122:39–50.
Zhang, Y., Chen, H., and Wu, Z. (2006). A social network-
based trust model for the semantic web. In Interna-
tional Conference on Autonomic and Trusted Comput-
ing, pages 183–192. Springer.
Zinoviev, D., Duong, V., and Zhang, H. (2010). A
game theoretical approach to modeling information
dissemination in social networks. arXiv preprint
arXiv:1006.5493.
Zuo, Y., Hu, W.-c., and O’Keefe, T. (2009). Trust comput-
ing for social networking. In 2009 Sixth International
Conference on Information Technology: New Gener-
ations, pages 1534–1539. IEEE.
COMPLEXIS 2020 - 5th International Conference on Complexity, Future Information Systems and Risk
42