Future Generation Computer Systems 84 (2018) 139–148

Contents lists available at ScienceDirect

Future Generation Computer Systems

journal homepage: www.elsevier.com/locate/fgcs

An efficient key agreement protocol for Sybil-precaution in online

social networks
Muhammad Al-Qurishi a , Sk Md Mizanur Rahman a,c , M. Shamim Hossain a,b, *,
Ahmad Almogren a , Majed Alrubaian a , Atif Alamri a,b , Mabrook Al-Rakhami a , B.B. Gupta d
a
Chair of Pervasive and Mobile Computing, College of Computer and Information Sciences, King Saud University, Riyadh, Saudi Arabia
b
Department of Software Engineering, College of Computer and Information Sciences, King Saud University, Riyadh, Saudi Arabia
c
Department of Information Systems, College of Computer and Information Sciences, King Saud University, Riyadh, Saudi Arabia
d
National Institute of Technology Kurukshetra, India

highlights

• An efficient key management protocol for defending against Sybil-attack in OSNs.

• An algorithm to generate different manipulated keys to improve the key exchange process.
• Key agreement and authentication that calculate pair wise secret session accurately.
• Demonstration of the key management using social networks.

article info a b s t r a c t
Article history: Identifying malicious users in online social networks (OSNs) is a challenging task that demands a great
Received 15 April 2017 deal of skill and knowledge because these users can have multiple forms: Sybils, bots, spammers, phishers,
Received in revised form 2 July 2017 impersonations or fake accounts. Different types of research methodologies have been proposed to solve
Accepted 21 July 2017
this problem; hence, there are varied solutions. Most of the work on OSNs has focused on trust, distrust
Available online 15 August 2017
to detect and preventing these types of attacks. Some researchers have found that a suspected node can
generate private/public keys to prevent its identity from being stolen by an adversary; however, they have
Keywords:
Pairing-based cryptography not explained how these keys are generated and managed. We propose a new and efficient centralized
Key agreement key management protocol to prevent Sybil attack and to provide a secure communication service among
Sybil attack users in OSNs. The core tenet of this method is the existence of a ‘roadblock’ that any user intending
Online social networks to join a group must go through, which includes a task that only a human user can accomplish. Hence,
automatically controlled accounts are prevented from joining, and the group will consist only of users
that have been confirmed as genuine. The mechanism is very effective in recognizing bot accounts, which
enables it to guard the network against malicious behavior by fake accounts.
© 2017 Elsevier B.V. All rights reserved.

1. Introduction First conceived by John Douceur [3] as a metaphor for a person

The most significant type of intrusion against online social
networks (OSN) is the so-called Sybil attack [1–3]. In this type
of operation, an attacker spawns a large number of automated
accounts with false credentials, that is, Sybil accounts, and initiates
the largest possible number of connections with genuine accounts,
thus artificially building up the relevance of the fake users on the
network [4]. Such user accounts are also often used for activities
that are contrary to the rules of the platform or outright illegal.
*
Corresponding author at: Department of Software Engineering, College of Com-
puter and Information Sciences, King Saud University, Riyadh, Saudi Arabia.
E-mail address: mshossain@ksu.edu.sa (M.S. Hossain).
with a split personality, the term ‘‘Sybil’’ is used in information
technology (IT) security circles to describe any attempts at system
manipulation by the means of opening multiple accounts under
fake names before using them to gain undeserved influence. This
is contrary to the stated rule that every user should be limited
to one account, and it relates to a realistic problem observed in
most real-life networks. One reason for this is that contemporary
social [networks, such as Twitter, Instagram and Facebook, allow
users to complete the registration process in minutes, with their
electronic mail accounts used as the only means of identification
[5]. This is not a serious obstacle to the creation of more than one
account, which leads to the existence of bots or Sybils whose only
real purpose is to initiate connections with legitimate members.

http://dx.doi.org/10.1016/j.future.2017.07.055
0167-739X/© 2017 Elsevier B.V. All rights reserved.
140 M. Al-Qurishi et al. / Future Generation Computer Systems 84 (2018) 139–148
Once Sybils become established in the community, they can be
used for a range of behaviors that goes against the network’s terms
of use, that is, tilting the balance of online polls by voting on
masse or propagating false information from their pseudonymous
accounts [6].
Determining which accounts are actually Sybils is a daunting
task. The issue is not new; fake accounts have haunted OSNs ever
since they were invented, and there are several works on this
topic, according to Google Scholar [7,8]. Such intense attention
quickly led to some very distinguished scientific efforts, starting
with the SybilGuard [9] solution that was first developed in 2006
and further enhanced two years later under the name SybilLimit
[10]. These solutions have proven to be effective regardless of the
level of centralization [11] and they use two major theoretical
assumptions. They assume that the community of honest users
in OSNs is tightly connected and fast mixing, whereas the Sybil
community is slow mixing [12,13] despite the fact that intruders
are able to create as many accounts as they want and can form an
indefinite number of links between them. The second assumption
is that the number of links between Sybils and legitimate users is
limited. However, these assumptions are not feasible and cannot
hold in real OSNs [6,8,12,14–16]. Later solutions, such as [17] and
Ostra [18] also apply those fundamental principles, while adding
additional solutions to allow for the improvement of the general
strategy by distributing credits between various users based on
their activities, thus boosting defensive capacities against Sybils.
None of the previous approaches have considered the problem
from the cryptographic [4,11] and key agreement points of view
in OSNs. For SybilLimit [10], a suspected node can generate pri-
vate/public keys to prevent its identity from being stolen by an
adversary; however, the authors have not explained how these
keys are generated and managed. This could be difficult because of
the nature of OSNs, in which a user might have a million followers;
thus, the traditional cryptographic technique might need to store
millions of keys for secure communication for each user. In this
paper, we focus on an innovative method for anti-Sybil protection
that introduces a trapdoor using pairing-based cryptography to
stop Sybil attacks on social networks, thereby building on similar
work on wireless sensor networks. In this methodology, users
are required to go through a ‘trapdoor’ to be verified as genuine
whenever they want to become members of online groups. Each
user needs to store its own unique secret, and can dynamically
generate a session secret based on this unique secret. Thus, storing
only one secret key would be sufficient for secure communication.
Because only legitimate users can complete the task demanded by
the trapdoor, the method is very effective for separating real users
from fake users and ensuring that the group consists exclusively of
legitimate users, while blocking Sybil-controlled accounts.
To the best of our knowledge, this is the first approach for
key management in OSNs for secure communication. We propose
a new and efficient centralized key management protocol that
is a modified version of the Tso et al. and Lim et al. protocols
[19–21]. The proposed protocol can be adopted by the service
provider to provide a secure communication service among users
in OSNs. In the proposed methodology, each user collects its own
private secret from the private key generator (PKG). The PKG server
can be maintained by the provider. The end user’s private secret
is derived from the master secret of the PKG, and it satisfies the
mathematical NP-complete problem, that is, the discrete logarithm
problem. As the process satisfies a mathematical hard problem,
by knowing the node’s secret, it is infeasible to determine the
master secret of the PKG. Any node can generate its dynamic
public component and shares this with its neighbor; an honest
neighbor node can generate a session key using the shared public
component combined with its own private secret using pairing-
based cryptography.
The remainder of the paper is organized as follows: In the next
subsection, we describe related works. In Section 2, we discuss
the preliminary concepts for understanding the proposed protocol.
In Section 3, we discuss the proposed protocol. In Section 4, we
present the experimental results. In Section 5, we discuss differ-
ent types of attacks that are protected by the proposed protocol.
Finally, in Section 6, we draw our conclusions.
1.1. Brief overview of the proposed protocol
In this paper, we propose an m-resilient authenticated key
agreement protocol for OSNs. The proposed protocol assumes that
OSN service providers can have available a cryptographic PKG.
At registration time, when a person registers as a new user, it
chooses a unique username and supplies it to the PKG. Based
on the unique identifier (ID), the PKG generates a cryptographic
ID by applying a special hash function. Using the cryptographic
ID, the PKG generates a private secret that is derived from the
PKG’s master secret combined with pairing-based cryptography.
With only one secret, a user can dynamically generate a random
session key using its neighbor nodes. Thus, the proposed protocol
substantially reduces key storage.
2. Related works
Practical concerns are driving scientific research in this area
because Sybil attacks are putting OSNs at serious risk [1]. Attacks
that involve large numbers of fake accounts are quite common
in practice, and overwhelm networks with malicious content and
challenge their integrity [22–24]. Numerous Sybil attack defenses
have been devised in response to the growing frequency of these
attacks. The solutions can be conveniently classified in two broad
categories based on the type of defensive mechanism they use.
One group of solutions attempts to identify the Sybils present in
the network [9,10,13,25–30], whereas the other attempts to set
the accepted level and maintain the number of Sybils under this
threshold [17,31,32]. Most solutions that inspired this work are
from the former group, and seek to devise ways to identify Sybil
users so that they can be removed from OSNs by administrators.
Solutions based on Sybil identification start from the following two
central assumptions:
• The graphical representation of the network contains re-
gions that are populated by legitimate accounts and bot
accounts, with a moderate amount of linking between them.
This means that the edges between the two regions are
weaker compared with those that span the inside of the
bot region. However, this claim cannot be automatically
accepted in real-world networks, which is why this principle
has limited value.
• The region populated by legitimate users is subject to rapid
reconfiguration, with the initial configuration changing as
the connections develop. This also cannot be fully proven
in real-world systems because networks are shaped by ran-
dom events and spread at great speed, and research into
the reconfiguration speed cannot appropriately account for
such intense dynamic growth and lacks the metrics to cap-
ture it.
Another group of techniques exists that place less emphasis on
the identification of individual users as legitimate or Sybils, but
rather on stopping Sybils from multiplying by opening additional
accounts. Even if solutions from this group change the defensive
mechanism, they also rely on the first assumption listed above for
their validity. To ensure that the system remains in a dynamic state
and that Sybil-initiated connections remain under the limit, these
solutions introduce a new principle that involves the distribution
 M. Al-Qurishi et al. / Future Generation Computer Systems 84 (2018) 139–148
of credits between various users. In sharp contrast to the first two
strategies, this strategy takes into account the previous behavior
of the users in question, in addition to past trends typical for the
network that needs to be defended, thus enhancing the precision
of Sybil identification to a considerable degree. The solutions oper-
ating on this basis can still be vulnerable if the principle does not
hold true, and the credit distribution method can be subject to a
denial-of-service type of attack, which causes the influence to be
unjustly assigned to Sybils as a consequence [33].
Two crucial aspects of any Sybil attack solution intended for
macro systems that must be balanced with other advantages are
efficiency and scalability [34]. There are solutions that display
solid performance, but require massive processing power because
of a large number of operations, which prevents their practical
use on systems that feature millions of individual users. Other
methods are sufficiently scalable, but have the tendency to return
too many false positives, which reduces their trustworthiness.
With a full understanding of all other methods, our work promotes
cryptanalysis based on pairings as an innovative mechanism for the
identification and elimination of Sybils from large social networks.
Every user that attempts to become a member of certain groups is
prompted to perform a security task (i.e., pass through a ‘trapdoor’)
to be confirmed as genuine. To complete the task, the user must
present authentic credentials, together with a confidential code
received from an administrator. Because accounts controlled by
Sybils cannot gain access to the confidential code, they are stopped
at this line of defense and cannot acquire membership to the
desired group. In effect, a large-scale invasion of OSNs by bots is
avoided by the application of this procedure.
3. Preliminaries
The procedure proposed in this paper depends on the following
theoretical positions and calculations.
3.1. Bilinear mapping
If G1 is defined as an addition-based group and G2 is defined
as a multiplication-based group that belongs to the same order
q, we can also use P to define a random initiator of G1 and aP as
the measure that describes the value of the initiator multiplied a
times. Under such conditions, G1 can be considered as a collection
of points that are grouped along an elliptic curve Fq , whereas G2 can
be considered to be a part of the enclosed area Fqk given parameter
k ∈ Zq∗ = {k|1 ≤ k ≤ q − 1}. If those parameters are combined
to form G1 × G1 → G2 , the resulting procedure can be described
as a cryptographic bilinear mapping. The resulting map has the
following features:
• Bilinearity: e(aP , bQ ) = e(P , Q )ab in the case in which
P , Q ∈ G1 and a, b ∈ Zq∗ . The same can also be expressed by
claiming that P , Q , R ∈ G1 , then e(P + Q , R) = e(P , R)e(Q , R)
and e(P , Q + R) = e(P , Q )e(P , R)
• Non-degeneracy: If P is a generator of G1 , then it must be
true that e(P , P) generates G2 Otherwise, it can be stated that
e(P , P)can never be equal to one.
• Computable: When e(P , P) can be calculated in polynomial
time, then for each value of P , Q ∈ G1 , the map can be
calculated in polynomial time.
3.2. Diffie–Hellman (DH) challenges
The network G1 that was described in the previous subsection
can be analyzed through a series of cryptographic challenges that
include the following:
141
• Discrete logarithm (DL) challenge: If we assume that P , Q ∈
G1 , the task is to identify an entire number n that conforms
to P = nQ in all cases.
• Computational DH (CDH) challenge: With defined elements
(P , aP , bP ∈ G1 ), the task is to calculate the value of abP.
• Decision-based DH challenge (DDH): With defined elements
(P , aP , bP , cP ∈ G1 ), where a, b, c ∈ Zq∗ , the task is to
determine whether it is true that c = ab mod q.
• Gap-based DH challenge (GDH): A type of DH challenge that
includes a CDH of high difficulty and DDH of low difficulty.
• Dual-linear DH challenge (DLDH): With defined elements
(P , aP , bP , cP ∈ G1 ), where a, b, c ∈ Zq∗ , the task is to
calculate the value of e(P , P)abc .
3.3. Connection patterns in OSNs
The basic outline of the OSN’s internal structure and defense
objectives needs to be formulated in such a way that demonstrates
that Sybil intrusions can be stopped in a centralized manner in
realistic circumstances. In real a social network, there are two types
of graphs: social and interaction. A social graph is represented by
normal follower or friend relationships, such as on Twitter and
Facebook, or a subscribing relation, such as on YouTube. By con-
trast, the interaction graph is represented by actions and behaviors,
such as replying, retweeting and mentioning someone on Twitter,
or commenting on a video or channel on YouTube.
3.3.1. Social graph
Generally, it can be safely assumed that any OSN includes a mul-
titude of proactive nodes, which can mutually interact by forming
connections that symbolize contact, familiarity or a friendly rela-
tionship. The structure of relationships is largely inherited from the
offline environment, in which people are connected through mem-
bership in a family, business entity or social club. Every interaction
between two users can be graphically summarized as an edge, with
users that share a friendly relationship on the network connected
by very strong links.
The graph used to visualize the network is centered on users
and projects them into two distinct node types: H-users are rare
but strongly linked individuals who are represented as central
nodes, and numerous L-users are followers of H-users, have fewer
contacts and are projected as peripheral nodes. It is possible for the
same user to be centrally represented in one situation and periph-
erally represented in another. In realistic systems, H-users typically
facilitate segmentation, with segments forming around them. It is
possible that malicious intruders could attempt to take advantage
of this feature and emulate or corrupt an H-user’s account. Even
if only an L-user is compromised, the intruder can use personal
data or existing connections to propagate its goals. It should also
be noted that the total number of H-users is comparatively small
when compared with the number of L-users, which makes the
structure of all networks asymmetric. Every user in the network
has a specific identification code, which is issued by the central
authority of the network (see Fig. 1).
A. Segmentation of networks: Once users become officially ac-
cepted by system administrators, they naturally gravitate towards
one segment of the network where their contacts are located. To
streamline this analysis, we allow each H-user to interact with
adjacent H-users, either directly or through intermediaries. Once
the network starts fragmenting, H-users typically act as center-
pieces of new segments, whereas their followers group around
them. Thus, it can be claimed that the activity of H-users is largely
responsible for the structural shape of the OSN.
B. Linking through the network: The total weight of all edges
running inside a single segment of the graph is greater than that of
142 M. Al-Qurishi et al. / Future Generation Computer Systems 84 (2018) 139–148
Fig. 1. Cluster formation in an online social network.
the edges connecting H-users with the remainder of the segment.
In the intervening time, edges that link multiple H-users can be
considered to form the highway running through the entire graph,
and for the purpose of this analysis, are considered to be represen-
tative of the interactions between two segments.
Sending messages between users can occur through two chan-
nels: internally within the single segment as L-users interact with
central users through an array of other followers, or between
segments as H-users accumulate information from their segment
before resending it to a corresponding central user via the infor-
mation highway.
3.3.2. Interaction graph
The interaction graph is a topic or event-driven graph that can
be constructed according to some event or a trend in a specific
time. The interaction graph can create a difference between active
users and those that simply associate with them merely based on
their name. For example, on YouTube, commenting on or even
replying to comments on a user’s video or channel, and on Twitter,
retweeting, mentioning or replying to tweets are all actions that
can create an interaction graph. These actions do not require any
social connection between the actors; they require only a Gmail or
Twitter account. These types of interactions are more vulnerable
to Sybil attacks, especially targeted attacks. Fig. 4 shows a retweet
graph constructed from almost 1000 nodes on Twitter regarding
the ‘‘USA elections 2016’’ topic.
4. Proposed methodology
In this section, we provide a description of the proposed key
agreement protocol. It is assumed that the provider of an OSN has a
PKG server that is responsible for providing a private secret to each
user during the registration phase. Table 1 describes the notation
used in the remainder of this paper.
4.1. Key exchange algorithm
The PKG server generates the following parameters:
• selects two random groups G1 and G2 that have the same
prime order q, where G1 is an additive group and G2 is a
multiplicative group;
• selects a generator P of G1 and bilinear map fˆ : G1 × G2 → G2
that maps from the Cartesian product of G1 and G2;
• determines two collision resistant cryptographic hash func-
tions Ĥ 1 and Ĥ 2 , where Ĥ1 : {0, 1}∗ → G1 , that is, a
mapping from arbitrary-length strings to points in G1 and
Ĥ2 : {0, 1}∗ → {0, 1}β , that is, a mapping from arbitrary-
length strings to β -bit fixed length output; and
• determines the signature function Sigi (.) for entity i .
4.1.1. Key generation
The PKG of the provider in an OSN selects m+1 random num-
bers η0 , η2 , η3 , . . . , ηm ∈ Zq∗ and generates a polynomial h(x) of
degree m, where h(x) = η0 + η1 x + η2 x2 + · · · + ηm xm ∈
Zq . The PKG then computes V0 = η0 P V1 = η1 P , . . . ., Vm =
ηm P. The public parameters of the system published by the PKG
are {G1 , G2 , fˆ , Ĥ1 , Ĥ2 , P , V , V 1 , V2 , . . . , Vm }and the PKG stores its
secret keys, which are {η0 , η1 , η2 , . . . , ηm }. In addition to the above
computations, the PKG also computes λi = h(IDi ) = η0 + η1 IDi +
η2 (IDi )2 + · · · + ηm (IDi )m mod q, where user UI is identified by its
unique real identity IDRI , such that IDi = Ĥ1 (IDRI ) ∈ Zq∗ . During
the registration phase, the user creates its real public identity and
sends the identity to the provider. The PKG generates a private
secret that corresponds to the public identity of the user and sends
it back to the user through a private channel. The public and private
keys of the user can be computed by the PKG as follows:
• For user UI :
– public key:IDI = Ĥ1 (IDRI ); secret key: λI = h(IDI ).
Example users are as follows:
• For user Alice:
– public key:IDA = Ĥ1 (IDALICE ); secret key: λA = h(IDA ).
• For user Bob:
– public key:IDB = Ĥ1 (IDBOB ); secret key: λB = h(IDB ).
• For user Eve:
– public key:IDE = Ĥ1 (IDEVE ); secret key: λE = h(IDE ).
4.1.2. Key agreement and authentication
To establish a shared secret session key between two neigh-
bor nodes (user, follower, friend), each node computes the fol-
lowing public parameters and sends them to the other nodes.
The user chooses a random rUS ∈ Zq∗ , computes the pub-
∑m
lic parameters YUS = rUS P and θUS = j
j=0 (IDUS ) Vj =
λUS P, generates signature SigUS (YUS ), and then sends the public
parameters and signature to its neighbor nodes (i.e., friends and/or
followers). The friend node chooses a random rFR ∈ Zq∗ , computes
∑m
the public parameters YFR = rFR P and θFR = j=0 (IDFR ) Vj = λFR P,
j
generates signatureSigFR (YFR ), and then sends the public parame-
ters and signature to the user. Similarly, the follower node chooses
a random rFL ∈ Zq∗ , computes the public parameters YFL = rFL P
∑m
and θFL = j=0 (IDFL ) Vj = λFL P, generates signature SigFL (YFL ), and
j
 M. Al-Qurishi et al. / Future Generation Computer Systems 84 (2018) 139–148 143

Table 1
Notation used in the proposed protocol.
Notation Description
Zq∗ A modular multiplicative set of order q
G1 As an additive group of prime order p
G2 As a multiplicative group of prime order p
fˆ : G1 × G2 → G2 Which maps from the Cartesian product between G1 to G2.
Ĥ 1 A hash function which is mapping from arbitrary-length strings to
points in G1
Ĥ 2 , A hash function which is a mapping from arbitrary-length strings to
β -bit fixed length output
Sigi (.) Signature function
h(x) A polynomial of degree m,
η0 , η2 , η3 , . . . , ηm ∈ Zq∗ Random secrets for PKG ∈ Zq∗
IDRI , Real ID of the nodes
IDi Cryptographic Mapped ID from the real Ids
V0 , V1 , V2 , . . . , Vm Cryptographic public parameters derived from the PKG’s private
combining with the nodes public IDs
λI Secret parameters for user I
IDA , IDB , IDE Public key for Alice, Bob and Eve respectively
θ, Y , r , λ, P Public parameters to calculate shared secret session key KUS −FR , KFR−US
between user and his friends or followers.

then sends the public parameters and signature to the user. Finally,
each node computes its pairwise secret session key as follows

• the shared secret key between the user and friend (cal-
culated by the user):KUS −FR = fˆ (θFR , rUS YFR )λUS = fˆ
(λFR P , rUS rFR P)λUS = fˆ (P , P)λUS λFR rUS rFR ; and
• the shared secret key between the user and friend (cal-
culated by the friend):KFR−US = fˆ (θUS , rFR YUS )λFR = fˆ
(λUS P , rFR rUS P)λFR = fˆ (P , P)λFR λUS rFR rUS .

Thus, KUS −FR = KFR−US

4.1.3. Consistency
In this section, we prove that the user and friend can generate
the same session key without knowing each other’s private secret
component. The user only collects the friend’s public components
θFR and YFR combines the user’s own private secret λUS with the ses-
sion dependent random rUS and generates the session key KUS −FR .
By contrast, the friend only collects the user’s public components
θUS and YUS combines its own private secret λFR with the session
dependent random rFR and generates the session key KFR−US . In the
following derivation, we show that KUS −FR = KFR−US :
KUS −FR
= fˆ (θFR , rUS YFR )λUS
m

(IDFR )j Vj , rUS YFR )λUS

∑
= fˆ (
j=0
= fˆ (V0 + (IDFR )V1 + (IDFR )2 V2 + · · · + (IDFR )m Vm , rUS rFR P)λUS
= fˆ (η0 P + η1 (IDFR )P + η2 (IDFR )2 P + · · · + ηm (IDFR )m P , P)λUS rUS rFR
= fˆ ([η0 + η1 IDFR + η2 (IDFR )2 + · · · + ηm (IDFR )m ]P , P)λUS rUS rFR
= fˆ (h(IDFR )P , P)λUS rUS rFR
= fˆ (λFR P , P)λUS rUS rFR
= fˆ (P , P)λFR λUS rFR rUS
= KFR−US .
Similarly, any two party combinations (user/follower, follower/user,
follower/friend and friend/follower) can generate the session key
between them.

4.2. Authentication between the ‘‘H’’ nodes and ‘‘L’’ nodes 5. Result analysis

The following algorithms illustrate the authentication process

between the ‘‘H’’ nodes and ‘‘L’’ nodes. The H node and L node were In this section, we discuss key storage for both types of graphs
described in Section 2.3. for OSNs.
144 M. Al-Qurishi et al. / Future Generation Computer Systems 84 (2018) 139–148

Fig. 2. Social graph for 10,000 nodes in the Twitter network.

Fig. 3. Social graph: key space.

5.1. Key storage for the social graph of 100 and maximum of 1000 connected nodes. The corresponding
key space is illustrated in Fig. 3. In the social graph representation,
We conducted experiments on Twitter and YouTube datasets. the proposed methodology provides very efficient key storage, and
Fig. 2 shows the social graph for 10,000 nodes, with a minimum thus provides effective key management. Analyzing the key space
 M. Al-Qurishi et al. / Future Generation Computer Systems 84 (2018) 139–148
Fig. 4. Interaction graph for 1007 nodes in the Twitter network.
graph, we can observe the number of nodes in the network and
number of nodes that are connected as neighbor nodes; there is
an effect of key storage and managing the key. Only one secret
key is required for the registered user at registration time, and
subsequently, the node dynamically generates the session key with
its neighbor nodes and communicates securely. The key generation
and authentication process were discussed in the previous sec-
tion.
5.2. Key storage for the interaction graph
The interaction graph for 1007 nodes with a minimum of two
actions and maximum of 118 actions is shown in Fig. 4, and the
corresponding key spaces are shown in Figs. 5 and 6. In social
networks such as Twitter and YouTube, users interact with each
other using different types of actions, for example, in the Twitter
network, a user might reply, mention or retweet.
Considering the interaction graph, the maximum key space is
required when users perform any action, which is a maximum
of three per user, because there are three types of action: reply,
mention and retweet. The maximum key space is illustrated in
Fig. 5 and the minimum key space is illustrated in Fig. 6.
6. Attack analysis
An OSN is susceptible to different types of attack; thus,
the proposed protocol must be robust against the following
attacks.
Tunneling attack: This type of attack is known as a ‘wormhole’ at-
tack in the research domain. This type of system invasion includes
the formation of a packet in two separate areas of the network, and
creation of a tunnel between them. The same packet is then sent all
over the section of the network that is infiltrated by the attackers.
This principle can be illustrated by introducing points P1 and P2
connected by a tunnel, followed by their effort to persuade nearby
145
users Q1 , Q2 , Q3 and so on, to transmit data through the tunnel, thus
falling under the influence of the attackers. Our solution eliminates
this problem by requiring any network point to be authenticated
before it can be described as a trusted member, which prevents the
intruders from invading data paths and acting as intermediaries,
and disables this type of attack.
Masquerade attack: The central premise of this strategy is that
the invader attempts to emulate the behavior of regular network
points and participates in the normal flow of data in an attempt to
redirect the routes to its advantage. Such behavior could change
the data accumulated from all users, and in certain situations,
could even bring down the entire system. For example, if point P1
assumes the identity of S1 , it could conceivably corrupt the data
acquired by Q1 , Q2 , Q3 and any other nearby nodes. Our solution re-
quires all network points to validate each other as communication
is established between them, which effectively blocks the infected
nodes from sending invalid data through the network. Because of
this requirement, a masquerade attack cannot be performed when
the solution is deployed.
Reply attack: In a similar setting as the previous attack, the in-
truder can recycle an old password or validation sequence to gain
access to the network and become involved in data exchange,
attempt to trick legitimate network points and become accepted
as one of them. This cannot happen with our solution for two
reasons: the invader would be unable to complete the validation
procedure, and old passwords would be useless because all codes
are based on a randomized value that is calculated dynamically
every time a new session is initiated, including both validations
(Ver0 , Ver1 , Aut0 , Aut1 ). Because of this security feature, repeating a
once-used key does not result in a successful validation because all
parameters are reset every time the procedure is run. Additionally,
traffic within the network is cryptographically protected, which
prevents any possibility of a replay attack occurrence.
Data interruption attack: With this approach, the intruders
damage, change or falsify data traffic, thus affecting the entire
146 M. Al-Qurishi et al. / Future Generation Computer Systems 84 (2018) 139–148

Fig. 5. Interaction graph: maximum key space.

Fig. 6. Interaction graph: minimum key space.

communication flow in a negative manner. For such an operation
to be executed, the invaders need to infiltrate the message ex-
change chain and play an active part in passing the data pack-
ets along. If the intruder manages to change the content of the
packet, this can result in a lost message somewhere down the
line. However, our solution forces the intruder to undergo the
validation procedure before gaining access to the message flow;
thus, it becomes impossible for the intruder to come into contact
with the data and corrupt it. This practical mechanism prevents
data interruption attacks.
 M. Al-Qurishi et al. / Future Generation Computer Systems 84 (2018) 139–148
7. Conclusion
Real-life OSN users have to manage the issue of bot attacks. If
intruders successfully compromise a participating user, they could
extend their influence to numerous other users spread through-
out the network. Multiple approaches to this issue have been
suggested, and we introduced a novel mechanism that efficiently
interferes with an intrusion by bots. This mechanism depends on
the installation of a roadblock based on cryptographic challenges
that stops Sybils and allows legitimate users to pass through.
Because bots are unable to become members of a group, they
cannot propagate the objectives of the intruders. The proposed
mechanism was tested on a realistic data sample, and its defensive
features were optimized to serve its intended purpose.
Acknowledgment
The authors are grateful to the Deanship of Scientific Research
at King Saud University for funding this paper through the Vice
Deanship of Scientific Research Chairs.
References
[1] M. Al-Qurishi, M. Al-Rakhami, A. Alamri, M. Alrubaian, S.M.M. Rahman, M.S.
Hossain, Sybil defense techniques in online social networks: a survey, IEEE
Access 5 (2017) 1200–1219.
[2] S. Misra, A.S.M. Tayeen, W. Xu, SybilExposer: An effective scheme to detect
Sybil communities in online social networks, in: Communications (ICC), 2016
IEEE International Conference on, 2016, pp. 1–6.
[3] J.R. Douceur, The sybil attack, in: International Workshop on Peer-to-Peer
Systems, 2002, pp. 251–260.
[4] P. Vijayakumar, M. Azees, A. Kannan, L. Jegatha Deborah, Dual authentication
and key management techniques for secure data transmission in vehicular ad-
hoc networks, IEEE Trans. Intell. Transp. Syst. 17 (4) (2015) 1015–1028.
[5] K. Regan, 10 Amazing Social Media Growth Stats From 2015. Available: http:
//www.socialmediatoday.com/social-networks/kadie-regan/2015-08-10/10-
amazing-social-media-growth-stats-2015.
[6] Z. Yang, J. Xue, X. Yang, X. Wang, Y. Dai, VoteTrust: Leveraging friend invitation
graph to defend against social network sybils, 2015.
[7] H. Yu, Sybil defenses via social networks: a tutorial and survey, ACM SIGACT
News 42 (2011) 80–101.
[8] B. Viswanath, A. Post, K.P. Gummadi, A. Mislove, An analysis of social network-
based sybil defenses, ACM SIGCOMM Computer Commun. Rev. 41 (2011)
363–374.
[9] H. Yu, M. Kaminsky, P.B. Gibbons, A.D. Flaxman, Sybilguard: defending against
sybil attacks via social networks, IEEE/ACM Trans. Netw. 16 (2008) 576–589.
[10] H. Yu, P.B. Gibbons, M. Kaminsky, F. Xiao, Sybillimit: a near-optimal social
network defense against sybil attacks, IEEE/ACM Trans. Netw. 18 (2010) 885.
[11] P. Vijayakumar, R. Naresh, L. Jegatha Deborah, S.K. Hafizul Islam, Computation
cost efficient group key agreement protocol for secure peer to peer communi-
cation, Secur. Commun. Netw. 9 (17) (2016) 3952–3965.
[12] A. Mohaisen, A. Yun, Y. Kim, Measuring the mixing time of social graphs, in:
Proceedings of the 10th ACM SIGCOMM conference on Internet measurement,
2010, pp. 383–389.
[13] Q. Cao, M. Sirivianos, X. Yang, T. Pregueiro, Aiding the detection of fake
accounts in large scale social online services, in: Proceedings of the 9th
USENIX conference on Networked Systems Design and Implementation, 2012,
pp. 15–15.
[14] Z. Yang, C. Wilson, X. Wang, T. Gao, B.Y. Zhao, Y. Dai, Uncovering social network
sybils in the wild, in: ACM Transactions on Knowledge Discovery from Data,
TKDD, Vol. 8, 2014, p. 2.
[15] G. Wang, T. Konolige, C. Wilson, X. Wang, H. Zheng, B.Y. Zhao, You are how you
click: Clickstream analysis for sybil detection, in: Proc. USENIX Security, 2013,
pp. 1–15.
[16] G. Wang, M. Mohanlal, C. Wilson, X. Wang, M. Metzger, H. Zheng, (2012) Social
turing tests: Crowdsourcing sybil detection, arXiv preprint arXiv:1205.3856,
2012.
[17] D.N. Tran, B. Min, J. Li, L. Subramanian, Sybil-Resilient Online Content Voting,
in: NSDI, 2009, pp. 15–28.
[18] A. Mislove, A. Post, P. Druschel, P.K. Gummadi, Ostra: Leveraging trust to
thwart unwanted communication, in: NSDI, 2008, pp. 15–30.
147
[19] R. Tso, T. Okamoto, T. Takagi, E. Okamoto, An ID-based non-interactive tripar-
tite key agreement protocol with K-resilience, communications and computer
networks, 2005, 38–42.
[20] Meng-Hui Lim, Sanggon Lee, An improved one-round ID-based tripartite au-
thenticated key agreement protocol, IACR Cryptology ePrint Archive 2007,
2007, p. 189.
[21] M.S. Hossain, G. Muhammad, Sk.M.M. Rahman, W. Abdul, A. Alelaiwi, A.
Almari, Towards end-to-end biometrics-based security for IoT infrastructure,
IEEE Wirel. Commun. Mag. 23 (5) (2016) 45–51.
[22] A. Chinchore, F. Jiang, G. Xu, Intelligent sybil attack detection on abnormal
connectivity behavior in mobile social networks, in: Knowledge Management
in Organizations, Springer, 2015, pp. 602–617.
[23] M. Al-Qurishi, R. Aldrees, M. AlRubaian, M. Al-Rakhami, S.M.M. Rahman, A.
Alamri, A new model for classifying social media users according to their
behaviors, in: Web Applications and Networking, WSWAN, 2015 2nd World
Symposium on, 2015, pp. 1–5.
[24] X. Yang, et al., Automatic visual concept learning for social event understand-
ing, IEEE Trans. Multimedia 17 (3) (2015) 46–58.
[25] H. Zhang, J. Zhang, C. Fung, C. Xu, Improving sybil detection via graph pruning
and regularization techniques, in: Proceedings of The 7th Asian Conference on
Machine Learning, 2015, pp. 189–204.
[26] W. Wei, F. Xu, C.C. Tan, Q. Li, Sybildefender: a defense mechanism for sybil
attacks in large social networks, IEEE Trans. Parallel Distrib. Syst. 24 (2013)
2492–2502.
[27] L. Shi, S. Yu, W. Lou, Y.T. Hou, Sybilshield: An agent-aided social network-based
sybil defense among multiple communities, in: INFOCOM, 2013 Proceedings
IEEE, 2013, pp. 1034–1042.
[28] Q. Cao, X. Yang, Sybilfence: Improving social-graph-based sybil defenses with
user negative feedback, arXiv preprint arXiv:13043819, 2013.
[29] F. Li, P. Mittal, M. Caesar, N. Borisov, Sybilcontrol: practical sybil defense with
computational puzzles, in: Proceedings of the seventh ACM workshop on
Scalable trusted computing, 2012, pp. 67–78.
[30] G. Danezis, P. Mittal, SybilInfer: detecting sybil nodes using social networks,
in: NDSS, 2009.
[31] N. Tran, J. Li, L. Subramanian, S.S. Chow, Optimal sybil-resilient node admission
control, in: INFOCOM, 2011 Proceedings IEEE, 2011, pp. 3218–3226.
[32] B. Viswanath, M. Mondal, K.P. Gummadi, A. Mislove, A. Post, Canal: Scaling
social network-based Sybil tolerance schemes, in: Proceedings of the 7th ACM
european conference on Computer Systems, 2012, pp. 309–322.
[33] M. AlRubaian, M. Al-Qurishi, S.M.M. Rahman, A. Alamri, A novel prevention
mechanism for Sybil attack in online social network, in: Web Applications and
Networking (WSWAN), 2015 2nd World Symposium on, 2015, pp. 1–6.
[34] P. Vijayakumar, Victor. Chang, L. Jegatha Deborah, Balamurugan. Balusamy,
P.G. Shynu, Computationally efficient privacy preserving anonymous mutual
and batch authentication schemes for vehicular ad hoc networks, Future
Gener. Comput. Syst. (2017) in press. https://doi.org/10.1016/j.future.2016.1
1.024.
Muhammad Al-Qurishi is a Ph.D. candidate in the Information Systems Depart-
ment in the College of Computer and Information Sciences, King Saud University,
Riyadh, Kingdom of Saudi Arabia. He received his master’s degree in information
systems from King Saud University, Kingdom of Saudi Arabia. He has published sev-
eral papers in refereed IEEE/ACM/Springer journals and conferences. His research
interests include online social networks, social media analysis and mining, human-
computer interaction, and health technology.
Sk Md Mizanur Rahman is an Assistant Professor in In-
formation System Department in the College of Computer
and Information Sciences at King Saud University, KSA.
Prior to his current appointment, he worked for several
years in cryptography and security engineering in the
high-tech industry in Ottawa, Canada. He also worked as
a postdoctoral researcher for several years in University
of Ottawa, University of Ontario Institute of Technology
(UOIT), and University of Guelph, Canada. He completed
a Ph.D. in Risk Engineering (Major: Cyber Security En-
gineering) in the Laboratory of Cryptography and Infor-
mation Security, Department of Risk Engineering, University of Tsukuba, Japan,
on March 2007. Information Processing Society Japan (IPSJ) awarded Dr. Rahman
with ‘‘IPSJ Digital Courier Funai Young Researcher Encouragement Award" for his
excellent contribution in IT security research. He awarded with ‘‘Gold Medal" for
the distinction marks in his undergraduate and graduate program. Primary research
interests of Dr. Rahman are Cryptography, Software and Network Security, Privacy,
148 M. Al-Qurishi et al. / Future Generation Computer Systems 84 (2018) 139–148

Cloud-security, Sensor Network Security, White box-cryptography, and Internet of Atif Alamri is an Associate Professor in the Information
Things security. Systems Department at the College of Computer and In-
formation Sciences, King Saud University. Riyadh, Saudi
Arabia. His research interests include multimedia-assisted
health systems, ambient intelligence, and service-oriented
M. Shamim Hossain (SM’09) is an Associate Professor at the King Saud Univer-
architecture. Dr. Alamri was Guest Associate Editor of the
sity, Riyadh, KSA. Dr. Shamim Hossain received his Ph.D. in Electrical and Com-
IEEE Transactions on Instrumentation and Measurement,
puter Engineering from the University of Ottawa, Canada. His research interests
a co-chair of the first IEEE International Workshop on Mul-
include serious games, social media, IoT, cloud and multimedia for healthcare,
timedia Services and Technologies for E-health, a technical
smart health, and resource provisioning for big data processing on media clouds.
program co-chair for the 10th IEEE International Sympo-
He has authored and coauthored around 120 publications including refereed
sium on Haptic Audio Visual Environments and Games,
IEEE/ACM/Springer/Elsevier journals, conference papers, books, and book chapters.
and serves as a program committee member for many conferences in multimedia,
He has served as a member of the organizing and technical committees of several
virtual environments, and medical applications.
international conferences and workshops. He has served as co-chair, general chair,
workshop chair, publication chair, and TPC for over 12 IEEE and ACM conferences
and workshops. Currently, he serves as a co-chair of the 6th IEEE ICME workshop
Mabrook AlRakhami is a Ph.D. student in the Computer
on Multimedia Services and Tools for E-health MUST-EH 2016. He is a recipient
Sciences Department in the College of Computer and Infor-
of a number of awards including, the Best Conference Paper Award, the 2016
mation Sciences, King Saud University, Riyadh, Kingdom
ACM Transactions on Multimedia Computing, Communications and Applications
of Saudi Arabia. He received his masters degree in infor-
(TOMM) Nicolas D. Georganas Best Paper Award, and the Research in Excellence
mation systems from King Saud University, Kingdom of
Award from King Saud University. He is on the editorial board of IEEE Access,
Saudi Arabia. He has authored several papers in refereed
Computers and Electrical Engineering (Elsevier), Games for Health Journal and
IEEE/ACM/Springer conferences and journals. His research
International Journal of Multimedia Tools and Applications (Springer). Previously,
interests include social networks, cloud computing, and
he served as a guest editor of IEEE Transactions on Information Technology in
health technology.
Biomedicine (currently JBHI), International Journal of Multimedia Tools and Appli-
cations (Springer), Cluster Computing (Springer), Future Generation Computer Sys-
tems (Elsevier), Computers and Electrical Engineering (Elsevier), Sensors (MDPI).
and International Journal of Distributed Sensor Networks. Currently, he serves as
a lead guest editor of IEEE Communication Magazine, IEEE Transactions on Cloud
B. B. Gupta received Ph.D. degree from Indian Institute of
Computing, Future Generation Computer Systems (Elsevier), IEEE and Access. Dr.
Technology Roorkee, India in the area of Information and
Shamim is a Senior Member of IEEE, a member of ACM and ACM SIGMM.
Cyber Security. He has published more than 90 research
papers (including 03 book and 14 chapters) in Interna-
tional Journals and Conferences of high repute including
IEEE, Elsevier, ACM, Springer, Wiley Inderscience, etc. He
Ahmad Almogren obtained his Ph.D. in Computer Sci-
has visited several countries, i.e. Canada, Japan, China,
ences from Southern Methodist University, Dallas, Texas,
Malaysia, Hong-Kong, etc. to present his research work.
USA in 2002. Previously, he served as the Dean of Com-
His biography was selected and publishes in the 30th Edi-
puter College and the Head of the Academic Accreditation
tion of Marquis Who’s Who in the World, 2012. He is also
Council at Al Yamamah University. Presently, he is the
working principal investigator of various R&D projects. He
Vice Dean for Development and Quality at the college of
is serving as associate editor of IEEE Access, Associate editor of IJICS, Inderscience
Computer and Information Sciences College at King Saud
and Executive editor of IJITCA, Inderscience, respectively. He is also serving as
University in Saudi Arabia. His research areas of interest
reviewer for Journals of IEEE, Springer, Wiley, Taylor & Francis, etc. Currently he
include networking, mobile computing, security and data
is guiding 10 students for their Masters and Doctoral research work in the area of
consistency.
Information and Cyber Security. He is also serving as guest editor of various reputed
Journals. Dr. Gupta is also holding position of editor of various International Journals
and magazines. He has also served as Technical program committee (TPC) member
of more than 20 International conferences worldwide. Dr. Gupta is member of IEEE,
Majed Al-Rubaian is a Ph.D. candidate in the Information
ACM, SIGCOMM, The Society of Digital Information and Wireless Communications
Systems Department in the College of Computer and Infor-
(SDIWC), Internet Society, Institute of Nanotechnology, Life Member, International
mation Sciences, King Saud University, Riyadh, Kingdom of
Association of Engineers (IAENG), Life Member, International Association of Com-
Saudi Arabia. He received his master’s degree in informa-
puter Science and Information Technology (IACSIT). He was also visiting researcher
tion systems from King Saud University, Kingdom of Saudi
with Yamaguchi University, Japan in January, 2015. His research interest includes
Arabia. He has authored several papers in refereed IEEE/
Information security, Cyber Security, Mobile security, Cloud Computing, Web secu-
ACM/Springer journals and conferences. He is a student
rity, Intrusion detection, Computer networks and Phishing.
member of the ACM and the IEEE. His research interests
include social media analysis, data analytics and mining,
social computing, information credibility, and cyber secu-
rity.