Interaction-based Reputation Model in Online Social Networks

Izzat Alsmadi

, Dianxiang Xu

and Jin-Hee Cho

Department of Computer Science, University of New Haven, West Haven, CT, U.S.A.

Department of Computer Science, Boise State University, Boise, ID, U.S.A.

US Army Research Laboratory, Adelphi, MD, U.S.A.

Keywords:

Online Social Networks, Reputation, User Attributes, Privacy, Information Credibility.

Abstract:

Due to the proliferation of using various online social media, individual users put their privacy at risk by

posting and exchanging enormous amounts of messages or activities with other users. This causes a seri-

ous concern about leaking private information out to malevolent entities without users’ consent. This work

proposes a reputation model in order to achieve efﬁcient and effective privacy preservation in which a user’s

reputation score can be used to set the level of privacy and accordingly to determine the level of visibility

for all messages or activities posted by the users. We derive a user’s reputation based on both individual and

relational characteristics in online social network environments. We demonstrate how the proposed reputation

model can be used for automatic privacy assessment and accordingly visibility setting for messages / activities

created by a user.

1 INTRODUCTION

As the use of online social network (OSN) appli-

cations becomes more and more popular than ever,

many people share their daily lives by posting stories,

comments, videos, and/or pictures, which may ex-

pose their private / personal information (Mansﬁeld-

Devine, 2008). In such contexts, privacy became a

critical issue because private information can be ex-

posed to the public (e.g., by search engines) without

the user’s consent.

This work proposes a reputation model that eval-

uates an individual’s reputation based on his/her in-

teractions with peers such as friends in OSNs. Rep-

utation is deﬁned as an opinion about an entity, ei-

ther something or someone including a person, orga-

nization, service, etc., based on third parties’ opinions

(e.g., recommendations). In particular, many OSN

sites have different mechanisms to estimate reputa-

tion. For example, Facebook uses the concept of a

friend with three different levels such as close friend,

friends, and acquaintances in order to categorize the

closeness-based friendship. This kind of friends can

be designated by an individual user and does not rep-

resent the degree of friendship based on the amount

of interactions between two individual users. Similar

to web ranking algorithms, some algorithms of scor-

ing friendship use the number of interactions (e.g.,

likes, tags, posts, comments) to calculate the degree

of friendship between two entities (Jones et al., 2013).

An individual user’s reputation can be determined

based on a variety of aspects of the user’s attributes.

One of the important attributes is whether the user

feeds quality information such as correct, accurate,

or credible information. While users are free to

post their opinions, OSNs should promote propa-

gation of credible information to maintain friendly,

healthy OSN environments. To this end, we pro-

pose our reputation model in which reputation is com-

puted based on a user’s personal attributes embrac-

ing his/her background and afﬁliations and relational

attributes including interactions with others and the

reputation levels of the user’s friends.

An example scenario can be as follows. User i

continuously posts highly helpful, relevant informa-

tion in OSNs. When user j continuously interacts

with those posts fed by users i and j will be listed

as one of top contributors for those activities. User

j can list user i as one of his/her top peers. In terms

of user j’s reputation, user i, one of j’s top peers, is

a key contributor to improve user j’s reputation. Ac-

cordingly, user i will be encouraged to post such types

of activities that may attract more interactions with

other peers, leading to increasing her/his reputation

indirectly.

Many OSN applications are equipped with their

Alsmadi, I., Xu, D. and Cho, J-H.

Interaction-based Reputation Model in Online Social Networks.

DOI: 10.5220/0005668302650272

In Proceedings of the 2nd International Conference on Information Systems Security and Privacy (ICISSP 2016), pages 265-272

ISBN: 978-989-758-167-0

265

own algorithms to estimate peer relationships. For in-

stance, Facebook calculates “close friends” or ranks

each friend based on interactions including proﬁle

view, group interactions, the number of likes and/or

recency of interactions. In this work, we aim to pro-

pose a generic model of reputation which can rep-

resent comprehensive aspects of an individual’s trust

based on collected information for statistical calcu-

lation / analysis but without revealing any private

content of an individual such as activity content, or

friends’ identities without the user’s consent. There-

fore, our proposed reputation model is to assess an in-

dividual’s reputation based on his/her individual and

relational attributes while preserving the individual’s

privacy.

This paper has the following contributions:

• We propose a reputation model in which reputa-

tion score is derived from multiple aspects of a

user, particularly, in terms of individual attributes

and relational attributes. This way of estimat-

ing reputation is to ensure the quality of an in-

formation provider by assessing a user’s personal

attribute (e.g., work, education, interest groups)

while assuring information credibility based on

the popularity of activities created by the user. In

addition, by assessing the quality of peers exhibit-

ing active interactions through the user’s activi-

ties, the reputation score of an entity can repre-

sent multidimensional aspects of trustworthiness

in each entity.

• Although an entity’s reputation value is de-

rived from multidimensional aspects of his/her at-

tributes, the proposed reputation model aims to

preserve the user’s privacy by not exposing any

information to any peers without the user’s con-

sent. In particular, since reputation is estimated

based on the amount of interactions but the con-

tent of activities / messages, an entity’s privacy

is well preserved unless the entity permits special

actions to manually set the privacy level (e.g., spe-

cial permission for a spouse to see all the activities

/ messages although he / she is not active on on-

line interactions).

• The proposed reputation model is adaptive to re-

ﬂect the dynamic evolution of an entity’s reputa-

tion. As an entity’s interactions or activities are

updated, the entity’s reputation is updated as well

and accordingly it will be reﬂected in updating

privacy setting for each peer on visibility of ac-

tivities / messages the user posts. In addition, our

reputation model allows human intervention (e.g.,

a user can manually modify privacy setting for a

particular peer) when the automated system set-

ting of privacy is not agreed with the need of a

user. The reputation-based privacy setting can be

an adaptive support mechanism for a user to set

his/her privacy setting more effectively and efﬁ-

ciently.

The rest of this work is organized as follows. Section

2 discusses how an entity’s reputation is derived by

describing each component of reputation. Section 3

describes how the proposed reputation model is ap-

plied to privacy assessment in OSNs. Section 4 gives

an overview of related work in terms of existing mod-

els of trust, reputation, and privacy in OSNs. Section

5 concludes this work and suggests future work direc-

tions.

2 REPUTATION MODEL

The focus of the proposed reputation system is on

how to derive reputation score for a user, either an in-

dividual user or an organization, who has an account

in social network media. In this work, we consider a

graph G where an entity is a vertex, v

and the social

relationship between two entities, i and j, is an edge,

i, j

, as many studies assume. If two entities are con-

nected with an edge, this means the two entities are

friends to each other. But note that the degree of trust

a different entity has towards a same third-party entity

is different, implying the trust relationship is subjec-

tive and asymmetric (Cho et al., 2011).

We estimate an individual user’s reputation score

based on the following three components:

• Personal attributes including occupation, educa-

tion level, afﬁliated groups, favorites, interest, or

political leaning which may represent an entity’s

social identity;

• Personal activities representing the quality and

quantity of interactions with other friends through

posts, likes, tweets, creation of interest-driven

groups;

• Characteristics associated with an individual’s

friends indicating the amount of interactions with

friends and the friends’ reputation levels;

The ﬁrst component is called individual attributes

while the other two components are called relational

attributes. Figure 1 describes what characteristics of

an individual user are considered to derive a reputa-

tion score as discussed above. To be speciﬁc, top L

personal attributes, top M personal activities, and top

K peers are considered to derive the user’s overall rep-

utation. Individual user i’s reputation score, R

, can be

given by:

= α × P

i,L

+ β × P

i,M

+ µ × R

i,K

(1)

ICISSP 2016 - 2nd International Conference on Information Systems Security and Privacy

266

Figure 1: Components of Reputation Score.

i,L

is i’s reputation score towards top L personal

attributes, P

i,M

is i’s reputation score towards top

M personal activities, and R

i,K

represents i’s top K

friends’ reputation scores. Each reputation compo-

nent is weighted by α, β, and µ which can be adjusted

based on the need of a given OSN application. Each

reputation score is computed by

i,L

∑

l=1

i,l

(2)

i,M

∑

m=1

i,m

(3)

i,K

∑

k=1

i,k

for e

i,k

> 0 (4)

Note that e

i,k

> 0 means users i and k are friends to

each other. P

i,l

is attribute l’s score, P

i,m

is activity

m’s score, and R

i,k

is user k’s reputation score where

i has attribute l, activity m, and friend k. The score of

each reputation component is computed as:

• The rank of top personal attributes is derived from

the reputation scores of the OSN websites of i’s

current profession, highest degree institution and

joined social network groups;

• The rank of top personal activities is determined

based on the amount of popular activities a user

has (e.g., popular posts, activities); and

• The rank of top peer’s reputation is calculated

based on reputation scores for peers who interact

most with individual i’s activities.

2.1 Top Personal Attributes

A user entity, either individual user or organization

(e.g., company, universities, or government), may

have its individual online social network page. All

the entities may have their gateways to the different

OSNs in addition to the main entity portal. For some

entities, their social network pages are visited as of-

ten as their main portals by having a great number of

members, friends, or subscribers. Estimated reputa-

tion scores of those entities can be used for several

possible use cases. For example, the rank of a univer-

sity (e.g., academic reputation) may be signiﬁcantly

impacted by the entity reputation score.

If an entity is an individual user (e.g., person, pro-

fessional service provider), we consider three per-

sonal attributes in terms of three categories includ-

ing personal background (e.g., education, occupation,

interest, political leaning), involved activities, and

peers’ activities, as discussed earlier. If an entity is an

organization or service provider, the personal back-

ground is replaced with the attributes and past reputa-

tion of an organization such as a number of employees

(e.g., students, members), quality of employees (e.g.,

acceptance rate for a university), salary levels, and/or

historical reputation (e.g., top ranked for best work

places).

2.2 Top Personal Activities

We consider what kinds of activities an entity

is in-

volved with and how popular the activities are among

its peers. For example, when entity i creates activity

l which receives a large amount of positive feedback

by entity i’s peers, referred by other individual enti-

ties j’s (e.g., likes, favorites, tweets, retweets, pos-

itive comments), entity i’s reputation may increase

due to its popular activity l. Each activity can be

weighted differently depending on importance or spe-

ciﬁc needs. For simplicity, we consider each activity

is weighted equally in this work. More speciﬁcally,

we consider positive popularity as a form of reputa-

tion in which we distinguish negative feedback from

positive feedback. To be speciﬁc, we calculate the

reputation or i’s personal activity l as

i,l

+ I

−

i,l

(5)

i,l

indicates the number of positive feedback and I

−

i,l

is the number of negative feedback on activity l, re-

spectively. The example can be ‘likes’ or ’dislikes’

for posts / activities.

Even if entity i’s peer, j, is listed in entity i’s top

peers, entity j is not necessarily highly active to in-

An entity refers to a user in OSNs. We interchangeably

use an entity or a user to refer to a person or an organization

in OSNs.

Interaction-based Reputation Model in Online Social Networks

267

Figure 2: Activity Rank.

Figure 3: Edge Rank.

volve a high ranked activity l led by entity i. Com-

monly, the highly ranked activity l tends to be liked

by most of peers, rather than a particular set of peers.

Activities are ranked based on the volume of in-

teractions with friends, as shown in Figure 2. In this

model, we assume that all activities and types of in-

teractions with friends have an equal weight.

2.3 Top Peers’ Reputation

We consider user i’s peers’, j’s, reputation in order

to derive user i’s reputation as well. User i assesses

j’s reputation based on the trust i has in j based on

the interactions between i and j. The interactions in-

clude the amount of interactions involved in the activ-

ities created by user i. That is, when user j provides

highly positive feedback towards i’s activity, l, such

as likes, tweets, retweets, favorites, trust, comments,

etc., i’s trust towards j increases. Note that the trust i

estimates towards j is about the relationship between

i and j, not based on j’s personal attributes. Trust is

asymmetric in relationships. Even if j provides a large

amount of highly positive feedback towards i’s activ-

ity, it does not mean user i also provides the exactly

the same amount of positive feedback towards user j’s

activity. There would be a different number of activi-

ties each i or j is involved with; i and j may have dif-

ferent personal disposition such as introvert or extro-

vert which may respond differently towards an activ-

ity. However, a certain level of reciprocal relationship

may exist which can be broadly said ‘friendly’, ‘indif-

ferent’, or ‘hostile’ relationship although the amount

of responsiveness towards each other may not be ex-

actly same. Edge e

i, j

indicates there exists an edge be-

tween i and j. Edge rank of e

i, j

indicates the amount

of interactions between these two users, by increas-

Figure 4: Edge Rank Table with Top Friends Markings.

ing the rank upon interaction. This is demonstrated in

Figure 3 where the thickness of a line between two en-

tities indicates the degree of trust from a trustor (eval-

uator) to a trustee (evaluatee). Note that as the trust

relationship is asymmetric, we distinguish edge e

i, j

from edge e

j,i

Each user may request a monthly report for edge

rank statistics. For simplicity, we represent the active

peer as binary such as active (i.e., existence of interac-

tions) vs. inactive (i.e., no interaction) per the user’s

activity.

3 REPUTATION-BASED PRIVACY

ASSESSMENT

We target our proposed reputation model to be applied

to privacy assessment in various social networking

sites (e.g., Facebook, Twitter, LinkedIn, Snapchat).

Here we discuss a case study for privacy assessment

based on activity logs collected at real time.

Figure 4 shows a dataset of 20 users (i.e., nodes)

that can be collected from a social networking site.

We assume that the social networking site allows

users to create different types of activities (e.g., text,

image, post, video, etc.) and to add other users to their

page as friends such as Facebook. Each user interacts

with his/her friends via activities it creates.

Each column represents three reputation compo-

nents discussed in Section 2, including personal at-

tributes, personal activities, and peers’ reputation.

For this case study, we pick a set of weights for rep-

utation components, α, β, and µ which are set to

0.8, 0.3, and 0.05 respectively, where each weight is

ranged in [0, 1].

ICISSP 2016 - 2nd International Conference on Information Systems Security and Privacy

268

Figure 5: Example of Reputation Score Table.

In this case study, we weigh more on personal

characteristics (i.e., individual reputation) than rela-

tional characteristics (i.e., interactions with friends

via activities & peers’ reputation). A weight of each

reputation component can be adjusted depending on

the need or requirement of a given OSN. α, β, and µ

are ﬁxed for all users for fairness. The weights can be

periodically updated based on historical data in order

to avoid any dominance of a single component.

To measure reputation based on personal at-

tributes, current work, highest education, and afﬁl-

iated interest groups are considered and their own

scores are included as part of individual score. The

reputation rank can be derived for those entities from

the same OSN. For top peers or activities, we use top

ﬁve peers or activities to represent an individual user’s

reputation. We equally weight each personal attribute,

peer, or activity for simplicity in this work. However,

it can be adjusted in order to reﬂect different criteria

of reputation.

In Figure 5, the following features are presented:

• The three constant weights (i.e., α,β,µ) can have a

signiﬁcant impact on the overall reputation value.

The values of those weight parameters should be

ﬁxed for all users. Ultimately, they should be cal-

culated based on the analysis of a large volume of

historical data in the same OSN in order to iden-

tify the importance of each attribute in determin-

ing the value of each weight.

• When users are in the top 5 peers of each other

(e.g., a spouse or a couple), we call it a circular

dependency problem. The reputation scores for

a couple listed in top peers of each other should

not be calculated simultaneously. We should en-

sure to avoid circular dependency problem so that

it is unlikely that top 5 activities for the same cou-

ple will be adjusted simultaneously. Similarly, a

circular dependency may occur between an orga-

nization (e.g., universities or companies) and its

afﬁliated individuals. Reputation ranks of such

entities (e.g., universities or companies) are ex-

pected to change less frequently in comparison

with those of individual users.

• Users may increase their reputation ranks trough

various ways including legitimate or illegiti-

mate activities. Reputation models should detect

anomaly behavior in the evolution of reputation.

One application for using friends’ automatic clas-

siﬁcation system is to decide visibility levels on users’

created activities. Close friends should be given

higher visibility access. Higher privacy corresponds

to lower visibility on sensitive information. Senti-

mental analysis can be used to automate the process

of privacy level classiﬁcation. This can work as an

alternative to users’ manual classiﬁcation for the pri-

vacy level of their posted activity. Friends with high

trust edges (e.g., spouse, couple, close friends) can be

assigned to high privacy/visibility while friends with

low trust edges (e.g., acquaintances) can be assigned

to low privacy/visibility. Thus, top K peers can have

the highest visibility/privacy.

The advantages of the proposed privacy assess-

ment model are summarized as follows:

• Privacy assessment can be automatically set based

on the calculated edge rank of each peer; users

only need to set privacy category (e.g., low,

medium, high) for an activity they create, and the

eligible peers for the post visibility will be auto-

matically determined based on the edge rank of

each peer.

• If the system’s automatic decision of post visibil-

ity is not agreed by a user, the user can manu-

ally modify the visibility to a particular peer. This

allows the ﬂexibility that accepts a user’s input

in order to be adaptive to the user’s need. That

is, privacy assessment system dynamically deter-

mines a user’s privacy setting based on top K

peers identiﬁed over time or upon activities / in-

teractions. But the system always allows the user

to manually change the privacy setting depending

on the need.

• Privacy assessment system allows a user to clas-

sify his/her friends based on the customized

friendship categories by setting the range of each

category. For example, based on the rank of each

friend, a user can categorize: 75% and above for

close friends (i.e., high visibility); 75% to 25 %

for normal friends (i.e., medium visibility); and

below 25% and below for acquaintances (i.e., low

Interaction-based Reputation Model in Online Social Networks

269

visibility). Such three level classiﬁcation can be

used to control privacy and visibility of posts. A

user can select the thresholds such as 75% and

25% as the above.

4 RELATED WORK

In this section, we discuss what trust, reputation, and

privacy are and how they are used, measured, and im-

plemented in OSN environments.

4.1 Trust Models in OSNs

Trust is deﬁned as “assured reliance on the character,

ability, strength, or truth of someone or something”

(Merriam and Webster Dictionary, 2015). Trust in-

dicates a relationship between two entities called a

trustor and trustee in which the trustor assesses its

subjective opinion towards the trustee based on given

criteria. Since trust is a multidisciplinary concept, dif-

ferent types of trust relationships are used depending

on a domain such as e-commerce settings, automatic

computing, or communication networks (Cho et al.,

2011).

The nature of trust has the following inherent

properties (Cho et al., 2011):

• Subjective: Since a human entity has a different

opinion based on its own subjective view, differ-

ent individual entities may have different views

towards a same entity;

• Asymmetric: When two entities are friends to each

other, they may not trust each other with a same

degree. That is, A may not trust B as much as

B trusts A. In an extreme case, trust between two

individuals can be directional such as the case that

A trusts B while B does not trust A (Hardin, 2002).

• Non-transitive: Trust is not completely transitive

or transferable. Although trust in communication

networks has used to be transitive or transferable

in PGP (Pretty Good Privacy) using the web of

trust (Stallings, 1995), it is not true in social net-

works where an entity is a human and may have a

different subjective perception or cognition in in-

formation processing.

• Dynamic: Trust decays over time without contin-

uous interactions or changes as context changes.

In social networks, the ﬂuctuations of the quality

and quantity of interactions hinder the continuous

evolution of trust.

• Context-dependent: Trust is affected by a do-

main context. In particular, when a situation

changes over time, the degree of perceived also

changes. Jøsang and Pope (2005) call the context-

dependent trust decision trust while the context-

independent trust reliability trust. Decision trust

is affected by what situation a trustor and trustee

are placed (e.g., whether to use a wore rope to es-

cape from a ﬁre situation in building).

The concept and properties of trust have been uti-

lized in various types of OSN applications in the liter-

ature. Ganzaroli (2002) and Murphy (2006) see trust

as the continuous process to be built in institutional

and structural contexts in addition to subjective view-

points. Grabner-Krauter (2009) treats trust as an in-

dicator of an individual’s conﬁdence in dynamic de-

cision making context in which trust is used as an in-

strumental support emphasizing its practical use, in

addition to subjective trust based on emotional nature.

Ferlander (2003) and Kavanaugh et al. (2005) dis-

cuss two types of trust in terms of the degree of infor-

mation sharing. Thick trust is formed based on strong

social ties through frequent direct interactions while

thin trust is a weak social tie which is formed only

based on a small amount of information sharing. In

this sense, how much information is shared between

two entities in social networks can be an indicator of

more interactions leading to thick trust. This is of-

ten called social capital when the social relationship

introduces productivity in real life (e.g., recruitment

through social networks such as LinkedIn) (Wool-

cook, 2001).

Graph theories have been popularly used to model

trust relationships in social networks (Marti et al.,

2005; Zhang et al., 2006). Common challenges in us-

ing graph theory are in three-fold: (1) graph complex-

ity where typical social networks are large in number

of nodes, volume of interactions, etc.; (2) scaling of

trust (e.g., binary, real number in [0, 1], nominal, cate-

gorical); and (3) validation of trust models in terms of

trust accuracy. Richardson et al. (2003) consider con-

tinuous trust values, rather than categorical or nomi-

nal trust scale.

Kuter and Golbeck (2007) evaluate interactions

between friends to rate movies by developing a trust

model called FilmTrust. Compared to product rat-

ings provided by many commercial websites, this rat-

ing can be seen to individuals as more “trustworthy”

because the rates are evaluated based on their friends.

In our reputation model, we add more dimensions of

personal and relational attributes to derive reputation

score which reﬂects the concept of multidimensional

trust or reputation in evaluating an entity or service.

Trust in social networks can be highly dependent

on density and centrality (Buskens, 1998). Trust be-

tween friends can increase or decrease based on com-

ICISSP 2016 - 2nd International Conference on Information Systems Security and Privacy

270

mon interest (Khambatti et al., 2004). Unlike this

work, although our reputation model considers groups

to evaluate a peer’s reputation rank, we also consider

other factors such as the amount of interactions in a

particular activity. The number of common groups

between two individuals may not necessary indicate a

strong tie. We consider both quality and quantity of

interactions in groups with common interest.

In the web context and network graphs, nodes may

not necessary be human individuals as they can be

web agents, crawlers, bots (Hang and Singh, 2010),

routers or routing protocols (Marti et al., 2005), web-

sites, articles, products. Trust between those agents

will increase or decrease over time based on historical

data. Malicious users can have negative impacts on

trust models (Caverlee et al., 2008) because their mis-

behavior can impact their own trust as well as other

nodes who are interacting with them. As a result, trust

model should include methods to enhance reliability

and counter malicious behaviors.

Although trust models have been used to esti-

mate trust of an individual entity (e.g., person, ma-

chine), the concept of trust has been extended to refer

to trust in an organization, not an individual person,

such as companies, groups, government, clubs, and

so on (Grabner-Krauter and Bitter, 2015; Granovet-

ter, 1992). The concept of trust used for organiza-

tions is called enterprise trust. Granovetter (1992)

deﬁne inﬂuence based on two aspects: (1) inﬂuence

based on relations between individuals; and (2) inﬂu-

ence based on social network structure. The former is

more related to trust in the behavior of an individual

while the latter refers to trust derived from where an

entity is located in a given social network. The met-

rics of measuring individual trust and enterprise trust

have been used differently. Individual trust is mea-

sured based on closeness, intimacy, or emotional sup-

port while enterprise trust has been estimated based

on density or cohesiveness of an associated network,

technology, software, system or network architectures

used in an organization.

While the concept of trust is used to indicate

the relationship between two entities, a more general

sense of trust towards a particular entity is estimated

based on opinions by multiple entities. We call it rep-

utation whose concept and models of reputation are

discussed as below.

4.2 Reputation Models in OSNs

Reputation is deﬁned as general beliefs or opinions

towards someone or something (Merriam and Web-

ster Dictionary, 2015). Although the concept of repu-

tation is overlapped with that of trust in terms of sub-

jective perception, expectation, or belief about capa-

bility, honesty, or reliability of something or someone,

it has a more aspect of objective concept than that of

trust because it tends to rely on more aggregated opin-

ions of multiple entities (Hussain and Chang, 2007).

In OSN environments, reputation is often mea-

sured based on the amount of interactions. In partic-

ular, the concept of friendship and its measurement

based on various behavioral attributes are adopted

in order to quantify the degree of friendship with

the goal of measuring an entity’s reputation (Traud

et al., 2011; Hossmann et al., 2011; Nguyen et al.,

2013). The example metrics of friendship are based

on the types of relationships including relatives,

spouse, neighborhood, friendship, work colleagues,

school alumni, common interests, or hobbies (Jones

et al., 2013). Reputation models using the concept

of friendship have been used for various social net-

work applications such as spam detection (Gao et al.,

2012), categorization of relationships based on inter-

actions (Akbas et al., 2013), and trust propagation

based on the amount of interactions between friends.

5 CONCLUSION

In this work, we proposed a reputation model where

reputation is derived from three main components:

personal attributes, personal activities, and peers’ rep-

utation. In addition, we showed an application of the

proposed reputation model for privacy assessment.

The key idea of this reputation-based privacy model

is that privacy levels for each friend or post can be au-

tomatically set based on dynamic estimation of rep-

utation score of each friend. That is, the estimated

reputation score is used to determine the visibility of

any posts or activities by a user.

We plan our future work directions as: (1) investi-

gating the circular dependency problem of reputation

scores; (2) examining how to distinguish popularity

obtained by high interactions with multiple friends

from that by high interactions with a single friend;

and (3) implementing / validating this model through

empirical studies.

REFERENCES

Akbas, M. I., Avula, R. N., Bassiouni, M. A., and Turgut, D.

(2013). Social network generation and friend ranking

based on mobile phone data. In Proceedings of the

International Conference on Communications, pages

1444–1448.

Buskens, V. (1998). The social structure of trust. Social

Networks, 20(3):265–289.

Interaction-based Reputation Model in Online Social Networks

271

Caverlee, J., Liu, L., , and Webb, S. (2008). Social

trust: Tamper-resilient trust establishment in online

communities. In Proceedings of the 8th ACM/IEEE-

CS Joint Conference on Digital Libraries (JCDL08),

pages 104–114.

Cho, J., Swami, A., and Chen, I. (2011). A survey of trust

management in mobile ad hoc networks. IEEE Com-

munications Surveys and Tutorials, 13(4):562–583.

Ferlander, S. (2003). The Internet, social capital and local

community. Technical report, University of Sterling.

Ganzaroli, A. (2002). Creating trust between local and

global systems. Technical report, Erasmus University

Rotterdam.

Gao, H., Chen, Y., Lee, K., Palsetia, D., and Choudhary,

A. N. (2012). Towards online spam ﬁltering in social

networks. In Proceedings of the Network and Dis-

tributed System Security Symposium (NDSS).

Grabner-Krauter, S. (2009). Web 2.0 social networks: The

role of trust. Journal of Business Ethics, 90:505–522.

Grabner-Krauter, S. and Bitter, S. (2015). Trust in online

social networks: A multifaceted perspective. Forum

for Social Economics, 44(1):48–68.

Granovetter, M. (1992). Problems of explanation in eco-

nomic sociology. Technical report, Harvard Business

School Press, Boston, MA, USA.

Hang, W. C. and Singh, M. P. (2010). Trust based recom-

mendation based on graph similarities.

Hardin, R. (2002). Trust and Trustworthiness. Russell Sage

Foundation.

Hossmann, T., Legendre, F., Nomikos, G., and Spyropou-

los, T. (2011). Stumbl: Using facebook to collect

rich datasets for opportunistic networking research.

In Proceedings of IEEE International Symposium on

a World of Wireless, Mobile and Multimedia Net-

works(WOWMOM), pages 1–6.

Hussain, F. K. and Chang, E. (2007). An overview of the

interpretations of trust and reputation. In The Third

Advanced International Conference on Telecommuni-

cations (AICT 2007).

Jones, J. J., Settle, J. E., Bond, R. M., Fariss, C. J., Marlow,

C., and Fowler, J. H. (2013). Trust in online social

networks: A multifaceted perspective. Plos One Jour-

nal.

Jøsang, A. and Pope, S. (2005). Semantic constraints for

trust transitivity. Proceedings of Asia-Paciﬁc Conf. of

Conceptual Modeling, 3.

Kavanaugh, A., Reese, D. D., Carroll, J. M., and Rosson,

M. (2005). Weak ties in networked communities.

The Information Society: An International Journal,

21(2):119–131.

Khambatti, M., Dasgupta, P., and Ryu, K. D. (2004). A role-

based trust model for peer-to-peer communities and

dynamic coalitions. In IEEE Information Assurance

Workshop.

Mansﬁeld-Devine, S. (2008). Anti-social networking: Ex-

ploiting the trusting environment of web 2.0. Network

Security, 2008(11):4–7.

Marti, S., Ganesan, P., and Garcia-Molina, H. (2005).

Sprout: P2p routing with social networks. In Cur-

rent Trends in Database Technology-EDBT. Springer

Berlin Heidelberg.

Merriam and Webster Dictionary (2015). Trust and Repu-

tation Deﬁnition.

Murphy, J. (2006). Building trust in economic space.

Progress in Human Geography, 30(4):427–450.

Nguyen, T., Chen, M., and Szymanski, B. (2013). An-

alyzing the proximity and interactions of friends in

communities in gowalla. In Proceedings of IEEE

13th International Conference Data Mining Work-

shops, pages 1036–1044, Dallas, Texas, USA.

Richardson, M., Agrawal, A., and Domingos, P. (2003).

Trust management for the semantic web. In Proceed-

ings of the Second International Semantic Web Con-

ference, Sanibel Island, Florida.

Stallings, W. (1995). The pgp web of trust. BYTE,

20(2):161–162.

Traud, A. L., Mucha, P. J., and Porter, M. A. (2011).

Social structure of facebook networks. Physica A,

391(16):4165–4180.

Woolcook, M. (2001). The place of social capital in under-

standing social and economic outcomes. Canadian

Journal of Policy Research, 2(1).

Zhang, Y., Chen, H., and Wu, Z. (2006). A social network-

based trust model for the semantic web. In Proceed-

ings of the 6th International Conference on Autonomic

and Trusted Computing, pages 183–192.

ICISSP 2016 - 2nd International Conference on Information Systems Security and Privacy

272