ArticlePDF Available

Abstract and Figures

In World Wide Web there are many online communities with a huge number of users and a great amount of data which are continuously increasing. In this context it is important for users to interact with resources and other users according to their preferences. On this direction of information filtering domain our work is oriented to trust based filtering. We have developed a model formed by three interconnected components: trust component which allows computation of trust levels among users which are not directly connected, a component which computes reputation of entities in the system, and a recommendation component. Several sets of tests of our model have been performed and we have the possibility to integrate it in various online communities.
Content may be subject to copyright.
Studies in Informatics and Control, Vol. 20, No. 2, June 2011 http://www.sic.ici.ro 143
1. Introduction
The existing technologies have led to an
increasing interaction between people who do
not know each other. In this case, online
interactions replace human interactions. An
important goal of our article is to improve
these interactions based on two important
concepts: social trust and reputation.
Trust and reputation are two interrelated
concepts. We can find trust at personal level.
Reputation expresses an opinion resulting
from collective opinions of community
members. This type of evaluation may lead to
risks such as penalty of innovative and
minority ideas, problem described in
(Tocqueville 1840), (Massa, Avesani 2007)
as “tyranny of the majority”. Naturally, the
opinions of minority groups matter and
should be seen as opportunities. But if
minority groups obtain a full priority, it is
obtained the other extreme, the so-called
phenomenon of "echo chamber". In this case,
as shown in (Sunstein 2009) will result a
fragmentation of society into micro-groups
that tend to sustain extremely their opinions.
Nowadays, there are many online
communities (community for sharing
resources, social networks, scientific
communities, etc.), that store a great amount
of data which are continuously increasing.
Anyone can publish any kind of resources: a
diary published within a blog, a track that a
user wants to make public, etc.
In this context in which users have to interact
with other users about whom they don’t have
any previous information, and in which the
overloaded information phenomenon brings a
major impact, this paper comes up with a
solution to improve the interactions among
users and resources management.
The proposed trust and reputation model
assures that users experience resulted from
the previous interactions are used to establish
user-user and user-resource evaluation levels.
Therefore, the purpose of the paper is to find
a solution based on trust and reputation to
provide users from online communities a
balanced combination of personal vision with
a global perspective on the community that
will provide the opportunity to interact with
users and resources that are relevant to them.
Section 2 presents the trust and reputation
concept and the main proposals that exist in
the scientific literature. Section 3 presents our
trust and reputation model based on a set of
published results (Alboaie 2008), (Alboaie,
Barbu 2008), (Alboaie, Vaida 2010).
Section 4 provides details about our model
architecture and the results obtained from
several tests are presented. A comparison with
the most relevant local trust metric, Mole
Trust has been performed. Section 5 will
contain the conclusions and the future work.
Trust and Reputation Model for Various
Online Communities
Lenuta Alboaie1, Mircea-F. Vaida2
1 Computer Science Department, Alexandru Ioan Cuza University of Iasi, Romania,
16,Berthelot, Iasi, Romania,
adria@infoiasi.ro
2 Communication Department, Technical University Of Cluj-Napoca,
26-28, Gh. Baritiu Street, Cluj-Napoca, Romania,
mircea.vaida@com.utcluj.ro, corresponding author
Abstract: In World Wide Web there are many online communities with a huge number of users and a great amount
of data which are continuously increasing. In this context it is important for users to interact with resources and other
users according to their
p
references. On this direction of information filtering domain our work is oriented to trust
b
ased filtering. We have developed a model formed by three interconnected components: trust component which
allows computation of trust levels among users which are not directly connected, a component which computes
reputation of entities in the system, and a recommendation component. Several sets of tests of our model have been
performed and we have the possibility to integrate it in various online communities.
Ke
y
words: online communit
y
, social trust, local trust metric, re
p
utation, recommendations
http://www.sic.ici.ro Studies in Informatics and Control, Vol. 20, No. 2, June 2011
144
2. Related Work
In each of the areas in which trust plays an
important role, e.g. sociology, psychology,
political science, economics, philosophy and
computer science, were given various
definitions of the trust concept.
The definition of trust concept accepted by
great majority of the authors is presented in
(Gambetta 1990): “Trust is the subjective
probability by which an individual, A,
expects that another individual, B, performs a
given action on which its welfare depends”.
This action is in online communities an
evaluation, an opinion that someone
expresses regarding someone else and is
quantified by numerical values.
The concept of social trust is associated with
four properties (Golbeck 2005b): transitivity,
composability, personalization, asymmetry.
These properties make it possible to calculate
trust. Trust is not perfectly transitive in the
mathematical sense, but trust can be
transmitted between entities.
The composability property specifies how the
associated ratings of trust are propagated
between entities which are not
directly connected.
The third property is the personalization of
trust, this means that on the entity C, A and B
may have different opinions. Another
property of trust is the asymmetry and means
that if A trusts B, B may not have the same
trust level in A. We are dealing with so-called
one-way trust (Hardin 2002). In essence, trust
is represented by a user judgment concerning
other user, sometimes carried out directly and
explicitly, sometimes indirectly through
assessment of various actions taken by the
same user.
Trust is calculated in a so-called trust
network. Trust network is a graph obtained
by aggregating users’ evaluations. These
evaluations can be quantified, so we have
different levels of trust that can be established
between users. In such a trust network can
run a so-called trust metric which is actually
an algorithm that receives input information
from the network and calculates various
values of trust among users (Massa 2006). In
the literature it is used the trust metric notion,
where metric does not signify as the
mathematical concept of metric, being a
distance function.
A first trust metric was developed in (Levien
2003) within the Advogato system where the
metric was used to determine how members
can trust among community members.
In this paper we use the notion of trust metric
or algorithm for measuring trust in the sense
described above.
Trust metrics are divided into local trust
metrics and global trust metrics. Global
metrics take into account all existing nodes
and links of trust. A global value is assigned
to an agent based on all network information.
Many global trust metrics such as (Sepander
et al 2003), (Guha 2003) were inspired from
the PageRank algorithm (Page et al 1998)
that calculates the reputation of Web pages.
Local trust metrics take into account personal
interactions. A local trust metric calculates
trust from subjective opinion of an entity.
Thus, the trust value associated to an entity
varies for each existing agent in the system.
For the concept of reputation we stopped on
the two definitions we encounter in Merriam-
Webster's dictionary and in the Compact
Oxford Dictionary:
Definition1: overall quality or character as
seen or judged by people in general.
Definition2: the beliefs or opinions that are
generally held about someone or something
In (Mui et al 2002) is identified a property
which characterizes the relationship between
trust and reputation: reciprocity. The
reciprocity is defined as the reciprocal
exchange of assessment (favourable or not).
Decrease any of these automatically conduct
to the reverse effect.
In (Massa, Avesani 2007) is clearly illustrated
the difference between a global and a local
metric when the entities over which there are
divergent points of view are highlighted. It is
natural that users have different views on a
user, but this does not mean that one opinion is
correct and the other not. Actually must be
considered that they just disagree. With a
global trust metric the controversial users’
aspect cannot be surprised.
In literature we find very few models of local
trust metric (Zieglar, Lausen 2004), (Golbeck
Studies in Informatics and Control, Vol. 20, No. 2, June 2011 http://www.sic.ici.ro 145
2005b), (Massa, Avesani 2006) or a proposed
trust metric (Zhili et al 2009) which extends
the metric from (Massa, Avesani 2006).
In majority of cases the computing models
are based on the global trust or reputation,
and in (Josang 2007) was made a
classification of reputation systems and used
calculation methods: calculation of the
ratings sum (e.g., eBay), averaging ratings
(e.g. Amazon, Epinions) using Bayesian
systems (e.g. systems proposed at the
theoretical level (Nurmi 2006), (Mui et al
2002 )), using discrete trust models (e.g. the
model proposed in (Rahman, Hailes 2000)),
using a fuzzy model (e.g. systems proposed
in (Sabater, Sierra 2002)), using flow models
(e.g., Google Page Rank (Page et al 1998),
Advogado (Levien 2003) Appleseed (Ziegler,
Lausen 2004)).
One of the novelties carry out in this paper is
a model that provides to users, both a local
personalized vision of the system provided by
our local trust metric, and a global vision
given by a mechanism that compute
the reputation.
3. Trust and Reputation
Proposed Model
In this section will be described the proposed
trust and reputation model, named StarTrust.
StarTrust is based on experiments and tests
realized with a previous model, StarWorth
(Alboaie 2008), (Alboaie, Barbu 2008),
(Alboaie, Vaida 2010). StarTrust contains in
addition a mechanism for trust propagation
that take into account the untrust factor that
may exist between two users. StarTrust will
contain a reputation component that provides
to systems that integrates our model, a
balance between two factors "echo chamber"
and "tyranny of majority." The StarTrust
model is made up of three main elements:
Trust component,
Resource recommendation component,
Reputation component.
We will consider the following terms:
Users - are members of an online community.
Resources - their definition is made
accordingly to the definition given by (T.
Berners-Lee 1998).
Worth - is a measure that signifies an
evaluation accorded by a user to another
user or resource. Also, the worth can be
obtained (quantized) indirectly as we will
see in the following paragraphs. In our
system we consider five evaluation levels
with the following semantic:
Table 1. Levels of evaluations with
their significance
Level1 Level
2
Level3 Level
4
Level5
(0,1] (1,2] (2,3] (3,4] (4,5]
useless/spa
m
poor worth
attention good exceptiona
l
We note the upper limit with MaxWorth,
where MaxWorth = 5 in our experiments.
Trust component
The purpose of this component is to provide
users from an online community a
personalized vision of the system. We
considered a set of constructions that will be
used in the following sections and which
have the following associated semantics. In
fact, these constructions can be
mathematically considered as functions or,
from the implementation point of view, they
are considered associative tables:
Explicit worth of a user:
WE_UU(useri,userj) – explicit worth,
represents the rating for userj, and the rating
is given manually by the useri to userj.
Implicit (deducted) worth of a user:
WE_UU(useri,userj) – measures how
close are of both preferences.
(The preference can be considered the
accepting degree of a point of view).
We consider the function WE(useri,userj) for
each pair of (useri,userj) that contains the
explicit and implicit evaluations:
(, )
( , ), if explicitly eval.
_
( , ), otherwise
ij
ij i j
ij
WU U U
WE UU U U U U
WI UU U U
(1)
We will define the manner of computation of
the implicit values introduced above.
http://www.sic.ici.ro Studies in Informatics and Control, Vol. 20, No. 2, June 2011
146
Let us consider two users Ui, Uj. The value of
WI_UU(Ui,Uj) indicates the deducted worth
based on explicit evaluations made by users
to each other. Let consider the users Ui, from
whom we have ratings to },...,{ 1k
ii UU . Also,
we consider having explicit ratings from
l
i
Uto Uj, lk so we have defined
),(_ j
l
iUUUUWE (see Figure 1).
In order to compute WI_UU we must to
compute the value of the weight
corresponding with the explicit ratings. We
denote this weight with ),( j
l
iE UUP . The
weight represents (from the point of view of
Ui) an explicit rating, in our case the rating
weight given by l
i
Uto j
U and it is computed
as follows:
M
axWorth
UUWU
UUP l
ii
j
l
iE ),(
),( (2)
We compute the implicit rating that user Ui
provided to Uj as:
k
lj
l
ij
l
iEji UUWEUUP
k
UUUUWI 1),(*),(*
1
),(_ (3)
where, kl 1, k is the number of the users
that were explicitly evaluate by i
U. From (2)
and (3) we obtain the implicit reputation
computing formula that we used in tests and
experiments realized in our previous works:
M
axWorthk
UUWEUUWU
UUUUWI
k
lj
l
i
l
ii
ji *
),(*),(
),(_ 1
(4)
We have ),(_),( l
ii
l
ii UUUUWEUUWU if
there exists an explicit evaluation from Ui to
l
i
U, otherwise we consider an implicit
evaluation from i
Uto l
i
U.
Our model in this phase only surprise the trust
that user Ui will accord to l
i
Uthat will made a
correct evaluation of Uj from his point of view.
But the model does not surprise the fact that
it is possible that l
i
U don’t realize a correct
evaluation from i
U point of view. For a
better understanding of the necessity to
surprise such an aspect will begin to examine
a particular case and we consider:
three users 321 ,, UUU
),(_ 21 UUUUWE is the trust value given
in an explicit mode by user 1
U to 2
U,
),(_ 32 UUUUWE is the trust value given
in an explicit mode by user 2
Uto 3
U.
We want to analyze the following (see Figure 2):
how to obtain the trust value
),(_ 31 UUUUWI (that represents the
trust value given in implicit mode of 1
U
to 3
U, that the system will compute)
how relevant is that value for 1
U
Figure 1. Implicit user-user evaluation computation
Studies in Informatics and Control, Vol. 20, No. 2, June 2011 http://www.sic.ici.ro 147
Considering relation (4) the value
),(_ 31 UUUUWI will be computed thus:
),(_*
*1 ),(_
),(_ 32
21
31 UUUUWE
M
axWorth
UUUUWE
UUUUWI
Ratio
M
axWorth
UUUUWE*1 ),(_ 21 represents the
probability that 2
Uwill make a correct
evaluation of 3
Ufrom the point of view of
1
U. We mention that any evaluations are
correct, that provides the existence of local
trust metrics.
User 1
Udoes not know that 2
Uwill make a
correct evaluation, and thus to surprise such a
possibility of an incorrect evaluation the
general relation to compute ),(_ 31 UUUUWI
must be as follows:
TC
P
M
axWorth
UUUUWE
UUUUWE
MaxWorth
UUUUWE
UUUUWI
*)
*1
),(_
1(),(_*
*
*1
),(_
),(_
21
32
21
31
Where: TC
P is named trust control
parameter, representing a probable implicit
trust that is accorded to any user in system.
The expression
M
axWorth
UUUUWE*1 ),(_
121
represents the probability that 2
Uto have no
right in evaluation of 3
Ufrom the point of
view of 1
U.
If ),(_ 21 UUUUWE has a big value ( 1
U has
a maximum trust in 2
U) the value of the
expression TC
P
M
axWorth
UUUUWE *)
*1 ),(_
1( 21
will
be close to 0 and the relation will be
simplified to formula (4).
We consider users Ui and Uj from a
community. We consider },...,{ 1k
ii UU the set
of users to whom there are explicit or implicit
ratings from user Ui. We also consider that
exist explicit ratings from l
i
U to j
U, kl .
The general formula to compute implicit trust
is (5):
),(*)
*
),(
1(
*
),(*),(
),(_
1
1
jiTC
k
l
l
ii
k
lj
l
i
l
ii
ji
UUP
MaxWorthk
UUWU
MaxWorthk
UUWEUUWU
UUUUWI
We used the notation PTC(Ui,Uj) to specify
that PTC is not absolute a constant, the value
may vary depending on the user that realize
the evaluation and the user that receive
the evaluation.
Therefore we have defined a general local
trust metric that may be adapted to any type
of community by means of the values that
may be associated to parameter TC
P. The
experiments from this paper will consider
values for parameter TC
P=0, and also
different of 0.
If TC
Pis 0, than the trust accorded to far
nodes is smaller as the distance to the source
node is bigger.
Depending on the community were we
integrating StarTrust, we may choose more
values of TC
P parameter:
in a close community (such as a scientific
community) the value can be considered
0. Considering this value, all ratings from
users who are at large distances in the
graph will be increasingly smaller. In this
way is minimized the influence of users
who are not very close to the source node.
Figure 2. Implicit evaluation between two entities
http://www.sic.ici.ro Studies in Informatics and Control, Vol. 20, No. 2, June 2011
148
in a community such as an online
community where it is not possible to
define the profile of the users, the value
for TC
Pcould be considered as the
reputation of the user.
in other communities we may consider
TC
Phaving as value the media value of
the given explicit evaluations or the
media of first 10% of good evaluations,
so depending of the type of the
community we may consider specific
values for this factor.
How the ratings are lower, the member
M
axWorthk
UUWU
k
l
l
ii
*
),(
11
will be close to 1, so that the
importance of PTC is essential (in this case the
distance between users doesn’t matter).
We consider as follows the pseudo code for
local trust metric from StarTrust. We have
the following notations:
sourceUser – is user for which is
calculated the vision over community;
WU – contains the evaluation from trust
network in community at a given time
WE – contains the explicit evaluations
from WU
sinkUsers – are users that received
evaluations from sourceUser
Input: sourceUser, WE, WU
Output: WU, sinkUsers
Step 1. add in sinkUsers all nodes accessible fro
m
sourceUser
Step 2. do savedWU = currentTrustNetwork;
Step 3. Foreach U în sinkUsers
Step 4. Find 1
{ ,..., }
k
UU
ii
satisfying the following
conditions: (there is an edge between sourceUser an
d
each k
i
Uin WU ) and (there is an edge between eac
h
k
i
Uand U in WE )
Step 5. Calculate implicit trust value betwee
n
sourceUser and Uusing (5):
),(*)
*
1),(
1(
*
1),(*),(
),(
UsourceUser
TC
P
MaxWorthk
k
ll
i
UsourceUsersavedWU
MaxWorthk
k
lU
l
i
UWE
l
i
UsourceUsersavedWU
UsourceUserWU
/* Update or insert an edge between sourceUser and
U
with capacity computed at Step 5 in WU */
while (savedWU != WU)
Algorithm 1. StarTrust – local trust metric
The stop condition ( WUsavedWU !) is
materialized in implementation through
election of a
value, so that two matrix
savedWU, WU are in different relationship if
there exists indices i and j so
that
]),[],[( jisavedWUjiWU .
Resource recommendation component
In the context of traditional recommendation
systems, users give ratings to resources and
based on these ratings the system will make
recommendations. The standard mechanism
used in recommendation systems (e.g. Person
Correlation) causes cases in which a
recommendation system does not provide
satisfactory results, and we mention (Massa,
Avesani 2006): extending the period of real
integration of new users or promotion of a
new resource added to the system is realized
for a long period.
The goal of the proposal component is to
build a flexible way to manage resources in a
personalized manner. In order to achieve this
we shall consider a mechanism that is based
on the trust relation among users, which
already have evaluated other resources.
We shall consider the following constructions:
WE_UR(useri,resourcej) - represents the
explicit evaluation given by useri
to resourcej.
WI_UR(useri,resourcej) - represents the
implicit evaluation value, computed by
system and that useri associate
to resourcej
Let us consider ),( resourceuserWR function,
for every pair (user, resource):
otherwise ),,(_
eval. explicitly if ,),(_
),( ji
jiji
ji RUURWI
RURUURWE
RUWR (6)
Using the same reasoning presented for
computing trust between users, we obtained
the implicit rating from user i
Uto j
R(7):
M
axWorth
k
RUURWEUUWU
RUURWI
k
lj
l
i
l
ii
ji *
),(_*),(
),(_ 1
Using this component a system can provide a
hierarchical personalized ranking of
resources based on user's vision, so we have a
recommendation based resource mechanism
Studies in Informatics and Control, Vol. 20, No. 2, June 2011 http://www.sic.ici.ro 149
that uses both the ratings given by users to
resources and the ratings given among
the users.
Reputation component
As presented in Section 2 the concepts of
trust and reputation are closely linked.
Reputation is a value that signifies the image
of a community concerning a user. For our
system we consider a general formula for
calculating the reputation, we justify the
choices and we show that it can be
customized for various online communities
that may exist.
We consider a user i
U. We note with
},...,,{ 21 k
lll
i
lEvEvEvN the set of ratings that
user i
U have received in the interval given by
level l, MaxWorthl 1, MaxWorth=5.
We note with i
l
Nthe cardinal of i
l
N that
represents the number of ratings that a user
received from the given interval. We consider
for a user the general reputation formula so:
l
RC
MaxWorth
lMaxWorth
j
i
j
j
LW
i
l
iP
NP
N
U*
*
)(Rep
1
1
(8)
where: l
RC
P is the reputation control
parameter, j
LW
P is the level weight parameter
Choosing justifications:
Intuitively, the global reputation of a user
should take into account the actual received
ratings and the number of ratings he received
for each rating level. In the calculation of
reputation, the ratio
MaxWorth
j
i
j
i
l
N
N
1
represents the
importance of the number of ratings that a
user was evaluated in an interval. Intuitively,
we consider that j
LW
Pis 1. This factor will
help us to adjust the importance of each
evaluation level to compute the reputation.
The factor l
RC
P goal is to adapt the
computation of reputation value to the
community profile.
We consider different possibilities to select
this factor. In previous studies it was imposed
the restriction that the value of reputation to
be in the interval [0, MaxWorth]. A method
to provide this thing is to choose the factor
l
RC
Pto have a value in the specified interval
of level l. So if we have the interval [m, M],
MmMaxWorthMMaxWorthm ,;1,0
must respect the condition: MPm l
RC . In
this case is easy to demonstrate that the
computed reputation will be in the interval
[0, MaxWorth].
We present a possible choosing values for
parameters for the adjusting the reputation
If l
RC
P has as value the ratings media on a
given interval l, the formula to compute
reputation is reduced to determine the
arithmetical media, this being a mechanism
consider by many functional real systems
(e.g. eBay). If we have a close community
where we are able to give a certain trust level
(as example a scientific community), than we
may consider i
l
l
RC NP max = . If we have an
open community where the community
members are able to easy change the identity
and it is not possible to establish a general
profile of the community users, we may
consider i
l
l
RC NP min = .
This parameterization to compute reputation
presents an aspect that is not considered by
the majority of the existed literature models.
Thus, the adaptation computing mechanism
of reputation can be modeled depending on
the community type where is used.
For the next studies we will eliminate these
restrictions and we will consider that only
trust, which is subjective, must be integrated in
such an interval. Reputation is better to do not
be limited to an interval. Thus, the reputation
will reflect a real unlimited value of the user.
We presented that reputation can be used in
the relation to compute the trust, and for that
we will provide a corresponding scaling.
StarTrust Implementation
StarTrust was developed as a model that once
integrated in a community is able to provide
the following services: trust relations among
users provided by trust component, users
reputation provided by reputation component
http://www.sic.ici.ro Studies in Informatics and Control, Vol. 20, No. 2, June 2011
150
and a service that is able to recommend
resources depending on preferences
expressed by users using the resource
recommendation component.
The model implementation was realized by
implementing the three components with
their interactions.
In the first step were considered formulas (1)-
(5). In this step the aim was to obtain user
trust vectors. We consider a set of
users

n
UU ,...
1. To each user from the
system is associated a trust vector. Also we
consider the matrix R that contains
the ratings:
],1[ , toby given rating),( ni,jUUjiR ji
The ratings from R are explicit or implicit
ratings obtained by relations (1)-(5).
We consider the following algorithm that
computes the elements of R matrix. We
consider the constructions:
RE – explicit rating matrix;
R – current ratings matrix of the system;
Input: RE
Output:R
Step 1. R=RE
Step 2. Insert the rating R(i,j) = MaxWorth for
ji , ni 1,nj
1
Step 3. For each i, ni 1
execute the Algorithm 1 using:
- i as sourceUser
- RE as WE
- R as WU
Algorithm 2. Algorithm of trust propagation in
whole community
The second step in the implementation
process of the ratings mechanism will
consider formulas (6)-(7). In this step the aim
is to obtain relevant resources concerning a
user point of view. So were computed the
vectors that will contain the given ratings by
a certain resource from system users. These
ratings represent actually the interest level
that a user has for a resource.
We consider:
n
UU ,...
1the users set
m
RR ,...
1the resources set
The next algorithm will compute the elements
of RR matrix that contains evaluations for
resources in [0, MaxWorth] range:
Step 1. Insert the explicit ratings in RR
Step 2. For each i, ni
1
Step 3. For each j, mj
1
Find },...,{ k
i
U
k
i
U satisfying the following conditions:
User i
U evaluates each of these users
There exists an explicit evaluation from users
k
i
k
iUU ,..., to j
R
We compute the value ),( jiRR using (7)
Algorithm 3. Resource associate ratings
compute algorithm
Remark: Obviously if does exist no user j
i
U,
mj
1 that evaluated resource j
R
than 0),(
j
RiRR . Finally we obtain the
value of RR that represents the implicit and
explicit evaluations that users associated to
the system resources at a given moment.
With these values the system can recommend
a hierarchy relevant resources for each user.
4. Experiments with StarTrust
To test the local trust algorithm we need the
communities for this testing. In the
specialized literature the local trust metric
tests proposed have been accomplished on
the dates from Epinions in (Massa, Avesani
2007) or in (Zieglar, Lausen 2004) have been
created their own dates for tests. In this work
we chose to analyze and to generate data that
are useful in obtaining conclusions on the
system, without introducing factors which
would not be necessary or could even impede
a correct analysis. The generation of the test’s
data is made by using a generator called
DataTestGenerator (M. Breaban, 2009),
which can be customized to generate
different cathegories of explicit evaluations
which mirror evaluations that can also be
found within the online communities.
Studies in Informatics and Control, Vol. 20, No. 2, June 2011 http://www.sic.ici.ro 151
Tests and results
Use Case. We want to study trust propagation
and how works StarTrust in a community of
15 users that have 20 resources.
Remarks: This small number of users and
resources was chosen considering graphic
limitations for visualization.
Using the data generator we obtained the
explicit user-user evaluations:
0.00 0.00 0.00 0.00 0.00 4.00 0.00 3.00 0.00 0.00 0.00 0.00 4.00 0.00
0.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 4.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 1.00
0.00 0.00 3.00 0.00 0.00 0.00 0.00 0.00 3.00 0.00 0.00 0.00 1.00 0.00
0.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 4.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 2.00 0.00
0.00 0.00 4.00 3.00 0.00 0.00 0.00 0.00 3.00 0.00 0.00 0.00 0.00 0.00
0.00 3.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 4.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 3.00 0.00 0.00 3.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 2.00
0.00 0.00 0.00 0.00 3.00 0.00 3.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 4.00 1.00
0.00 0.00 0.00 0.00 0.00 4.00 3.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
0.00 3.00 0.00 1.00 2.00 0.00
0.00 0.00 0.00 0.00 0.00 0.00 0.00 3.00 0.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 0.00
0.00 4.00 0.00 0.00 3.00 3.00 0.00 0.00 0.00 0.00 0.00 2.00 0.00 0.00
1.00 2.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 3.00 0.00 0.00 0.00
1.00 0.00 0.00 0.00 1.00 0.00
4.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 1.00
3.00 0.00 3.00 0.00 0.00 4.00 4.00 0.00 3.00 3.00 0.00 0.00 2.00 0.00
0.00 0.00 4.00 1.00 0.00 4.00
If we consider that the community is a new
one, composed for example with students that
will not easily change the identity, we can
consider TC
P=2.5. Using trust computing
component from StarTrust we obtain the
implicit evaluations.
Also, using trust recommendation component
we obtain the matrix that contains the explicit
generated ratings for resources and implicit
computed ratings for resources:
0.00 0.00 0.00 0.00 3.00 4.00 3.00 3.00 0.00 0.00 0.00 0.00 4.00 0.00
0.00 0.00 0.00 0.00 4.00 1.00
0.00 0.00 2.90 0.00 0.00 0.00 0.00 0.00 4.00 0.00 0.00 0.00 1.30 0.00
0.00 0.00 0.00 0.00 0.00 1.00
0.00 0.00 3.00 0.00 0.00 2.90 0.00 0.00 3.00 0.00 0.00 0.00 1.00 0.00
0.00 0.00 0.00 0.00 0.00 2.10
0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 4.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 2.00 0.00
0.00 0.00 0.00 0.00 0.00 3.40 2.80 0.00 0.00 0.00 0.00 0.00 0.00 0.00
0.00 2.80 0.00 1.60 2.00 0.00
0.00 0.00 4.00 3.00 0.00 0.00 0.00 0.00 3.00 0.00 0.00 0.00 0.00 0.00
0.00 3.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 4.00 0.00 3.00 0.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 3.00 0.00 0.00 3.00 2.90 0.00 0.00 0.00 0.00 0.00 0.00 0.00
0.00 2.90 0.00 1.30 2.10 2.00
0.00 0.00 2.90 0.00 3.00 0.00 3.00 0.00 2.90 0.00 0.00 0.00 1.30 0.00
0.00 0.00 0.00 0.00 4.00 1.00
0.00 0.00 0.00 0.00 0.00 4.00 3.00 0.00 0.00 0.00 2.70 0.00 0.00 0.00
1.90 3.00 0.00 1.00 2.00 0.00
2.80 0.00 0.00 0.00 0.00 0.00 0.00 3.00 0.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 2.20
0.00 4.00 0.00 0.00 3.00 3.00 0.00 0.00 0.00 0.00 3.00 2.00 0.00 0.00
1.00 2.00 0.00 0.00 1.00 0.00
0.00 2.80 0.00 0.00 2.60 2.60 0.00 0.00 0.00 0.00 3.00 2.40 0.00 0.00
1.00 2.40 0.00 0.00 1.00 0.00
4.00 0.00 0.00 0.00 2.90 0.00 2.90 0.00 0.00 0.00 0.00 0.00 0.00 0.00
0.00 0.00 0.00 0.00 3.70 1.00
3.00 2.80 3.00 0.00 2.60 4.00 4.00 2.80 3.00 3.00 0.00 2.40 2.00 0.00
2.20 2.40 4.00 1.00 0.00 4.00
From this matrix we extract hierarchy of
relevant resources for each user.
Table 2. Hierarchy resources vision of every
member of community
User Identification Resources sorted by relevance
User Id=1 6 13 19 5 7 8 20
User Id=2 9 3 13 20
User Id=3 3 9 6 20 13
User Id=4 9 19
User Id=5 6 7 16 19 18
User Id=6 3 4 9 16
User Id=7 6 8
User Id=8 3 6 7 16 19 20 18
User Id=9 19 5 7 3 9 13 20
User Id=10 6 7 16 11 19 15 18
User Id=11 8 1 20
User Id=12 2 5 6 11 12 16 15 19
User Id=13 11 2 5 6 12 16 15 19
User Id=14 1 19 5 7 20
User Id=15 6 7 17 20 1 3 9 10 2 8 5 12
16 15 13 18
We notice that in spite that the user with
ID=2 ( 2
U) has evaluated in explicit mode
only one resource (resource with ID=9 -
9
Rwith a rating with value 4 depicted in
matrix with user-user explicit evaluations),
the system will be able to suggest a resource
hierarchy suitable with his preferences,
considering trust evaluations that he associate
to other users.
We analyze the recommendations provided
by the system to user 2
U. The resource
3
Rwas recommended because user
2
Uevaluated 3
Uwith a 4 rating and this user
have already an evaluation for resource 3
R,
therefore StarTrust uses this experience to
recommend new resources.
We remark that in Table 2 we have the entire
resource hierarchy accessible for users, and
from the matrix of explicit and implicit user-
resource evaluations, we may identify the
computed ratings for each resource and easily
we may establish a threshold of evaluations,
that will give the possibility to obtain the two
sets of resources recommended or
un-recommended.
If for our example we use a threshold with
value 2 than from matrix of explicit and
http://www.sic.ici.ro Studies in Informatics and Control, Vol. 20, No. 2, June 2011
152
implicit user-resource evaluations we may
observe that the recommended resources for
user 2
U will be 9
Rand 3
Rin this order. Thus
the system ensures that the user will not have
to interact with irrelevant resources for him
and his actions in the community can be safer
and more efficient. The user also, has its own
image of the users from the system.
Using reputation component from StarTrust
we obtain the following results:
As we specified previously, we considered a
student community and for such a profile we
may consider the following control parameters
associated to reputation compute value:
reputation control parameter l
RC
P will
take as value the media of ratings for
level l, where },...,,{ 21 k
lll
i
lEvEvEvN is
the set of ratings that user i
U received in
the interval specified by level l,
MaxWorthl 1, MaxWorth=5
for level importance control parameter
j
LW
P we considered value 1 for levels.
Therefore, using the functionality offered by
all 3 components, StarTrust will provide in
the context of a huge amount of data and a
huge number of users, an interaction with
resources and relevant users for each user, or
for new users a rapid integration in
the system.
Comparison with MoleTrust
The proposed system from this paper is based
on a local trust metric. The majority of other
systems are based on a global metric (e.g.
PageRank, eBay, Amazon, etc.). We
mentioned in section 2 the proposals for local
trust metrics. Some research works were
realized concerning the personalization of
PageRank (Haveliwala et al 2003) algorithm.
The closest metric to our research is
MoleTrust proposed by (Massa, Avesani
2006). Analyzing this metric we have some
observations that sustain our modeling point
of view in StarTrust.
In StarTrust, we consider that to eliminate the
graph circuits will affect the results accuracy.
We consider that the MoleTrust argument
regarding computing time reduction will
affect the trust propagation. In StarTrust the
computing algorithm is equivalent with
solving a linear system of equations (Alboaie
2009). To analyze the effect of the results
affected by eliminating circuits we consider
the following example: user A may be
interested that B and C are evaluated in a
contradictory mode. This aspect is reflected
in the trust value that A will associate to C. If
C is evaluated in an unjust mode by B, in
system will be other evaluations that may be
favorable to C and will diminish the effect of
B rating; also the reputation of C will not be
strong affected. But to destroy a circuit from
B to C may represents an eliminated
information source for A, and we will obtain
values that will not totally reflect the reality.
In MoleTrust, is introduced a factor named
trust_threshold, that represents the threshold
among the considered and un-considered
ratings in trust computing. We mention that
this factor will also affect trust propagation
and the access to suitable resources.
Let us consider the following situation: a user
A has no evaluation (explicit or implicit)
associated to users B, C, D. A does not
evaluated a resource R. In this case A may
consider the reputation of system users.
We consider that B has evaluated with a big
rating the resource R and B has a good
reputation. Focusing on that, A will consider
that R resource is good and will try to use it.
But will discover that the resource is not that
he need and then will evaluate B with 0.1. In
MoleTrust such an evaluation (scaled in
interval 0-1 become 0.02) will be effectively
ignored and in this case the penalty of an
evaluation is not possible to be realized. So,
the preference of the user to do not take into
account some resources evaluated by B is an
aspect unsurprising in MoleTrust.
Other analyze was realized concerning the
MoleTrust trust computing formula:
Table 3. Reputation values for community members
User Id 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Reputation
value 3.00 0.00 3.31 0.00 5.00 0.00 0.00 3.12 3.53 3.05 5.00 1.93 2.85 1.38 0.00
Studies in Informatics and Control, Vol. 20, No. 2, June 2011 http://www.sic.ici.ro 153
rspredecessoi
rspredecessoi itrust
itrustuiedgetrust
utrust ))((
))(*),(_(
)(
Use-case 1: we consider the following case:
Table 4. Explicit evaluations
Source User Id 1 2
Evaluated User Id 2 3
MoleTrust/StarTrust
Rating 0.2/ 1 1/5
Using MoleTrust formula, trust that user 1
will provide to user 3 will be computed so:
1
)2( )3,2(_*)2(
)3( trust edgetrusttrust
trust , in
evaluation interval [0, 1].
As we observe in this case when only one arc
will enter in the evaluated node, the trust
value will be the trust value of the node that
has a direct arc in 3, i.e. node 2. The trust of 1
in 2 is not used, being simplified.
In StarTrust in this case the value determined
by 1 to 3 in [0, 5] interval will be:
1)),(_*
),((
*1 1
),(_
32
2131
UUUUWE
UUWU
MaxWorth
UUUUWI ,
To compute the factor we used was PTC = 0.
If we consider that to the users we can grant a
certain trust level we may use for PTC =1.5. In
this case: WI_UU(U1,U3)=1,85.
Intuitive, the value obtained with StarTrust is
more relevant than the value obtained with
MoleTrust. If a user A will evaluate a user B
with 1 (specifying the distrust in B) and B
will evaluate C with 5, then is sure that A
will not wish to evaluate C with 5 using only
the experience of B.
Use-case 2: This example is assumed from
(Massa, Avesani 2006) with MoleTrust
metric and we specify some remarks using
Star Trust metric.
In (Massa, Avesani 2006) using MoleTrust
from the point of view of user 3
U we have:
767.0
18.0 9.0*16.0*8.0
)1()2( )3,1(_*)1()3,2(_*)2(
)3(
trusttrust edgetrusttrustedgetrusttrust
trust
or 3.83 in scale [0, 5].We remark that metric
does not compute values less than a certain
threshold, so the evaluation from user 5 to 4
is ignored. If the value will be considered,
then
)3(trust 0.778 or 3.89 in scale [0, 5].
Table 5. User-User explicit evaluations from (Massa, Avesani 2006)
Source User
Id 1 2 2 4 4 5 5 4 6 7 8 8 9 10
Evaluated
User Id 3 3 4 3 2 4 2 7 7 6 7 9 10 9
MoleTrust/
StarTrust
Rating
0.9/
4.5 0.6/
3 0.5/
2.5 1.0/
5 0.9/
4.5 0.1/
0.5 0.8/
4 0.9/
4.5 0.9/
4.5 0.5/
2.5 1/
5 0.8/
4 0.7/
3.5 0.3/
1.5
Figure 3. The graph represents the community composed by 10 users. A is the graph with
explicit ratings with values for arcs in [0, 5] scale (see Table 5 for Mole Trust scale); B is the
graph for explicit and implicit ratings obtained with StarTrust; continue arcs represents explicit
evaluations and implicit evaluations are represented by dot arcs
http://www.sic.ici.ro Studies in Informatics and Control, Vol. 20, No. 2, June 2011
154
Using StarTrust, we obtain trust propagation
as in the Figure 3. The value
64,2),(_ 35 UUUUWI or 0.528 in interval
[0, 1], for TC
P=1.5.
Using an empiric analyze, we remark that U5
will express the distrust in U4 giving a 0.5
rating. In the value of ),(_ 35 UUUUWI it can
be find this evaluation which will determine a
lower implicit rating of 5
U for 3
U. For
TC
P=4 we obtain 95,3),(_ 35
UUUUWI .
This big value of the parameter specifies that
we have a community with a high trust level
for members.
We also remark that depending on different
online communities with different profiles
StarTrust allows to realize an adaptation of
trust computing depending on these situations.
5. Conclusions and future work
The paper presents a trust and reputation based
model, able to help users from online
communities to interact with appropriate users
and resources. In these mode good decisions
and few time-consuming actions concerning
resource management can be realized.
We enumerate a set of consequences
resulting from how the system was modeled.
Resources relevant to a user (even those
new) are visible in the top list of resources
The system ensures that a user will see
resources prioritized, in a similar manner
with those which resemble
Users who add spam resources will see
more spam because the system groups
users according to their preferences
Users are encouraged to make
proper evaluations
It will not happen as in eBay, where users
give the most positive ratings of fear of
possible revenge. In systems based on
StarTrust metric, there are no good or bad
ratings, there are interesting or uninteresting
ratings from the point of view of users. The
rating given by a user is pursue its goal,
namely to quickly access important resources
for it.
As we have seen with a trust metric, the trust
can be propagated in the community. This is
an advantage that can be used by a
recommendation system. If we have a new
user U evaluating many resources, through
standard mechanisms those evaluations cannot
be propagated. But if user U has evaluated a
set of users, than using a trust metric as used in
Star Trust, the system can associate a greater
number of resources and the recommendation
mechanism is more efficient.
Moreover, taking into account user-user
evaluations, the system ensure a faster
integration of new users in the system.
Additionally, if they entered in the
community by an invitation of an older
member of the community, this invitation can
be considered an explicit evaluation between
users. Using StarTrust trust metric the system
will be capable to recommend resources from
the first moment.
The design of StarTrust will allow that our
system can be integrated in different online
communities as: education, e-health
(Chiorean et all 2010), social networks, etc.
As future research direction we will study the
behaviour of the model in real communities
as medical and educational domains. Also,
we shall study the reputation in an online
community not limited to a given domain.
REFERENCES:
1. ALBOAIE, L. PReS – Personalized
Evaluation System in a WEB
Community, Proceedings of the
International Conference on E-Business,
2008, pp. 64-69.
2. ALBOAIE, L., T. BARBU, An
Automatic User Recognition Approach
within a Reputation System Using a
Nonlinear Hausdorff Derived Metric,
Numerical Functional Analysis and
Optimization, Pub. Taylor & Francis,
Vol. 29, Issue 11 & 12, 2008,
pp. 1240-1251
3. ALBOAIE, L., Studies on Trust
Modeling and Computation of
Reputation in Online Communities,
Ph.D. thesis, Al. I. Cuza University,
Iasi, 2009
4. ALBOAIE, L., S. BURAGA, Trust and
Reputation in e-Health Systems,
Proceedings of the International
Studies in Informatics and Control, Vol. 20, No. 2, June 2011 http://www.sic.ici.ro 155
Conference on Advancements of
Medicine and Health Care through
Technology; IFMBE Cluj-Napoca, vol.
26, 2009, pp. 43-46.
5. ALBOAIE, L., M.-F. VAIDA, Modeling
of Trust to Provide Users Assisted
Secure Actions in Online
Communities, NDT, Communications in
Computer and Information Science of
Springer Lecture Notes, Prague, Series-
87, 2010, pp. 369-382.
6. BERNERS-LEE, T., et al., Uniform
Resource Identifiers (URI), General
Syntax, RFC 2396, IETF, 1998.
7. BREABAN, M., L. ALBOAIE, H.
LUCHIAN, Guiding Users within Trust
Networks Using Swarm Algorithms,
IEEE Congress on Evolutionary
Computation (IEEE CEC 2009), 2009,
pp. 1770-1777.
8. CHIOREAN, L. D., L. SUTA, M.-F.
VAIDA, M. HEDESIU, Medical Fusion
Components for a Web Dedicated
Application, Studies in Informatics and
Control, Vol. 19, Issue 4, 2010, pp. 435-
444, ISSN 1220-1766.
9. GAMBETTA, D., Trust: Making and
Breaking Cooperative Relations; Can
We Trust Trust?, Basil Blackwell, 2000,
pp. 213-237.
10. GOLBECK, J., Personalizing
Applications through Integration of
Inferred Trust Values in Semantic
Web-Based Social Networks,
Proceedings of the Semantic Network
Analysis Workshop at the 4th Intl.
Semantic Web Conference, Galway 2005.
11. GOLBECK, J., Computing and
Applying Trust in Web-based Social
Networks PhD thesis, Univ. of
Maryland, 2005.
12. GOLBECK, J., Computing with Social
Trust, Springer Publisher, 2009, pp. 338.
13. GUHA, R., Open Rating Systems,
Technical report, Stanford Knowledge
Systems Laboratory, Stanford, CA,
USA, 2003.
14. HARDIN, R., Trust and
Trustworthiness, Russell Sage
Foundation, New York, 2002.
15. HARMON, A., Amazon Glitch
Unmasks War of Reviewers, The New
York Times, 14 February 2004.
16. JOSANG, A., R. ISMAIL, C. BOYD, A
Survey of Trust and Reputation
Systems for Online Service Provision,
Decision Support Systems, Vol.43, no.2,
Elsevier Science Publishers., 2007,
pp. 618-644
17. LEVIEN, R., Attack Resistant Trust
Metrics, PhD thesis, UC Berkeley,
Berkeley, CA, USA, 2003.
18. MASSA, P., P. AVESANI,
Controversial Users Demand Local
Trust Metrics: an Experimental Study
on Epinions.com Community,
Proceeding of the 25th Conference, 2005,
pp. 121-126.
19. MASSA, P., P. AVESANI, Trust-Aware
Bootstrapping of Recommender
Systems, ECAI'06 Workshop on
Recommender Systems, 2006, pp. 29-33.
20. MASSA, P., P. AVESANI, Trust
Metrics on Controversial Users:
Balancing Between Tyranny of The
Majority and Echo Chambers, Intl.
Journal on Semantic Web and
Information Systems, vol. 3(1), 2007,
pp. 39-64.
21. MUI, L., M. MOHTASHEMI, A.
HALBERSTADT, A Computational
Model of Trust and Reputation,
Proceedings of the 35th Annual Hawaii
Intl. Conference on System Sciences,
2002, pp. 2431- 2439.
22. NURMI, P., A Bayesian Framework for
Online Reputation Systems, AICT-
ICIW’06, 2006, p. 121.
23. PAGE, L., S. BRIN, R. MOTWANI, T.
WINOGRAD, The PageRank Citation
Ranking: Bringing Order to the Web,
Technical report, Stanford Digital Library
Technologies Project, 1998.
24. RAHMAN, A., S. HAILES, Supporting
Trust in Virtual Communities, System
Sciences, Proceedings of the 33rd Annual
Hawaii Intl. Conference on System
Sciences, vol. 1, 2000, p. 9.
25. SABATER, J., C. SIERRA, Social
ReGreT, a Reputation Model Based on
http://www.sic.ici.ro Studies in Informatics and Control, Vol. 20, No. 2, June 2011
156
Social Relations, SIGecom Exchanges
3.1, 2002, pp. 44-56.
26. SEPANDAR, D. K., M. T.
SCHLOSSER, H. GARCIA-MOLINA,
The Eigentrust Algorithm for
Reputation Management in P2P
Networks, 12th Intl. Conf. on World
Wide Web, 2003, pp. 640-651.
27. SUNSTEIN, C. R., Republic.com 2.0,
Princeton University Press, 2009, p. 241.
28. DE TOCQUEVILLE, A., Democracy in
America, The Library of America, New
York, 1835-1840.
29. ZIEGLER, C., G. LAUSEN, Analyzing
Correlation between Trust and User
Similarity in Online Communities, 2nd
Intl. Conf. on Trust Management,
Springer, 2004, pp. 251-265.
30. WU, Z., X. YU, J. SUN, An Improved
Trust Metric for Trust-Aware
Recommender Systems, First Intl.
Workshop on Education Technology and
Computer Science, vol. 1, 2009,
pp. 947-951.
... The cloud is also an important ERP trend, especially today. The industry has shown itself sceptical about adopting this delivery model at first, refusing to place sensitive data outside the company firewall [1]. However, as the advantages of this delivery model became obvious, hesitations have been evaporating either naturally or by migrating through intermediate solutions, as a private cloud deployed inside the company environment [11,15,21]. ...
... Traditionally, on premise solutions require additional infrastructure and costs compared to cloud based offerings, from IT employees and specific technical skill, to deployment and maintenance overhead. Nevertheless, usually caused by privacy concerns towards having their data system outside the company firewall [1], a lot of modern organizations choose the private cloud as a stepping stone, compromising responsibility externalization in favour of increased control and security metrics. Cloud-based integration options do not always imply not needing additional infrastructure, but rather provide the option. ...
Full-text available
Article
The Enterprise Resource Planner is a half-centenary giant that gained popularity in the industry-software market segment throughout the years. Influenced by the latest technology trends, such as mobility and cloud, it keeps software integration as one of its greatest challenges. The present article introduces a cloud-based Field Service Management solution, meant to empower mobility, manage workflows and enhance process automation for ERP enabled companies. Motivated by ERP vendor dependence and a limited adoption rate, the solution's objective is to design and implement optimized cloud generic integration capabilities. Using a commercial, world-wide-distributed environment as a continuous feedback and validation method, the results of our implementations show good application performance and feature offering, but rather low usability metrics and sub-optimal commercial adoption. Based on a post-implementation critical analysis of the existing integration applications and the legacy solution infrastructure, the paper proposes an improved integration strategy, zooming in from business requirements and API constraints, to application architecture and data flows that would overcome technical limitations, reduce prototyping and integration time and decrease maintenance efforts. The implementations and improvement model represent specific practical contributions to a productive, enterprise-oriented solution, proven both successful and highly competitive in respect to its market competitors.
... An LMS (Nagarajan & Jiji, 2010;Lonn & Teasley, 2009) allows the management of learning resources for the various actors who are part of an organization, enabling the formation of a virtual learning community (Alboaie & Vaida, 2011;Ribón & de Miguel & Ortíz, 2009), where the actors interact with each other, to carry activities where they're required and information of territory is captured. For example, academic institutions have LMS applications with documentary information in the territories (geography, hydrography, economics, etc.). ...
... CC refers to applications and services running on distributed networks that use resource virtualization techniques and can generally be accessed via the Internet using protocols and standard network services. CC ensures the transparency of physical resources and their configurations, with end users having the perception that the resources they have are theoretically unlimited [1]. ...
Full-text available
Article
Cloud Computing solutions are nowadays one of the leading issues that small, middle and large enterprises face. The paper presents the benefits of the implementation of cloud solutions in small and medium businesses, factors which can have an influence on the level of the use of cloud systems and appropriate areas in which cloud systems can be used. Moreover, further development of this approach has been described. For academia, the results of the paper can serve as a useful extension for future works in this field. For businessmen, the findings would benefit them in terms of arriving at a decision toward cloud computing implementation and defining better strategies for cloud computing deployments.
... An interesting use case of using PDS are social networks and other systems which, besides private data, manage also trust and reputation data [5]. ...
Full-text available
Conference Paper
With the increased use of Internet, governments and large companies store and share massive amounts of personal data in such a way that leaves no space for transparency. When a user needs to achieve a simple task like applying for college or a driving license, he needs to visit a lot of institutions and organizations, thus leaving a lot of private data in many places. The same happens when using the Internet. These privacy issues raised by the centralized architectures along with the recent developments in the area of serverless applications demand a decentralized private data layer under user control. We introduce the Private Data System (PDS), a distributed approach which enables self-sovereign storage and sharing of private data. The system is composed of nodes spread across the entire Internet managing local key-value databases. The communication between nodes is achieved through executable choreographies, which are capable of preventing information leakage when executing across different organizations with different regulations in place. The user has full control over his private data and is able to share and revoke access to organizations at any time. Even more, the updates are propagated instantly to all the parties which have access to the data thanks to the system design. Specifically, the processing organizations may retrieve and process the shared information, but are not allowed under any circumstances to store it on long term. PDS offers an alternative to systems that aim to ensure self-sovereignty of specific types of data through blockchain inspired techniques but face various problems, such as low performance. Both approaches propose a distributed database, but with different characteristics. While the blockchain-based systems are built to solve consensus problems, PDS's purpose is to solve the self-sovereignty aspects raised by the privacy laws, rules and principles.
... ),Benkler (2006),Castells (2010) care dezbat formele de organizare a societății prezentului sau viitorului precum și implicațiile Internetului și a tehnologiilor de rețea în activitatea economico-socială.Încrederea și reputația sunt aspecte esențiale ale existenței și continuității în bune condiții a unui model de cloud de comunitate. Experiența din domeniul comunităților on-line(Alboaie & Vaida, 2011) poate fi preluată, adaptată și formalizată și în domeniul cloud-ului de comunitate stabilind regulile generale de operare și reacție în cazul abaterilor de la normele standard de funcționare a sistemului. ...
Full-text available
Book
Apariția fenomenului, cunoscut sub denumirea de cloud computing, reprezintă o schimbare fundamentală în felul în care sunt dezvoltate, livrate, actualizate, întreținute și plătite serviciile IT din cadrul organizațiilor moderne. Cloud computing semnifică convergența a două tendințe majore ale IT-ului zilelor noastre: eficiența IT - unde puterea calculatoarelor moderne este utilizată mai eficient printr-o scalare înaltă a resurselor de hardware și software și agilitatea de business – unde tehnologia informațională poate fi folosită ca instrument competitiv pe piață prin livrare rapidă, loturi paralele de procesare, utilizarea instrumentelor de inteligență a afacerilor, care necesită calcul intensiv și aplicații mobile interactive și care răspund în timp real cerințelor utilizatorului.
... Trust is part of our daily life and thus can be relied upon when making access control decisions, which can be realized by using trust to provide network security 2,3 . In some early work 4,5 , trust models have been proposed to control anonymity, unpredictability and uncertainty. In recent years, many researchers have applied trust in dynamic environments. ...
Full-text available
Article
Access control is an important element in network security. Meanwhile, trust provides a new direction for access control in open network environments. Based on the dynamic nature of trust, we study the temporal and spatial characteristics in the security of society and propose the concept of scenario trust in which we consider four factors: access time, place, history behavior and risk control strategy. First, we apply fuzzy clustering method and information entropy theory to the design of an algorithm for allocating weights to the factors of scenario trust and, consequently, quantify scenario trust. Then, we introduce the notion of trust threshold for fine-grained access permission management and describe the rules that map scenario trust values to access permissions. On this basis, we propose a dynamic access control model based on scenario trust. Simulation results show that the proposed method is effective and can defend joint fraud. The model exhibits good scalability and can meet the need of dynamic access control in open network environments.
Article
The paper presents a unified solution for detecting, processing, validating, and exporting various data expressed by different existing semantic markups (microformats, HTML5 microdata, and RDFa) embedded into Web pages. The developed tool is available as a browser extension and a bookmarklet, providing a modular architecture.
Article
There is a range of organizations which have expertise related to conflict areas (NGOs, governments, etc.). Many of these organizations have put their expertise in digital media that can be accessed by its members through technology platforms. The information in conflict areas is useful for decision making, to generate public policies for the control and humanitarian support. Organizations are willing to cooperate and share their expertise with others, but there is a technological problem that is making the above impossible and therefore causing isolation. This is caused by the inability of the technology platforms that each organization has, to locate and interoperate with expertise from other platforms. This paper describes an experience in which mobile technologies are used as sensing devices to identify positioning and movement of people who are in conflict areas and to capture multimedia information in such territory. These sensing devices allow access to a portfolio of services from a cooperating cloud of territorial information in the region of Montes de Maria (Colombia), which is a territory with political and public order conflicts. With this experience, a solution to the problem of technological isolation is offered and access to the expertise shared by peers in the community.
Article
Opinion dynamics is a complex collective behavior in human societies. When individuals exchange opinions with others, they usually adopt a bounded confidence rule and only accept the opinions within the confidence range. Furthermore, individuals are heterogeneous in real social systems. Thus, they have distinct confidence levels, and also play different roles in the collective opinion formation. In this paper, a leader-follower bounded confidence model is proposed for a group of social agents, who have heterogeneous confidence levels, at the same time, come from two subgroups: opinion leaders and opinion followers. Simulation results are obtained for the collective opinion evolution influenced by three factors: the leader fraction, the group size and the trust degree. The results show that the roles of opinion leaders are remarkable when the opinion followers have high confidence levels and trust degrees.
Article
Patients rely on the physician for recommending an adequate treatment path for their medical condition. Since there are several possible paths to choose from, the optimal solution should be selected. This paper proposes a method to optimize medical services as business workflows. Such a workflow consists in a set of medical services. Each medical service has different attributes such as: average wait time, rating, distance and cost. For each service type several alternatives of concrete medical services exists. Determining the optimal combination of medical services for a business workflow is an NP-hard problem. Genetic Algorithms (GA) and Differential Evolution (DE) are used to find the optimal solution. The results indicate that DE approach converges faster.
Full-text available
Article
A web dedicated application is very useful for assisted diagnosis in healthcare domain. Medical images acquired with different medical image modalities contain important diagnostic information. The combination of complementary data from different images can lead to more important information. The fusion process allows combination of salient feature of these images. In this paper we present methods for medical image fusion, based on discrete wavelet transform. These methods are implemented in a web distributed application, using Java technology. The web dedicated application will integrate components able to be discovered in a ubiquitous computing system. The components are developed in dedicated Java packages that allow physicians to realize a fusion from a remote place. Considering radiology images we approve the functionality of the application by calculating significant quantitative parameters for fusion process evaluation.
Full-text available
Article
In today's connected world it is possible and indeed quite common to interact with unknown people, whose reliability is unknown. Trust Met- rics are a technique for answering questions such as "Should I trust this person?". However, most of the current research assumes that every user has a global quality score everyone agree on and the goal of the technique is just to predict this correct value. We show, on data from a real and large user community, Epinions.com, that such an assumption is not re- alistic because there is a significant portion of what we call controversial users, users who are trusted by many and distrusted by many: a global agreement about the trustworthiness value of these users does not exist. We argue, using computational experiments, that the existence of contro- versial users (a normal phenomenon in complex societies) demands local trust metrics, techniques able to predict the trustworthiness of a user in a personalized way, depending on the very personal views of the judging user as opposed to most commonly used global trust metrics which as- sume a unique value of reputation for every single user. The implications of such an analysis deal with the very foundations of what we call society and culture and we conclude discussing this point by comparing the two extremes of culture that can be induced by the two different kinds of trust metrics: tyranny of the majority and echo chambers.
Article
In today’s connected world, it is possible and indeed quite common to interact with un-known people whose reliability is unknown. Trust metrics are a technique for answering questions such as “Should I trust this person?” However, most of the current research assumes that every user has a global quality score everyone agrees on and the goal of the technique is just to predict this correct value. We show on data from a real and large user community, Epinions.com, that such an assumption is not realistic because there is a significant portion of what we call controversial users, users who are trusted by many and distrusted by many: a global agreement about the trustworthiness value of these users does not exist. We argue, using computational experiments, that the existence of controversial users (a normal phenomenon in complex societies) demands local trust metrics, techniques able to predict the trustworthiness of a user in a personalized way, depending on the very personal views of the judging user as opposed to most commonly used global trust metrics, which assume a unique value of reputation for every single user. The implications of such an analysis deal with the very foundations of what we call society and culture and we conclude discussing this point by comparing the two extremes of culture that can be induced by the two different kinds of trust metrics: tyranny of the majority and echo chambers.
Article
In this paper, we provide an automatic unsupervised recognition technique for Web community user reputations that uses a special nonlinear metric. First we describe the general framework for reputation systems. Then, we propose a feature extraction approach for the reputation system users. The resulting feature vectors (reputations) are clustered with an unsupervised classification algorithm using a nonlinear distance, derived from the Hausdorff metric for sets.
Chapter
The Internet is an amazingly powerful tool for connecting people together, unmatched in human history. Yet, with that power comes great potential for spam and abuse. Trust metrics are an attempt to compute the set of which people are trustworthy and which are likely attackers. This chapter presents two specific trust metrics developed and deployed on the Advogato Website, which is a community blog for free software developers. This real-world experience demonstrates that the trust metrics fulfilled their goals, but that for good results, it is important to match the assumptions of the abstract trust metric computation to the real-world implementation.