PRES – PERSONALIZED EVALUATION SYSTEM IN A WEB
COMMUNITY
A Conceptual Model Designed to Evaluate Reputation in Order to Achive a
Personalised View on the System for Each User
Lenuţa Alboaie
Department of Computer Science, “A.I. Cuza” University of Iaşi – 16, Berthelot, 700483 Iaşi, Romania
Keywords: Reputation systems, evaluation, resource, Web community.
Abstract: The purpose of the PRES model is to build a flexible and easy way to manage resources in a personalized
manner. Our proposed model assures for every user that his preferences are important and permits the
formation of some homogenous groups on the basis of these preferences. The homogeneity is due by the
relations resulted from the explicit and implicit evaluations of resources. The purpose of the proposed model
is to build a flexible way to filter irrelevant resources for users. In this way, a user which is member to a
community based on the PRES model will dynamically see information that he/she is most interested in.
1 PREAMBLE
In this moment the WWW space stores large
amounts of data which are continuously growing.
The main problem that appears is to find solution to
use efficiently the existent resources.
A first step to solve this problem is to associate
metadata to resources. As a fact, it is a manual
classification process performed by the user (E.g.
delicious, digg.com). This direction is a part of
explicit Web that is realized through explicit
activities as tagging or digging.
An important direction, using the above solution, is
to obtain data/information by observing and
analyzing the user actions. Thus, we enter the space
that is known as implicit Web (O’Reilly, 2005). An
important drift of it is collective intelligence domain
(T. Segaran, 2007).
In this context, this work proposes the analysis
and projection of a prototype of a reputation’s
personalized evaluation system in a Web community
(PRES - Personalized Resource Evaluation System).
The originality of this approach consists in the
chosen perspective to accomplish the evaluation.
This work is structured as follows: in section 2
we describe a short survey on the present situation
(O’Reilly, 2005, H.Zhuge, 2008). In the next section
we present the problem, we explain why such a
system is necessary and we present the proposed
model. In the fourth section we analyze the benefits
of the proposed system. The article ends with an
overview on the discussed domain, mentioning the
future directions.
2 ACTUAL SITUATION
At this moment there are many sites that collect
various information about thousands or even
millions of people on the Web. This information is
obtained often without even interrupt user actions
with questions. His behavior and profile can be
obtained from this information using different
techniques like machine learning and statistical
methods
In the collective intelligence spectrum we have
two different approaches, one exists due the
information furnished by users (e.g. Wikipedia). The
other part of the spectrum is based on different
algorithms which allow obtaining new information
that enhance the user experience. An important
example in this sense is Google, which uses links to
rank web pages, but also collects and process data
obtained from situations when advertisements are
clicked.
Other examples consist of web communities that
use recommendation systems (Massa,
B.Bhattacharjee, 2004). In this cases there are
64
Alboaie L. (2008).
PRES – PERSONALIZED EVALUATION SYSTEM IN A WEB COMMUNITY - A Conceptual Model Designed to Evaluate Reputation in Order to Achive
a Personalised View on the System for Each User.
In Proceedings of the International Conference on e-Business, pages 64-69
DOI: 10.5220/0001914100640069
Copyright
c
 SciTePress

collected information like purchasing history and
user characteristics, and the system make proper
recommendations based on them (e.g. Amazon,
Netflix).
Other examples consist of web systems which
use reputation systems (
Golbeck, Hendler, 2005).
Reputation systems are extremely useful in those
communities where the users have to interact with
some resources posted by other users or they have to
interact with other users. (E.g. YouTube, Slashdot,
Flicker). In these situations, using experience of
other users would be very useful. Also, reputation
systems are useful in setting some evaluation levels
for users and resources (e.g. more or less interesting
resources). There are a variety of reputation systems.
A well-known system, mentioned before, is Google
Page Rank (A. Langville, C. Meyer, 2006) that is
based on complex algorithms that assure the web
page ranking.
Another reputation system is that used by eBay.
The system assures a feedback profile for each
member.
Each feedback consists of a positive, negative or
neutral value (these values are obtained from the
ratings of the transaction partners) and a short
comment.
Everything2 is a knowledge base that contains
reputations system both for users and their posted
articles. The system is based on anonymous votes of
other users which determine positive or negative
ratings. Negative evaluated articles are deleted. The
users are evaluated on the basis of the number of
their submitted articles (and not deleted) and on the
average of their associated values.
Such a system implies some problems: new users
posting articles that receive negative feedbacks may
appear. These articles will be deleted, thus
discouraging new postings by such users. Even the
experienced users hesitate to post new articles which
they consider as being not very good, because the
received negative feedbacks are not deleted. Also, in
this kind of system the re-actualization of older
articles is less appreciated.
Slashdot has a reputation system named karma.
In this system there are moderators that can make
the evaluations in a similar way to the system
Everything2. Every user may become moderator if
he has a good karma obtained on the basis of the
ratings associated to their comments. But this
moderator state is temporary until he uses the
available votes. This evaluation system is criticized
because it is weak on issues like Anonymous
Coward or sock puppets (R. Falcone, S.Barber, L.
Korba, M. Singh, 2002).
Another system we referred here before is
Wikipedia that represents an online community
containing a great number of users, but not using a
formal reputation computation mechanism.
As in the previously discussed systems, a less
visible user hierarchy exists. All users, on the basis
of their contribution, may receive the so-called
barnstar acknowledgement. Although one can
follow each user posting history, it does not exist a
particular rating system.
3 PRES MODEL PROPOSAL
3.1 Context
In section 2 we have discussed a set of reputation
systems (R. Falcone, S.Barber, L. Korba, M. Singh,
2002), but in all these related approaches we do not
find a personalized evaluation. In this section we
explain what a personalized evaluation means, from
our point of view.
In a Web community there exist a lot of
resources. There are human resources and other
types of resources. The people have either different
or similar profiles. Therefore, they are interested in
either different or similar resources.
We quantize this interest with values which are
provided by the user for other users or resources.
Also, this interest will have an indirectly computed
component. We give a simple example here, the
other cases being analyzed in section 3.2. We have
the situation when a user evaluates favorably one or
more users. These users evaluate favorably a given
resource. Even if the user does not evaluate directly
that resource we will consider an implicit favorable
evaluation (J.Golbeck, J. Hendler, 2006). Thus, the
user has the chance to access more relevant
resources for him.
In our system there is no it does not exist an
absolute value of good or bad resource
characteristic. A resource can be good for a set of
users but not useful for other set of users.
In section 3.2 we establish a set of metrics (J.
L. Mui,
2002), taken into account by the evaluation
mechanism, for the purpose of measuring the
usefulness of a resource for a given user.
Whenever new users become community
members they can interact with the users
corresponding to their preferences. Also, they will
be able to access much faster the proper resource set.
This represents the general direction our system is
based on.
PRES – PERSONALIZED EVALUATION SYSTEM IN A WEB COMMUNITY - A Conceptual Model Designed to
Evaluate Reputation in Order to Achive a Personalised View on the System for Each User
65

3.2 The Proposed Model
First we define the vocabulary used in the
developing model. We also specify the used
notations and their semantics. The system will
contain:
 Users which know other users.
 The list of the users considered to be
interesting for a user.
 Users nominated by a community as
evaluators. We use notations E
1
...E
n
to
indicate the community evaluators. These
evaluators are in fact some reviewers. They
will be useful for the new users which have
not established their own knowledge list
yet.
 Known person list of a user. Initially, it
contains the community reviewers list only.
 Resources – their definition is made
accordingly to the definition given by (T.
Berners-Lee, 1998).
So, in our system one considers as resources
everything having an identity (e.g. electronic
document, an image, a service and eventually a
collection of other resources). There are considered
as resources those that cannot be accessed directly
via Internet (e.g. human beings, organizations)
 Worth – this parameter is a metric. This
metric signifies a given rating, according by
a user to a resource or a user. Also, the
worth can be obtained (quantized)
indirectly.
This parameter – Worth – takes the following
values:
 1 = useless/spam
 2 = poor
 3 = worth attention
 4 = good
 5 = exceptional. We note this limit with
MaxWorth.
We think of using the 1-10 interval for possible
values for Worth metric, this approach assuring
higher granularity in resource evaluation. We prefer
the above specified selection to simplify the model.
In future works we will analyze if this aspect has a
major influence on the resource evaluation manner.
We will use a set of constructions which have the
following associated semantics. In fact, these
constructions can be mathematically considered as
functions (eventually partial functions) or, from the
implementation point of view, they are considered
associative tables:
 Explicit worth of a resource: WE_UR
(user, resource) – explicit worth, represents
the rating for a resource, this rating being
given manual by a user
 Implicit worth of a resource: WI_UR (user,
resource) – implicit worth, represents a
rating inferred from the set of existing
ratings from the known person list of a user
 Explicit worth of a user: WE_UU (user,
user) – explicit worth, represents the rating
for a user, and the rating is given manual by
the user to another user
 Implicit (deducted) worth of a user:
WI_UU (user, user) – measure how close
are his preferences to the others preferences
(The preference can be considered: the
accepting degree of a point of view or the
appreciation degree of a piece of art).
Implicit we consider that we have:
WI_UU (user, evaluator) = MaxWorth;
If an user evaluates an evaluator in an explicit
manner, then this evaluation - WE_UU (user,
evaluator) – will have priority.
 we consider the function
WU(user, user)
for every pair of (user, user) from a Web
community
Its value will be WE_UU (user, user) if there is
an explicit evaluation (different from 0), otherwise
its value will be WI_UU (user, user). So, let us
consider the users: U
x
and U
y
.
If the user Ux evaluates explicitly the user U
y
then the function WE_UU has a value different of 0
and the value of WU(U
x
,U
y
) will be WE_UU(U
x
,U
y
).
If U
x
does not make an explicit evaluation for
user U
y
then WU (U
x
,U
y
) value will be the inferred
value which is actually the value of WI_UU(U
x
,U
y
).
 we consider a function WR(user, resource)
for every pair (user, resource)
WR (user, resource) value will be WE_UR if the
user evaluates explicitly the resource, thus the value
of WE_UR exists. Otherwise WR value will be the
value of WI_UR.
Therefore, let us consider the user U
x
and the
resource Ry. WR (U
x
, R
y
) value will be WE_UR
(U
x
,R
y
) if user U
x
has explicitly evaluated the
resource Ry. Otherwise WR value will be WI_UR
(U
x
, R
y
) if the user U
x
did not evaluate the resource.
This value is based on the ratings to R
y
made by
users which are in known list of the user U
x
.
We will define the manner of computation of the
implicit values introduced above.
ICE-B 2008 - International Conference on e-Business
66

Implicit WI_UU Value Computation
Let us consider two users U
x
, U
y
from the Web
community. In order to define WI_UU(U
x
,U
y
) we
introduce the following partial functions:
 WIU_UU(U
x
,U
y
) – its value indicates the
deducted worth based on explicit
evaluations made by users to each other
 WIR_UU(U
x
, U
y
) – its value signifies the
deducted worth based on evaluations that
users do to the same resources
Defining WIU_UU value on the basis of the
explicit values
Let the users, whom we have explicit ratings
from user User
1
to be U
i
,
ki ≤
≤
1
, be U
1
,…U
k
.
Therefore we have the definition WE_UU
(User
1
, U
i
). Also we have explicit ratings from U
i
to
User
2
, so we have defined WE_UU (U
i
, User
2
) (see
Figure 1).
Figure 1: WI_UU computation based on explicit
evaluations.
To evaluate the WIU_UU value we must
consider which is the value of the weight
corresponding to the explicit ratings.
We denote this weight with PE (User
1
, U
j
). It
represents an explicit rating weight, in our case the
weight of the rating provided by User
1
.
The value of this weight is computed as a ratio of
the explicitly user established value and the sum of
all explicit ratings provided by him. We will have:
∑
=
=
k
1i
1
1
1
),(_
),(
),UWE_UU(User
UUserUUWE
UUserPE
i
j
j
(1)
where
ki ≤≤1
,
ki ≤≤1
.
In this moment we prepare the context to
compute implicit rating whom user U
x
provided to
U
y
:
∑
=
=
k
j
yjjxyx
UUUUWEUUPEUUUUWIU
1
),(_*),(),(_
(2)
where
kj ≤≤1
.
Defining WIR_UU value on the basis of resource
evaluation
The need of partial function WIR_UU when the set
of users used for defining of WIU_UU is the empty
set. This means that we do not have a set of users
U
1
, .., U
k
whom we have explicit ratings from User
1
to U
i
,
ki
≤
≤
1
and also we do not have explicit
ratings from U
i
to User
2
. In this case we can obtain
information on the basis of the worth of a set of
resources evaluated by users. These resources are
required to be evaluated by both users. Thus, on the
basis of the evaluations of the same resource, one
can obtain a mutual evaluation of two users.
Let us consider: U
x
, U
y
and the resources
R
1
,...,R
n
. If there exists WE_UR(U
x
,R
i
) and
WE_UR(U
y
,R
i
) , for all
ni ≤
≤
1
, then the value of
WIR_UU(U
x
,U
y
) will exist and it will be equal with
WIR_UU(U
y
,U
x
). We define WIR_UU (U
x
,U
y
) as
follows:
n
RUURWERUURWE
MaxWorthUUUUWIR
n
i
iyix
yx
∑
=
−
−
=
1
|),(_),(_|
),(_
(3)
where
ni
≤
≤
1
. The demonstration of the assertion:
),(_),(_
xyyx
UUUUWIRUUUUWIR
=
(4)
is obvious. Therefore, in the case when we want to
obtain WI_UU on the basis of resource evaluation
WI_UU has the value of WIR_UU (U
1
, U
2
).
Finally the worth of WI_UU (U
1
, U
2
) will be
WIU_UU (U
1
, U
2
), if defined, or WIR_UU, if
defined, or it will be a default value fixed in the
system configuration.
In a future work, we will present a mechanism to
obtain a complete function WIU_UU without this
implicit value. In addition, if we have both user-user
and user-resource evaluations, then we can foresee a
given priority between them.
Implicit WI_UR Value Computation
We will define the manner to obtain the worth of the
implicit evaluation - WI_UR(U
x
,R
z
) – whom a user
U
x
provides to a resource R
z
, 1<z<n, where n is the
resource number of the system. We consider that the
user U
x
has in his known person list the following
users: U
1
,...,U
k
. These users have evaluated the
considered resource. This means we have defined
the following relations: WE_UU(U
x
,U
i
),
ki
≤
≤
1
and also WE_UR(U
i
,R
z
),
ki ≤
≤
1
The implicit rating provide by U
x
to resource R
z
is represented by the proportion between: sum of the
product of the rating weights of the user U
x
for each
user from his list and the value provided by him to
PRES – PERSONALIZED EVALUATION SYSTEM IN A WEB COMMUNITY - A Conceptual Model Designed to
Evaluate Reputation in Order to Achive a Personalised View on the System for Each User
67

resource R
z
, and the number of users (which is k in
our situation)
k
RUURWEUUPE
RUURWI
k
i
ziix
zx
∑
=
=
1
),(_*),(
),(_
(5)
where
ki ≤≤1
. We introduce the worth average
provided to a resource and we denote it with WA.
The value of WA (Resource) represents relevant
statistical average provided to a resource by all
users. WA for a resource inside a Web community
plays the same role that page rank plays in Web
page evaluation. This metric is necessary in case we
do not have enough trustworthy evaluators in the
community.
4 ASPECTS REGARDING THE
PRES BENEFITS
In this section we discuss shortly a set of
consequences, due to the way the system has been
modeled. We will argue our assertions through few
examples and in a next paper we will give the
appropriate algorithms used for these cases.
 The system assures the property to see the
things prioritized the same way as similar
users.
 The spammers will see more spam because
the system groups the users by their
preferences.
Let us consider a web community with users
U
1
,…,U
k
. We can consider that a new user U
x
joins
the community and posts a new resource - R
x
.
The resource posted by U
x
will be evaluated by
the users from community with worth values (
implicitly WE_UR(U
x
,R
x
)=5).
If R
x
is a spam resource, it will be explicitly
evaluated by users U
i
which are not interested in
spam resources with WE_UR(U
i
,R
x
)=1 or it will be
explicitly evaluated by users Uj which are interested
by this kind of resources with WE_UR(U
j
,R
x
) = 5,
where
ki ≤≤1 ,
kj ≤≤1
, i ≠ j .
Also, let us consider the case when a user U
y
evaluates the users U
i
. Because users U
i
have
evaluated resource Rx with low worth than the sum
∑
=
k
i
ziix
RUURWEUUPE
1
),(_*),(
has a low value.
Than the value of WI_UR(U
y
,R
x
) will be low and in
this case the spam resource R
x
will not be considered
interesting for the user U
y
.
In other case when user U
y
will evaluate users U
j
with worth metric with a higher value than the sum
∑
=
k
i
ziix
RUURWEUUPE
1
),(_*),(
has a higher value
and in this case the resource R
x
will be automatically
consider useful for user U
y
. We argue with this
example one case from a set of possible use-cases.
We will discuss in detail these cases and the used
algorithms in our next paper.
 The resources which are relevant for the
user are on top of the list of visible
resources. In this moment we know that
Google uses Page Rank system. The new
resources, even valuable, will reach hardly
on the top, because it takes long until they
receive links. And worse it is the fact that if
they are not on the top, they do not receive
links.
Therefore there exists a very high probability
that a good resource is not used.
In our system the new valuable resources appear
quickly on top when they are evaluated first by an
honest community member (one who tries always to
evaluate correctly). If somebody over evaluate his
own resource and the others rate it with low marks,
the mark WI_UU will drop, therefore those who add
resources are required to give right marks.
 The users will be required to do a fair
evaluation.
It will not happen like in the eBay system. In this
system, one assures a feedback for each user. The
feedback value is obtaining from other users
evaluations. One observed that the users are afraid of
obtaining a negative feedback. For this reason they
post positive feedbacks in a high proportion, hoping
that they will obtain positive feedbacks.
 The system can be easily integrated in
different Web communities.
Let us consider a real community like LinkedIn.
There exist in this moment some posted
announcements which offer jobs for this community
only. Our system would give the possibility that this
announcements to be visualized only by the users
with a given profile, the announcement being not
useful for other users types.
Thus our system makes it more efficient the
information management that is visible to the user.
ICE-B 2008 - International Conference on e-Business
68

5 CONCLUSIONS
Reputation system gives people information about
past activities of the other users. It can enhance an
on-line interaction environment by: helping people
decide who to interact, encouraging people to be
more honest, discouraging those who are not
responsible from participating. The actions, the
behavior, the user preferences can be regarded as
resources on which one can initiate interpretation
and processing mechanisms. PRES system allows
each user to have its own evaluation of the resources
and of the other users. The proposed metrics can be
used for implementation in real Web communities.
In this work we have presented the basic
elements of PRES model. For the future
development of the prototype we will perform a
detailed analyzes of the system properties.
In a real system the resources are changing in time.
This problem will be studied in our system thus
foreseeing the possibility that the users can see and
change the given ratings.
Another problem related to the reputation
computation that will be studied is a complexity of
the algorithm of performing the entire calculus in the
system. The computation of WR and WU can be
easily performed for a proper number of resources
and users. For hundreds thousands of users and
resources we need a parallel algorithm to compute
periodically the WR and WU values.
REFERENCES
Amy N. Langville, Carl D. Meyer, 2006. Google's
PageRank and Beyond: The Science of Search Engine
Rankings , Princeton University Press
Hai Zhuge, The Web Resource Space Model, 2008,
Springer.
J. Golbeck, J. Hendler. 2006. Inferring Trust Relationships
in Web-Based Social Networks, ACM Transactions on
Internet Technology, 6(4).
J. Golbeck, J. Hendler. 2004. Accuracy of Metrics for
Inferring Trust and Reputation in Semantic Web-based
Social Networks, Proceedings of 14th International
Conference on Knowledge Engineering and
Knowledge Management, Northamptonshire, U.K.
Paolo Massa, Bobby Bhattacharjee, 2004, Using trust in
recommender systems: an extol analysis, in
Proceedings of iTrust 2004
O’Reilly, T., 2005. What is Web 2.0 – Design Patterns
and Business Models for the Next Generation of
Software. O’Reilly.
R. Falcone, S.Barber, L. Korba, M. Singh (Eds.), 2002,
Trust, reputation, and Security: Theories and Practice.
AAMAS 2002 International Workshop, Springer
Ross A. Malaga, 2004. Web-Based Reputation
Management Systems: Problems and Suggested
Solutions, Springer
L. Mui, PhD thesis: Computational models of trust and
reputation: agents, evolutionary, games and social
networks, MIT, 2002
Jennifer Golbeck, James Hendler: Reputation network
analysis for email filtering, for the 1st Conf. on Email
and Anti-Spam, 2004
T. Berners-Lee et al. (eds.), Uniform Resource Identifiers
(URI): General Syntax, RFC 2396, IETF, 1998
Toby Segaran, 2007. Programming Collective
Intelligence, O’Reilly.
Web-o-Trust effort by Russell Nelson, http://www.web-o-
trust.org/.
del.icio.us. http://del.icio.us/.
dig.com. http://digg.com/.
Slashdot. http://slashdot.org/.
eBay. http://www.ebay.com/.
LinkedIn. http://www.linkedin.com
Everything2.http://everything2.com/.
Wikipedia. http://en.wikipedia.org.
PRES – PERSONALIZED EVALUATION SYSTEM IN A WEB COMMUNITY - A Conceptual Model Designed to
Evaluate Reputation in Order to Achive a Personalised View on the System for Each User
69