=Paper=
{{Paper
|id=Vol-1382/paper6
|storemode=property
|title=The Positive Power of Prejudice: A Computational Model for MAS
|pdfUrl=https://ceur-ws.org/Vol-1382/paper6.pdf
|volume=Vol-1382
|dblpUrl=https://dblp.org/rec/conf/woa/SapienzaFC15
}}
==The Positive Power of Prejudice: A Computational Model for MAS==
https://ceur-ws.org/Vol-1382/paper6.pdf
Proc. of the 16th Workshop “From Object to Agents” (WOA15) June 17-19, Naples, Italy
The Positive Power of Prejudice:
A Computational Model for MAS
Alessandro Sapienza, Rino Falcone and Cristiano Castelfranchi
Institute of Cognitive Science and Technologies, ISTC-CNR, Rome, Italy
{alessandro.sapienza, rino.falcone, cristiano.castelfranchi}@istc.cnr.it
Abstract— In MAS studies on Trust building and dynamics hundreds of things about it). At a social level this means that I
the role of direct/personal experience and of recommendations can know a lot of things on people that I never met; it is social
and reputation is proportionally overrated; while the importance "prejudice" with its good side and fundamental contribution to
of inferential processes in deriving the evaluation of trustees’ social exchange. How can I trust (for drugs prescription) a
trustworthiness is underestimated and not enough exploited.
medical doctor that I never met before and nobody of my
In this paper we focus on the importance of generalized friends knows? Because he is a doctor!
knowledge: agents' categories. The cognitive advantage of Of course we are underlining the positive aspects of
generalized knowledge can be synthesized in this claim: "It generalized knowledge, its essential role for having
allows us to know a lot about something/somebody we do not information on people never met before and about whom no
directly know". At a social level this means that I can know a lot one gave testimony. The more rich and accurate this
of things on people that I never met; it is social "prejudice" with knowledge is, the more it is useful. It offers huge opportunity
its good side and fundamental contribution to social exchange. In both for realizing productive cooperation and for avoiding
this study we experimentally inquire the role played by risky interactions. The problem is when the uncertainty about
categories' reputation with respect to the reputation and opinion
on single agents: when it is better to rely on the first ones and
the features of the categories is too large or it is too wide the
when are more reliable the second ones. Our claim is that: the variability of the performers within them. In our culture we
larger the population and the ignorance about the attribute a negative sense to the concept of prejudice, and this
trustworthiness of each individual (as it happens in an open because we want to underline how generalized knowledge can
world) the more precious the role of trust in categories. In produce unjust judgments against individuals (or groups)
particular, we want investigate how the parameters defining the when superficially applied (or worst, on the basis of precise
specific environment (number of agents, their interactions, discriminatory intents). Here we want rather to point out the
transfer of reputation, and so on) determine the use of categories' positive aspects of the prejudice concept.
reputation. In this study we intend to explain and experimentally show the
This powerful inferential device has to be strongly present in advantage of trust evaluation based on classes' reputation with
WEB societies. respect to the reputation and opinion on single potential agents
(partners). In an open world or in a broad population how can
I. INTRODUCTION we have sufficient direct or reported experience on
everybody? The quantity of potential agents in that population
In MultiAgent Systems (MAS) and Online Social Networks
(OSN) studies on Trust building and dynamics the role of or net that might be excellent partners but that nobody knows
direct/personal experience and of recommendations and enough can be high.
reputation (although important) is proportionally overrated; Our claim is that: the larger the population and the ignorance
while the importance of inferential processes in deriving the about the trustworthiness of each individual the more precious
evaluation of trustee's trustworthiness is underestimated and the role of trust in categories. If I know (through signals,
not sufficiently exploited (a part from the so called marks, declaration, ...) the class of a given guy/agent I can
“transitivity”, which is also, very often, wrongly founded). have a reliable opinion of its trustworthiness derived from its
In particular, generalization and instantiation from classes, class-membership.
categories [8] and analogical reasoning (from task to task and It is clear that the advantages of such cognitive power
from agent to agent) really should receive much more provided by categories and prejudices does not only depend on
attention. In this paper we focus on the importance of recommendation and reputation about categories. We can
generalized knowledge: agents' categories. The cognitive personally build, by generalization, our evaluation of a
advantage of generalized knowledge (building classes, category from our direct experience with its members (this
prototypes, categories, etc.), can be synthesized in this happens in our experiments for the agents that later have to
obvious claim: "It allows us to know a lot about propagate their recommendation about). However, in this
something/somebody we do not directly know" (for example, I simulation we have in the trustor (which has to decide whom
never saw Mary's dog, but - since it is a dog - I know rely on) only a prejudice based on recommendations about that
39
Proc. of the 16th Workshop “From Object to Agents” (WOA15) June 17-19, Naples, Italy
category and not its personal experience. different cluster views and combining them to obtain better
After a certain degree of direct experiences and circulation of results.
recommendations, the performance of the evaluation based on Another example is [6] where authors use information
classes will be better; and in certain cases there will be no regarding social friendships in order to provide users with
alternative at all: we do not have any evaluation on that more accurate suggestions and rankings on items of their
individual, a part from its category; either we work on interest.
inferential instantiation of trustworthiness or we loose a lot of A classical decentralized approach is referral systems [20],
potential partners. This powerful inferential device has to be where agents adaptively give referrals to one another.
strongly present in WEB societies supported by MAS. We Information sources come into play in FIRE [12], a trust and
simplify here the problem of the generalization process, of reputation model that use them to produce a comprehensive
how to form judgement about groups, classes, etc. by putting assessment of an agent’s likely performance. Here authors
aside for example inference from other classes (higher or sub); take into account open MAS, where agents continuously enter
we build opinion (and then its transmission) about classes on and leave the system. Specifically, FIRE exploits interaction
the bases of experience with a number of subjects of a given trust, role-based trust, witness reputation, and certified
class. reputation to provide trust metrics.
First of all, we want to clarify that here we are not interested The described solutions are quite similar to our work, although
in steretypes, but in categories. We define steretypes as the set we contextualized this problem to information sources.
of features that, in a given culture/opinion, characterize and However we do not investigate recommendations with just the
distinguish that specific group of people. aim of suggesting a particular trustee, but also for inquiring
Knowing the stereotype of an agent could be expensive and categories’ recommendations.
time consuming. Here we are just interested in the fact that an
agent belongs to a category: it has not to be a costly process II. RECOMMENDATION AND REPUTATION: DEFINITIONS
and the recognition must be well discriminative and not- Let us consider a set of agents Ag1, ..., Agn in a given world
cheating. There should be visible and reliable "signals" of that (for example a social network). We consider that each agent in
membership. In fact, the usefulness of categories, groups, this world could have trust relationships with anyone else. On
roles, etc. makes fundamental the role of the signs for the basis of these interactions the agents can evaluate the trust
recognizing or inferring the category of a given agent. That's degree of their partners, so building their judgments about the
why in social life are so important coats, uniforms, titles, trustworthiness of the agents with whom they interacted in the
badges, diplomas, etc. and it is crucial their exhibition and the past.
assurance of their authenticity (and, on the other side, the The possibility to access to these judgements, through
ability to falsify and deceive). In this preliminary model and recommendations, is one of the main sources for trusting
simulation let us put aside this crucial issue of indirect agents outside the circle of closer friends. Exactly for this
competence and reliability signaling; let us assume that the reason recommendation and reputation are the more studied
membership to a given class or category is true and and diffused tools in the trust domain [15].
transparent: the category of a given agent is public, common We introduce
knowledge. Recx,y,z (t ) (1)
Differently from [2][10][17] in this work we do not address
where x, y, zÎ { Ag1 , Ag2,...., Agn } , we call D the specific set of
the problem of learning categorical knowledge and we assum
that the categorizzation process is objective. agents: D º { Ag1 , Ag2,...., Agn }
Similarly to [3], we give agents the possibility to recommend and 0 £ Recx,y,z (t ) £1
categories and this is the key point of this paper.
In the majority of the cases available in the literature, the , as established in the trust model of [4], is the task on which
concept of recommendation is used concerning recommender the recommender x expresses the evaluation about y.
systems [1]. These ones can be realized using both past In words: Recx,y,z (t ) is the value of x’s recommendation about
experience (content-based RS)[13] or collaborative filtering, y performing the task , where z is the agent receiving this
in which the contribute of single agents/users is used to recommendation. In this paper, for sake of simplicity, we do
provide group recommendations to other agents/users. not introduce any correlation/influence between the value of
Focusing on collaborative filtering, the concepts of similarity the recommendations and the kind of the agent receiving it:
and trust are often exploited (together or separately) to the value of the recommendation does not depend from the
determine which contributes are more important in the agent to whom it is communicated.
aggregation phase [14][18]For instance, in [7] authors provide So (1) represents the basic expression for recommendation.
a system able to recommend to users group that they could We can also define a more complex expression of
join in Online Social Network. Here it is introduced the recommendation, a sort of average recommendation:
concepts of compactness of a social group, defined as the Agn
weighted mean of the two dimensions of similarity and trust. å
Rec (t ) / n
x,y,z
(2)
Even in [11] authors present a clustering-based recommender x=Ag1
system that exploits both similarity and trust, generating two in which all the agents in the defined set of agents express
their individual recommendation on the agent y with respect
40
Proc. of the 16th Workshop “From Object to Agents” (WOA15) June 17-19, Naples, Italy
the task and the total value is divided by the number of In words: Recx,Cy,z (t ) is the value of x’s recommendation
agents. about the agents included in category Cy when they perform
We consider the expression (2) as the reputation of the agent y the task , (as usual z is the agent receiving this
with respect to the task in the set D. recommendation).
Of course the reputation concept is more complex than the We again define a more complex expression of
simplified version here introduced [5][16]. recommendation, a sort of average recommendation:
It is in fact the value that would emerge in the case in which Agn
we receive from each agent in the world its recommendation å Rec x,Cy,z (t ) / n (6)
about y (considering each agent as equally reliable). x=Ag1
In the case in which an agent has to be recommended not only in which all the agents in the domain express their individual
on one task but on a set of tasks ( 1 , ..., k), we could define recommendation on the category Cy with respect the task and
instead of (1) and (2) the following expressions: the total value is divided by the number of the recommenders.
k
We consider the expression (6) as the reputation of the
å Re c x,y,z (t i ) / k (3)
category Cy with respect the task in the set D.
i=1
that represents the x’s recommendation about y performing the Now we extend to the categories, in particular to Cy, the
set of tasks (1,...,k), where z is the agent receiving this recommendations on a set of tasks (1, ...,k):
k
recommendation. å
Rec (t ) / k
x,Cy,z i
(7)
Imagine having to assign a meta-task (composed of a set of i=1
tasks) to just one of several agents. In this case the information that represents the recommendation value of the x's agent
given from the formula (3) could be useful for selecting (given about the agents belonging to the category Cy when they
the x's point of view) on average (with respect to the tasks) the perform the set of tasks (1,...,k).
more performative agent y. Finally, we define:
Agn k Agn k
å å
Rec (t ) / nk
(4)
x,y,z i å å Rec x,Cy,z (t i ) / nk (8)
x=Ag1 i=1
x=Ag1 i=1
that represents a sort of average recommendation from the set that represents the value of the reputation of the category Cy
of agents in D, about y performing the set of tasks ( 1 , ..., k). (of all the agents y included in Cy) with respect the set of tasks
We consider the expression (4) as the reputation of the agent y (1,...,k), in the set D.
with respect the set of tasks (1 , ...,k), in the set D.
Having to assign the meta-task proposed above, the B. Definition of Interest for this Work
information given from the formula (4) could be useful for In this paper we are in particular interested in the case in
selecting on average (with respect to both the tasks and the which z (a new agent introduced in the world) asks for
agents) the more performative agent y. recommendation to x ( x Î D ) about an agent belonging to its
A. Using Categories domain Dx for performing the task (Dx is a subset of D, it is
composed by the agents that x knows). x will select the best
As described above, an interesting approach for evaluating evaluated y, with y Î Dx on the basis of formula:
agents is to classify them in specific categories already pre-
judged/rated and as a consequence to do inherit to the agents max yÎD (Recx,y,z (t ))
x
(9)
the properties of their own categories. where Dx º { Ag1 , Ag2,...., Agm} , Dx includes all the agents
So we can introduce also the recommendations about
categories, not just about agents (we discuss elsewhere how evaluated by x. They are a subset of D: Dx Í D .
these recommendations are formed). In this sense we define: In general D and Dx are different because x does not
Recx,Cy,z (t ) (5) necessarily know (has interacted with) all the agents in D.
z asks for recommendations not only to one agent, but to a set
where x Î { Ag1 , Ag2 ,...., Agn } as usual, and we characterize the
of different agents: x Î Dz (Dz is a subset of D, to which z asks
categories {C1 ,....,Cl }through a set of features { fy1 ,..., fym} : for reputation), and selects the best one on the basis of the
"y Î { Ag1 ,..., Agn } $cy Î {C1 ,...,Cl } | (Cy º { fy1 ,..., fym})Ù({ fy1 ,..., fym} Î y) value given from the formula:
it is clear that there is a relationship between task , and the max xÎD (max yÎD (Recx,y,z (t )))
z x
(10)
features { fy1 ,..., fym} of the Cy category. In words we can say Dz Í D , z could ask to all the agents in the world or to a
that each agent in D is classified in one of the categories defined subset of it (see later).
{C1 ,....,Cl } that are characterized from a set of features We are also interested to the case in which z ask for
recommendations to x about a specific agents’ category for
{ f1 ,..., fm} ; as a consequence each agent belonging to a performing the task . x has to select the best evaluated Cy
category owns the features of that category. among the different Cy Î {C1 ,....,Cl } x has interacted with (we
0 £ Recx,Cy,z (t ) £1
are supposing that each agent in the world D, belongs to a
category in the set {C1 ,....,Cl }).
41
Proc. of the 16th Workshop “From Object to Agents” (WOA15) June 17-19, Naples, Italy
In this case we have the following formulas: particular, the trustor will ask them for the best category and
maxCyÎDx (Recx,Cy,z (t )) (11) the best trustee they have experienced.
In this way, the trustor is able to collect information about
that returns the category best evaluated from the point of view
both the best recommended category and agent.
of an agent (x). And
It is worth noting that the trustor collects information from the
max xÎD (maxCyÎD (Recx,Cy,z(t )))
z x
(12)
agents considering them equally trustworthy with respect to
that returns the category best evaluated from the point of view the task of "providing recommendations". Otherwise it should
of all the agents included in Dz . weigh differently these recommendations. In practice the
agents are sincere.
III. COMPUTATIONAL MODEL Then it will select a randomly chosen agent belonging to the
best recommended category and it will compare it, in terms of
A. General Setup
objective trustworthiness, with the best recommended
In order to realize our simulations, we exploited the software individual agent (trustee).
NetLogo [19]. The possible outcomes are:
In every scenario there are four general categories, called trustee wins (t_win): the trustee selected with
Cat1, Cat2, Cat3 and Cat4, composed by 100 agents per individual recommendation is better than the one
category. selected by the means of category; then this method
Each category is characterized by: gets one point;
1. an average value of trustworthiness, in range
[0,100]; category wins (c_win): the trustee selected by the
means of category is better than the one selected with
2. an uncertainty value, in range [0,100]; this value individual recommendation; then this method gets
represents the interval of trustworthiness in which the one point;
agents can be considered as belonging to that category.
equivalent result: if the difference between the two
These two values are exploited to generate the objective trustworthiness values is not enough (it is under a
trustworthiness of each agent, defined as the probability that, threshold), we consider it as indistinguishable result.
concerning a specific kind of required information, the agent In particular, we considered the threshold of 3% as,
will communicate the right information. on the basis of previous test simulations, it has
resulted a resonable value.
Of course the trustworthiness of categories and agents is
strongly related to the kind of requested information/task. These two phases are repeated 500 times for each setting.
Nevertheless, for the purpose of our it is enough to use just
IV. SIMULATIONS RESULTS
one kind of information (defined by ) in the simulations. The
categories’ trustworthiness of Cat1, Cat2, Cat3 and Cat4 are In these simulations we present a series of scenarios with
fixed respectively to 80, 60, 40 and 20% for . What changes different settings to show when it is more convenient to
through scenarios is the uncertainty value of the categories: 1, exploit recommendations about categories rather than
20, 50, and 80%. recommendations about individuals, and vice versa.
We also present the “all-in-one” scenario, whose peculiarity is
B. How the simulations work that the exploration lasts just 1 tick and in that tick every agent
Simulations are mainly composed by two main steps that are experiences all the others. Although this is a limit case, very
repeated continuously. In the first step, called exploration unlikely in the real world, it is really interesting as each agent
phase, agents without any knowledge about the world start has not a good knowledge of the other agent as individual
experiencing other agents, asking to a random 3% of the elements (it experienced them just one time), but it is able to
population for the information P. Then they memorize the get a really good knowledge of their categories, as it has
performance of each queried agent both as individual element experienced them as many times as the number of agents for
and as a member of its own category. each category. This is an explicit case in which agents’
The performance of a agent can assume just the two values 1 recommendations about categories are surely more
or 0, with 1 meaning that the agent is supporting the informative than the ones about individuals.
information P and 0 meaning that it is opposing to P. For sake In particular, we will represent this value:
of simplicity, we assume that P is always true. c _ win
The exploration phase has a variable duration, going from 100 (13)
ticks to 1 tick. Depending on this value, agents will have a
c _ win t _ win
better or worse knowledge of the other agents. In words, this ratio shows how much categories’
Then, in a second step (querying phase) we introduce in the recommendation is useful if compared to individual
world a trustor (a new agent with no knowlegde about the recommendation.
trustworthiness of other agents and categories, and that has the Simulations’ results are presented in a graphical way,
necessity to trust someone reliable for a given informative exploiting 3D shapes to represent all the outcomes. These
task: in our case ). It will select a given subset of the shapes are divided into two area and represented with two
population, going from 100% to 5%, and it will query them. In different colors:
42
Proc. of the 16th Workshop “From Object to Agents” (WOA15) June 17-19, Naples, Italy
the part over 0.5, in which prevails the category
Queried trustee %
recommendation; 1
the one below 0.5, in which prevails the individual
recommendation. 0,5
5 0,5-1
These graphs represent an useful view about the utility of the 0
categorial role in the different interactional and social 0-0,5
3 all-in
contexts. 50 10
For each value of uncertainty, we explored 40 different
settings, considering all the possible couple of exploration Exploration phase
phase and queried trustee percentage, where:
exploration phase {all-in,1,3,5,10,25,50,100};
Figure 4. Outcomes for 80% of categories' uncertainty
queried trustee percentage {5,10,25,50,100}.
The part in which category recommendation wins over
individual recommendation is represented in light grey.
Queried trustee %
1 Conversely, the part in which individual recommendation
wins is represented in dark grey.
0,5 Through these graphs we identify three effects that influence
0,5-1 the outcome. The first effect is due to categories' uncertainty:
5
0 the less it is, the more is the utility of using categories; the
0-0,5 more it is, the less categories will be useful. It is not possible
3 all-in
50 10 to notice this effect just looking at one picture. On the
contrary, looking at the overal picture one can notice that the
curves of the graphs lower, going from a maximal value in
Exploration phase
Figure 1 to a minimal value in Figure 4.
The second effect is due to exploration phase. The longer it is
Figure 1. Outcomes for 1% of categories' uncertainty the more individual recommendations are useful; the less it
lasts the more category recommendations are useful.
The third effect is introduced by the queried trustee
Queried trustee %
1 percentage, that acts exactly as the exploration phase: the
higher the percentage of queried agents, the more individual's
0,5 recommendations are useful; the less it is, the more categories'
0,5-1 recommendations are useful.
5
0 The exploration phase’s length and the queried agents’
0-0,5
3 all-in percentage occur in all the four graphs and cooperate in
50 10
determining respectevely the degree of knowledge (or
Exploration phase ignorance) in the world and the level of inquire about this
knowledge. In particular, with "the knowledge in the world"
we intend how the agents can witness the trustworthiness of
Figure 2. Outcomes for: 20% of categories' uncertainty the other agents or their aggregate, given the constraints
defined from the external circumstances (number and kind of
interactions, kind of categories, and so on).
Queried trustee %
1 In practice, both these elements seem to suggest how the role
of categories becomes relevant when either decreases and
0,5 degrades the knowledge within the analyzed system (before
0,5-1 the interaction with the trustor) or is reduced the transferred
5
0 knowledge (to the trustor).
0-0,5
3 all-in Let us explain better. The first effect shows how the reliability
50 10 of category's trustworthiness (that will be inherited by its
Exploration phase members) depends, of course, from the variability of the
behavior among the class members. There may be classes
where all the members are very correct and competent, other
Figure 3. Outcomes for: 50% of categories' uncertainty classes where there is a very high variance: in this last case
our betting on a member of that class is quite risky.
The second effect can be described with the fact that each
agent, reducing the number of interactions with the other
agents in the explorative phase, will have relevantly less
information with respect to the individual agents. At the same
43
Proc. of the 16th Workshop “From Object to Agents” (WOA15) June 17-19, Naples, Italy
time its knowledge with respect to categories does not undergo works we have to consider how, starting from the analysis of
a significant decline given that categories' performances derive this study, could change the role of knowledge about
from several different agents. categories in a situation of open world. We have also to
The third effect can be explained with the fact that reducing consider the cases in which the recommendations are not so
the number of queried trustees, the trustor will receive with transparent but influenced by specific goals of the agents.
decreasing probability information about the more trustworthy
individual agents in the domain, while information on
categories, maintains a good level of stability also reducing ACKNOWLEDGMENTS
the number of queried agents, thanks to greater robustness of This work is partially supported both by the Project PRISMA
these structures. (PiattafoRme cloud Interoperabili per SMArt-government;
Resuming, the above pictures clearly show how, when the Cod. PON04a2 A) funded by the Italian Program for Research
quantity of information (about the agents' trustworthiness and Innovation (Programma Operativo Nazionale Ricerca e
exchanged in the system) decreases, it is better to rely on the Competitività 2007-2013) and by the project CLARA—
categorial recommendations rather than individual CLoud plAtform and smart underground imaging for natural
recommendations. Risk Assessment, funded by the Italian Ministry of Education,
This result reaches the point of highest criticality in the “all- University and Research (MIUR-PON).
in-one” case in which, as expected, the relevance of categories
reach its maximal value.
REFERENCES
V. CONCLUSION [1] Adomavicius, G., Tuzhilin, A. Toward the next generation of
Other works [9][2] show the advantages of using recommender systems: A survey of the state-of-the-art and possible
extensions. IEEE Transactions on Knowledge and Data Engineering
categorization to select trustworthy agents. In particular, how (TKDE) 17, 734–749, 2005
it were possible to attribute to a certain unknown agent, a [2] Burnett, C., Norman, T., and Sycara, K. 2010. Bootstrapping trust
value of trustworthiness with respect to a specific task, on the evaluations through stereotypes. In Proceedings of the 9th International
basis of its classification in, and membership to, one (/or Conference on Autonomous Agents and Multiagent Systems
(AAMAS'10). 241248.
more) category/ies. In practice, the role of generalized
[3] C. Burnett, T. J. Norman, and K. Sycara. Stereotypical trust and bias in
knowledge and prejudice (in the sense of pre-established dynamic multiagent systems. ACM Transactions on Intelligent Systems
judgment on the agents belonging to that category) has proven and Technology (TIST), 4(2):26, 2013.
to determine the possibility to anticipate the value of unknown [4] Castelfranchi C., Falcone R., Trust Theory: A Socio-Cognitive and
agents. Computational Model, John Wiley and Sons, April 2010.
In this paper we have investigated the different roles that [5] Conte R., and Paolucci M., 2002, Reputation in artificial societies.
recommendations can play about individual agents and about Social beliefs for social order. Boston: Kluwer Academic Publishers.
categories of agents. [6] P. De Meo, E. Ferrara, G. Fiumara, and A. Provetti. Improving
Recommendation Quality by Merging Collaborative Filtering and Social
In this case the new agent introduced (called trustror) has a Relationships. In Proc. of the International Conference on Intelligent
whole world of agents completely unknown to it, and ask for Systems Design and Applications (ISDA 2011) , Córdoba, Spain, IEEE
recommendations to a (variable) subset of agents for selecting Computer Society Press, 2011
an agent to whom delegate a task. The information received [7] P De Meo, E Ferrara, D Rosaci, and G Sarné. Trust and Compactness of
Social Network Groups. IEEE Transactions on Cybernetics, PP:99, 2014
regards both individual agents and agents' categories. The
[8] Falcone R, Castelfranchi C, Generalizing Trust: Inferencing
informative power of these two kinds of recommendations Trustworthiness from Categories. In: TRUST 2008 - Trust in Agent
depends on the previous interactions among the agents and Societies, 11th International Workshop, TRUST 2008. Revised Selected
also on the number agents queried by the trustor. However, and Invited Papers (Estoril, Portugal, 12-13 May 2008). Proceedings, pp.
there are cases in which information about categories is more 65 - 80. R. Falcone, S. K. Barber, J. Sabater-Mir, M. P. Singh (eds.).
(Lecture Notes in Artificial Intelligence, vol. 5396). Springer, 2008.
useful that information towards individual agents. In some
[9] Falcone R., Piunti, M., Venanzi, M., Castelfranchi C., (2013), From
sense this result complements the results achieved in [9][2] Manifesta to Krypta: The Relevance of Categories for Trusting Others,
because here we have a more strict match between information in R. Falcone and M. Singh (Eds.) Trust in Multiagent Systems, ACM
on individual agents and information about categories of Transaction on Intelligent Systems and Technology, Volume 4 Issue 2,
agents: We are measuring the quantity of information, about March 2013
individual agents and categories, for evaluating when is better [10] H. Fang, J. Zhang, M. Sensoy, and N. M. Thalmann. A generalized
stereotypical trust model. In Proceedings of the 11th International
using direct information rather than generalized information Conference on Trust, Security and Privacy in Computing and
or, vice versa, when is better using the positive power of Communications (TrustCom), pages 698–705, 2012.
prejudice. Our results show how in certain cases becomes [11] G. Guo, J. Zhang and N. Yorke-Smith, Leveraging Multiviews of Trust
essential the use of categorial knowledge for selecting and Similarity to Enhance Clustering-based Recommender Systems,
Knowledge-Based Systems, accepted, 2014
qualified partners.
[12] Huynh, T.D., Jennings, N. R. and Shadbolt, N.R. An integrated trust and
In this work we have in fact considered a closed world, with a reputation model for open multi-agent systems. Journal of Autonomous
fixed set of agents. This choice was based on the fact that we Agents and Multi-Agent Systems, 13, (2), 119-154., 2006
were interested to evaluate the relationships between [13] P. Lops, M. Gemmis, and G. Semeraro, “Content-based
knowledge about individual and knowledge about categories, recommender systems: State of the art and trends,” in Recommender
for calibrating their roles and reciprocal influences. In future Systems Handbook. Springer, pp. 73–105, 2011.
44
Proc. of the 16th Workshop “From Object to Agents” (WOA15) June 17-19, Naples, Italy
[14] P. Massa, P. Avesani, Trust-aware recommender systems, RecSys '07:
Proceedings of the 2007 ACM conference on Recommender systems,
2007
[15] S. Ramchurn, N. Jennings, Carles Sierra, and Lluis Godo. Devising a
trust model for multi-agent interactions using confidence and reputation.
Applied Artificial Intelligence, 18(9-10):833-852, 2004.
[16] Sabater-Mir, J. 2003. Trust and reputation for agent societies. Ph.D.
thesis, Universitat Autonoma de Barcelona.
[17] M. Sensoy, B. Yilmaz, and T. J. Norman. STAGE: Stereotypical trust
assessment through graph extraction. Computational Intelligence, 2014.
[18] C. Than and S. Han, Improving Recommender Systems by
Incorporating Similarity, Trust and Reputation, Journal of Internet
Services and Information Security (JISIS), volume: 4, number: 1, pp. 64-
76, 2014
[19] Wilensky, U. (1999). NetLogo. http://ccl.northwestern.edu/netlogo/.
Center for Connected Learning and Computer-Based Modeling,
Northwestern University, Evanston, IL.
[20] Yolum, P. and Singh, M. P. 2003. Emergent properties of referral
systems. In Proceedings of the 2nd International Joint Conference on
Autonomous Agents and MultiAgent Systems (AAMAS'03).
45