Wang Y. Vassileva J. (2003) Trust and Reputation Model in Peer-to-Peer Networks, Proc. of IEEE Conference on P2P Computing,
Linkoeping, Sweden, September 2003.
Trust and Reputation Model in Peer-to-Peer Networks
Yao Wang, Julita Vassileva
University of Saskatchewan, Computer Science Department,
Saskatoon, SK, S7N 5A9, Canada
{yaw181, jiv}@cs.usask.ca
Abstract
It is important to enable peers to represent and update
their trust in other peers in open networks for sharing files,
and especially services. In this paper, we propose a
Bayesian network-based trust model and a method for
building reputation based on recommendations in peer-to-
peer networks. Since trust is multi-faceted, peers need to
develop differentiated trust in different aspects of other
peers’ capability. The peer’s needs are different in
different situations. Depending on the situation, a peer may
need to consider its trust in a specific aspect of another
peer’s capability or in multiple aspects. Bayesian networks
provide a flexible method to present differentiated trust and
combine different aspects of trust. The evaluation of the
model using a simulation shows that the system where peers
communicate their experiences (recommendations)
outperforms the system where peers do not share
recommendations with each other and that a differentiated
trust adds to the performance in terms of percentage of
successful interactions.
1. Introduction
Peer-to-peer networks are networks in which peers
cooperate to perform a critical function in a decentralized
manner [6]. All peers are both consumers and providers of
resources and can access each other directly without
intermediary peers. Compared with a centralized system, a
peer-to-peer (P2P) system provides an easy way to
aggregate large amounts of resources residing on the edge
of Internet or in ad-hoc networks with a low cost of system
maintenance. P2P systems have attracted increasing
attention from researchers recently, but they also bring up
some problems. Since peers are heterogeneous, some peers
might be benevolent in providing services. Some might be
buggy or malicious and cannot provide services with the
quality that they advertise. Since there is no centralized
node to serve as an authority to monitor and punish the
peers that behave badly, malicious peers have an incentive
to provide poor quality services for their benefit because
they can get away. Some traditional security techniques,
such as service providers requiring access authorization, or
consumers requiring server authentication, are used as
protection from known malicious peers. However, they
cannot prevent from peers providing variable-quality
service, or peers that are unknown. Mechanisms for trust
and reputation can be used to help peers distinguish good
from bad partners. This paper describes a trust and
reputation mechanism that allows peers to discover
partners who meet their individual requirements through
individual experience and sharing experiences with other
peers with similar preferences.
The rest of this paper is organized as follows: section 2
discusses the definitions of trust and reputation and their
characteristics. Section 3 introduces our approach to
developing a Bayesian network-based trust model and a
method for building reputation based on recommendations.
The experiment design and results are presented in Sections
4 and 5. Section 6 discusses related work on trust and
reputation. In the last section, we present conclusions and
directions for future work.
2. Trust and reputation
Trust and reputation mechanisms have been proposed
for large open environments in e-commerce, peer-to-peer
computing, recommender systems [4, 13, 14, 17, 18, 19].
However, there is no universal agreement on the definition
of trust and reputation. In this paper, we adopt the
following working definitions:
Trust – a peer’s belief in another peer’s capabilities,
honesty and reliability based on its own direct experiences;
Reputation – a peer’s belief in another peer’s
capabilities, honesty and reliability based on
recommendations received from other peers. Reputation
can be centralized, computed by a trusted third party, like a
Better Business Bureau; or it can be decentralized,
computed independently by each peer after asking other
peers for recommendations.
Although trust and reputation are different in how they
are developed, they are closely related. They are both used
Wang Y. Vassileva J. (2003) Trust and Reputation Model in Peer-to-Peer Networks, Proc. of IEEE Conference on P2P Computing,
Linkoeping, Sweden, September 2003.
to evaluate a peer’s trustworthiness, so they also share
some common characteristics [1, 8, and 12].
Ø Context specific.
Trust and reputation both depend on some context. For
example, Mike trusts John as his doctor, but he does not
trust John as a mechanic who can fix his car. So in the
context of seeing a doctor, John is trustworthy. But in the
context of fixing a car, John is untrustworthy.
Ø Multi-faceted.
Even in the same context, there is a need to develop
differentiated trust in different aspects of the capability of a
given peer. The same applies for reputation. For instance, a
customer might evaluate a restaurant from several aspects,
for example, the quality of food, the price, and the service.
For each aspect, she develops a kind of trust. The overall
trust depends on the combination of the trusts in each
aspect. While the context-specificity of trust accentuates
that trust in an identical peer can be different in different
situations, the characteristic, multi-faceted, emphasizes that
trust has multiple aspects, which can play a role in deciding
whether a peer is trustworthy to interact with.
Ø Dynamic.
Trust and reputation increase or decrease with further
experience (direct interaction). They also decay with time.
3. Bayesian network-based trust model
3.1 Trust and reputation mechanism
In our model a peer builds two kinds of trust in another
peer, say peer A and peer B respectively. The first one is
the trust that peer A has in peer B’s capability in providing
services. The other is the trust that peer A has in peer B’s
reliability in providing recommendations about other
peers. Here the reliability includes two aspects:
Ø Truthfulness – whether peer B is truthful in telling its
information
Ø Similarity – whether peer B is similar to peer A in
preferences and ways of judging issues.
Reliability = Truthfulness ? Similarity, i.e. a peer B’s
reliability as a referee depends on both being truthful and
similar in its preferences to the peer requesting the
recommendation. Since peers are heterogeneous, they may
have different preferences and judge issues by different
criteria. For example, some peers may consider a movie
provider good because it provides movies with high
quality, while others may consider the movie provider bad
because the speed of download from it is very slow. If two
peers A and B are similar in their evaluation criteria, peer
A can trust peer B’s recommendations, if it knows that peer
B is truthful. However, if the peers have different
evaluation criteria, peer A cannot trust peer B’s
recommendations even when peer B tells the truth.
It is important for a peer to develop trust in other peers
as references in a decentralized system, since when a peer
is not sure about the trustworthiness of a service provider,
it can ask for recommendations only those few peers that it
trusts most instead of asking a large number of peers, which
not only helps the peer get more reliable recommendations,
but also saves time and communication costs.
We will use a peer-to-peer file sharing application as an
example in the discussion, however the method is general
and can be applied to other applications, like web-
services, e-commerce, recommender systems or peer-to-
peer distributed computing.
In the area of file sharing in peer-to-peer networks, all
the peers are both providers and users of shared files. Each
peer plays two roles, the role of file provider offering files
to other peers and the role of user searching and
downloading files provided by other peers. In order to
distinguish the two roles of each peer, in the rest of paper,
when a peer acts as a file provider, we call it file provider;
otherwise, we call it simply peer. Peers will develop two
kinds of trust, the trust in the file providers’ capability (in
providing files) and the trust in the other peers’ reliability
in making recommendations. We assume all the peers are
truthful in telling their evaluations. However, the peers may
have different ways of evaluating other peers’ performance,
which reflect different user preferences.
A search request in file sharing peer-to-peer
applications usually results in a long list of providers for an
identical file. If a peer happens to select a provider of files
with bad quality or slow download speed, the peer will
waste time and effort, which may lead to user frustration
and abandoning the system. In order to solve the problem,
we use the mechanism of trust and reputation as shown in
Figure 1. Once a peer receives a list of file providers for a
given search, it can arrange the list according to its trust in
these file providers. Then the peer chooses one of the file
providers on top of the list. If the file provider is
trustworthy according to the peer’s previous experiences,
the peer will interact with the file provider (download
files). If the file provider is not trustworthy, the peer will
select another file provider to interact with. If the peer is
not sure about the trustworthiness of the file provider, for
example, the peer has no interactions or only a few
interactions with the file provider, it can ask other peers to
make recommendations for it. How the peer uses the
reputation and its own trust to make a decision with which
file provider to interact is an open question. Some peers
may prefer to trust their own experience and rely on their
trust even if they had very few interactions with the service
provider. Others may be more cautious and rely on the
reputation of the service provider. After each interaction,
the peer updates its trust in the file provider according to its
evaluation of the interaction. If the interaction is satisfying,
it will increase its trust in the file provider; if the
interaction is not satisfying, it will decrease its trust in the
file provider. If the decision of interaction is based on other
peers’ recommendations, the peer will also update its trust
Wang Y. Vassileva J. (2003) Trust and Reputation Model in Peer-to-Peer Networks, Proc. of IEEE Conference on P2P Computing,
Linkoeping, Sweden, September 2003.
in each of the peers that give recommendations (we call
these peers “referees”). If the referee’s recommendation is
consistent with the peer’s evaluation of the interaction, the
peer will increase its trust in the referee; otherwise, it will
decrease its trust.
Select a file
provider
Interaction
Evaluation of the
Interaction
Trustworthy
Update trust in the
file provider
No
Yes
Reputaion
Not Sure
Update trust in
references
Recommendations
from references
Evaluation of
references
Trust in
references
DB
Trust in file
providers
DB
Reputation MechanismTrust Mechanism
Figure 1. Functionality of the trust and
reputation mechanism on board of the peer
3.2 Trust in a file provider’s capability
In a peer-to-peer network, file providers’ capabilities
are not uniform. For example, some file providers (FP) may
be connecting through a high-speed network, while others
connect through a slow modem. Some file providers might
like music, so they share a lot of music files. Some may be
interested in movies and share more movies. Some may be
very picky about file quality, so they only keep and share
files with high quality. Therefore, the file provider’s
capability can be presented in various aspects, such as the
download speed, file quality and file type.
FTFQDS
T
Trust in a FP
Download speed
File quality
File type
Figure 2. A Bayesian network model
The peer’s needs are also different in different
situations. Sometimes, it may want to know the file
provider’s overall capability. Sometimes it may only be
interested in the file provider’s capability in some
particular aspect. For instance, a peer wants to download a
music file from a file provider. At this time, knowing the
file provider’s capability in providing music files is more
valuable for the peer than knowing the file provider’s
capability in providing movies.
Peers also need to develop differentiated trust in the file
providers’ capabilities. For example, the peer who wants
to download a music file from the file provider cares about
whether the file provider is able to provide the music file
with good quality at a fast speed, which involves the file
provider’s capabilities in two aspects, quality and speed.
How does the peer combine its two separated trust
representations together, the trust in the file provider’s
capability in providing music files with good quality and
the trust in the file provider’s capability in providing a fast
download speed, in order to decide whether the file
provider is trustworthy or not?
A Bayesian network provides a flexible method to solve
the problem. It is a relationship network that uses statistic
methods to represent probability relationships between
different elements [10]. We use a naïve Bayesian network
to represent the trust of a peer in a file provider. Every peer
develops a naive Bayesian network for each file provider
that it has interacted with. Each Bayesian network (see
Figure 2) has a root node T that represents the peer’s trust
in the file provider’s capability in providing files. It is the
percentage of interactions that are satisfying. The leaf nodes
under the root node represent the file provider’s capability
in different aspects. The node, denoted by FT, represents
the set of file types. Suppose it includes five values,
“Music”, “Movie”, “Document”, “Image” and
“Software”. The node “DS” denotes the set of download
speeds. It has three values, “Fast”, “Medium” and
“Slow”, each of which covers a range of download speeds.
The node “FQ” denotes the set of file qualities. It also has
three values, “High”, “Medium” and “Low”.
Wang Y. Vassileva J. (2003) Trust and Reputation Model in Peer-to-Peer Networks, Proc. of IEEE Conference on P2P Computing,
Linkoeping, Sweden, September 2003.
Here we only take three aspects of trust into account.
More relevant aspects can be added in the Bayesian
network later to account for user preferences with respect
to service.
According to a Bayesian network, a peer can infer the
trustworthiness of a file provider in different conditions,
such as the trustworthiness of the file provider in providing
music files, the trustworthiness of the file provider in
providing files with high quality, the trustworthiness of the
file provider in providing music files with high quality. The
condition can be any combination of the aspects. The
method will save peers effort in building different trusts
separately, or developing new trust when conditions
change.
3.3 Evaluating interactions and updating trust in
file providers
After each interaction, peers make an evaluation of it.
Peers might have different criteria to judge an interaction.
Some peers might be very picky. Some might be generous.
So they might have different evaluations of an identical
interaction. The overall evaluation of an interaction is a
combination of evaluations of each aspect related to the
interaction, such as download speeds, file quality. How to
combine evaluations of each aspect depends on each peer’s
preference. For example, some peers more care about the
download speed. Some more care about the quality of
downloaded files. Some may equally care about both of
them.
The result of the overall evaluation, “the interaction is
satisfying” or “not satisfying”, is used to update the peer’
trust in the file provider involved. The update is
implemented by adding the new experience into the peer’s
corresponding Bayesian network. The details are shown in
[16].
3.4 Handling recommendations
When a peer is not sure about the trustworthiness of a
file provider, it can ask other peers for recommendations.
The recommendation requests can vary according to the
peer’s needs. For example, if the peer is going to download
a movie, it may care about the movie’s quality. Another
peer may care about the download speed. So the request
can be “Does the file provider have movies with good
quality?” If the peer cares both about the quality and the
download speed, the request will be something like “Does
the file provider offer files with good quality and fast
download speed?” When other peers receive these
requests, they will check their trust representations, i.e.
their Bayesian networks, to see if they can answer such
questions. If a peer has downloaded movies form the file
provider before, it will answer the first question with its
trust in the file provider under the condition that the file
provider providers files with good quality and the second
question with its trust under the condition that the file
provider provides files with good quality and fast
download speed according to its Bayesian network.
The peer might receive several such recommendations at
the same time from trustworthy, untrustworthy
acquaintances, or strangers. If the references are
untrustworthy, the peer can discard their recommendations
immediately. Then the peer needs to combine the
recommendations from trustworthy references and from
unknown references to get the total recommendation for the
file provider. Peers may value the importance of the
recommendation from trustworthy references and from
unknown references differently. Since peers often have
different preferences and points of view, the peer’s
trustworthy acquaintances are those peers that share similar
preferences and viewpoints with the peer most of time. The
peer should weight the recommendations from its
trustworthy acquaintances higher than the recommendations
from strangers. Given a threshold θ , if the total
recommendation value is greater than θ , the peer will
interact with the file provider; otherwise, not.
If the peer interacts with the file provider, it will not
only update its trust in the file provider, i.e. its
corresponding Bayesian network, but also its trust in the
referee-peers that provide recommendations by the
following reinforcement learning formula:
α
αα etrtr
o
ij
n
ij
*)1(* −+= (1)
n
ij
tr denotes the new trust value that the
th
i peer has in
the
th
j referee after the update;
o
ij
tr denotes the old trust
value.
α
is the learning rate – a real number in the interval
[0,1].
α
e is the new evidence value, which can be -1 or 1.
If the value of recommendation is greater than θ and the
interaction with the file provider afterwards is successful,
α
e is equal to 1. If there is a mismatch between the
recommendation and the actual experience with the file
provider, the evidence is negative, so
α
e is -1.
Another way to find if a peer is reliable in making
recommendations is the comparison between two peers’
Bayesian networks relevant to an identical file provider.
When peers are idle, they can “gossip” with each other
periodically, exchange and compare their Bayesian
networks. This can help them find other peers who share
similar preferences more accurately and faster. After each
comparison, the peers will update their trusts in each other
according the formula:
β
ββ etrtr
o
ij
n
ij
*)1(* −+= (2)
The result of the comparison
β
e is a number in the
interval [-1, 1]. β is the learning rate – a real number in
the interval [0,1], which follows the constraint αβ > . This
is because the Bayesian network collectively reflects a
Wang Y. Vassileva J. (2003) Trust and Reputation Model in Peer-to-Peer Networks, Proc. of IEEE Conference on P2P Computing,
Linkoeping, Sweden, September 2003.
peer’s preferences and viewpoints based on all its past
interactions with a specific file provider. Comparing the
two peers’ Bayesian networks is tantamount to comparing
all the past interactions of the two peers. The evidence
α
e in formula (1) is only based on one interaction. The
evidence
β
e should affect the peer’s trust in another peer
more than
α
e .
How do the peers compare their Bayesian networks and
how is
β
e computed? First, we assume all peers have the
same structure of Bayesian networks. We only compare the
values in their Bayesian networks. Suppose peer 1 will
compare its Bayesian network with the corresponding
Bayesian network of peer 2. Peer 1 gets the degree of
similarity between the two Bayesian networks by
computing the similarity of each pair of nodes (T, DS, FQ
and FT), according to the similarity measure based on
Clark’s distance [7], and then combining the similarity
results of each pair of nodes with different weight in order
to take into account peers’ preferences. So peers with
similar preferences, such as the importance of file type,
quality, and download speed, will weight each other’s
opinions higher.
In the above discussion, we assume all the peers are
truthful in making recommendations. In the situation that
peers are not truthful, our method is still suitable. Since a
file provider’s reputation is built on a collection of
recommendations, even if a few peers lie, it will not
influence the overall reputation of the file provider. If a
peer does lie to another peer, for example, peer A lies to
peer B, peer B’s trust in peer A as a referee will decrease
quickly because peer A’s recommendation does not match
peer B’s evaluation of the involved interaction.
4. Experiments
In order to evaluate this approach, we developed a
simulation of a file sharing system in a peer-to-peer
network. The system is developed on the JADE 2.5. For the
sake of simplicity, each node in our system plays only one
role at a time, either the role of file provider or the role of
a peer. At the beginning every peer knows only peers
directly connected with it and a few file providers.
Every peer has an interest vector. The interest vector is
composed of five elements: music, movie, image,
document and software. The value of each element
indicates the strength of the peer’s interests in the
corresponding file type. The files the peer wants to
download are generated based on its interest vector. Every
peer keeps two lists. One is the peer list that records all the
other peers that the peer has interacted with and its trust
values in these peers. The other is the file provider list that
records the known file providers and the corresponding
Bayesian networks representing the peer’s trusts in these
file providers. Each file provider has a capability vector
showing its capabilities in different aspects, i.e. providing
files with different types, qualities and download speeds.
Our experiments involve 10 different file providers and
40 peers. Peers will gossip with other peers periodically
(after every 5 interactions) to exchange their Bayesian
networks. The total number of interactions is 1000. We run
each configuration for 10 times and use means for the
evaluation criteria.
5. Results
0
10
20
30
40
50
60
70
100
200
300
400
500
600
700
800
900
1000
The number of interactions
Successful recommendation(%)
Trust and reputation system with BN
Trust and reputation system without BN
Figure 3. Trust and reputation system with BN
vs. trust and reputation system without BN
The goal of the first experiment is to see if a Bayesian
network-based trust model helps peers to select file
providers that match better their preferences. Therefore we
measure the system performance in terms of percentage of
successful recommendations. A successful recommendation
is defined as a positive recommendation about a file
provider such that, after receiving it and interacting with the
file provider, the peer is satisfied with the interaction. The
percentage of successful recommendation is the number of
successful recommendations divided by the number of
positive recommendations because if a peer gets a negative
recommendation for a file provider, it will not interact with
the file provider. So we are looking at the proportion of
satisfactory performance over unsatisfactory performance
after positive recommendation.
We compare the performance of a system consisting of
peers with Bayesian network-based trust models and a
system consisting of peers without Bayesian networks (BN)
trust model. These peers represent general trust only, which
is not differentiated into different aspects. So, we have two
configurations in this experiment:
Ø Trust and reputation system with BN: the system
consists of peers with Bayesian networks-based trust