Trust - Peer-to-Peer Computing: Applications, Architecture, Protocols, and Challenges - page 141

1. If the transaction amount is very high and the transaction time is new,
then the aggregation weight is very large.

2. If the transaction amount is very low or the transaction time is very old,
then the aggregation weight is small.

3. If a peer's reputation is good and the transaction amount is high, then
the aggregation weight is very large.

4. If a peer's reputation is good and the transaction amount is low, then
the aggregation weight is medium.

5. If a peer's reputation is bad, then the aggregation weight is very small.

FuzzyTrust then computes the global reputation using the following defi-
nition:

w j

R i =

j∈S w j t ji

(6.10)

j∈S

where R i is the global reputation of peer i, S is the set of peers with whom
peer i has conducted transactions, t ji is the local trust score of peer i rated
by peer j, w j is the aggregation weight of t ji . The global aggregation process
runs multiple iterations until each R i converges to a stable global reputation
rating for peer i.
Song et al. implemented the prototype FuzzyTrust system on a DHT-based
P2P overlay network, with an architecture similar to that of Chord. This
DHT ring provides fast trust aggregation and secure message transmission.
The Chord system is highly scalable, robust to failure, and self-organizing in
that it handles peer join and leave from the system. Figure 6.8 shows the
DHT-based FuzzyTrust system architecture.
Each peer maintains two tables: a transaction record table to maintain
transaction records with remote peers, and a local score table to main-
tain remote peers' evaluated trust scores. Based on the transaction records,
FuzzyTrust infers the global aggregation weights through the fuzzy inference
system. When performing global reputation aggregation, each peer queries the
trust scores from remote peers. To tackle the hot-spot issue, the system par-
tially queries qualified peers that meet an aggregation threshold. Figure 6.9
shows an example of global reputation aggregation based on the DHT config-
uration in Figure 6.8.
In Song et al.'s simulation study based on eBay trace data, FuzzyTrust
consistently outpeformed EigenTrust in terms of convergence time, error in
detecting malicious users, and message overheads.
Gri ths et al. [Gri ths et al., 2006] extended Song et al.'s FuzzyTrust
framework by incorporating a new notion called undistrust. Gri ths et al.
observed that most previous work on trust has concentrated on the positive
aspect of trust but largely ignored the notion of distrust.
Gri ths et al. insightfully observed that distrust is not simply the negation