Information Technology Reference
In-Depth Information
ing bogus prior transaction records). To tackle this problem, Feldman et al.
suggested a graph theoretic technique. To illustrate, consider the reputation
graph shown in Figure 5.2. Here, each node in the graph represents a peer (C
denotes a colluder) and each directed edge represents the perceived reputation
value (i.e., the reputation value of the node incident by the edge as perceived
by the node originating the edge). We can see that the colluders give each
other a high reputation value. On the other hand, a contributing peer (e.g.,
the top node) gives a reputation value of 0 to each colluder because the con-
tributing peer does not have any prior successful transaction carried out with
a colluder. With this graph, we can apply the maxflow algorithm to compute
the reputation value of a destination peer as perceived by a source peer. For
instance, peer B's (the destination) perceived reputation value with respect
to peer A (the source) is 0 despite that many colluders give a high reputation
value to B.
20
0
0
0
0
0
20
A
C
C
C
C
C
100
100
100
100
100
0
C
B
FIGURE 5.2: A graph depicting the perceived reputation values among
peers (C denotes a colluder) [Feldman et al., 2004a].
Finally, an adaptive stranger policy is proposed to deal with whitewashing.
Instead of always penalizing a new user (which would discourage expansion of
the P2P network), the proposed policy requires that each existing peer, before
deciding whether to do a sharing transaction with a new user, computes a ratio
of amount of services provided to amount of services consumed by a new user.
If this ratio is great than or equal to 1, then the existing peer will work with
the new user. On the other hand, if the ratio is smaller than 1, then the ratio
is treated as a probability of working with this new user.
Sun and Garcia-Molina [Sun and Garcia-Molina, 2004] suggested an in-
Search WWH ::
Custom Search