Özet:

Nowadays trust has become one of the main challenges in every online social network. The open and decentralized nature of P2P network makes it vulnerable to be exploited by malicious users to distribute tampered contents and harmful programs. We propose a dynamic computational trust model based on a neuro-fuzzy network and implement it on Gnutella network which is an unstructured peer-to-peer file sharing system. The model is a novel application of control system in evaluating trust and reputation in peer-to-peer networks. The neuro-fuzzy network we are using is the Local Linear Model Tree (LOLIMOT); an incremental algorithm which partitions the input space in every iteration. We show the performance of our trust and reputation system by simulation, we model the behavior of malicious peers in the simulated network. Results show that the neuro-fuzzy model detects and follows up the behavior of malicious users in an accurate, fast way in comparison with simple models of trust and reputation. It also better estimates the value of trust and thus reduces the mean-square error of the output.

Anahtar Kelimeler: