In peer-to-peer (P2P) service-oriented environments, a peer may need to interact with unknown peers for the services or products provided. Thus the trust evaluation prior to and posterior to interactions becomes a very important issue, which may be based on other peers’ recommendations/evaluations. This chapter presents a dynamic peer trust evaluation model, which aims to measure responding peers’ recommendation trust, and hence filter out low credibility recommendations and obtain more accurate and objective trust values. In our model, prior to any interaction with an unknown peer (target peer), the mean trust value results from the evaluations (recommendations) given by responding peers. Posterior to interactions with the target peer, the trust values are aggregated from both responding peers’ recommendations and the requesting peer’s experience. On aggregating trust values, the weight to the requesting peer’s evaluation becomes bigger and bigger. Meanwhile, during this process, the credibility (recommendation trust) of each responding peer’s recommendation can be measured round by round. This helps filter out low credibility peers and improve the trust evaluation accuracy.
TopIntroduction
Peer-to-Peer (P2P) network is an infrastructure where each peer can play the role of a client and a service provider at the same time. P2P technique has been widely used in the field of information-sharing systems. For example, Gnutella and Napster are the most well-known systems for sharing music or movie files. Meanwhile, P2P e-commerce systems (Anancha 2003; eBay.com; P2PBazaar.com) let P2P networks go beyond the scope of information sharing systems like Napster and GNutella.
However, as most P2P systems lack the central management, the dynamic status of each peer causes trust evaluation a very complex issue. For example, in the information-sharing systems, it may consume several hours and some monetary cost of Internet access to download a large file (e.g. a video file). The provided file may be incomplete leading to a low quality of the service. In e-commerce environments, trust is also a critical issue (Resnick 2000). Before interacting with an unknown seller, it is rational for a buyer to doubt the trustworthiness of the seller, which results from the quality of previous transactions. Therefore, trust evaluations are useful and can make a new transaction securer.
Traditionally in security area, the common mechanism that has been used to identify the interacting entity is based on identity-based certificates. A registered peer should apply for a certificate from a Certificate Authority (CA) that can be used for authenticating the peer itself to other peers. Authentication has been used to form the basis of “trust” in deciding whether to carry out a transaction or not. However it is clear that such an approach is very limited as certificates do not necessarily convey much about the level of trust that one peer is willing to place on the other.
An alternative technique is to take into account of previous history of interactions with the target peer (Kamvar et al, 2003; Marti & Garcia-Molina 2004; Xiong & Liu 04). For example, each peer (e.g., a buyer) can rate the seller after each transaction. The rating is used to reflect the quality of the transaction delivered. Assuming that a lot of peers maintain this kind of local ratings, when an end-peer (referred to as a requesting peer) is going to have an interaction with a serving peer (referred to as a target peer), it can send a request to other peers. Once receiving the request, the peer (referred to as a responding peer) with a transaction with the target peer can respond to the request with its ratings. Thus, by collecting feedbacks from responding peers about their previous interactions, the end-peer may analyze and determine the trust value of the target peer being investigated.
The above method is the typical process for trust evaluation. However, in such a method, another issue arises. Each responding peer may be trustworthy or not from the point view of the requesting peer. In general, it is very difficult for a requesting peer to measure the trustworthiness of a responding peer if the requesting peer has no knowledge about each responding peer. But this is also a critical issue to the responding peer. On one hand, good recommendation trust of responding peer can lead to objective trust result. On the other hand, as the requesting peer may need to know the trust level of other target peers. Knowledge of recommendation trust of other peers can avoid obtaining misleading feedback in later trust evaluations. Namely, measuring the recommendation trust of a responding peer is a critical and challenging issue in the field, which is also the focus of this work.