Research On Several Key Issues In Complex Online Reputation System

With the development of Internet technologies, it emerges a lot of large scale distributed systems, such as Ubiquitous computing, Grid computing, Peer-to-Peer (P2P) computing, Ad hoc networks, etc. Because of the dynamics, openness and autonomy nature of these systems, it may harm users' benefit under the increasing risks. Therefore, there are big challenges how to solve the trust issues (e.g., security, fairness) among participants in this kind of environment. Reputation systems which designed to address these problems represent a significant trend in decision support for internet mediated service provision. The basic idea is to let parties rate each other after the completion of an interaction, and use the aggregated ratings about a given party to derive a reputation score, which can assist other parties in deciding whether or not interact with that party in near future. Facing the weakness of the existing work, in this dissertation, we highlight several key research topics, such as, granularity of trust evaluation, cooperation theory and trust overlay.The contributions of this dissertation are list as follows:(1) Granularity of Trust Evaluation. We propose a fine-grained trust computation model for P2P networks. Our model defines the service as a fined-grained quality-of-service (QoS) (N-dimensional vector), and in order to accurate the recommendation trust computing, several concepts are introduced to reflect the recommenders'current status, history behavior, and the gap between these two behaviors. Also, we firstly introduce the Gauss-bar function to measure the preference similarity between peers. All these will result in a flexible model which represents trust in a manner more close to human intuitions and satisfies the diverse QoS requirements of peers in P2P networks.(2) Cooperation Theory. We propose a group-based scheme to enhance cooperation for online reputation systems. In this section, we define related concepts to capture the social structure and ties among participants in reputation system, e.g., group, assortativity, etc. Also, we use Femi distribution function to reflect the bounded rationality of participants and the existing of stochastic factors in evolutionary process.(3) Trust Overlay. We propose two methods to analyze structured trust overlay and un-structured trust overlay, repectively. In the first method, we analyze the security issues in PKI trust model, especially for Bridge Certificate Authorities (BCA), by using the concept of "impersonation risk", and prove the upper of "impersonation risk" of BCA trust model is 1-(n-1)/2mn. Based on that, we finally propose a new modified BCA model, which enhances its security, but a bit more complex in certificate path building and implementation than existing one. In the seconde method, we develop a mathematical framework to analyze the stochastic dynamics of trust structure, which enabling agents to change. Our method can instruct to how enhance agent cooperation, or design reward and punishment mechanisms in open networks.