Research On Multi-Dimensional Trust Sequential Patterns Mining And Defense Mechanism In Trust Networks

In recent years, many applications in open networks, such as P2P sharing, e-commerce and social services, have changed the ways of people’s social activities and interactions. However, lots of fraud behaviors and unreliable services exist in such an open and virtual social networks, which have severely affected the qualities of interaction among users. So the security issue has become a critical factor for the development of these applications.Trust networks can effectively reduce the risks of the interaction between both sides, and improve the quality of services and reliability of system, that currently becomes a hot research issue. Due to having potential great benefits, trust networks also suffer a variety of malicious attacks, leading to unreliable trust relationships established between users, which greatly affect the security of trust networks and the performance of trust evaluation mechanism.Most current researches focus on two aspects:trust relationship computing and trust reasoning, which lack of the discussion about node importance and the correlations between nodes, and meanwhile ignore to explore the security issues of trust networks from the perspective of network structure. Moreover, the studies on the security and protection issue of trust networks have not yet drawn enough attentions. Therefore, this paper will propose a multi-dimensional sequential patterns mining method to analyze the importance of nodes in trust networks and their association structures, and deeply investigate and analyze the security influenced by different nodes in trust networks and the roles to defense malicious attacks. The main contributions of this paper are as follows:1. A multi-dimensional trust sequential patterns mining algorithm called MTrustSeq is proposed, which mainly includes two sub processes:mining the frequent trust sequences and then filtering the multi-dimensional patterns. With multiple factors such as trust strength, length of sequences and node credibility taken into account, this algorithm can effectively mine the important nodes as well as their association structures included in the multi-dimensional trust sequential patterns. The simulation experiments show that the results of the proposed MTrustSeq algorithm can comprehensively and accurately reflect the characteristics of the important nodes and correlations between them in trust networks.2. With considering the impact of node importance on security and defense, a hierarchical nodes classification algorithm called levelDiv is also proposed, which has combined with the network structural characteristics. This algorithm is used to analyze the importance of nodes in affecting trust propagation and overall network security. According to the level of node importance, the structural trust networks are divided into four categories:core nodes, important nodes, associated nodes and other nodes. Experiments show that our method not only overcomes the shortcomings of the traditional evaluation methods about node importance, such as degree, betweenness and so on, but also effectively improves the capabilities of trust network to detect and defend malicious attacks.