Malicious Nodes Identification for Complex Network Based on Local Views

Several social, biological and information systems can be described through complex network models. All complex networks display common structural features, such as the small-world and scale-free properties. However, the presence of selfish and/or malicious nodes can damage the network operation, as they may attack the network in several different ways, like not cooperating, or inserting, modifying or eliminating information in the network. Trust evaluation algorithms are a useful incentive for encouraging selfish nodes to collaborate or even to isolate malicious ones. Nodes that refrain from cooperation or present a malicious behavior get lower trust value and may be penalized as other nodes tend to cooperate only with highly trusted ones. This paper presents an algorithm to calculate the number of malicious and/or selfish nodes in a network based on the local trust views that each node has about their neighbors. The algorithm points out to the network manager exactly which nodes they are. Simulation results over four real complex networks demonstrate the effectiveness of the proposed approach. In fact, it presents an error margin smaller than 15% for 35 000 malicious or selfish nodes in a network of 70 000 nodes. If the number of malicious nodes goes under 5000, the error margin is around one node.