No abstract

Abstract-this article hopes to learn rules of driver behavior habit, which is useful for early warning. Those rules are divided into single behavior rules and multiple behavior rules. C4.5 is used to learn single behavior rules; resolution principle and correlation principle are combined and used to learn multiple behavior rules. At last rules that have conflict are filtered by a multi-attribute compositional method. Experiments show that the obtained rules are both efficient and comprehensive.

A local immunization strategy, which is effective in huge no-scale network, is proposed in the paper. When there is an outbreak of virus in network, vaccinating limit nodes can reduce the damage of virus by preventing the propagation. Immunization strategy, which has been a hot topic for years, manages to choose the proper nodes to vaccinate. However, with the modern network growing large rapidly, the classic immunization strategies, such as degree-based strategies and betweenness-based strategies, will face the efficiency problem and accuracy problem. To solve the problem, we have pro-posed a local immunization strategy, which performs effectively in huge no-scale network. Firstly, we train an unsupervised graph neural network to get the embedding of nodes. Then with the help of node's embedding, we find the nodes exposed to virus spreader most severely and choose them as the vaccinated target. The calculation of node's exposure is local and fast, and it can depict the node's suspicious exposure to the virus from the global view of the network, which makes our immunization strategy approximate the global optimal solution. In addition, the vaccinating targets are calculated timely according to real-time distribution of virus spreaders, by this way, our strategy is adaptive to the fast changes in the propagation of virus. At last, we refine the time complexity to make our strategy practical in huge network. Compared with current method, our network immunization strategy shows privilege in the simulation experiment.

No abstract