大规模并行网络动态演化特征挖掘技术研究

传统技术无法适应动态变化的网络演化特征,容易引入很多无关节点连接信息,合理设定参数非常困难,导致动态演化特征挖掘结果不可靠。为此提出一种新的大规模并行网络动态演化特征挖掘技术。在建立的大规模并行网络中,把网络节点划分成普通节点和簇头节点,普通节点加入大规模并行网络后,被看作簇头节点,只和某个簇头构建链路,通过多跳实现数据转发,依据择优添加连接和反择优过滤节点演化。针对大规模并行网络动态演化特征,提出挖掘模型,通过初始权重对节点在网络中的初始化状态进行描述,利用突发权重,依据时间独立性对动态演化特征的突发性进行描述,采用密集权重对网络在局部时间内节点连接的密集程度进行描述,通过连续权重对网络在相同演化期间体现的连续性进行描述,依据总权重值实现动态演化特征的挖掘。实验结果表明,所提技术挖掘可靠性和实用性强。

Research on dynamic evolution feature mining in large scale parallel networks

traditional technology can not adapt to the dynamic evolution of network evolution. It is easy to introduce many irrelevant nodes to connect information. Setting parameters reasonably is very difficult, resulting in dynamic evolution characteristics mining results are not reliable. In this paper, a new dynamic evolution feature mining technique for large-scale parallel networks is proposed. In a massively parallel network, the network node is divided into ordinary nodes and cluster head nodes and common nodes join a massively parallel network, as the cluster head node, and a cluster head construction link data by multi hop forwarding, on the basis of merit add connection and anti preferential filtering node evolution. For large-scale parallel network dynamic evolution characteristics, put forward the mining model of the nodes in the network initialization state described by the initial weights, the weight of the burst, according to sudden dynamic characteristics to describe the independence of time by intensive intensive weight of node network in local time connection described by the continuity is reflected in the same network during the evolution of the continuous weight, total weight value according to the dynamic evolution characteristics of mining. The experimental results show that the proposed technology is reliable and practical.