基于深度学习的机会网络链路预测方法研究

发布时间：2018-02-23 19:39

本文关键词： 机会网络链路预测相似性指标条件深度信念网络　出处：《南昌航空大学》2017年硕士论文　论文类型：学位论文

【摘要】：链路预测的目标是根据节点间已知链路及节点的属性,估计节点间链路存在的可能性。近年来,链路预测问题正受到越来越多国内外学者的关注,这是因为准确的链路预测算法可为研究一些基本的网络演化机制及路由等上层协议提供支撑,并能应用于推荐系统、公共安全领域的舆情监控等领域。由于机会网络具有节点移动性、节点间间歇性连接和节点资源有限等特点,机会网络的链路预测是机会网络研究中的热点和难点。本文研究机会网络的链路预测方法,针对机会网络拓扑随时间变化频繁的特点,论文结合网络中链路历史信息及二阶邻居信息,构建了反映机会网络链路随时间动态变化的相似性指标O_AA;采用条件深度信念网络(Conditional Deep Belief Network,CDBN)构建链路预测模型,提取链路随时间变化的特征;以所提出的相似性指标作为基础样本,采用时间序列法构造样本空间;通过实验确定网络参数,提出单层自适应学习率,以优化训练过程;采用Logistic Regression分类器对机会网络的链路进行预测。论文采用受试者工作特征曲线(Receiver Operating Characteristic Curve,ROC)中的AUC和Precision作为评价指标,在MIT Reality和Infocom05数据集下,对本文提出的相似性指标O_AA的可行性进行了验证;采用ROC中的Precision、Accuracy作为评价指标,在MIT Reality和Infocom05数据集下设计多组对比实验,验证CDBN预测模型的有效性。实验结果表明,相似性指标O_AA能够更好地反应机会网络链路的变化情况;单层自适应学习率能够加快CDBN预测模型的收敛速度,提高了CDBN预测模型的计算效率;与深度信念网络模型相比,CDBN预测模型能够更好地提取链路随时间变化的特征,获得了更好的预测效果。
[Abstract]:The goal of link prediction is to estimate the possibility of the existence of links between nodes according to the known links between nodes and the attributes of nodes. In recent years, the problem of link prediction has been paid more and more attention by scholars at home and abroad. This is because accurate link prediction algorithms can provide support for the study of some basic network evolution mechanisms and routing protocols, and can be applied to recommendation systems. Public opinion monitoring in the field of public safety. Because of the mobility of nodes, intermittent connection between nodes and limited resources of nodes, the opportunistic network is characterized by the mobility of nodes, the intermittent connection between nodes, and the limited resources of nodes, etc. The link prediction of opportunistic networks is a hot and difficult point in the research of opportunistic networks. In this paper, the link prediction methods of opportunistic networks are studied, and the topology of opportunistic networks varies frequently with time. Combining the link history information and the second-order neighbor information in the network, this paper constructs a link prediction model using conditional Deep Belief Network (CDBNs), which reflects the dynamic changes of the opportunistic network links with time, and uses the conditional Deep Belief Network (CDBNs) to construct a link prediction model. The feature of link changing with time is extracted, the time series method is used to construct the sample space based on the proposed similarity index, the network parameters are determined by experiments, and the single-layer adaptive learning rate is proposed to optimize the training process. Logistic Regression classifier is used to predict the link of opportunistic network. In this paper, the AUC and Precision in receiver Operating Characteristic Curveroc are used as evaluation indexes under MIT Reality and Infocom05 data sets. In this paper, the feasibility of the similarity index O 'AA-presented is verified, and the prediction accuracy in ROC is used as the evaluation index. Under the MIT Reality and Infocom05 data sets, the validity of the CDBN prediction model is verified by designing a number of comparative experiments. The similarity index can better reflect the change of the opportunistic network link, and the single-layer adaptive learning rate can accelerate the convergence speed of the CDBN prediction model and improve the computational efficiency of the CDBN prediction model. Compared with the deep belief network model, the CDBN prediction model can extract the characteristics of the link with time better and obtain better prediction results.
【学位授予单位】：南昌航空大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN929.5;TP181

【相似文献】