A Privacy Preservation Method for Attributed Social Network Based on Negative Representation of Information

Due to the presence of a large amount of personal sensitive information in social networks,privacy preservation issues in social networks have attracted the attention of many scholars.Inspired by the self-nonself discrimination paradigmin the biological immune system,the negative representation of information indicates features such as simplicity and efficiency,which is very suitable for preserving social network privacy.Therefore,we suggest a method to preserve the topology privacy and node attribute privacy of attribute social networks,called AttNetNRI.Specifically,a negative survey-based method is developed to disturb the relationship between nodes in the social network so that the topology structure can be kept private.Moreover,a negative database-based method is proposed to hide node attributes,so that the privacy of node attributes can be preserved while supporting the similarity estimation between different node attributes,which is crucial to the analysis of social networks.To evaluate the performance of the AttNetNRI,empirical studies have been conducted on various attribute social networks and compared with several state-of-the-art methods tailored to preserve the privacy of social networks.The experimental results show the superiority of the developed method in preserving the privacy of attribute social networks and demonstrate the effectiveness of the topology disturbing and attribute hiding parts.The experimental results show the superiority of the developed methods in preserving the privacy of attribute social networks and demonstrate the effectiveness of the topological interference and attribute-hiding components.

Heterogeneous graph construction and node representation learning method of Treatise on Febrile Diseases based on graph convolutional network

Objective To construct symptom-formula-herb heterogeneous graphs structured Treatise on Febrile Diseases(Shang Han Lun,《伤寒论》)dataset and explore an optimal learning method represented with node attributes based on graph convolutional network(GCN).Methods Clauses that contain symptoms,formulas,and herbs were abstracted from Treatise on Febrile Diseases to construct symptom-formula-herb heterogeneous graphs,which were used to propose a node representation learning method based on GCN−the Traditional Chinese Medicine Graph Convolution Network(TCM-GCN).The symptom-formula,symptom-herb,and formula-herb heterogeneous graphs were processed with the TCM-GCN to realize high-order propagating message passing and neighbor aggregation to obtain new node representation attributes,and thus acquiring the nodes’sum-aggregations of symptoms,formulas,and herbs to lay a foundation for the downstream tasks of the prediction models.Results Comparisons among the node representations with multi-hot encoding,non-fusion encoding,and fusion encoding showed that the Precision@10,Recall@10,and F1-score@10 of the fusion encoding were 9.77%,6.65%,and 8.30%,respectively,higher than those of the non-fusion encoding in the prediction studies of the model.Conclusion Node representations by fusion encoding achieved comparatively ideal results,indicating the TCM-GCN is effective in realizing node-level representations of heterogeneous graph structured Treatise on Febrile Diseases dataset and is able to elevate the performance of the downstream tasks of the diagnosis model.

Modularity-based representation learning for networks

Network embedding aims at learning low-dimensional representation of vertexes in a network and effectively preserving network structures.These representations can be used as features for many complex tasks on networks such as community detection and multi-label classification.Some classic methods based on the skip-gram model have been proposed to learn the representation of vertexes.However,these methods do not consider the global structure(i.e.,community structure)while sampling vertex sequences in network.To solve this problem,we suggest a novel sampling method which takes community information into consideration.It first samples dense vertex sequences by taking advantage of modularity function and then learns vertex representation by using the skip-gram model.Experimental results on the tasks of community detection and multi-label classification show that our method outperforms three state-of-the-art methods on learning the vertex representations in networks.

Improved Density Peaking Algorithm for Community Detection Based on Graph Representation Learning

There is a large amount of information in the network data that we canexploit. It is difficult for classical community detection algorithms to handle network data with sparse topology. Representation learning of network data is usually paired with clustering algorithms to solve the community detection problem.Meanwhile, there is always an unpredictable distribution of class clusters outputby graph representation learning. Therefore, we propose an improved densitypeak clustering algorithm (ILDPC) for the community detection problem, whichimproves the local density mechanism in the original algorithm and can betteraccommodate class clusters of different shapes. And we study the communitydetection in network data. The algorithm is paired with the benchmark modelGraph sample and aggregate (GraphSAGE) to show the adaptability of ILDPCfor community detection. The plotted decision diagram shows that the ILDPCalgorithm is more discriminative in selecting density peak points compared tothe original algorithm. Finally, the performance of K-means and other clusteringalgorithms on this benchmark model is compared, and the algorithm is proved tobe more suitable for community detection in sparse networks with the benchmarkmodel on the evaluation criterion F1-score. The sensitivity of the parameters ofthe ILDPC algorithm to the low-dimensional vector set output by the benchmarkmodel GraphSAGE is also analyzed.

Research on Node Classification Based on Joint Weighted Node Vectors

Node of network has lots of information, such as topology, text and label information. Therefore, node classification is an open issue. Recently, one vector of node is directly connected at the end of another vector. However, this method actually obtains the performance by extending dimensions and considering that the text and structural information are one-to-one, which is obviously unreasonable. Regarding this issue, a method by weighting vectors is proposed in this paper. Three methods, negative logarithm, modulus and sigmoid function are used to weight-trained vectors, then recombine the weighted vectors and put them into the SVM classifier for evaluation output. By comparing three different weighting methods, the results showed that using negative logarithm weighting achieved better results than the other two using modulus and sigmoid function weighting, and was superior to directly concatenating vectors in the same dimension.

基于Node2vec的改进算法的研究

针对图节点的多标签分类任务,在Node2vec算法的基础上进行了改进,在原来随机游走的基础上加上部分标签信息的指导,然后对节点进行向量表示。算法首先根据每一个图节点及其邻居节点的标签信息和事先设定好的游走参数的值,计算当前节点的邻居节点被游走的概率;然后由概率值和其他设定好的游走的参数开始游走,得到若干条路径;之后再调用Word2vec方法对若干条游走路径进行训练,将每个图节点表示成向量。最后,通过使用逻辑分类模型对节点的特征表示进行多标签分类来验证算法的有效性。实验结果证明,通过使用标签信息的指导,多标签分类的准确率有明显提升。

融合二连通模体结构信息的节点分类算法

节点表示学习将图结构数据信息编码到低维的潜在空间中,在节点分类、聚类、链路预测等机器学习任务中被广泛应用。在复杂网络中,节点与节点之间不仅存在直接相连的低阶结构,也存在以特殊连接模式形成的高阶结构,称为模体。提出一种融合二连通模体结构信息的节点分类算法(FMI),利用节点间高阶二连通模体信息学习节点表示,完成节点分类任务。首先,统计网络中的二连通模体,利用其中信息提出一个节点重要性的度量指标——模体比值。根据模体比值计算采样概率进行邻域采样;构造一个带权辅助图以融合网络节点连接的低阶关系与高阶关系,对节点进行加权邻域聚合以得到节点表示。在5个数据集Cora、Citeseer、Pubmed、Wiki和DBLP上执行节点分类任务,与5种经典基准算法进行对比,所提算法FMI在准确度和F1-分数等指标上表现良好。

基于深度学习节点表示的谣言源定位方法

随着互联网的普及,网络上的信息以惊人的速度传播给公众。然而,由于级联效应,虚假信息和谣言同时也在迅速蔓延,对社会造成了巨大的危害。在社交网络上找到谣言的传播源头,对抑制谣言的传播起到至关重要的作用。传统的谣言源定位方法大多未能够融合多源特征且定位准确率还需进一步提高,因此,提出一种基于深度学习的谣言源定位方法,根据观测受谣言影响的节点多源特征来识别谣言源。首先,根据节点与观测节点之间的影响力相似度得到节点的影响力向量。然后,利用自编码网络对节点的影响力向量进行编码,得到包含节点信息、扩散路径和传播时间信息在内的节点的新的嵌入表示。最后,根据节点新的影响力向量计算节点为谣言源的概率,以定位谣言源。在2个模拟网络和4个真实网络上的实验结果表明,与其他方法相比,所提方法能够以更快的速度定位谣言源,且谣言源定位的准确率提升了25%以上。

融合用户传播风险和节点影响力分析的虚假信息传播控制方法

在线社交网络中虚假信息传播蔓延成为当前网络空间安全治理面临的重要挑战。提出一种融合用户传播风险和节点影响力分析的虚假信息传播控制方法DDC-UPRNI(disinformation diffusion control method integrating user propagation risk and node influence analysis)。综合考虑虚假信息传播特征空间的多样性和复杂性,通过自注意力机制实现用户传播虚假信息行为维度、时间维度和内容维度特征的嵌入表示,运用改进的无监督聚类K-means++算法实现不同用户传播风险等级的自动划分;设计一种自适应加权策略实现对离散粒子群优化算法的改进,进而提出一种基于离散粒子群优化的虚假信息传播关键节点选取方法,用于从具有特定传播风险等级的用户节点集合中选取若干个具有影响力的控制驱动节点,从而实现精准、高效的虚假信息传播控制;基于现实在线社交网络平台上开展试验,结果表明,所提出的DDC-UPRNI方法与现有算法相比,在控制效果和时间复杂度等重要指标上具有明显优势。该方法为社会网络空间中的虚假信息管控治理提供重要参考。

NLGAE:一种基于改进网络结构及损失函数的图自编码器节点分类模型

利用图嵌入方法将图的拓扑结构、节点属性等高维异构信息映射到稠密的向量空间,是解决图数据由非欧空间性带来的计算不友好、邻接矩阵的高度空间复杂性等问题的主流方法。在对经典图自编码器模型GAE与VGAE所存在的问题进行分析的基础上,尝试从编码器、解码器及损失函数3个方面对基于图自编码器的图嵌入方法进行改进,提出一种基于改进网络结构及损失函数的图自编码器模型NLGAE。首先,在模型结构设计上,一方面将编码器中堆叠的图卷积层倒置,以解决GAE与VGAE中无参Decoder缺乏灵活性并且表达能力不足的问题,另一方面引入注意力机制的图卷积网络GAT来解决节点之间的权重系数固化的问题;其次,重新设计的损失函数能够同时考虑到图结构与节点特征属性两部分信息。对比实验结果表明:NLGAE作为一种无监督模型,能够学习到高质量的节点嵌入特征,在下游节点分类任务上优于DeepWalk,GAE,GrpahMAE,GATE等经典无监督模型,并且在选择合适分类模型的情况下,甚至优于GAT和GCN等有监督的图神经网络模型。

一种用于军事目标发现的舰船知识图谱表示学习框架

我国舰船编队持续扩充,在众多舰船数据中,影响较大的目标数据需要被关注。针对舰船数据难以及时分析的问题,将表示学习框架(relationalgraph Transformer network,RGTN)引入舰船知识图谱分析领域,根据舰船知识图谱的结构及语义特征,研究了一种基于表示学习的节点重要性评估方法对舰船知识图谱进行处理,实现对舰船知识图谱中节点重要性的预测。较之前舰船知识图谱的节点重要性评估算法有更好的表现,更适用于舰船知识图谱分析领域。

基于互信息最大化和聚类感知的节点表示学习

节点表示学习是研究各类图结构数据的基础.图结构数据具有复杂的结构关系和丰富的节点信息,因此如何融合图结构和节点信息学习高质量的节点表示仍是一个挑战性问题.为此,提出一种基于互信息最大化和聚类感知的节点表示学习模型.首先,对原始图使用图扩散方法构造扩散图;然后,使用图卷积网络编码两个图到低维特征空间获得节点表示和全局表示;最后,基于互信息最大化原理,最大化一个图的节点表示和另一个图的全局表示间的一致性,反之亦然.同时,将语义相似的节点聚类到同一个簇,并最大化两个图的节点表示间的聚类一致性.在两个引文数据集上的节点分类和节点聚类的实验结果表明,该模型的性能在多项指标上都优于基线方法.以Cora数据集为例,在节点分类任务上,该模型对比基线方法在准确率和F1值指标上分别提高了2.7和0.6个百分点.

基于三元互信息的图对比学习方法研究

最近,图对比学习成为一种成功的无监督图表示学习方法,大多数方法都基于最大化互信息原则,通过数据增强来得到两个视图,并最大化两个视图的互信息。然而,两个视图的互信息可能包含不利于下游任务的信息。为了克服这些缺陷,提出基于三元互信息的图对比学习框架。该框架首先对输入图进行随机数据增强来生成两个视图,使用权重共享的编码器获得两个节点表示矩阵,随后使用共享权重解码器解码两个视图的节点表示。通过对比损失函数分别计算视图之间和视图与原图之间的损失,以最大化视图之间和视图与原图之间的互信息。实验结果表明,该方法在节点分类准确性方面的表现优于基线方法,甚至超过部分监督学习方法,验证了框架的有效性。

结构影响力及标签冲突感知的图课程学习方法

近年来,图神经网络(GNNs)已成为图学习领域的热点研究问题。受益于消息传递机制,GNNs在各类基于图的任务上均取得了优越的性能。现有的GNNs方法大多基于图中所有节点的训练难度相同的假设,然而,节点在结构影响力和邻域标签异配性等方面具有明显的差异。为此,提出了一种结构影响力及标签冲突感知的图课程学习方法(SILC-GCL),基于节点的训练难度对GNNs模型进行课程学习。首先,设计了一种综合考虑节点的PageRank影响力值以及邻域标签冲突程度的训练难度测量器;其次,采用了一个训练调度器,用于在每个训练阶段选择训练难度合适的节点并生成一个由易到难的训练节点序列;最后在GNNs骨架模型上进行训练。在6个现实网络数据集上进行的节点分类实验均验证了SILC-GCL方法的有效性。

RWK-GNN:基于特征增强与子核分解的非平衡图欺诈检测算法

金融欺诈对经济和社会稳定造成了严重的威胁,因此开发有效的欺诈检测算法对于保护金融系统的完整性至关重要.目前已有多种基于图学习的欺诈检测算法应用于实际场景当中,这些方法或针对图的结构信息开展分类,或通过图卷积神经网络学习节点的嵌入式表示进行欺诈检测工作,关注角度相对单一,无法完备地在非平衡多关系图上开展欺诈检测分析.针对以上问题,本论文提出了一种结合随机游走下的特征增强与子核分解的图神经网络欺诈检测算法(Random Walk feature enhancement and Kcore subkernel decomposition Graph Neural Network,RWKGNN),该算法能够高效地挖掘出多关系不平衡图中节点层级与全局网络层级的拓扑信息,并通过子核分解算法优化图结构特征在社区演进角度上的传播与聚合,最终完成欺诈检测与识别.为验证RWK-GNN算法性能,本文使用了图神经网络欺诈检测任务常用的公开数据集进行模型训练与测试.实验结果表明,在同一评价指标下,该方法较相关机器学习算法与图神经网络算法有着较大提升,与CARE-GNN算法相比,该方法的AUC值提升了17%;与PC-GNN算法相比,该方法的AUC值提升了8%;与SIGN算法相比,该方法的AUC值提升了7%.

面向动态交通流多步预测的时空图模型

为更好表征交通路网中节点之间的动态隐式关系,提出一种基于时空数据嵌入的动态图卷积交通流预测模型。基于路网中节点之间的共现关系,利用深度游走算法将时空数据映射到嵌入空间中学习节点的向量表示;引入时隙嵌入特征与二维空间嵌入特征共同构建三维嵌入邻接张量,用于捕获时空依赖关系;在图卷积网络中添加自适应更新机制,利用循环组件演化图卷积网络的参数,以捕获图序列的动态性。将所提模型应用于基于真实交通数据集的交通流预测,结果验证了其有效性和提取路网隐式关系的准确性。

Adaptive multi-channel Bayesian graph attention network for IoT transaction security

With the rapid advancement of 5G technology,the Internet of Things(IoT)has entered a new phase of appli-cations and is rapidly becoming a significant force in promoting economic development.Due to the vast amounts of data created by numerous 5G IoT devices,the Ethereum platform has become a tool for the storage and sharing of IoT device data,thanks to its open and tamper-resistant characteristics.So,Ethereum account security is necessary for the Internet of Things to grow quickly and improve people's lives.By modeling Ethereum trans-action records as a transaction network,the account types are well identified by the Ethereum account classifi-cation system established based on Graph Neural Networks(GNNs).This work first investigates the Ethereum transaction network.Surprisingly,experimental metrics reveal that the Ethereum transaction network is neither optimal nor even satisfactory in terms of accurately representing transactions per account.This flaw may significantly impede the classification capability of GNNs,which is mostly governed by their attributes.This work proposes an Adaptive Multi-channel Bayesian Graph Attention Network(AMBGAT)for Ethereum account clas-sification to address this difficulty.AMBGAT uses attention to enhance node features,estimate graph topology that conforms to the ground truth,and efficiently extract node features pertinent to downstream tasks.An extensive experiment with actual Ethereum transaction data demonstrates that AMBGAT obtains competitive performance in the classification of Ethereum accounts while accurately estimating the graph topology.

基于空间点云数据的异质材料零件动态建模方法

针对含有多种材料,且材料相呈非均质分布或异常分布的异质材料零件,提出一种基于空间点云数据集的异质材料零件动态建模方法。该方法采用了细化STL模型和空间微四面体重构的几何建模方法;通过引入描述各材料物理特征的材料分布特征值、各材料的体分量和材料分布向量,对材料切片内的特征节点进行材料赋值,并对节点间的材料进行插值运算,层层遍历赋值,从而实现了对异质材料零件中任意点结构与材料信息的表述。在开发的异质材料零件成形系统上,通过实例测试了该建模方法的有效性和实用性。

一种大规模网络中基于节点结构特征映射的链接预测方法

网络链接预测能够获取网络中丢失链接的重要信息或进行网络的动态演变分析.现有的基于节点相似性的网络链接预测方法往往针对简单的一(多)阶邻居信息或特定类型的小型网络,设计较为复杂的计算方法,其扩展性和大规模网络中的可计算性都受到了严峻的挑战.文中基于深度学习在神经网络语言模型中应用的启发,提出了一个LsNet2Vec(Large-scale Network to Vector)模型.通过结合随机游走的网络数据集序列化方法,进行大规模的无监督机器学习,从而将网络中节点的结构特征信息映射到一个连续的、固定维度的实数向量.然后,使用学习到的节点结构特征向量,就可以迅速计算大规模网络中任意节点之间的相似度,以此来进行网络中的链接预测.通过在16个大规模真实数据集上和目前的多个基准的最优预测算法对比发现,LsNet2Vec模型所得到的预测总体效果是最优的:在保证了大规模网络中链接预测计算可行性的同时,于多个数据集上相对已有方法呈现出较大的AUC值提升,最高达8.9%.

基于图编码网络的社交网络节点分类方法

针对如何融合节点自身属性以及网络结构信息实现社交网络节点分类的问题,提出了一种基于图编码网络的社交网络节点分类算法。首先,每个节点向邻域节点传播其携带的信息;其次,每个节点通过神经网络挖掘其与邻域节点之间可能隐含的关系,并且将这些关系进行融合;最后,每个节点根据自身信息以及与邻域节点关系的信息提取更高层次的特征,作为节点的表示,并且根据该表示对节点进行分类。在微博数据集上,与经典的深度随机游走模型、逻辑回归算法有以及最近提出的图卷积网络算法相比,所提算法分类准确率均有大于8%的提升;在DBLP数据集上,与多层感知器相比分类准确率提升4.83%,与图卷积网络相比分类准确率提升0.91%。