Graph Transformers研究进展综述

随着图结构数据在各种实际场景中的广泛应用,对其进行有效建模和处理的需求日益增加。Graph Transformers(GTs)作为一类使用Transformers处理图数据的模型,能够有效缓解传统图神经网络(GNN)中存在的过平滑和过挤压等问题,因此可以学习到更好的特征表示。根据对近年来GTs相关文献的研究,将现有的模型架构分为两类:第一类通过绝对编码和相对编码向Transformers中加入图的位置和结构信息,以增强Transformers对图结构数据的理解和处理能力;第二类根据不同的方式(串行、交替、并行)将GNN与Transformers进行结合,以充分利用两者的优势。介绍了GTs在信息安全、药物发现和知识图谱等领域的应用,对比总结了不同用途的模型及其优缺点。最后,从可扩展性、复杂图、更好的结合方式等方面分析了GTs未来研究面临的挑战。

基于Graph Transformer的半监督异配图表示学习模型

现有的图卷积网络(GCN)模型基于同配性假设,无法直接应用于异配图的表示学习,且许多异配图表示学习的研究工作受消息传递机制的限制,导致节点特征混淆和特征过度挤压而出现过平滑问题。针对这些问题,提出一种基于Graph Transformer的半监督异配图表示学习模型HPGT(HeteroPhilic Graph Transformer)。首先,使用度连接概率矩阵采样节点的路径邻域,再通过自注意力机制自适应地聚合路径上的节点异配连接模式,编码得到节点的结构信息,用节点的原始属性信息和结构信息构建Transformer层的自注意力模块;其次,将每个节点自身的隐层表示与它的邻域节点的隐层表示分离更新以避免节点通过自注意力模块聚合过量的自身信息,再把每个节点表示与它的邻域表示连接,得到单个Transformer层的输出,另外,将所有的Transformer层的输出跳连到最终的节点隐层表示以防止中间层信息丢失;最后,使用线性层和Softmax层将节点的隐层表示映射到节点的预测标签。实验结果表明,与无结构编码(SE)的模型相比,基于度连接概率的SE能为Transformer层的自注意力模块提供有效的偏差信息,HPGT平均准确率提升0.99%~11.98%;与对比模型相比,在异配数据集(Texas、Cornell、Wisconsin和Actor)上,模型节点分类准确率提升0.21%~1.69%,在同配数据集(Cora、CiteSeer和PubMed)上,节点分类准确率分别达到了0.8379、0.7467和0.8862。以上结果验证了HPGT具有较强的异配图表示学习能力,尤其适用于强异配图节点分类任务。

GraphMLP-Mixer:基于图-多层感知机架构的高效多行为序列推荐方法

在多行为序列推荐领域,图神经网络(GNNs)虽被广泛应用,但存在局限性,如对序列间协同信号建模不足和处理长距离依赖性等问题.针对这些问题,提出了一种新的解决框架GraphMLP-Mixer.该框架首先构造全局物品图来增强模型对序列间协同信号的建模,然后将感知机-混合器架构与图神经网络结合,得到图-感知机混合器模型对用户兴趣进行充分挖掘.GraphMLP-Mixer具有2个显著优势:一是能够有效捕捉用户行为的全局依赖性,同时减轻信息过压缩问题;二是其时间与空间效率显著提高,其复杂度与用户交互行为的数量成线性关系,优于现有基于GNN多行为序列推荐模型.在3个真实的公开数据集上进行实验,大量的实验结果验证了GraphMLP-Mixer在处理多行为序列推荐问题时的有效性和高效性.

基于GraphSAGE网络的藏文短文本分类研究

文本分类是自然语言处理领域的重要研究方向,由于藏文数据的稀缺性、语言学特征抽取的复杂性、篇章结构的多样性等因素导致藏文文本分类任务进展缓慢。因此,该文以图神经作为基础模型进行改进。首先,在“音节-音节”“音节-文档”建模的基础上,融合文档特征,采用二元分类模型动态网络构建“文档-文档”边,以充分挖掘短文本的全局特征,增加滑动窗口,减少模型的计算复杂度并寻找最优窗口取值。其次,针对藏文短文本的音节稀疏性,首次引入GraphSAGE作为基础模型,并探究不同聚合方式在藏文短文本分类上的性能差异。最后,为捕获节点间关系的异质性,对邻居节点进行特征加权再平均池化以增强模型的特征提取能力。在TNCC标题文本数据集上,该文模型的分类准确率达到了62.50%,与传统GCN、原始GraphSAGE和预训练语言模型CINO相比,该方法在分类准确率上分别提高了2.56%、1%和2.4%。

基于Graph-LSTMs的双重位置感知方面级情感分类

目前针对用户评论中方面词项情感分类任务的研究大多忽略了依存句法信息,或并未建立依存句法结构与单词之间的联系。为此,提出一种基于Graph-LSTMs的双重位置感知方面级情感分类方法。通过Graph-LSTMs学习词项的上下文语境特征;在双向GRU的输入中拼接具有双重位置信息的位置向量,优化句子情感编码;利用注意力机制捕获关键的情感特征,实现分类。在SemEval2014的两个数据集上的实验结果表明,该模型相比几种基线模型在准确率和Macro-F1这两个指标上提升明显。

基于路径存储表的Hashgraph共识算法优化与实现

Hashgraph是一种数据采用有向无环图(DAG)结构的区块链共识算法,Hashgraph引入了虚拟投票的概念,允许节点在无额外通信开销的情况下并发出块,实现异步场景下的拜占庭容错。然而,Hashgraph提出的虚拟投票算法存在算法时间复杂度较高、共识运行逻辑过于复杂等问题。为此,提出一种基于路径存储表的Hashgraph优化方案。首先,提出一种基于顶点可达表的见证人判定方法,通过存储路径的方式实时记录生成事件与历史事件的可达关系,在轮次划分阶段,通过查询顶点事件的可达信息取代回溯算法,降低见证人判断算法的时间复杂度;其次,针对顶点可达表无法跨轮次判断事件关系的问题,提出一种基于历史可达表的知名见证人判定方法,历史可达表将存储见证人与历史事件之间的可达关系,通过查询历史可达表解决知名见证人判定阶段需要反复回溯视图的问题;最后,根据顶点可达表和历史可达表改进Hashgraph中复杂的共识计算,提升算法效率,加快事件确认速度。实验结果表明,所提优化方案与Hashgraph原共识算法相比,算法运行效率提升65.76%,在吞吐量方面平均提升41.27%。

A STABILITY RESULT FOR TRANSLATINGSPACELIKE GRAPHS IN LORENTZ MANIFOLDS

In this paper,we investigate spacelike graphs defined over a domain Ω⊂M^(n) in the Lorentz manifold M^(n)×ℝ with the metric−ds^(2)+σ,where M^(n) is a complete Riemannian n-manifold with the metricσ,Ωhas piecewise smooth boundary,and ℝ denotes the Euclidean 1-space.We prove an interesting stability result for translating spacelike graphs in M^(n)×ℝ under a conformal transformation.

Heterophilic Graph Neural Network Based on Spatial and Frequency Domain Adaptive Embedding Mechanism

Graph Neural Networks(GNNs)play a significant role in tasks related to homophilic graphs.Traditional GNNs,based on the assumption of homophily,employ low-pass filters for neighboring nodes to achieve information aggregation and embedding.However,in heterophilic graphs,nodes from different categories often establish connections,while nodes of the same category are located further apart in the graph topology.This characteristic poses challenges to traditional GNNs,leading to issues of“distant node modeling deficiency”and“failure of the homophily assumption”.In response,this paper introduces the Spatial-Frequency domain Adaptive Heterophilic Graph Neural Networks(SFA-HGNN),which integrates adaptive embedding mechanisms for both spatial and frequency domains to address the aforementioned issues.Specifically,for the first problem,we propose the“Distant Spatial Embedding Module”,aiming to select and aggregate distant nodes through high-order randomwalk transition probabilities to enhance modeling capabilities.For the second issue,we design the“Proximal Frequency Domain Embedding Module”,constructing adaptive filters to separate high and low-frequency signals of nodes,and introduce frequency-domain guided attention mechanisms to fuse the relevant information,thereby reducing the noise introduced by the failure of the homophily assumption.We deploy the SFA-HGNN on six publicly available heterophilic networks,achieving state-of-the-art results in four of them.Furthermore,we elaborate on the hyperparameter selection mechanism and validate the performance of each module through experimentation,demonstrating a positive correlation between“node structural similarity”,“node attribute vector similarity”,and“node homophily”in heterophilic networks.

Weighted Forwarding in Graph Convolution Networks for Recommendation Information Systems

Recommendation Information Systems(RIS)are pivotal in helping users in swiftly locating desired content from the vast amount of information available on the Internet.Graph Convolution Network(GCN)algorithms have been employed to implement the RIS efficiently.However,the GCN algorithm faces limitations in terms of performance enhancement owing to the due to the embedding value-vanishing problem that occurs during the learning process.To address this issue,we propose a Weighted Forwarding method using the GCN(WF-GCN)algorithm.The proposed method involves multiplying the embedding results with different weights for each hop layer during graph learning.By applying the WF-GCN algorithm,which adjusts weights for each hop layer before forwarding to the next,nodes with many neighbors achieve higher embedding values.This approach facilitates the learning of more hop layers within the GCN framework.The efficacy of the WF-GCN was demonstrated through its application to various datasets.In the MovieLens dataset,the implementation of WF-GCN in LightGCN resulted in significant performance improvements,with recall and NDCG increasing by up to+163.64%and+132.04%,respectively.Similarly,in the Last.FM dataset,LightGCN using WF-GCN enhanced with WF-GCN showed substantial improvements,with the recall and NDCG metrics rising by up to+174.40%and+169.95%,respectively.Furthermore,the application of WF-GCN to Self-supervised Graph Learning(SGL)and Simple Graph Contrastive Learning(SimGCL)also demonstrated notable enhancements in both recall and NDCG across these datasets.

KGTLIR:An Air Target Intention Recognition Model Based on Knowledge Graph and Deep Learning

As a core part of battlefield situational awareness,air target intention recognition plays an important role in modern air operations.Aiming at the problems of insufficient feature extraction and misclassification in intention recognition,this paper designs an air target intention recognition method(KGTLIR)based on Knowledge Graph and Deep Learning.Firstly,the intention recognition model based on Deep Learning is constructed to mine the temporal relationship of intention features using dilated causal convolution and the spatial relationship of intention features using a graph attention mechanism.Meanwhile,the accuracy,recall,and F1-score after iteration are introduced to dynamically adjust the sample weights to reduce the probability of misclassification.After that,an intention recognition model based on Knowledge Graph is constructed to predict the probability of the occurrence of different intentions of the target.Finally,the results of the two models are fused by evidence theory to obtain the target’s operational intention.Experiments show that the intention recognition accuracy of the KGTLIRmodel can reach 98.48%,which is not only better than most of the air target intention recognition methods,but also demonstrates better interpretability and trustworthiness.

AFSTGCN:Prediction for multivariate time series using an adaptive fused spatial-temporal graph convolutional network

The prediction for Multivariate Time Series(MTS)explores the interrelationships among variables at historical moments,extracts their relevant characteristics,and is widely used in finance,weather,complex industries and other fields.Furthermore,it is important to construct a digital twin system.However,existing methods do not take full advantage of the potential properties of variables,which results in poor predicted accuracy.In this paper,we propose the Adaptive Fused Spatial-Temporal Graph Convolutional Network(AFSTGCN).First,to address the problem of the unknown spatial-temporal structure,we construct the Adaptive Fused Spatial-Temporal Graph(AFSTG)layer.Specifically,we fuse the spatial-temporal graph based on the interrelationship of spatial graphs.Simultaneously,we construct the adaptive adjacency matrix of the spatial-temporal graph using node embedding methods.Subsequently,to overcome the insufficient extraction of disordered correlation features,we construct the Adaptive Fused Spatial-Temporal Graph Convolutional(AFSTGC)module.The module forces the reordering of disordered temporal,spatial and spatial-temporal dependencies into rule-like data.AFSTGCN dynamically and synchronously acquires potential temporal,spatial and spatial-temporal correlations,thereby fully extracting rich hierarchical feature information to enhance the predicted accuracy.Experiments on different types of MTS datasets demonstrate that the model achieves state-of-the-art single-step and multi-step performance compared with eight other deep learning models.

An End-To-End Hyperbolic Deep Graph Convolutional Neural Network Framework

Graph Convolutional Neural Networks(GCNs)have been widely used in various fields due to their powerful capabilities in processing graph-structured data.However,GCNs encounter significant challenges when applied to scale-free graphs with power-law distributions,resulting in substantial distortions.Moreover,most of the existing GCN models are shallow structures,which restricts their ability to capture dependencies among distant nodes and more refined high-order node features in scale-free graphs with hierarchical structures.To more broadly and precisely apply GCNs to real-world graphs exhibiting scale-free or hierarchical structures and utilize multi-level aggregation of GCNs for capturing high-level information in local representations,we propose the Hyperbolic Deep Graph Convolutional Neural Network(HDGCNN),an end-to-end deep graph representation learning framework that can map scale-free graphs from Euclidean space to hyperbolic space.In HDGCNN,we define the fundamental operations of deep graph convolutional neural networks in hyperbolic space.Additionally,we introduce a hyperbolic feature transformation method based on identity mapping and a dense connection scheme based on a novel non-local message passing framework.In addition,we present a neighborhood aggregation method that combines initial structural featureswith hyperbolic attention coefficients.Through the above methods,HDGCNN effectively leverages both the structural features and node features of graph data,enabling enhanced exploration of non-local structural features and more refined node features in scale-free or hierarchical graphs.Experimental results demonstrate that HDGCNN achieves remarkable performance improvements over state-ofthe-art GCNs in node classification and link prediction tasks,even when utilizing low-dimensional embedding representations.Furthermore,when compared to shallow hyperbolic graph convolutional neural network models,HDGCNN exhibits notable advantages and performance enhancements.

Position-Aware and Subgraph Enhanced Dynamic Graph Contrastive Learning on Discrete-Time Dynamic Graph

Unsupervised learning methods such as graph contrastive learning have been used for dynamic graph represen-tation learning to eliminate the dependence of labels.However,existing studies neglect positional information when learning discrete snapshots,resulting in insufficient network topology learning.At the same time,due to the lack of appropriate data augmentation methods,it is difficult to capture the evolving patterns of the network effectively.To address the above problems,a position-aware and subgraph enhanced dynamic graph contrastive learning method is proposed for discrete-time dynamic graphs.Firstly,the global snapshot is built based on the historical snapshots to express the stable pattern of the dynamic graph,and the random walk is used to obtain the position representation by learning the positional information of the nodes.Secondly,a new data augmentation method is carried out from the perspectives of short-term changes and long-term stable structures of dynamic graphs.Specifically,subgraph sampling based on snapshots and global snapshots is used to obtain two structural augmentation views,and node structures and evolving patterns are learned by combining graph neural network,gated recurrent unit,and attention mechanism.Finally,the quality of node representation is improved by combining the contrastive learning between different structural augmentation views and between the two representations of structure and position.Experimental results on four real datasets show that the performance of the proposed method is better than the existing unsupervised methods,and it is more competitive than the supervised learning method under a semi-supervised setting.

GATiT:An Intelligent Diagnosis Model Based on Graph Attention Network Incorporating Text Representation in Knowledge Reasoning

The growing prevalence of knowledge reasoning using knowledge graphs(KGs)has substantially improved the accuracy and efficiency of intelligent medical diagnosis.However,current models primarily integrate electronic medical records(EMRs)and KGs into the knowledge reasoning process,ignoring the differing significance of various types of knowledge in EMRs and the diverse data types present in the text.To better integrate EMR text information,we propose a novel intelligent diagnostic model named the Graph ATtention network incorporating Text representation in knowledge reasoning(GATiT),which comprises text representation,subgraph construction,knowledge reasoning,and diagnostic classification.In the text representation process,GATiT uses a pre-trained model to obtain text representations of the EMRs and additionally enhances embeddings by including chief complaint information and numerical information in the input.In the subgraph construction process,GATiT constructs text subgraphs and disease subgraphs from the KG,utilizing EMR text and the disease to be diagnosed.To differentiate the varying importance of nodes within the subgraphs features such as node categories,relevance scores,and other relevant factors are introduced into the text subgraph.Themessage-passing strategy and attention weight calculation of the graph attention network are adjusted to learn these features in the knowledge reasoning process.Finally,in the diagnostic classification process,the interactive attention-based fusion method integrates the results of knowledge reasoning with text representations to produce the final diagnosis results.Experimental results on multi-label and single-label EMR datasets demonstrate the model’s superiority over several state-of-theart methods.

BLOW-UP CONDITIONS FOR A SEMILINEAR PARABOLIC SYSTEM ON LOCALLY FINITE GRAPHS

In this paper, we investigate a blow-up phenomenon for a semilinear parabolic system on locally finite graphs. Under some appropriate assumptions on the curvature condition CDE’(n,0), the polynomial volume growth of degree m, the initial values, and the exponents in absorption terms, we prove that every non-negative solution of the semilinear parabolic system blows up in a finite time. Our current work extends the results achieved by Lin and Wu (Calc Var Partial Differ Equ, 2017, 56: Art 102) and Wu (Rev R Acad Cien Serie A Mat, 2021, 115: Art 133).

Heuristic Expanding Disconnected Graph:A Rapid Path Planning Method for Mobile Robots

Existing mobile robots mostly use graph search algorithms for path planning,which suffer from relatively low planning efficiency owing to high redundancy and large computational complexity.Due to the limitations of the neighborhood search strategy,the robots could hardly obtain the most optimal global path.A global path planning algorithm,denoted as EDG*,is proposed by expanding nodes using a well-designed expanding disconnected graph operator(EDG)in this paper.Firstly,all obstacles are marked and their corners are located through the map pre-processing.Then,the EDG operator is designed to find points in non-obstruction areas to complete the rapid expansion of disconnected nodes.Finally,the EDG*heuristic iterative algorithm is proposed.It selects the candidate node through a specific valuation function and realizes the node expansion while avoiding collision with a minimum offset.Path planning experiments were conducted in a typical indoor environment and on the public dataset CSM.The result shows that the proposed EDG*reduced the planning time by more than 90%and total length of paths reduced by more than 4.6%.Compared to A*,Dijkstra and JPS,EDG*does not show an exponential explosion effect in map size.The EDG*showed better performance in terms of path smoothness,and collision avoidance.This shows that the EDG*algorithm proposed in this paper can improve the efficiency of path planning and enhance path quality.

A graph neural network approach to the inverse design for thermal transparency with periodic interparticle system

Recent years have witnessed significant advances in utilizing machine learning-based techniques for thermal metamaterial-based structures and devices to attain favorable thermal transport behaviors.Among the various thermal transport behaviors,achieving thermal transparency stands out as particularly desirable and intriguing.Our earlier work demonstrated the use of a thermal metamaterial-based periodic interparticle system as the underlying structure for manipulating thermal transport behavior and achieving thermal transparency.In this paper,we introduce an approach based on graph neural network to address the complex inverse design problem of determining the design parameters for a thermal metamaterial-based periodic interparticle system with the desired thermal transport behavior.Our work demonstrates that combining graph neural network modeling and inference is an effective approach for solving inverse design problems associated with attaining desirable thermal transport behaviors using thermal metamaterials.

Dynamic interwell connectivity analysis of multi-layer waterflooding reservoirs based on an improved graph neural network

The analysis of interwell connectivity plays an important role in the formulation of oilfield development plans and the description of residual oil distribution. In fact, sandstone reservoirs in China's onshore oilfields generally have the characteristics of thin and many layers, so multi-layer joint production is usually adopted. It remains a challenge to ensure the accuracy of splitting and dynamic connectivity in each layer of the injection-production wells with limited field data. The three-dimensional well pattern of multi-layer reservoir and the relationship between injection-production wells can be equivalent to a directional heterogeneous graph. In this paper, an improved graph neural network is proposed to construct an interacting process mimics the real interwell flow regularity. In detail, this method is used to split injection and production rates by combining permeability, porosity and effective thickness, and to invert the dynamic connectivity in each layer of the injection-production wells by attention mechanism.Based on the material balance and physical information, the overall connectivity from the injection wells,through the water injection layers to the production layers and the output of final production wells is established. Meanwhile, the change of well pattern caused by perforation, plugging and switching of wells at different times is achieved by updated graph structure in spatial and temporal ways. The effectiveness of the method is verified by a combination of reservoir numerical simulation examples and field example. The method corresponds to the actual situation of the reservoir, has wide adaptability and low cost, has good practical value, and provides a reference for adjusting the injection-production relationship of the reservoir and the development of the remaining oil.

Feature Matching via Topology-Aware Graph Interaction Model

Feature matching plays a key role in computer vision. However, due to the limitations of the descriptors, the putative matches are inevitably contaminated by massive outliers.This paper attempts to tackle the outlier filtering problem from two aspects. First, a robust and efficient graph interaction model,is proposed, with the assumption that matches are correlated with each other rather than independently distributed. To this end, we construct a graph based on the local relationships of matches and formulate the outlier filtering task as a binary labeling energy minimization problem, where the pairwise term encodes the interaction between matches. We further show that this formulation can be solved globally by graph cut algorithm. Our new formulation always improves the performance of previous localitybased method without noticeable deterioration in processing time,adding a few milliseconds. Second, to construct a better graph structure, a robust and geometrically meaningful topology-aware relationship is developed to capture the topology relationship between matches. The two components in sum lead to topology interaction matching(TIM), an effective and efficient method for outlier filtering. Extensive experiments on several large and diverse datasets for multiple vision tasks including general feature matching, as well as relative pose estimation, homography and fundamental matrix estimation, loop-closure detection, and multi-modal image matching, demonstrate that our TIM is more competitive than current state-of-the-art methods, in terms of generality, efficiency, and effectiveness. The source code is publicly available at http://github.com/YifanLu2000/TIM.

Survey and Prospect for Applying Knowledge Graph in Enterprise Risk Management

Enterprise risk management holds significant importance in fostering sustainable growth of businesses and in serving as a critical element for regulatory bodies to uphold market order.Amidst the challenges posed by intricate and unpredictable risk factors,knowledge graph technology is effectively driving risk management,leveraging its ability to associate and infer knowledge from diverse sources.This review aims to comprehensively summarize the construction techniques of enterprise risk knowledge graphs and their prominent applications across various business scenarios.Firstly,employing bibliometric methods,the aim is to uncover the developmental trends and current research hotspots within the domain of enterprise risk knowledge graphs.In the succeeding section,systematically delineate the technical methods for knowledge extraction and fusion in the standardized construction process of enterprise risk knowledge graphs.Objectively comparing and summarizing the strengths and weaknesses of each method,we provide recommendations for addressing the existing challenges in the construction process.Subsequently,categorizing the applied research of enterprise risk knowledge graphs based on research hotspots and risk category standards,and furnishing a detailed exposition on the applicability of technical routes and methods.Finally,the future research directions that still need to be explored in enterprise risk knowledge graphs were discussed,and relevant improvement suggestions were proposed.Practitioners and researchers can gain insights into the construction of technical theories and practical guidance of enterprise risk knowledge graphs based on this foundation.