The Factorization of Adjoint Polynomials of E^G（i）-class Graphs and Chromatically Equivalence Analysis

Let Sn be the star with n vertices, and let G be any connected graph with p vertices. We denote by Eτp＋（r-1）^G（i） the graph obtained from Sr and rG by coinciding the i-th vertex of G with the vertex of degree r - 1 of S,, while the i-th vertex of each component of （r - 1）G be adjacented to r - 1 vertices of degree 1 of St, respectively. By applying the properties of adjoint polynomials, We prove that factorization theorem of adjoint polynomials of kinds of graphs Eτp＋（r-1）^G（i）∪（r - 1）K1 （1 ≤i≤p）. Furthermore, we obtain structure characteristics of chromatically equivalent graphs of their complements.

Ontological similarity network reasoning framework

To properly compute the ontological similarity, an ontological similarity network-based reasoning framework is proposed. It structurally integrates extension-based approach, intension-based approach, the similarity network-based reasoning to exploit the implicit similarity, and the feedback from the context to validate the similarity measures. A new similarity measure is also presented to construct concept similarity network, which scales the similarity using the relative depth of the least common super-concept between any two concepts. Subsequently, the graph theory, instead of predefined knowledge rules, is applied to perform the similarity network-based reasoning such that the knowledge acquisition can be avoided. The framework has been applied to text categorization and visualization of high dimensional data. Theory analysis and the experimental results validate the proposed framework.

An Intelligent Graph Edit Distance-Based Approach for Finding Business Process Similarities

There are numerous application areas of computing similarity between process models.It includes finding similar models from a repository,controlling redundancy of process models,and finding corresponding activities between a pair of process models.The similarity between two process models is computed based on their similarity between labels,structures,and execution behaviors.Several attempts have been made to develop similarity techniques between activity labels,as well as their execution behavior.However,a notable problem with the process model similarity is that two process models can also be similar if there is a structural variation between them.However,neither a benchmark dataset exists for the structural similarity between process models nor there exist an effective technique to compute structural similarity.To that end,we have developed a large collection of process models in which structural changes are handcrafted while preserving the semantics of the models.Furthermore,we have used a machine learning-based approach to compute the similarity between a pair of process models having structural and label differences.Finally,we have evaluated the proposed approach using our generated collection of process models.

THE REGULARITY OF RANDOM GRAPH DIRECTED SELF-SIMILAR SETS

A set in Rd is called regular if its Hausdorff dimension coincides with its upper box counting dimension. It is proved that a random graph-directed self-similar set is regular a.e..

Similarity matching method of power distribution system operating data based on neural information retrieval

Operation control of power systems has become challenging with an increase in the scale and complexity of power distribution systems and extensive access to renewable energy.Therefore,improvement of the ability of data-driven operation management,intelligent analysis,and mining is urgently required.To investigate and explore similar regularities of the historical operating section of the power distribution system and assist the power grid in obtaining high-value historical operation,maintenance experience,and knowledge by rule and line,a neural information retrieval model with an attention mechanism is proposed based on graph data computing technology.Based on the processing flow of the operating data of the power distribution system,a technical framework of neural information retrieval is established.Combined with the natural graph characteristics of the power distribution system,a unified graph data structure and a data fusion method of data access,data complement,and multi-source data are constructed.Further,a graph node feature-embedding representation learning algorithm and a neural information retrieval algorithm model are constructed.The neural information retrieval algorithm model is trained and tested using the generated graph node feature representation vector set.The model is verified on the operating section of the power distribution system of a provincial grid area.The results show that the proposed method demonstrates high accuracy in the similarity matching of historical operation characteristics and effectively supports intelligent fault diagnosis and elimination in power distribution systems.

Graph Ranked Clustering Based Biomedical Text Summarization Using Top k Similarity

Text Summarization models facilitate biomedical clinicians and researchers in acquiring informative data from enormous domain-specific literature within less time and effort.Evaluating and selecting the most informative sentences from biomedical articles is always challenging.This study aims to develop a dual-mode biomedical text summarization model to achieve enhanced coverage and information.The research also includes checking the fitment of appropriate graph ranking techniques for improved performance of the summarization model.The input biomedical text is mapped as a graph where meaningful sentences are evaluated as the central node and the critical associations between them.The proposed framework utilizes the top k similarity technique in a combination of UMLS and a sampled probability-based clustering method which aids in unearthing relevant meanings of the biomedical domain-specific word vectors and finding the best possible associations between crucial sentences.The quality of the framework is assessed via different parameters like information retention,coverage,readability,cohesion,and ROUGE scores in clustering and non-clustering modes.The significant benefits of the suggested technique are capturing crucial biomedical information with increased coverage and reasonable memory consumption.The configurable settings of combined parameters reduce execution time,enhance memory utilization,and extract relevant information outperforming other biomedical baseline models.An improvement of 17%is achieved when the proposed model is checked against similar biomedical text summarizers.

Railway Passenger Flow Forecasting by Integrating Passenger Flow Relationship and Spatiotemporal Similarity

Railway passenger flow forecasting can help to develop sensible railway schedules,make full use of railway resources,and meet the travel demand of passengers.The structure of passenger flow in railway networks and the spatiotemporal relationship of passenger flow among stations are two distinctive features of railway passenger flow.Most of the previous studies used only a single feature for prediction and lacked correlations,resulting in suboptimal performance.To address the above-mentioned problem,we proposed the railway passenger flow prediction model called Flow-Similarity Attention Graph Convolutional Network(F-SAGCN).First,we constructed the passenger flow relations graph(RG)based on the Origin-Destination(OD).Second,the Passenger Flow Fluctuation Similarity(PFFS)algorithm is used to measure the similarity of passenger flow between stations,which helps construct the spatiotemporal similarity graph(SG).Then,we determine the weights of the mutual influence of different stations at different times through an attention mechanism and extract spatiotemporal features through graph convolution on the RG and SG.Finally,we fused the spatiotemporal features and the original temporal features of stations for prediction.The comparison experiments on a railway bureau’s accurate railway passenger flow data show that the proposed F-SAGCN method improved the prediction accuracy and reduced the mean absolute percentage error(MAPE)of 46 stations to 7.93%.

基于双节点-双边图神经网络的茶叶病害分类方法

传统茶叶病害分类主要依赖人工方法,此类方法费工费时,同时茶叶病害样本较少使得现有的机器学习方法的模型训练不充分,病害分类准确率不够高。针对茶炭疽病、茶黑煤病、茶饼病和茶白星病4类病害,提出一种基于双节点-双边图神经网络的茶叶病害分类方法。首先通过两分支卷积神经网络提取RGB茶叶病害特征和灰度茶叶病害特征,两分支均采用ResNet12作为骨干网络,参数独立不共享,两类特征作为图神经网络的两个子节点,以获得不同域样本所包含的病害信息;其次构建相对度量边和相似性边两类边,从而强化节点对相邻节点所含病害特征的聚合能力。最后,经过双节点特征和双边特征更新模块,实现双节点和双边交替更新,提高边特征对节点距离度量的准确性,从而实现训练样本较少条件下对茶叶病害的准确分类。本文方法和小样本学习方法进行了对比实验,结果表明,本文方法获得更高的准确率,在miniImageNet和PlantVillage数据集上5way-1shot的准确率分别达到69.30%和88.42%,5way-5shot准确率分别为82.48%和93.04%。同时在茶叶数据集TeaD-5上5way-1shot和5way-5shot准确率分别达到84.74%和86.34%。

Graph Similarity Join with K-Hop Tree Indexing

Graph similarity join has become imperative for integrating noisy and inconsistent data from multiple data sources. The edit distance is commonly used to measure the similarity between graphs. To accelerate the similarity join based on graph edit distance, in the paper, we make use of a preprocessing strategy to remove the mismatching graph pairs with significant differences. Then a novel method of building indexes for each graph is proposed by grouping the nodes which can be reached in k hops for each key node with structure conservation, which is the k-hop-tree based indexing method. Experiments on real and synthetic graph databases also confirm that our method can achieve good join quality in graph similarity join. Besides, the join process can be finished in polynomial time.

动量余弦相似度梯度优化图卷积神经网络

传统梯度下降算法仅对历史梯度进行指数加权累加,没有利用梯度的局部变化,造成优化过程越过全局最优解,即使收敛到最优解也会在最优解附近震荡,其训练图卷积神经网络会造成收敛速度慢、测试准确度低。利用相邻两次梯度的余弦相似度,动态调整学习率,提出余弦相似度梯度下降(SimGrad)算法。为进一步提升图卷积神经网络训练的收敛速度和测试准确度,减少震荡,结合动量思想提出动量余弦相似度梯度下降(NSimGrad)算法。通过收敛性分析,证明SimGrad算法、NSimGrad算法都具有O(√T)的遗憾界。在构建的三个非凸函数进行测试,并结合图卷积神经网络在四个数据集上进行实验,结果表明SimGrad算法保证了图卷积神经网络的收敛性,NSimGrad算法进一步提高图卷积神经网络训练的收敛速度和测试准确度,SimGrad、NSimGrad算法相较于Adam、Nadam具有更好的全局收敛性和优化能力。

Cantor Type Fixed Sets of Iterated Multifunction Systems Corresponding to Self-Similar Networks

We propose a new approach to the investigation of deterministic self-similar networks by using contractive iterated multifunction systems (briefly IMSs). Our paper focuses on the generalized version of two graph models introduced by Barabási, Ravasz and Vicsek ([1] [2]). We generalize the graph models using stars and cliques: both algorithm construct graph sequences such that the next iteration is always based on n replicas of the current iteration, where n is the size of the initial graph structure, being a star or a clique. We analyze these self-similar graph sequences using IMSs in function of the size of the initial star and clique, respectively. Our research uses the Cantor set for the description of the fixed set of these IMSs, which we interpret as the limit object of the analyzed self-similar networks.

基于相似网络和联合注意力的图嵌入模型

图注意力网络(graph attention network, GAT)将注意力机制与图神经网络融合,但模型只关注节点的一阶邻域节点,缺乏对高阶相似节点的考虑,同时在计算注意力分数时缺乏对节点结构特征的关注.为此提出一种基于相似网络和联合注意力的图嵌入模型.首先计算网络中的节点相似性,并将高相似度且未连接的节点对构建新边以形成相似网络.其次,引入结构相关性和内容相关性的概念,分别用于表征节点之间的结构关系和内容特征.通过融合两种相关性得分计算得到联合注意力分数.最后使用联合注意力分数对节点特征加权聚合,得到最终的节点嵌入表示.将本文所提算法在Cora、Citeseer和Pubmed 3个数据集上进行节点分类任务,准确率分别达到85.70%、74.30%、84.10%,与原始图注意力网络模型相比分别提高了2.70%、3.94%和2.60%.可见,所提出的算法可以得到更好的节点嵌入表示.

自适应相似图联合优化的多视图聚类

相比于单一视图学习,多视图学习往往可以获得学习对象更全面的信息,因而在无监督学习领域,多视图聚类受到了研究者的极大关注,其中基于图的多视图聚类,近年来取得了很大的研究进展.基于图的多视图聚类一般是先从各个视图原始数据学习相似图,再进行视图间相似图的融合来获得最终聚类结果,因此,多视图聚类的效果是由相似图质量和相似图融合方法共同决定的.然而,现有基于图的多视图聚类方法几乎都聚焦在视图间相似图的融合方法研究上,而缺乏对相似图本身质量的关注.这些方法大多数都是孤立地从各视图的原始数据中学习相似图,并且在后续图融合过程中保持相似图不变.这样得到的相似图不可避免地包含噪声和冗余信息,进而影响后续的图融合和聚类.而少量考虑相似图质量的研究,要么相似图构造和图融合过程是直接联立迭代的,要么在预定义相似图过程中提前利用秩约束进一步初始化,要么就是利用相似图存在的一些底层结构来获取融合图的.这些方法对相似图本身改进很小,最终聚类性能提升也十分有限.同时现有基于图的多视图聚类流程也缺乏对各视图间一致性和不一致性的综合考虑,这也会严重影响最终的多视图聚类性能.为了避免低质量预定义相似图对聚类结果的不利影响以及综合考虑视图间一致性与不一致性来提升最终聚类效果,本文提出了一种自适应相似图联合优化的多视图聚类方法.首先通过Hadamard积来获得视图间高质量一致性部分信息,再将每个预定义相似图和这部分信息对标,重构各个视图的预设相似图.这个过程强化了各视图间的一致性部分,弱化了不一致性部分.其次设计了相似图重构改进和图融合联合迭代优化框架,实现了相似图的自适应改进,最终达到相似图和聚类结果共同提升的效果.该方法将相似图改进过程与图融合过程联合起来进行自适应迭代优化,并且在迭代优化中不断强化各视图间的一致性,弱化视图间的不一致性.此外,本文的方法也集成了现有多视图聚类方法的一些优点,自加权以及无需额外聚类步骤等.在九个基准数据集上与八个对比方法的实验验证了本文方法的有效性与优越性.

基于图结构增强的图神经网络方法

针对图卷积网络(GCNs)在面对低同质性的图结构时性能骤降问题,提出了一种新颖的基于图结构增强的图神经网络方法,用于学习改善的图节点表示。首先将节点信息通过消息传播和聚合,得到节点的初始表示;然后计算节点表示的相似性度量,得到图的同质结构;最后融合图的原始结构和同质结构进行节点的信息传递得到节点表示用于下游任务。结果表明:在6个公开的数据集上,所提算法在节点分类的多个指标上均优于对比算法,特别是在同质性较低的4个数据集上,所提算法的准确度(ACC)分数分别超过最高基准5.53%、6.87%、3.08%、4.00%,宏平均(F1)值分别超过最高基准5.75%、8.06%、6.46%、5.61%,获得了远高于基准的优越表现,表明所提方法成功改善了图数据的结构,验证了该算法对图结构优化的有效性。

基于知识图谱的番茄种植管理可视化查询

为提高获取番茄种植管理知识的速度与准确率,研究了以图形式描述番茄在不同环境的种植管理,并基于知识图谱构建了番茄种植管理可视化查询系统。该方法利用“自顶向下”和“自底向上”的模块化CREATE解决了Neo4j的缓慢和准确率问题,并利用PyQt框架构建可视化查询界面,通过问题预处理和语义相似度计算输出最合适的番茄种植管理知识。试验结果表明:该方法的平均响应时间和平均准确率比Cypher查询语言分别提高88.33%及1.97%,可操性比Cypher语言友好。研究结果可以在不同环境下为番茄生产管理提供高质量的种植管理建议。

基于PathSim的MOOCs知识概念推荐模型

大规模开放在线课程提供大规模开放式在线学习平台,为推进现代教育发挥关键作用。然而,减少用户学习盲区和改善用户体验方面的研究仍具有挑战性:交互数据稀疏;难以扩展到大型推荐任务上;用户需求不单由用户喜好决定,还受到不同教师、课程影响;以统一的方式对课程学习事件中不同类型实体及关系进行建模并不妥靠。基于此,引入相关性度量,依据全图结构信息计算各边权重,提出采用相关性度量算法PathSim进行邻域采样的知识概念推荐模型PathSimSage。各实体间相关性得分可在本地离线计算,将神经网络与传播过程分离,保证神经网络的堆叠层数和传播过程的独立性,大幅减少模型所需训练时间。在公开的MoocCube数据集上进行了综合实验,PathSimSage降低了不相关的信息甚至噪声的影响,解决随机游走采样所引发的高度节点偏差问题,并在一定程度上缓解了过平滑效应。

基于知识图谱的专利侵权风险预警研究

同一领域产品的专利技术具有高技术关联度等特征,企业在生产经营活动中面临着专利侵权的潜在风险,立足于企业专利侵权预警的实际需求,高效、准确地检测产品存在的专利侵权风险具有重要意义。由此,本文提出了专利侵权风险预警模型,该模型重新定义了领域专利知识图谱、产品技术方案图谱的模式层,涵盖了组件实体、结构实体和功效实体三类实体类型,以及组成关系、相对位置关系、连接关系和功效达成关系四类实体关系;基于BERT(bidirectional encoder representations from transformers)和BiLSTM(bi-directional long short-term memory)模型构建专利知识图谱和产品技术方案知识图谱;基于ComplEx模型实现知识图谱的嵌入,实现产品和专利技术之间相似度的量化计算,并根据专利侵权风险指数做出侵权预警。以空气加湿器和耳机两类产品进行实证研究,专利侵权预警准确率为86.67%,具有一定的应用价值。

基于动态异构网络的股价预测

股票预测通常被形式化为非线性的时间序列预测任务,但很少有研究者试图通过技术面数据去系统地揭示股票市场内在结构,例如股票上涨或下跌背后的原因可能是业务领域之间的合作或冲突,这些额外信息的增加有助于判断股票的未来趋势。为了充分真实刻画股票市场的交易状态,表达股票之间显式或隐式的关系,提出一种基于动态异构网络的股价预测模型sDHN(stock dynamic heterogeneous network),综合股票以及所属行业和地域,将其建模为动态异构网络。该模型在网络上引入动态时序特征,创新融合股票节点的四种不同技术层面的相似性图,生成富信息异构图,最后聚合不同元路径中隐含的语义信息生成嵌入,从异构图的角度充分探索股票之间的潜在关联。此外,在三个真实世界的股票数据集上进行了大量实验,所提出的模型准确率比所有基线模型均高出5%~34%,F_(1)-score则高出11.5%~37%,并且在图解释上证明了该方法的有效性。

融合知识推理与相似度检索的民众诉求大模型构建与应用

高效回复民众诉求是实现智能化管理、提升民众满意度的必要措施,将智能问答应用于民众诉求能有效节约人力和时间资源。然而,智能问答中基于规则和检索的模型依赖预设知识,当诉求超出预设知识范围时无法提供有效回复,在处理多轮对话时也无法保持对话连贯性。现有的大语言模型可以和用户流畅对话,但通用大语言模型缺乏诉求领域知识。由于训练数据中问答对的信息没有覆盖回答用户问题所需要的知识,导致通用大语言模型生成错误回复或答非所问,产生幻觉。针对上述问题,构建了面向民众诉求领域的智能问答大语言模型(PC-LLM)。设计基于BERT-BiLSTM-CRF的实体关系抽取模型获得诉求工单中实体及其关系,进而构建诉求知识图谱,使用BERT模型对诉求工单向量化并构建诉求工单向量索引库;回复生成阶段,抽取用户诉求的实体和关系,在诉求知识图谱中通过实体链接进行知识推理,获取潜在关系提示,同时在诉求工单向量索引库内对诉求进行快速检索,获取相似诉求并构建相似诉求提示;将潜在关系提示、相似诉求提示与用户诉求融合形成综合提示,引导大语言模型生成准确的回复。实验分析显示,该大语言模型在诉求数据集中的表现明显优于ChatGPT4o、文心一言、通义千问等大语言模型。

基于图卷积的无监督跨模态哈希检索算法

为解决当前无监督跨模态哈希检索在全局相似性矩阵构建和异构数据语义信息融合中存在的困难,提出一种基于图卷积的无监督跨模态哈希检索算法(GCUH)。采用分层次聚合的方式,将各个模态的相似性结构编码到全局相似性矩阵中,获得跨模态的成对相似性信息来指导学习。使用图卷积模块融合跨模态信息,消除邻居结构中的噪声干扰,形成完备的跨模态表征,提出两种相似性保持的损失函数约束哈希码的一致性。与基线模型相比,GCUH在NUS-WIDE数据集上使用64位哈希码执行文本检索图片任务的检索精度提升了6.3%。