WO2020113437A1 - Graph structure processing method and system, network device and storage medium - Google Patents

Graph structure processing method and system, network device and storage medium Download PDF

Info

Publication number
WO2020113437A1
WO2020113437A1 PCT/CN2018/119227 CN2018119227W WO2020113437A1 WO 2020113437 A1 WO2020113437 A1 WO 2020113437A1 CN 2018119227 W CN2018119227 W CN 2018119227W WO 2020113437 A1 WO2020113437 A1 WO 2020113437A1
Authority
WO
WIPO (PCT)
Prior art keywords
sub
graph structure
subgraph
network device
graph
Prior art date
Application number
PCT/CN2018/119227
Other languages
French (fr)
Chinese (zh)
Inventor
袁振南
朱鹏新
Original Assignee
区链通网络有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 区链通网络有限公司 filed Critical 区链通网络有限公司
Priority to CN201880002415.9A priority Critical patent/CN109952742B/en
Priority to PCT/CN2018/119227 priority patent/WO2020113437A1/en
Publication of WO2020113437A1 publication Critical patent/WO2020113437A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/02—Knowledge representation; Symbolic representation
    • G06N5/022—Knowledge engineering; Knowledge acquisition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models

Definitions

  • the present application relates to the technical field of graph structures, and in particular, to a graph structure processing method, a graph structure processing system, a network device, and a computer-readable storage medium.
  • the partitioning method Before solving various problems with graph-structured data, such as anomaly detection, clustering, and label propagation, it is usually necessary to process graph-structured data, such as dividing the entire graph structure into smaller subgraph structures. Due to the existing graph structure processing, for example, the partitioning method usually needs to store the entire graph structure in one network device, which causes great difficulty in partitioning the graph structure, especially the large-scale partitioning of the graph structure.
  • the present application provides a processing method for a graph structure.
  • the processing method is separately executed by a plurality of network devices in a network.
  • the processing method includes: a network device acquiring a sub-graph structure, wherein the sub-graph structure belongs to the A part of the graph structure; acquiring optimized feature parameters of the sub-graph structure, wherein the optimized feature parameters are used to determine the size of the sub-graph structure in the graph structure; according to the optimized feature parameters Adjust the graph structure.
  • the present application also provides a graph-structured processing system.
  • the processing system includes multiple network devices, and the multiple network devices form a network, wherein: each network device is used to perform the following steps: obtain a sub-graph structure, wherein, The sub-graph structure belongs to a part of the graph structure; obtaining optimized feature parameters of the sub-graph structure, wherein the optimized feature parameters are used to determine the size of the sub-graph structure in the graph structure; according to the Optimize the feature parameters to adjust the sub-graph structure.
  • the present application also provides a network device.
  • the network device includes a processor and a memory.
  • the memory stores a computer program.
  • the computer program is executed by the processor to implement the following processing method: obtain a subgraph structure, where , The sub-graph structure belongs to a part of the graph structure; acquiring optimized feature parameters of the sub-graph structure, wherein the optimized feature parameters are used to determine the size of the sub-graph structure in the graph structure;
  • the optimization feature parameter adjusts the sub-graph structure.
  • the present application also provides a network device.
  • the network device includes: a first acquisition module: used to acquire a sub-graph structure, wherein the sub-graph structure belongs to a part of the graph structure; a second acquisition module is used to acquire the The optimized feature parameters of the sub-graph structure, wherein the optimized feature parameters are used to determine the size of the sub-graph structure in the graph structure; the processing module is used to perform the sub-graph structure according to the optimized feature parameters Adjustment.
  • the present application provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above division method are implemented.
  • Each network device of the present application separately obtains the partial sub-picture structure in the graph structure, and then processes the sub-graph structure, so that the processing work of the entire graph structure can be distributed to multiple network devices, each network device only needs It is sufficient to process the partial sub-graph structure in the entire graph structure, and the processing complexity is low.
  • FIG. 1 is a schematic structural diagram of a graph structure dividing system provided by an embodiment of the present application
  • FIG. 2 is a schematic flowchart of a method for dividing a graph structure according to an embodiment of the present application
  • FIG. 3 is a schematic flowchart of step 101 in the division method shown in FIG. 2;
  • FIG. 4 is a partial structural schematic diagram of the graph structure
  • FIG. 5 is a schematic diagram of a hardware structure of a network device provided by an embodiment of the present application.
  • FIG. 6 is a schematic diagram of a software structure of another network device provided by an embodiment of the present application.
  • the data can be characterized by a graph structure.
  • the graph structure includes vertices (or nodes) and edges, which are composed of a finite non-empty set of vertices and a set of edges between vertices, usually expressed as: G(V,E) , Where G represents a graph, V is the set of vertices in graph G, and E is the set of edges in graph G. Based on this, it can be converted to analyze the graph structure to detect various problems, such as anomaly detection, clustering, and label propagation. Before analyzing the graph structure, the graph structure usually needs to be processed, for example, divided into smaller graph structures. The following embodiments of the present application will provide a processing system and processing method for a graph structure to implement processing of the graph structure.
  • FIG. 1 is a schematic structural diagram of a processing system with a graph structure provided by an embodiment of the present application.
  • the processing system 10 of this embodiment includes multiple network devices 12.
  • a plurality of network devices 12 form a network system, that is to say, the processing system 10 of this embodiment can be regarded as a network system.
  • the network device 12 may be a processing node in the network system.
  • the network device 12 includes a server, a computer, and other electronic devices with computing and storage capabilities. It should be noted that the server may include a physical server and a virtual machine running in the physical server.
  • the network device 12 and the network device 12 can be freely connected to each other.
  • the influence between the network device 12 and the network device 12 will form a non-linear causal relationship through the network.
  • each network device 12 can maintain its own graph structure.
  • the graph structure can be processed by each network device 12, that is, the processing task of the graph structure is distributed to each network device 12.
  • this embodiment greatly reduces the complexity of the processing by the scheme for processing the graph structure by each network device 12 in the network system.
  • each network device 12 may be the same. For details, refer to the following embodiments.
  • FIG. 2 is a schematic flowchart of a graph structure processing method according to an embodiment of the present application. This processing method is used to be separately executed by multiple network devices 12 to jointly complete the sub-graph division of the graph structure. As shown in FIG. 2, the processing method of this embodiment includes:
  • Step 101 The network device obtains the sub-graph structure. Among them, the sub-graph structure is part of the graph structure.
  • each network device 12 When determining the processing system 10, that is, determining each network device 12, each network device 12 stores the sub-picture structure in the original state.
  • the sub-picture structure in the original state may be formed according to preset rules, for example, it may be formed according to data input by a user, or based on data sent by other communication devices, or the network device 12 itself may perform data such as historical processing records. Formed by analysis.
  • the sub-picture structure acquired by each network device 12 in step 101 is not the above-mentioned sub-picture structure in the original state, which can be understood as a sub-picture structure after reprocessing the sub-picture structure in the original state.
  • the following will describe in detail how each network device obtains the reprocessed sub-picture structure.
  • the following sub-picture structures unless otherwise specified, are all re-processed sub-picture structures.
  • the special limit can be understood as including adding the first, second, or certain similar qualifiers before the sub-picture structure.
  • the subgraph structure in the original state After the subgraph structure in the original state is obtained, it can be further processed to obtain a more suitable subgraph structure, for example, a subgraph structure with more similar characteristics is found and reintegrated together.
  • each network device 12 may obtain the sub-graph structure according to the conductivity.
  • the conductivity can represent the correlation between elements in the graph structure, for example, between nodes, between edges, and so on.
  • each network device 12 may adjust the reference subgraph according to the conductivity of the subgraph structure composed of neighboring nodes of each node in the reference subgraph structure to obtain the subgraph structure of the network device 12.
  • the reference sub-picture structure may be understood as the sub-picture structure currently being processed.
  • the network device 12 stores two sub-graph structures A and B in the original state.
  • the sub-graph structure A can be used as a reference sub-graph structure to calculate its respective The conductivity of the subgraph structure formed by the neighboring nodes of the node adjusts the reference subgraph.
  • Step 201 For each node in the reference subgraph structure, calculate the conductivity of the subgraph structure composed of adjacent nodes of the node. Specifically, the conductivity can be calculated according to the following formula:
  • ⁇ is the conductivity
  • S is the subgraph structure composed of adjacent nodes
  • E(S,VS) is the complement between the subgraph structure S composed of adjacent nodes and the complement set of the subgraph structure S composed of adjacent nodes
  • A is the degree matrix of the subgraph structure S composed of adjacent nodes
  • a VS is the degree matrix of the complementary set of the subgraph structure S composed of adjacent nodes.
  • this step can sequentially calculate the conductivity of the subgraph structure composed of different adjacent nodes. The following steps are performed once after each conductivity calculation.
  • Step 202 Determine whether the conductivity meets a preset conductivity threshold.
  • the conductivity threshold can be determined by a preset subgraph structure processing tree. Specifically, before processing the graph structure, a sub-graph structure processing tree may be preset according to requirements, and then a corresponding conductivity threshold may be formed according to the sub-graph structure processing tree.
  • Step 202 can indicate that the correlation between the subgraph structure of the adjacent node and the subgraph structure of the current node is relatively high when the judgment result is yes, and then jump to step 203; when the judgment structure is negative, it can indicate the adjacent node The correlation between the subgraph structure of and the subgraph structure of the current node is relatively low. At this time, jump to step 204.
  • Step 203 Expand the subgraph structure composed of adjacent nodes to the reference subgraph structure.
  • One way to expand may be to merge the subgraph structures of adjacent nodes into the subgraph structure of the current node to form a new subgraph structure. Therefore, if the subgraph structure of the neighbor node has a high correlation with the subgraph structure of the current node, the two subgraph structures are tried to be merged and updated to a new subgraph structure.
  • Step 204 Stop expanding its own reference sub-picture structure.
  • the expansion at this time is zero, that is, the subgraph structure after expansion is the subgraph structure before expansion.
  • the subgraph structure of the current node can be expanded in turn using the subgraph structures of different adjacent nodes until the preset expansion conditions are met and the final expanded subgraph structure is used as the subgraph after reprocessing by the network device 12 structure.
  • node A is used as the current node for explanation.
  • the sub-graph structure composed of node A is expanded in turn using the sub-graph structure composed of its neighboring nodes B-F according to the expansion conditions of the conductivity described above.
  • recalculation and expansion may be omitted.
  • a preset threshold of the number of expansion times and/or a threshold of the number of update times may be further set. It should be noted that the update represents the case where the expansion is not zero, that is, there is a case where the subgraph structure composed of adjacent nodes is merged into the subgraph structure composed of the current node.
  • the preset threshold is the number of update times
  • the update frequency threshold may be defined as a ratio of the update times among the preset calculation times. For example, if the threshold for the number of updates is set to 20%, and in step 201 of the conductivity calculation, if the number of expansions that can be expanded is only 5 of the 200 calculations, then the filtering of the sub-graph structure has been considered to be basically Meet the requirements.
  • step 201 If the determined structure is no, it may return to step 201.
  • the preset is the expansion times threshold, it is determined whether it is greater than the preset expansion times threshold.
  • step 204 If the result of the judgment is yes, it may return to step 204, that is, when the threshold of the number of expansion times is large, it may be deemed that the filtering of the sub-picture structure has basically met the requirements.
  • the threshold of the number of expansion times can be defined as the number of times the conductivity is calculated in step 201, because each time the conductivity calculation is expanded once. If the determined structure is no, it may return to step 201.
  • each network device obtains the expanded sub-graph structure through the conductivity. It should be understood that in other embodiments, each network device may also obtain the sub-graph structure through other correlation parameters of the graph structure.
  • the sub-graph structure acquired by each network device in step 101 is only a rough division process of the graph structure in the processing system 10, and the sub-graph structure acquired in step 101 is not yet suitable for analyzing the graph structure, so It is necessary to further perform detailed processing on the sub-picture structure obtained in step 101 through steps 102 and 103, that is, each network device adjusts the size of the sub-picture structure obtained in step 101 in the graph structure to achieve the final
  • the division relationship between the sub-graph structures of network equipment is a relatively accurate and reasonable division of the graph structure.
  • Step 102 Obtain the optimized feature parameters of the subgraph structure, where the optimized feature parameters are used to determine the size of the subgraph structure in the graph structure.
  • each network device 12 may convert and store the acquired feature information of the subgraph structure, for example, feature information such as edges and nodes in the subgraph structure through a matrix or a linked list. In order to get the characteristics.
  • the result of step 101 is a sub-graph structure after preliminary processing, and its features are extracted based on the sub-graph structure after initial processing, which may not be the objective optimal feature, so it needs to be optimized After operation, the optimized characteristic parameters are obtained.
  • the characteristic parameters of the subgraph structure include, for example, edges and nodes in the subgraph structure, which are used to determine the size of the subgraph structure in the graph structure.
  • the characteristic parameters of the subgraph structure include nodes A, B, C and their edge sets, and the range of the node structure A, B, C and their edge sets in the graph structure can be defined as the subgraph structure. range.
  • each network device can obtain the optimized feature parameters of the sub-graph structure stored by itself by combining the feature parameters of the sub-graph structure of the neighbor network device and the feature parameters of the sub-graph structure stored by itself.
  • Step 103 will adjust the sub-graph structure according to the optimized feature parameters.
  • each network device 12 may use the feature parameters of the subgraph structure of its neighboring network device to iterate over the feature parameters of its own subgraph structure to obtain the optimized feature parameters of its own subgraph structure. That is to say, by using the characteristic parameters of the sub-graph structure of the neighboring network device to continuously update the characteristic parameters of the sub-graph structure of itself, it gradually approaches the final optimized characteristic parameter during the iteration process.
  • K is the number of previous iterations
  • i is the serial number of the network device
  • j is the serial number of other network devices other than network device i
  • j in the above formula can vary with the calculation of the network device
  • X i is the characteristic parameter of the subgraph structure of the network device i
  • x j is the characteristic parameter of the subgraph structure of the other network device j.
  • x j is also variable, and w ij is relative to the network of other network devices j
  • the weight of device i, when i and j are non-adjacent network devices, w ij 0, therefore, the effective value of ⁇ j w ij x j (k) in the above formula is only the value of the adjacent network device, ⁇ is Iterative gradient stride, Is the gradient of the objective function.
  • the value of the weight w ij of each adjacent network device may be the same. That is, the average optimal characteristic parameters of the adjacent network devices are used to iterate the objective function gradient of the subgraph structure.
  • each network device When K tends to ⁇ , the calculated optimized characteristic parameters of each network device converge to the overall optimal characteristic parameters Field, where ⁇ is the largest second-order eigenvalue of the weight w. Therefore, when the number of iterations is sufficient, each network device obtains an optimized feature parameter close to the overall optimal feature parameter.
  • step 102 is to use the gradient descent method to optimize the feature parameters to form optimized feature parameters.
  • step 102 may also use optimization algorithms such as Newton's method, conjugate gradient method, Levenberg-Marquardt Algorithm (Levenberg-Marquardt method) to optimize the characteristic parameters to form optimized characteristic parameters.
  • optimization algorithms such as Newton's method, conjugate gradient method, Levenberg-Marquardt Algorithm (Levenberg-Marquardt method) to optimize the characteristic parameters to form optimized characteristic parameters.
  • Step 103 Adjust the subgraph structure according to the optimized feature parameters.
  • step 103 is specifically that each network device processes the sub-graph structure according to the final optimal processing parameter formed after optimization. Specifically, each network device can reasonably determine the size of its sub-graph structure through the above steps, that is, by performing the above steps through each network device, the graph structure can be accurately divided.
  • the above processing methods can all be implemented by a network device. Specifically, a computer program is used to represent the steps of the method, a software structure is constructed to implement the computer program, and a hardware device is used to execute the computer program to implement the method.
  • the computer program when implemented in software and sold or used as an independent product, can be stored in a readable storage medium of an electronic device, that is, the present invention also provides a computer-readable storage medium in which a computer program is stored When the computer program is executed by the processor, the steps of the above method are realized.
  • the computer-readable storage medium may be a U disk, an optical disk, a server, or the like.
  • FIG. 1 can be implemented by a processing system 10.
  • the steps performed by each network device 12 in the processing system 10 are the same, as shown in the graph structure processing steps described above.
  • FIG. 5 is a schematic diagram of a hardware structure of a network device 12 according to an embodiment of the present application.
  • the network device 12 of this embodiment includes a processor 121 and a memory 122.
  • the memory 122 stores a computer program, and the computer program is executed by the processor 121 to implement the steps of the following processing method.
  • the processor 121 in the network device can acquire the sub-graph structure, where the sub-graph structure belongs to a part of the graph structure, and obtain the optimized feature parameters of the sub-graph structure, and finally perform the sub-graph structure according to the optimized feature parameters Adjustment.
  • the network device 12 further includes a transceiver 123 for communicating with neighboring network devices and receiving data such as characteristic parameters of the neighboring network devices.
  • the transceiver 123 and the network device 12 of this embodiment cooperate with all the steps of the processing method described above.
  • FIG. 6 is a schematic diagram of a software structure of another network device provided by an embodiment of the present application.
  • the network device 60 of this embodiment includes:
  • the first acquiring module 601 is used to acquire a sub-graph structure, where the sub-graph structure belongs to a part of the graph structure.
  • the second obtaining module 602 is configured to obtain optimized feature parameters of the sub-graph structure.
  • the optimized feature parameters are used to determine the size of the sub-graph structure in the graph structure;
  • the processing module 603 is used to adjust the sub-graph structure according to the optimized feature parameters.

Abstract

Provided in the present application are a graph structure processing method and system, a network device and a computer-readable storage medium. The graph structure processing method is configured to be performed respectively by a plurality of network devices in a network and comprises the following steps: acquiring a sub-graph structure, wherein the sub-graph structure is part of the graph structure; acquiring an optimizing feature parameter of the sub-graph structure, wherein the optimizing feature parameter is configured to determine the size of the sub-graph structure in the graph structure; and adjusting the sub-graph structure according to the optimizing feature parameter. Accordingly, the present application can reduce the processing complexity of a network device.

Description

图结构处理方法、***、网络设备及存储介质Graph structure processing method, system, network equipment and storage medium 【技术领域】【Technical Field】
本申请涉及图结构技术领域，特别是涉及一种图结构处理方法、图结构处理***、网络设备以及计算机可读存储介质。The present application relates to the technical field of graph structures, and in particular, to a graph structure processing method, a graph structure processing system, a network device, and a computer-readable storage medium.
【背景技术】【Background technique】
在解决具有图结构数据的各种问题，例如异常检测、聚类、标签传播前通常需要对图结构数据进行处理，例如对整图结构进行划分为更小的子图结构等。由于现有的图结构处理，例如划分方法通常需要在一个网络设备中存储整个图结构，对图结构的划分特别是大规模的图结构的划分实现造成了很大的困难。Before solving various problems with graph-structured data, such as anomaly detection, clustering, and label propagation, it is usually necessary to process graph-structured data, such as dividing the entire graph structure into smaller subgraph structures. Due to the existing graph structure processing, for example, the partitioning method usually needs to store the entire graph structure in one network device, which causes great difficulty in partitioning the graph structure, especially the large-scale partitioning of the graph structure.
【发明内容】[Invention content]
本申请提供一种图结构的处理方法，所述处理方法用于被网络中的多个网络设备分别执行，该处理方法包括：网络设备获取子图结构，其中，所述子图结构属于所述图结构的一部分；获取所述子图结构的优化特征参数，其中，所述优化特征参数用于确定所述子图结构在所述图结构中的大小；根据所述优化特征参数对所述子图结构进行调整。The present application provides a processing method for a graph structure. The processing method is separately executed by a plurality of network devices in a network. The processing method includes: a network device acquiring a sub-graph structure, wherein the sub-graph structure belongs to the A part of the graph structure; acquiring optimized feature parameters of the sub-graph structure, wherein the optimized feature parameters are used to determine the size of the sub-graph structure in the graph structure; according to the optimized feature parameters Adjust the graph structure.
本申请还提供一种图结构的处理***，处理***包括多个网络设备，多个所述网络设备形成一网络，其中：每个网络设备用于执行以下步骤：获取子图结构，其中，所述子图结构属于所述图结构的一部分；获取所述子图结构的优化特征参数，其中，所述优化特征参数用于确定所述子图结构在所述图结构中的大小；根据所述优化特征参数对所述子图结构进行调整。The present application also provides a graph-structured processing system. The processing system includes multiple network devices, and the multiple network devices form a network, wherein: each network device is used to perform the following steps: obtain a sub-graph structure, wherein, The sub-graph structure belongs to a part of the graph structure; obtaining optimized feature parameters of the sub-graph structure, wherein the optimized feature parameters are used to determine the size of the sub-graph structure in the graph structure; according to the Optimize the feature parameters to adjust the sub-graph structure.
本申请还提供一种网络设备，网络设备包括处理器和存储器，所述存储器存储有计算机程序，所述计算机程序被所述处理器执行以实现以下的所述处理方法：获取子图结构，其中，所述子图结构属于所述图结构的一 部分；获取所述子图结构的优化特征参数，其中，所述优化特征参数用于确定所述子图结构在所述图结构中的大小；根据所述优化特征参数对所述子图结构进行调整。The present application also provides a network device. The network device includes a processor and a memory. The memory stores a computer program. The computer program is executed by the processor to implement the following processing method: obtain a subgraph structure, where , The sub-graph structure belongs to a part of the graph structure; acquiring optimized feature parameters of the sub-graph structure, wherein the optimized feature parameters are used to determine the size of the sub-graph structure in the graph structure; The optimization feature parameter adjusts the sub-graph structure.
本申请还提供一种网络设备，网络设备包括：第一获取模块：用于获取子图结构，其中，所述子图结构属于所述图结构的一部分；第二获取模块，用于获取所述子图结构的优化特征参数，其中，所述优化特征参数用于确定所述子图结构在所述图结构中的大小；处理模块，用于根据所述优化特征参数对所述子图结构进行调整。The present application also provides a network device. The network device includes: a first acquisition module: used to acquire a sub-graph structure, wherein the sub-graph structure belongs to a part of the graph structure; a second acquisition module is used to acquire the The optimized feature parameters of the sub-graph structure, wherein the optimized feature parameters are used to determine the size of the sub-graph structure in the graph structure; the processing module is used to perform the sub-graph structure according to the optimized feature parameters Adjustment.
本申请提供一种计算机可读存储介质，其上存储有计算机程序，计算机程序被处理器执行时实现上述划分方法的步骤。The present application provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above division method are implemented.
本申请的每个网络设备分别获取图结构中的局部子图结构，继而对子图结构进行处理，由此可将整个图结构的处理工作分配到多个网络设备当中，每个网络设备仅需对整个图结构当中的局部子图结构进行处理即可，处理复杂度较低。Each network device of the present application separately obtains the partial sub-picture structure in the graph structure, and then processes the sub-graph structure, so that the processing work of the entire graph structure can be distributed to multiple network devices, each network device only needs It is sufficient to process the partial sub-graph structure in the entire graph structure, and the processing complexity is low.
【附图说明】【Explanation】
为了更清楚地说明本申请实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。In order to more clearly explain the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without paying any creative work, other drawings can be obtained based on these drawings.
图1是本申请实施例提供的一种图结构的划分***的结构示意图；FIG. 1 is a schematic structural diagram of a graph structure dividing system provided by an embodiment of the present application;
图2是本申请实施例提供的一种图结构划分方法的流程示意图；2 is a schematic flowchart of a method for dividing a graph structure according to an embodiment of the present application;
图3是图2所示的划分方法中步骤101的流程示意图；3 is a schematic flowchart of step 101 in the division method shown in FIG. 2;
图4是图结构的局部结构示意图；FIG. 4 is a partial structural schematic diagram of the graph structure;
图5是本申请实施例提供的一种网络设备的硬件结构示意图；5 is a schematic diagram of a hardware structure of a network device provided by an embodiment of the present application;
图6是本本申请实施例提供的另一网络设备的软件结构示意图。6 is a schematic diagram of a software structure of another network device provided by an embodiment of the present application.
【具体实施方式】【detailed description】
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进 行清楚、完整地描述。可以理解的是，此处所描述的具体实施例仅用于解释本申请，而非对本申请的限定。另外还需要说明的是，为了便于描述，附图中仅示出了与本申请相关的部分而非全部结构。基于本申请中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. It can be understood that the specific embodiments described here are only used for explaining the present application, rather than limiting the present application. In addition, it should be noted that, for ease of description, the drawings only show parts, but not all structures related to the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.
在实际应用中，数据分析可帮助人们作出判断，以便采取适当行动。而数据可通过图结构表征，图结构包括顶点(或称为节点)和边，其是由顶点的有穷非空集合和顶点之间边的集合组成，通常表示为：G(V,E)，其中，G表示一个图，V是图G中顶点的集合，E是图G中边的集合。基于此，可转换为对图结构进行分析，以检测各种问题，例如，异常检测、聚类和标签传播等。在对图结构进行分析前，通常需要对图结构进行处理，例如划分为更小的图结构。本申请以下实施例将提供一种图结构的处理***和处理方法来实现对图结构的处理。In practical applications, data analysis can help people make judgments in order to take appropriate actions. The data can be characterized by a graph structure. The graph structure includes vertices (or nodes) and edges, which are composed of a finite non-empty set of vertices and a set of edges between vertices, usually expressed as: G(V,E) , Where G represents a graph, V is the set of vertices in graph G, and E is the set of edges in graph G. Based on this, it can be converted to analyze the graph structure to detect various problems, such as anomaly detection, clustering, and label propagation. Before analyzing the graph structure, the graph structure usually needs to be processed, for example, divided into smaller graph structures. The following embodiments of the present application will provide a processing system and processing method for a graph structure to implement processing of the graph structure.
首先请参阅图1，图1是本申请实施例提供的一种图结构的处理***的结构示意图。如图1所示，本实施例的处理***10包括多个网络设备12。多个网络设备12组成一个网络***，也就是说本实施例的处理***10可看作为一个网络***。其中，网络设备12可为该网络***中的一个处理节点，网络设备12包括服务器，计算机等具备计算能力和存储能力的电子设备。需要说明的是，服务器可包括物理服务器以及运行在物理服务器中的虚拟机。Please refer to FIG. 1 first. FIG. 1 is a schematic structural diagram of a processing system with a graph structure provided by an embodiment of the present application. As shown in FIG. 1, the processing system 10 of this embodiment includes multiple network devices 12. A plurality of network devices 12 form a network system, that is to say, the processing system 10 of this embodiment can be regarded as a network system. The network device 12 may be a processing node in the network system. The network device 12 includes a server, a computer, and other electronic devices with computing and storage capabilities. It should be noted that the server may include a physical server and a virtual machine running in the physical server.
在网络***中，网络设备12与网络设备12之间彼此可以自由连接。网络设备12与网络设备12之间的影响，会通过网络而形成非线性因果关系。In the network system, the network device 12 and the network device 12 can be freely connected to each other. The influence between the network device 12 and the network device 12 will form a non-linear causal relationship through the network.
在这个网络***中，每个网络设备12均可维护自身的图结构。In this network system, each network device 12 can maintain its own graph structure.
本实施例可通过各网络设备12对图结构进行处理，即将图结构的处理任务分配到各个网络设备12中。相对于现有技术的通过一个服务器对整个图结构进行处理的方案，本实施例通过网络***中的各网络设备12对图结构处理的方案大大降低了处理的复杂度。In this embodiment, the graph structure can be processed by each network device 12, that is, the processing task of the graph structure is distributed to each network device 12. Compared with the prior art scheme for processing the entire graph structure by one server, this embodiment greatly reduces the complexity of the processing by the scheme for processing the graph structure by each network device 12 in the network system.
各网络设备12的处理方法均可相同，具体可参阅以下实施例。The processing method of each network device 12 may be the same. For details, refer to the following embodiments.
请一并参阅图2，图2是本申请实施例提供的一种图结构处理方法的流 程示意图。该处理方法用于被多个网络设备12分别执行，以共同完成对图结构的子图划分。如图2所示，本实施例的处理方法包括：Please refer to FIG. 2 together. FIG. 2 is a schematic flowchart of a graph structure processing method according to an embodiment of the present application. This processing method is used to be separately executed by multiple network devices 12 to jointly complete the sub-graph division of the graph structure. As shown in FIG. 2, the processing method of this embodiment includes:
步骤101：网络设备获取子图结构。其中，子图结构属于图结构的一部分。Step 101: The network device obtains the sub-graph structure. Among them, the sub-graph structure is part of the graph structure.
在确定处理***10，即确定各网络设备12时，各网络设备12中均存储有原始状态下的子图结构。该原始状态下的子图结构可以依照预设规则形成的，例如可根据用户输入的数据来形成，或者根据其他通信设备发送的数据来形成，又或者网络设备12自身对历史处理记录等数据进行分析而形成。When determining the processing system 10, that is, determining each network device 12, each network device 12 stores the sub-picture structure in the original state. The sub-picture structure in the original state may be formed according to preset rules, for example, it may be formed according to data input by a user, or based on data sent by other communication devices, or the network device 12 itself may perform data such as historical processing records. Formed by analysis.
步骤101中的各网络设备12分别获取的子图结构并非是上述的原始状态下的子图结构，其可理解为对原始状态下的子图结构进行重新处理后的子图结构。以下将详述各网络设备如何获取重新处理后的子图结构，为了简化描述，以下的子图结构，若没有特别限定，均为重新处理后的子图结构。其中，特别限定可理解为包括在子图结构前加上第一、第二或某某的等类似限定词。The sub-picture structure acquired by each network device 12 in step 101 is not the above-mentioned sub-picture structure in the original state, which can be understood as a sub-picture structure after reprocessing the sub-picture structure in the original state. The following will describe in detail how each network device obtains the reprocessed sub-picture structure. To simplify the description, the following sub-picture structures, unless otherwise specified, are all re-processed sub-picture structures. Among them, the special limit can be understood as including adding the first, second, or certain similar qualifiers before the sub-picture structure.
在得到原始状态下的子图结构后，可以进一步对其进行处理以获得更为合适的子图结构，例如找寻到特征更为相似的子图结构，将其重新整合在一起。After the subgraph structure in the original state is obtained, it can be further processed to obtain a more suitable subgraph structure, for example, a subgraph structure with more similar characteristics is found and reintegrated together.
在一实施例中，各网络设备12可根据导率来获取子图结构。其中，导率可表征图结构中的元素间，例如节点与节点之间、边与边之间等的相关性。In an embodiment, each network device 12 may obtain the sub-graph structure according to the conductivity. Among them, the conductivity can represent the correlation between elements in the graph structure, for example, between nodes, between edges, and so on.
具体而言，各网络设备12可根据该参考子图结构中各节点的相邻节点组成的子图结构的导率，对该参考子图进行调整，以获得网络设备12的子图结构。其中，参考子图结构可理解为当前进行处理的子图结构。在一实际应用中，网络设备12存储了两个原始状态下的子图结构A和B，则在对子图结构A进行处理时，可将子图结构A作为参考子图结构，计算其各节点的相邻节点组成的子图结构的导率，对该参考子图进行调整。Specifically, each network device 12 may adjust the reference subgraph according to the conductivity of the subgraph structure composed of neighboring nodes of each node in the reference subgraph structure to obtain the subgraph structure of the network device 12. The reference sub-picture structure may be understood as the sub-picture structure currently being processed. In an actual application, the network device 12 stores two sub-graph structures A and B in the original state. When processing the sub-graph structure A, the sub-graph structure A can be used as a reference sub-graph structure to calculate its respective The conductivity of the subgraph structure formed by the neighboring nodes of the node adjusts the reference subgraph.
具体过程请一并参阅图3，包括以下子步骤：Please refer to Figure 3 for the specific process, including the following sub-steps:
步骤201：对参考子图结构中的每一节点，计算该节点的相邻节点组成的子图结构的导率。具体可根据以下公式计算导率：Step 201: For each node in the reference subgraph structure, calculate the conductivity of the subgraph structure composed of adjacent nodes of the node. Specifically, the conductivity can be calculated according to the following formula:
Figure PCTCN2018119227-appb-000001
Figure PCTCN2018119227-appb-000001
其中，Φ为导率，S为相邻节点组成的子图结构，E(S,V-S)为相邻节点组成的子图结构S和相邻节点组成的子图结构S的补集之间的相连接边数，A为相邻节点组成的子图结构S的度矩阵，A V-S为相邻节点组成的子图结构S的补集的度矩阵。 Where Φ is the conductivity, S is the subgraph structure composed of adjacent nodes, and E(S,VS) is the complement between the subgraph structure S composed of adjacent nodes and the complement set of the subgraph structure S composed of adjacent nodes The number of connected edges, A is the degree matrix of the subgraph structure S composed of adjacent nodes, and A VS is the degree matrix of the complementary set of the subgraph structure S composed of adjacent nodes.
基于以上计算公式，本步骤可依次计算不同相邻节点组成的子图结构的导率。在每次导率的计算后均执行一次以下步骤。Based on the above calculation formula, this step can sequentially calculate the conductivity of the subgraph structure composed of different adjacent nodes. The following steps are performed once after each conductivity calculation.
步骤202：判断导率是否满足预设的导率阈值。Step 202: Determine whether the conductivity meets a preset conductivity threshold.
导率阈值可由预设的子图结构处理树决定。具体而言，在对图结构进行处理前，可首先根据要求预设一个子图结构处理树，再根据子图结构处理树形成对应的导率阈值。The conductivity threshold can be determined by a preset subgraph structure processing tree. Specifically, before processing the graph structure, a sub-graph structure processing tree may be preset according to requirements, and then a corresponding conductivity threshold may be formed according to the sub-graph structure processing tree.
步骤202在判断的结果为是时可表示相邻节点的子图结构与当前节点的子图结构相关性比较高，此时跳转到步骤203；在判断的结构为否时可表示相邻节点的子图结构与当前节点的子图结构相关性比较低，此时跳转到步骤204。Step 202 can indicate that the correlation between the subgraph structure of the adjacent node and the subgraph structure of the current node is relatively high when the judgment result is yes, and then jump to step 203; when the judgment structure is negative, it can indicate the adjacent node The correlation between the subgraph structure of and the subgraph structure of the current node is relatively low. At this time, jump to step 204.
步骤203：将相邻节点组成的子图结构对参考子图结构进行扩展。Step 203: Expand the subgraph structure composed of adjacent nodes to the reference subgraph structure.
扩展的一种方式可为将相邻节点的子图结构合并到当前节点的子图结构当中，形成一个新的子图结构。由此，若邻居节点的子图结构与当前节点的子图结构相关性较高，则将两个子图结构尝试合并，更新为新的子图结构。One way to expand may be to merge the subgraph structures of adjacent nodes into the subgraph structure of the current node to form a new subgraph structure. Therefore, if the subgraph structure of the neighbor node has a high correlation with the subgraph structure of the current node, the two subgraph structures are tried to be merged and updated to a new subgraph structure.
步骤204：停止对自身的参考子图结构进行扩展。Step 204: Stop expanding its own reference sub-picture structure.
可以理解的是，此时的扩展为零，即扩展后的子图结构为扩展前的子图结构。It can be understood that the expansion at this time is zero, that is, the subgraph structure after expansion is the subgraph structure before expansion.
基于以上思想，可依次利用不同相邻节点的子图结构对当前节点的子图结构进行扩展，直到满足预设的扩展条件后将最终扩展的子图结构作为网络设备12重新处理后的子图结构。如图4所示，以节点A为当前节点进行阐述。相对于节点A，根据前文所述的导率的扩展条件依次利用其相邻节点B—F组成的子图结构对节点A的组成子图结构进行扩展。Based on the above ideas, the subgraph structure of the current node can be expanded in turn using the subgraph structures of different adjacent nodes until the preset expansion conditions are met and the final expanded subgraph structure is used as the subgraph after reprocessing by the network device 12 structure. As shown in FIG. 4, node A is used as the current node for explanation. Relative to node A, the sub-graph structure composed of node A is expanded in turn using the sub-graph structure composed of its neighboring nodes B-F according to the expansion conditions of the conductivity described above.
进一步的，为了节约计算成本，在扩展到一定程度后可以省略再计算、 扩展。具体可进一步设置预设的扩展次数阈值和/或更新次数阈值。应该说明的是更新代表的是扩展不为零的情况，即存在将相邻节点组成的子图结构合并到当前节点组成的子图结构中的情况。Further, in order to save the calculation cost, after calculation to a certain extent, recalculation and expansion may be omitted. Specifically, a preset threshold of the number of expansion times and/or a threshold of the number of update times may be further set. It should be noted that the update represents the case where the expansion is not zero, that is, there is a case where the subgraph structure composed of adjacent nodes is merged into the subgraph structure composed of the current node.
具体而言，若预设的是更新次数阈值时，可在执行了步骤203或204后进一步判断是否小于预设的更新次数阈值。Specifically, if the preset threshold is the number of update times, after step 203 or 204 is executed, it may be further determined whether it is less than the preset threshold of the number of update times.
若判断的结果为是时可返回执行步骤204，也就是说，当更新次数阈值较小时可认定为当前的子图结构的筛选已经基本符合要求。其中，更新次数阈值可定义为预设计算次数中的更新次数的比率。例如，更新次数阈值若设为20％，而在导率计算的步骤201中，若200次的计算次数中，可进行扩展的次数仅为5次，则可认定为子图结构的筛选已经基本符合要求。If the result of the judgment is yes, it may return to step 204, that is to say, when the threshold of the number of update times is small, it may be deemed that the filtering of the current sub-picture structure has basically met the requirements. The update frequency threshold may be defined as a ratio of the update times among the preset calculation times. For example, if the threshold for the number of updates is set to 20%, and in step 201 of the conductivity calculation, if the number of expansions that can be expanded is only 5 of the 200 calculations, then the filtering of the sub-graph structure has been considered to be basically Meet the requirements.
若判断的结构为否时，在可返回到步骤201。If the determined structure is no, it may return to step 201.
若预设的是扩展次数阈值时，判断是否大于预设的扩展次数阈值。If the preset is the expansion times threshold, it is determined whether it is greater than the preset expansion times threshold.
若判断的结果为是时可返回到步骤204，也就是说，当扩展次数阈值较大时可认定为子图结构的筛选已经基本符合要求。其中，扩展次数阈值可定义为步骤201中的计算导率的次数，这是因为每次导率的计算都会进行扩展一次。若判断的结构为否时，可返回到步骤201。If the result of the judgment is yes, it may return to step 204, that is, when the threshold of the number of expansion times is large, it may be deemed that the filtering of the sub-picture structure has basically met the requirements. The threshold of the number of expansion times can be defined as the number of times the conductivity is calculated in step 201, because each time the conductivity calculation is expanded once. If the determined structure is no, it may return to step 201.
以上，各网络设备通过导率获取到了扩展后的子图结构。应理解，在其他实施例中，各网络设备还可以通过图结构的其他相关性的参数来获取子图结构。Above, each network device obtains the expanded sub-graph structure through the conductivity. It should be understood that in other embodiments, each network device may also obtain the sub-graph structure through other correlation parameters of the graph structure.
各网络设备在步骤101中获取到的子图结构仅为对处理***10中的图结构的一个粗略的划分处理，步骤101中获取的子图结构还不适合用于对图结构进行分析，因此需要通过步骤102和103进一步对步骤101获取到的子图结构进行进一步的细处理，即每个网络设备对其步骤101获取到的子图结构在图结构中的大小进行调整，以实现最终各网络设备的子图结构间的划分关系是对图结构较为准确、合理的划分。The sub-graph structure acquired by each network device in step 101 is only a rough division process of the graph structure in the processing system 10, and the sub-graph structure acquired in step 101 is not yet suitable for analyzing the graph structure, so It is necessary to further perform detailed processing on the sub-picture structure obtained in step 101 through steps 102 and 103, that is, each network device adjusts the size of the sub-picture structure obtained in step 101 in the graph structure to achieve the final The division relationship between the sub-graph structures of network equipment is a relatively accurate and reasonable division of the graph structure.
步骤102：获取子图结构的优化特征参数，其中，该优化特征参数用于确定子图结构在图结构中的大小。Step 102: Obtain the optimized feature parameters of the subgraph structure, where the optimized feature parameters are used to determine the size of the subgraph structure in the graph structure.
首先各网络设备12可将获取的子图结构的特征，例如子图结构中的边、节点等特征信息通过矩阵、连表等方式进行转换并存储。以便于得到其中的特征。First, each network device 12 may convert and store the acquired feature information of the subgraph structure, for example, feature information such as edges and nodes in the subgraph structure through a matrix or a linked list. In order to get the characteristics.
由前文所述，步骤101得到的是一个初略处理后的子图结构，其各特征是基于初略处理后的子图结构提炼出来的，可能并不是客观的最优特征，因此需要进行优化运算，得到优化特征参数。本文中，该子图结构的特征参数例如包括子图结构中的边、节点等，用于确定出该子图结构在图结构中的大小。例如，该子图结构的特征参数包括节点A、B、C及其边集合，即可定义出该图结构中由该节点A、B、C及其边集合组成的范围即为子图结构的范围。As mentioned above, the result of step 101 is a sub-graph structure after preliminary processing, and its features are extracted based on the sub-graph structure after initial processing, which may not be the objective optimal feature, so it needs to be optimized After operation, the optimized characteristic parameters are obtained. In this paper, the characteristic parameters of the subgraph structure include, for example, edges and nodes in the subgraph structure, which are used to determine the size of the subgraph structure in the graph structure. For example, the characteristic parameters of the subgraph structure include nodes A, B, C and their edge sets, and the range of the node structure A, B, C and their edge sets in the graph structure can be defined as the subgraph structure. range.
基于此，各网络设备可结合邻居网络设备的子图结构的特征参数以及自身存储的子图结构的特征参数获取自身存储的子图结构的优化特征参数。步骤103将根据该优化特征参数进行子图结构的调整。Based on this, each network device can obtain the optimized feature parameters of the sub-graph structure stored by itself by combining the feature parameters of the sub-graph structure of the neighbor network device and the feature parameters of the sub-graph structure stored by itself. Step 103 will adjust the sub-graph structure according to the optimized feature parameters.
具体各网络设备12可利用其相邻网络设备的子图结构的特征参数对自身子图结构的特征参数进行迭代，以获取自身子图结构的优化特征参数。也就是说通过利用相邻网络设备的子图结构的特征参数对自身的子图结构的特征参数进行不断的迭代更新，使得其在迭代过程中慢慢接近于最后的优化特征参数。Specifically, each network device 12 may use the feature parameters of the subgraph structure of its neighboring network device to iterate over the feature parameters of its own subgraph structure to obtain the optimized feature parameters of its own subgraph structure. That is to say, by using the characteristic parameters of the sub-graph structure of the neighboring network device to continuously update the characteristic parameters of the sub-graph structure of itself, it gradually approaches the final optimized characteristic parameter during the iteration process.
对于每一步迭代，首先根据自身前一次迭代得到的子图结构的特征参数获取子图结构的目标函数梯度，然后利用相邻网络设备的前一次迭代得到的特征参数和子图结构的目标函数梯度，得到自身子图结构当前次迭代的子图结构的特征参数。其对应的迭代公式如下：For each iteration, first obtain the objective function gradient of the subgraph structure according to the characteristic parameters of the subgraph structure obtained by the previous iteration of itself, and then use the characteristic parameters obtained by the previous iteration of the adjacent network device and the objective function gradient of the subgraph structure, The characteristic parameters of the subgraph structure of the current iteration of the subgraph structure are obtained. The corresponding iteration formula is as follows:
Figure PCTCN2018119227-appb-000002
Figure PCTCN2018119227-appb-000002
其中，K为前一次迭代次数，i为网络设备的序号，j为网络设备i以外的其他网络设备的序号，应理解，j在上述公式中是可随着计算的网络设备的不同而变化的，x i为网络设备i的子图结构的特征参数，x j为其他网络设备j的子图结构的特征参数，同理，x j也是可变化的，w ij为其他网络设备j相对于网络设备i的权重，当i和j为非相邻网络设备时，w ij＝0，因此，上述公式中∑ jw ijx j(k)的有效值仅为相邻网络设备的值，α为迭代梯度步幅，
Figure PCTCN2018119227-appb-000003
为目标函数梯度。 Where K is the number of previous iterations, i is the serial number of the network device, j is the serial number of other network devices other than network device i, it should be understood that j in the above formula can vary with the calculation of the network device , X i is the characteristic parameter of the subgraph structure of the network device i, x j is the characteristic parameter of the subgraph structure of the other network device j. Similarly, x j is also variable, and w ij is relative to the network of other network devices j The weight of device i, when i and j are non-adjacent network devices, w ij = 0, therefore, the effective value of ∑ j w ij x j (k) in the above formula is only the value of the adjacent network device, α is Iterative gradient stride,
Figure PCTCN2018119227-appb-000003
Is the gradient of the objective function.
进一步的，各相邻网络设备的权重w ij的值可相同。即利用相邻网络设备的平均最优特征参数来对子图结构的目标函数梯度进行迭代。 Further, the value of the weight w ij of each adjacent network device may be the same. That is, the average optimal characteristic parameters of the adjacent network devices are used to iterate the objective function gradient of the subgraph structure.
当K趋于∞时，计算得到的各网络设备的优化特征参数都收敛于整体最优特征参数的
Figure PCTCN2018119227-appb-000004
领域，其中β为权重w的最大二阶特征值。因此当迭代次数足够时，各网络设备都得到一个接近与整体最优特征参数的优化特征参数。 When K tends to ∞, the calculated optimized characteristic parameters of each network device converge to the overall optimal characteristic parameters
Figure PCTCN2018119227-appb-000004
Field, where β is the largest second-order eigenvalue of the weight w. Therefore, when the number of iterations is sufficient, each network device obtains an optimized feature parameter close to the overall optimal feature parameter.
以上，本实施例中，步骤102是利用梯度下降法对特征参数进行优化形成优化特征参数。As mentioned above, in this embodiment, step 102 is to use the gradient descent method to optimize the feature parameters to form optimized feature parameters.
在其他实施例中，步骤102还可利用牛顿法、共轭梯度法、Levenberg–Marquardt Algorithm(莱文贝格－马夸特方法)等优化算法对特征参数进行优化形成优化特征参数。In other embodiments, step 102 may also use optimization algorithms such as Newton's method, conjugate gradient method, Levenberg-Marquardt Algorithm (Levenberg-Marquardt method) to optimize the characteristic parameters to form optimized characteristic parameters.
步骤103：根据优化特征参数对子图结构进行调整。Step 103: Adjust the subgraph structure according to the optimized feature parameters.
结合步骤102，步骤103具体是各网络设备根据优化后形成的最终最优处理参数对子图结构进行处理。具体而言，各网络设备可通过上述步骤合理确定出各自的子图结构的大小，即通过各网络设备执行上述步骤，可实现对图结构进行精确划分。With reference to step 102, step 103 is specifically that each network device processes the sub-graph structure according to the final optimal processing parameter formed after optimization. Specifically, each network device can reasonably determine the size of its sub-graph structure through the above steps, that is, by performing the above steps through each network device, the graph structure can be accurately divided.
上述处理方法均可通过网络设备实现，具体来说，通过一段计算机程序来表示方法的步骤，构建软件结构以实现该计算机程序，并利用硬件设备来执行该计算机程序从而实现上述方法。The above processing methods can all be implemented by a network device. Specifically, a computer program is used to represent the steps of the method, a software structure is constructed to implement the computer program, and a hardware device is used to execute the computer program to implement the method.
对于计算机程序，以软件形式实现并作为独立的产品销售或使用时，可存储在一个电子设备可读取存储介质中，即，本发明还提供一种计算机可读存储介质，其中存储有计算机程序，该计算机程序被处理器执行时实现上述方法的步骤。计算机可读存储介质可以为U盘、光盘、服务器等。The computer program, when implemented in software and sold or used as an independent product, can be stored in a readable storage medium of an electronic device, that is, the present invention also provides a computer-readable storage medium in which a computer program is stored When the computer program is executed by the processor, the steps of the above method are realized. The computer-readable storage medium may be a U disk, an optical disk, a server, or the like.
对于硬件结构，请参阅图1，其可通过一个处理***10实现。而处理***10中的每个网络设备12所执行的步骤均相同，如前文所述的图结构处理步骤。For the hardware structure, please refer to FIG. 1, which can be implemented by a processing system 10. The steps performed by each network device 12 in the processing system 10 are the same, as shown in the graph structure processing steps described above.
对于网络设备12的硬件结构，请进一步参阅图5，图5是本申请实施例提供的一种网络设备12的硬件结构示意图。如图5所示，本实施例的网络设备12包括处理器121和存储器122，存储器122存储有计算机程序，计算机程序被处理器121执行以实现以下的上述处理方法的步骤。For the hardware structure of the network device 12, please refer to FIG. 5 further. FIG. 5 is a schematic diagram of a hardware structure of a network device 12 according to an embodiment of the present application. As shown in FIG. 5, the network device 12 of this embodiment includes a processor 121 and a memory 122. The memory 122 stores a computer program, and the computer program is executed by the processor 121 to implement the steps of the following processing method.
具体来说，网络设备中的处理器121能够获取子图结构，其中，子图结构属于所述图结构的一部分，并获取子图结构的优化特征参数，最后根 据优化特征参数对子图结构进行调整。Specifically, the processor 121 in the network device can acquire the sub-graph structure, where the sub-graph structure belongs to a part of the graph structure, and obtain the optimized feature parameters of the sub-graph structure, and finally perform the sub-graph structure according to the optimized feature parameters Adjustment.
进一步的，网络设备12还包括收发器123，用于与相邻网络设备通讯，接收相邻网络设备的特征参数等数据。Further, the network device 12 further includes a transceiver 123 for communicating with neighboring network devices and receiving data such as characteristic parameters of the neighboring network devices.
类似于上述的过程，本实施例的收发器123和网络设备12配合能够前文所述的处理方法的所有步骤。Similar to the above process, the transceiver 123 and the network device 12 of this embodiment cooperate with all the steps of the processing method described above.
对于软件结构，上述处理方法不同的步骤对应不同的程序数据，相应的需要构建不同的软件结构。具体请参阅图6，图6是本本申请实施例提供的另一网络设备的软件结构示意图。For the software structure, different steps of the above processing method correspond to different program data, and accordingly different software structures need to be constructed. Please refer to FIG. 6 for details. FIG. 6 is a schematic diagram of a software structure of another network device provided by an embodiment of the present application.
本实施例的网络设备60包括：The network device 60 of this embodiment includes:
第一获取模块601：用于获取子图结构，其中，子图结构属于图结构的一部分。The first acquiring module 601 is used to acquire a sub-graph structure, where the sub-graph structure belongs to a part of the graph structure.
第二获取模块602，用于获取子图结构的优化特征参数其中，所述优化特征参数用于确定所述子图结构在所述图结构中的大小；The second obtaining module 602 is configured to obtain optimized feature parameters of the sub-graph structure. The optimized feature parameters are used to determine the size of the sub-graph structure in the graph structure;
处理模块603，用于根据优化特征参数对子图结构进行调整。The processing module 603 is used to adjust the sub-graph structure according to the optimized feature parameters.
上述各个模块能够前文所述的处理方法的各个步骤，具体不再赘述。Each of the above modules can perform the steps of the processing method described above, and the details are not repeated here.
以上所述仅为本申请的实施方式，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。The above are only the embodiments of the present application, and therefore do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by the description and drawings of this application, or directly or indirectly used in other related technologies In the field, the same reason is included in the scope of patent protection of this application.

Claims (19)

  1. 一种图结构的处理方法，其特征在于，所述处理方法用于被网络中的多个网络设备分别执行，所述处理方法包括：A processing method with a graph structure is characterized in that the processing method is used to be executed by multiple network devices in a network, and the processing method includes:
    所述网络设备获取子图结构，其中，所述子图结构属于所述图结构的一部分；The network device obtains a sub-graph structure, wherein the sub-graph structure belongs to a part of the graph structure;
    获取所述子图结构的优化特征参数，其中，所述优化特征参数用于确定所述子图结构在所述图结构中的大小；Acquiring optimized feature parameters of the sub-graph structure, wherein the optimized feature parameters are used to determine the size of the sub-graph structure in the graph structure;
    根据所述优化特征参数对所述子图结构进行调整。Adjusting the sub-graph structure according to the optimized feature parameters.
  2. 根据权利要求1所述的处理方法，其特征在于，所述网络设备获取子图结构包括：The processing method according to claim 1, wherein the network device acquiring sub-graph structure comprises:
    所述网络设备根据导率来获取所述子图结构。The network device obtains the sub-graph structure according to the conductivity.
  3. 根据权利要求2所述的处理方法，其特征在于，所述网络设备根据导率来获取所述子图结构包括：The processing method according to claim 2, wherein the network device acquiring the sub-graph structure according to the conductivity includes:
    所述网络设备根据参考子图结构中各节点的相邻节点组成的子图结构的导率，对所述参考子图结构进行调整，以获得所述网络设备的所述子图结构。The network device adjusts the reference subgraph structure according to the conductivity of the subgraph structure composed of neighboring nodes of each node in the reference subgraph structure to obtain the subgraph structure of the network device.
  4. 根据权利要求3所述的处理方法，其特征在于，所述网络设备根据参考子图结构中各节点的相邻节点组成的子图结构的导率，对所述参考子图结构进行调整，以获得所述网络设备的所述子图结构，包括：The processing method according to claim 3, wherein the network device adjusts the reference subgraph structure according to the conductivity of the subgraph structure composed of adjacent nodes of each node in the reference subgraph structure, to Obtaining the sub-graph structure of the network device includes:
    对所述参考子图结构中的每一节点，计算所述节点的相邻节点组成的子图结构的导率；For each node in the reference subgraph structure, calculate the conductivity of the subgraph structure composed of adjacent nodes of the node;
    判断所述导率是否满足预设的导率阈值，并在判断的结果为是时，将所述相邻节点组成的子图结构对所述参考子图结构进行扩展。It is judged whether the conductivity meets a preset conductivity threshold, and when the judgment result is yes, the subgraph structure composed of the neighboring nodes is expanded to the reference subgraph structure.
  5. 根据权利要求3或4所述的处理方法，其特征在于，所述导率至少由以下步骤形成：The processing method according to claim 3 or 4, wherein the conductivity is formed by at least the following steps:
    根据以下公式获得所述导率：The conductivity is obtained according to the following formula:
    Figure PCTCN2018119227-appb-100001
    Figure PCTCN2018119227-appb-100001
    其中，Φ为所述导率，S为所述相邻节点组成的子图结构，E(S,V-S)为 所述相邻节点组成的子图结构S和所述相邻节点组成的子图结构S的补集之间的相连接边数，A为所述相邻节点组成的子图结构S的度矩阵，A V-S为所述相邻节点组成的子图结构S的补集的度矩阵。 Where Φ is the conductivity, S is the subgraph structure composed of the adjacent nodes, and E(S,VS) is the subgraph structure S composed of the adjacent nodes and the subgraph composed of the adjacent nodes The number of connected edges between the complementary sets of the structure S, A is the degree matrix of the subgraph structure S of the adjacent nodes, and A VS is the degree matrix of the complementary set of the subgraph structures S of the adjacent nodes .
  6. 根据权利要求1所述的处理方法，其特征在于，所述获取所述子图结构的优化特征参数包括：The processing method according to claim 1, wherein the acquiring the optimized feature parameters of the sub-graph structure comprises:
    所述网络设备利用其相邻网络设备的所述子图结构的特征参数对自身所述子图结构的特征参数进行迭代，以获取自身所述子图结构的优化特征参数。The network device uses the feature parameters of the sub-graph structure of its neighboring network devices to iterate the feature parameters of the sub-graph structure of itself to obtain optimized feature parameters of the sub-graph structure of itself.
  7. 根据权利要求6所述的处理方法，其特征在于，所述利用其相邻网络设备的所述子图结构的特征参数对自身所述子图结构的特征参数进行迭代包括：The processing method according to claim 6, wherein the using the characteristic parameters of the subgraph structure of the neighboring network device to iterate the characteristic parameters of the subgraph structure of itself includes:
    根据自身前一次迭代得到的所述子图结构的特征参数获取所述子图结构的目标函数梯度；Acquiring the target function gradient of the sub-graph structure according to the characteristic parameters of the sub-graph structure obtained by the previous iteration of itself;
    利用相邻网络设备的前一次迭代得到的特征参数和所述子图结构的目标函数梯度，得到自身所述子图结构当前次迭代的子图结构的特征参数；Use the characteristic parameters obtained by the previous iteration of the adjacent network device and the objective function gradient of the subgraph structure to obtain the characteristic parameters of the subgraph structure of the current iteration of the subgraph structure;
    重复执行上述步骤以进行多次迭代，以得到自身所述子图结构的优化特征参数。Repeat the above steps to perform multiple iterations to obtain the optimized feature parameters of the sub-graph structure.
  8. 根据权利要求7所述的处理方法，其特征在于，所述利用相邻网络设备的前一次迭代得到的特征参数和所述子图结构的目标函数梯度，得到自身所述子图结构当前次迭代的子图结构的特征参数包括：The processing method according to claim 7, wherein the characteristic parameter obtained by the previous iteration of the adjacent network device and the objective function gradient of the subgraph structure are used to obtain the current iteration of the subgraph structure The characteristic parameters of the subgraph structure include:
    根据以下公式进行迭代：Iterate according to the following formula:
    Figure PCTCN2018119227-appb-100002
    Figure PCTCN2018119227-appb-100002
    其中，K为前一次迭代次数，i为所述网络设备的序号，j为所述网络设备i以外的其他网络设备的序号，x i为网络设备i的所述子图结构的特征参数，x j为其他网络设备的子图结构的特征参数，w ij为其他网络设备j相对于所述网络设备i的权重，且当i和j为非相邻网络设备时，w ij＝0，α为迭代梯度步幅，
    Figure PCTCN2018119227-appb-100003
    为所述目标函数梯度。     Where K is the number of previous iterations, i is the serial number of the network device, j is the serial number of other network devices other than the network device i, x i is the characteristic parameter of the sub-graph structure of the network device i, x j is a characteristic parameter of the subgraph structure of other network devices, w ij is the weight of other network devices j relative to the network device i, and when i and j are non-adjacent network devices, w ij =0 and α is Iterative gradient stride,
    Figure PCTCN2018119227-appb-100003
    Is the gradient of the objective function.
  9. 根据权利要求8所述的处理方法，其特征在于，各相邻网络设备的权重w ij的值相同。 The processing method according to claim 8, wherein the weight w ij of each adjacent network device has the same value.
  10. 一种图结构的处理***，其特征在于，所述处理***包括多个网络设备，多个所述网络设备形成一网络，其中：A processing system with a graph structure is characterized in that the processing system includes multiple network devices, and the multiple network devices form a network, wherein:
    每个网络设备用于执行以下步骤：Each network device is used to perform the following steps:
    获取子图结构，其中，所述子图结构属于所述图结构的一部分；Obtaining a sub-graph structure, wherein the sub-graph structure belongs to a part of the graph structure;
    获取所述子图结构的优化特征参数，其中，所述优化特征参数用于确定所述子图结构在所述图结构中的大小；Acquiring optimized feature parameters of the sub-graph structure, wherein the optimized feature parameters are used to determine the size of the sub-graph structure in the graph structure;
    根据所述优化特征参数对所述子图结构进行调整。Adjusting the sub-graph structure according to the optimized feature parameters.
  11. 根据权利要求10所述的处理***，其特征在于，各网络设备根据参考子图结构中各节点的相邻节点组成的子图结构的导率，对所述参考子图结构进行调整，以获得所述网络设备的所述子图结构。The processing system according to claim 10, wherein each network device adjusts the reference subgraph structure according to the conductivity of the subgraph structure composed of adjacent nodes of each node in the reference subgraph structure to obtain The subgraph structure of the network device.
  12. 根据权利要求11所述的处理***，其特征在于，The processing system according to claim 11, wherein:
    对所述参考子图结构中的每一节点，各网络设备计算所述节点的相邻节点组成的子图结构的导率，进一步判断所述导率是否满足预设的导率阈值，并在判断的结果为是时，将所述相邻节点组成的子图结构对所述参考子图结构进行扩展。For each node in the reference subgraph structure, each network device calculates the conductivity of the subgraph structure composed of neighboring nodes of the node, and further determines whether the conductivity meets a preset conductivity threshold, and When the judgment result is yes, the sub-graph structure composed of the neighboring nodes is expanded to the reference sub-graph structure.
  13. 根据权利要求11或12所述的处理***，其特征在于，根据以下公式获得所述导率：The processing system according to claim 11 or 12, wherein the conductivity is obtained according to the following formula:
    Figure PCTCN2018119227-appb-100004
    Figure PCTCN2018119227-appb-100004
    其中，Φ为所述导率，S为所述相邻节点组成的子图结构，E(S,V-S)为所述相邻节点组成的子图结构S和所述相邻节点组成的子图结构S的补集之间的相连接边数，A为所述相邻节点组成的子图结构S的度矩阵，A V-S为所述相邻节点组成的子图结构S的补集的度矩阵。 Where Φ is the conductivity, S is the subgraph structure composed of the adjacent nodes, and E(S,VS) is the subgraph structure S composed of the adjacent nodes and the subgraph composed of the adjacent nodes The number of connected edges between the complementary sets of structure S, A is the degree matrix of the subgraph structure S of the adjacent nodes, and A VS is the degree matrix of the complementary set of the subgraph structures S of the adjacent nodes .
  14. 根据权利要求10所述的处理***，其特征在于，各网络设备利用其相邻网络设备的所述子图结构的特征参数对自身所述子图结构的特征参数进行迭代，以获取自身所述子图结构的优化特征参数。The processing system according to claim 10, characterized in that each network device uses the characteristic parameters of the sub-graph structure of its neighboring network devices to iterate the characteristic parameters of the sub-graph structure of itself to obtain its own Optimized characteristic parameters of subgraph structure.
  15. 根据权利要求14所述的处理***，其特征在于，所述网络设备进一步根据自身前一次迭代得到的所述子图结构的特征参数获取所述子图结构的目标函数梯度；进而利用相邻网络设备的前一次迭代得到的特征参数和所述子图结构的目标函数梯度，得到自身所述子图结构当前次迭代的子 图结构的特征参数；以此重复执行以进行多次迭代，以得到自身所述子图结构的优化特征参数。The processing system according to claim 14, wherein the network device further obtains the target function gradient of the subgraph structure according to the characteristic parameters of the subgraph structure obtained by the previous iteration of itself; The feature parameters obtained from the previous iteration of the device and the target function gradient of the sub-graph structure, to obtain the feature parameters of the sub-graph structure of the current iteration of the sub-graph structure itself; this is repeated to perform multiple iterations to obtain The optimized feature parameters of the sub-graph structure described by itself.
  16. 根据权利要求15所述的处理***，其特征在于，根据以下公式进行迭代：The processing system according to claim 15, wherein the iteration is performed according to the following formula:
    Figure PCTCN2018119227-appb-100005
    Figure PCTCN2018119227-appb-100005
    其中，K为前一次迭代次数，i为所述网络设备的序号，j为所述网络设备i以外的其他网络设备的序号，x i为所述网络设备i的所述子图结构的特征参数，x j为其他网络设备的子图结构的特征参数，w ij为其他网络设备j相对于网络设备i的权重，且当i和j为非相邻网络设备时，w ij＝0，α为迭代梯度步幅，
    Figure PCTCN2018119227-appb-100006
    为所述目标函数梯度。     Where K is the number of previous iterations, i is the serial number of the network device, j is the serial number of other network devices other than the network device i, and x i is a characteristic parameter of the subgraph structure of the network device i , X j is the characteristic parameter of the subgraph structure of other network devices, w ij is the weight of other network devices j relative to the network device i, and when i and j are non-adjacent network devices, w ij =0, α is Iterative gradient stride,
    Figure PCTCN2018119227-appb-100006
    Is the gradient of the objective function.
  17. 一种网络设备，其特征在于，所述网络设备包括处理器和存储器，所述存储器存储有计算机程序，所述计算机程序被所述处理器执行以实现以下的所述处理方法：A network device, characterized in that the network device includes a processor and a memory, and the memory stores a computer program, and the computer program is executed by the processor to implement the following processing method:
    获取子图结构，其中，所述子图结构属于所述图结构的一部分；Obtaining a sub-graph structure, wherein the sub-graph structure belongs to a part of the graph structure;
    获取所述子图结构的优化特征参数，其中，所述优化特征参数用于确定所述子图结构在所述图结构中的大小；Acquiring optimized feature parameters of the sub-graph structure, wherein the optimized feature parameters are used to determine the size of the sub-graph structure in the graph structure;
    根据所述优化特征参数对所述子图结构进行调整。Adjusting the sub-graph structure according to the optimized feature parameters.
  18. 一种网络设备，其特征在于，所述网络设备包括：A network device, characterized in that the network device includes:
    第一获取模块：用于获取子图结构，其中，所述子图结构属于所述图结构的一部分；The first acquiring module: used to acquire a sub-graph structure, wherein the sub-graph structure belongs to a part of the graph structure;
    第二获取模块，用于获取所述子图结构的优化特征参数，其中，所述优化特征参数用于确定所述子图结构在所述图结构中的大小；A second obtaining module, configured to obtain optimized feature parameters of the sub-graph structure, wherein the optimized feature parameters are used to determine the size of the sub-graph structure in the graph structure;
    处理模块，用于根据所述优化特征参数对所述子图结构进行调整。The processing module is configured to adjust the sub-graph structure according to the optimized feature parameter.
  19. 一种计算机可读存储介质，其上存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现权利要求1-9中任一项所述处理方法的步骤。A computer-readable storage medium on which a computer program is stored, characterized in that when the computer program is executed by a processor, the steps of the processing method according to any one of claims 1 to 9 are realized.
PCT/CN2018/119227 2018-12-04 2018-12-04 Graph structure processing method and system, network device and storage medium WO2020113437A1 (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN201880002415.9A CN109952742B (en) 2018-12-04 2018-12-04 Graph structure processing method, system, network device and storage medium
PCT/CN2018/119227 WO2020113437A1 (en) 2018-12-04 2018-12-04 Graph structure processing method and system, network device and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2018/119227 WO2020113437A1 (en) 2018-12-04 2018-12-04 Graph structure processing method and system, network device and storage medium

Publications (1)

Publication Number Publication Date
WO2020113437A1 true WO2020113437A1 (en) 2020-06-11

Family

ID=67006582

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/119227 WO2020113437A1 (en) 2018-12-04 2018-12-04 Graph structure processing method and system, network device and storage medium

Country Status (2)

Country Link
CN (1) CN109952742B (en)
WO (1) WO2020113437A1 (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111191715A (en) * 2019-12-27 2020-05-22 深圳市商汤科技有限公司 Image processing method and device, electronic equipment and storage medium
CN112990332B (en) * 2021-03-26 2023-06-02 杭州海康威视数字技术股份有限公司 Sub-graph scale prediction and distributed training method and device and electronic equipment

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1571363A (en) * 2004-05-13 2005-01-26 上海交通大学 Network partitioning method for large-scale complicated network
CN106054875A (en) * 2016-05-25 2016-10-26 北京航空航天大学 Distributed multi-robot dynamic network connectivity control method
CN107480213A (en) * 2017-07-27 2017-12-15 上海交通大学 Community's detection and customer relationship Forecasting Methodology based on sequential text network
CN107798146A (en) * 2017-12-01 2018-03-13 江苏理工学院 A kind of education resource Ontology Matching method based on subgraph approximation isomorphism
WO2018077039A1 (en) * 2016-10-27 2018-05-03 腾讯科技（深圳）有限公司 Community discovery method, apparatus, server, and computer storage medium
CN108600013A (en) * 2018-04-26 2018-09-28 北京邮电大学 The overlapping community discovery method and device of dynamic network

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103310453B (en) * 2013-06-17 2015-12-09 北京理工大学 A kind of fast image registration method based on subimage Corner Feature
EP4202782A1 (en) * 2015-11-09 2023-06-28 Google LLC Training neural networks represented as computational graphs
US10409828B2 (en) * 2016-07-29 2019-09-10 International Business Machines Corporation Methods and apparatus for incremental frequent subgraph mining on dynamic graphs
CN108182265B (en) * 2018-01-09 2021-06-29 清华大学 Multilayer iterative screening method and device for relational network
CN108257036A (en) * 2018-01-12 2018-07-06 西安电子科技大学 Discovery method, the Web Community's system of overlapping community are extended based on seed node
CN108537452A (en) * 2018-04-13 2018-09-14 中山大学 It is a kind of to be overlapped community division method towards the intensive of large-scale complex network

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1571363A (en) * 2004-05-13 2005-01-26 上海交通大学 Network partitioning method for large-scale complicated network
CN106054875A (en) * 2016-05-25 2016-10-26 北京航空航天大学 Distributed multi-robot dynamic network connectivity control method
WO2018077039A1 (en) * 2016-10-27 2018-05-03 腾讯科技（深圳）有限公司 Community discovery method, apparatus, server, and computer storage medium
CN107480213A (en) * 2017-07-27 2017-12-15 上海交通大学 Community's detection and customer relationship Forecasting Methodology based on sequential text network
CN107798146A (en) * 2017-12-01 2018-03-13 江苏理工学院 A kind of education resource Ontology Matching method based on subgraph approximation isomorphism
CN108600013A (en) * 2018-04-26 2018-09-28 北京邮电大学 The overlapping community discovery method and device of dynamic network

Also Published As

Publication number Publication date
CN109952742A (en) 2019-06-28
CN109952742B (en) 2022-02-22

Similar Documents

Publication Publication Date Title
CN109194707B (en) Distributed graph embedding method and device
Mei et al. Signal processing on graphs: Estimating the structure of a graph
WO2021089013A1 (en) Spatial graph convolutional network training method, electronic device and storage medium
US20230385652A1 (en) System and Method of Federated Learning with Diversified Feedback
TWI700599B (en) Method and device for embedding relationship network diagram, computer readable storage medium and computing equipment
CN105677489B (en) The dynamic of batch gap size sets system and method under discrete stream process model
CN103838803A (en) Social network community discovery method based on node Jaccard similarity
WO2020113437A1 (en) Graph structure processing method and system, network device and storage medium
TWI700647B (en) Electronic apparatus and compression method for artificial neural network
CN107623639A (en) Data flow distribution similarity join method based on EMD distances
CN111460234A (en) Graph query method and device, electronic equipment and computer readable storage medium
CN111178261A (en) Face detection acceleration method based on video coding technology
CN110795599B (en) Video emergency monitoring method and system based on multi-scale graph
Zhang et al. Grading learning for blind source separation
CN108470251B (en) Community division quality evaluation method and system based on average mutual information
US10320636B2 (en) State information completion using context graphs
CN116468967B (en) Sample image screening method and device, electronic equipment and storage medium
WO2023143080A1 (en) Data processing method and related device
CN115913995A (en) Cloud service dynamic QoS prediction method based on Kalman filtering correction
Li et al. High resolution radar data fusion based on clustering algorithm
US11386335B2 (en) Systems and methods providing evolutionary generation of embeddings for predicting links in knowledge graphs
WO2022100118A1 (en) Model processing method and related device
CN114881124B (en) Causal relation graph construction method and device, electronic equipment and medium
Gunarathne et al. Distributing deep learning inference on edge devices
Liu et al. A fuzzy density peak optimization initial centers selection for k-medoids clustering algorithm

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18942521

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18942521

Country of ref document: EP

Kind code of ref document: A1