62 results about "Connectivity" patented technology

Methods and systems for compressing and decompressing 3-D geometry data which includes regularly tiled surface portions. One compression method includes representing a surface portion as a "vertex raster", which comprises specifying an extent value and encoding the vertex parameter values of vertices within the surface portion. The extent of the surface portion specifies the arrangement of vertices within the surface portion, and allows the vertices to be properly assembled into drawing primitives during decompression. The encoded vertex parameter values may be encoded globally (by setting initial values and corresponding delta values), locally (on a per-vertex basis), or using a combination of these techniques. Absolute, delta encoding, or delta-delta encoding may be utilized for these parameter values. Vertex parameters which may be encoded in this manner include position, color, normals, z-displacement values, texture map coordinates, and surface material properties. Additionally, connectivity information may also be encoded using this compression method by specifying quad split bits and half-resolution edges. Quad split bits are used to tessellate a quadrilateral formed by neighboring vertices of a surface portion according to the direction of the strongest color change. Half-resolution edges are utilized to gradually shift from an area of high resolution to an adjacent surface portion represented in lower resolution. For graphical objects which include a plurality of adjacent surface portions, a step command is disclosed which allows data from one surface portion to be advantageously reused. Decompression of a vertex raster representation may comprise decoding the extent value, global parameter values, and a per-vertex stream of local parameter values.

Methods and systems for compressing and decompressing 3-D geometry data which includes regularly tiled surface portions. One compression method includes representing a surface portion as a "vertex raster" by specifying an extent value and encoding the vertex parameter values of vertices within the surface portion. The extent value specifies the arrangement of vertices within the surface portion and allows the vertices to be properly assembled into drawing primitives during decompression. The encoded vertex parameter values (such as position, color, normals, z-displacement values, texture map coordinates, and surface material properties) may be encoded globally (by setting initial values and corresponding delta values), locally (on a per-vertex basis), or using a combination of these techniques. Absolute, delta, or delta-delta encoding may be utilized for these parameter values. Connectivity information may also be encoded using this compression method by specifying quad split bits and half-resolution edges. For graphical objects which include a plurality of adjacent surface portions, a step command may be used to allow reuse of data for more than one surface.

The embodiment of the invention provides a binding recommendation method and system based on a graph convolutional neural network. The method comprises the steps: acquiring historical interaction dataof a user and binding, historical interaction data of the user and an article and affiliation data of the binding and the article; inputting the data into the binding recommendation model to obtain auser and binding interaction possibility recommendation result output by the binding recommendation model, wherein the bundled recommendation model is obtained by constructing a unified heterogeneousgraph based on a user data set, a bundled data set and an article data set, extracting article hierarchical graph convolution propagation features and bundled hierarchical graph convolution propagation features, then performing joint prediction and feature connection, and performing training based on a difficult negative sample learning strategy. According to the embodiment of the invention, theinteraction relationship and subordination relationship among the user, the bundle and the article are reconstructed into the graph, and three associated entity representations are learned from a complex topological structure and high-order connectivity by utilizing the powerful capability of the graph neural network, so that better recommendation performance can be achieved.

The invention discloses a comprehensive resistance weight and graph theory-based river network connectivity assessment method. The method includes the following steps that: a river network is generalized into a directed river network graph model which adopts river sections as edges and adopts the sources, convergences and intersections of rivers, as well as gate dams as vertexes; the resistance items, namely, hydrology, hydrodynamic force, water quality and human activities of any of river sections are calculated, the comprehensive resistance weight (CRW) of the river section is calculated; and the CRW of each river section in the river network is calculated as the previous step, and the CRWs of the river sections are adopted as the weights of the edges, so that a river network weighted adjacency matrix can be established, and a minimum comprehensive resistance matrix can be also obtained, and therefore, the minimum comprehensive resistance of the river network can be obtained, and the reciprocal of the minimum comprehensive resistance of the river network is solved, so that the comprehensive connectivity index (CCI) of the river network can be obtained. According to a method for optimizing a river network gate dam system by using the above method, any one gate dam in the river network is removed sequentially, and comprehensive connectivity indexes (CCI) are calculated again, so that the obtained comprehensive connectivity indexes (CCI) are arranged according to a large-to-small order, and the arrangement order is a priority removing order, and the river network gate dam system is optimized according to the priority removing order.

A method and system provide the ability to optimize a structural engineering design. A dataset is synthesized by acquiring a structural skeleton design of an entire building. The skeleton defines locations and connectivities of bars that represent columns or beams. The skeleton design is represented as a structural graph with each bar represented as a graph node and edges connecting graph nodes. Structural simulation results are computed for the synthetic dataset based on the structural graph, various loads, and a structural analysis. A simulation model and a size optimization model are trained based on the structural simulation results with the size optimization model determining cross-section sizes for the bars to satisfy a building mass objective, building constraints, and output from the simulation model. The structural engineering design is output from the size optimization model.

The embodiment of the invention provides an anti-fraud detection method and system based on graph analysis, and the method comprises the steps: 1), building a relation graph corresponding to a user incommunication data based on the communication data which is provided by a communication operator and comprises victim communication data; 2) obtaining victim user characteristics from the relation graph, training a random forest classification model by utilizing the victim user characteristics, and then identifying whether a user to be identified is a victim result by utilizing the trained randomforest classification model; 3) shrinking other nodes except the number nodes in the relation graph into corresponding edges or nodes to obtain a target relation graph; and 4) obtaining the number node having the largest influence on other number nodes in each sub-connected graph in the target relation graph, and taking the user corresponding to the number node having the largest influence on other number nodes as a fraud criminal. By applying the embodiment of the invention, the fraud criminal can be identified.

The embodiment of the invention discloses a method and device for determining user community information. According to one specific embodiment, the method comprises the steps: obtaining user information of a preset user set, and constructing a connected graph; extracting connected sub-graphs from the connected graphs through sampling, and extracting user relationship features and community features of the connected sub-graphs; based on the user relationship characteristics and the community characteristics corresponding to the connected sub-graphs, respectively determining user relationship prior information and community prior information corresponding to the connected sub-graphs, and estimating distribution information of communities to which users in the user subsets corresponding to the connected sub-graphs belong by adopting a preset topic model based on the user relationship prior information, the community prior information and the connected sub-graphs; and determining communityaffiliation information of the corresponding users based on the distribution information of the communities to which the users in the user subsets corresponding to the connected subgraphs belong. According to the embodiment, the distribution information of the community to which the user belongs estimated by the topic model is coupled with the community feature and the relationship feature of theuser, so the accuracy of determining the community attribution information of the user is improved.

The embodiment of the invention discloses a graph data partitioning method, device and equipment. The method comprises the steps: obtaining to-be-processed graph data; scattering the node data of theto-be-processed graph data to obtain a first partition result of the to-be-processed graph data; based on the first partitioning result, performing partitioning, performing clustering analysis on eachcorresponding node in the first partitioning result; and obtaining a second partition result of the to-be-processed graph data, so that vertex data and/or edge data of adjacent connected graphs in the second partition result are/is stored in the same partition, and the second partition result is a final partition result of the to-be-processed graph data. By adopting the partitioning method of thegraph data provided by the embodiment of the invention, the storage load of the graph data can be uniform, the hot spot problem is avoided, and the calculation efficiency of the graph data can be improved.

The present invention establishes a process for creating a set of structures that allows efficient storage and subsequent handling. The multigraph is represented using bitmaps with element counters and mappings between values and organized bitmaps to faciliate the handling of the multigraphs. The bits in the bitmaps represent two aspects of the multigraph: 1) indexing of all the objects of the multigraph as a function of their identifiers and 2) connectivity between objects of the multigraph, whether they are vertices or edges. Mappings allow, given a value, accessing the objects of the multigraph which contain such value. Multigraph operations are solved by accessing the mappings and applying logical operations on the bitmaps. This way of representing a graph allows efficiently performing graph operations such as: inserting a vertex or an edge, inserting an attribute, acquiring the incoming and outgoing edges of an attribute, etc.

The invention discloses a text graph construction method based on text content features, which is characterized in that when edges of a text graph are constructed, dependence on a co-occurrence relationship is avoided, and meanwhile, semantic relationships of word nodes are reserved, so that text semantic features can be accurately expressed; in addition, two graph construction methods are provided, an appropriate method can be selected according to practical application, the graph obtained through the first method can have a global node, namely the node with the maximum degree, and the globalnode has a connecting edge with other remaining nodes; however, if the number of nodes in the graph is large and the weight difference of the nodes is not large, the difference between the degree setvalue of the intermediate node and the node weight value is too large in the mode; and the second method can solve the defects of the method to a certain extent, but the constructed graph may not beconnected, and if a subsequently adopted learning algorithm has a requirement on the connectivity of the graph or global node features need to be utilized to represent a graph feature method, the graph feature method can be flexibly selected according to actual requirements, so that the flexibility of text graph construction is improved.

The invention belongs to the field of graph data security, and particularly relates to a graph network classification model training method, device and system and electronic equipment. The method comprises the following steps: acquiring a graph data set, and dividing the graph data set into label nodes and label-free nodes; inputting the graph data with the label nodes into a classification model for training; using the trained classification model to classify graph data without label nodes; randomly selecting adversarial nodes from label nodes and label-free nodes, and calculating the influence of the adversarial nodes on a loss function of an adversarial sample generator so as to generate an adversarial sample; inputting the adversarial sample into the trained classification model for training again to obtain an enhanced classification model. According to the invention, the trained classification model is used to predict the labels of the label-free nodes, so that the connectivity problem and the low label rate problem in semi-supervised learning can be solved. According to the method, the robustness of the node classification model can be effectively improved.

The invention discloses a transformer substation secondary circuit fault positioning method and system based on a graph neural network, wherein the method comprises the steps: analyzing a configuration description file of an intelligent transformer substation, storing an analysis result into a graph database, and building a corresponding relation between a physical circuit and a virtual circuit of secondary equipment; making a training set by using a historical database or a fault emergence method, and training a graph neural network model offline; finding out all associated fault equipment by using an alarm signal, preprocessing the alarm signal, judging whether the associated fault equipment forms a connected graph or not, and if the connected graph is not formed, dividing the the associated fault equipment into independent connected graphs, representing topological information and alarm signals of the independent connected graphs, and inputting the topological information and alarm signals into a trained graph neural network model; and predicting the fault type of the associated fault equipment by using the graph neural network model. According to the method, the graph neural network is used for building the fault positioning model, so that the accuracy of the model is improved under the condition that the networking mode is changed.

The invention relates to the field of artificial intelligence, and provides a word sense disambiguation method and device based on a graph expansion structure, equipment and a medium, which can extract candidate nodes in an initial graph according to a created window, realize the expansion of the graph, and then improve the connectivity. The weight of each edge is calculated according to the text similarity between the two candidate nodes corresponding to each edge and the word meaning similarity between the two candidate nodes corresponding to each edge, and the importance of each candidate node is determined according to the weight of each edge, so that high semantic and emotion information is explored and captured, and the word meaning disambiguation accuracy is improved; and according to the importance degree of each candidate node, interference removal processing is performed on the candidate map to obtain a target map, so that the word sense disambiguation efficiency is improved through interference removal processing, center measurement is carried out on the target map to obtain a target word sense, and the improvement of the word sense disambiguation effect is realized. In addition, the invention also relates to a block chain technology, and the target map can be stored in a block chain node.

The invention belongs to the technical field of face clustering, and particularly relates to a face clustering method based on a graph convolutional neural network. According to the method, firstly, feature extraction is performed on face data, face features are regarded as nodes, and local density values of all the nodes are calculated; then, the data is divided into high-density nodes and low-density nodes based on the local density value, and the high-density nodes are connected with the nodes which are high in density in the nearest neighbor to form a plurality of clustering centers; constructing a self-adaptive sub-graph for the low-density nodes, taking the self-adaptive sub-graph as the input of a graph convolutional neural network, and predicting the connectivity between the nodes; and finally, after the two parts are combined, edges which do not meet requirements are cut off by using pseudo-label propagation, and a final clustering result is obtained. According to the method, the data is divided into two parts based on the density, only the sub-graph is constructed for the low-density part for reasoning, the clustering efficiency is improved, meanwhile, the adaptive sub-graph can extract richer context information, reasoning on the sub-graph is more accurate, and the clustering accuracy is improved.

The invention discloses a multi-unmanned aerial vehicle cooperative tracking and positioning method. The method comprises the following steps of: establishing a directed connected graph based on a graph theory method according to an unmanned aerial vehicle communication network topological structure graph, and obtaining connection information of unmanned aerial vehicle sensor nodes and adjacent nodes; establishing a linear continuous time system model; estimating unknown information of a target by using an IKCF filtering algorithm; and carrying out data fusion on the estimation result of each node by adopting a sequential fast covariance cross fusion algorithm to obtain a determined target position. According to the multi-unmanned aerial vehicle cooperative tracking and positioning method of the invention, the estimation information of the adjacent unmanned aerial vehicle nodes is fully utilized, the real-time performance of the system is improved, and the target position estimation is more accurate.

The invention discloses an e-commerce network platform user community discovery method based on a graph convolutional neural network and similarity, and the method comprises the steps of building an e-commerce platform user network according to the user data of an e-commerce network, and calculating the similarity of nodes in the network based on a node similarity index; constructing a preferencenetwork according to the node similarity matrix; dividing initial communities through the connectivity of the preference network, and distinguishing strong communities according to the community scalecoefficient and the community sparsity coefficient; selecting the maximum node in the strong community as a leader node of the community and marking the leader node with a community label; and in combination with the advantages of the graph convolutional neural network in irregular graph data processing, using the graph convolutional neural network model to train the community label of the e-commerce network prediction node, and forming a final e-commerce network user community structure. According to the invention, the node similarity and the graph convolutional neural network are combined,so that the community discovery speed and quality are improved.

The invention discloses a method for simulating linear and nonlinear electrical properties of a random silver nanowire network. The method comprises the following steps: step 1, constructing a randomsilver nanowire network microcosmic model; 2, describing the microcosmic model by utilizing a graph theory idea and a double-node model to obtain directed graph matrix representation of the microcosmic model; 3, judging the connectivity of the nanowire network according to a connectivity algorithm of the graph; 4, abstracting the nanowire network directed graph into an electric network model by adopting an equivalent circuit method/electric network theory; and step 5, performing electrical simulation and analysis on the nanowire network circuit model by means of an HSPICE circuit simulation tool. According to the method, under the condition that the topological structure of the random silver nanowire network is kept unchanged, the resistance circuit model and the resistance-memristor hybrid circuit model are freely converted, accurate simulation of linear and nonlinear electrical behaviors of the same random silver nanowire network is achieved, and design and optimization of the randomsilver nanowire network in practical application can be guided.

The invention relates to a community discovery method based on a Louvain algorithm, computer equipment and a readable storage medium thereof. The method comprises the following steps: S1, generating agraph representing a network structure according to input data, wherein the graph comprises nodes and edges; S2, taking each node of the graph as an independent community; S3, performing inner-layercirculation and updating the affiliated community of each node; S4, repeating the step S3 until the percentage of the modularity change of the graph is less than a first threshold value and the current cycle index is an even number, or the inner-layer cycle index is greater than a second threshold value and the current cycle index is an even number, and ending the inner-layer cycle; S5, performing connectivity check on each community; if the communities are not connected, segmenting the communities into a plurality of connected sub-graphs, and taking each connected sub-graph as an independentcommunity; S6, compressing all the communities, and compressing each community into a node; and S7, inputting the result of the step S6 into the step S2, repeating the steps S3 to S6, and outputtingthe result until the modularity of the graph does not change any more or the percentage of the change is smaller than a third threshold.

The invention provides a cross connection line processing method and device for a quantum connected graph, a terminal and a storage medium, and the method comprises the steps: obtaining a quantum connected graph of a target quantum algorithm, the quantum connected graph comprising a plurality of graph nodes and a connection line between two graph nodes; judging whether a cross connection line exists in the quantum connection map or not; and when the cross connection line exists in the quantum connection map, determining an intermediary node in the quantum connection map, and retaining connection lines between the intermediary node and other map nodes to obtain a target quantum connection map without the cross connection line. The intermediary node in the quantum connection map is determined to optimize and eliminate the cross connection line, so that when the target quantum connection map without the cross connection line is realized on the quantum chip of the two-dimensional structure, the depth of the quantum algorithm can be effectively shortened, the number of times of applying the swap gate can be reduced in the realization process, and the implementation efficiency is improved. And the execution effect of the quantum algorithm is improved.

The invention discloses an electric power CPS structure communication vulnerability evaluation method based on a graph theory. The evaluation method comprises the following steps: 1, determining electric power CPS topological structure data and an interaction relationship between a power side and an information side; 2, identifying an initial risk source emergence node of the power side; 3, determining a maximum connected set in the power side network after the initial risk source node emerges; 4, removing nodes which do not belong to the maximum connected set in the power side; 5, removing the nodes losing the interaction relationship with the information side from the information side; 6, determining a maximum connected set in the information side network; 7, removing nodes which do notbelong to the maximum connected set in the information side; 8, rejecting the nodes losing the interaction relationship with the connected set in the power side; 9, outputting the number of failure nodes in the topological structure after risk fault transmission and the proportion of the number of failure nodes to the total number of initial power information nodes; and 10, evaluating the connectivity vulnerability of the electric power CPS topological network structure.

The invention discloses a recommendation method based on a graph convolutional network and an attention mechanism, and the method comprises the steps: constructing a user-article high-order connectivity interaction graph according to the interaction data of a user and an article; modeling according to high-order connection information of the high-order connectivity interaction graph, and generating a recommendation model; training a recommendation model according to the user and article interaction data; and after the training is completed, the recommendation model obtains an association score between the user and the article through the inner product, and finally, whether the article is recommended to the user or not is judged according to the association score. According to the method, embedded representation is learned in a user-article interaction bigraph, interaction information between a user and an article is considered in a model embedding layer, and high-order connectivity is modeled on a user-article interaction graph through embedding propagation, so that the model can learn high-dimensional feature information, more effective embedding is obtained, and the user-article interaction efficiency is improved. And the recommendation accuracy is improved.

The invention discloses a general hydropower station group topological graph generation method based on an improved graph theory algorithm and application, belongs to the field of hydropower energy optimization operation, and particularly relates to a hydropower station group topological relation generation method. The method comprises the steps of abstracting and generalizing hydropower station group topological relation analysis, analyzing a development mode of each hydropower station, a relation between dam sites and plant sites and a runoff relation between upstream and downstream hydropower stations, and generating a hydropower station group topological graph by adopting a weighted directed graph algorithm. The method aims at solving the problems that in runoff topological relation setting of an existing hydropower station group or a water conservancy and hydropower engineering group, expression is tedious, calculation efficiency is low, and universality is insufficient. A graph theory algorithm is introduced, the application range of the topological relation is further expanded, and the method can be widely applied to simulation operation scheduling, accessibility analysis, connectivity analysis, the shortest path considering cost and the like in a water conservancy and hydropower engineering system with numerous nodes.

The invention discloses a code attribute graph compression method and device for source code vulnerability detection. The method comprises the following steps of: calculating node neighborhood information gains based on first K-hop neighbors according to a code attribute graph; performing local normalization processing on the node neighborhood information gains; selecting nodes with low node neighborhood information gains after normalization processing to form a candidate deleted node set, judging whether cut points exist in the candidate deleted node set or not, and removing the cut points from the candidate deleted node set to finally obtain a deleted node set; and removing nodes in the deleted node set and edges connected with the nodes from the code attribute graph to obtain a code attribute compressed graph. According to the method, the first K-hop neighbor gain information of the nodes is calculated, the nodes with low gain information are selected and deleted, meanwhile, the connectivity of the compressed graph is guaranteed, the complexity is reduced under the condition that the node attributes and structural features of the code attribute graph are kept as much as possible, and therefore the space-time efficiency of subsequent model training is improved.

The invention provides a risk judgment method and device for relational network topology and a medium. The method comprises the following steps: obtaining a plurality of relational network topologies which are judged as risks, the relational network topologies being connected graphs formed by nodes and directed edges among the nodes; extracting common sub-topologies in the plurality of relational network topologies which are judged as risks; adding public elements of the nodes or the directed edges in the relation network topologies judged as risks to the nodes or the directed edges in the common sub-topologies, and the common sub-topology with the added nodes or the directed edges serving as risk mode sub-topologies; and if the relational network topology to be judged contains the risk mode sub-topology, judging that the relational network topology to be judged is at risk. According to the embodiment of the invention, the risk judgment of the relational network topology is more efficient and accurate.

The invention provides a group target clustering method based on an algebraic graph theory. Constructing a connected graph in the monitoring space by utilizing the algebraic graph theory and the relation information between the measurements; real-time grouping of the group targets is completed by searching the connected graph in the monitoring space, the limitation condition that the targets in the same group are required to be completely similar in a traditional method is broadened, and grouping errors caused by the fact that the similarity of the targets in the group is not completely 1 canbe avoided; in addition, the data dimension is not changed when the similarity and the clustering matrix are calculated, that is, the method can be used for clustering processing of the group targetson the premise of not changing an existing radar data processing structure.