143 results about "Graph structured data" patented technology

A dynamic foldering system automatically manages the creation and deletion of run-time dynamic folders and selection of documents, items, or object graphs found in the run-time dynamic folders. The system comprises a query / predicate for a design-time folder group based on which dynamic folders are automatically managed, a variable binding mechanism, automatic identifications of dynamic folders containing a particular object, parallel navigation, and customized combination of foldering results. The query / predicate for a design-time folder group can be a parameterized query. The dynamic folder hierarchy is defined by a hierarchy of design-time folder groups defined by a query / predicate on a document comprising structured or semi-structured data. Modifiable criteria for creating the dynamic folder hierarchy are provided by the user. Variable binding provides powerful query / predicate definitions on hierarchical data or graph-structured data. Given a document, the system automatically identifies which dynamic folders contain the document. Parallel navigation allows a user to navigate along additional paths in a hierarchy and combine the navigation results using set operations.

A recommendation system based on graph convolution technology comprises a preprocessing module, a heterogeneous graph generation module, a model training module and a recommendation result generationmodule, wherein, the preprocessing module cleans the interaction records of the user and the article and performs the data cleaning and the format standardization operation, and generates the interaction sequence for each user and outputs the interaction sequence to the heterogeneous graph generation module; The heterogeneous graph generation module constructs three heterogeneous graphs representing user preferences, dependencies among items and similarities among users according to user interaction sequence data, and outputs the generated graph structure data to the model training module. Themodel training module trains the graph convolution model based on graph structure data and generates vector representation for each user and object. The recommendation result generation module calculates the user's preference for all items according to the vector expression, and generates the final recommendation result. The invention solves the problem that the number of the neighbors of each node is not equal, and the information of the neighbors of the nodes in the heterogeneous graph is mined by the convolution operation, so that the recommendation effect is improved.

Arrangements and methods for providing for the efficient implementation of ranked keyword searches on graph-structured data. Since it is difficult to directly build indexes for general schemaless graphs, conventional techniques highly rely on graph traversal in running time. The previous lack of more knowledge about graphs also resulted in great difficulties in applying pruning techniques. To address these problems, there is introduced herein a new scoring function while the block is used as an intermediate access level; the result is an opportunity to create sophisticated indexes for keyword search. Also proposed herein is a cost-balanced expansion algorithm to conduct a backward search, which provides a good theoretical guarantee in terms of the search cost.

The invention relates to a method of visualizing hierarchical graph structure data, a device thereof and a computer program product thereof. The hierarchical graph structure data comprises a plurality of nodes distributed on a plurality of hierarchies and the relations among a plurality of nodes. The method comprises presenting the thinning of the hierarchical graph structure data on the nodes of an appointed hierarchy and the relation thereof in a focus region (411), and presenting the nodes under the appointed hierarchy of the hierarchical graph structure data and the relation thereof on a subgraph region (412). Therefore, the nodes of the appointed hierarchy of the hierarchical graph structure data and the relation thereof can be clearly and visually displayed even if the hierarchy of the hierarchical graph structure data and or graph information is much, and the specific detailed information of the hierarchical graph structure data can be dynamically displayed according to the requirements of users.

The invention provides a traffic flow prediction system and method, and a model training method. The traffic flow prediction system comprises a first space-time convolution network, a second space-time convolution network, and an output layer, wherein the first space-time convolution network is used for generating a first space-time feature vector according to the graph structure data; the secondspace-time convolution network is used for generating a second space-time feature vector according to the graph structure data and the first space-time feature vector; and the output layer is used forgenerating a prediction value according to the second spatial-temporal feature vector. In the present invention, the traffic flow prediction system processes the graph structure data; and on the premise of ensuring the time attribute of the traffic data, the graph structure data gives the spatial attribute to the traffic data, and the spatial-temporal dependence relationship is obtained by superposing the two spatial-temporal convolutional networks, so that the traffic flow prediction system provided by the invention truly realizes prediction of the traffic flow by combining the spatial feature and the temporal feature of the traffic data.

A graph generating device which can layout common nodes so as to have the same positional relationship as positional relationship in a graph before updating when content of graph structure data is updated, is provided. In a storage device 30, first graph structure data and layout original data with respect to the first graph structure data are stored. When second graph structure data obtained by updating content of the first graph structure data is input, data updating and registering means 51 stores the second graph structure data in the storage device 30. When a graph corresponding to the second graph structure data is generated, layout position calculating means 52 specifies common nodes described common to the second graph structure data and the first graph structure data with respect to the second graph structure data and performs layout of the graph under the conditions that concerning each of the common nodes, relative positional relationship defined by the layout original data with respect to the first graph structure data is maintained.

The efficiency of an AGM algorithm is further improved. For an AGM algorithm that can efficiently extract, from a graph database including graph structured data, graph (frequent graph) data having a support level equal to or greater than the minimum support level, a function "relabe" for ordering the vertex labels and edge labels of the graph is executed (step 1). Further, for a function "Newjoin", for employing a set Fk of adjacency matrixes that represent a size k frequent graph, for generating a set Ck+1 of adjacency matrixes, which represent a size k+1 candidate frequent graph, a fourth condition for coupling a first generator matrix to a second generator matrix is added to the three conditions of the AGM algorithm only when the first generator matrix is a canonical form.

The invention provides a graph neural network training method, graph neural network equipment and device and a medium. The graph neural network training method comprises the steps: obtaining graph structure data of a graph neural network, the graph structure data comprising representation vectors of nodes and representation vectors of edges, and the edges representing association relationships between connected nodes; utilizing the graph neural network to generate a transfer matrix based on the characterization vector of the edge, the transfer matrix representing an information transfer mode between connected nodes; determining mutual information between the representation vector of the edge and the transfer matrix; and training the graph neural network by using the mutual information.

The efficiency of an AGM algorithm is further improved. For an AGM algorithm that can efficiently extract, from a graph database including graph structured data, graph (frequent graph) data having a support level equal to or greater than the minimum support level, a function “relabel” for ordering the vertex labels and edge labels of the graph is executed (step 1). Further, for a function “Newjoin”, for employing a set Fk of adjacency matrixes that represent a size k frequent graph, for generating a set Ck+1 of adjacency matrixes, which represent a size k+1 candidate frequent graph, a fourth condition for coupling a first generator matrix to a second generator matrix is added to the three conditions of the AGM algorithm only when the first generator matrix is a canonical form.

The invention discloses a method, device and system for processing massive data of a graph structure. Data computational efficiency can be improved, and operational reliability of a system can be improved. The method for processing the massive data of the graph structure comprises the steps of using a slave node to read the graph structure data to a memory; preprocessing the graph structure data in the memory to obtain at least one data piece in the graph structure data, wherein every two adjacent vertexes of the data piece are arranged on the same data piece; mapping the data piece obtained through preprocessing to the slave node; using the slave node to compute the data piece which is mapped to the slave node through an iterative algorithm. Due to the fact that information of every two adjacent vertexes is mapped to the same slave node instead of different slave nodes, when the slave node computes the data piece which is mapped to the slave node through the iterative algorithm, the slave node does not need to be in communication with other slave nodes, the traffic is reduced, the requirement for bandwidth resources in a cluster is reduced, and computational efficiency of a cluster system and computational efficiency of nodes of the cluster system are improved.

Arrangements and methods for providing for the efficient implementation of ranked keyword searches on graph-structured data. Since it is difficult to directly build indexes for general schemaless graphs, conventional techniques highly rely on graph traversal in running time. The previous lack of more knowledge about graphs also resulted in great difficulties in applying pruning techniques. To address these problems, there is introduced herein a new scoring function while the block is used as an intermediate access level; the result is an opportunity to create sophisticated indexes for keyword search. Also proposed herein is a cost-balanced expansion algorithm to conduct a backward search, which provides a good theoretical guarantee in terms of the search cost.

The invention discloses a fault positioning method based on a finite-state machine and a graph neural network. The method comprises the steps that fault information is converted into graph structure data, reasoning is conducted on the graph structure data through a graph neural network model, conversion is achieved through a finite-state machine model, and the accuracy, reliability and timelinessof fault diagnosis in the industrial process can be remarkably improved.

The invention provides a graph embedding method and device, and a storage medium, and the method comprises the steps: reading graph structure data and node feature values in a target graph, and building a graph structure model; regarding each node in the graph structure model as a target node, and sampling a first-order neighbor node of each target node according to the non-uniform neighbor node sampling function to obtain a first-order neighborhood of each target node; constructing second-order neighborhoods of the target nodes according to the first-order neighborhoods of the target nodes, aggregating the second-order neighborhoods to the first-order neighborhoods corresponding to the target nodes, and inputting the aggregated features of the second-order neighborhoods into the fully-connected neural network to obtain new features of the first-order neighborhoods of the target nodes; and aggregating the new features to the corresponding target nodes, and inputting the aggregated newfeatures of the first-order neighborhood into the fully-connected neural network to obtain output features of the target nodes. The neighborhood can be flexibly and effectively constructed for each node in the graph, and feature aggregation can be rapidly carried out, so that the graph embedding effect based on the graph neural network is improved.

The invention discloses a visual analysis method for understanding a graph convolutional neural network. The method comprises the following steps of step 1, dividing an input graph structure data setinto a training set, a verification set and a test set; step 2, defining a parameter set, wherein the parameter set comprises a hidden layer number set and a hidden neuron number set; step 3, based onthe defined parameter set, training to obtain a series of graph convolutional neural network models; step 4, designing a hidden layer analysis view, and displaying the influence of hidden layer parameters on the classification accuracy; step 5, designing a loss and accuracy view, and displaying the change of the loss and classification accuracy in the iterative training process of the model; andstep 6, calculating the positions of the nodes in the graph by adopting a GraphTSNE visualization method, designing a graph layout view, and presenting the prediction condition of the nodes under different training step numbers and the difference of the prediction condition of the nodes between the two training step numbers.

The invention discloses an urban area tail gas pollution prediction method, which belongs to the technical field of environment detection, and comprises the steps of creating a training sample set byusing historical tail gas space-time distribution diagram sequence data and external environment feature data; training the deep space-time diagram convolutional network model by using the training sample set to obtain a deep space-time diagram convolutional network prediction model; and based on the deep space-time diagram convolutional network prediction model, predicting the regional exhaust emission at the moment t by using the external environment feature data at the current moment t and the historical exhaust space-time distribution diagram sequence data before the moment t-1. Accordingto the method, the exhaust space-time distribution is constructed into the graph structure data by utilizing the road network connectivity data, so that the regional exhaust emission prediction problem is converted into the space-time graph sequence prediction problem, and the regional exhaust prediction with higher precision can be realized on the real telemetry data by adopting the deep space-time graph convolution model.

The invention relates to a graph-convolutional-neural-network-based auxiliary diagnosis method for the Alzheimer's disease. The method comprises the following steps: processing a brain function magnetic resonance image to obtain a time sequence of each brain region; calculating a Pearson correlation coefficient between any two time sequences in the time sequences of the brain regions to obtain a brain function connection network using each time sequence as a node and the Pearson correlation coefficient as a weight of a connection edge between the two nodes; removing all edges with the weightssmaller than a set threshold and simplifying the brain function connection network to obtain graph structure data; and designing a graph convolutional neural network model, training the designed graphconvolutional neural network model by using the graph structure data, using a training result with the best performance in a verification set as an auxiliary diagnosis model, and outputting a diseasestate corresponding to the whole graph structure. Therefore, compared with the traditional method, the method has the advantages that the detection accuracy is higher, and the advanced auxiliary diagnosis classification level on the disease is obtained.

The embodiment of the invention discloses an image region segmentation model training method and device based on artificial intelligence, and an image region segmentation method and device based on artificial intelligence. In the model training process, a sample image set comprising at least one sample image is acquired, wherein the sample image has first annotation information, and the first annotation information can be annotation information with large granularity such as an image level. For each sample image in the sample image set, graph structure data corresponding to the sample image isgenerated, and each vertex in the graph structure data comprises at least one pixel point in the sample image. Second annotation information of the vertex is determined according to the graph structure data and the first annotation information through a graph convolution network model, wherein the granularity of the second annotation information is smaller than the granularity of the first annotation information. Since the vertexes are actually superpixel points and the second annotation information is superpixel-level annotation, in the training process, due to intervention of pixel-level annotation, strong supervision can be achieved, and the accuracy of the model is improved, and then the accuracy of image segmentation is improved.

The invention discloses an intelligent contract reentrancy vulnerability detection method based on a graph neural network, which realizes intelligent contract reentrancy vulnerability detection through optimized graph neural network training and learning, and specifically comprises the following steps: collecting an intelligent contract source code data set; extracting and constructing a corresponding graph structure model by the smart contract source code; standardizing the graph structure; constructing a graph neural network of the time message flow; and inputting the standardized intelligent contract graph structure data set, training an intelligent contract reentrant vulnerability detection model, and realizing intelligent contract reentrant vulnerability detection through the model. The static source code is converted into a message flow graph structure with a time sequence, reentrant and non-reentrant labels can be automatically output through the training model, the vulnerability detection accuracy is improved, a new method thought is provided for intelligent contract vulnerability detection, and the method has good practical value.

The invention discloses a graph neural network (GNN)-based emotional tendency prediction method, device and system, and a computer storage medium. The graph neural network-based emotional tendency prediction method comprises the steps of firstly, obtaining a segment of text information; then, converting the text information into the graphic structure data taking the words as the nodes and taking the similarity among the words as edges; converting the graphic structure data into an adjacent matrix formed by the similarity between words; and then performing emotional tendency prediction on the adjacent matrix through an emotional classification model. According to the emotion tendency prediction method, when a long text is processed, the text information is converted into the graphic structure data through the similarity between words, so that on one hand, the advantages of a graphic neural network can be utilized to contain the semantics as many as possible, and on the other hand, by aggregating the similar words together, the emotional tendency characteristics with the most representative significance can be extracted more easily, and the prediction result is more accurate.

The invention discloses a relation discovery method and device and a computer readable storage medium, and the method comprises the steps: converting event data into personnel relation data, the personnel relation data comprising n sets, and each set comprising m * (m- 1) / 2 subsets, each subset comprising two persons in an event, n being the number of events, and m being the number of persons in each set; Converting the personnel relationship data into graph structure data G = (V, E), a set V of vertexes in the graph G being all personnel in the n sets, and a set E of sides being a connectionline between two personnel belonging to one subset; And in the graph structure data, mining a frequent subgraph with the frequency meeting a preset minimum frequency threshold value, and obtaining a partner relationship among the personnel according to the mined frequent subgraph. According to the method, the event data is converted into the graph structure data, the frequent subgraphs are mined according to the minimum frequency threshold value, the granularity of relation discovery is better adjusted, and therefore the hidden relation in the event data is more accurately obtained.

The invention provides a similar medical record searching method, device and equipment and a readable storage medium. The method comprises the steps: obtaining query medical record data and multiple pieces of historical medical record data; obtaining query graph structure data corresponding to the query medical record data and historical graph structure data corresponding to the historical medicalrecord data, wherein the query graph structure data and the historical graph structure data both comprise first-class sub-graphs and second-class sub-graphs, and intermediate nodes and leaf nodes ofthe second-class sub-graphs are obtained by conducting feature recognition on the first-class sub-graphs; according to the root node similarity, the first type of sub-graph similarity and the second type of sub-graph similarity, obtaining the similarity degree of each historical graph structure data and the query graph structure data; according to a preset selection rule and the similarity degree,determining a similar medical record search result for querying the medical record data, so that inherent and recognizable sub-graphs in the medical record data are extracted, the relevance of corresponding sub-graph contents is measured in comparison, and the similar medical record searching accuracy is improved.

The invention discloses a data processing method and device based on a graph convolutional neural network, equipment and a medium. The method comprises the steps of obtaining first graph structure data to be processed and inputting the first graph structure data to the pre-trained graph convolutional neural network; calling the graph convolutional neural network to perform neighbor node sampling operation and neighborhood information aggregation operation on each first node in the first graph structure data so as to update each first node; and calling the graph convolutional neural network toobtain a processing result of the first graph structure data based on the updated first nodes. According to the method, the difference of the relationship between the nodes is mined by the graph convolutional neural network, so that a more accurate processing result can be made according to mined more sufficient information.

The invention provides a relational reinforcement learning system and method based on graph structure data, and the system encodes a real world image and an unmatched high-resolution image into degradation representations, and then processes the positive and negative sample relationship between the representations through comparative learning, thereby obtaining an encoder capable of correctly extracting image degradation features. The method is used by a subsequent down-sampling network. A low-resolution image obtained by processing a high-resolution image through a linear down-sampling network has degradation characteristics of a real world, and a finally generated data set is input into a reconstruction network for training to complete an image super-division task. Compared with an existing super-division algorithm based on fuzzy kernel estimation, the method has the advantages that implicit expression between images can be learned without introducing extra degradation models and prior knowledge, and better generalization is achieved while the same performance as a supervised method is achieved.

A data processing apparatus according to an embodiment includes a data acquisition unit, a setting unit, and a reinforcement learning unit. The data acquisition unit acquires graph-structured data describing a connection relation between nodes. The setting unit sets a first network representing the graph-structured data acquired by the data acquisition unit. The reinforcement learning unit derives a parameter of the first network such that a feature quantity of an output layer of an evaluation target node in the first network approaches a reward and a feature quantity of an output layer of an operation node becomes a feature quantity causing the feature quantity of the output layer of the evaluation target node to approach the reward.

The invention discloses a cellular network fault diagnosis method. The method comprises the following steps: the step 1, determining a network fault data set; the step 2, obtaining a network fault data set after dimension reduction; the step 3, representing the network fault data set after dimension reduction in the step 2 in the form of a feature matrix, expressing label information of the network fault data set after dimension reduction in the step 2 in a label matrix form, and converting the introduced weight matrix into an adjacent matrix with matrix elements only being 0 and 1; and the step 4, carrying out fault diagnosis based on the graph convolutional neural network. According to the novel cellular network fault diagnosis method, intelligent fault diagnosis of a heterogeneous wireless network is deeply studied, similar characteristics among samples are analyzed in combination with a big data processing method, an existing network fault parameter data set is converted into graph structure data, and features are extracted from the graph structure data by using a graph convolutional neural network, so that the classification task of the sample nodes is completed, and the fault type of the cell is predicted.

The invention provides a group behavior identification method based on channel information fusion and group relationship spatial structural modeling. The method comprises the following steps: firstly,segmenting a to-be-identified video, sampling a plurality of frames at equal intervals, and extracting fusion features containing space-time and motion information through an improved STM network module; performing intra-frame region division and high-dimensional mapping on the fusion feature of each frame to form graph structure data; and finally, through a graph convolution-LSTM network containing a core group relationship evolution model, integrating a global behavior discrimination feature and a local behavior discrimination feature as a group behavior descriptor to discriminate behaviorclassification, and obtaining a final behavior label through softmax. According to the scheme, a channel selection module is added to fuse space and motion features, so that feature representation containing space and motion information is extracted at the same time, and the relevance of the features is enhanced; the spatial structural modeling of the group relationship is combined to ensure the integrity and comprehensiveness of the extracted spatiotemporal information features, and the key object of the group interaction relationship which plays a decisive role in behavior discrimination isemphatically considered, so that the recognition precision can be effectively improved.

The invention discloses a dominant instability mode recognition model construction and application method based on a graph neural network and belongs to the field of power system stability judgment. The weighted graph structure constructed by the method can better reflect the topology of the power grid. Before model training, a corresponding map structure is pre-constructed according to the powergrid topology of the sample set; in the training process, the effect of the graph structure is equivalent to the effect of converting original matrix type data into graph structure data, and a test result shows that compared with a convolutional neural network method without considering topology, the graph neural network method considering the topological structure of the power grid has higher discrimination precision; according to the method, weighting processing is performed on the graph structure formed according to the power grid topology by adopting the admittance of each transmission line, so the topological information of the power grid can be further enhanced, the model is enabled to better fit the special graph structure of the power grid, the power grid topological characteristics contained in the input graph structure are enhanced, and the model is enabled to have better judgment accuracy.

The invention discloses a physical circuit diagram recognition method based on deep learning and an application thereof. The method comprises the steps of obtaining an image of a to-be-recognized physical circuit diagram and performing image enhancement processing on the image of the to-be-recognized physical circuit diagram; identifying the binary image by using the trained component identification neural network model to obtain all components of the to-be-identified physical circuit diagram, each component corresponding to an identification ID and a component name; generating graph structuredata corresponding to the to-be-identified physical circuit diagram, wherein the Graph structure data comprises a vertex set and an edge set, the vertex set is an intersection set of component connecting lines, and the edge set is a connecting line set between vertexes; performing component detection and Graph simplification on the generated Graph structure data to output an associated componentsequence, the associated component sequence comprising a component connection type and a component ID, and calculating a physical attribute of a target component by using the associated component sequence to achieve classification and identification of all circuit components of the circuit diagram; and extracting the connection relationship among the components.