193 results about "Graph Edge" patented technology

The invention discloses a cockpit active recommendation system and method based on a knowledge graph and semantic reasoning, and relates to the technical field of intelligent cockpits. The system receives natural language voice input of a user, executes voice recognition and semantic analysis, extracts user intention, keywords and slot entities, generates structured semantic information, constructs or calls a knowledge graph structure with semantic relation edges in combination with environment context information, and obtains the knowledge graph structure with the semantic relation edges. Semantic path reasoning is carried out based on the path dependence weight and the semantic similarity, a semantic edge label guided graph attention mechanism is introduced to calculate a path consistency score, a candidate recommendation set is generated, the semantic fitting degree and the path score are fused to sort and output recommendation content, and the graph edge weight and the user portrait are updated based on user feedback. According to the method, semantic understanding precision, recommendation path interpretability and system adaptive capacity are improved, and the method is suitable for personalized voice recommendation, man-machine interaction and scene linkage control tasks in an intelligent cockpit.

The invention relates to the technical field of big language models, and discloses a big language model training method based on knowledge graph enhancement, comprising the following steps: S1, constructing a multi-source heterogeneous knowledge graph; s2, coding the mixed attention heterogeneity map; s3, bidirectionally mapping a pre-training task; and S4, position specific gating fusion. According to the big language model training method and system based on knowledge graph enhancement, a same proton graph is established for a structured triple and text entity description, nodes are connected across graph edges to form a heterogeneous graph, associated edges are established through entity linking and syntactic analysis, multi-source knowledge is modeled in a unified mode, and the problem of low fusion efficiency is solved; mixed attention coding adopts a layering mechanism, a semantic level calculates weights according to type compatibility, a node level calculates similarity aggregation features through cosine distance and path length, entity vectors are generated through pooling, map structures and semantics are explicitly learned, reasoning accuracy is improved, and the problem of knowledge understanding superficial layer is solved.

The invention discloses a PHP taint type vulnerability detection method based on a heterogeneous graph neural network, and belongs to the field of software security. The method comprises the following steps: performing annotation removal, variable naming standardization and character string standardization processing on a PHP source code through a code preprocessing module to generate a standardized code; based on a vulnerability sub-attribute graph extraction module, reversely tracking vulnerability sinks to a taint source, extracting a simplified vulnerability sub-attribute graph, and removing redundant nodes and edges; fusing BERT semantic features and node type features through a graph node embedding module to generate an initial embedding vector, and constructing a heterogeneous graph comprising an abstract syntax tree edge, a program flow graph edge and a control dependence graph edge; a heterogeneous graph neural network vulnerability detection module is adopted to perform independent feature aggregation on multiple types of edges, dynamic weighted fusion is performed in combination with an attention mechanism, and key nodes are screened through Top-k graph pooling; and finally, inputting the graph-level features into a classifier to realize vulnerability detection.

The invention discloses a multi-layer graph multi-mode sentiment analysis method and device based on meta-information, and the method comprises the steps: firstly, extracting the high-level features of text, video and audio data through BERT, OpenFace 2.0 and Librosa, carrying out the time sequence coding through BiLSTM, and then mapping each mode feature to a unified space through MLP; then, respectively capturing modal invariant and specific features by using a shared multi-modal encoder and a private multi-modal encoder; for modal specific features, adopting GNN to construct a modal graph, and combining a grammar dependency tree and self-attention to calculate a graph edge weight so as to align meta-information in a modal; video and audio features are aligned with a text through Cross-Attention, a new multi-modal graph is constructed, and multi-level features are extracted; then, applying convolution, BiLSTM and GNN to extract local, time sequence and global features, and realizing multi-level feature fusion through a cross-modal attention mechanism; and after each modal feature graph and each modal invariant feature are input into a graph knowledge distillation module, dimension reduction processing is carried out, features are further fused through Transform and a multi-head attention mechanism, and finally an emotion prediction value is obtained.

The invention provides a multi-parameter automobile body flow field prediction method and device and a server, and relates to the technical field of vehicle fluid mechanics, and the method comprises the steps: carrying out the data preprocessing of a three-dimensional CAD model and a preset physical parameter set, and determining target input data; the method comprises the following steps: constructing a dynamic graph structure of a vehicle body flow field by performing dynamic graph edge convolution processing on target input data, and performing edge convolution processing on the dynamic graph structure to extract a local feature vector of the dynamic graph structure; and performing layer-by-layer aggregation processing and global feature aggregation processing on the local feature vectors, determining global feature vectors, and sending the global feature vectors to a preset full-connection network to predict various physical quantities in the automobile body flow field, and determining a target flow field parameter prediction result. According to the method, the expression ability of the model to a complex flow field can be enhanced based on dynamic graph edge convolution, so that the simulation efficiency and the prediction accuracy are remarkably improved.

The invention discloses an industry standard document-oriented deep semantic entity and relationship automatic extraction method, which relates to the technical field of document intelligent processing, and comprises the following steps of: inputting a document page image into a multi-modal document understanding model for processing to obtain a multi-modal document heterogeneous graph; processing the converter architecture model to obtain a semantic entity; generating a training data set by the multistage noise reduction neural network, and extracting a heterogeneous graph to obtain the heterogeneous graph; fusing the edge-oriented graph attention network model to obtain a relation result of heterogeneous graph entity recognition; and the co-reference resolution model processes a relation result identified by the heterogeneous graph, the semantic entity, the heterogeneous graph and the heterogeneous graph entity of the multi-modal document, and a global knowledge network is obtained by combining calculation of the trained link prediction model. According to the method, a purer and more reliable training data set is provided, and the extraction precision of the model is improved.

The invention provides a medical consumable management method and device based on an ant colony algorithm, and relates to the technical field of ant colony algorithms, and the method comprises the steps: constructing a medical consumable turnover frequency weight model, and generating a consumable turnover priority function; generating a goods allocation weight mapping matrix through an ant colony algorithm in combination with a priority function, wherein the mapping matrix is established according to the medical consumables and the access areas; simulating ant colony individuals to perform multi-path search in a hospital channel topological structure, and dynamically updating path pheromones to form a shortest path set; a storage space is constructed into a heterogeneous graph structure with storage locations as graph nodes and channels as graph edges, and an embedding result is fed back to an ant colony optimization algorithm as a goods allocation initial constraint through embedding attribute vectors of each node by a graph neural network; goods allocation initial constraint and a trafficability score and a blocking factor generated by camera identification data are introduced, and a consumable transportation path under an emergent task is dynamically adjusted and realized. According to the invention, full-life-cycle management of the medical consumables can be realized.

An attack path reasoning method based on attack technique and tactics scoring comprises the following steps: (1) malicious behavior traceability graph construction: constructing an original traceability graph based on system log data, detecting abnormal nodes, identifying attack techniques and tactics of the abnormal nodes, and retaining the abnormal nodes to form a malicious behavior traceability graph; (2) extracting an attack technology sequence mode: extracting an attack technology sequence from the threat intelligence, and constructing an attack technology sequence mode tree; and (3) scoring the edges of the malicious behavior traceability graph: scoring the edges in the malicious behavior traceability graph by comprehensively considering attack tactics and technologies. And (4) attack path reasoning based on the malicious behavior traceability graph: sampling of candidate attack paths is realized based on edge scores, scores of the candidate attack paths are calculated in combination with node feature information, and screening of the attack paths is realized. According to the method, attack path reasoning is carried out by comprehensively considering an execution sequence mode of attack techniques and tactics, so that more accurate attack path reasoning is realized; and the problem of alarm fatigue of the existing threat detection system is reduced.

The invention discloses an evaluation method for dynamically monitoring carbon sink of a grassland and wetland ecosystem, which relates to the field of ecological environment monitoring and comprises the following steps: acquiring historical carbon flux observation data of a target area; dividing the target area into a plurality of space grid units, taking each grid unit as a node in the graph structure, and establishing graph edge connection between adjacent nodes; extracting ecological features for each node; calculating the edge weight of each edge in the graph structure according to the ecological characteristic difference between the adjacent nodes; inputting the graph structure containing the node features and the edge weights into a graph neural network model, and training the graph neural network model by using historical carbon flux observation data as labels; and predicting the carbon flux of an unobserved node or a future time period node by using the trained graph neural network model. The method solves the problem that a traditional model is large in estimation error of the carbon flux in the ecological boundary region, and can remarkably improve the carbon exchange prediction precision of the boundary region.

The invention provides an intelligent data consanguinity tracking and visualization method based on graph calculation, and the method comprises the steps: carrying out the data structure analysis and metadata extraction of original data assets, so as to generate a standardized data asset package; performing graph node attribute definition on the data entities in the standardized data asset package, and performing graph edge attribute definition on the association relationship between the data entities to generate a first data blood relationship model; performing blood relationship path mining on graph nodes and graph edges in the first data blood relationship graph model to obtain a basic blood relationship path set, and performing quantitative calculation and feature labeling on path association strength in the basic blood relationship path set to generate a second data blood relationship graph model; and performing visual rule mapping on graph node attributes and graph edge attributes in the second data blood relationship graph model to construct a standardized visual data set and generate a data blood relationship visual interaction interface according to the standardized visual data set.

The invention relates to the technical field of infrared light machine system thermal deformation prediction, in particular to a multi-material structure thermally induced stress deformation prediction method based on a graph neural network. The method comprises the steps of data set establishment, graph structure establishment, graph neural network model establishment and training and model and parameter optimization. Finite element nodes correspond to graph nodes, finite element edges correspond to graph edges, an encoder-message passing-decoder architecture model is established, and node states are updated through a three-layer physical symmetry message passing mechanism. Physical constraint loss including minimum displacement smoothness constraint and stress continuity constraint is innovatively added into a loss function. Compared with traditional finite element calculation, the method has the advantages that the speed is increased by more than 100 times, high hardware adaptability is achieved, the black box limitation of a data-driven neural network model is broken through, thermally induced stress deformation analysis caused by different material coefficients can be processed, the adaptability to geometric changes is high, and good engineering application value is achieved.

The invention discloses a power distribution network topological structure identification method based on a Bayesian double aggregation graph neural network. The method comprises the following steps: abstracting equipment in a power distribution network as nodes, establishing a node set, collecting node voltage data, and forming an input feature vector; constructing a Bayesian double aggregation graph neural network; jointly training a Bayesian double-aggregation graph neural network by adopting cross entropy loss and a BNN regular term; and collecting real-time data to realize real-time identification of the topological structure of the power distribution network. According to the method, the Bayesian double aggregation graph neural network is adopted to accurately identify the topological structure of the power distribution network in real time, and only node voltage data of a single time section is depended on, so that the complexity of data acquisition is greatly reduced; meanwhile, global and local complex feature information among nodes in the power distribution network is effectively captured through a dual aggregation strategy of a global graph convolution network and local dynamic graph edge convolution.

The invention discloses a multi-terminal collaborative modular power operation and maintenance method and system based on digital twinning, and the method comprises the following steps: constructing a global power topological graph of a power energy site, mapping power equipment, sensors and connecting lines into graph nodes and graph edges, and binding the static attributes of the equipment with the topological structure; and acquiring real-time operation data of the power equipment, and distributing a unified time label for the multi-source heterogeneous data. According to the method, a unified twinborn model of cross-voltage class, cross-region and cross-equipment type is established in a digital twinborn platform, and operation data of different equipment such as photovoltaic equipment, battery energy storage equipment, an inverter, a protection device, an electric energy meter and temperature control equipment are expressed in a unified data semantic mode; according to the invention, the system can still automatically establish a consistent device portrait under the conditions of a large number of multi-terminal devices and diversified communication protocols, and the multi-terminal cooperative computing capability is significantly improved.

The invention belongs to the technical field of engineering simulation data processing, and particularly discloses a finite element grid graph structure construction method and system, a terminal and a medium. Comprising the following steps: analyzing original full-amount grid data exported by finite element simulation, segmenting unstructured grid data into a grid vertex coordinate set and a grid unit mark number set, and constructing a graph edge topological structure corresponding to a grid based on a unit mark number relationship; on the basis, loading physical field data and adopting a tolerance-based coordinate matching algorithm to realize accurate mapping of physical field labels and material attributes with grid nodes; and generating standardized finite element grid graph data which can be directly used for graph neural network processing. By means of the method, high-consistency and high-physical-reliability conversion from the finite element simulation data to the graph structure data is achieved, and the physical field modeling and simulation acceleration capacity based on graph learning is improved.

The invention relates to the technical field of network security, in particular to a network security threat perception recognition response method based on a security knowledge graph, and the method comprises the steps: obtaining data; determining temporary nodes and concerned nodes; determining a marked node; determining risk nodes and suspicious nodes; obtaining a corrected risk node; connecting map edges; determining an abnormal node; adjusting the threshold value and giving an alarm. According to the method, various key indexes are obtained in real time, marked nodes are determined according to the flow change rate in combination with the inlet and outlet flow ratio and the source I P address, risk nodes and suspicious nodes are further determined, the suspicious nodes are corrected through the inlet and outlet flow ratio, map edges are connected, abnormal nodes are determined, and then the threshold value is dynamically adjusted. And finally, giving an alarm based on the abnormal node, and optimizing the judgment standard in real time according to the change of the network environment, thereby effectively solving the problems of low danger identification accuracy and slow response speed caused by overlarge data volume and excessive dependence on the model.

The invention discloses a graph structure-based domain penetration attack path generation method, which belongs to the technical field of network security, and comprises the following steps of: constructing a multi-level relation graph comprising a host node, a service node and a user node through automatic detection by taking any host in a domain as a starting point; assigning a weight attribute to the atlas edge based on a vulnerability library and an attack pattern library; an improved heuristic graph search algorithm is adopted, all feasible attack paths and threat scores thereof are generated by integrating the path length, attack difficulty and permission improvement effect, and the problem that the threat scores of all the feasible attack paths are influenced in various scenes and dynamic change domain environments is solved. The technical problems of realizing comprehensive automatic penetration testing, accurately identifying potential attack paths and establishing a systematic threat assessment mechanism are solved, the automation, intelligence and high-efficiency level of domain penetration testing is remarkably improved, and the method has good adaptability, expansibility and practical value and is suitable for popularization and application. And attack path discovery and risk early warning work in a dynamic network environment with high security requirements can be effectively supported.

The invention provides a code knowledge graph generation method and device, a code generation method and device, equipment and a medium, and relates to the field of data processing, in particular to the technical field of intelligent search, man-machine interaction, artificial intelligence and large models. According to the specific implementation scheme, the method comprises the steps that a code tree is obtained, wherein the code tree comprises first code elements and the structural relation between the first code elements; wherein the code tree is generated by performing content analysis on a source code; generating first map node information according to the first code element; according to the structural relationship, generating first map edge information between the corresponding map nodes; and generating a code knowledge graph according to the first graph node information and the first graph edge information. According to the embodiment of the invention, the integrity of the knowledge graph can be improved, the storage capacity and the data processing capacity of the knowledge graph are reduced, and the knowledge graph construction efficiency is improved.

The invention provides a reef limestone crack connectivity prediction method and device in combination with a graph neural network, and the method comprises the following steps: S1, taking the intersection points of reef limestone crack line segments as graph nodes, and recording the three-dimensional space coordinates of each node; taking actual crack line segments between adjacent nodes as edges of the graph, and endowing each edge with an edge feature vector; the edge feature vector at least comprises a crack trend azimuth angle, a crack section length, a crack section curvature and a minimum included angle formed by the crack section and all adjacent crack sections; s2, constructing an adjacent matrix and a node feature matrix, and constructing structured graph data for describing the geometrical morphology and topological connection of the crack network in combination with the edge feature vectors; and S3, inputting the structured graph data into the graph neural network to obtain a connected probability graph. A reef limestone fracture network is converted into a graph structure and a graph neural network, and collaborative breakthrough of precision and efficiency of seepage capacity evaluation and decision reliability is achieved.

The invention discloses a regional public opinion propagation mode mining method based on a heterogeneous graph neural network. The method comprises the following specific steps: S1, constructing a heterogeneous graph; s2, weight definition of edge semantics; s3, node feature propagation and updating are carried out based on the graph neural network, and node embedding representation is obtained; and S4, node embedding is clustered, and regional public opinion propagation mode recognition is carried out in an embedding space. According to the method, joint modeling of users, media and regional nodes in a multi-relation heterogeneous graph structure is realized, and cross-semantic-edge information aggregation is completed by using a multi-relation attention mechanism, so that node embedding representation capable of representing multi-level propagation characteristics is obtained; the expressions provide basic support for subsequent public opinion propagation link identification, diffusion range prediction and opinion leader mining.

The invention discloses a bridge structure risk safety identification method based on artificial intelligence, and relates to the technical field of bridge structure risk identification, and the method comprises the steps: building a topological graph model which reflects the geometric and mechanical connection relation of a bridge, taking piers, main beam sections and supports as graph nodes, and taking the physical connection between components as graph edges; structure response signals and environment load parameters in the service period of the bridge are collected, and the collected data are synchronized according to time and then mapped to corresponding nodes and edges in the topological graph model; based on the mapped data and a topological graph model, establishing a graph neural network embedded with structural dynamic constraints, and outputting the risk probability of a component corresponding to each node by jointly optimizing the consistency of a monitoring data fitting error and a physical rule; and performing causal relationship analysis on the structural response signal and the environmental load parameter, identifying a causal path between environmental interference and structural abnormality, and separating an abnormal component caused by structural degradation from the original response according to an identification result.

A computer-implemented method comprising: generating a main graph based on a coarse mesh, the coarse mesh comprising a plurality of mesh elements arranged to fill an area or volume defined by a geometry; generating a vertex graph based on the geometry, wherein the vertex graph comprises vertex graph nodes corresponding to vertices of the area or volume defined by the geometry and vertex graph edges corresponding to links between the vertices in the geometry; generating, using a first graph neural network (GNN), an embedding of the vertex graph; generating, using a second GNN and based on the main graph and the embedding of the vertex graph, a prediction indicative of a refinement of the coarse mesh for generating a refined mesh.

The invention discloses a graph-driven self-attention compressible memory management method, and belongs to the technical field of penetration testing. According to the graph-driven self-attention compressible memory management method, an attention graph is constructed, semantic dependence and attention flow directions among multiple Agent nodes are accurately described by utilizing edge weights, the relevance and the accessibility of information are improved, and in the aspects of screening and loading, the expandability of the memory is improved. According to the method, key memory fragments are screened on the basis of concerned graph edge weights and task related features, only the loaded information is loaded, redundancy is effectively eliminated, the data volume processed by a system is reduced, and the system operation efficiency is improved; the semantic compression module can perform abstract processing on historical information based on various elements, and on the premise of ensuring that key semantic information is not lost, the abstract length is controlled within the window limit, so that the model can normally process information, and the adaptability and stability of the system to different data volumes are improved.

The invention provides a tunnel grouting fracture network generation and risk dynamic assessment method and system, and belongs to the technical field of tunnel engineering grouting reinforcement and numerical simulation. Comprising the following steps: performing fracture topological parameter extraction on acquired multi-modal data in a tunnel by adopting a three-dimensional semantic segmentation method to construct a fracture feature database; the fracture statistical features in the fracture feature database are input into the generative adversarial network to generate a tunnel grouting fracture network; based on the tunnel grouting fracture network, graph nodes are selected, and graph edges are constructed; a graph neural network is utilized to learn a connection relation between graph nodes to obtain a fracture connectivity coefficient; constructing a Bayesian network according to the fracture connectivity coefficient; and based on the Bayesian network, constructing a continuous three-dimensional risk field distribution map to realize risk dynamic assessment. According to the method, the three-dimensional random fracture model conforming to the special geological statistical characteristics can be constructed, and minute-level accurate evaluation of the grouting risk probability in the tunnel under the geology is realized.

The invention relates to the technical field of semiconductor manufacturing, and particularly provides a mask pattern edge compensation and correction method, which comprises the following steps of: constructing a training data set; training the deep learning model based on the training data set; obtaining a target mask plate design graph, and determining an open circuit position coordinate set of the target mask plate design graph based on optical proximity effect correction simulation of a historical defect database and / or a photoetching imaging model; respectively inputting the target mask design graph, the open circuit position coordinate set of the target mask design graph, photoetching machine model parameters and optical model physical constraint parameters into an open circuit analysis model, and outputting an open circuit probability distribution graph and a graph edge correction rule of the target mask design graph; and inputting the target mask design graph, the open circuit probability distribution graph and the graph edge correction rule into a photoetching imaging model to generate a corrected mask design graph. According to the method, the limitation of traditional correction is broken through, the mask correction precision is remarkably improved, and the circuit breaking risk in chip production is reduced.

The disclosure concerns a machine-learning method. The method comprises obtaining a training dataset of B-rep graphs. Each B-rep graph represents a respective B-rep. Each B-rep graph comprises graph nodes each representing an edge, a face or a co-edge of the respective B-rep and being associated with one or more geometrical and / or topological features. Each B-rep graph further comprises graph edges each between a respective first graph node representing a respective co-edge and a respective second graph node representing a face, an edge, an adjacent co-edge, or a mating co-edge associated with the respective co-edge. The method further comprises learning, based on the training dataset, a local Deep CAD neural network. The local Deep CAD neural network is configured to take as input a B-rep graph and to output, for each graph node of the input B-rep graph, a local topological signature of the B-rep element represented by the graph node.

The invention is suitable for the technical field of data analysis, and provides an intelligent decision graph construction method based on dynamic time sequence event data, which comprises the following steps: reading and cleaning original event data, processing fields, expanding nested information through a structured analysis and entity alignment technology, and generating cleaned structured data; the frequency of different field combinations is calculated, a frequency feature column is generated in combination with a timestamp and a weight factor, and a multi-dimensional relation is combined; screening event pairs based on a dynamic frequency threshold, constructing a directed graph data structure, taking unique identifiers of the events as nodes, establishing edges of a graph, setting edge weights, and optimizing a time sequence relationship between the events through time sequence matching and a dynamic weighting mechanism; and outputting a graph containing the node number, the edge number and the graph edge data screened based on the frequency. According to the method, the accurate, efficient and intelligent graph construction method is realized, the technical progress in the field of dynamic event data analysis is promoted, and the method can be applied to an actual scene in which dynamic event data needs to be processed and an associated graph needs to be constructed.

The invention discloses a cloud platform access path optimization method based on a graph convolutional network, and the method comprises the following steps: S1, collecting user access behavior data and resource state data in a cloud platform, and constructing an original access data set; s2, constructing an access path graph; s3, graph structure features, access behavior features and resource state features are extracted, and node input vectors are generated; s4, inputting to an improved GraphSAGE model, and outputting a path scoring result; s5, identifying bottleneck nodes and high-risk path segments; s6, constructing an optimized path candidate set, and generating an optimal access path combination; and S7, collecting an actual access result of the user to construct access feedback data, updating graph edge attributes and graph nodes, and adjusting aggregation parameters and scoring weights to realize adaptive updating of the path scoring network. The resource scheduling efficiency of the cloud platform can be effectively improved, the access conflict risk is reduced, and the system response performance is enhanced.

The invention provides an AI-based low-code aid decision-making method and system, and the method comprises the steps: collecting an event stream of a user on a low-code platform, and constructing an event vector sequence and a behavior component mapping table; obtaining a low-code configuration graph to construct a behavior-structure fusion graph, generating a component complexity score table and determining a high-complexity component set; according to the component complexity score table and the high-complexity component set, constructing a local cause-effect sub-graph and generating a path semantic nested matrix, generating a high-risk cause-effect path set and endowing a path level risk score, and generating a sub-graph edge score matrix; determining an optimized suggested draft set and a suggested positioning node set by taking the consistency loss in the minimized path as a target function; and mapping the optimized suggested draft set and the suggested positioning node set to the low-code configuration graph for visual labeling, automatically executing configuration modification after structural constraint verification, and recording a change log to support rollback and version tracing.

The invention discloses a target abnormal movement early-warning method based on eye walking along with a hook, and the method comprises the following steps: collecting an eye movement data stream, constructing a fixation behavior data set, and generating a fixation track sequence; constructing the fixation points as graph nodes, generating graph edges according to a time sequence and spatial proximity, and forming a graph structure sequence; inputting to the improved ST-GCN model, and outputting a target prediction vector; identifying offset candidate segments; performing trajectory morphological analysis on the offset candidate segments, and judging whether formation conditions of a trajectory loopback structure are met or not; if not, calculating an access frequency domain; if the access frequency is greater than a preset frequency threshold, outputting a low early warning signal; and screening the high-weight fixation segment based on the offset candidate segment, carrying out similarity matching, and outputting formal early warning information according to a matching result. According to the invention, multi-level accurate early warning of the abnormal motion state of the target is realized, and the method has the advantages of clear structure, strong real-time performance, good adaptability and the like.

A machine-learning method including obtaining a training dataset of B-rep graphs. Each B-rep graph represents a respective B-rep. Each B-rep graph includes graph nodes each representing an edge, a face or a co-edge of the respective B-rep and being associated with one or more geometrical and / or topological features. Each B-rep graph further comprises graph edges each between a respective first graph node representing a respective co-edge and a respective second graph node representing a face, an edge, an adjacent co-edge, or a mating co-edge associated with the respective co-edge. The method further includes learning, based on the training dataset, a local Deep CAD neural network. The local Deep CAD neural network takes as input a B-rep graph and to output, for each graph node of the input B-rep graph, a local topological signature of the B-rep element represented by the graph node.