1451 results about "Graph neural networks" patented technology

The invention discloses a graph neural network traffic flow prediction method and system based on an attention mechanism. The method comprises the steps of obtaining to-be-predicted urban traffic flowdata; constructing a road network map according to the road connection relationship; preprocessing the urban traffic flow data to be predicted; and inputting the road network map and the preprocessedresult into a pre-trained neural network based on an attention mechanism, and finally outputting a prediction result of the urban traffic flow. Roads and checkpoints are encoded according to the roadnetwork information, a road network graph structure is established according to the upstream and downstream relationship of the roads, vehicle passing data of the checkpoints is counted under different time dimensions nd summarized to form a road network traffic flow data table; a graph neural network formed by stacking multiple layers of attention modules is constructed, a time sequence attention mechanism and the graph attention network are used for modeling the traffic flow in the whole road network, and the future traffic flow condition of a specified checkpoint is predicted.

The invention relates to a text data-oriented threat intelligence knowledge graph construction method. The functions of automatically extracting the key information from the text threat intelligence data and constructing the threat intelligence knowledge graph are realized. The invention provides a threat intelligence knowledge graph construction method for text data. The method comprises the following steps: defining an ontology structure in the threat intelligence field; using a threat intelligence named entity recognition model based on multiple factors and a threat intelligence entity relation extraction model based on a graph neural network to obtain threat intelligence entities and relation triples from text data, and finally storing information through a graph database to form a threat intelligence knowledge graph.

The invention discloses a session-based project recommendation method, device and apparatus and a computer readable storage medium. In the scheme of the invention, the method comprises the following steps of: constructing a directed session structure diagram through a historical session sequence of a user; based on the session structure diagram, enabling a graph neural network to capture conversion between adjacent projects; generating implicit state vectors of all nodes in the graph; establishing long-distance dependence by using a self-attention mechanism, and finally predicting the probability of next click by taking the linear combination of the global preference and the current local preference of a user as the hidden vector of a current session. The method utilizes the complementaryadvantages of the self-attention network and a graph neural network to realize the accurate recommendation of a project.

The invention discloses a knowledge graph entity semantic space embedding method based on graph second-order similarity, and the method comprises the steps: (1) inputting a knowledge graph data set and a maximum number of iterations; (2) calculating first-order and second-order similarity vector representations through first-order and second-order similarity feature embedding processing by considering a relation between entities through a graph attention mechanism to obtain first-order and second-order similarity semantic space embedding representations; (3) carrying out weighted summation onthe final first-order similarity vector and the final second-order similarity vector of the entity to obtain a final vector representation of the entity, inputting a translation model to calculate a loss value to obtain a graph attention network and a graph neural network residual, and iterating the network model; And (4) performing link prediction and classification test on the network model. According to the method, the relation between entities is mined by using a graph attention mechanism for the first time, and patents have a relatively good effect in the application fields of link prediction, classification and the like of the knowledge graph.

The invention provides a method for automatically generating Java code annotations based on Graph2Seq of deep reinforcement learning At the encoder end, codes are firstly converted into a form of a code graph, a graph structure is utilized to express structured information of the codes, and then a graph neural network is utilized to encode a program graph to obtain structured semantic informationof the codes. At the decoder end, the decoder improved based on the deep reinforcement learning framework is used for obtaining the natural language annotation content corresponding to the decoder, the problem of high deviation caused by the fact that no real sequence word guidance exists in the reasoning stage in a traditional method is solved, and therefore code annotation natural generation isaccurate and natural.

The present invention relates generally to object annotation, specifically to polygonal annotations of objects. Described are methods of annotating an object including steps of receiving an image depicting an object, generating a set of image features using a CNN encoder implemented on one or more computers, and producing a polygon object annotation via a recurrent decoder or a Graph Neural Network. The recurrent decoder may include a recurrent neural network, a graph neural network or a gated graph neural network. A system for annotating an object and a method of training an object annotation system are also described.

The invention discloses a visual dialogue generation method based on a context perceptual map neural network. The visual dialogue generation method comprises the following steps of 1, preprocessing the text input in a visual dialogue and constructing a word list; 2, extracting the features of a dialogue image and the features of a dialogue text; 3, obtaining a context feature vector of the historical dialogue; 4, constructing a context perceptual map; 5, iteratively updating the context perceptual map; 6, carrying out attention processing on the nodes of the context perceptual map based on a current problem; 7, performing multi-modal semantic fusion and decoding to generate an answer feature sequence; 8, generating the parameter optimization of a network model based on the visual dialogueof the context perceptual map neural network; 9, generating a prediction answer. According to the method, the context perceptual map neural network is constructed on the visual dialogue, and the implicit relationship between different objects in the image can be reasoned by using the text semantic information with finer granularity, so that the reasonability and accuracy of the answers generated by an intelligent agent for question prediction are improved.

A method for detecting malicious program behavior includes performing program verification based on system activity data, analyzing unverified program data identified from the program verification to detect abnormal events, including analyzing host-level events to detect abnormal host-level events by learning a program representation as a graph embedding through an attentional architecture based on an invariant graph between different system entities, generating detection results based on the analysis, and performing at least one corrective action based on the detection results.

The invention relates to a source code multi-tag graph neural network-based program code copying type detection method. The method comprises the following steps of S1, for a code text, generating a copying version by utilizing a self-defined code micro-confusion tool, and recording a copying type; S2, performing code attribute graph eigenvector extraction on the code text and the copying version;S3, integrating code attribute graph eigenvectors of the code text and the copying version, providing a good input for a neural network, and setting integrated code attribute graph eigenvectors of thecode text and the copying version as positive examples; S4, performing integration by utilizing the methods in the steps S2-S3 to obtain code attribute graph eigenvectors of the code text-the code text, and setting the code attribute graph eigenvectors of the code text-the code text as negative examples; and S5, defining a multi-task learning network model by utilizing the neural network, training 10 classifiers at the same time for each positive example/negative example, and finally outputting a 10-dimension vector, thereby providing a copying evidence for an assessor, wherein each dimensionrepresents a defined copying type.

The invention discloses a session recommendation method based on a space-time sequence diagram convolutional network. The method comprises the following steps: S1, modeling all session sequences intoa directed session graph; S2, constructing a global graph by taking common commodities in the session as links; S3, embedding an ARMA filter into a gated graph neural network, extracting a topologicalgraph signal which changes over time from the graph model, and obtaining a feature vector of each node involved in the session graph; S4, obtaining global preference information from historical sessions of the user by adopting an attention mechanism; S5, obtaining local preference information of the user from the last session clicked by the user, and obtaining final preference information of theuser in combination with the global preference information; S6, predicting the probability of possible occurrence of the next clicked commodity in each session, and giving a Top-K recommended commodity. According to the method, rich context relationships of clicked commodities can be captured from the global graph, global and local preferences of the user are accurately learned, the time attenuation effect of historical preferences of the user on current preferences is effectively evaluated, and accurate commodity prediction is provided.

The invention discloses a knowledge reasoning method based on a multi-modal knowledge graph, and aims to enable knowledge reasoning reliability and accuracy to be higher and enable the knowledge reasoning method to have stronger modeling and reasoning capabilities. The method is realized through the following technical scheme: different information is fused based on multi-hop reasoning of a large-scale knowledge base; attribute completion is performed on the attribute missing graph through attribute graph embedding, structured information is extracted from unstructured and semi-structured documents or sentences, and a dynamic heterogeneous graph embedding model is constructed for multi-type characteristics of the multi-modal knowledge graph through heterogeneous graph embedding; feature learning of semi-structured knowledge, structured knowledge and different types of non-structured knowledge is achieved, and multi-modal knowledge graph features are obtained and serve as input for knowledge reasoning based on a graph neural network GNN; an inference path is generated, and a plurality of types of inference paths are constructed; and classification, edge prediction and frequent subgraphs of node types are calculated on the graph, a knowledge reasoning task is generated, and multi-step complex knowledge reasoning is completed.

The invention provides an interest point recommendation method based on a graph neural network and user long-term and short-term preferences. The method comprises the following steps: taking a point-of-interest sequence accessed every day in historical sign-in data of a user as a session sequence; creating a directed graph based on the sessions, where each session sequence is regarded as a sub-graph, indicating that each node represents a point of interest, and each directed edge represents that a user accesses a pointed point of interest after accessing a source point of interest of each edge. Based on this graph, relationships between points of interest are captured by a graph neural network and vector representations of the points of interest are accurately generated. Based on the representation vectors of the interest points, the interest points to be accessed in the next step are recommended for the user by combining an attention mechanism. According to the invention, a better geographic information model is fused from the perspectives of users and the interest points. Therefore, the geographic distance between the users and the interest points and the sign-in frequency of theusers on the adjacent interest points are used in the model, and the problem of sign-in data sparseness is solved.

The invention belongs to the technical field of wireless communication, and discloses a communication signal modulation mode identification method based on a graph neural network. The method includes:transmitting modulation signals of various modulation modes at a transmitting end to obtain a communication signal modulation mode identification data set; dividing the data set according to the number of graph neural network interfaces to obtain a plurality of training sample subsets, inputting the training sample subsets into a feature embedding network one by one, outputting a feature embedding vector of a modulation signal, inputting the feature embedding vector set into the graph neural network, and outputting feature vectors of test samples; and finally, mapping the feature vectors of the test samples into classification results, training the feature embedding network and the graph neural network according to the classification results, and identifying a modulation mode of an unknown modulation signal after the training is completed. According to the method, the problem that an additional data preprocessing means is needed in the prior art is solved, so that the recognition efficiency is improved, and the system complexity is reduced.

The invention provides a multi-relation collaborative filtering algorithm based on a graph neural network, and the algorithm comprises the steps: 1, processing the historical interaction data of a user and an article, and extracting all user sequences UU and article sequences SI from the historical interaction data; and step 2, respectively constructing each user sequence SU and each article sequence SI into a user relationship graph GU = (VU, EU) and an article relationship graph GI = (VI, EI). The invention provides a multi-relation collaborative filtering algorithm based on a graph neural network. The article relationship and the user relationship are modeled at the same time by constructing the article relationship graph and the user relationship graph from the user and article interaction data, the multiple relationships are fused into the learning process of user and article interaction, and the influence of the multiple relationships on user and article interaction is learned, so that the model can better model user preferences.

The invention discloses a session recommendation method based on graph representation. According to the method, based on historical interaction data of a given target user, the next item most likely to be interacted by the target user is found out. The method comprises the following steps: firstly, constructing a directed graph neural network of an article based on a user historical session sequence; and capturing a transfer relationship between the articles through a graph neural network; then modeling the current session of the user by using a long-short memory network to extract short-terminterests, and meanwhile, obtaining the long-term interests of the user from the recent session sequence of the user by using a maximum pool method; and finally, combining the short-term interests andthe long-term interests of the users to recommend the articles. The method disclosed by the invention overcomes the defects in the existing method: (1) the complex transfer relationship of articles in the session sequence cannot be captured; and (2) the long-term interest of the user is not considered to change along with time. Therefore, compared with an existing method, the recommendation effect implemented by the method is obviously improved.

Methods and apparatus are provided for generating a graph neural network (GNN) model based on an entity-entity graph. The entity-entity graph comprising a plurality of entity nodes in which each entity node is connected to one or more entity nodes of the plurality of entity nodes by one or more corresponding relationship edges. The method comprising: generating an embedding based on data representative of the entity-entity graph for the GNN model, wherein the embedding comprises an attention weight assigned to each relationship edge of the entity-entity graph; and updating weights of the GNN model including the attention weights by minimising a loss function associated with at least the embedding; wherein the attention weights indicate the relevancy of each relationship edge between entitynodes of the entity-entity graph. The entity-entity graph may be filtered based on the attention weights of a trained GNN model. The filtered entity-entity graph may be used to update the GNN model or train another GNN model. The trained GNN model may be used to predict link relationship between a first entity and a second entity associated with the entity-entity graph.

The embodiment of the invention provides a commodity recommendation method and device. The commodity recommendation method comprises the steps: inputting a target user into a trained recommendation system, and obtaining a candidate commodity set; reordering each candidate commodity in the candidate commodity set based on a nearest neighbor search algorithm according to the target user representation vector and the target commodity representation vector, and taking a plurality of candidate commodities sorted in the top as optimal recommended commodities; and recommending all the optimal recommended commodities to the user. The commodity recommendation method and device generate the target user representation vector and the target commodity representation vector according to the similarity between the target user and each commodity, based on the recommendation system constructed by the graph neural network, so that the diversity of the candidate commodities is improved, and the diversityof the best recommended commodities is improved, and the commodity with the maximum probability is recommended to the user through the commodity recall stage, and the recommendation accuracy is improved.

The invention discloses a fine-grained source code vulnerability detection method based on a graph neural network, which comprises the following steps: A, for all source codes of software, extractinga code attribute graph of a function in the source codes, and performing vectorization representation on the code attribute graph, the vectorization representation comprising vectorization representation of node attributes and graph structures; B, extracting code features by using a graph attention network, and processing all statements to obtain a final data set; and C, training a vulnerability detection model, and determining a final model through the effect on the test set. The source code vulnerability detection method provided by the invention has the advantages that the automation degreeis higher; dependence on domain expert knowledge is reduced, code auditing cost is greatly saved, code auditing efficiency is improved, grammar and semantic information in source codes are better reserved, meanwhile, vectorized representation of statements can be learned purposefully through a graph attention network, and vulnerability detection of a finer-grained statement level is achieved.

The embodiment of the invention provides a behavior prediction model training method. The method comprises the steps: firstly determining a plurality of sample users for a target object, enabling anyfirst sample user to correspond to a first sample hard tag, and indicating whether a specific behavior is carried out on the target object or not; performing graph embedding processing on the user-object bipartite graph based on the trained graph neural network to determine an embedding vector set; determining a sample user feature vector corresponding to the first sample user and a target objectfeature vector corresponding to the target object, determining a specific behavior probability that the first sample user makes a specific behavior to the target object as a first sample soft label, and inputting the sample user feature vector into a first behavior prediction model to obtain a behavior prediction result; and training the first behavior prediction model by using a first loss item determined based on the prediction result and the first sample hard tag and a second loss item determined based on the prediction result and the first sample soft tag.

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.

An embodiment of the invention provides a transaction data exception detection method, a medium, a device and computing equipment. The method comprises the steps of generating a knowledge graph basedon transaction data obtained in advance; wherein nodes of the knowledge graph are used for representing account entities in the transaction data, and edges between the two nodes are used for representing transaction relationships between the account entities corresponding to the two nodes respectively; performing graph deep learning on the knowledge graph by using a graph neural network to obtainfeature representation of each edge in the knowledge graph, and determining the feature representation of the edge as a feature vector of transaction data corresponding to the edge; and inputting a predetermined feature vector of the transaction data to be detected into a neural network model obtained by training the feature vector of the transaction data, and outputting a detection result of thetransaction data to be detected after processing of the neural network model. According to the invention, exception detection of transaction data can be automatically completed, and compared with theprior art, the method is advantageous in that the problem of inaccuracy caused by a manual mode is avoided.

The invention provides a commodity recommendation method and system based on a user session and a graph convolutional neural network. The commodity recommendation method comprises the following steps:receiving a clicked commodity sequence in the same time unit, taking the clicked commodity sequence as a session, modeling in the form of a directed graph, and constructing a session graph, wherein the session graph is expressed in the form of an embedded vector; inputting the plurality of session graphs in the embedded vector form into a gated graph neural network, and outputting posterior probabilities of all commodities contained in the session graphs; and according to the descending order, screening out the commodities corresponding to the posterior probabilities ranked at the front preset digits as user preference commodity prediction results, and recommending the commodities one by one. On the premise of considering the commodity relevance, the commodity recommendation accuracy andspeed are improved.

The invention belongs to the technical field of Internet of Vehicles edge computing, and particularly relates to an Internet of Vehicles edge computing task unloading method based on hierarchical reinforcement learning. Firstly, a task unloading problem in an edge computing network of the Internet of Vehicles is modeled as an optimization problem with a minimum delay-energy consumption-cost joint loss function as a target, wherein optimization parameters are a task execution sequence, a computing decision, local resource allocation and transmission power control; then, the application with task relevance is expressed in the form of a directed acyclic graph, implicit features in the application are mined through a graph neural network, and meanwhile the discrete continuous mixed action space is processed through a layered reinforcement learning algorithm; and a simulation experiment is carried out by taking the automobile speed adopted in a real environment as a data set, and a result shows that compared with a heuristic algorithm, the method disclosed by the invention can adaptively adjust task unloading and resource allocation strategies under various environmental parameters, so that a system loss function is more effectively reduced.