105 results about "Graph classification" patented technology

The invention provides a software defect prediction method for open source software defect feature deep learning, and belongs to the technical field of software engineering. The method comprises the steps of collecting open source software defect information, constructing a software defect database, and generating an abstract syntax tree from source codes; pruning the abstract syntax tree by usinga community detection algorithm to obtain a defect sub-tree, establishing an information corpus of the defect sub-tree in combination with the repair description, the project basic information and the source code, extracting theme words from the information corpus, converting the theme words into vector representation, and taking the vector representation as attributes of nodes in the defect sub-tree; finally, establishing a software defect prediction model of the convolutional neural network based on graph classification, expressing the defect subtree as an adjacent matrix and an attribute matrix to serve as input of the model to train the convolutional neural network, and recognizing whether the source code of the to-be-predicted software module has defect tendency or not. According tothe method, the defect depth features are directly extracted from the structured software codes by using a deep learning method, so that a better defect recognition effect can be achieved.

Disclosed is system and method of connection information regularization, graph feature extraction and graph classification based on adjacency matrix. By concentrating the connection information elements in the adjacency matrix into a specific diagonal region of the adjacency matrix in order to reduce the non-connection information elements in advance. The subgraph structure of the graph is further extracted along the diagonal direction using the filter matrix. Then a stacked convolutional neural network is used to extract a larger subgraph structure. On the one hand, it greatly reduces the amount of computation and complexity, solving the limitations of the computational complexity and the limitations of window size. And on the other hand, it can capture large subgraph structure through a small window, as well as deep features from the implicit correlation structures at both vertex and edge level, which improves the accuracy and speed of the graph classification.

The invention provides a graph classification method based on graph set reconstruction and graph kernel dimensionality reduction. The method comprises the steps of: 1) performing frequent sub-graph mining on a graph data set used for training, and performing discriminative sub-graph screening on obtained frequent sub-graphs with the emerging frequentness differences of the sub-graphs in a positive class and a negative class; 2) reconstructing the original graph set with selected discriminative frequent sub-graphs; 3) obtaining a kernel matrix for describing the similarity between every two graphs in the newly-reconstructed graph set by using a Weisfeiler-Lehman shortest path kernel method, and based on class label information of training graphs, performing dimensionality reduction on high-dimensionality kernel matrixes by using a KFDA method; 4) training graph data projected to a low-dimensionality vector space based on an extreme learning machine to build a classifier; 5) standardizing graph data requiring classification, projecting the data to a low-dimensionality space obtained through training and inputting the projected data to the classifier to obtain a classification result. The method can directly classify graph data without class labels and guarantee high classification accuracy.

The invention relates to a modulated signal time-frequency diagram classification system based on a generative adversarial network and an operation method of the modulated signal time-frequency diagram classification system. The modulated signal time-frequency diagram classification system comprises an IQ signal time-frequency diagram conversion module, a generator module, a discriminator module and an auxiliary classifier module, wherein the IQ signal time-frequency diagram conversion module converts an original signal into a signal short-time Fourier time-frequency diagram form; the generator module maps an original noise vector and category information input by means of the original noise vector into a corresponding short-time Fourier time-frequency graph; the discriminator module is used for receiving real picture data and picture data generated by a discriminator, and outputting confidence probabilities which are different in input and correspond to each other and are discriminated to be true; and the auxiliary classifier module receives a high-dimensional feature map extracted by a convolution layer and outputs corresponding category information. The operation method is usedfor solving the problem that a deep neural network model for signal classification is trained but the number of data set samples is insufficient.

The invention discloses an entity relationship joint extraction method based on span and knowledge enhancement, and belongs to the technical field of information extraction and natural language processing. The method comprises the following steps: firstly, constructing a sample data set and labeling the data set; carrying out entity identification and relationship classification and specifically,for the labeled data, mapping words in a high-dimensional discrete space to a low-dimensional continuous space vector by a pre-training language model; carrying out span identification, filtering andrelationship classification by a span-based model; converting relationship classification into graph classification by utilizing a graph-based model, and introducing a syntactic dependency relationship so as to assist relationship judgment and classification; and performing joint training on an output result of the span-based model and an output result of the graph-based model, and identifying entities contained in the data and relationships among the entities. Syntactic information such as the dependency relationship is introduced into the end-to-end neural network model, so that the overlapping relationship is effectively identified, and the joint extraction accuracy of the entity relationship is improved.

The invention discloses an electrocardiosignal graph classification method based on deep learning, and belongs to the field of image recognition. A traditional single convolutional neural network or single recurrent neural network has the problem of poor classification in part of categories, and has the problem of poor generalization ability of an automatic classification algorithm caused by different representations under the same disease due to individual differences such as heartbeat intensity, heart rate and the like. The invention discloses an electrocardiosignal graph classification method based on deep learning. The method comprises the following steps: preprocessing acquired image data of the electrocardiosignal graph, wherein the preprocessing includes wavelet denoising of the image data of the electrocardiosignal graph and data enhancement of the image data of the electrocardiosignal graph; designing a convolutional neural network, and classifying the electrocardiosignal graph by utilizing the designed convolutional neural network. The electrocardiosignal recognition and classification effect can be remarkably improved.

The invention discloses a hyperspectral image semi-supervised classification method based on comprehensive confidence. The method comprises the following steps: reading a hyperspectral image; Calculating a graph weight matrix; 8, performing adjacent connection on the sparse graph weight matrix; Calculating a normalized graph weight matrix; Obtaining an initial training set and a candidate set; Setting collaborative training iteration times and starting a training process; Training a polynomial logic regression classifier; Obtaining a prediction label of the candidate set sample by using a polynomial logic regression classifier; Obtaining prediction tags of the candidate set samples by using a semi-supervised graph classification method; Selecting two candidate samples with consistent prediction tags and corresponding prediction tags to form a protocol set, and forming a comprehensive confidence set by corresponding confidence coefficients; Screening out a protocol set sample with a comprehensive confidence coefficient higher than 99% and a corresponding prediction label, and forming an amplification set and adding the amplification set into a training set; Removing an amplificationset sample in the candidate set; And judging whether the training reaches a set number of times, if not, continuing iteration, and if yes, stopping iteration, and classifying the hyperspectral imagesby using the semi-supervised graph.

The invention discloses a graph theoretical method MalZero for quickly and accurately detecting zero-day malicious software. The method is characterized in that features are extracted from API calling graphs to establish a classifier model, benign software and malicious software have different graph theoretical properties in respective API calling graphs, and the graph theoretical features extracted from the API calling graphs can be used for distinguishing the malicious software from the benign software effectively and efficiently. Three modules, namely a graph construction module, a feature extraction module and a graph classification module are involved in the method. Through the MalZero, the malicious software can be quickly and accurately detected on a terminal host, high efficiency is achieved in terms of storage space and detection time, and the MalZero can supplement an existing malicious software detection scheme of the terminal host due to low overhead.

The invention discloses an Ethereum phishing fraud detection method and device based on graph classification, and the method comprises the steps: extracting a target node and preset-order neighbor nodes from an Ethereum network, and enabling the preset-order neighbor nodes to comprise a first-order neighbor node and a second-order neighbor node; constructing a second-order transaction sub-graph network taking the target node as a center node according to the first-order neighbor node and the second-order neighbor node; refining the second-order transaction sub-graph network according to the related transaction information data of each node in the second-order transaction sub-graph network to obtain a target transaction sub-graph network; extracting features in the target transaction sub-graph network by adopting a preset graph embedding algorithm to obtain a network representation vector; and inputting the network representation vector into a preset classifier for binary classificationprocessing to obtain a target fishing node. The technical problems that in the prior art, manual feature limitation is obvious, the processing process is complex, and consequently time and operationcost is high can be solved.

The invention discloses a graph classification method of a cyclic neural network model based on Attention. The Attension idea is applied in the graph classification problem, and the graph classification problem is regarded as a decision process of interaction between a machine and a graph environment in reinforcement learning. Based on the Attention idea, the machine preferentially observes the target area of the graph classification task instead of directly processing the whole graph, so that the target area can be preferentially processed by ignoring the nodes irrelevant to the classification task, and the visual angle movement direction of the machine observation graph can be trained and determined by reinforcement learning rules. At the same time, the model can control the parameters and computational load, and get rid of the constraint on the size of graph data. The invention constructs a cyclic neural network, which integrates the local information of the graph observed before bythe machine through the hidden layer of the circulating neural network, and is used for assisting the decision of the angle of view movement and the graph classification. The invention avoids the problem of subgraph isomorphism in frequent subgraph mining and the problem that the graph kernel function method lacks scalability.

The invention provides an independent decoupling convolutional neural network representation algorithm for graph data. A brand-new neural network structure decoupled by using independent factors is provided, decoupling representation learning is carried out by using a neighborhood routing mechanism, then independence of potential factor representation between nodes and neighbor nodes is enhanced by using an HSIC algorithm, and the independence is integrated into a convolutional neural network as a regularization item. Through the method, independence among potential factors of the nodes can beenhanced, and better graph node separation representation can be obtained. Through verification of different graph data, the method can be applied to three tasks including semi-supervised graph classification, graph clustering and graph visualization, and has good performance and obvious advantages.

A classification model training method is provided. The method includes selecting a training dataset, determining a category of a sketch in the training dataset according to a sketch classification model, to obtain a first category processing result, and analyzing, according to a second feature analysis model, a feature of the sketch extracted by a first feature extracting model, to obtain a second analysis result of the sketch; then obtaining a function value of a first loss function according to the first category processing result and the second analysis result of the sketch; and finally adjusting a first model parameter value of the sketch classification model according to the function value of the first loss function.

The invention provides a graph classification method and device based on a graph neural network technology of sub-graph division and inter-sub-graph pooling. The graph classification method comprisesthe steps of selecting a sub-graph extension center based on the node degree; according to a mode based on breadth-first traversal, obtaining sub-graphs through expansion; training a corresponding graph convolution network, and obtaining an intra-sub-graph feature vector of each sub-graph containing node features and adjacency relation information from the adjacency matrix and the feature matrix of each sub-graph; performing maximum pooling by taking the modulus lengths of the feature vectors in the sub-graphs as metrics, and selecting part of the feature vectors in the sub-graphs as GAT input; training a corresponding GAT, and taking the feature vector in the sub-graph selected by maximum pooling as a node input to obtain a sub-graph feature vector containing information between the sub-graphs; and utilizing a classifier to classify the sub-graph feature vectors to obtain sub-graph categories, and determining the graph categories according to a maximum voting principle. According to the technical scheme, the classification principle has good interpretability and a good classification effect.

The invention discloses a heterogeneous graph classification method based on a double-layer attention mechanism, and the method comprises the steps: S11, carrying out the mapping operation of the feature vectors of different types of nodes, and obtaining the feature vectors of a target node and a neighborhood node corresponding to the target node in the same entity space, S12, learning weights ofdifferent types of neighbors and different adjacent nodes from top to bottom from the type-level attention to the node-level attention according to the obtained feature vectors and based on the type-level attention and the node-level attention in the double-layer attention, S13, constructing a heterogeneous graph according to the obtained weight to obtain a classification model, and S14, inputtingtest data into the obtained classification model, and outputting a final classification result.

The invention provides a link prediction method based on a graph neural network and a capsule network, and the method comprises the steps: carrying out the representation learning of a network through GNN, and generating corresponding node features; secondly, through a conversion block designed by the patent of the invention, converting the learned node features into an Edge feature map so that a link prediction problem is converted into a graph classification problem; thirdly, performing feature representation learning on the node pair feature graph by means of CapsNet, and capturing attributes of node pairs from different aspects for graph classification; and finally, performing wide evaluation on six networks without node attributes and three networks with node attributes, and analyzing the feasibility and effectiveness of the proposed GCCL method. Experimental results show that the GCCL method provided by the invention is obviously superior to the compared method not only in a network containing node attributes, but also in a network not containing node attributes, and the accuracy is averagely improved by about 20%. The reasonability and the effectiveness of the GCCL method in the link prediction task can be proved.

The invention discloses an MAC protocol identification method based on a convolutional neural network, and the method comprises the following steps: (1) generating training data: firstly obtaining labeled original sampling data, the data form being a time-power value sequence, and then converting the original sampling data into a time-frequency graph; (2) convolutional neural network training: inputting the generated time-frequency graph into a convolutional neural network for training to obtain a trained convolutional neural network model; and (3) MAC protocol identification: carrying out data acquisition on a signal of a network to be identified, converting the signal into a time-frequency graph, and inputting the time-frequency graph into the trained convolutional neural network model to realize protocol identification. According to the method, the original communication data of the communication network is converted into the time-frequency graph by adopting a graphical recognition thought, and the time-frequency graph is classified by utilizing the convolutional neural network, so that the MAC protocol type of the target network is recognized.

The invention discloses a graph classification method based on a U-shaped nested network. According to the method, starting from extraction of more complex multi-scale targets, Graph U-Nets are nested, so that the structural information of overall data can be captured, and local information and global information can be considered at the same time. The invention provides an improved graph neural network for a graph classification task, and the improved graph neural network can extract complex structure features in data, thereby improving the prediction accuracy.

The invention provides an identification method and device for an Internet of Things identifier and terminal equipment. The identification method for the Internet of Things identifier comprises the following steps: acquiring a to-be-identified Internet of Things identifier; converting the to-be-identified Internet of Things identifier into a polygonal graph; identifying the polygonal graph by using a pre-trained graph classification model to obtain an identification result representing a coding category of the to-be-identified Internet of Things identifier; and analyzing the to-be-identified Internet of Things identifier according to the coding category of the to-be-identified Internet of Things identifier. According to the embodiment of the invention, the invention can achieve the automatic recognition of IOT identifiers of different coding systems through the pre-trained graph classification model, simplifies the recognition process of the heterogeneous IOT identifiers, and improvesthe recognition efficiency.

The invention relates to the field of graph classification and recognition, in particular to an image classification and recognition method and device based on a self-adaptive dynamic convolutional network and computer equipment, and the method comprises the steps: obtaining a to-be-detected image, inputting the to-be-detected image into a preprocessing block, and obtaining a shallow feature graph and graph parameter information of the image; combining the image parameter information obtained after preprocessing with the to-be-detected image, and inputting the combined image parameter information and the to-be-detected image into an adaptive dynamic convolutional network of a backbone network to obtain image global features; wherein the adaptive dynamic convolution is to select a convolution kernel with a corresponding shape according to the corresponding parameter information; inputting an image shallow layer feature map obtained after preprocessing into a branch network, and extracting local features of the to-be-detected image; and carrying out feature fusion on the local features and the global features, inputting the fused features into a classification network, and outputting classification identification information of the to-be-detected image. According to the method, the calculation cost required for image classification and recognition is low, the precision is high, and the applicability of related products is high.

The invention discloses a graph classification method based on an adaptive multi-channel cross graph convolutional network. The method comprises the following steps: 1) constructing an initial graph based on a node feature matrix X; (2) carrying out graph convolution operation on the input graph, (3) carrying out a cross network, (4) carrying out a graph convolution module, and (5) carrying out a full connection layer with an attention mechanism. The invention fully utilizes information in a space, eliminates the requirement of searching a plurality of multi-task network system structures on the basis of each task, and ensures the consistency of learning embedding.

The invention discloses a graph classification method. A graph as a commonly used dataset structure can be used for expressing various complex relations among data objects in many scientific applications. According to the method disclosed in the invention, a new graph characteristic which is an ordinal pattern is put forward, and weight information and a local topology structure in the graph can be retained via the ordinal pattern which is therefore an ideal graph character. Via a series of algorithms, a distinguishing ordinal pattern is extracted from the graph and is used as a character. Finally, based on a support vector machine, a classifier used for graph classification is built. Via the method disclosed in the invention, efficient and accurate graph classification can be realized.

The invention provides a multi-view image classification method and system, computer equipment and a storage medium. The method comprises the steps of obtaining a multi-view image classification dataset; constructing a multi-view classification model based on a graph convolutional neural network according to the classification data set; wherein the multi-view classification model comprises an input layer, a spectrogram convolution layer, a batch regularization layer, a graph pooling layer, a full connection layer and a softmax function multi-classification output layer; and performing classification training on the multi-view classification model according to the classification data set to obtain a classification result. According to the embodiment of the invention, on the application ofmulti-view image classification, the processing of any data structure is supported, the universality of the model is improved, the low dimensionality of training data is ensured, the storage space andcomputing resources are reduced, and the convergence rate and classification precision of the model are improved under the condition that pre-training is not needed.

The invention discloses a graph classification method for a multi-layer MLP network, a medium and equipment.The method comprises the steps: constructing a graph, and converting a node connection edge of the graph into an adjacent matrix; inputting the minimum batch of feature matrix into an input layer; inputting the minimum batch of representation vectors output by the input layer into a BatchnNormal layer for mean value normalization; multiplying a normalized minimum batch representation vector output by the BatchnNormal layer by the minimum batch adjacency matrix, inputting the product into the middle layer, and outputting a minimum batch representation vector; inputting the output minimum batch representation vector into a BatchNorm layer for normalization, multiplying the normalized representation vector by a minimum batch attention adjacency matrix, and inputting the product into an output layer; establishing a network model and training; and inputting to-be-predicted similar graphs into the trained neural network model, outputting graph labels, and completing a graph classification task. According to the method, different importance is distributed to different nodes in a neighborhood by adopting an attention mechanism according to the characteristics of the nodes, the characteristic vectors are aggregated for multiple times by using MLP, graph labels are better classified, and the classification precision is high.

The invention relates to a microseism event detection and positioning method and system, and the method comprises the steps: carrying out the SET imaging of collected microseism signals of a plurality of monitoring stations in a fracturing process, dividing an image according to whether a microseism event occurs, and building a training data set and a test data set; inputting a training data set sample into the established residual error network model for training, and inputting a test data set sample into the trained residual error network model for checking the performance of the residual error network model; storing parameters of the trained residual network model; and collecting micro-seismic signals of a plurality of monitoring stations in the real-time fracturing process, establishing a to-be-tested data set, inputting the data of the to-be-tested data set into the trained residual network model for detection to obtain an imaging graph classification result, and determining whether a micro-seismic event exists and the position of a seismic source exists or not according to the imaging graph classification result. According to the invention, real-time on-line detection of the micro-seismic event and the seismic source position can be realized, the identification speed is fast, and the identification precision is high.

The embodiment of the invention provides an information management method and system based on big data and the Internet, and the method comprises the steps that the image classification labels under each artificial intelligence recognition result are classified based on the preset big data collection and classification, thereby considering the difference of different big data collection and classification and improvingthe condition of multi-classification conflicts in the classification process; in addition, the user search behavior information and the historical grading and classification information of the target graph search object are combined, after the information management index sequences of the graph classification and grading labels of theuser search behavior information and the historical grading and classification information are compared, big data analysis is carried out on each graph unit library in the target graph search object based on each corresponding graph classification and grading label of the artificial intelligence recognition result, so that based on the historical classification condition during the previous big data analysis, more accurate and rapid big data processing is carried out on the image unit library based on some key graph classification and grading tags, the big data analysis efficiency is improved, and the buffer time is shortened.

The embodiment of the invention provides a data analysis method and system based on artificial intelligence and the Internet. The method comprises the steps that a graph classification grading bookmark corresponding to a target graph search object initiated by Internet access equipment is collected; the graph classification grading bookmark is used for representing graph classification grading information of the target graph search object under the big data analysis bookmark, then big data analysis is carried out on each graph unit library in the target graph search object according to the graph classification grading bookmark, and therefore information recommendation is carried out after portrait recognition is carried out. Compared with a traditional scheme in which big data analysis iscarried out in the interaction process of a graphic unit library in a generalized range, the situation that portrait analysis accuracy is not high due to introduction of excessive noise data can be avoided, pertinence of big data analysis is improved. Meanwhile, cloud computing resources can be saved, cloud computing performance is improved, and the computing amount is effectively reduced.

The invention discloses a method for improving the classification accuracy of a vector graph bitmap. The method comprises the following steps: obtaining the number of element paths of a classificationgraph and the size of the classification graph by reading a binary byte stream of each classification graph; Dividing the input matrix into a training set, a test set and a verification set accordingto a ratio of 7: 2: 1, determining a residence point of a minimized objective function by using the input matrix, and determining classification center points of the two types of classification graphs; obtaining clustering classification center by minimizing the objective function, obtaining and marking then the two types of classification graphs respectively, and achieving rapid and accurate classification. According to the method, the clustering model is trained by introducing the element paths and the classification graph size tags, and the classification graph classification accuracy canbe effectively improved.

The invention discloses a Chinese speech decoding nursing system based on transfer learning. The Chinese speech decoding nursing system comprises that an Inception-V3 neural network obtained by transfer learning is taken as a basis; the auxiliary classifier at the front layer position is deleted, anddeleting all network structures behind the auxiliary classifier at the middle layer position are deleted; an original full connection layer is deleted at the middle layer position of the neural network and then a new seven-layer residual neural network structure is added to construct a new convolutional neural network suitable for electroencephalogram time-frequency graph classification; only the residual neural network is trained by using the local first data set to obtain a trained first convolutional neural network, and then only the residual neural network is trained by using the local second data set to obtain a trained second convolutional neural network; the first convolutional neural network judges whether the paralyzed patient feels uncomfortable or not according to the electroencephalogram time-frequency diagram obtained by the signal preprocessing module, and if yes, the second convolutional neural network continues to be used for further judging the type of discomfort of the patient; the second convolutional neural network will judge whether the paralyzed patient is hungry or cold or other.

The invention discloses a graph classification method based on an attention mechanism. The method comprises the steps of obtaining an original graph network and corresponding graph network data information; initializing feature embedding of nodes, constructing a high-order graph network, and initializing node embedding in the high-order graph network; capturing and absorbing neighbor node information and interaction information between neighbor nodes by nodes in the graph network of each order by adopting a substructure interaction attention network, and training and obtaining new feature embedding of each node in the graph network; adopting a node attention network to embed and fuse node features in the graph network into graph network embedding, and after cascading each order of graph network embedding, performing dimension reduction classification through a multi-layer perceptron to obtain a final graph network classification result. The invention also discloses a compound toxicity prediction method comprising the attention mechanism-based graph classification method. According to the method, the accuracy of graph network classification and the accuracy of compound toxicity prediction are effectively improved; and the method is relatively high in efficiency, relatively good in accuracy and easy to implement.