41 results about "Nearest neighbor graph" patented technology

The invention discloses a mass spectrum data high-throughput alignment and parallel qualitative method based on deep characterization learning, and relates to the field of chromatography-mass spectrometry data processing and calculation mass spectrography. The method comprises the following steps: performing streaming, micro-batch and single traversal analysis on original mass spectrum data files of a plurality of samples to be detected; extracting metadata and main data by taking the fragmented spectrogram as a unit, inputting the metadata and the main data into a pre-trained depth representation learning model to generate a high-dimensional embedded vector, and incrementally writing the metadata and the main data into a column type storage database; constructing a sparse approximate nearest neighbor graph; executing a clustering algorithm on the graph to generate a consensus spectrogram set, and generating a consensus feature vector for each clustering cluster and a single example cluster; determining the nature of the consensus feature vector; and constructing a sample * feature matrix based on the cluster affiliation relationship and the qualitative result. According to the method, the technical problems of I / O bottleneck, memory overflow, combinatorial explosion, computing power waste and the like in large-scale mass spectrum data analysis are effectively solved.

The invention discloses a novel graph contrast learning framework with mutual information maximization, and relates to the technical field of graph contrast learning, multi-modal learning and mutual information maximization. The framework comprises the following steps: constructing a nearest neighbor graph, spreading similar samples to data of a missing mode, optimizing a loss function, and performing clustering optimization. The existing method has two main problems: (1) the existing method is mostly suitable for complete data and has a poor processing effect on data of a missing mode; and (2) the current palmprint and palm vein fusion recognition technology does not completely capture the complex relationship between the palmprint and palm vein features, resulting in inaccurate clustering result. According to the method, the modal loss in the incomplete palm print and palm vein fusion recognition task is effectively reduced, and the clustering performance is enhanced. According to the method, the mutual information maximization loss function is introduced, the adaptability and robustness of the model to incomplete data are effectively enhanced, and the method has high recognition precision and clustering performance when processing incomplete multi-modal data.

This invention relates to a method and system for mining potential user needs based on hierarchical temporal memory enhancement. First, a dynamic time window hierarchical slicing mechanism is used to decouple the cleaned multimodal user behavior sequence into long-term historical sequences and short-term real-time sequences. A semantically approximate nearest neighbor graph is constructed from the long-term historical sequence, and adaptive community discovery is performed. This is combined with a large language model to generate a high-semantic-density long-term memory abstract representation. Then, a hybrid prompt template integrating system instructions, long-term memory representation, and short-term sequences is constructed to drive the large language model to complete a three-stage thought chain reasoning process: focused denoising, scene instantiation, and function completion, generating a structured reasoning data package. A dual-tower semantic retrieval and scene-constrained filtering architecture is employed to achieve accurate mapping between the reasoning results and the product database, outputting a product list with recommendation reasons. This invention endows the recommendation system with deep logical reasoning and cross-domain association capabilities, improving the accuracy and interpretability of e-commerce recommendations.

This invention discloses a method for quantifying anxiety and depression comorbidities using partial label learning based on brain imaging, belonging to the fields of medical image processing and artificial intelligence. The method is based on a dataset constructed using brain imaging measurements. It selects representative typical anxiety and depression samples with distinctive brain imaging features from clinically diagnosed groups and constructs a sparse nearest neighbor graph by comprehensively considering the relationships between each sample and its nearest ordinary and typical neighbor samples. Partial label learning modeling and optimization are performed through an iterative label propagation mechanism, ultimately outputting label confidence matrices for clinical anxiety patients, depression patients, and patients with comorbid anxiety and depression. This enables a quantitative assessment of the anxiety and depression levels in patients with comorbid anxiety and depression and completes the classification of comorbidity subtypes. This invention improves the accuracy and objectivity of anxiety and depression comorbidity assessment, providing auxiliary support for the clinical diagnosis, pathological mechanism research, and intervention target screening of anxiety and depression comorbidities.

The present application relates to a kind of hard disk and the near neighbor graph vector retrieval method and device of memory, first according to the distance of each feature vector in data set and constructs near neighbor graph, then it is saved in hard disk and is released in memory according to 4K alignment mode optimization its storage and resource, then according to the structure information of near neighbor graph saved in hard disk, calculate sample set, again construct the near neighbor graph of sample set object as sample near neighbor graph.Using query vector on sample near neighbor graph executes nearest neighbor search to obtain the preliminary more similar query target set, finally with the more similar query target set obtained as the initial solution of near neighbor graph in hard disk, executes nearest neighbor search to obtain the most similar query target.The present application reduces the memory occupation in large-scale data search process under the condition of guaranteeing accuracy, and improves the search efficiency.

The invention discloses a high-dimensional vector clustering algorithm based on GPU acceleration, and the algorithm comprises the steps: obtaining a high-dimensional space vector set, accelerating the construction process of a k-nearest neighbor graph through a GPU, and completing the construction of a global k-nearest neighbor graph; obtaining a high-dimensional space vector set, recursively executing k-means clustering on the high-dimensional space vector set, and completing partition operation of vector objects; independently executing DBSCAN clustering in parallel among different partitions by using a global k-neighbor graph, a partitioning scheme and a GPU thread block to complete partition clustering; vector objects conforming to core point definition are obtained, a core neighbor graph is constructed based on the density direct relation among the vector objects, the core neighbor graph and the partition clustering results are utilized, the partition clustering results are combined, and a global clustering result of the high-dimensional space vector set is obtained. According to the method, the execution time of the DBSCAN algorithm can be remarkably shortened, and the efficiency of large-scale high-dimensional vector clustering is greatly improved on the premise that the clustering precision is guaranteed.

The invention discloses a data acquisition visualization traceability method, which comprises the following steps of: performing real-time acquisition on target data through an equipment sensor, and performing preprocessing operation on the acquired data; generating a unique identifier for each piece of collected data by using a Hash identification method, and storing the identified data into a database; performing normalization processing on the sensor data, converting the data into feature vectors, calculating function association vectors among the data by adopting a gating circulation unit model, and determining the association of the sensor data in combination with a geographic position; the method comprises the following steps: calculating a spatial distance between sensors, constructing a graph model, calculating a function association vector, endowing a weight to an edge in the graph model, adjusting the weight of the edge by adopting a multi-layer dynamic weighted nearest neighbor graph optimization algorithm, screening out a path with the minimum total weight from candidate nodes and the edge, and visually displaying a traceability path in a coordinate system. The weights of the nodes and the edges are distinguished by using different colors and sizes, so that the visual traceability analysis of the data is realized.

This invention proposes a collaborative evolutionary classification method based on cross-view structural attribute dual-domain semantic mining, belonging to the field of multi-view semi-supervised classification. This invention acquires multi-view features from training samples, constructs K-nearest neighbor graphs for each view, and adaptively prunes based on the shared support relationships of the graph structure to obtain a purified adjacency matrix. Subsequently, structural and attribute representations are extracted through a shared graph convolutional network, and the consistency between the two in a unified space is constrained using a structure-attribute consistency loss. Simultaneously, learnable view position encoding and masked self-attention mechanisms are introduced for progressive fusion, and cross-view evolutionary patterns are learned through adjacent view mapping to form an evolutionary loss. Finally, the evolutionary representations of each view are concatenated and mapped to a sample-level fused representation, which is then input into the classifier for prediction. The model is jointly optimized by the classification loss, consistency loss, and evolutionary loss. Compared with other methods, this invention significantly improves the accuracy of classification results with a small amount of labeled data.

The application belongs to the technical field of databases, and specifically discloses an efficient GPU near neighbor graph index construction method based on data locality, which comprises the following steps: dividing index original data into multiple subsets, and performing data processing on the multiple subsets to obtain rearranged data; constructing subgraphs corresponding to each subset in the rearranged data, and constructing a graph index based on the subgraphs; and refining the structure of the graph index to obtain a final graph index. The method can effectively alleviate the problems of insufficient memory resources and excessive calculation cost in the subgraph merging stage, significantly reduce the construction cost of the graph index, and effectively extend to large-scale data sets to improve scalability.

This invention relates to the field of voice interaction technology, and more particularly to a voice dialogue system based on a large model, comprising: a feature extraction module for acquiring samples containing multi-turn dialogue information and extracting multi-dimensional features; an alignment and processing module for achieving fine alignment through cross-modal alignment matrices and path search, and calculating a contextual consistency score based on role transition states; a stability calculation module for determining sample stability scores based on a teacher-student model mechanism; an anomaly assessment module for constructing a sample nearest neighbor graph based on semantic embedding vectors, calculating structural anomaly scores and sample retention values, and rescuing correctable samples by supplementing with nearest neighbor information; and a training module for selecting voice dialogue training samples according to sample retention values ​​and training the model. This invention provides high-quality samples for model training through a multi-dimensional quality assessment and data filtering mechanism, improving the accuracy and fluency of voice dialogue.

The invention discloses a federated graph network fault diagnosis method, device and equipment fusing nearest neighbor graph composition and adaptive clustering, and aims to solve the problems of poor data privacy, weak sample relevance and insufficient model generalization ability in the prior art. On the premise of strictly protecting data privacy, relation information of graph structure data is effectively utilized, an incidence relation between samples is constructed through a nearest neighbor graph composition technology based on a federated learning framework, structured features are extracted through a graph neural network, and then the problem of non-independent identically distributed data is solved through a self-adaptive clustering technology; and finally, high-precision and high-generalization-ability fault diagnosis is realized.

The application discloses a communication product configuration method based on multi-modal analysis and vector matching, and particularly relates to the technical field of intelligent configuration of communication products, and is used for solving the problem that the efficiency of manually reading documents and manually selecting and configuring is low, errors are easy to occur, and it is difficult to trace back when the customer demand is proposed in the form of multi-source and multi-format documents, the field caliber is complex, and the rule constraints are numerous in the group and government and enterprise customer scenarios of operators. By taking the field dictionary and structured constraints as the basis, the multi-source and multi-format communication product demand document is sequentially converted into demand parameters, demand configuration graphs and multi-channel feature vectors, and vector matching of rule perception such as field mandatory, mutual exclusion of rights and interests, region, contract period and cost is performed on the hierarchical near neighbor graph index in combination with the template graph, so that the communication product configuration scheme meeting the business constraints is automatically output under the unified caliber, the workload of manually reading documents and selecting by experience is significantly reduced, and the accuracy and consistency of the configuration result in the complex scenario are improved.

Disclosed are a method and apparatus for learning equivariant and invariant representations for rotation of an image based on a graph convolutional network. The method of learning an equivariant and invariant representation for rotation of an image based on a graph convolutional network performed by a computer device includes learning an equivariant representation for rotation of an image by using a self-weighted nearest neighbors graph convolutional network (SWN-GCN); and finally obtaining the equivariant representation for the rotation of the image obtained from the self-weighted nearest neighbors graph convolutional network as an invariant representation of the rotation of the image by using permutation invariance of global average pooling (GAP).

This application relates to the field of new pollutant source tracing technology in water bodies, and provides a method and related equipment for predicting new pollutant sources. The method includes: acquiring mass spectrometry data from multiple locations in a target water body area; constructing a material screening matrix for each location based on all mass spectrometry data; constructing a heterogeneous graph structure based on the material screening matrix, and constructing a nearest neighbor graph for the target water body locations; calculating the prior knowledge view embedding corresponding to the target water body area based on the heterogeneous graph structure, and calculating the data similarity view embedding corresponding to the target water body area based on the nearest neighbor graph; and performing a fusion prediction on the target water body area based on the prior knowledge view embedding and the data similarity view embedding to obtain the new pollutant source tracing prediction result for the target water body area. The method of this application can improve the accuracy and interpretability of new pollutant source tracing prediction in water bodies.

The invention provides an intelligent identification method and system for abnormal auditing data of a hospital. The intelligent identification method comprises the steps of obtaining a multi-dimensional data set of the hospital, calculating a local outlier factor LOF of each data record, distributing a differential privacy budget and generating a privacy data set; constructing a weighted K nearest neighbor graph based on the privacy data set and the LOF value, and performing community division on the graph; calculating weighted intermediary centrality, community connection strength and average path length of each node, fusing the weighted intermediary centrality, the community connection strength and the average path length into a structural anomaly score, and screening data records of which the scores are higher than a first threshold value as first-level candidate anomaly data; for each piece of first-level candidate abnormal data, mapping the data structure abnormal score into a time window, extracting a time sequence track, and calculating an average time sequence track of the community as a prototype track; outputting a time sequence similarity distance; constructing a confirmation threshold value, wherein the threshold value is adjusted based on community structure characteristics, data differential privacy budget and a structure anomaly score; and when the time sequence similarity distance is greater than the threshold value, identifying the data as abnormal data.

The invention provides a federated learning-based neighbor graph indexing method and system and electronic equipment. The method comprises the following steps: each client participating in federated learning trains a local performance prediction model used for predicting the retrieval performance of the neighbor graph based on a local sample data set of the client to obtain a corresponding local model parameter; the federation server participating in federation learning obtains local model parameters and test data sets of all the clients, a global test data set is obtained based on the test data sets of all the clients, and contribution quantized values corresponding to all the clients are calculated through the global test data set; and the federation server side aggregates all the local model parameters based on the contribution quantized value to obtain global model parameters, so that each client side updates a local performance prediction model based on the global model parameters, and constructs a corresponding neighbor graph index for the to-be-queried data set by using the updated local performance prediction model. According to the technical scheme, neighbor graph intelligent indexing supporting distributed data can be realized.

The application discloses a cell space polarity quantification analysis method based on optimal transmission theory, and the steps include: 1) performing slice quality filtering, feature screening, total amount standardization and logarithmic conversion on an original space omics count matrix to obtain a standardized expression matrix; 2) performing principal component analysis on RNA and protein data respectively to construct a low-dimensional embedding for similarity calculation; 3) calculating the difference between the actual distribution of each molecule and the uniform distribution based on the first Wasserstein distance to realize continuous gradient quantification along the plant-animal axis; 4) constructing a k-nearest neighbor graph based on the distance between molecules to identify molecular communities with common spatial positioning patterns; and 5) calculating the matching significance of the observed distribution and the ideal pattern, and then controlling the false discovery rate through the Benjamini-Hochberg procedure to screen out molecules with significant gradient distribution characteristics. The application significantly improves the spatial distribution pattern recognition and quantification ability.

The invention provides a multimodal transport graph structure generation method and system fused with online map service, and the method comprises the steps: S1, adding site nodes in a railway transport network and port nodes in a shipping network into a graph structure, and forming a basic topology skeleton; s2, the two ends of the belt are directly connected to the associated industrial point location, railway station or port node; s3, for the road nodes in belt connection, searching the nearest neighbor nodes in the graph structure through the geographic buffer area to establish connection; s4, dynamically connecting the industrial point locations which are not accessed into the graph structure to nearest neighbor graph nodes through a geographic buffer area; and S5, calling an online map API (Application Program Interface) for the connecting lines generated in the step S3 and the step S4 to carry out actual path planning, and updating the weight in the topology. According to the method, through dynamic buffer connection and an API path post-processing mechanism, the generation efficiency of the multimodal graph topology, the path planning accuracy and the system expansibility are remarkably improved.

The invention discloses a vector retrieval method and device for a balanced graph topological structure, and relates to the field of vector retrieve.The method comprises the steps that firstly, a graph is initialized into a minimum directed complete graph with any vector meeting the target in-out degree, a gradual vector adding mode is adopted, a forward optimization multi-level edge selection strategy and an edge refining mode are combined, and the minimum directed complete graph with the target in-out degree is obtained; and finally, constructing a neighbor graph of a balanced graph topological structure, and then performing greedy approximate nearest neighbor retrieval on the neighbor graph for the query vector to obtain a query result. According to the method, the balance of the topological structure of the neighbor graph is realized, the problem of deflection of the topological structure of the neighbor graph caused by unbalanced vector distribution is eliminated, and the accuracy and efficiency of vector retrieval are effectively improved. Compared with a current mainstream neighbor graph composition method, the composition method does not depend on reverse edge adding operation to update neighbors for existing vectors in the graph, and composition complexity and time expenditure are reduced.

The invention relates to a graph database query method and device based on vector indexes and electronic equipment, and belongs to the technical field of graph databases. The method comprises the steps of determining node embedding vectors corresponding to nodes in a graph database based on original feature vectors of the nodes in the graph database and a topological connection relation of the nodes in the graph database; determining a nearest neighbor node of each node based on the node embedding vector to obtain an approximate nearest neighbor graph; and in response to a received query request, searching in the graph database based on the query request and the approximate nearest neighbor graph to obtain a query result corresponding to the query request, the query request including a query vector and at least one graph operator filtering condition. And the index retrieval accuracy in the structured and unstructured multi-dimensional constrained graph database retrieval scene is improved. In this way, graph structure query and similarity retrieval can be fused, semantics and local contexts of the processed nodes are dynamically fused, and the accuracy of query results is effectively improved.

The application discloses a backdoor defense method of text generation graph diffusion model based on Ollivier-Ricci curvature, and belongs to the technical field of artificial intelligence security. The method constructs a concept database to store text encoder features corresponding to predefined concepts and intermediate layer features of a denoising network; a to-be-detected text prompt is input into a diffusion model to extract text encoder features and intermediate layer features of the denoising network, and reference features are obtained based on semantic similarity matching. A K nearest neighbor graph is constructed based on the text encoder features, Ollivier-Ricci curvature is calculated, and abnormality is determined. If no abnormality is detected, global features are constructed at multiple time steps, curvature is calculated in the same way, and an abnormality proportion is counted. When abnormality is detected at any stage, fine-grained curvature analysis is performed on local intermediate layer features at multiple time steps, and whether a backdoor trigger is contained is determined according to the abnormality proportion, so that backdoor defense is achieved. The method is suitable for backdoor detection of various diffusion models and samplers.

The invention discloses a drug screening-oriented small molecule graph similarity calculation and retrieval method and device. Comprising the following steps: analyzing a chemical structure of a small molecule compound, converting into a standard small molecule graph, and constructing a graph pair data set; extracting structure and semantic information in the graph by utilizing a graph isomorphic network, and generating node and edge embedding; based on node and edge embedding, semantic similarity and structure consistency information are fused, and an alignment result is calculated; performing embedding exchange fusion and secondary coding, and enhancing embedding re-interaction to obtain alignment cost; carrying out model training by utilizing a mean square error of a true value and a predicted value; on the basis of an anchor point atlas selected iteratively and node embedding pooling, generating graph embedding containing global features; constructing a weighted HNSW neighbor graph index according to graph embedding and positive and negative samples; and carrying out rough arrangement and fine arrangement collaborative retrieval by utilizing a weighted HNSW neighbor graph index and a graph similarity calculation model to obtain the most similar K small molecules. The efficiency and precision of small molecule drug screening are improved.

The application belongs to the technical field of biometric recognition, and specifically discloses a K-nearest neighbor graph iteration vein recognition method based on deep learning, which comprises the following steps: obtaining a feature vector of a to-be-tested vein image and a feature vector set of a vein image training set, a distance between the feature vector and each feature vector in the feature vector set, and selecting k feature vectors with the smallest distance from the feature vector set; inputting the k+1 feature vectors and corresponding label information into a graph iteration model to iteratively learn the correlation between the feature vectors and simultaneously predict the label information of the to-be-tested vein image, so as to realize the final prediction of the classification recognition of the to-be-tested vein image. The application also discloses a vein recognition system based on the method. The application solves the technical problem that the prior art does not fully utilize the distance information and correlation information between the feature vectors of corresponding samples, so that the prediction label accuracy is not high.

This invention relates to the field of hyperspectral remote sensing image classification, and specifically to a hyperspectral band selection method based on an improved iterative greedy algorithm. The method includes: obtaining clustering results of hyperspectral image data using a k-means clustering algorithm; generating a nearest neighbor graph between bands by calculating the Euclidean distance between bands within each cluster; extracting the bands in each cluster with the closest Euclidean distance to the cluster center, and using the extracted band combinations as initial solutions; using the information entropy and mutual information of the band combinations as the objective function of an improved iterative greedy algorithm based on simulated annealing; iteratively optimizing the initial solution through destructive reconstruction and neighborhood search operations to find the maximum value of the objective function; the band combination corresponding to the maximum value of the objective function is the optimal solution. This method effectively improves the efficiency and accuracy of hyperspectral image band selection and classification.

The specification discloses a similar data search method, system and electronic equipment, which can efficiently and accurately determine similar data, reduce computing cost and improve search efficiency. The method comprises: acquiring specified query data and a given data set, the given data set comprising a plurality of basic data; performing feature vector extraction on the specified query data and the plurality of basic data to generate corresponding multi-dimensional feature vectors, determining a query point corresponding to the specified query data and a plurality of data points corresponding to the plurality of basic data in a multi-dimensional vector space; generating corresponding vector approximate coding by interval division of a plurality of dimensions of the multi-dimensional vector space; constructing a nearest neighbor graph for the plurality of data points, performing nearest neighbor search in the plurality of data points based on the nearest neighbor graph and the vector approximate coding, determining the nearest neighbor data point of the query point, and thereby determining similar data.

The invention discloses a tobacco field contour extraction method based on a graph neural network, and belongs to the field of artificial intelligence and image processing, and the method comprises the steps: carrying out the preprocessing of a tobacco field remote sensing image; inputting the preprocessed image into an encoder to extract multi-scale features to obtain a high-level feature map; dividing the high-level feature map after space flattening into a plurality of sub-regions to construct a local k neighbor map, and calculating center feature vectors of the sub-regions; constructing a full-connection global graph based on the central feature vector, and performing cross-regional feature interaction based on a graph attention mechanism; the updated global features are broadcasted to the corresponding sub-regions, and global-local feature fusion is carried out in combination with neighbor feature differences; and reconstructing the fused features into an image feature map, outputting a tobacco field segmentation mask map from the image feature map through a decoder, and obtaining a pixel-level extraction result of the tobacco field contour based on the tobacco field segmentation mask map. According to the method, the global perception and boundary connectivity modeling capability of the model on the outline of the tobacco field is effectively enhanced.

The application relates to the technical field of medical image processing, in particular to a multi-modal lower limb medical image registration method and system based on manifold learning. The method first performs left-right separation on the whole lower limb CT to obtain independent left and right candidate regions; the local MRI is extracted to obtain a limb mask and determine the side. The local image is resampled with the whole resolution as a reference to realize spatial resolution unification. Local block high-dimensional features are extracted, a near neighbor graph is constructed, and a graph Laplace eigenvalue equation is solved to obtain a multi-channel manifold feature map. The representation ambiguity of manifold embedding is eliminated through symbol correction and orthogonal alignment, then Fourier-Mellin transformation is adopted to realize coarse registration, and local refinement is realized in a limited neighborhood. The application completely separates and independently registers the left and right sides, avoids interference of the opposite structure, realizes stable positioning and accurate alignment of the local MRI in the whole lower limb CT.

This invention relates to a range-filtering nearest neighbor search method, electronic devices, readable storage media, and program products for timely knowledge retrieval of large language models, encompassing computer data retrieval, vector databases, and high-dimensional similarity search technologies. It addresses the challenge of existing technologies in large language model retrieval enhancement scenarios, where handling vector retrieval with ordered attributes makes it difficult to simultaneously balance index size and query efficiency. The method includes database index construction and query phases. By constructing a segmented nearest neighbor graph index based on ordered attribute values ​​and combining it with user-given attribute ranges for efficient vector similarity search and filtering, the timeliness and accuracy of retrieval during the query phase are improved. This invention is primarily used to retrieve external knowledge applied to large language models.

This invention discloses a multimodal data fusion method based on local cross-attention and contrastive learning, belonging to the fields of deep learning and space biology. The method effectively fuses spatial transcriptomic and proteomic data by designing a local cross-attention module and a hard negative sample mining strategy, improving spatial domain recognition accuracy. The local cross-attention module constructs a k-nearest neighbor graph using cell coordinates, restricting cross-omics interactions to spatially adjacent cells and reducing noise. A dual-channel spatial regularization module ensures spatial structure consistency, and hard negative sample mining enhances the ability to capture subtle phenotypic differences. Experimental results show that SpaMOCA performs excellently in spatial domain recognition, rare cell subpopulation identification, and tissue boundary delineation, exhibiting strong robustness against noise and batch effects. This method provides a powerful tool for a deeper understanding of cell biology and disease development.

The application discloses a learner emotion analysis method and device based on a cross-time-space graph representation, and the method comprises the following steps: dividing a plurality of input learner expression video frames into a plurality of subblocks of equal size and without overlapping, and performing graph node feature embedding processing on the subblocks to obtain initial features of the graph nodes; constructing a cross-time-space nearest neighbor graph structure of each subblock; classifying output emotion space-time graph features of a learner emotion cross-time-space graph representation module to predict the emotion category of the input learner expression video frame; the application can divide the local area of the learner expression video frame without using additional auxiliary information, so that the cross-time-space graph convolution network can directly process the learner expression video frame data; in addition, the application can obtain the space-time correlation graph representation between different regions and different video frames of the learner expression through the cross-time-space graph construction and the graph convolution operator with correlation perception, and can be widely applied to the technical field of computers.