230 results about "Information propagation" patented technology

The invention relates to a gated multi-graph convolution perception modeling method for traffic flow prediction. The method integrates multi-graph structure construction, gating graph convolution and time feature extraction, and aims to solve the problems of strong time fluctuation and heterogeneous spatial relationship in traffic data. The method comprises the following steps of: firstly, respectively constructing a geographic map and a semantic map according to the maximum mutual information measurement between the spatial distribution information of a sensor and historical traffic data; and then, designing a dual-adaptive gating graph convolution module, and dynamically adjusting an information propagation path of a multi-graph structure by introducing an attention mechanism and a gating factor, thereby improving the modeling performance of the model on spatial isomerism dependence. On the time dimension, a time sequence interactive sensing module is constructed in combination with multi-scale causal convolution and an attention mechanism, time dependence characteristics of a short period and a long period are captured, and fusion and expression of time characteristics are completed. According to the method, the modeling precision and stability of the traffic prediction model in a complex traffic scene can be effectively enhanced, and the method has relatively high practical application value.

The embodiment of the invention provides a computer communication method and system based on the Internet of Things, and the method comprises the steps: constructing a multi-level system architecture, carrying out the preprocessing and feature extraction of an original data flow, recognizing the data characteristics through a time sequence analysis method, and constructing a dynamic data model; deploying a monitoring agent at a transmission node to collect network performance indexes in real time, and constructing a network quality evaluation model; for an incomplete sensor data flow, prior probability distribution is constructed based on a dynamic data model and a network quality grade, and an optimal estimation value of missing data is calculated by adopting a Bayesian reasoning framework and an iterative algorithm; establishing a mapping relation between a network state and an optimal parameter through reinforcement learning to realize self-adaptive adjustment and optimization; and grading the data according to reliability, extracting high-reliability data points as anchor points, designing an iterative refinement algorithm to realize information propagation, and fusing to obtain a complete sensor data stream. According to the method, the problems of poor data recovery accuracy, static parameter configuration and insufficient adaptability in a complex network environment are solved.

The invention relates to the technical field of network information propagation prediction, and discloses a topic propagation prediction method and system based on multi-dimensional feature fusion, and the method comprises the steps: extracting multi-dimensional features, inputting the multi-dimensional features to a decoder layer of Transform for fusion, and constructing a prediction model for topic propagation prediction; the multi-dimensional features comprise user comprehensive influence, text emotion features, time dynamic features and user interaction behaviors. According to the method, through collaborative optimization of the dynamic user modeling module, the multi-modal feature fusion module and the intelligent time sequence analysis module, the social network propagation prediction accuracy is remarkably improved, and accurate prediction of the topic propagation trend in the social network is achieved.

The invention relates to an information-oriented network malicious behavior message backtracking generation method, and relates to the technical field of network security, and the method comprises the steps: constructing a network node topology according to a target power grid network; constructing a node anomaly identifier of each network node and carrying out node anomaly detection to obtain a node detection result; when an abnormal node exists, analyzing information guidance from the abnormal node based on an information flow direction, and determining an information propagation path; and based on the information propagation path, reconstructing a complete attack chain of the malicious behavior according to a time sequence, and generating a malicious behavior message backtracking result. According to the method, the problems of inaccurate abnormal node positioning, incomplete information propagation path tracing and disordered attack chain time sequence in traditional network malicious behavior backtracking are solved, and the integrity, accuracy and timeliness of malicious behavior backtracking are improved.

The invention relates to the field of enterprise risk assessment, and discloses an enterprise risk assessment method and system based on a large language model and high-frequency graph convolution, and the method comprises the steps: S1, building an enterprise risk association graph: generating a structured joint weight between enterprises, and building the enterprise risk association graph; fusing the semantic weight of the large language model and the structured joint weight into a final edge weight; jointly inputting the fused risk map and the multi-modal feature into two-stage attention to generate a multi-modal attention feature; and S2, enterprise risk assessment: on the enterprise risk association graph endowed with the edge weight, carrying out information propagation and aggregation by adopting high-frequency graph convolution. According to the method, edge-level gating and fusion edge weight guide risk information to be preferentially diffused along a more credible and more relevant path, high-frequency signals such as financial mutation, negative news and major events can be quickly captured, and blind diffusion of the information in a network is reduced. And the amplification effect of market systematic impact is reduced.

The invention discloses a recommendation method based on semantic enhancement and a heterogeneous hypergraph network. The recommendation method comprises the following steps that semantic information in an explicit feedback text is coded and serves as an auxiliary signal of a recommendation task; classifying the articles into predefined categories by using LLM, constructing article-category association, and mining a potential co-occurrence relationship of the articles; constructing a heterogeneous hypergraph network; spreading and aggregating hypergraph information; performing semantic alignment and model training; and performing recommendation calculation based on the final representation of the user and the representation of the article, and outputting a recommendation result. According to the method, through technical paths of semantic coding, hypergraph modeling, information spreading and alignment supervision, comment semantics of LLM coding are aligned to the recommendation space through GAE, and the problem of degradation of LLM representation in the recommendation space is effectively solved.

The invention discloses an intelligent decision-making method and system based on deep learning, and relates to the field of data processing. The method comprises the steps of obtaining place data and corresponding attribute data of a to-be-decided project; establishing a fuzzy relation matrix between the places and the attributes; generating a network model reflecting trust relationship strength among users through trust propagation operation of the graph neural network; identifying a community structure containing a community overlapping degree through a community discovery algorithm; and according to the trust relationship strength between the users and the community overlapping degree, calculating an influence weight of a decision maker, forming a group consensus through a robust optimization method, and generating a decision result of the project to be decided. Aiming at low network relation modeling precision caused by multi-source heterogeneous data in bus station layout decision making in the prior art, the method and the device have the advantages that the network relation modeling precision is low through accurate modeling of an information propagation path in a complex trusted network, effective dimension reduction representation of a high-dimensional feature space and robust optimization solution in an uncertain environment; therefore, the calculation precision and robustness of the bus station layout intelligent decision-making system are improved.

The invention discloses a multi-modal fine-grained semantic alignment method and device based on a graph neural network, and relates to the field of multi-modal deep learning. Firstly, deep feature extraction is performed on input multi-modal original data, and then word-level text features and local image features are constructed into a cross-modal graph structure. And performing weighted aggregation on node neighborhood information of the cross-modal graph structure through the graph attention network. And finally, carrying out weighted fusion on the text alignment features and the image alignment features. According to the cross-modal feature fusion method, the word-level text features and the local image features are uniformly abstracted into the graph structure nodes for refined alignment, a more accurate cross-modal semantic corresponding relation can be captured, the heterogeneity problem in expression modes and semantic structures is effectively relieved, and the accuracy and reliability of cross-modal feature fusion are improved. The graph attention network can adaptively adjust the weight distribution of information propagation, highlights the effect of key features in the alignment process, and ensures that the model makes full use of important semantic relationships.

The invention relates to a graph convolution multi-modal dialogue emotion recognition method based on emotion dimension compensation. The method belongs to the field of multi-modal emotion recognition. Comprising the following steps: respectively extracting original features of texts, audios and visual modalities by utilizing a pre-training model; detecting a mode missing condition, and generating a pseudo feature by using an available mode feature and context information; the multi-modal features are mapped to a VAD three-dimensional space, and prediction and enhancement are carried out; the global level is based on VAD similarity to connect cross-utterance nodes, and emotional consistency is quantified to capture long-distance interaction. In the local level, information spreading and node updating are carried out by calculating the similarity between different modal features in the same utterance; neighbor information is accumulated through multiple layers of image volumes, node weights are dynamically adjusted to suppress noise propagation, and finally utterance-level emotional representation is generated by fusing multi-modal features through mean pooling operation. According to the method, the emotion recognition performance in a modal missing scene is remarkably improved.

The invention discloses an efficient graph comparison learning method for social recommendation. As an emerging self-supervised learning normal form, graph contrast learning is excellent in response to data sparseness and cold start due to the fact that the graph contrast learning can effectively capture similarity and heterogeneity characteristics in a graph structure, although the learning normal form achieves a good effect in a recommendation system, the graph contrast learning can be used for solving the problems of data sparseness and cold start. However, the method still faces three defects: (1) average neighbor aggregation and a non-adaptive representation reading mechanism are adopted in a message propagation process, and high-quality node representation is difficult to learn; (2) a visual angle is enhanced by depending on a random disturbance generation graph during intervention of comparative learning, which may destroy the inherent structure of graph data and further weaken the accuracy of the model; and (3) equally treating all observation samples during parameter optimization, and neglecting the difference influence of positive samples in different training stages. Specifically, aiming at the problems, the invention provides an efficient graph contrast learning method (EGCL for short). The method comprises the following steps: firstly, designing a graph adaptive propagation module, improving an information propagation rule of a graph neural network by referring to a thermonuclear thought and an attention mechanism, and realizing differentiated aggregation of neighbor nodes by adopting a learnable weight distribution strategy; secondly, designing a double contrast learning normal form which does not need graph enhancement, and realizing mutual promotion of node characterization through intra-domain contrast learning (inter-CL) and inter-domain contrast learning (inter-CL); and finally, introducing a sample weight adaptive efficient optimization algorithm, converting the training process into a double-layer optimization problem, and adaptively adjusting the contribution degree of each sample to model optimization in different stages.

The invention relates to the field of influence measurement, and discloses a social media-based content influence measurement method, which comprises the following steps of: constructing a social network graph according to collected user social relation data; each user is a node, interaction between the users is an edge, and features in the social network are analyzed; modeling a propagation path of the information by adopting a propagation model, and weighting the weight of the information propagation according to the social relationship between the users; establishing a weight model, endowing each edge in the propagation path with different weights according to the strength of different social relationships, and strengthening the influence of the core social group on content propagation; evaluating the propagation breadth and propagation speed of the content in the social network, analyzing the participation behavior of the user, and evaluating the propagation effects of the content in different social circles; and training the data by using a machine learning algorithm, and optimizing the propagation effect prediction model. The method has the advantage of accurately evaluating the actual propagation effect and influence of the content.

The invention belongs to the technical field of pipeline safety assessment, and discloses a multi-dimensional information interconnection and autonomous evolution collaborative pipeline full-state safety assessment method. And capturing a high-order relationship of data through double hypergraph reasoning of the instance-level hypergraph and the modal-level hypergraph to realize efficient interconnection. According to the method, mode-level and instance-level hypergraph information features are extracted through hypergraph information propagation, high-order correlation is mined through double-graph information aggregation, cross-mode and cross-instance consistency information and exclusive information are output after feature recombination, multi-dimensional data deep fusion is promoted, and high-quality data support is provided for follow-up pipeline full-state safety assessment. A two-stage autonomous evolution mechanism of intra-class progressive calibration and inter-class knowledge migration is respectively adapted to slight fluctuation and significant change scenes of the deep sea environment: precise adaptation of environment perturbation is realized through dual-branch feature extraction and dynamic weight adjustment in a domain; model parameter dynamic optimization is completed between domains through spatial-temporal feature clustering and cross-domain knowledge migration, and dynamic environment self-adaption can be achieved without manual intervention.

The invention relates to the technical field of epilepsy detection, in particular to a self-supervised graph neural network epilepsy detection method based on Transform. The method comprises the following steps: S1, preprocessing original EEG data, and constructing an EEG graph; s2, building a graph neural network based on DCTran, and respectively capturing a space-time dependency relationship of the EEG signal through diffusion convolution and a Transform structure; and S3, training the model through spatio-temporal joint complementary double-branch pre-training based on the self-supervised prediction pre-training task and the mask reconstruction pre-training task. According to the epilepsy detection method based on the self-supervised graph neural network of the Transform, provided by the invention, EEG data is modeled into a graph structure, and a diffusion convolution space-time network DCTran based on the Transform is provided; the diffusion convolution accurately captures a complex spatial relationship between the electrodes through multi-order neighbor information propagation; and meanwhile, global context information of the EEG signal is fully utilized, and DCTran is introduced into a Transform structure to model a time sequence, so that the long-distance time dependency relationship in the signal is effectively captured.

The invention provides a multi-mode sentiment analysis method and device based on sentiment dynamic tracking. The method and device can be used for automatic quality inspection of call scenes in the electric power customer service industry. According to the method, a multi-mode sentiment analysis technology is adopted, the defect of a single mode is overcome, sentiment features in voice and sentiment information in a text are fused through a graph neural network (GNN), and a graph structure effectively integrates internal relations and interaction among different modes by means of a unique topological structure and a dynamic information spreading mechanism. In order to fit a customer service-customer multi-round dialogue scene, an emotion dynamic tracking module is added, splicing features between utterance pairs are constructed, the module can effectively and dynamically track changes of customer emotional states, and even if the customer emotional states suddenly change, flexible recognition can be achieved. The module not only deepens the understanding of the model on the mood dynamics, but also restrains the network parameters of the main task through a strategy similar to comparative learning.

The invention provides a low-code and component intelligent collaborative application construction method based on a dual-drive architecture, and belongs to the technical field of computer programming. The method comprises the following steps: establishing a component dependency digraph, optimizing a component collaborative relationship by adopting topological sorting, graph coloring and a minimum spanning tree algorithm, collecting multi-dimensional performance indexes to establish a monitoring matrix, processing monitoring data by utilizing a collaborative optimization model containing a gating information propagation mechanism, and generating a dynamic scheduling decision vector; the load change of the system is judged through a change rate threshold equation, quick or slow response adjustment is executed, and finally, a low-code component library and a visual configuration interface are established to realize intelligent collaboration and dynamic resource allocation among components. The technical problems that the collaboration efficiency among low-code platform components is low and intelligent dynamic scheduling cannot be achieved according to system load changes are solved.

The invention relates to the technical field of thermal power plant abnormal data processing, in particular to a thermal power plant production monitoring system network security data set abnormal data identification and cleaning method, which comprises the following steps: mapping a spatial-temporal feature vector to a preset thermal power plant multi-level causal graph to obtain a data causal graph; performing information propagation and node updating on the data causal graph, and calculating an abnormal score of each node in a preset thermal power plant multilevel causal graph; identifying abnormal data according to the abnormal score and the network security data set; and cleaning the abnormal data according to the abnormal type and a preset data cleaning strategy. According to the method, correlation modeling of data on time and a system topological structure is realized through a multi-level cause and effect graph; an abnormal score is calculated through information spreading and node updating, and an abnormal event in the network security data can be accurately identified; the abnormal data is cleaned based on the preset data cleaning strategy, redundant, wrong and abnormal data can be effectively removed, and meanwhile key risk information is reserved.

The invention relates to the field of network public opinion management and control, in particular to a cross-modal network emergency command system based on a network space model. The method comprises the following steps: acquiring a network public opinion multi-modal data stream, positioning a malicious public opinion propagation source account based on the network public opinion multi-modal data stream, and performing network space cross-layer holographic portraying based on the malicious public opinion propagation source account to generate a source account holographic portraying information set; on the basis, multi-modal feature extraction and network space behavior model construction are carried out on hidden negative guidance behaviors, and an implicit negative information propagation mode model is generated; on the basis, potential malicious account active tracking and risk propagation network mining are carried out, and a potential risk propagation network information set is generated; on the basis, multi-level differential intervention strategies are generated and executed, and a network public opinion intervention strategy instruction set is generated. In the network public opinion emergency command process, pollution of false information to network space is reduced, and construction of healthy and sustainable network ecology is assisted.

The invention discloses a method for maximizing the influence of information competition propagation in a social network, and relates to the technical field of information propagation. Comprising the following steps: constructing an extended independent cascade model; aiming at each initial information seed node, constructing an influence subgraph taking the initial seed information node as a root node; constructing an alternative target node set by taking the initial information seed node as a starting node; according to the frequency of occurrence of the root node and the non-root node in the influence subgraph of the originating node, obtaining a contribution degree; selecting an initial information seed node having the maximum influence gain on adjacent nodes in the three-hop range from the alternative node set to obtain a plurality of intermediate information seed nodes, and determining the comprehensive influence of each intermediate information seed node; and arranging the comprehensive influence of the intermediate information seed node in a descending order, and selecting a final information seed node. According to the method, the target node in the competitive environment can be accurately identified, and the propagation of competitors is restrained while the information of the target node is maximized.

The invention relates to the technical field of social network information propagation prediction, in particular to an event propagation prediction method based on group influence. The method comprises the following steps: constructing a social network graph of target event participating users, calculating an activation probability matrix among the users by using historical propagation data, and performing social graph structure enhancement based on a threshold value to obtain a de-noised social graph; applying a graph neural network on the de-noised social graph and introducing information bottleneck constrained self-supervised contrast learning optimization node embedding, and reserving features useful for group division and propagation prediction; dynamically dividing user groups related to propagation according to the similarity and propagation context of de-noising node embedding to form a group-level propagation sequence; based on the group representation, utilizing an attention mechanism and a group relation graph to carry out modeling on propagation influence among the groups to obtain global group representation of the event; global group representation is combined with candidate user embedding, propagation probability distribution is calculated, and future participating users of corresponding events are predicted.

The invention discloses a geometric topology fusion-based part assembly method, which comprises the following steps of: acquiring boundary representation models of at least two to-be-assembled CAD parts, and converting the boundary representation model of each CAD part into a structured heterogeneous geometric topology graph; performing feature coding on each node in the heterogeneous geometric topological graph to form node feature representation; inputting the node feature representation into a graph attention reasoning module, and obtaining a node embedding representation representing a geometrical relationship and a topological relationship in the CAD part through multi-layer residual graph structure information propagation and feature aggregation; based on the node embedding representations of the different parts, calculating association scores of node pairs among the different parts; and performing supervised training on the feature coding module and the graph attention reasoning module by using the labeled assembly constraint sample data. According to the method, fine geometric features can be extracted from a boundary representation B-Rep model, and a complex topological dependency relationship is inferred, so that robust assembly constraint inference is realized.

The invention provides a public opinion thermodynamic statistical method and system based on multi-AI agent collaboration, and the method comprises the steps: collecting public opinion data from a plurality of heterogeneous data sources, and carrying out the cleaning and standardization processing of the public opinion data, and obtaining the standardized public opinion data; identifying public opinion transaction events in the standardized public opinion data; for a public opinion transaction event, constructing a multi-dimensional thermal factor based on a large language model; wherein the multi-dimensional thermal factors comprise a volume factor based on an information propagation map, an emotion intensity factor based on text fine-grained emotion analysis and a source weight factor based on information source influence evaluation; according to the characteristics of the public opinion transaction event, configuring weights of the volume factor, the emotion intensity factor and the source weight factor; performing weighted fusion on the multi-dimensional thermal factors after weight configuration to obtain a comprehensive thermal value; and comparing the comprehensive thermodynamic value with a preset thermodynamic threshold value to obtain the public opinion thermodynamic level, thereby improving the accuracy of public opinion thermodynamic statistics.

The invention discloses a forum public opinion monitoring method and device and electronic equipment. The method comprises the following steps: acquiring forum event data and constructing a target propagation tree; and inputting the target propagation tree into a bidirectional hierarchical semantic interaction neural network model to obtain a credibility classification result of the forum event, thereby realizing forum public opinion monitoring. The bidirectional hierarchical semantic interaction neural network model is a deep learning model for outputting forum public opinion event credibility classification based on a tree propagation structure through an interaction mechanism of top-to-bottom multi-branch semantic coding and bottom-to-top global semantic representation in combination with transverse semantic feature extraction between same-level nodes. According to the method, multi-dimensional semantic information of a propagation tree can be accurately covered, and semantic features of information propagation can be comprehensively reflected, so that the core capability of public opinion monitoring is remarkably improved, a high-precision credibility classification result is output, the capability of identifying abnormal public opinions and normal public opinions is enhanced, and the accuracy and reliability of monitoring are practically improved.

The invention belongs to the technical field of false information spreading and screening, and particularly discloses and provides a network news false information spreading and screening method, which comprises the following steps: screening key evaluation dimensions based on historical data, determining weights, generating evaluation rules, and determining the weight of each evaluation rule; determining a conventional or emergency monitoring mode according to the amplification of the forwarding quantity of the information to be screened and the number of matched words in a preset keyword library, identifying an initial publishing source through a propagation path, evaluating source authority, content consistency and propagation abnormality in the conventional mode, and outputting comprehensive credibility; in the emergency mode, the emergency credibility is directionally and rapidly evaluated and output, and finally, the credibility is matched with the early warning interval to obtain an early warning level; according to the method, the source authority is assessed based on the source information, the content consistency is judged through the text content, the propagation abnormality is assessed through the propagation data, the comprehensive credibility is output in combination with the assessment dimension rule, and multi-dimensional collaborative assessment of the source, the content and the propagation is achieved.

A network device may have an interface configured as a trunk interface. The network device may receive information indicative of the state of the interface for enabling or disabling VLAN information propagation using the interface. Based on the received information, the network device may appropriately configure the interface to participate in a VLAN information propagation protocol.

The invention discloses a social media key user identification system based on information non-uniform propagation characteristics, and relates to the technical field of key user identification, and the system comprises a dynamic module which is used for calling an information propagation model through a weighted directed graph of N time windows, simulating information propagation, obtaining initial influence scores of a user in the N time windows, and obtaining initial influence scores of the user in the N time windows; and a cross-layer module which constructs a multi-layer network, quantifies the influence of the user in different layers in the multi-layer network based on the dynamic influence score, obtains a cross-layer influence score, and performs weighted average on the initial influence score of the user in N time windows to obtain a dynamic influence score. The recognition module is used for calculating comprehensive scores according to the initial influence scores, the dynamic influence scores and the cross-layer influence scores of the users, and the comprehensive scores are arranged in a descending order to recognize key users; through the key user identification method, the social platform can automatically identify key users really having transmission force and guiding force.

The invention relates to the technical field of social network application, in particular to a social media viewpoint evolution simulation method and device based on coupling dynamics, and the method comprises the steps: extracting event topic interaction network data based on social network public information, carrying out the statistics of user historical interaction behaviors, and determining an information transmission network and an internal association network according to the data, the two forms a multi-layer association network; determining a transmission state of forwarding and a text viewpoint baseline for calculating a transmission state and a viewpoint value of a user in the social network data set; and inputting the propagation state and the viewpoint value into a coupling dynamics simulation model, carrying out iterative calculation until the viewpoint value converges so as to obtain a final propagation state and a final viewpoint value of the user, and generating forwarding situation distribution and user viewpoint distribution. Therefore, the problems that errors are generated, public opinion monitoring and early warning and public opinion guide strategy construction are affected and the like due to the fact that a viewpoint evolution model adopts a single-layer propagation structure and coupling modeling is not carried out on propagation and an internal correlation structure in the related technology are solved.

The invention relates to the technical field of information processing, in particular to an information cascade sequence prediction method and device, equipment and a medium, and the method comprises the steps: constructing multi-level data association through extracting a propagation track of target information, a historical behavior track of a user and a diffusion track of associated information; then, the steps of updating the information influence and updating the user influence are executed through iteration, so that the information spreading value and the user spreading capability are mutually enhanced in a loop: in each round of iteration, the information spreading potential is re-evaluated based on the updated user influence, and meanwhile, the user spreading efficiency is re-evaluated based on the updated information influence; and a continuously optimized feedback loop is formed. The problem of information loss of traditional one-way modeling is effectively solved, dynamic interaction of propagation content features, propagator attribute features and network structure features can be captured at the same time, and the information propagation prediction effect is improved.

The invention discloses an intelligent tumor image segmentation method and system based on deep learning, and belongs to the field of tumor image segmentation, and the method comprises the steps: carrying out the feature extraction of a tumor image through a preset U-Net model, obtaining a multilayer convolution feature map, and determining a preliminary candidate point set of a boundary region; if the initial connection weight is higher than a preset threshold value, node features are updated through an information propagation mechanism, and positioning coordinates of the key control points are judged; after positioning coordinates are obtained, a topological path is calculated according to a distance matrix between the key control points, and a shortest connection sequence is determined; fusing space consistency constraint through the shortest connection sequence, and adopting iterative optimization to adjust path nodes to obtain an optimized topological connection relationship; according to the optimized topological connection relation, closed filling is carried out on the segmented contour, the consistency of the inner area of the contour is judged, and a complete contour boundary is obtained; and verifying the segmentation precision of the blood vessel interweaving region through pixel analysis, and determining a final tumor segmentation result.

The application provides a visual token pruning method based on graph information propagation, comprising the following steps: visual extraction is performed on an input image to obtain visual tokens; importance scores of the visual tokens are initialized; a graph structure about the visual tokens is constructed, each visual token is taken as a node, an adjacency matrix is calculated to construct connections between the visual tokens, and the graph structure is initialized; the adjacency matrix is updated through a preset similarity threshold to obtain visual token subgraph structures of different regions; each row of the adjacency matrix is normalized, node information is iteratively propagated, and final scores of each visual token are calculated; k visual tokens with the highest scores are selected according to the final scores of the visual tokens and are projected; the visual tokens obtained through the projection are spliced with text tokens, input into a large language model, and output results are obtained. The method can improve the calculation efficiency of the model, significantly reduces the calculation cost while maintaining the performance of the visual task.