An agent scene double-dimensional visual analysis method, system, device and medium

Abstract

The application belongs to the technical field of agent scene analysis, and discloses an agent scene double-dimension visual analysis method, system, device and medium, which comprises the following steps: reading scene basic data, business dimension data, technical dimension data and scene hierarchical relationship data in an agent scene library to form standardized scene records; constructing a scene-business correlation matrix and a scene-technology correlation matrix, and splicing to obtain a joint correlation matrix; performing non-negative matrix tri-decomposition on the joint correlation matrix to generate a scene joint embedding vector; calculating the shortest path distance of a scene node according to the scene hierarchical relationship data, and combining the scene joint embedding vector to generate a hyperbolic embedding coordinate; and outputting an agent scene double-dimension visual analysis result according to the hyperbolic embedding coordinate. The application can express business dimensions, technical dimensions and scene hierarchical relationships, reduce the fragmentation degree of visual analysis results, and improve the consistency of agent scene analysis.

Description

Technical Field

[0001] This application relates to the field of intelligent agent scene analysis technology, and more specifically, to a method, system, device, and medium for two-dimensional visualization analysis of intelligent agent scenes. Background Technology

[0002] As intelligent agent technology is increasingly used in scenarios such as business processing, knowledge services, process scheduling, and decision support, the same intelligent agent scenario usually involves business attributes, technical capability attributes, and scenario hierarchy relationships simultaneously. The source and expression form of scenario data are also gradually expanding from a single list-style record to a multi-dimensional associated record.

[0003] Existing intelligent agent scenario analysis methods typically focus on business category statistics, technical capability statistics, and scenario list display. Business and technical dimensions are often processed separately, and the hierarchical relationships between scenarios are often displayed in an independent directory format, resulting in a lack of unified data connection between business categories, technical capability categories, and scenario levels.

[0004] Under the above processing method, when it is necessary to analyze which technical capabilities a certain business category covers, which business scenarios a certain technical capability is associated with, and the relationship distribution between different levels of scenarios, existing technologies often need to switch between multiple statistical results and multiple display pages, making it difficult to form a joint analysis result that can be continuously transmitted in the same processing chain. As a result, the two-dimensional relationship and hierarchical distribution relationship of the intelligent agent scenario are easily fragmented in the visualization expression.

[0005] Therefore, a visualization analysis method is needed that can organize business dimensions, technical dimensions, and scenario hierarchical relationships into continuously processable data relationships, and simultaneously express the business and technical connections and hierarchical distribution relationships between scenarios in the visualization results.

[0006] In view of this, this application proposes a two-dimensional visualization analysis method, system, device and medium for intelligent agent scenes to solve the above problems. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this application aims to provide a method, system, device, and medium for two-dimensional visualization analysis of intelligent agent scenarios. By uniformly organizing the basic scenario data, business dimension data, technical dimension data, and scenario hierarchical relationship data in the intelligent agent scenario library, a scenario-business association matrix, a scenario-technology association matrix, and a joint association matrix are constructed. Combined with non-negative matrix trifactorization, scenario joint embedding, and hyperbolic embedding coordinate generation processing, the output includes Venn diagram views, hierarchical tree views, and three-dimensional block views of intelligent agent scenarios, achieving the goal of expressing business dimensions, technical dimensions, and scenario hierarchical relationships within the same processing chain. This reduces the fragmentation of two-dimensional association analysis results and improves the consistency of visualization analysis of complex intelligent agent scenarios.

[0008] To achieve the above objectives, this application adopts the following technical solution:

[0009] Firstly, this application provides a two-dimensional visualization analysis method for intelligent agent scenes, including:

[0010] Read the basic scenario data, business dimension data, technical dimension data and scenario hierarchy relationship data from the intelligent agent scenario library, and form standardized scenario records based on the basic scenario data;

[0011] Business categories are generated based on business dimension data, and technical capability categories are generated based on technical dimension data. Business categories and technical capability categories are used as column objects, and scenario-business association matrices and scenario-technology association matrices are constructed based on standardized scenario records.

[0012] By merging the scenario-business association matrix and the scenario-technology association matrix, a joint association matrix is ​​obtained. The joint association matrix is ​​then subjected to non-negative matrix tri-decomposition to obtain the scenario implicit feature matrix, the association strength matrix, and the dimension implicit feature matrix.

[0013] A joint embedding vector for the scene is generated based on the scene latent feature matrix, the association strength matrix, and the dimension latent feature matrix;

[0014] A scene tree is built based on the scene hierarchy relationship data. The shortest path distance between scene nodes is calculated. Hyperbolic embedding coordinates are generated based on the scene joint embedding vector and the shortest path distance. Based on the hyperbolic embedding coordinates, the results of the intelligent agent scene two-dimensional visualization analysis, including Venn diagram view, hierarchical tree view and three-dimensional block view, are output.

[0015] Furthermore, methods for constructing scenario-business correlation matrices and scenario-technology correlation matrices include:

[0016] The matrix row objects are determined according to the scene identifiers in the standardized scene records, the column objects of the scene-business association matrix are determined by the business category, and the column objects of the scene-technology association matrix are determined by the technical capability category. When the business dimension data record corresponds to an intelligent agent scene and a corresponding business category, the corresponding element in the scene-business association matrix is ​​assigned a value of 1, and when no corresponding intelligent agent scene and corresponding business category are recorded, the value is assigned a value of 0. When the technical dimension data record corresponds to an intelligent agent scene and a corresponding technical capability category, the corresponding element in the scene-technology association matrix is ​​assigned a value of 1, and when no corresponding intelligent agent scene and corresponding technical capability category are recorded, the value is assigned a value of 0.

[0017] Furthermore, methods for performing nonnegative matrix ternary decomposition on the joint incidence matrix include:

[0018] Based on the pre-set number of scene latent categories and dimension latent categories, initialize the scene latent feature matrix, association strength matrix, and dimension latent feature matrix; update the scene latent feature matrix, association strength matrix, and dimension latent feature matrix sequentially using an alternating multiplication update method, and calculate the reconstruction error between the joint association matrix and the product of the scene latent feature matrix, association strength matrix, and dimension latent feature matrix; stop iterating when the absolute value of the difference between the reconstruction error of the current iteration and the reconstruction error of the previous iteration is less than or equal to the preset reconstruction error change threshold, and also stop iterating when the number of iterations reaches the preset maximum number of iterations.

[0019] Furthermore, the method for generating the scene joint embedding vector includes:

[0020] Read the scene latent feature row vector corresponding to each agent scene in the scene latent feature matrix; multiply the scene latent feature row vector with the association strength matrix to obtain the dimension latent association vector of the corresponding agent scene; multiply the dimension latent association vector with the transpose of the dimension latent feature matrix to obtain the scene joint embedding vector, where the first part of the scene joint embedding vector corresponds to the business category and the last part corresponds to the technical capability category.

[0021] Furthermore, methods for calculating the shortest path distance include:

[0022] Read the connection path between any two intelligent agent scene nodes in the scene tree;

[0023] Count the number of edges traversed by the connecting path, and determine the connecting path with the smallest number of edges as the shortest connecting path;

[0024] The number of edges corresponding to the shortest connection path is determined as the shortest path distance between the corresponding two intelligent agent scene nodes.

[0025] Furthermore, the method for generating hyperbolic embedding coordinates includes:

[0026] The scene joint embedding vector is projected onto a two-dimensional space, and the projection result is scaled to the inside of a Poincaré sphere to obtain the initial hyperbolic coordinates corresponding to the scene nodes of each agent. The root scene node in the scene tree is determined, and the hyperbolic embedding coordinates of the root scene node are fixed as the origin of the Poincaré sphere. Hierarchical distance constraints are constructed based on the shortest path distance, and association preservation constraints are constructed based on the initial hyperbolic coordinates. Based on the hierarchical distance constraints and association preservation constraints, the initial hyperbolic coordinates of all nodes except the root scene node are iteratively adjusted using the Riemann stochastic gradient descent method. The iteration stops when the absolute value of the difference between the hyperbolic embedding error of the current iteration and the hyperbolic embedding error of the previous iteration is less than or equal to a preset embedding error change threshold. The iteration also stops when the number of iterations reaches the preset number of embedding iterations, and the coordinates at the time of stopping the iteration are determined as the hyperbolic embedding coordinates.

[0027] Furthermore, the method for generating the Venn diagram view includes:

[0028] Read the user-selected business category and technical capability category, and form a business-related scenario set and a technical-related scenario set respectively based on the association values ​​of the corresponding business category column and technical capability category column in the joint association matrix; determine the Venn diagram overlapping area, business category area and technical capability category area based on the intersection and difference between the business-related scenario set and the technical-related scenario set; determine the display position of each intelligent agent scenario node in the corresponding Venn diagram area based on the hyperbolic embedding coordinates of each intelligent agent scenario node, and generate a Venn diagram view.

[0029] Furthermore, the method for generating the three-dimensional block view includes:

[0030] The planar position of the 3D block corresponding to each agent scene is determined based on hyperbolic embedding coordinates;

[0031] The number of technical capability categories with a correlation value greater than zero corresponding to each intelligent agent scene is counted based on the scene-technology association matrix, and the height of the three-dimensional block is determined based on the number of technical capability categories.

[0032] Generate a 3D block view based on planar position and 3D block height.

[0033] Secondly, this application provides a two-dimensional visualization analysis system for intelligent agent scenes, including:

[0034] The scene organization module is used to read the basic scene data, business dimension data, technical dimension data and scene hierarchy relationship data in the intelligent agent scene library, and form standardized scene records based on the basic scene data.

[0035] The matrix construction module is used to generate business categories based on business dimension data and technical capability categories based on technical dimension data. It uses business categories and technical capability categories as column objects and constructs scenario-business association matrices and scenario-technology association matrices based on standardized scenario records.

[0036] The matrix decomposition module is used to concatenate the scenario-business association matrix and the scenario-technology association matrix to obtain a joint association matrix, and to perform non-negative matrix ternary decomposition on the joint association matrix to obtain the scenario implicit feature matrix, the association strength matrix, and the dimension implicit feature matrix.

[0037] The joint embedding module is used to generate scene joint embedding vectors based on the scene latent feature matrix, the association strength matrix, and the dimension latent feature matrix;

[0038] The hyperbolic output module is used to build a scene tree based on scene hierarchy relationship data, calculate the shortest path distance between scene nodes, generate hyperbolic embedding coordinates based on the scene joint embedding vector and the shortest path distance, and output the two-dimensional visualization analysis results of the agent scene, including Venn diagram view, hierarchical tree view and three-dimensional block view, based on the hyperbolic embedding coordinates.

[0039] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executed by the processor, wherein the processor executes the computer program to implement the steps of the above-described intelligent agent scene two-dimensional visualization analysis method.

[0040] Fourthly, this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described intelligent agent scene two-dimensional visualization analysis method.

[0041] The technical effects and advantages of the intelligent agent scene two-dimensional visualization analysis method, system, device and medium proposed in this application are as follows:

[0042] Compared with existing technologies, this application reads basic scene data, business dimension data, technical dimension data and scene hierarchical relationship data from the intelligent agent scene library, and forms standardized scene records based on the basic scene data. This organizes the originally scattered intelligent agent scene information into a unified input that can continue to build an association matrix, thereby reducing the subsequent analysis breaks caused by inconsistent scene identification and scattered data sources.

[0043] This application generates business categories based on business dimension data and technical capability categories based on technical dimension data. It then constructs scenario-business association matrices and scenario-technology association matrices using business categories and technical capability categories as column objects. This transforms the relationships between intelligent agent scenarios and business categories and technical capability categories into structured matrix data, thereby providing a clear data foundation for joint analysis of business and technical dimensions.

[0044] This application obtains a joint correlation matrix by splicing the scenario-business correlation matrix and the scenario-technology correlation matrix, and performs a non-negative matrix ternary decomposition on the joint correlation matrix to obtain the scenario implicit feature matrix, the correlation strength matrix, and the dimension implicit feature matrix. This allows the correlation information that exists scattered in the business dimension and the technology dimension to be expressed in the same matrix decomposition process, thereby improving the organization and analyzability of the two-dimensional correlation relationship of the intelligent agent scenario.

[0045] This application generates a joint embedding vector of scenarios based on the scenario implicit feature matrix, the association strength matrix, and the dimension implicit feature matrix. It also builds a scenario tree by combining scenario hierarchical relationship data, calculates the shortest path distance between scenario nodes, and generates hyperbolic embedding coordinates. This allows the business and technical association results and the scenario hierarchical distance to be expressed in the same coordinate system, thereby reducing the result fragmentation caused by separately displaying business statistics, technical statistics, and hierarchical directories.

[0046] This application outputs two-dimensional visualization analysis results of intelligent agent scenes based on hyperbolic embedded coordinates, including Venn diagram views, hierarchical tree views, and three-dimensional block views. This enables the overlapping business and technology relationships, hierarchical relationships, and technology coverage of the same intelligent agent scene to be displayed in association with different views, thereby improving the readability and consistency of two-dimensional relationship analysis results in complex intelligent agent scene libraries. Attached Figure Description

[0047] Figure 1 This is a flowchart of a two-dimensional visualization analysis method for intelligent agent scenes according to Embodiment 1 of this application;

[0048] Figure 2 This is a data link diagram for the two-dimensional visualization analysis of the intelligent agent scene in Embodiment 1 of this application;

[0049] Figure 3 This is a flowchart of hyperbolic embedded coordinate generation and multi-view output in Embodiment 1 of this application. Detailed Implementation

[0050] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] Example 1:

[0052] See Figures 1-3 As shown, this embodiment provides a two-dimensional visualization analysis method for intelligent agent scenarios, including:

[0053] Read all basic scenario data from the intelligent agent scenario library, organize each intelligent agent scenario according to the scenario identifier, and write the scenario identifier, scenario name, scenario description, and scenario level identifier into the same standardized scenario record.

[0054] The intelligent agent scenario library stores multiple intelligent agent scenarios, each with a unique scenario identifier. The basic scenario data includes the scenario identifier, scenario name, scenario description, and scenario level identifier. The business dimension data includes the scenario identifier, business specialty identifier, business cluster identifier, and business category identifier. The technical dimension data includes the scenario identifier, technical capability category identifier, and technical capability tag. Based on the scenario identifiers in the business dimension data and technical dimension data, the business dimension data and technical dimension data are grouped into standardized scenario records with the same scenario identifier.

[0055] Scene hierarchy relationship data includes the inclusion relationship between the original scene and the micro scene, as well as the inclusion relationship between the micro scene and the meta scene; in another possible implementation, scene hierarchy relationship data also includes the reference reuse relationship between different agent scenes.

[0056] The total number of agent scenarios is determined based on standardized scenario records, and the agent scenarios are denoted as follows: ,in, Indicates the first Individual agent scenarios This indicates the scene number of the intelligent agent. This represents the total number of intelligent agent scenarios.

[0057] Read the business dimension data corresponding to each standardized scenario record. Combine the business specialty identifier, business cluster identifier, and business category identifier in the same business dimension data into a business category identifier in the order of business specialty identifier, business cluster identifier, and business category identifier. Remove duplicate business category identifiers and sort them according to the business category identifier. Assign column numbers to each deduplicated business category in sequence. Each business category is denoted as: ,in, Indicates the first Each business category The number representing the business category. This indicates the total number of business categories.

[0058] Read the technical dimension data corresponding to each standardized scenario record, group technical dimension data with the same technical capability category identifier into the same technical capability category, deduplicate each technical capability category identifier, sort them according to the technical capability category identifier, and assign column numbers to each deduplicated technical capability category in sequence. Each technical capability category is denoted as follows: ,in, Indicates the first Each technical capability category The number representing the category of technical capability. This indicates the total number of technical capability categories.

[0059] Construct a scenario-business association matrix based on the intelligent agent scenario and business category. The scenario-business association matrix is ​​represented as follows:

[0060] ;

[0061] in, Represents the scenario-business relationship matrix. Indicates the first The first intelligent agent scenario and the first Business association values ​​between business categories This represents the total number of intelligent agent scenarios. This indicates the total number of business categories.

[0062] The method for determining the scenario-business related value is as follows: when the first... The intelligent agent scenario belongs to the first... When considering individual business categories, Set to one; when the first The intelligent agent scenario was not categorized into the first... When considering individual business categories, Set to zero.

[0063] A scenario-technology correlation matrix is ​​constructed based on the intelligent agent scenario and technology capability categories. The scenario-technology correlation matrix is ​​represented as follows:

[0064] ;

[0065] in, Representing the scenario-technology correlation matrix, Indicates the first The first intelligent agent scenario and the first Technical correlation values ​​between each technical capability category This represents the total number of intelligent agent scenarios. This indicates the total number of technical capability categories.

[0066] The method for determining the scenario-technology correlation value is as follows: when the first... The first intelligent agent scenario requires the first When categorizing technical capabilities, Set to one; when the first The first intelligent agent scenario does not require the first When categorizing technical capabilities, Set to zero.

[0067] In another possible implementation, when the scenario library already stores business association weights and technical capability usage intensity, the elements in the scenario-business association matrix and the scenario-technology association matrix are represented by corresponding non-negative weight values, where the non-negative weight values ​​are existing field values ​​in the scenario library that have been normalized to between zero and one.

[0068] In another possible implementation, before performing nonnegative matrix tri-decomposition on the joint association matrix, the columns in the scenario-business association matrix and the scenario-technology association matrix are normalized. The normalization process uses column vector L2 normalization to bring the business category column and the technology capability category column to a uniform numerical scale.

[0069] The scenario-business association matrix and the scenario-technology association matrix are concatenated along their columns to obtain the joint association matrix. The formula for calculating the joint association matrix is ​​as follows:

[0070] ;

[0071] in, Represents the joint incidence matrix. Represents the scenario-business relationship matrix. Representing the scenario-technology correlation matrix, This indicates that the scenario-business association matrix and the scenario-technology association matrix are concatenated according to the column direction.

[0072] The joint incidence matrix satisfies:

[0073] ;

[0074] in, Represents the joint incidence matrix. Denotes the set of nonnegative real numbers. This represents the total number of intelligent agent scenarios. Indicates the total number of business categories. This indicates the total number of technical capability categories.

[0075] Each row in the joint association matrix corresponds to an intelligent agent scenario, and each column corresponds to a business category or a technical capability category. The column represents information related to business dimensions, followed by... The columns represent information related to the technical dimensions, thus allowing the business affiliation and technical capability requirements of the same intelligent agent scenario to be expressed in the same matrix structure.

[0076] The joint association matrix is ​​decomposed into a nonnegative matrix tri-factor to obtain the scene implicit feature matrix, the association strength matrix, and the dimension implicit feature matrix, which transforms the scene row objects, business dimension column objects, and technical dimension column objects in the joint association matrix into low-dimensional implicit structure expressions.

[0077] The formula for calculating the trifactor of a nonnegative matrix is:

[0078] ;

[0079] in, Represents the joint incidence matrix. Represents the latent feature matrix of the scene. Represents the correlation strength matrix. The dimension of the latent feature matrix is ​​represented. This represents the transpose of the dimension-hidden feature matrix.

[0080] The scene latent feature matrix, the association strength matrix, and the dimension latent feature matrix satisfy the following:

[0081] ;

[0082] in, Indicates the number of implicit categories in the scene. This indicates the number of implicit categories in a dimension.

[0083] Number of implicit categories in a scene and the number of implicit categories in the dimension Based on the pre-set size of the intelligent agent scenario library, the number of implicit categories in a scenario is used to control the granularity of scenario clustering, and the number of implicit categories in a dimension is used to control the granularity of capability stack merging for business categories and technical capability categories.

[0084] It should be noted that in this embodiment, the preset reconstruction error change threshold, preset maximum number of iterations, preset embedding error change threshold, preset number of embedding iterations, preset collaboration threshold, association preservation coefficient, distance decay coefficient, boundary interval parameter, as well as each scaling parameter and spacing parameter are all algorithm parameters configured before performing visualization analysis. The algorithm parameters are determined according to the size of the intelligent agent scene library, display resolution, and interaction response requirements, and remain unchanged during the same visualization analysis process.

[0085] The scene latent feature matrix F, the correlation strength matrix S, and the dimension latent feature matrix G are non-negatively initialized by generating random non-negative values ​​between zero and one, and each column is normalized to ensure that the initial scale of each column is consistent.

[0086] The Frobenius norm squared between the products of the joint correlation matrix and the scene latent feature matrix, the correlation strength matrix, and the dimensional latent feature matrix is ​​used as the optimization objective of the nonnegative matrix trifactoring, and the formula is:

[0087] ;

[0088] in, Represents the joint incidence matrix. Represents the latent feature matrix of the scene. Represents the correlation strength matrix. The dimension of the latent feature matrix is ​​represented. This represents the transpose of the dimension-wise latent feature matrix. This represents the Frobenius norm.

[0089] An alternating multiplication update method is used to iteratively solve the scene latent feature matrix, the correlation strength matrix, and the dimension latent feature matrix. In each iteration, two of the matrices are fixed and the remaining matrix is ​​updated. Then, the three matrices are updated in the same way.

[0090] The update formula for the scene's latent feature matrix is:

[0091] ;

[0092] in, A definite expression represents element-wise multiplication, while a fraction represents element-wise division. Represents the latent feature matrix of the scene. Represents the joint incidence matrix. The dimension of the latent feature matrix is ​​represented. This represents the transpose of the correlation strength matrix. This represents a constant used to prevent the denominator from being zero.

[0093] The update formula for the correlation strength matrix is:

[0094] ;

[0095] in, Represents the correlation strength matrix. This represents the transpose of the scene's latent feature matrix. Represents the joint incidence matrix. The dimension is represented by the latent feature matrix.

[0096] The update formula for the dimension-wise latent feature matrix is:

[0097] ;

[0098] in, The dimension of the latent feature matrix is ​​represented. This represents the transpose of the joint incidence matrix. Represents the latent feature matrix of the scene. This represents the correlation strength matrix.

[0099] Calculate the first The reconstruction error corresponding to the next iteration is calculated using the following formula:

[0100] ;

[0101] in, Indicates the first The reconstruction error corresponding to the next iteration. , and They represent the first The scene latent feature matrix, correlation strength matrix, and dimension latent feature matrix obtained in each iteration.

[0102] The iteration stops when the absolute value of the difference between the reconstruction error of the current iteration and the reconstruction error of the previous iteration is less than or equal to the preset reconstruction error change threshold; the iteration also stops when the number of iterations reaches the preset maximum number of iterations, and the scene hidden feature matrix, correlation strength matrix and dimension hidden feature matrix at the time of stopping the iteration are taken as the result of non-negative matrix tri-decomposition.

[0103] The scene latent category corresponding to each agent's scene is determined based on the scene latent feature matrix, and the calculation formula is as follows:

[0104] ;

[0105] in, Indicates the first The implicit category of a scenario corresponding to a given intelligent agent scenario. Indicates the first The first intelligent agent scenario in the... The implicit association values ​​of each scenario's categories This indicates the operation of retrieving the category corresponding to the maximum associated value.

[0106] The dimensional latent category corresponding to each business category and technical capability category is determined based on the dimensional latent feature matrix. The calculation formula is as follows:

[0107] ;

[0108] in, Indicates the first The implicit category of each dimension object. Indicates the first The dimension object in the first... Each dimension implies associated values ​​on a category, and the dimension object includes business categories and technical capability categories.

[0109] The scene joint embedding vector is generated based on the scene latent feature matrix, the association strength matrix, and the dimensional latent feature matrix. The calculation formula is as follows:

[0110] ;

[0111] in, Indicates the first The scene joint embedding vector corresponding to each agent scene. Represents the first element in the scene's latent feature matrix. OK, Represents the correlation strength matrix. This represents the transpose of the dimension-hidden feature matrix.

[0112] Scene joint embedding vectors include There are 10 components, of which the first 10 is the first 10. Each component corresponds to a different business category. Each component corresponds to a different technical capability category, enabling the implicit correlation information of each intelligent agent scenario in the business and technical dimensions to form a vector expression corresponding to the column objects of the joint correlation matrix.

[0113] Hyperbolic embedding coordinates are generated based on the joint embedding vectors of each agent scenario and the scene hierarchy relationship data, so that the implicit business and technical relationships and scene hierarchy distance relationships between agent scenarios are jointly mapped into the Poincaré sphere hyperbolic space.

[0114] Read the scene hierarchy relationship data, and build a scene tree based on the inclusion relationship between the original scene and the micro scene, and the inclusion relationship between the micro scene and the meta scene. The nodes in the scene tree are the intelligent agent scene nodes, and the edges in the scene tree are the inclusion edges between scenes of different levels.

[0115] The root scene node is determined based on the top-level scene node in the scene tree. When there are multiple top-level scene nodes in the scene tree, a virtual root scene node that does not participate in the visualization output is created, and containment edges are created between the virtual root scene node and each top-level scene node. The virtual root scene node is then determined as the root scene node of the scene tree.

[0116] When building the scene tree, only the containment relationship between the original scene and the micro scene, and the containment relationship between the micro scene and the meta scene are used as edges in the scene tree. The reference reuse relationship is not included in the shortest path distance calculation of scene nodes, and is output as a reuse connection relationship when generating a hierarchical tree view or a 3D block view.

[0117] Calculate the shortest path distance between any two agent scene nodes based on the scene tree. The shortest path distance is expressed as: , indicating the first The intelligent agent scene node and the first The shortest path distance between each intelligent agent scene node in the scene tree.

[0118] The initial embedding vector for each agent scene node is determined based on the joint embedding vector of each agent scene, and the first embedding vector is then set to the first embedding vector. The joint embedding vector of each agent scenario As the first The initial embedding vector of each intelligent agent scene node.

[0119] The initial embedding vectors of each intelligent agent scene node are mapped to the Poincaré sphere hyperbolic space, and the Poincaré sphere is defined by the following formula:

[0120] ;

[0121] in, express Verponcaré ball, This represents the point in the Poincaré sphere. express 3D real space, Point The Euclidean norm.

[0122] In the visualization display scenario, the Poincaré sphere dimension is set to two dimensions, so that each intelligent agent scene node corresponds to a two-dimensional hyperbolic embedding coordinate.

[0123] The joint embedding vectors corresponding to each agent's scenario are arranged into a joint embedding matrix according to the order of the agent's scenario. The joint embedding matrix is ​​centered and the covariance matrix of the centered joint embedding matrix is ​​calculated. The first two eigenvectors are selected in descending order of eigenvalues, and the transposes of the first two eigenvectors are used as the two rows of the linear projection matrix.

[0124] When the dimension of the scene joint embedding vector is equal to two dimensions, the two-dimensional identity matrix is ​​used as the linear projection matrix; when the dimension of the scene joint embedding vector is one dimension, zero components are added after the one-dimensional projection result to form a two-dimensional projection result.

[0125] Scaling the linear projection result to the interior of the Poincaré sphere, the initial hyperbolic coordinates are calculated using the following formula:

[0126] ;

[0127] in, Indicates the first The initial hyperbolic coordinates of each intelligent agent scene node Indicates the first The scene joint embedding vector corresponding to each agent scene. Represents the linear projection matrix. This represents a pre-configured scaling factor that is greater than zero and less than one. This represents a constant used to prevent the denominator from being zero.

[0128] The initial hyperbolic coordinates of the root scene node are set to the origin of the Poincaré sphere, and the hyperbolic embedding coordinates corresponding to the root scene node are kept unchanged during subsequent hyperbolic embedding optimization.

[0129] The formula for calculating the hyperbolic distance between two intelligent agent scene nodes in a Poincaré sphere is:

[0130] ;

[0131] in, Indicates the first The intelligent agent scene node and the first Hyperbolic distance between intelligent agent scene nodes Indicates the first Hyperbolic embedding coordinates of each intelligent agent scene node Indicates the first Hyperbolic embedding coordinates of each agent scene node This represents the inverse hyperbolic cosine function.

[0132] The distance weight of node pairs is determined based on the shortest path distance in the scene tree. The formula for calculating the distance weight is as follows:

[0133] ;

[0134] in, Indicates the first The intelligent agent scene node and the first Distance weights between scene nodes of each intelligent agent Indicates the first The intelligent agent scene node and the first The shortest path distance between each intelligent agent scene node in the scene tree. This represents the distance attenuation coefficient that is pre-configured by the system and is greater than zero.

[0135] Using the difference between the hyperbolic distance and the shortest path distance in the scene tree as the hierarchical distance constraint, and the distance between the adjusted hyperbolic embedding coordinates and the initial hyperbolic coordinates as the association preservation constraint, the formula for the hyperbolic embedding optimization objective is:

[0136] ;

[0137] in, This represents the target value for hyperbolic embedding optimization. and They represent the first The first intelligent agent scene node and the first Hyperbolic embedding coordinates of each agent scene node Represents a two-dimensional Poincaré sphere. Represents the hyperbolic embedding coordinates of the root scene node. Represents the origin of the Poincaré ball. Indicates the first The intelligent agent scene node and the first Distance weights between scene nodes of each intelligent agent This represents the hyperbolic distance between two nodes. Indicates the first The intelligent agent scene node and the first The shortest path distance between each intelligent agent scene node in the scene tree. Indicates the first The initial hyperbolic coordinates of each intelligent agent scene node This represents the association retention coefficient that is greater than zero.

[0138] Riemann stochastic gradient descent is used to solve the hyperbolic embedding optimization objective. In each iteration, node pair samples are selected from the intelligent agent scene node pairs, the Euclidean gradient corresponding to the node pair sample is calculated, and the Euclidean gradient is converted into the Riemann gradient.

[0139] The formula for converting the Euclidean gradient to the Riemann gradient is:

[0140] ;

[0141] in, Indicates the objective function at the th Riemann gradient at each intelligent agent scene node Indicates the objective function at the th The Euclidean gradient at each scene node of the intelligent agent. Indicates the first Hyperbolic embedding coordinates of each agent scene node denoted by the Euclidean norm of hyperbolic embedded coordinates.

[0142] Update the hyperbolic embedding coordinates of all agent scene nodes except the root scene node according to the Riemann gradient. After updating, project the coordinates. If the Euclidean norm of the updated hyperbolic embedding coordinates is greater than or equal to... When the coordinate is scaled to a value equal to the Euclidean norm, the coordinate is scaled to a value equal to the Euc The location.

[0143] The projection processing formula is:

[0144] ;

[0145] in, Represents the hyperbolic embedding coordinates after projection processing. Indicates the hyperbolic embedding coordinates before projection processing. This represents a boundary interval parameter that is pre-configured by the system and is greater than zero and less than one.

[0146] After each hyperbolic embedding iteration, the th Substitute the hyperbolic embedding coordinates corresponding to each agent scene node obtained in the iteration into the hyperbolic embedding optimization objective, and determine the obtained hyperbolic embedding optimization objective value as the th iteration. The hyperbolic embedding error corresponding to the next iteration is calculated using the following formula:

[0147] ;

[0148] in, Indicates the first The hyperbolic embedding error corresponding to the next iteration. This represents the aforementioned hyperbolic embedding optimization objective. Indicates the first During the nth iteration Hyperbolic embedding coordinates corresponding to each intelligent agent scene node.

[0149] When the absolute value of the difference between the hyperbolic embedding error of the current iteration and the hyperbolic embedding error of the previous iteration is less than or equal to the preset embedding error change threshold, the hyperbolic embedding optimization is stopped; when the number of iterations reaches the preset number of embedding iterations, the hyperbolic embedding optimization is also stopped, and the coordinates corresponding to the scene nodes of each agent at the time of stopping the iteration are determined as the hyperbolic embedding coordinates.

[0150] Since the hyperbolic embedding coordinates of the root scene node are fixed at the origin of the Poincaré sphere, and the hyperbolic embedding optimization objective includes the hierarchical distance constraint corresponding to the shortest path distance of the scene tree, the hyperbolic distance of each agent scene node relative to the origin of the Poincaré sphere is used to characterize the hierarchical position of the corresponding agent scene node in the scene hierarchy structure; the association preservation constraint restricts the degree of offset of the hyperbolic embedding coordinates after adjustment relative to the initial hyperbolic coordinates, so that the differences in business and technical associations between agent scene nodes continue to be preserved in the hyperbolic embedding coordinates.

[0151] Multi-view visualization results are generated based on hyperbolic embedded coordinates. These multi-view visualization results include Venn diagram views, hierarchical tree views, and three-dimensional block views, enabling the same group of intelligent agent scene nodes to express business dimensions, technical dimensions, and scene hierarchy relationships in different display modes.

[0152] Read the display mode and focus center selected by the user in the interactive interface. When the display mode is Venn diagram view, perform Möbius transformation on the hyperbolic embedding coordinates of each intelligent agent scene node according to the focus center, so that the business category, technical capability category or target scene node that the user is interested in moves to the center of the display area.

[0153] The formula for calculating the Möbius transform is:

[0154] ;

[0155] in, Indicates the first Complex representation of each intelligent agent scene node after Möbius transformation Indicates the first Complex number representations of hyperbolic embedded coordinates of each intelligent agent scene node. This indicates the focus point selected by the user, and For the point inside the Poincaré sphere, The conjugate complex number representing the center of focus.

[0156] The two-dimensional display coordinates in the Venn diagram view are determined based on the transformed complex number representation. The formula for calculating the two-dimensional display coordinates is as follows:

[0157] ;

[0158] in, Indicates the first The two-dimensional display coordinates of each intelligent agent scene node in the Venn diagram view. Representing complex numbers The real part, Representing complex numbers The imaginary part.

[0159] Read the user's selected number The business category, based on the scenario-business association matrix. The business association values ​​in the columns corresponding to each business category form a set of business association scenarios, calculated using the following formula:

[0160] ;

[0161] in, This represents the set of business-related scenarios corresponding to the j-th business category. Indicates the first Each intelligent agent scene node. Indicates the first The first intelligent agent scenario and the first Business association values ​​between business categories.

[0162] Read the user's selected number The first technical capability category, based on the scenario-technology correlation matrix. The technology association values ​​corresponding to each technology capability category form a set of technology association scenarios, calculated using the following formula:

[0163] ;

[0164] in, Indicates the first A set of technology-related scenarios corresponding to each technology capability category. Indicates the first The first intelligent agent scenario and the first Technical correlation values ​​between each technical capability category.

[0165] Calculate the intersection of the business-related scenario set and the technology-related scenario set, and display the intelligent agent scenario nodes corresponding to the intersection in the overlapping area of ​​the Venn diagram; calculate the difference between the business-related scenario set and the technology-related scenario set, and display the intelligent agent scenario nodes corresponding to the difference in the business category area; calculate the difference between the technology-related scenario set and the business-related scenario set, and display the intelligent agent scenario nodes corresponding to the difference in the technology capability category area.

[0166] Based on the business-related scenario set, technology-related scenario set, and their intersection, the corresponding Venn diagram region is determined for each intelligent agent scenario node. The two-dimensional display coordinates obtained by the Möbius transformation of each intelligent agent scenario node are scaled and translated proportionally according to the display range of the corresponding Venn diagram region, so that each intelligent agent scenario node falls into the corresponding Venn diagram region, while maintaining the relative positional relationship between intelligent agent scenario nodes within the same Venn diagram region.

[0167] When the display mode is a hierarchical tree view, the node hierarchy depth is determined based on the hyperbolic radial distance of each agent scene node relative to the origin of the Poincaré sphere, the arrangement position within the same level is determined based on the angular coordinates of each agent scene node, and the original scene node, micro scene node, and meta scene node are connected based on the scene hierarchy relationship data.

[0168] Calculate the first The Euclidean radial modulus of each intelligent agent scene node corresponding to hyperbolic embedded coordinates is calculated using the following formula:

[0169] ;

[0170] in, Indicates the first The Euclidean radial modulus of hyperbolic embedded coordinates corresponding to each intelligent agent scene node. Indicates the first Hyperbolic embedding coordinates of each intelligent agent scene node.

[0171] Calculate the first based on the Euclidean radial modulus. The hyperbolic radial distance of each intelligent agent scene node relative to the origin of the Poincaré sphere is calculated using the following formula:

[0172] ;

[0173] in, Indicates the first The hyperbolic radial distance of each intelligent agent scene node relative to the origin of the Poincaré sphere This represents the inverse hyperbolic cosine function.

[0174] The first is determined based on the hyperbolic radial distance. The hierarchical depth of a scene node of an intelligent agent in the hierarchical tree view is calculated using the following formula:

[0175] ;

[0176] in, Indicates the first The hierarchical depth of each intelligent agent scene node in the hierarchical tree view. This indicates a hierarchical spacing greater than zero. This represents the floor function.

[0177] According to the The angular coordinates of a scene node of an intelligent agent are calculated using the two hyperbolic embedded coordinate components, and the calculation formula is as follows:

[0178] ;

[0179] in, Indicates the first Angular coordinates of each intelligent agent scene node and They represent the first Two hyperbolic embedded coordinate components of a scene node for an intelligent agent. This represents a function that calculates the angle based on two coordinate components.

[0180] Calculate the first based on the hierarchy depth The display radius corresponding to each intelligent agent scene node is calculated using the following formula:

[0181] ;

[0182] in, The hierarchical depth is The display radius corresponding to the scene node of the intelligent agent. Indicates the initial display radius that is greater than or equal to zero. This indicates the display spacing between adjacent levels that is greater than zero.

[0183] The formula for calculating the two-dimensional display coordinates in the hierarchical tree view is:

[0184] ;

[0185] in, Indicates the first The two-dimensional display coordinates of each intelligent agent scene node in the hierarchical tree view. The hierarchical depth is The display radius corresponding to the intelligent agent scene node.

[0186] Read the parent-child node relationship based on the scene hierarchy data. When there is an inclusion relationship between the original scene and the micro scene or between the micro scene and the original scene between two intelligent agent scene nodes, draw the hierarchical connection line between the two intelligent agent scene nodes in the hierarchy tree view.

[0187] In another possible implementation, when the scene hierarchy relationship data contains reference reuse relationships, the reference reuse relationships are displayed as visual relationships, and the corresponding reuse connection relationships are displayed in the hierarchy tree view or the 3D block view.

[0188] When the display mode is a 3D block view, the 3D radial display distance is determined based on the Euclidean radial modulus of the hyperbolic embedded coordinates corresponding to each intelligent agent scene node, the 3D circumferential position is determined based on the angular coordinates, the 3D block height is determined based on the number of technical capability categories associated with each intelligent agent scene, and the 3D block display identifier is determined based on the business category or the scene implicit category.

[0189] Calculate the first The radial display distance of each intelligent agent scene node in the 3D block view is calculated using the following formula:

[0190] ;

[0191] in, Indicates the first The radial display distance of each intelligent agent scene node in a 3D block view. Indicates the first The Euclidean radial modulus of hyperbolic embedded coordinates corresponding to each intelligent agent scene node.

[0192] According to the scenario-technology correlation matrix, the first... The number of technical capability categories associated with a given intelligent agent scenario is calculated using the following formula:

[0193] ;

[0194] in, Indicates the first The number of technical capability categories associated with each intelligent agent scenario Indicates the first The first intelligent agent scenario and the first Technical correlation values ​​between each technical capability category This represents an indicator function, which takes the value of 1 when the condition within the parentheses is true, and takes the value of zero when the condition within the parentheses is false. This indicates the total number of technical capability categories.

[0195] Determine the number of technical capability categories. The formula for calculating the height of the 3D block corresponding to a scene of an intelligent agent is:

[0196] ;

[0197] in, Indicates the first The height of the 3D block corresponding to each intelligent agent scene. This represents a height scaling factor greater than zero.

[0198] The formula for calculating the display coordinates in a 3D block view is:

[0199] ;

[0200] in, Indicates the first The display coordinates of each intelligent agent scene node in a 3D block view. Indicates the first The angular coordinates of each intelligent agent scene node.

[0201] In the 3D block view, each intelligent agent scene node is displayed as a 3D block. The planar position of the 3D block is determined by hyperbolic embedded coordinates, the height of the 3D block is determined by the number of technical capability categories associated with the corresponding intelligent agent scene, and the display identifier of the 3D block is determined by the business category or the scene implicit category.

[0202] In another possible implementation, when different technical capability categories have different technical weights, the technical complexity is calculated based on the technical correlation values ​​in the scenario-technology correlation matrix and the technical weights corresponding to each technical capability category. The calculation formula is as follows:

[0203] ;

[0204] in, Indicates the first The technical complexity of each intelligent agent scenario Indicates the first The non-negative technical weights corresponding to each technical capability category.

[0205] The height of a 3D block is determined using technical complexity, and the calculation formula is as follows:

[0206] ;

[0207] in, Indicates the use of technical weights when the first The height of the three-dimensional block corresponding to each intelligent agent scene.

[0208] When a user switches between Venn diagram view, hierarchical tree view, and 3D block view, the hyperbolic embedding coordinates of the same agent scene nodes remain unchanged, and the visualization coordinates are recalculated only according to the target display mode, so that the relative relationships of nodes in different views remain consistent.

[0209] When a user performs drag, zoom, filter, or node selection operations, the focus center, display range, business category filter conditions, and technical capability category filter conditions are updated according to the user's operation parameters, and the visualization coordinates in the current display mode are regenerated.

[0210] When a user selects any intelligent agent scene node, the system reads the scene identifier, scene name, business category, technical capability category, scene implicit category, dimension implicit category, number of technical capability categories and level depth corresponding to that intelligent agent scene node, and outputs the corresponding node details information in the interactive display interface; in the implementation method that adopts technical weights, the node details information also includes technical complexity.

[0211] Based on the association strength matrix, business technology collaboration patterns are extracted. When the association strength between any scene latent category and any dimension latent category in the association strength matrix S is greater than or equal to a preset collaboration threshold, the scene latent category, the corresponding dimension latent category, and the covered intelligent agent scene nodes are determined as a set of business technology collaboration patterns. The covered intelligent agent scene nodes are those that satisfy... The preset collaboration threshold is a collaborative mode extraction parameter pre-configured by the system, and the preset collaboration threshold is greater than or equal to zero. It is equal to the implicit category of the scene and there is at least one dimension object with an association value greater than zero in the joint association matrix A that belongs to the agent scene node of the implicit category of the dimension.

[0212] The output of the business-technology collaboration mode includes scenario implicit category identifiers, dimension implicit category identifiers, association strength, number of covered scenarios, main business categories, and main technical capability categories. Among them, the number of covered scenarios is the number of intelligent agent scenario nodes covered by the business-technology collaboration mode. The main business category is the business category with the highest frequency of business association value of 1 among the covered intelligent agent scenario nodes. The main technical capability category is the technical capability category with the highest frequency of technical association value of 1 among the covered intelligent agent scenario nodes. When multiple business categories have the same frequency, the main business category is determined by sorting according to the business category identifier. When multiple technical capability categories have the same frequency, the main technical capability category is determined by sorting according to the technical capability category identifier.

[0213] Generate a two-dimensional visualization analysis result of the intelligent agent scene, which includes scene node display result, business technology association result, scene hierarchy display result, and node detail display result.

[0214] The scene node display results include the display coordinates, display identifier, business category, technical capability category, scene implicit category, and layer depth of each intelligent agent scene node in the current display mode. The business and technology association results include the association strength between the scene implicit category and the dimension implicit category, the number of covered scenes, the main business category, and the main technical capability category.

[0215] The scene hierarchy display results are generated based on the scene hierarchy relationship data. The scene hierarchy display results include the parent-child relationship between the original scene node, micro scene node and meta scene node, the hierarchy depth and the hierarchy connection line identifier.

[0216] When the user selects the first option on the interactive display interface When selecting a business category, filter to meet the following criteria. The intelligent agent scene nodes are selected and highlighted in the current display mode, covering a number of scenes that meet the requirements. The number of intelligent agent scene nodes, and the main technical capability category is the technical capability category with the largest number of times the technical association value is greater than zero in the intelligent agent scene nodes. When there are multiple technical capability categories with the same number of times, the main technical capability category is determined by sorting according to the technical capability category identifier.

[0217] When the user selects the first option on the interactive display interface When selecting a technical capability category, filter those that meet the criteria. The intelligent agent scene nodes are selected and highlighted in the current display mode, covering a number of scenes that meet the requirements. The number of intelligent agent scene nodes is determined by the number of times the business association value is greater than zero among the intelligent agent scene nodes. When multiple business categories have the same number of times, the main business category is determined by sorting according to the business category identifier.

[0218] When a user selects any business technology collaboration mode on the interactive display interface, the system filters the intelligent agent scene nodes covered by that business technology collaboration mode and displays the scene implicit category, dimension implicit category, correlation strength, main business category, main technical capability category, and covered scene list corresponding to that business technology collaboration mode.

[0219] Example 2:

[0220] This embodiment provides a two-dimensional visualization analysis system for intelligent agent scenes, including: a scene organization module, a matrix construction module, a matrix decomposition module, a joint embedding module, and a hyperbolic output module.

[0221] The scenario organization module is used to read the basic scenario data, business dimension data, technical dimension data and scenario hierarchy relationship data from the intelligent agent scenario library, and form standardized scenario records based on the basic scenario data.

[0222] The matrix construction module is used to generate business categories based on business dimension data and technical capability categories based on technical dimension data. It uses business categories and technical capability categories as column objects and constructs scenario-business association matrices and scenario-technology association matrices based on standardized scenario records.

[0223] The matrix decomposition module is used to concatenate the scenario-business association matrix and the scenario-technology association matrix to obtain a joint association matrix. The joint association matrix is ​​then decomposed into a non-negative matrix tri-decomposition to obtain the scenario implicit feature matrix, the association strength matrix, and the dimension implicit feature matrix.

[0224] The joint embedding module is used to generate scene joint embedding vectors based on the scene latent feature matrix, the association strength matrix, and the dimension latent feature matrix.

[0225] The hyperbolic output module is used to build a scene tree based on scene hierarchy relationship data, calculate the shortest path distance between scene nodes, generate hyperbolic embedding coordinates based on the scene joint embedding vector and the shortest path distance, and output the two-dimensional visualization analysis results of the agent scene, including Venn diagram view, hierarchical tree view and three-dimensional block view, based on the hyperbolic embedding coordinates.

[0226] Example 3:

[0227] A computer device includes a processor and a memory, wherein the memory stores a computer program for the processor to execute, and the processor executes the computer program to implement the above-mentioned intelligent agent scene two-dimensional visualization analysis method.

[0228] When the processor executes the computer program, it reads the basic scene data, business dimension data, technical dimension data, and scene hierarchical relationship data from the intelligent agent scene library, constructs a scene-business association matrix, a scene-technology association matrix, and a joint association matrix, performs non-negative matrix ternary decomposition on the joint association matrix, generates scene joint embedding vectors corresponding to each intelligent agent scene based on the decomposition results, maps each scene joint embedding vector to the Poincaré sphere hyperbolic space, and generates at least one of a Venn diagram view, a hierarchical tree view, and a three-dimensional block view based on the hyperbolic embedding coordinates.

[0229] The computer device also includes a display device that receives visual coordinates and view data generated by the processor, and displays intelligent agent scene nodes, business categories, technical capability categories, scene hierarchy relationships, and user interaction selection results.

[0230] Example 4:

[0231] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aforementioned two-dimensional visualization analysis method for intelligent agent scenarios.

[0232] When the computer program is executed by the processor, it reads the basic scene data, business dimension data, technical dimension data, and scene hierarchical relationship data from the intelligent agent scene library, constructs the scene-business association matrix and the scene-technology association matrix, and concatenates the two into a joint association matrix. The joint association matrix is ​​then subjected to non-negative matrix tri-decomposition to obtain the scene latent feature matrix, the association strength matrix, and the dimension latent feature matrix. Based on the scene latent feature matrix, the association strength matrix, and the dimension latent feature matrix, the joint embedding vector of each intelligent agent scene is generated. Based on the joint embedding vector of each scene and the scene hierarchical relationship data, hyperbolic embedding coordinates in the hyperbolic space of the Poincaré sphere are generated. Based on the hyperbolic embedding coordinates, Venn diagram visualization coordinates, hierarchical tree visualization coordinates, and three-dimensional block visualization coordinates are generated.

[0233] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A two-dimensional visualization analysis method for intelligent agent scenarios, characterized in that, include: Read the basic scenario data, business dimension data, technical dimension data and scenario hierarchy relationship data from the intelligent agent scenario library, organize each intelligent agent scenario according to the scenario identifier, write the scenario identifier, scenario name, scenario description and scenario hierarchy identifier into the same standardized scenario record, and classify the business dimension data and technical dimension data into the standardized scenario record with the same scenario identifier according to the scenario identifier in the business dimension data and technical dimension data. Business categories are generated based on business dimension data, and technical capability categories are generated based on technical dimension data. Business categories and technical capability categories are used as column objects, and scenario-business association matrices and scenario-technology association matrices are constructed based on standardized scenario records. By merging the scenario-business association matrix and the scenario-technology association matrix, a joint association matrix is ​​obtained. The joint association matrix is ​​then subjected to non-negative matrix tri-decomposition to obtain the scenario implicit feature matrix, the association strength matrix, and the dimension implicit feature matrix. A joint embedding vector for the scene is generated based on the scene latent feature matrix, the association strength matrix, and the dimension latent feature matrix; A scene tree is built based on the inclusion relationship between the original scene and the micro-scene, as well as the inclusion relationship between the micro-scene and the meta-scene in the scene hierarchy relationship data. The shortest path distance between the scene nodes of the agent in the scene tree is calculated. Hyperbolic embedding coordinates are generated based on the scene joint embedding vector and the shortest path distance. Based on the hyperbolic embedding coordinates, the two-dimensional visualization analysis results of the agent scene, including Venn diagram view, hierarchical tree view and three-dimensional block view, are output.

2. The method for two-dimensional visualization analysis of intelligent agent scenes according to claim 1, characterized in that, The methods for constructing the scenario-business correlation matrix and the scenario-technology correlation matrix include: The matrix row objects are determined according to the scene identifiers in the standardized scene records, the column objects of the scene-business association matrix are determined by the business category, and the column objects of the scene-technology association matrix are determined by the technical capability category. When the business dimension data record corresponds to an intelligent agent scene and a corresponding business category, the corresponding element in the scene-business association matrix is ​​assigned a value of 1, and when no corresponding intelligent agent scene and corresponding business category are recorded, the value is assigned a value of 0. When the technical dimension data record corresponds to an intelligent agent scene and a corresponding technical capability category, the corresponding element in the scene-technology association matrix is ​​assigned a value of 1, and when no corresponding intelligent agent scene and corresponding technical capability category are recorded, the value is assigned a value of 0.

3. The method for two-dimensional visualization analysis of intelligent agent scenes according to claim 1, characterized in that, Methods for performing nonnegative matrix ternary decomposition on the joint incidence matrix include: Based on the pre-set number of scene latent categories and dimension latent categories, initialize the scene latent feature matrix, association strength matrix, and dimension latent feature matrix; update the scene latent feature matrix, association strength matrix, and dimension latent feature matrix sequentially using an alternating multiplication update method, and calculate the reconstruction error between the joint association matrix and the product of the scene latent feature matrix, association strength matrix, and dimension latent feature matrix; stop iterating when the absolute value of the difference between the reconstruction error of the current iteration and the reconstruction error of the previous iteration is less than or equal to the preset reconstruction error change threshold, and also stop iterating when the number of iterations reaches the preset maximum number of iterations.

4. The method for two-dimensional visualization analysis of intelligent agent scenes according to claim 1, characterized in that, The method for generating the joint embedding vector of the scene includes: Read the scene latent feature row vector corresponding to each agent scene in the scene latent feature matrix; multiply the scene latent feature row vector with the association strength matrix to obtain the dimension latent association vector of the corresponding agent scene; multiply the dimension latent association vector with the transpose of the dimension latent feature matrix to obtain the scene joint embedding vector, where the first part of the scene joint embedding vector corresponds to the business category and the last part corresponds to the technical capability category.

5. The method for two-dimensional visualization analysis of intelligent agent scenes according to claim 1, characterized in that, Methods for calculating the shortest path distance include: Read the connection path between any two intelligent agent scene nodes in the scene tree; Count the number of edges traversed by the connecting path, and determine the connecting path with the smallest number of edges as the shortest connecting path; The number of edges corresponding to the shortest connection path is determined as the shortest path distance between the corresponding two intelligent agent scene nodes.

6. The method for two-dimensional visualization analysis of intelligent agent scenes according to claim 5, characterized in that, The method for generating hyperbolic embedding coordinates includes: The scene joint embedding vector is projected onto a two-dimensional space, and the projection result is scaled to the inside of a Poincaré sphere to obtain the initial hyperbolic coordinates corresponding to the scene nodes of each agent. The root scene node in the scene tree is determined, and the hyperbolic embedding coordinates of the root scene node are fixed as the origin of the Poincaré sphere. Hierarchical distance constraints are constructed based on the shortest path distance, and association preservation constraints are constructed based on the initial hyperbolic coordinates. Based on the hierarchical distance constraints and association preservation constraints, the initial hyperbolic coordinates of all nodes except the root scene node are iteratively adjusted using the Riemann stochastic gradient descent method. The iteration stops when the absolute value of the difference between the hyperbolic embedding error of the current iteration and the hyperbolic embedding error of the previous iteration is less than or equal to a preset embedding error change threshold. The iteration also stops when the number of iterations reaches the preset number of embedding iterations, and the coordinates at the time of stopping the iteration are determined as the hyperbolic embedding coordinates.

7. The method for two-dimensional visualization analysis of intelligent agent scenes according to claim 1, characterized in that, The method for generating the Venn diagram view includes: Read the user-selected business category and technical capability category, and form a business-related scenario set and a technical-related scenario set respectively based on the association values ​​of the corresponding business category column and technical capability category column in the joint association matrix; determine the Venn diagram overlapping area, business category area and technical capability category area based on the intersection and difference between the business-related scenario set and the technical-related scenario set; determine the display position of each intelligent agent scenario node in the corresponding Venn diagram area based on the hyperbolic embedding coordinates of each intelligent agent scenario node, and generate a Venn diagram view.

8. The method for two-dimensional visualization analysis of intelligent agent scenes according to claim 1, characterized in that, The method for generating the three-dimensional block view includes: The planar position of the 3D block corresponding to each agent scene is determined based on hyperbolic embedding coordinates; The number of technical capability categories with a correlation value greater than zero corresponding to each intelligent agent scene is counted based on the scene-technology association matrix, and the height of the three-dimensional block is determined based on the number of technical capability categories. Generate a 3D block view based on planar position and 3D block height.

9. A two-dimensional visualization analysis system for intelligent agent scenarios, characterized in that, include: The scenario organization module is used to read the basic scenario data, business dimension data, technical dimension data and scenario hierarchy relationship data in the intelligent agent scenario library, organize the scenarios of each intelligent agent according to the scenario identifier, write the scenario identifier, scenario name, scenario description and scenario hierarchy identifier into the same standardized scenario record, and classify the business dimension data and technical dimension data into the standardized scenario record with the same scenario identifier according to the scenario identifier in the business dimension data and technical dimension data. The matrix construction module is used to generate business categories based on business dimension data and technical capability categories based on technical dimension data. It uses business categories and technical capability categories as column objects and constructs scenario-business association matrices and scenario-technology association matrices based on standardized scenario records. The matrix decomposition module is used to concatenate the scenario-business association matrix and the scenario-technology association matrix to obtain a joint association matrix, and to perform non-negative matrix ternary decomposition on the joint association matrix to obtain the scenario implicit feature matrix, the association strength matrix, and the dimension implicit feature matrix. The joint embedding module is used to generate scene joint embedding vectors based on the scene latent feature matrix, the association strength matrix, and the dimension latent feature matrix; The hyperbolic output module is used to build a scene tree based on the inclusion relationship between the original scene and the micro scene and the inclusion relationship between the micro scene and the meta scene in the scene hierarchy relationship data. It calculates the shortest path distance between the scene nodes of the agent in the scene tree, generates hyperbolic embedding coordinates based on the scene joint embedding vector and the shortest path distance, and outputs the two-dimensional visualization analysis results of the agent scene, including Venn diagram view, hierarchical tree view and three-dimensional block view, based on the hyperbolic embedding coordinates.

10. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executed by the processor, characterized in that the processor executes the computer program to implement a two-dimensional visualization analysis method for intelligent agent scenes according to any one of claims 1-8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements a two-dimensional visualization analysis method for intelligent agent scenes as described in any one of claims 1-8.

Images