Efficient Query Method, Device and Computer Equipment for Enterprise Information Review

Abstract

The present application relates to an efficient query method, device, and computer equipment for enterprise information review. The method includes: obtaining the query information of a user and the data domain knowledge graph of multiple enterprises, and based on the query information, identifying the content of the user's query requirements through a semantic recognition network; generating graph query content and risk index recommendation information based on the content of the query requirements, and generating a feature graph query scheme for the user based on the graph query content and the risk index recommendation information; collecting adjustment information of the user for the feature graph query scheme, and adjusting the feature graph query scheme based on the adjustment information to obtain a target feature graph scheme; querying the target query result of the user through the data domain knowledge graph based on the target feature graph scheme. Using this method can improve the accuracy of risk identification and analysis of enterprise association structures.

Description

Technical Field

[0001] This application relates to the fields of risk analysis and natural language technology, and particularly to an efficient query method, device, and computer equipment for enterprise information review. Background Art

[0002] With the improvement of enterprise information digitization and regulatory compliance requirements, knowledge graphs based on enterprise relationships play an important role in fields such as risk control, due diligence, industry analysis, supply chain monitoring, and investment and financing decisions. Enterprise knowledge graphs include equity structures, affiliated company relationships, upstream and downstream information in the industrial chain, financial data, legal litigation information, and risk annotation information. In actual business scenarios, there are various complex risk feature patterns, such as multi-layer nested shareholdings, circular equity chains, cross-shareholding networks, etc. These features are often the basis for the hidden behaviors of high-risk enterprises. Quickly identifying and retrieving similar feature graphs helps improve the efficiency of risk review and prediction. Therefore, how to improve the efficiency of risk review and prediction of enterprises is the current research focus.

[0003] The existing technology constructs an ultra-large-scale enterprise knowledge graph through the use of graph databases such as Neo4j and JanusGraph, and represents enterprise entities and their relationships such as equity, investment, and supply chain through node and edge models. It also vectorizes enterprise nodes and edges based on embedding algorithms such as TransE, DeepWalk, Node2Vec, and GraphSAGE, and then uses cosine similarity or Euclidean distance to implement the retrieval of similar nodes or similar subgraphs. However, in the face of an ultra-large-scale knowledge graph, it is very time-consuming to find substructures similar to a certain high-risk subgraph, and most retrievals can only cover relatively limited patterns, lacking flexible scalability. Moreover, such technologies are usually weak in interpretability and customization, resulting in poor accuracy in risk identification and analysis of enterprise association structures. Summary of the Invention

[0004] Based on this, it is necessary to provide an efficient query method, device, computer equipment, computer-readable storage medium, and computer program product for enterprise information review to address the above technical problems.

[0005] In a first aspect, this application provides an efficient query method for enterprise information review, including:

[0006] Obtain the user's query information and the data domain knowledge graphs of multiple enterprises, and based on the query information, identify the content of the user's query requirements through a semantic recognition network;

[0007] Based on the content of the query requirements, generate graph query content and risk index suggestion information, and based on the graph query content and the risk index suggestion information, generate the user's feature graph query plan;

[0008] Collect the adjustment information of the user for the feature map query scheme, and based on the adjustment information, adjust the feature map query scheme to obtain a target feature map scheme;

[0009] Based on the target feature map scheme, query the target query result of the user through the data domain knowledge graph.

[0010] Optionally, the obtaining of the data domain knowledge graphs of multiple enterprises includes:

[0011] Obtain the enterprise entity information of each enterprise and the enterprise relationship information of each enterprise, and for each enterprise, extract the enterprise node vectors of the enterprise entity information of the enterprise and the enterprise relationship edge vectors of the enterprise relationship information of the enterprise;

[0012] Collect the enterprise similarity weight information of the enterprise and the minimum feature subgraphs of the enterprise features, and based on the enterprise similarity weight information of the enterprise, construct the similarity measurement information of the enterprise;

[0013] Based on the minimum feature subgraphs of the enterprise features, generate the embedded subgraph information of the enterprise, and based on the minimum feature subgraphs of the feature representations, the enterprise node vectors, the enterprise relationship edge vectors, and the similarity measurement information of the enterprise, construct the data domain knowledge graphs of multiple enterprises.

[0014] Optionally, the identifying of the query requirement content of the user through the semantic recognition network based on the query information includes:

[0015] Identify the query text information corresponding to the query information, and through the semantic recognition network, identify the text semantic content in the query text information;

[0016] Through the semantic parsing strategy, identify the key entity content and the relationship content in the text semantic content, and use the key entity content and the relationship content as the query requirement content of the user.

[0017] Optionally, the generating of the graph query content and the risk index suggestion information based on the query requirement content includes:

[0018] Based on the key entity content and the relationship content, filter the query statement information and the attribute association information of the user;

[0019] Convert the query statement information into each query feature condition, and use all the query feature conditions as the graph query content;

[0020] Based on the attribute association information of the user, through the risk index recommendation network, generate the attribute index information of each risk index recommendation type of the user, and use all the attribute index information as the risk index recommendation information.

[0021] Optionally, the generating the feature graph query scheme of the user based on the graph query content and the risk index recommendation information includes:

[0022] Based on each of the query feature conditions, identify the node type and the relationship type, and based on the attribute index information of each risk index recommendation type, identify the feature graph query condition information;

[0023] Based on the node type, the relationship type, and the feature graph query condition information, generate the initial feature graph query scheme of the user, and through the visualization generation strategy, perform visualization display processing on the initial feature graph query scheme to obtain the feature graph query scheme.

[0024] Optionally, the adjusting the feature graph query scheme based on the adjustment information to obtain the target feature graph scheme includes:

[0025] Identify each adjustment content in the adjustment information, and based on the adjustment range corresponding to each adjustment content, query the target adjustment amount corresponding to each adjustment content and the adjustment position information corresponding to each adjustment content in the feature graph query scheme;

[0026] Based on the target adjustment amount corresponding to each adjustment content and the adjustment position information corresponding to each adjustment content, perform adjustment processing on the feature graph query scheme to obtain the target feature graph scheme.

[0027] In a second aspect, the present application also provides an efficient query device for enterprise information review, including:

[0028] An acquisition module, configured to acquire the query information of the user and the data domain knowledge graph of multiple enterprises, and based on the query information, through the semantic recognition network, identify the query requirement content of the user;

[0029] A generation module, configured to generate graph query content and risk index recommendation information based on the query requirement content, and generate the feature graph query scheme of the user based on the graph query content and the risk index recommendation information;

[0030] An adjustment module, configured to collect the adjustment information of the user for the feature graph query scheme, and based on the adjustment information, adjust the feature graph query scheme to obtain the target feature graph scheme;

[0031] A query module, configured to query the target query result of the user through the data domain knowledge graph based on the target feature map scheme.

[0032] Optionally, the obtaining module is specifically configured to:

[0033] Obtain the enterprise entity information of each enterprise and the enterprise relationship information of each enterprise, and for each enterprise, extract the enterprise node vectors of the enterprise entity information of the enterprise and the enterprise relationship edge vectors of the enterprise relationship information of the enterprise;

[0034] Collect the enterprise similarity weight information of the enterprise and the minimum feature subgraphs of the enterprise features of the enterprise, and construct the similarity measurement information of the enterprise based on the enterprise similarity weight information of the enterprise;

[0035] Generate the embedded subgraph information of the enterprise based on the minimum feature subgraphs of the enterprise features, and construct the data domain knowledge graph of multiple enterprises based on the minimum feature subgraphs of each feature representation, each enterprise node vector, each enterprise relationship edge vector, and the similarity measurement information of the enterprise.

[0036] Optionally, the obtaining module is specifically configured to:

[0037] Identify the query text information corresponding to the query information, and identify the text semantic content in the query text information through a semantic recognition network;

[0038] Identify the key entity content in the text semantic content and the relationship content in the text semantic content through a semantic parsing strategy, and use the key entity content and the relationship content as the query requirement content of the user.

[0039] Optionally, the generating module is specifically configured to:

[0040] Filter the query statement information and attribute association information of the user based on the key entity content and the relationship content;

[0041] Convert the query statement information into each query feature condition, and use all the query feature conditions as the graph query content;

[0042] Generate the attribute index information of each risk index recommendation type of the user through a risk index recommendation network based on the attribute association information of the user, and use all the attribute index information as the risk index recommendation information.

[0043] Optionally, the generating module is specifically configured to:

[0044] Based on each of the query feature conditions, identify the node type and the relationship type, and based on the attribute index information of each of the risk index suggestion types, identify the feature map query condition information;

[0045] Based on the node type, the relationship type, and the feature map query condition information, generate an initial feature map query scheme for the user, and through a visualization generation strategy, perform a visualization display process on the initial feature map query scheme to obtain a feature map query scheme.

[0046] Optionally, the adjustment module is specifically configured to:

[0047] Identify each adjustment content in the adjustment information, and based on the adjustment range corresponding to each adjustment content, query the target adjustment amount corresponding to each adjustment content and the adjustment position information corresponding to each adjustment content in the feature map query scheme;

[0048] Based on the target adjustment amount corresponding to each adjustment content and the adjustment position information corresponding to each adjustment content, perform an adjustment process on the feature map query scheme to obtain a target feature map scheme.

[0049] In a third aspect, the present application provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the steps of the method described in any one of the first aspects are implemented.

[0050] In a fourth aspect, the present application provides a computer-readable storage medium. A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the method described in any one of the first aspects are implemented.

[0051] In a fifth aspect, the present application provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the method described in any one of the first aspects are implemented.

[0052] The above-mentioned efficient query method, device, and computer equipment for enterprise information review obtain the user's query information and the data domain knowledge graph of multiple enterprises, and based on the query information, use a semantic recognition network to identify the content of the user's query requirements; based on the content of the query requirements, generate graph query content and risk index recommendation information, and based on the graph query content and the risk index recommendation information, generate the user's feature graph query plan; collect the adjustment information of the user for the feature graph query plan, and based on the adjustment information, adjust the feature graph query plan to obtain a target feature graph plan; based on the target feature graph plan, query the user's target query result through the data domain knowledge graph. In this solution, by leveraging the powerful understanding and generation capabilities of large language models, users can describe complex query requirements in natural language, and the system automatically converts them into precise graph query conditions and feature graph plans. At the same time, through the automatic suggestion of attribute filtering conditions and risk indicators, the technical threshold for business personnel is significantly reduced, and the query efficiency and accuracy are improved. The generated target feature graph plan introduces visual information, enabling business personnel to intuitively edit and optimize the feature graph plan, further enhancing the flexibility and interpretability of the system. The continuous optimization of the reinforcement learning mechanism ensures that the system can continuously adapt to and meet the dynamically changing business needs, providing an efficient, intelligent, and reliable enterprise risk analysis solution. Thus, the accuracy of risk identification and analysis of the enterprise association structure is comprehensively improved. Brief Description of the Drawings

[0053] To more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0054] Figure 1 It is a schematic flowchart of an efficient query method for enterprise information review in an embodiment;

[0055] Figure 2 It is a schematic flowchart of an efficient query example for enterprise information review in an embodiment;

[0056] Figure 3 It is a block diagram of the structure of an efficient query device for enterprise information review in an embodiment;

[0057] Figure 4 It is an internal structure diagram of a computer device in an embodiment. Detailed Description of the Embodiments

[0058] In order to make the objectives, technical solutions, and advantages of this application clearer and more understandable, the following further details this application in combination with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0059] The efficient query method for enterprise information auditing provided by the embodiments of this application can be applied to the application environment of efficient query of enterprise information auditing. Among them, this method can be applied to a terminal, a server, or a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. Among them, the terminal can be, but is not limited to, various personal computers, laptop computers, mid-range computers, etc. By leveraging the powerful understanding and generation capabilities of large language models, users can describe complex query requirements in natural language, and the system automatically converts them into precise graph query conditions and feature graph schemes. At the same time, through the automatic suggestion of attribute filtering conditions and risk indicators, the technical threshold for business personnel is significantly reduced, and the query efficiency and accuracy are improved. The generated target feature graph scheme introduces visual information, enabling business personnel to intuitively edit and optimize the feature graph scheme, further enhancing the flexibility and interpretability of the system. The continuous optimization of the reinforcement learning mechanism ensures that the system can continuously adapt to and meet the dynamically changing business needs, providing an efficient, intelligent, and reliable enterprise risk analysis solution. Thereby comprehensively improving the risk identification and analysis accuracy of the enterprise association structure.

[0060] In an exemplary embodiment, as Figure 1 shown, an efficient query method for enterprise information auditing is provided. Taking the application of this method to a terminal as an example, it includes the following steps S101 to S104. Among them:

[0061] Step S101, obtain the user's query information and the data domain knowledge graph of multiple enterprises, and based on the query information, identify the content of the user's query requirements through a semantic recognition network.

[0062] In this embodiment, the terminal obtains the user's query information in response to the user's query information upload operation. The query information can be audio information or text information. Then, the terminal obtains the data domain knowledge graphs of each enterprise pre-constructed. The data domain knowledge graph is a knowledge graph in which enterprise-specific relationships and attributes (such as equity, voting rights, holding, senior executives, registered capital, etc.) are introduced into the mathematical model and are emphasized in graph embedding and similarity measurement. The specific construction process will be described in detail later. The knowledge graph can accurately measure and identify similar high-risk structures in the enterprise data space. The predefined seed subgraph is used as the basic feature unit, and through mathematical linear combination, graph kernel method and GNN hierarchical expansion mechanism, the knowledge graph has high scalability and business adaptability. This mathematical model not only improves the efficiency and accuracy of similarity retrieval, but also ensures the stability and reliability of the system in a complex enterprise data environment. Then, based on the query information, the terminal identifies the content of the user's query needs through a semantic recognition network. The semantic recognition network is a text semantic recognition network based on a large language model.

[0063] Step S102: Generate graph query content and risk index recommendation information based on the content of the query requirement, and generate a feature graph query scheme for the user based on the graph query content and the risk index recommendation information.

[0064] In this embodiment, the terminal generates graph query content and risk index recommendation information based on the content of the query requirement, and generates a feature graph query scheme for the user based on the graph query content and the risk index recommendation information. The graph query content is the association query content for the user to query the association information between each enterprise, and the risk index recommendation information is the index information related to the risk index between the two enterprises concerned by the user. The specific generation process will be described in detail later.

[0065] Step S103: Collect the adjustment information of the user for the feature graph query scheme, and adjust the feature graph query scheme based on the adjustment information to obtain a target feature graph scheme.

[0066] In this embodiment, the terminal collects the adjustment information of the user for the feature graph query scheme, and adjusts the feature graph query scheme based on the adjustment information to obtain a target feature graph scheme. The feature graph query scheme includes the node type, relationship type, attribute filtering conditions, and similarity setting information of the enterprises to be queried. The adjustment information includes the adjustment content of one or more of the above contents in the feature graph query scheme.

[0067] Step S104: Query the target query result of the user through the data domain knowledge graph based on the target feature graph scheme.

[0068] In this embodiment, the terminal queries the target query result of the user through the data domain knowledge graph based on the target feature map solution. Among them, the knowledge graph query method is to extract the enterprise information of the target enterprise that the user needs to query, and based on the target feature map solution, query the target similar enterprises in the data domain knowledge graphs of each enterprise whose similarity to the target enterprise is greater than the preset similarity threshold, and summarize the similarity content between the target similar enterprises and the target enterprise to obtain the target query result of the user.

[0069] Based on the above solution, by leveraging the powerful understanding and generation capabilities of the large language model, users can describe complex query requirements in natural language, and the system automatically converts them into precise graph query conditions and feature map solutions. At the same time, through the automatic suggestion of attribute filtering conditions and risk indicators, the technical threshold for business personnel is significantly reduced, and the query efficiency and accuracy are improved. The generated target feature map solution introduces visual information, enabling business personnel to intuitively edit and optimize the feature map solution, further enhancing the flexibility and interpretability of the system. The continuous optimization of the reinforcement learning mechanism ensures that the system can continuously adapt to and meet the dynamically changing business needs, providing an efficient, intelligent, and reliable enterprise risk analysis solution. Thus, the risk identification and analysis accuracy of the enterprise association structure are comprehensively improved.

[0070] Optionally, obtaining the data domain knowledge graphs of multiple enterprises includes: obtaining the enterprise entity information of each enterprise and the enterprise relationship information of each enterprise, and for each enterprise, extracting the enterprise node vectors of the enterprise entity information of the enterprise and the enterprise relationship edge vectors of the enterprise relationship information of the enterprise; collecting the enterprise similarity weight information of the enterprise and the minimum feature subgraphs of the enterprise features, and constructing the similarity measurement information of the enterprise based on the enterprise similarity weight information of the enterprise; generating the embedded subgraph information of each enterprise based on the minimum feature subgraphs of each enterprise feature, and constructing the data domain knowledge graphs of multiple enterprises based on the minimum feature subgraphs of each feature representation, the enterprise node vectors, the enterprise relationship edge vectors, and the similarity measurement information of the enterprise.

[0071] In this embodiment, the terminal obtains the enterprise entity information of each enterprise and the enterprise relationship information of each enterprise, and for each enterprise, extracts the enterprise node vectors of the enterprise entity information of the enterprise and the enterprise relationship edge vectors of the enterprise relationship information of the enterprise. Then, the terminal collects the enterprise similarity weight information of the enterprise and the minimum feature subgraphs of the enterprise features, and constructs the similarity measurement information of the enterprise based on the enterprise similarity weight information of the enterprise. Among them, the similarity measurement information includes pre-formaldehyde similarity, enterprise-specific distance measurement, and business weight optimization information, etc.

[0072] After renting, the terminal generates the embedded subgraph information of each enterprise based on the minimum feature subgraph of each enterprise's characteristics, and constructs a data domain knowledge graph of multiple enterprises based on the minimum feature subgraph of each feature representation, each enterprise node vector, each enterprise relationship edge vector, and the similarity measurement information of the enterprises.

[0073] Specifically,

[0074] 1. Enterprise data modeling:

[0075] First, the terminal needs to clearly represent enterprise-specific entities and relationships in the knowledge graph. This includes:

[0076] Entities (Nodes):

[0077] · Company.

[0078] · Executive.

[0079] · Shareholder.

[0080] Relationships (Edges):

[0081] · Ownership: Represents the equity holding relationship between enterprises.

[0082] · VotingRights: Represents the voting rights of shareholders in an enterprise.

[0083] · Control: Represents the control of one enterprise over another.

[0084] · Executive Role: Represents the position of an executive in an enterprise.

[0085] · Registered Capital: Represents the amount of registered capital of an enterprise.

[0086] These entities and relationships can be described in detail through attribute vectors.

[0087] For example:

[0088] · The attribute vector (ф(v)) of the enterprise node (v) includes:

[0089] · Registered capital (C v )

[0090] · Number of executives (E v )

[0091] · Holding ratio (H v )

[0092] · The attribute vector (ψ(e)) of the equity relationship edge (e=(u,v)) contains

[0093] · Equity ratio (O u,v )

[0094] · Voting right ratio (V u,v )

[0095] 2. Feature vector definition:

[0096] In the enterprise data space, the feature vectors of each node and edge not only contain basic attributes but also enterprise-specific dimensions.

[0097] These vectors will be used as the input for graph embedding.

[0098] Attribute vector of enterprise node:

[0099] ;

[0100] Among them:

[0101] · (C v ) represents the registered capital of enterprise (v).

[0102] · (E v ) represents the number of senior executives of enterprise (v).

[0103] · (H v ) represents the holding ratio of enterprise (v).

[0104] Attribute vector of equity relationship edge:

[0105] ;

[0106] Among them:

[0107] · (O u，v ) represents the equity ratio of enterprise (u) to enterprise (v).

[0108] · (V u，v ) represents the voting right ratio of enterprise (u) to enterprise (v).

[0109] These attribute vectors will be input into the graph embedding model to generate more business-semantic embedding representations.

[0110] 3. Embedding function extension:

[0111] Traditional graph embedding methods (such as TransE, GraphSAGE, Graph Transfommer, Graph2Vec) can be extended to make full use of enterprise-specific attribute vectors.

[0112] 3.1 Embedding Function Definition:

[0113] Define an enhanced embedding function (f) that not only considers the structural information of nodes and edges but also their attribute vectors:

[0114] ;

[0115] where (G’) is a subgraph of the knowledge graph and (D) is the dimension of the embedding vector.

[0116] 3.2 Aggregation Function Extension:

[0117] The traditional aggregation function (f(G')) needs to be extended to include enterprise-specific attributes:

[0118] ;

[0119] Specifically, the aggregation function can be defined as:

[0120] ;

[0121] where (GNN) represents a graph neural network used to generate subgraph embeddings by integrating the attribute vectors of nodes and edges.

[0122] 4. Similarity Metric Optimization:

[0123] In the vector space, when defining the similarity metric, the weights and impacts of enterprise-specific dimensions need to be considered.

[0124] 4.1 Weighted Cosine Similarity:

[0125] To emphasize enterprise-specific relationships, different weights can be assigned to different dimensions of the vector space. Assume the embedding vector is ( ), and the weight vector is ( ), then the weighted cosine similarity is defined as:

[0126] ;

[0127] where ( ) represents the element-wise Hadamard product.

[0128] 4.2 Enterprise-Specific Distance Metric:

[0129] Define a distance metric based on enterprise-specific dimensions, such as the weighted Euclidean distance:

[0130] ;

[0131] where (w i ) is the weight of the (i)-th dimension, reflecting the importance of this dimension in the similarity metric.

[0132] 4.3 Business Weight Optimization

[0133] The weight vector ( ) can be optimized according to business requirements. For example:

[0134] The equity ratio (O) and voting right ratio (V) can be given higher weights to highlight the importance of control and decision-making power.

[0135] The registered capital (C) and the number of senior executives (E) are given medium weights to reflect the enterprise scale and management complexity.

[0136] 5. Seed Subgraph Expansion:

[0137] 5.1 Definition of Seed Subgraph:

[0138] In the enterprise data space, a set of representative minimum feature subgraphs (seed subgraphs) are predefined. For example:

[0139] Three-node closed loop: ( ).

[0140] Four-node clique loop: ( ).

[0141] Cross-shareholding pattern: ( ) and ( ).

[0142] These seed subgraphs represent common high-risk equity structure patterns.

[0143] 5.2 Vectorization of Seed Subgraph

[0144] For each seed subgraph (S i ), calculate its embedding vector (S i ):

[0145] ;

[0146] Store these embedding vectors in a vector database (such as Faiss, HNSWlih, Mivus) for fast retrieval and similarity matching.

[0147] 5.3 Expansion Mechanism of Seed Subgraph:

[0148] To expand the seed subgraph to cover more complex structures, the following mathematical mechanisms are introduced at the terminal:

[0149] 5.3.1 Subgraph Decomposition and Recombination:

[0150] Any complex subgraph (G’) can be decomposed into a combination of several seed subgraphs. Suppose (G’) contains (n) seed subgraphs (S1, S2, …, S n ), then its embedding can be approximately represented as:

[0151] ;

[0152] where, ( ) is the combination coefficient, and (∈) is the error term.

[0153] 5.3.2 Feature Mapping and Weighted Combination:

[0154] Using the feature mapping method, the subgraph embedding vectors are combined through weighted combination to form the embedding of the complex subgraph:

[0155] ;

[0156] where, ( ) is the weight matrix, which is used to adjust the contribution of different seed subgraphs in the embedding.

[0157] 5.3.3 Graph Kernel Method:

[0158] Define a graph kernel based on seed subgraphs :

[0159] ;

[0160] where, measures the matching degree of (G1) and (G2) on the seed subgraph (S i ), for example:

[0161] ;

[0162] And the graph kernel is converted into the embedding space through feature mapping:

[0163] ;

[0164] Then it is converted into a low-dimensional embedding vector through dimensionality reduction or non-linear mapping.

[0165] 5.3.4 GNN Hierarchical Expansion:

[0166] When using the graph neural network (GNN), the seed subgraph embedding is used as the detector for the initial layer or the intermediate layer:

[0167] ;

[0168] where, ( ) is the representation of node (v) at the first layer, ( ) is the set of neighbor nodes of node (v). Through multi - layer propagation, the features of the seed sub - graph will be reflected and extended in the embedding space through the message - passing mechanism.

[0169] 6.1 Definition of the enterprise data vector space:

[0170] Define the vector representation of each enterprise (v) in the enterprise data space , where (D) contains the following dimensions:

[0171] ;

[0172] Where:

[0173] ·(C v ): Registered capital;

[0174] ·(E v ): Number of senior executives;

[0175] ·(H v ): Shareholding ratio;

[0176] ·(ф(v)): High - dimensional feature vector generated by graph embedding, containing relationship information such as equity and voting rights.

[0177] 6.2 Definition of similarity measure in the enterprise data space:

[0178] In the enterprise data space, define a comprehensive similarity measure (Sim(u, v)), considering the weights of multiple enterprise dimensions:

[0179] ;

[0180] Where:

[0181] ·( ) is the similarity measure of the (i) - th dimension. For example:

[0182] ·( )(Cosine similarity of registered capital).

[0183] ·( )(Negative distance of shareholding ratio).

[0184] ·(w i ) is the weight of the (i) - th dimension, reflecting its importance in the overall similarity measure.

[0185] 7.1 Embedding representation of the Seed sub - graph prototype:

[0186] The embedding vector ( ) of each seed sub - graph ( ) contains the comprehensive features of enterprise - specific relationships:

[0187] ;

[0188] Among them:

[0189] ·( ): The average registered capital of enterprises in the seed sub - graph;

[0190] ·( ): The average number of senior executives of enterprises in the seed sub - graph;

[0191] ·( ): The average shareholding ratio of enterprises in the seed sub - graph;

[0192] ·( ): The high - dimensional feature vector generated by the graph embedding model.

[0193] 7.2 Linear combination and extension of the Seed sub - graph:

[0194] For any complex sub - graph (G’), its embedding vector (f(G)) can be expressed as a weighted combination of several seed sub - graph vectors:

[0195] ;

[0196] Among them:

[0197] ·( ) is the weight coefficient, reflecting the contribution degree of the seed sub - graph (Si) in the complex sub - graph (G’);

[0198] ·(k) is the total number of seed sub - graphs.

[0199] 7.3 Optimization and extension mechanism:

[0200] When some complex sub - graphs (G’) cannot be fully represented by the existing seed sub - graphs ( ), new seed sub - graphs ( ) need to be introduced:

[0201] ;

[0202] By continuously introducing new seed sub - graphs and retraining the graph embedding model, the model's ability to represent complex structures can be enhanced.

[0203] 8. Complete expression of the mathematical model:

[0204] The mathematical model combining enterprise - specific relationships can be summarized as follows:

[0205] 8.1 Attribute vectors of nodes and edges

[0206] Attribute vector of enterprise node (v):

[0207] ф(v) = [C v , E v , H v , ф struct (v)] ∈ R d

[0208] Where:

[0209] ·(C v ): Registered capital;

[0210] ·(E v ): Number of senior executives;

[0211] ·(H v ): Shareholding ratio;

[0212] ·(ф struct (v)): Structural feature vector obtained through the graph embedding model;

[0213] ·Attribute vector of equity relationship edge (e=(u,v)):

[0214] ψ(e) = [O u，v , V u，v , ф struct (v)] ∈ R d

[0215] Where:

[0216] ·(O u，v ) represents the equity ratio of enterprise (u) to enterprise (v).

[0217] ·(V u，v ) represents the voting right ratio of enterprise (u) to enterprise (v).

[0218] ·(ф struct (v)): Structural feature vector obtained through the graph embedding model.

[0219] 8.2 Subgraph embedding function:

[0220] ;

[0221] 8.3 Similarity measure:

[0222] Based on weighted cosine similarity:

[0223] ;

[0224] Where, ( ) represents the element-wise Hadamard product.

[0225] 8.4 Seed Subgraph Expansion:

[0226] ;

[0227] Among them, ( ) is the embedding vector of the predefined seed subgraph ( ), and ( ) is the combination coefficient.

[0228] Based on the above scheme, by introducing enterprise-specific relationships and attributes (such as equity, voting rights, shareholding, executives, registered capital, etc.) into the mathematical model and highlighting them in graph embedding and similarity measurement, this system can accurately measure and identify similar high-risk structures in the enterprise data space. The predefined seed subgraph, as the basic feature unit, enables the system to have high scalability and business adaptability through mathematical linear combination, graph kernel methods, and GNN hierarchical expansion mechanisms. This mathematical model not only improves the efficiency and accuracy of similarity retrieval but also ensures the stability and reliability of the system in a complex enterprise data environment.

[0229] Optionally, based on the query information, through the semantic recognition network, identify the content of the user's query needs, including: identifying the query text information corresponding to the query information, and through the semantic recognition network, identifying the text semantic content in the query text information; through the semantic parsing strategy, identifying the key entity content in the text semantic content and the relationship content in the text semantic content, and taking the key entity content and the relationship content as the content of the user's query needs.

[0230] In this embodiment, the terminal identifies the query text information corresponding to the query information and, through the semantic recognition network, identifies the text semantic content in the query text information. Among them, the semantic recognition network is a neural network based on a large language model. Then, the terminal, through the semantic parsing strategy, identifies the key entity content in the text semantic content and the relationship content in the text semantic content. Among them, the semantic parsing strategy is a parsing strategy based on a large language model, and the key entity content is the similar enterprise characteristics between two enterprises. For example, "enterprise characteristics, executives, shareholders, etc.", and the relationship content is the similar enterprise structure information between two enterprises, such as "equity relationship, voting rights relationship, shareholding relationship, executive relationship, and registered capital relationship, etc."

[0231] Finally, the terminal takes the key entity content and the relationship content as the content of the user's query needs.

[0232] Based on the above scheme, by combining the large language model, the terminal can efficiently and accurately extract the content of the user's query needs, thereby improving the extraction accuracy of the query need content.

[0233] Optionally, based on the query requirement content, generate graph query content and risk indicator recommendation information, including: filtering the user's query statement information and attribute association information based on the key entity content and relationship content; converting the query statement information into each query feature condition, and using all the query feature conditions as the graph query content; based on the user's attribute association information, generating the attribute index information of each risk indicator recommendation type of the user through the risk indicator recommendation network, and using all the attribute index information as the risk indicator recommendation information.

[0234] In this embodiment, the terminal filters the user's query statement information and attribute association information based on the key entity content and relationship content. Among them, the query statement information is the similar statement information between the two queried enterprises. For example, an enterprise that is a high-risk enterprise and has a similar structure to enterprise A, and the attribute association information is the similar information that needs to be concerned between the two enterprises, such as the holding ratio exceeding a percentage, the equity cycle depth, and the cosine similarity, etc.

[0235] Subsequently, the terminal converts the query statement information into each query feature condition. For example, converting "similar high-risk enterprise structures" into a query condition including features such as equity cycle and holding ratio. And the terminal uses all the query feature conditions as the graph query content.

[0236] Finally, the terminal generates the attribute index information of each risk indicator recommendation type of the user through the risk indicator recommendation network based on the user's attribute association information, and uses all the attribute index information as the risk indicator recommendation information. Among them, each risk indicator recommendation type includes but is not limited to types such as attribute filtering conditions, risk indicators, and similarity parameters. For example, the terminal automatically recommends relevant attribute filtering conditions (such as "holding ratio exceeding 30%") and risk indicators (such as "equity cycle depth of 3 layers"), and similarity parameters (such as "cosine similarity ≥ 0.8") through a large language model based on the business knowledge base and historical data.

[0237] Based on the above solution, by leveraging a large language model (LLM), the system can convert the user's natural language input into an executable graph query or sub-graph feature description, automatically recommend attribute filtering conditions, risk indicators, and similarity parameters, which not only reduces the technical threshold but also improves the flexibility and user experience of the system.

[0238] Optionally, based on the graph query content and the risk index recommendation information, a feature graph query scheme for the user is generated, including: identifying the node type and the relationship type based on each query feature condition, and identifying the feature graph query condition information based on the attribute index information of each risk index recommendation type; generating an initial feature graph query scheme for the user based on the node type, the relationship type, and the feature graph query condition information, and performing a visualization display process on the initial feature graph query scheme through a visualization generation strategy to obtain the feature graph query scheme.

[0239] In this embodiment, the terminal identifies the node type and the relationship type based on each query feature condition, and identifies the feature graph query condition information based on the attribute index information of each risk index recommendation type. Then, the terminal generates an initial feature graph query scheme for the user based on the node type, the relationship type, and the feature graph query condition information, and performs a visualization display process on the initial feature graph query scheme through a visualization generation strategy to obtain the feature graph query scheme. Specifically, the generated feature graph scheme is, for example:

[0240] · The system comprehensively generates the following feature graph scheme;

[0241] · Node type: Company;

[0242] · Relationship type: OWNERSHIP;

[0243] Attribute filtering:

[0244] · Shareholding ratio (Ownership. Control ratio) > 30%;

[0245] · Equity cycle depth ≥ 3 layers;

[0246] · Risk indicators: Equity cycle depth, shareholding ratio;

[0247] · Similarity parameter: Cosine similarity ≥ 0.85.

[0248] Based on the above scheme, through the introduction of the visualization business verification layer, business personnel can intuitively edit and optimize the feature graph scheme, further improving the flexibility and interpretability of the system. The continuous optimization of the reinforcement learning mechanism ensures that the system can continuously adapt to and meet the dynamically changing business needs, providing an efficient, intelligent, and reliable enterprise risk analysis solution.

[0249] Optionally, based on the adjustment information, adjust the feature map query scheme to obtain a target feature map scheme, including: identifying each adjustment content in the adjustment information, and based on the adjustment range corresponding to each adjustment content, query the target adjustment amount corresponding to each adjustment content and the adjustment position information corresponding to each adjustment content in the feature map query scheme; based on the target adjustment amount corresponding to each adjustment content and the adjustment position information corresponding to each adjustment content, perform adjustment processing on the feature map query scheme to obtain a target feature map scheme.

[0250] In this embodiment, the terminal identifies each adjustment content in the adjustment information, and based on the adjustment range corresponding to each adjustment content, queries the target adjustment amount corresponding to each adjustment content and the adjustment position information corresponding to each adjustment content in the feature map query scheme. Then, the terminal performs adjustment processing on the feature map query scheme based on the target adjustment amount corresponding to each adjustment content and the adjustment position information corresponding to each adjustment content to obtain a target feature map scheme.

[0251] Based on the above scheme, by visually presenting the feature map query scheme to the user, the user can accurately and efficiently determine the adjustment content, so as to efficiently adjust the scheme, improving the accuracy and efficiency of the obtained target feature map scheme.

[0252] This application also provides an efficient query example for enterprise information review, as Figure 2 shown, and the specific processing process includes the following steps:

[0253] Step S201, obtain the query information of the user.

[0254] Step S202, obtain the enterprise entity information of each enterprise and the enterprise relationship information of each enterprise, and for each enterprise, extract each enterprise node vector of the enterprise entity information of the enterprise and each enterprise relationship edge vector of the enterprise relationship information of the enterprise.

[0255] Step S203, collect the enterprise similarity weight information of the enterprise and the minimum feature subgraph of each enterprise feature of the enterprise, and based on the enterprise similarity weight information of the enterprise, construct the similarity measurement information of the enterprise.

[0256] Step S204, generate each embedded subgraph information of the enterprise based on the minimum feature subgraph of each enterprise feature, and construct a data domain knowledge graph of multiple enterprises based on the minimum feature subgraph of each feature representation, each enterprise node vector, each enterprise relationship edge vector, and the similarity measurement information of the enterprise.

[0257] Step S205, identify the query text information corresponding to the query information, and identify the text semantic content in the query text information through a semantic recognition network.

[0258] Step S206: Identify the key entity content and relationship content in the text semantic content through a semantic parsing strategy, and use the key entity content and relationship content as the query requirement content of the user.

[0259] Step S207: Filter the query statement information and attribute association information of the user based on the key entity content and relationship content.

[0260] Step S208: Convert the query statement information into each query feature condition, and use all the query feature conditions as the graph query content.

[0261] Step S209: Based on the attribute association information of the user, generate the attribute index information of each risk index recommendation type of the user through a risk index recommendation network, and use all the attribute index information as the risk index recommendation information.

[0262] Step S210: Identify the node type and relationship type based on each query feature condition, and identify the feature graph query condition information based on the attribute index information of each risk index recommendation type.

[0263] Step S211: Generate an initial feature graph query scheme for the user based on the node type, relationship type, and feature graph query condition information, and perform a visualization display process on the initial feature graph query scheme through a visualization generation strategy to obtain a feature graph query scheme.

[0264] Step S212: Collect the adjustment information of the user for the feature graph query scheme.

[0265] Step S213: Identify each adjustment content in the adjustment information, and query the target adjustment amount corresponding to each adjustment content and the adjustment position information corresponding to each adjustment content in the feature graph query scheme based on the adjustment range corresponding to each adjustment content.

[0266] Step S214: Perform an adjustment process on the feature graph query scheme based on the target adjustment amount corresponding to each adjustment content and the adjustment position information corresponding to each adjustment content to obtain a target feature graph scheme.

[0267] Step S215: Query the target query result of the user through the data domain knowledge graph based on the target feature graph scheme.

[0268] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are sequentially shown according to the indications of the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of the steps or stages in other steps or other steps.

[0269] Based on the same inventive concept, an embodiment of the present application further provides an efficient query device for enterprise information review for implementing the efficient query method for enterprise information review involved above. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the efficient query device for enterprise information review provided below can refer to the limitations on the efficient query method for enterprise information review in the above text, and will not be repeated here.

[0270] In an exemplary embodiment, as Figure 3 shown, an efficient query device for enterprise information review is provided, including: an acquisition module 310, a generation module 320, an adjustment module 330, and a query module 340, where:

[0271] The acquisition module 310 is used to acquire the query information of the user and the data domain knowledge graph of multiple enterprises, and based on the query information, identify the content of the user's query requirements through a semantic recognition network;

[0272] The generation module 320 is used to generate graph query content and risk index recommendation information based on the query requirement content, and generate a feature graph query scheme for the user based on the graph query content and the risk index recommendation information;

[0273] The adjustment module 330 is used to collect the adjustment information of the user for the feature graph query scheme, and based on the adjustment information, adjust the feature graph query scheme to obtain a target feature graph scheme;

[0274] The query module 340 is used to query the target query result of the user through the data domain knowledge graph based on the target feature graph scheme.

[0275] Optionally, the acquisition module 310 is specifically used for:

[0276] Obtain the enterprise entity information of each enterprise and the enterprise relationship information of each enterprise, and for each enterprise, extract the enterprise node vectors of the enterprise entity information of the enterprise and the enterprise relationship edge vectors of the enterprise relationship information of the enterprise;

[0277] Collect the enterprise similarity weight information of the enterprise and the minimum feature subgraphs of the enterprise features, and construct the similarity measurement information of the enterprise based on the enterprise similarity weight information of the enterprise;

[0278] Generate the embedded subgraph information of the enterprise based on the minimum feature subgraphs of the enterprise features, and construct the data domain knowledge graph of multiple enterprises based on the minimum feature subgraphs characterized by the features, the enterprise node vectors, the enterprise relationship edge vectors, and the similarity measurement information of the enterprise.

[0279] Optionally, the obtaining module 310 is specifically configured to:

[0280] Identify the query text information corresponding to the query information, and identify the text semantic content in the query text information through a semantic recognition network;

[0281] Identify the key entity content and the relationship content in the text semantic content through a semantic parsing strategy, and use the key entity content and the relationship content as the query requirement content of the user.

[0282] Optionally, the generating module 320 is specifically configured to:

[0283] Filter the query statement information and the attribute association information of the user based on the key entity content and the relationship content;

[0284] Convert the query statement information into each query feature condition, and use all the query feature conditions as the graph query content;

[0285] Generate the attribute index information of each risk index recommendation type of the user through a risk index recommendation network based on the attribute association information of the user, and use all the attribute index information as the risk index recommendation information.

[0286] Optionally, the generating module 320 is specifically configured to:

[0287] Identify the node type and the relationship type based on each of the query feature conditions, and identify the feature graph query condition information based on the attribute index information of each risk index recommendation type;

[0288] Generate an initial feature map query scheme for the user based on the node type, the relationship type, and the feature map query condition information, and perform a visualization display process on the initial feature map query scheme through a visualization generation strategy to obtain a feature map query scheme.

[0289] Optionally, the adjustment module 330 is specifically configured to:

[0290] Identify each adjustment content in the adjustment information, and based on the adjustment range corresponding to each adjustment content, query the target adjustment amount corresponding to each adjustment content and the adjustment position information corresponding to each adjustment content in the feature map query scheme;

[0291] Adjust the feature map query scheme based on the target adjustment amount corresponding to each adjustment content and the adjustment position information corresponding to each adjustment content to obtain a target feature map scheme.

[0292] Each module in the above high-efficiency query device for enterprise information review can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0293] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as Figure 4As shown in the figure. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface, the display unit, and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements an efficient query method for enterprise information review. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0294] Those skilled in the art can understand that Figure 4 the structure shown in the figure is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0295] In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps corresponding to the efficient query method for enterprise information review are implemented.

[0296] In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, the steps corresponding to the efficient query method for enterprise information review are implemented.

[0297] In an embodiment, a computer program product is provided, including a computer program. When the computer program is executed by the processor, the steps corresponding to the efficient query method for enterprise information review are implemented.

[0298] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0299] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., and are not limited thereto. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., and are not limited thereto.

[0300] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0301] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. An efficient query method for enterprise information audit, characterized in that: The method comprises: Get the user's query information; Acquire enterprise entity information of each enterprise and enterprise relationship information of each enterprise, and for each enterprise, extract each enterprise node vector of the enterprise entity information of the enterprise and each enterprise relationship edge vector of the enterprise relationship information of the enterprise; Collecting enterprise similarity weight information of the enterprise and the minimum feature subgraph of each enterprise feature of the enterprise, and constructing similarity measurement information of the enterprise based on the enterprise similarity weight information of the enterprise; Based on the minimum feature subgraph of each enterprise feature, each embedded subgraph information of the enterprise is generated, and based on the minimum feature subgraph represented by each feature, each enterprise node vector, each enterprise relationship edge vector, and the similarity measurement information of the enterprise, the data domain knowledge graph of the multiple enterprises is constructed; The data domain knowledge graph includes: Node and edge attribute vectors: Attribute vector of enterprise node v: f(v)= [C v , IN v , H v , f struct (v)]∈ R D ; Where: C v : Registered capital; E v : Number of senior executives; H v : controlling stake; ф struct (v): structural feature vector obtained by graph embedding model; D is the dimension of embedding vector; Attribute vector of equity relationship edge e=(u,v): ψ(e)= [O u，v , V u，v ,ф struct (v)]∈ R D 100. Among them: u，v V represents the equity ratio of enterprise u to enterprise v; u，v represents the voting rights ratio of enterprise u to enterprise v; struct (v): structural feature vector obtained by graph embedding model; D is the dimension of embedding vector; Subgraph embedding function: f(G')=p({ф(v):v∈V'},{ψ(e):e∈E'})=GNN(∑ v∈V' ,ф(v)+∑ e∈E' ,ψ(e); Among them, f(G') is the traditional aggregation function, is a subgraph of the knowledge graph, p is an aggregation function that defines the attribute vector of the enterprise node and the attribute vector of the equity relationship edge, V' is the node set in the subgraph, including the set of all enterprise nodes v, and E' is the edge set in the subgraph, including the set of all equity relationship edges e; Similarity measure: Based on weighted cosine similarity: ； in, is the embedding vector, is the weight vector, D is the dimension of the embedding vector, u is the enterprise, represents the element-wise Hadamard product; Seed subgraph extension: ； in, is the weight matrix, is the embedding vector in the predefined seed subgraph, i∈(1,2,3…), D is the dimension of the embedding vector, is the combination coefficient; Based on the query information, identifying the query demand content of the user through a semantic recognition network; Generate graph query content and risk indicator suggestion information based on the query requirement content, and generate a feature graph query solution for the user based on the graph query content and the risk indicator suggestion information; Collecting adjustment information of the user on the feature graph query scheme, and adjusting the feature graph query scheme based on the adjustment information to obtain a target feature graph scheme; Based on the target feature graph solution, the target query result of the user is queried through the data domain knowledge graph.

2. The method according to claim 1, characterized in that: The identifying the query requirement content of the user through a semantic recognition network based on the query information includes: Identifying query text information corresponding to the query information, and identifying text semantic content in the query text information through a semantic recognition network; Through the semantic parsing strategy, the key entity content in the text semantic content and the relationship content in the text semantic content are identified, and the key entity content and the relationship content are used as the query demand content of the user.

3. The method according to claim 2, characterized in that The generating of graph query content and risk indicator suggestion information based on the query requirement content includes: Based on the key entity content and the relationship content, filter the user's query statement information and attribute association information; Convert the query statement information into query feature conditions, and use all query feature conditions as graph query content; Based on the attribute association information of the user, attribute indicator information of each risk indicator suggestion type of the user is generated through a risk indicator suggestion network, and all attribute indicator information is used as risk indicator suggestion information.

4. The method according to claim 3, characterized in that The generating the user's characteristic graph query solution based on the graph query content and the risk indicator suggestion information includes: Based on each of the query feature conditions, identifying the node type and the relationship type, and based on the attribute indicator information of each of the risk indicator suggestion types, identifying the feature graph query condition information; Based on the node type, the relationship type, and the feature graph query condition information, the user's initial feature graph query solution is generated, and the initial feature graph query solution is visualized and displayed through a visualization generation strategy to obtain a feature graph query solution.

5. The method according to claim 1, characterized in that The step of adjusting the feature graph query scheme based on the adjustment information to obtain a target feature graph scheme includes: Identify each adjustment content in the adjustment information, and based on the adjustment range corresponding to each adjustment content, query the target adjustment amount corresponding to each adjustment content and the adjustment position information corresponding to each adjustment content in the feature map query scheme; Based on the target adjustment amount corresponding to each of the adjustment contents and the adjustment position information corresponding to each of the adjustment contents, the feature map query scheme is adjusted to obtain a target feature map scheme.

6. An efficient query device for enterprise information audit, characterized in that: The device comprises: The acquisition module is used to obtain the query information of the user; obtain the enterprise entity information of each enterprise and the enterprise relationship information of each enterprise, and for each enterprise, extract each enterprise node vector of the enterprise entity information of the enterprise and each enterprise relationship edge vector of the enterprise relationship information of the enterprise; collect the enterprise similarity weight information of the enterprise and the minimum feature subgraph of each enterprise feature of the enterprise, and construct the similarity measurement information of the enterprise based on the enterprise similarity weight information of the enterprise; generate each embedded subgraph information of the enterprise based on the minimum feature subgraph of each enterprise feature, and construct the data domain knowledge graph of the multiple enterprises based on the minimum feature subgraph represented by each feature, each enterprise node vector, each enterprise relationship edge vector, and the similarity measurement information of the enterprise; the data domain knowledge graph includes: attribute vectors of nodes and edges: attribute vector of enterprise node v: ф(v)= [C v , E v , H v ,ф struct (v)]∈ R D ; Among them: C v : Registered capital; E v : Number of senior executives; H v : controlling stake; ф struct (v): Structural feature vector obtained by graph embedding model; D is the dimension of embedding vector; Attribute vector of equity relationship edge e=(u,v): ψ(e)=[O u，v , V u，v , ф struct (v)]∈ R D ; Among them: O u，v V represents the equity ratio of enterprise u to enterprise v; u，v represents the voting rights ratio of enterprise u to enterprise v; struct (v): structural feature vector obtained by graph embedding model; D is the dimension of embedding vector; subgraph embedding function: f(G')=p({ф(v):v∈V'},{ψ(e):e∈E'})=GNN(∑ v∈V' ,ф(v)+∑ e∈E' ,ψ(e)); where f(G') is the traditional aggregation function, is a subgraph of the knowledge graph, p is an aggregation function that defines the attribute vector of the enterprise node and the attribute vector of the equity relationship edge, V' is the node set in the subgraph, including the set of all enterprise nodes v, E' is the edge set in the subgraph, including the set of all equity relationship edges e; Similarity measure: based on weighted cosine similarity: ;in, is the embedding vector, is the weight vector, D is the dimension of the embedding vector, u is the enterprise, Represents the element-wise Hadamard product; Seed subgraph expansion: ;in, is the weight matrix, is the embedding vector in the predefined seed subgraph, i∈(1,2,3…), D is the dimension of the embedding vector, is the combination coefficient; Based on the query information, identifying the query demand content of the user through a semantic recognition network; A generation module, configured to generate graph query content and risk indicator suggestion information based on the query requirement content, and generate a characteristic graph query solution for the user based on the graph query content and the risk indicator suggestion information; An adjustment module, used for collecting the user's adjustment information on the feature graph query scheme, and adjusting the feature graph query scheme based on the adjustment information to obtain a target feature graph scheme; A query module is used to query the user's target query result based on the target feature graph scheme through the data domain knowledge graph.

7. The device according to claim 6, characterized in that The acquisition module is specifically used for: Identifying query text information corresponding to the query information, and identifying text semantic content in the query text information through a semantic recognition network; Through the semantic parsing strategy, the key entity content in the text semantic content and the relationship content in the text semantic content are identified, and the key entity content and the relationship content are used as the query demand content of the user.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images