Enterprise actual holder identification and risk conduction relation mining method and equipment

Abstract

The invention discloses an enterprise actual holder identification and risk conduction relation mining method and equipment. The method comprises the following steps: performing data acquisition from a data source, preprocessing the acquired data, and performing entity identification and relationship extraction according to the acquired data; according to the identified entity and the extracted relationship, constructing a relationship knowledge graph; deducing the relation knowledge graph based on a graph neural network model to mine a potential relation of an undominant record; analyzing a risk conduction path and quantifying conduction influence; and a visual interaction interface is provided, and result display and model optimization are realized. According to the method, an AI large model is taken as a core, knowledge graph construction and graph neural network reasoning technologies are combined, enterprise actual holder identification and risk conduction relation mining are completed through a multi-step process, diffusion paths and intensity changes of risks in a multi-level stockholder network can be simulated, key risk nodes and high-influence paths can be identified, and the risk prediction accuracy is improved. And risk conduction prediction and early warning capabilities are provided for a supervision mechanism.

Description

Technical Field

[0001] The present invention relates to the technical field of financial risk control, and particularly relates to a method and device for identifying the actual holders of an enterprise and mining the relationship of risk conduction. Background Art

[0002] With the advancement of economic globalization, the shareholder structure and relationship network of enterprises have become increasingly complex. In the fields of enterprise management and supervision, the transparency of shareholder relationships is one of the key issues. However, the types of relationships among enterprise shareholders are diverse, including both direct equity relationships and potential connections such as indirect investments and contract collaborations. This complex relationship network poses extremely high technical requirements for the identification of the actual holders of an enterprise and the analysis of risk conduction paths.

[0003] In recent years, the development of artificial intelligence (AI) technology has provided new solutions for this field. For example, knowledge graph technology has been widely used in the structured expression of information and relationship modeling; deep learning models have also demonstrated powerful capabilities in the mining of unstructured data. However, the existing technologies and systems still have difficulty in comprehensively solving the following problems:

[0004] 1. Shareholder relationship data is multi-source heterogeneous and dynamically changing;

[0005] 2. Potential relationships (such as implicit investments or agency relationships) are difficult to directly identify from traditional data;

[0006] 3. The shareholder network structure is complex, and traditional analysis methods are difficult to measure its multi-dimensional interaction effects.

[0007] Although the existing knowledge graph and AI technologies have made breakthroughs in single scenarios, when applied to the identification of the actual holders of an enterprise and risk conduction analysis, they still face the following challenges:

[0008] 1. The data sources are scattered and the formats are diverse

[0009] The shareholder information of an enterprise is distributed in multiple data sources, including enterprise registration information, public news reports, contract records, and financial statements, etc. These data have significant heterogeneity and lack of a unified format, resulting in low efficiency of traditional methods in data fusion and processing.

[0010] 2. Hidden relationships are difficult to discover

[0011] Existing technologies usually rely on structured enterprise data or explicit relationship networks and are difficult to mine potential relationships from unpublicized information (such as implicit agency or informal contracts), resulting in blind spots in risk analysis.

[0012] 3. Limitations in relationship network analysis

[0013] The shareholder relationship network has multi-dimensional attributes (such as shareholding ratio, time dimension, industry distribution, etc.) and dynamic change characteristics. Traditional relationship diagrams or statistical analysis tools lack the ability to model and analyze multi-dimensional dynamic networks and cannot reveal the risk conduction mechanism in complex relationships.

[0014] 4. Insufficient technical interpretability

[0015] Although deep learning and graph neural network technologies perform excellently in relationship inference, their black-box attributes make the output results lack transparency, and it is difficult for regulatory agencies or enterprise managers to directly understand and trust the model results.

[0016] 5. Insufficient real-time performance and scalability

[0017] In regulatory and enterprise governance scenarios, the analysis system needs to update the shareholder network in real time and adapt to scale expansion. However, existing technologies are less efficient in processing large-scale data and it is difficult to achieve dynamic updates and real-time analysis. Summary of the Invention

[0018] The object of the present invention is to provide a method and device for identifying actual holders of an enterprise and mining risk conduction relationships in view of the deficiencies of the existing technology.

[0019] To achieve the above object, in the first aspect, the present invention provides a method for identifying actual holders of an enterprise and mining risk conduction relationships, including:

[0020] Step 1: Collect data from a data source, preprocess the collected data, and perform entity recognition and relationship extraction based on the collected data;

[0021] Step 2: Construct a relationship knowledge graph based on the identified entities and extracted relationships;

[0022] Step 3: Infer the relationship knowledge graph based on a graph neural network model to mine potential relationships that are not explicitly recorded;

[0023] Step 4: Analyze the risk conduction path and quantify the conduction impact;

[0024] Step 5: Provide an intuitive interaction interface to realize result display and model optimization.

[0025] Further, the method of entity recognition is specifically as follows:

[0026] P(e|x) = softmax(W h ·h e )

[0027] where P(e|x) represents the probability of entity e in text x, softmax(.) represents the normalized exponential function, and h eis the hidden representation of the entity, W h is the mapping matrix.

[0028] Furthermore, the method of relationship extraction is as follows:

[0029]

[0030] Among them, R(X) represents the relationship between entity e i and entity e j The argmax(.) represents the maximum value parameter, and r is the category or type of the candidate relationship.

[0031] Furthermore, step 2 specifically includes:

[0032] 1. Definition of nodes and edges:

[0033] Nodes: enterprises, shareholders, legal persons, investment institutions;

[0034] Edges: direct equity relationship, agency relationship, investment relationship, contract association;

[0035] 2. Process of constructing the relationship knowledge graph:

[0036] Entity alignment: Eliminate entity ambiguities in the data to ensure the merger of the same entity in different data sources;

[0037] Attribute completion: Fill in the missing relationship attributes through inference or completion algorithms;

[0038] 3. Storage and query:

[0039] Use a graph database to store the knowledge graph;

[0040] Provide a graph query language interface for subsequent modules to call.

[0041] Furthermore, the update method of the graph neural network model in step 3 is as follows:

[0042]

[0043] Among them, is the embedding representation of node v at the t+1 layer, σ is the activation function, W (t) is the weight matrix in the graph neural network, h u (t) is the embedding representation of node u at the t layer, u is the neighbor node of node v, and N(v) represents the set of neighbor nodes of node v;

[0044] Combine prior knowledge to evaluate the confidence of the inference result, specifically as follows:

[0045]

[0046] Among them, represents the relationship confidence between node i and node j, is the embedding representation of node i, and h j is the feature of node j, and M is the relationship feature matrix.

[0047] Furthermore, step 4 specifically includes:

[0048] Using a diffusion model to simulate the spread of risk from one node to surrounding nodes, and the diffusion model is expressed as:

[0049] S(t + 1) = S(t) - β * S(t) * I(t)

[0050] where S(t + 1) represents the proportion of healthy nodes at time t + 1, S(t) represents the proportion of healthy nodes at time t, β is the transmission coefficient, and I(t) is the core variable of transmission;

[0051] Perform centrality analysis on key nodes, specifically as follows:

[0052]

[0053] where C(v) is the centrality of node v, d(v, u) is the shortest path distance between node v and node u, and G is the entire network used to calculate centrality.

[0054] Furthermore, in step 5, D3.js or ECharts is used to implement the dynamic display of the relationship network, and provide interactive filtering and risk path highlighting functions.

[0055] Furthermore, the data source includes enterprise registration information, contract texts, news and public information, and financial statements.

[0056] Furthermore, the preprocessing includes removing redundant information and unifying the data format.

[0057] In a second aspect, the present invention provides an enterprise actual holder identification and risk conduction relationship mining device, including a storage medium and a processor, where the storage medium stores a computer program, and when the computer program is executed by the processor, it is used to implement the above method.

[0058] Beneficial effects: 1. The present invention innovatively applies the AI large model to the identification of actual enterprise holders and the mining of complex shareholder relationships. By deeply integrating multi-source heterogeneous data and conditional generation technology, it identifies the actual controllers hidden in the multi-layer relationship network. The large model dynamically combines industrial and commercial data, contract information, news texts, financial data, etc., generates a multi-dimensional dynamic knowledge graph of enterprise shareholder relationships, and conducts potential relationship inference through graph neural network (GNN) technology. Compared with the traditional static shareholder structure analysis method, the present invention breaks through the limitations of single data and the inability to handle complex dynamic relationships, providing a comprehensive and dynamic solution for shareholder relationship analysis;

[0059] 2. The present invention proposes a dynamic risk conduction analysis method for enterprise shareholder networks. By combining a knowledge graph and a risk propagation model (such as the SIR model), it simulates the diffusion path and intensity change of risks in a multi-level shareholder network; systematically quantifies the process of risk propagation from high-risk nodes (such as holding groups) to other enterprises (such as subsidiaries or affiliated enterprises), and identifies key risk nodes and high-impact paths; in addition, the invention dynamically captures risk exposure points in the enterprise network through time series analysis, providing the regulatory agency with the ability of risk conduction prediction and early warning;

[0060] The present invention constructs a full-link system from enterprise relationship data extraction, knowledge graph generation to risk analysis and interpretation, and provides intuitive visualization support; the system displays the multi-level shareholder relationships and risk conduction paths of enterprises through a dynamic interactive graph, and at the same time combines the attention mechanism to explain the influence weights of key relationships and paths, improving the transparency and understandability of the analysis results. This closed-loop design provides full-process support for enterprise risk management and regulatory decision-making. Brief description of the drawings

[0061] Figure 1 It is a schematic flow chart of a method for identifying actual enterprise holders and mining risk conduction relationships. Detailed implementation manners

[0062] The following further clarifies the present invention in conjunction with the drawings and specific embodiments. These embodiments are implemented on the premise of the technical solution of the present invention. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention.

[0063] As Figure 1 shown, the embodiment of the present invention provides a method for identifying actual enterprise holders and mining risk conduction relationships, including:

[0064] Step 1: Collect data from data sources, preprocess the collected data, and perform entity recognition and relationship extraction based on the collected data. The above data sources include enterprise registration information (such as equity structure, legal representative), contract texts (such as investment agreements, related party transactions), news and public information (such as enterprise cooperation, financial analysis), and financial statements (such as cash flow, balance sheet). The above preprocessing includes removing redundant information and unifying data formats. The method of entity recognition is to use named entity recognition (NER) technology to extract information such as enterprise names, shareholder names, and equity ratios. The method of relationship extraction is to extract relationship types and attributes (such as "direct shareholding", "agency relationship") from the text based on a large AI model (such as Wenxin Yiyan).

[0065] Specifically, the method of the above entity recognition is as follows:

[0066] P(e|x) = softmax(W h ·h e )

[0067] Among them, P(e|x) represents the probability of entity e in text x, softmax(.) represents the normalized exponential function, h e is the hidden representation of the entity, and W h is the mapping matrix.

[0068] The method of the above relationship extraction is as follows:

[0069]

[0070] Among them, R(X) represents the relationship between entity e i and entity e j , argmax(.) represents the maximum value parameter, and r is the category or type of the candidate relationship.

[0071] Step 2: Construct a relationship knowledge graph based on the identified entities and extracted relationships. Furthermore, convert the structured data into the form of a knowledge graph to display the relationship network among shareholders. The specific method is as follows:

[0072] 1. Definition of nodes and edges:

[0073] Nodes: enterprises, shareholders, legal persons, investment institutions;

[0074] Edges: direct equity relationship, agency relationship, investment relationship, contract association;

[0075] 2. Process of constructing the relationship knowledge graph:

[0076] Entity alignment: Eliminate entity ambiguities in the data to ensure the merger of the same entity in different data sources;

[0077] Attribute completion: fill in missing relation attributes through inference or completion algorithms;

[0078] 3. Storage and query:

[0079] Use graph databases (such as Neo4j) to store knowledge graphs;

[0080] Provides a graph query language (Cypher) interface for subsequent modules to call.

[0081] Step 3: Infer the relationship knowledge graph based on the graph neural network model to mine potential relationships that are not explicitly recorded. The specific method is as follows:

[0082] Graph Neural Network (GNN): Uses embedded representations of nodes and edges to infer potential relationships;

[0083] Bayesian inference: Combine prior knowledge to assess the confidence of inference results.

[0084] The graph neural network model is updated as follows:

[0085]

[0086] in, is the embedding representation of node v at layer t+1, σ is the activation function used to introduce nonlinearity so that the model can capture more complex relationships and patterns, and W (t) is the weight matrix in the graph neural network, which is used to linearly transform the node embedding representation, h u (t) is the embedding representation of node u at the tth layer, u is the neighbor node of node v, and N(v) represents the set of neighbor nodes of node v.

[0087] Combined with prior knowledge, the confidence of the inference results output by the graph neural network is evaluated as follows:

[0088]

[0089] in, represents the confidence of the relationship between node i and node j, is the embedding representation (transposed vector) of node i, h j is the feature of node j, To capture the feature h j is the embedding representation of node j, and M is the relationship feature matrix. If the relationship confidence is above the set threshold, the inference result output by the graph neural network is considered reliable.

[0090] Step 4: Analyze the risk transmission path and quantify the transmission impact. The specific method is as follows:

[0091] Propagation model: The diffusion model (such as the SIR model) is used to simulate the spread of risks from one node to surrounding nodes. The above diffusion model is expressed as:

[0092] S(t + 1) = S(t) - β * S(t) * I(t)

[0093] Among them, S(t + 1) represents the proportion of healthy nodes at time t + 1, S(t) represents the proportion of healthy nodes at time t, β is the propagation coefficient, and I(t) is the core variable of propagation, which determines the propagation intensity of currently affected nodes to healthy nodes.

[0094] Influence analysis: Use PageRank or centrality metrics to calculate key nodes. Taking centrality analysis as an example, the centrality analysis of key nodes is as follows:

[0095]

[0096] Among them, C(v) is the centrality of node v, d(v, u) is the shortest path distance between node v and node u, and G is the entire network used to calculate centrality, which is a network structure composed of nodes and edges.

[0097] Step 5: Provide an intuitive interactive interface to achieve result display and model optimization. For example, use D3.js or ECharts to achieve the dynamic display of the relationship network, and provide interactive filtering and risk path highlighting functions. Model optimization is achieved by collecting user feedback on the system inference results and using reinforcement learning to optimize the model.

[0098] Based on the above embodiments, those skilled in the art can easily understand that the present invention also provides an enterprise actual holder identification and risk conduction relationship mining device, including a storage medium and a processor. The storage medium stores a computer program, and when the computer program is executed by the processor, it is used to implement the above method.

[0099] Illustrative example:

[0100] Case 1: Identify the actual holder of the enterprise

[0101] 1. User background

[0102] A certain financial regulatory agency needs to analyze the actual holder structure of "X Holdings Co., Ltd.". The existing public information shows that its shareholder hierarchy is complex, involving multiple levels of corporate shareholders and potential shareholding relationships.

[0103] Through multi-source data integration, knowledge graph construction, graph neural network reasoning, and finally visual display, the identification and analysis of the actual holders of the enterprise are completed.

[0104] 2. Detailed steps

[0105] Step 1: Data Collection and Preprocessing

[0106] 1. Data Sources:

[0107] Enterprise business registration information (shareholder names, shareholding ratios).

[0108] Contract texts (investment agreements, nominee agreements).

[0109] News reports (clues of hidden relationships).

[0110] Financial statements (fund flows of related enterprises).

[0111] 2. Specific Operations:

[0112] Use natural language processing (NLP) technology to perform entity recognition (Named Entity Recognition, NER) on unstructured data (news, contracts), and extract key information such as shareholder names, equity ratios, and relationship types.

[0113] Data cleaning: Unify shareholder names, remove redundant information, and ensure entity alignment.

[0114] Output Results: After data collation, the following structured information is obtained:

[0115] Shareholders of X Holding Co., Ltd.:

[0116] Company Y holds 40%;

[0117] Company Z holds 60%.

[0118] Contract analysis shows that the shares held by Company Y are held in trust by "Zhang".

[0119] News analysis finds that there are potential related-party transactions between Company Z and Company Y.

[0120] Step 2: Construct a Knowledge Graph

[0121] Convert the above structured data into a knowledge graph, representing a multi-dimensional network of enterprise shareholder relationships.

[0122] 1. Definitions of Nodes and Edges:

[0123] Nodes: X Holding, Company Y, Company Z, Zhang.

[0124] Edges: Direct shareholding (shareholding ratio), nominee relationship, contract association.

[0125] 2. Graph Example:

[0126]

[0127] 3. Storage Method:

[0128] Store using a graph database (such as Neo4j).

[0129] Provide a query interface to quickly retrieve key relationships through Cypher statements.

[0130] Step 3: Inference of potential relationships

[0131] Use a graph neural network (GNN) model to infer the relationship graph and determine Zhang's final control ratio. Specifically as follows:

[0132] 1. Graph neural network inference formula:

[0133]

[0134] 2. Propagation model:

[0135] S(t + 1) = S(t) - β * S(t) * I(t) 3. Calculation process:

[0136] Initial conditions:

[0137] Zhang's initial control right: Initial state of Company Y:

[0138] Uncontrolled ratio: Controlled ratio: Initial state of X Holdings:

[0139] Uncontrolled ratio: Controlled ratio: Propagation parameter:

[0140] Zhang controls Company Y through a nominee relationship: β = 0.4.

[0141] Company Y affects X Holdings through direct shareholding: β = 0.4.

[0142] 4. Iterative calculation process:

[0143] First round of propagation (t = 1)

[0144] (1) Zhang → Company Y:

[0145] Zhang's control right propagates to Company Y:

[0146]

[0147] Remaining uncontrolled ratio of Company Y:

[0148]

[0149] (2) Company Y → X Holdings:

[0150] The control right of Company Y spreads to X Holdings:

[0151]

[0152] The remaining uncontrolled proportion of X Holdings:

[0153]

[0154] The second round of spread (t = 2)

[0155] (1) Zhang → Company Y:

[0156] The control right of Zhang spreads to Company Y:

[0157] The control right of Zhang has been fully spread to Company Y, and there is no new spread:

[0158] (2) Company Y → X Holdings:

[0159] Company Y continues to spread its remaining control right to X Holdings::

[0160] The remaining uncontrolled proportion of X Holdings:

[0161]

[0162] The third round of spread (t = 3)

[0163] (1) Zhang → Company Y:

[0164] The control right of Zhang spreads to Company Y:

[0165] The control right of Zhang has been fully spread to Company Y, and there is no new spread:

[0166] (2) Company Y → X Holdings:

[0167] Company Y continues to spread its remaining control right to X Holdings::

[0168] The remaining uncontrolled proportion of X Holdings:

[0169]

[0170] The fourth round of spread (t = 4)

[0171] (1) Zhang → Company Y:

[0172] The control right of Zhang spreads to Company Y:

[0173] Zhang's control rights have been fully transmitted to Company Y, with no new transmission:

[0174]

[0175] (2) Company Y → X Holdings:

[0176] Company Y continues to transmit its remaining control rights to X Holdings:

[0177]

[0178] The remaining uncontrolled proportion of X Holdings:

[0179]

[0180] The 5th round of transmission (t = 5)

[0181] (1) Zhang → Company Y:

[0182] Zhang's control rights are transmitted to Company Y:

[0183] Zhang's control rights have been fully transmitted to Company Y, with no new transmission:

[0184]

[0185] (2) Company Y → X Holdings:

[0186] Company Y continues to transmit its remaining control rights to X Holdings::

[0187]

[0188] (3) Convergence determination:

[0189]

[0190] The control ratio reaches stability, and the convergence value is: I X控股 = 0.891136

[0191] 3. Final result

[0192] Zhang's final control ratio is to indirectly control 40% of the equity of X Holdings through Company Y, and the control ratio of the final comprehensive influence is 0.891 (89.1%). There may be implicit related-party transactions between Company Y and Company Z.

[0193] 4. Analysis

[0194] Zhang's great influence first stems from his absolute control over Company Y through a nominee shareholding relationship. Although Zhang does not directly hold shares in X Holdings, through the nominee agreement with Company Y, he has complete control over the 40% equity held by Company Y in X Holdings. This nominee relationship gives Zhang absolute decision-making and control power over the equity portion of X Holdings, which is the fundamental source of his control and the key breakthrough point for identifying the actual controller.

[0195] In the corporate relationship network, Zhang's influence is not limited to the direct effect of the nominee relationship, but is gradually amplified through multi-level paths in the network. Company Y's shareholding relationship with X Holdings further transmits Zhang's control power to X Holdings. In addition, there may be implicit relationships between Company Z, another major shareholder of X Holdings (holding 60% of the shares), and Company Y, such as capital cooperation, business collaboration, or contractual agreements. The potential existence of these relationships provides a path for Zhang's indirect control and further expands his sphere of influence.

[0196] The multi-level and complex nature of this control path makes Zhang a key node in the corporate network. In the corporate network, Zhang directly controls Company Y and indirectly influences Company Z, thus having significant power over the governance and decision-making of X Holdings. This power not only stems from the explicit shareholding relationship but is also transmitted through implicit associated relationships. The inferences of knowledge graphs and graph neural networks further reveal this complex hierarchical amplification effect.

[0197] The final inference result shows that Zhang's comprehensive control ratio over X Holdings reaches 89.1%. This ratio reflects Zhang's absolute say in the decision-making of X Holdings through nominee shareholding, related-party transactions, and implicit network relationships. This control ratio is significantly higher than the direct shareholding ratio, indicating that there may be a huge difference between the actual controller of the enterprise and the equity distribution. Zhang's control power has penetrated multiple levels of X Holdings and he has become the core figure in corporate decision-making.

[0198] This implicit control relationship has important practical implications for regulatory authorities. Although Zhang does not appear in the shareholder registration information, his actual control power achieved through nominee agreements and network paths may have a profound impact on the corporate governance structure and financial stability. Regulatory authorities must use more in-depth data analysis and relationship mining tools to effectively identify this hidden actual controller and its potential risks.

[0199] From the perspective of enterprise risk assessment, Zhang's control power means that if he has legal disputes or financial problems, X Holdings and its associated enterprises may face major operational and decision-making risks. This risk is transmitted through network paths and may further affect the financial condition and business stability of associated enterprises. Therefore, clearly identifying this implicit control relationship is of great value for improving corporate governance and strengthening risk management.

[0200] Case 2: Risk Transmission Path Analysis

[0201] 1. User Background

[0202] A manufacturing enterprise, "P Company", has recently fallen into a financial crisis. Its major shareholder, "Q Group", faces bankruptcy risks due to high debt and poor management. The regulatory agency hopes to analyze whether the risks of "Q Group" will be transmitted to "P Company" and its downstream affiliated enterprises (such as "A Company"), and quantify the intensity and scope of risk transmission.

[0203] 2. Detailed Steps

[0204] Step 1: Data Collection and Modeling

[0205] Data Sources:

[0206] Business data: Q Group directly holds 50% of the equity of P Company.

[0207] Financial data: There is a relatively large proportion of accounts receivable (30%) between P Company and A Company.

[0208] News information: Q Group has recently faced bankruptcy risks due to high debt and has attracted market attention.

[0209] Data Processing:

[0210] Use NLP technology to extract key entities and relationships from unstructured data (news, contracts), such as "Q Group - High Debt", "P Company - Accounts Receivable - A Company".

[0211] After data structuring, the following main relationships are formed:

[0212] Q Group → Shareholding → P Company (50%).

[0213] P Company → Accounts Receivable → A Company (30%).

[0214] Modeling:

[0215] Build a knowledge graph with nodes including Q Group, P Company, and A Company.

[0216] Edges include direct shareholding relationships, accounts receivable relationships, and potential risk relationships.

[0217] Step 2: Risk Propagation Model Design

[0218] (1) Model Selection:

[0219] Adopt the SIR risk propagation model, where:

[0220] S(t): The proportion of nodes not affected by risks.

[0221] I(t): The proportion of nodes affected by risks.

[0222] β: The risk propagation coefficient, assumed to be 0.2.

[0223] (2) Initial conditions

[0224]

[0225]

[0226] Step 3: Iterative propagation calculation

[0227] The first round of propagation (t = 1)

[0228] (1) The risk of Group Q propagates to Company P:

[0229]

[0230] The controlled proportion of Company P is 20%.

[0231]

[0232] (2) Company P propagates risks to Company A:

[0233]

[0234] The controlled proportion of Company A is 4%.

[0235]

[0236] The second round of propagation (t = 2)

[0237] (1) The risk of Group Q propagates to Company P:

[0238]

[0239] (2) Company P continues to propagate risks to Company A:

[0240]

[0241] The third round of propagation (t = 3), and so on...

[0242] Continue to propagate risks according to the above formula until the change in each round of propagation is less than the set threshold:

[0243] |I(t + 1) - I(t)| < 0.001

[0244] Final convergence value:

[0245] The controlled proportion of Company P: I P公司= 0.5 (The final conduction ratio of Q Group's risk in Company P is 50%).

[0246] Controlled ratio of Company A: I A公司 = 0.12 (The final impact ratio of Company A affected by Q Group's risk is 12%).

[0247] Step 4: Visualization and Analysis

[0248] (1) Visualization of risk path:

[0249] The dynamic knowledge graph shows the risk propagation chain of Q Group → Company P → Company A.

[0250] Highlight the risk intensity and mark the key nodes (such as the controlled ratio of Company P is 50%, which is the core node of risk conduction).

[0251] (2) Analysis of key nodes:

[0252] Q Group is the risk source, and its high debt problem is directly transmitted to Company P.

[0253] As a bridge node, Company P further spreads the risk to downstream enterprises (such as Company A).

[0254] 3. Analysis

[0255] The risk of Q Group firstly comes from its own high debt and management crisis. As the major shareholder of Company P, Q Group directly holds 50% of the shares of Company P. This direct shareholding relationship enables the financial problems of Q Group to be directly transmitted to Company P and have a significant impact on its financial stability and business decisions. Since Q Group is the core shareholder of Company P, its bankruptcy risk becomes the main source of Company P's operation risk and is also an important basis for analyzing the risk conduction of enterprises.

[0256] In the enterprise relationship network, the risk conduction of Q Group not only occurs through the direct shareholding relationship but also further spreads through the accounts receivable relationship between Company P and its downstream enterprise Company A. Company P relies on Company A to pay the accounts receivable to maintain its cash flow, and this capital chain relationship enables the risk of Q Group to indirectly affect Company A through Company P. Through the analysis of the knowledge graph and risk propagation model, the financial problems of Q Group are gradually amplified in a multi-level path, and the impact on the enterprise network is not limited to the shareholder relationship but also involves the cross-enterprise capital chain and business cooperation relationship.

[0257] The multi-level and complex nature of this risk path makes Q Group a key risk node in the corporate network. As the source of risk, the bankruptcy crisis of Q Group significantly affects Company P through direct equity relationships (the ultimate controlled ratio reaches 50%), and further indirectly affects Company A through accounts receivable relationships (the ultimate controlled ratio is 12%). This multi-level risk conduction effect is quantified through knowledge graphs and propagation models, revealing the path and intensity of risk diffusion, providing an important basis for risk monitoring and assessment in the corporate network.

[0258] The final risk assessment shows that the crisis of Q Group has a profound impact on Company P and Company A, and may further spread to the associated corporate network. This multi-level risk conduction model reflects the complexity of the relationships between enterprises, and also indicates that risk management cannot only focus on a single enterprise, but needs to be analyzed from the dynamic perspective of the entire relationship network. Through the visual display of knowledge graphs, regulatory agencies can more clearly understand the risk propagation path and identify key node enterprises, so as to take more targeted regulatory measures.

[0259] This risk conduction analysis has important practical significance for regulatory agencies and corporate governance. For regulatory agencies, the high debt problem of Q Group is not only its own crisis, but also a potential risk source for the entire corporate network, and a systematic analysis of its scope of influence and conduction path is required. For corporate managers, Company P and Company A need to strengthen the risk monitoring of Q Group and take countermeasures to reduce its impact on operations and finance.

[0260] From the perspective of risk management, the risk conduction phenomenon in this complex network reveals the importance of relationships between enterprises. Identifying and quantifying risk conduction paths not only helps to timely discover potential risk nodes, but also provides strong support for building a more stable corporate network and improving corporate governance. This analysis model can be widely applied to corporate risk assessment, financial regulation and industry monitoring, and has important practical value.

[0261] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, the other parts not specifically described belong to the prior art or common general knowledge. Without departing from the principle of the present invention, several improvements and refinements can still be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for identifying the actual owner of an enterprise and mining the risk transmission relationship, characterized in that: include: Step 1: Collect data from the data source, pre-process the collected data, and perform entity recognition and relationship extraction based on the collected data; Step 2: Construct a relational knowledge graph based on the identified entities and extracted relationships; Step 3: Inferring the relational knowledge graph based on the graph neural network model to mine potential relationships that are not explicitly recorded; Step 4: Analyze the risk transmission path and quantify the transmission impact; Step 5: Provide an intuitive interactive interface to display results and optimize the model.

2. According to claim 1, a method for identifying actual owners of an enterprise and mining risk transmission relationships is characterized in that: The entity identification method is as follows: P(e|x)=softmax(W h ·h e ) Among them, P(e|x) represents the probability of entity e in text x, softmax(.) represents the normalized exponential function, and h e is the hidden representation of the entity, W h is the mapping matrix.

3. According to claim 2, a method for identifying actual owners of an enterprise and mining risk transmission relationships is characterized in that: The relationship extraction method is as follows: Among them, R(X) represents entity e i With entity e j The relationship between , argmax(.) means taking the maximum parameter, and r is the category or type of the candidate relationship.

4. According to claim 1, a method for identifying actual owners of an enterprise and mining risk transmission relationships is characterized in that: The step 2 specifically includes:

1. Definition of nodes and edges: Nodes: enterprises, shareholders, legal persons, investment institutions; Side: direct equity relationship, proxy holding relationship, investment relationship, contractual relationship; 2. Relational knowledge graph construction process: Entity alignment: eliminate entity ambiguity in the data and ensure that the same entity in different data sources is merged; Attribute completion: fill in missing relation attributes through inference or completion algorithms; 3. Storage and query: Use graph database to store knowledge graph; Provides a graph query language interface for subsequent modules to call.

5. According to claim 1, a method for identifying actual owners of an enterprise and mining risk transmission relationships is characterized in that: The updating method of the graph neural network model in step 3 is as follows: in, is the embedding representation of node v at layer t+1, σ is the activation function, W (t) is the weight matrix in the graph neural network, h u (t) is the embedding representation of node u at layer t, u is the neighbor node of node v, and N(v) represents the set of neighbor nodes of node v; Combined with prior knowledge, the confidence of the inference results is evaluated as follows: in, represents the confidence of the relationship between node i and node j, is the embedding representation of node i, h j is the feature of node j, and M is the relationship feature matrix.

6. According to claim 1, a method for identifying actual owners of an enterprise and mining risk transmission relationships is characterized in that: The step 4 specifically includes: A diffusion model is used to simulate the diffusion of risk from one node to surrounding nodes. The diffusion model is expressed as: S(t+1)=S(t)-β*S(t)*I(t) Among them, S(t+1) represents the proportion of healthy nodes at time t+1, S(t) represents the proportion of healthy nodes at time t+1, β is the propagation coefficient, and I(t) is the core variable of propagation; The centrality analysis of key nodes is as follows: Among them, C(v) is the centrality of node v, d(v,u) is the shortest path distance between node v and node u, and G is the entire network used to calculate the centrality.

7. According to claim 1, a method for identifying actual owners of an enterprise and mining risk transmission relationships is characterized in that: In step 5, D3.js or ECharts is used to realize the dynamic display of the relationship network, and to provide interactive filtering and risk path highlighting functions.

8. According to claim 1, a method for identifying actual owners of an enterprise and mining risk transmission relationships is characterized in that: The data sources include enterprise registration information, contract texts, news and public information, and financial statements.

9. The method for identifying the actual owner of an enterprise and mining the risk transmission relationship according to claim 1, characterized in that: The preprocessing includes removing redundant information and unifying data formats.

10. A device for identifying actual owners of an enterprise and mining risk transmission relationships, comprising a storage medium and a processor, wherein the storage medium stores a computer program, characterized in that: When the computer program is executed by a processor, it is used to implement the method according to any one of claims 1 to 9.

Images