A method and device for deep mining of suspected actual controllers of an enterprise

Abstract

The present application discloses a method for deep mining of suspected actual controllers of an enterprise. Extract information from the business data of the enterprise to construct a graph. For each node of the enterprise type, the actual controller of the node is initialized to the node itself, and a message is sent from the node to the pointed node along the first type of directed edge connected to the node, the second type of directed edge, and the third type of directed edge. Follow the principle of uniqueness to calculate the actual controller of each current node calculated by this superstep, and the control path from the node of the actual controller of each current node calculated by this superstep to each current node. Repeat the iterative process until all nodes participating in the iterative calculation are inactive or the calculation ends when the maximum number of iterations N is reached. The actual controller data mined by this application is more accurate and complete.

Description

Technical Field

[0001] The present application relates to a graph computing method, and more particularly to a method for mining suspected actual controllers of an enterprise using graph computing. Background Art

[0002] The actual controller of an enterprise (referred to as the actual controller) refers to a person who has actual control over the enterprise and can actually control the behavior of the enterprise. The person here includes natural persons, legal persons or other organizations. The actual controller of an enterprise usually includes: controlling shareholders formed by direct holding, indirect holding, proxy holding, etc.; people who have a significant influence or dominant role in the operation and management of the enterprise, the resolution of the enterprise or the selection of senior management personnel; people who have control over the financial behavior of the enterprise, etc. According to the requirements of the regulatory authorities, the actual controller of a listed company will be announced when the information is disclosed, but there is no way to query the actual controller of other types of enterprises.

[0003] The Chinese invention patent application "A method for calculating the actual controller and control path of an enterprise supporting distributed computing" with application publication number CN110991803A and application publication date April 10, 2020 discloses a method for calculating the actual controller and control path of an enterprise. The method uses an iterative idea based on equity relations to calculate the actual controller of the enterprise. This method has the following two defects. First, this method only considers equity relationship factors to mine the controller of the enterprise. In special scenarios, such as when two people hold 50% of the equity of a certain enterprise at the same time, it is impossible to determine whether to continue iterative calculation. Second, the condition for the method to determine the end of iterative update is that the controller and the status of the controller of the enterprise no longer change, without considering the change of the control path. After a certain iterative calculation, the controller of the enterprise may remain unchanged, but the control path of the controller of the enterprise changes. At this time, the method ends the iterative calculation, which may result in the loss of the control path. Summary of the invention

[0004] The technical problem to be solved by this application is to deeply explore the suspected actual controller of an enterprise based on the investment relationship, partnership agreement or control relationship of the enterprise, and ensure that the suspected actual controller of the enterprise can ultimately be traced back to a natural person, state-owned assets management department or other units or organizations.

[0005] In order to solve the above technical problems, this application proposes a method for deep mining of suspected actual controllers of an enterprise, including the following steps. Step S10: Extract information from the business data of the enterprise to construct a graph; the constructed graph includes nodes and edges; all organizations or individuals involved in the extracted information are used as nodes in the graph; from a certain enterprise node to the node of the branch of the enterprise, or from the node of the executive partner of a partnership to the node of the partnership, a first type of directed edge is constructed, and the first type of directed edge contains the attributes of the branch relationship or the executive partner relationship; from the node of the direct shareholder of a certain enterprise to the node of the enterprise, a second type of directed edge is constructed, and the second type of directed edge contains the attributes of the direct shareholding ratio; from a certain enterprise node to the node of the senior executives of the enterprise, a third type of directed edge is constructed, and the third type of directed edge contains the attributes of the position relationship. Step S30: For each enterprise-type node, the actual controller of the node is initialized to be the node itself, and a message is sent from the node to the pointed node along the first type of directed edge, the second type of directed edge, and the third type of directed edge connected to the node; this step is called the 0th superstep; the message includes the actual controller of the node calculated by this superstep, and the control path from the node of the actual controller calculated by this superstep to the node. Step S40: Each current node reads the received message, thereby learning the actual controller of all upstream nodes calculated by the previous superstep, and the control path of all upstream nodes from the node of the actual controller of the previous superstep calculation to the upstream node, and combining the types of edges pointing to the current node from all upstream nodes, following the uniqueness principle, calculate the actual controller of each current node calculated by this superstep, and the control path of each current node from the node of the actual controller of this superstep calculation to each current node; if the actual controller of the current node calculated by this superstep changes, or the control path of the current node from the node of the actual controller of this superstep calculation to the current node changes, then mark the current node as active, otherwise mark the current node as inactive; each active node sends the message to the pointed node along the direction of the first type of directed edge, the second type of directed edge, and the third type of directed edge connected to the node; this step is called the 1st-Nth superstep; inactive nodes no longer send messages downstream. Repeat step S40 to start the next superstep iteration process until all nodes participating in the iterative calculation are in an inactive state or the calculation is terminated when the maximum number of iterations N is reached; at this time, each node participating in the iterative calculation is used by the actual controller of this superstep calculation as the suspected actual controller of the node, and the control path from the node of the actual controller of this superstep calculation to the node is used as the control path from the node of the suspected actual controller of the node to the node; the node participating in the iterative calculation refers to a node that sends at least one message or receives at least one message during the entire calculation process.The above method has two conditions for determining whether to end the iterative calculation: (1) the actual controller of all nodes participating in the iterative calculation does not change; (2) the control path from the node of the actual controller of the current superstep calculation to the node of all nodes participating in the iterative calculation does not change; this makes the mined control path more comprehensive.

[0006] Preferably, in step S10, the extracted information includes: the name of the enterprise, the registered capital of the enterprise, the type of enterprise, the industry to which the enterprise belongs, the direct shareholders of the enterprise, the names of the senior executives of the enterprise, the branches of the enterprise, and the executive partners of the partnership.

[0007] Preferably, in step S10, data cleaning is performed before constructing the graph, including any one or more of equity ratio legitimacy verification, senior management data legitimacy cleaning, data consistency checking, and filling in missing data, and then the graph is constructed based on the cleaned data.

[0008] Preferably, in step S10, after constructing the graph, isolated nodes are excluded, where the isolated nodes refer to single nodes that are not connected to any edges.

[0009] Optionally, after step S10 and before step S30, step S20 is further included. Step S20: Evenly distribute the nodes and the edges connected to them to each server partition; the edges starting from a node A and pointing to other nodes are allocated to the server partition allocated to the node A.

[0010] Furthermore, in step S40, the calculation rules are as follows: (1) if there is a first-class directed edge pointing from an upstream node of the current node to the current node, then the actual controller of the current node calculated by this superstep is directly determined to be the actual controller of the upstream node; (2) if there is no first-class directed edge among all the edges pointing to the current node, then the actual controller of the current node calculated by this superstep is determined according to the absolute controlling shareholder and the largest sole shareholder principle through the direct shareholding ratio attribute of the second-class directed edge; (3) if there are multiple shareholders with the same shareholding ratio at the current node, then the actual controller of the current node is determined in combination with the position relationship of the third-class directed edge; (4) the control path of all upstream nodes from the node of the actual controller of the previous superstep calculation to the upstream node is known, the actual controller of the current node calculated by this superstep is known, and the path of all upstream nodes to the current node is known, and the three are combined to obtain the control path of the current node from the node of the actual controller of this superstep calculation to the current node. The above calculation rules are based on the company's branch relationships, the company's executive affairs partner relationships, the company's direct shareholding data, and the company's senior management positions to identify the company's actual controllers. The factors considered are more comprehensive and can be better applied to some special scenarios.

[0011] Preferably, in step S40, different weight values ​​are given to different employment relationships. For example, if there are multiple shareholders with the same shareholding ratio in the current node and there are multiple nodes of the shareholders connected to the current node by directed edges of the third type, the actual controller of the current node calculated by this super-step is determined based on the employment relationship with the highest weight value.

[0012] Furthermore, if during the execution of the 0th superstep in step S30 or the 1st to Nth superstep in step S40, a node finds that it has appeared in the control path transmitted from the upstream node of the node, it indicates that a closed loop has occurred; if this occurs in step S30, the node is marked as inactive, and the inactive node no longer sends messages to the downstream; if this occurs in step S40, the control path from the actual controller of this superstep calculation to the node and the node from the actual controller of this superstep calculation to the node remain unchanged, and the node is marked as inactive. This is a solution for the closed loop situation that occurs during iterative calculation, which prevents the iterative calculation from falling into an infinite loop as a whole.

[0013] Optionally, step S50 and step S60 are also included after step S40. Step S50: Use a breadth-first traversal algorithm to search for second-class directed edges between a certain node and the node of its suspected actual controller, and obtain all paths between the two nodes formed by second-class directed edges. Step S60: Calculate the comprehensive shareholding ratio of the node of the suspected actual controller to the node based on all paths formed by second-class directed edges between a certain node and the node of its suspected actual controller. Use a breadth-first traversal algorithm to mine all equity relationship paths between two nodes and then calculate the comprehensive shareholding ratio, and the calculation result will be more accurate and more in line with the actual situation.

[0014] The present application also proposes a device for deep mining of suspected actual controllers of an enterprise, including a graph construction unit, a 0th superstep calculation unit, and a 1-Nth superstep calculation unit. The graph construction unit is used to extract information from the business data of an enterprise to construct a graph; the constructed graph includes nodes and edges; all organizations or individuals involved in the extracted information are used as nodes in the graph; a first type of directed edge is constructed from a certain enterprise node to the node of the branch of the enterprise, or from the node of the executive partner of a partnership to the node of the partnership, and the first type of directed edge contains the attributes of the branch relationship or the executive partner relationship; a second type of directed edge is constructed from the node of a direct shareholder of an enterprise to the node of the enterprise, and the second type of directed edge contains the attributes of the direct shareholding ratio; a third type of directed edge is constructed from a certain enterprise node to the node of the senior executives of the enterprise, and the third type of directed edge contains the attributes of the position relationship. The 0th superstep calculation unit is used to initialize the actual controller of each enterprise type node to the node itself, and send a message from the node to the pointed node along the first type of directed edge, the second type of directed edge, and the third type of directed edge connected to the node; the message includes the actual controller of the node calculated by this superstep, and the control path from the node of the actual controller of the current superstep calculation to the node. The 1-Nth superstep calculation unit is used to read the received message by each current node, so as to know the actual controller of all upstream nodes by the previous superstep calculation, and the control path of all upstream nodes from the node of the actual controller of the previous superstep calculation to the upstream node, and combine the type of the edge pointing to the current node from all upstream nodes, and follow the uniqueness principle to calculate the control path of each current node by the actual controller of this superstep calculation, and each current node from the node of the actual controller of this superstep calculation to each current node; if the actual controller of the current node by the current superstep calculation changes, or the control path of the current node from the node of the actual controller of the current superstep calculation to the current node changes, then mark the current node as active, otherwise mark the current node as inactive; each active node sends the message to the pointed node along the direction of the first type of directed edge connected to the node, the second type of directed edge, and the third type of directed edge; the inactive node no longer sends messages downstream. The 1st to Nth superstep calculation units are repeatedly executed until all nodes participating in the iterative calculation are in an inactive state or the calculation is terminated when the maximum number of iterations N is reached; at this time, each node participating in the iterative calculation is regarded as the suspected actual controller of the node by the actual controller of this superstep calculation, and the control path of the node from the node of the actual controller of this superstep calculation to the node is regarded as the control path from the node of the suspected actual controller of the node to the node; the node participating in the iterative calculation refers to a node that sends at least one message or receives at least one message during the entire calculation process.The above-mentioned device has two conditions for determining whether to terminate the iterative calculation: (1) the actual controller of all nodes participating in the iterative calculation does not change; (2) the control path from the node of the actual controller of the current superstep calculation to the node of all nodes participating in the iterative calculation does not change; this makes the excavated control path more comprehensive.

[0015] The technical effect achieved by this application is: to efficiently and accurately calculate the actual controller and control path of an enterprise based on various relationships (edges) such as branches, direct shareholdings, and senior executives' positions, and to calculate all paths between two points based on the actual controller's data, to obtain the actual controller's comprehensive shareholding and all shareholding paths, thereby ensuring the accuracy and completeness of the actual controller's data. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flowchart of the deep mining method for suspected actual controllers of an enterprise proposed in this application.

[0017] Figure 2 It is a structural schematic diagram of the deep excavation device of the suspected actual controller of the enterprise proposed in this application.

[0018] Explanation of the reference numerals in the figure: 10 is a graph construction unit, 30 is a 0th superstep calculation unit, and 40 is a 1st to Nth superstep calculation unit. DETAILED DESCRIPTION

[0019] See also Figure 1 The in-depth mining method for suspected actual controllers of an enterprise proposed in this application includes the following steps.

[0020] Step S10: extract information from the enterprise's business data to construct a graph.

[0021] The information extracted in this step includes, for example: the name of the enterprise, the registered capital of the enterprise, the type of enterprise, the industry to which the enterprise belongs, the direct shareholders of the enterprise, the names of the senior executives of the enterprise (such as legal representatives, directors, and supervisors), the branches of the enterprise, the executive partners of the enterprise, etc. There is very little information about the branches of the enterprise contained in the business data of the enterprise. This application is based on the data of the existing branches and supplements the branch data according to the name of the enterprise (including former names) and the regional code of the enterprise. For example, "XX Bank Co., Ltd. XX Province Branch" is a branch of "XX Bank Co., Ltd.". The partners of the enterprise are for partnerships. Partnerships are divided into limited partnerships and general partnerships, both of which have executive partners.

[0022] The graph constructed in this step includes nodes and edges. All organizations or individuals involved in the extracted information (such as enterprises, direct shareholders of enterprises, natural persons, branches of enterprises, etc.) are taken as nodes in the graph, and the nodes have unique ID (identification), name, type and other attributes. The name of the node is, for example, the name of the organization or the name of a natural person. The types of nodes include, for example, E, P, G, S, Z, LT, U, etc. Among them, E represents various types of enterprises such as individual industrial and commercial households, sole proprietorships, cooperatives, partnerships, companies, state-owned enterprises, collective enterprises, etc.; P represents natural persons; G represents government agencies; S represents public institutions; Z represents social organizations; LT represents listed companies; U represents other types. Obviously, a node can have one or more types. The direct shareholding relationship, personnel position relationship, branch relationship, and partnership relationship in the extracted information are taken as edges in the graph, that is, lines are created between nodes. For example, based on the relationship between the branch offices of an enterprise or the relationship between the executive affairs partners of an enterprise, a first type of directed edge is constructed between the nodes of an enterprise and its branches, or between the nodes of a partnership and its executive affairs partners. The first type of directed edge contains the attributes of the branch relationship or the executive affairs partner relationship, and the direction of the first type of directed edge is from a certain enterprise node to the node of the branch office of the enterprise, or from the node of the executive affairs partner of a certain partnership to the node of the partnership. Based on the direct shareholding data of an enterprise, a second type of directed edge is constructed between the nodes of an enterprise and its direct shareholders, and the second type of directed edge contains the attributes of the direct shareholding ratio, and the direction of the second type of directed edge is from the node of the direct shareholder of a certain enterprise to the node of the enterprise. Based on the position relationship of the senior executives of an enterprise, a third type of directed edge is constructed between the nodes of an enterprise and its senior executives, and the third type of directed edge contains the attributes of the position relationship, and its content is, for example, the description of the position, and the direction of the third type of directed edge is from a certain enterprise node to the node of the senior executives of the enterprise.

[0023] Preferably, data cleaning is performed before constructing the graph, such as checking the legality of equity ratios, cleaning the legality of executive data, checking data consistency, filling in missing data, etc., and then the graph is constructed based on the cleaned data to remove invalid and erroneous data.

[0024] Preferably, after constructing the graph, isolated nodes are excluded to reduce subsequent unnecessary data transmission and calculation. The isolated node refers to a single node data that is not connected to any edge.

[0025] Step S20: Evenly distribute the nodes and the edges connected to them to each server partition. For example, evenly distribute the nodes to each server partition according to the node ID. Assuming that the number of server partitions is M, the node ID is composed of numbers, and the node ID is divided by M and then a remainder operation is performed. The server partition to which the node is assigned is determined according to the remainder. Preferably, the edges from a certain node A to other nodes are assigned to the server partition assigned to the node A. This is an optional step.

[0026] Step S30: For each enterprise type node (the type of the node includes E), the actual controller of the node is initialized to the node itself, and a message is sent from the node to the pointed node along the first type of directed edge connected to the node, the second type of directed edge, and the third type of directed edge. This step is called the 0th superstep. Superstep refers to an iterative calculation in the bulk synchronous parallel (BSP) calculation. The node that sends the message is called the upstream node, and the node that receives the message is called the downstream node. The relationship between the upstream node and the downstream node is that the upstream node is connected to the downstream node through the first type of directed edge, the second type of directed edge, or the third type of directed edge, and the direction of one of the edges between the upstream node and the downstream node points to the downstream node. The message includes the actual controller of the node calculated by this superstep, and the path from the node of the actual controller of the node calculated by this superstep to the node (called the control path).

[0027] Step S40: Each downstream node in the previous superstep is called the current node. Each current node reads the received message, thereby knowing the actual controller of all upstream nodes calculated by the previous superstep, and the control path of all upstream nodes from the node of the actual controller of the previous superstep calculation to the upstream node. Combined with the type of edges pointing to the current node from all upstream nodes, the actual controller of each current node calculated by this superstep, and the control path of each current node from the node of the actual controller of this superstep calculation to each current node are calculated according to the uniqueness principle. The calculation rules are as follows: (1) If there is a first-class directed edge pointing from an upstream node of the current node to the current node, the actual controller of the current node calculated by this superstep is directly determined to be the actual controller of the upstream node. The logic is that the actual controller of the head office can be directly transferred to the subsidiary (branch) or the executive partner of the partnership to bear unlimited joint and several liability for the partnership and be responsible for executing the partnership affairs. (2) If there is no first-class directed edge among all the edges pointing to the current node, the actual controller of the current node calculated by this superstep is determined according to the absolute controlling shareholder and the largest sole shareholder principle through the direct shareholding ratio attribute of the second-class directed edge. For example, all upstream nodes of the current node V are A, B, and C, among which A points to V through a second-type directed edge with a direct shareholding ratio of 30%, B points to V through a second-type directed edge with a direct shareholding ratio of 30%, and C points to V through a second-type directed edge with a direct shareholding ratio of 40%. It is known that the actual controllers of A and B are both P, and the actual controller of C is Q. Therefore, it can be determined that the actual controller of V is P, because P has the greatest control over V. (3) If there are multiple shareholders with the same shareholding ratio (referring to the shareholding ratio after calculating equity penetration) at the current node, the actual controller of the current node is determined in combination with the position relationship of the third-type directed edge. For example, among multiple shareholders with the same shareholding ratio, one shareholder is also the legal representative of the current node, then the actual controller of the current node is determined to be the legal representative by the current superstep calculation. Preferably, different weight values ​​are given to different positions. For example, if there are multiple shareholders with the same shareholding ratio (referring to the shareholding ratio after calculating equity penetration) at the current node and there are multiple nodes of the shareholders connected to the current node by a third type of directed edge, the actual controller of the current node calculated by this superstep is determined according to the position relationship with the highest weight value. (4) The control path from the node of the actual controller of the previous superstep calculation to the upstream node of all upstream nodes is known, the actual controller of the current node calculated by this superstep is known, and the path from all upstream nodes to the current node is known. The three are combined to obtain the control path from the node of the actual controller of the current superstep calculation to the current node.

[0028] As the operation progresses, the actual controller of the current node may not change, but the control path from the node of the actual controller of the current node to the current node may change. For example, continuing the previous example, the actual controller of the current node V is P, and the control paths from P to V are P->A->V and P->B->V. If the actual controller of C is also calculated as P, the control path from P to V will increase to P->C->V. Changes in the control path need to be radiated downstream, because the control path of the downstream node will change as the control path of the upstream node changes.

[0029] If the actual controller of the current superstep calculation changes, or the control path from the node of the actual controller of the current superstep calculation to the current node changes, the current node is marked as active, otherwise the current node is marked as inactive. Each active node sends a message to the pointed node along the direction of the first type of directed edge, the second type of directed edge, and the third type of directed edge connected to the node. This step is called the 1-Nth superstep. The node that sends the message is called the upstream node, and the node that receives the message is called the downstream node. The message contains the actual controller of the node in this superstep calculation, and the control path from the node of the actual controller of the current superstep calculation to the node. Nodes in the inactive state no longer send messages downstream.

[0030] Repeat step S40 to start the next superstep iteration process until all nodes participating in the iterative calculation are in an inactive state or the calculation ends when the maximum number of iterations N is reached. At this time, each node participating in the iterative calculation is used as the suspected actual controller of the node by the actual controller of this superstep calculation, and the control path from the node of the actual controller of this superstep calculation to the node is used as the control path from the node of the suspected actual controller of the node to the node. The node participating in the iterative calculation refers to a node that sends at least one message or receives at least one message during the entire calculation process. When all nodes participating in the iterative calculation are in an inactive state, it indicates that the actual controller of all nodes participating in the iterative calculation no longer changes, and the control path from the node of the actual controller of this superstep calculation to the node of all nodes participating in the iterative calculation no longer changes.

[0031] Furthermore, if during the execution of the 0th superstep in step S30 or the 1st to Nth superstep in step S40, a node finds that it has appeared in the control path transmitted from the upstream node of the node, it indicates that a closed loop has occurred. This indicates that the node points to the upstream node of the node through one or more edges, and the upstream node of the node also points to the node through one or more edges. In order to avoid entering an infinite loop, if this occurs in step S30, the node is marked as inactive, and the inactive node no longer sends messages to the downstream; if this occurs in step S40, the control path from the actual controller of this superstep calculation to the node and the node from the actual controller of this superstep calculation to the node remain unchanged, and the node is marked as inactive.

[0032] Optionally, step S50 and step S60 are further included after step S40.

[0033] Step S50: Use a breadth-first traversal (also known as a BFS) algorithm to search for the second type of directed edges between a node and the node of the suspected actual controller, and obtain all paths between the two nodes that are formed by the second type of directed edges. After step S40 is executed, the control path from the node of the suspected actual controller of a node to the node is obtained. The path formed by the second type of directed edges in the control path may not reflect all the paths formed by the second type of directed edges between the two nodes, so a more comprehensive mining is performed in step S50.

[0034] Step S60: Calculate the comprehensive shareholding ratio of the node of the suspected actual controller to the node based on all paths formed by the second type of directed edges between the node and the node of the suspected actual controller.

[0035] For example, it is known that there is an enterprise node S and its suspected actual controller node AC. There are three paths from node AC to node S composed of second-type directed edges, namely P1, P2, and P3. Path P1 includes: from node AC to node V2 through a second-type directed edge with a direct shareholding ratio attribute of K3, from node V2 to node V1 through a second-type directed edge with a direct shareholding ratio attribute of K2, and from node V1 to node S through a second-type directed edge with a direct shareholding ratio attribute of K1. Then the single-path shareholding ratio of path P1 PK1 = K1×K2×K3. Similarly, the single-path shareholding ratio PK2 of path P2 and the single-path shareholding ratio PK3 of path P3 are calculated respectively. The sum of each single-path shareholding ratio PK1+PK2+PK3 is the comprehensive shareholding ratio of node AC to node S.

[0036] See also Figure 2The deep mining device for suspected actual controllers of an enterprise proposed in this application includes a graph construction unit 10, a 0th superstep calculation unit 30, and a 1st-Nth superstep calculation unit 40.

[0037] The graph construction unit 10 is used to extract information from the business data of an enterprise to construct a graph. The constructed graph includes nodes and edges. All organizations or individuals involved in the extracted information are used as nodes in the graph. A first type of directed edge is constructed from a certain enterprise node to the node of the branch of the enterprise, or from the node of the executive partner of a partnership to the node of the partnership, and the first type of directed edge contains the attributes of the branch relationship or the executive partner relationship. A second type of directed edge is constructed from the node of a direct shareholder of an enterprise to the node of the enterprise, and the second type of directed edge contains the attributes of the direct shareholding ratio. A third type of directed edge is constructed from a certain enterprise node to the node of the senior executives of the enterprise, and the third type of directed edge contains the attributes of the position relationship.

[0038] The 0th superstep calculation unit 30 is used to initialize the actual controller of each enterprise type node to the node itself, and send a message from the node to the directed node along the first type of directed edge, the second type of directed edge, and the third type of directed edge connected to the node. The message contains the actual controller of the node calculated by this superstep, and the control path of the node from the node of the actual controller of the current superstep calculation to the node.

[0039] The 1-Nth superstep calculation unit 40 is used to read the received message by each current node, so as to know the control path of all upstream nodes from the actual controller of the previous superstep calculation and the node of the actual controller of the previous superstep calculation to the upstream node, and the control path of each current node from the actual controller of the current superstep calculation to each current node is calculated according to the uniqueness principle in combination with the type of the edge of all upstream nodes pointing to the current node. If the actual controller of the current superstep calculation changes, or the control path of the current node from the node of the actual controller of the current superstep calculation to the current node changes, the current node is marked as active, otherwise the current node is marked as inactive. Each active node sends the message to the pointed node along the first type of directed edge connected to the node, the second type of directed edge, and the third type of directed edge. The inactive node no longer sends messages downstream.

[0040] The 1-Nth superstep calculation unit 40 is repeatedly executed until all nodes participating in the iterative calculation are in an inactive state or the calculation ends when the maximum number of iterations N is reached. At this time, the actual controller of each node participating in the iterative calculation is used as the suspected actual controller of the node, and the control path from the node of the actual controller of the superstep calculation to the node is used as the control path from the node of the suspected actual controller of the node to the node. The node participating in the iterative calculation refers to a node that sends at least one message or receives at least one message during the entire calculation process.

[0041] The deep mining method of suspected actual controllers of an enterprise proposed in this application has the following beneficial technical effects.

[0042] First, this application simultaneously mines the actual controller of an enterprise based on the enterprise's branch relationship, the enterprise's executive affairs partner relationship, the enterprise's direct shareholding data, and the enterprise's senior management position relationship (i.e., three types of edges). The factors considered are relatively comprehensive. In special scenarios, for example, when two people hold 50% of the shares of a certain enterprise at the same time, it can also determine whether to continue iterative calculations.

[0043] Second, the condition for this application to determine whether to end the iterative calculation is that the actual controller of all nodes participating in the iterative calculation does not change from this super-step calculation, and the control path from the node of the actual controller of this super-step calculation to the node of all nodes participating in the iterative calculation does not change. After a certain iterative calculation, it may happen that the controller of the enterprise remains unchanged, but the control path of the controller of the enterprise changes. At this time, this application will not end the iterative calculation, but continue the iterative calculation, which makes the excavated control path more comprehensive.

[0044] Third, the existing technology usually calculates the comprehensive shareholding ratio based on the control path from the actual controller node of a certain enterprise to the enterprise node, which may miss some indirect shareholding paths. After obtaining the suspected actual controller of the enterprise, this application uses the breadth-first traversal algorithm to further explore all equity relationship paths between the two nodes, and then calculates the comprehensive shareholding ratio, so that the calculation result is more accurate and more in line with the actual situation.

[0045] Fourth, a solution is provided to prevent the iterative calculation from falling into an infinite loop, that is, a node points to its upstream node through one or more edges, and the upstream node of the node also points to the node through one or more edges.

[0046] The above are only preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should all be included in the protection scope of the present application.

Claims

1. A method for in-depth mining of suspected actual controllers of an enterprise, characterized by: The method comprises the following steps: Step S10: extract information from the enterprise business data to construct a graph; the constructed graph includes nodes and edges; all organizations or individuals involved in the extracted information are used as nodes in the graph; A first type of directed edge is constructed from a certain enterprise node to the node of the branch of the enterprise, or from the node of the executive partner of a certain partnership to the node of the partnership. The first type of directed edge contains the attributes of the branch relationship or the executive partner relationship. A second type of directed edge is constructed from the node of the direct shareholder of a certain enterprise to the node of the enterprise. The second type of directed edge contains the attributes of the direct shareholding ratio. A third type of directed edge is constructed from a certain enterprise node to the node of the senior executives of the enterprise. The third type of directed edge contains the attributes of the position relationship. Step S30: for each enterprise-type node, the actual controller of the enterprise-type node is initialized to the enterprise-type node itself, and a message is sent from the enterprise-type node to the directed node along the directions of the first type of directed edge, the second type of directed edge, and the third type of directed edge connected to the enterprise-type node; This step is called the 0th superstep; the node that sends the message is called the upstream node, and the node that receives the message is called the downstream node; the message contains the actual controller of the enterprise type node calculated by this superstep, and the control path of the enterprise type node from the node of the actual controller of this superstep calculation to the node of the enterprise type; Step S40: Each current node reads the received message, thereby learning the control paths of all upstream nodes from the actual controller of the previous superstep calculation, and from the node of the actual controller of the previous superstep calculation to the upstream node, and combining the types of the edges pointing to the current node from all upstream nodes, and following the uniqueness principle, calculating the control paths of each current node from the actual controller of this superstep calculation, and from the node of the actual controller of each current node to each current node; If the actual controller of the current super-step calculation of the current node changes, or the control path from the node of the actual controller of the current super-step calculation to the current node changes, the current node is marked as active, otherwise the current node is marked as inactive; each active node sends the message to the pointed node along the direction of the first type of directed edge, the second type of directed edge, and the third type of directed edge connected to the active node; This step is called the 1-N superstep; inactive nodes no longer send messages downstream; Repeat step S40 to start the next superstep iteration process, until all nodes participating in the iterative calculation are in an inactive state or the calculation is terminated when the maximum number of iterations N is reached; at this time, the actual controller of each node participating in the iterative calculation in this superstep calculation is used as the suspected actual controller of the active node, and the control path of the active node from the node of the actual controller of this superstep calculation to the active node is used as the control path from the node of the suspected actual controller of the active node to the active node; The nodes participating in the iterative calculation refer to nodes that send at least one message or receive at least one message during the entire calculation process.

2. The deep mining method for suspected actual controllers of an enterprise according to claim 1 is characterized in that: In step S10, the extracted information includes: the name of the enterprise, the registered capital of the enterprise, the type of the enterprise, the industry to which the enterprise belongs, the direct shareholders of the enterprise, the names of the senior executives of the enterprise, the branches of the enterprise, and the executive partners of the partnership.

3. The deep mining method for suspected actual controllers of an enterprise according to claim 1 is characterized in that: In the step S10, data cleaning is performed before constructing the graph, including any one or more of equity ratio legitimacy verification, senior management data legitimacy cleaning, data consistency checking, and filling in missing data, and then the graph is constructed based on the cleaned data.

4. The deep mining method for suspected actual controllers of an enterprise according to claim 1 is characterized in that: In the step S10, after the graph is constructed, isolated nodes are excluded, where the isolated nodes refer to single nodes that are not connected to any edges.

5. The deep mining method for suspected actual controllers of an enterprise according to claim 1 is characterized in that: After step S10 and before step S30, step S20 is also included; Step S20: Evenly distribute the nodes and the edges connected to them to each server partition; the edges starting from a certain node A and pointing to other nodes are allocated to the server partition allocated to the node A.

6. The deep mining method for suspected actual controllers of an enterprise according to claim 1 is characterized in that: In step S40, the calculation rules are as follows: (1) if there is a first-class directed edge pointing from an upstream node of the current node to the current node, the actual controller of the current node calculated by this superstep is directly determined to be the actual controller of the upstream node; (2) if there is no first-class directed edge among all the edges pointing to the current node, the actual controller of the current node calculated by this superstep is determined according to the absolute controlling shareholder and the largest sole shareholder principle through the direct shareholding ratio attribute of the second-class directed edge; (3) If there are multiple shareholders with the same shareholding ratio at the current node, the actual controller of the current node is determined by combining the employment relationship of the third type of directed edges; (4) The control paths from the node of the actual controller of the previous superstep calculation to the upstream node of all upstream nodes are known, the actual controller of the current superstep calculation of the current node is known, and the paths from all upstream nodes to the current node are known. The three are combined to obtain the control path from the node of the actual controller of the current superstep calculation to the current node.

7. The deep mining method for suspected actual controllers of an enterprise according to claim 6 is characterized in that: In the step S40, different weight values ​​are given to different employment relationships. For example, if there are multiple shareholders with the same shareholding ratio at the current node and there are multiple nodes of shareholders with the same shareholding ratio connected to the current node by a third type of directed edge, the actual controller of the current node calculated by this superstep is determined based on the employment relationship with the highest weight value.

8. The deep mining method for suspected actual controllers of an enterprise according to claim 1 is characterized in that: If, during the execution of the 0th superstep in step S30, a node of a certain enterprise type finds that the node of the enterprise type has appeared in the control path transmitted from the upstream node of the node of the enterprise type, it indicates that a closed loop has occurred; the node of the enterprise type is marked as inactive, and the inactive node no longer sends messages to the downstream; If, during the execution of the 1st to Nth supersteps in step S40, a node finds that the active node has appeared in the control path transmitted from the upstream node of the active node, it indicates that a closed loop has occurred; the control path from the actual controller of the active node in this superstep calculation to the node of the active state and from the node of the actual controller of the superstep calculation to the node of the active state are kept unchanged, and the active node is marked as inactive.

9. The method for deep mining of suspected actual controllers of an enterprise according to claim 1 is characterized in that The step S40 also includes steps S50 and S60; Step S50: using a breadth-first traversal algorithm to search for second-type directed edges between a node of a certain enterprise type and a node of a suspected actual controller, and obtaining all paths between the two nodes formed by second-type directed edges; Step S60: Calculate the comprehensive shareholding ratio of the node of a certain enterprise type to the node of the suspected actual controller based on all paths formed by the second type of directed edges between the node of the suspected actual controller and the node of the suspected actual controller.

10. A device for deep mining of suspected actual controllers of an enterprise, characterized in that: It includes a graph construction unit, a 0th superstep calculation unit, and a 1st to Nth superstep calculation unit; The graph construction unit is used to extract information from the business data of the enterprise to construct a graph; the constructed graph includes nodes and edges; all organizations or individuals involved in the extracted information are used as nodes in the graph; A first type of directed edge is constructed from a certain enterprise node to the node of the branch of the enterprise, or from the node of the executive partner of a certain partnership to the node of the partnership. The first type of directed edge contains the attributes of the branch relationship or the executive partner relationship. A second type of directed edge is constructed from the node of the direct shareholder of a certain enterprise to the node of the enterprise. The second type of directed edge contains the attributes of the direct shareholding ratio. A third type of directed edge is constructed from a certain enterprise node to the node of the senior executives of the enterprise. The third type of directed edge contains the attributes of the position relationship. The 0th superstep calculation unit is used for initializing the actual controller of each enterprise type node to the enterprise type node itself, and sending a message from the enterprise type node to the directed node along the first type of directed edge, the second type of directed edge, and the third type of directed edge connected to the enterprise type node; the node sending the message is called the upstream node, and the node receiving the message is called the downstream node; the message contains the actual controller of the enterprise type node calculated by this superstep, and the control path of the enterprise type node from the node of the actual controller of this superstep calculation to the enterprise type node; The 1-Nth superstep calculation unit is used to read the received message by each current node, so as to know the control path of all upstream nodes from the actual controller of the previous superstep calculation, and the control path of all upstream nodes from the node of the actual controller of the previous superstep calculation to the upstream node, and calculate the control path of each current node from the actual controller of this superstep calculation, and each current node from the node of the actual controller of this superstep calculation to each current node in accordance with the uniqueness principle in combination with the type of the edge pointing to the current node from all upstream nodes; If the actual controller of the current superstep calculation of the current node changes, or the control path from the node of the actual controller of the current superstep calculation to the current node changes, the current node is marked as active, otherwise the current node is marked as inactive; each active node sends the message to the pointed node along the direction of the first type of directed edge, the second type of directed edge, and the third type of directed edge connected to the active node; this is called the 1st-Nth superstep; the inactive node no longer sends messages downstream; The 1st to Nth superstep calculation units are repeatedly executed until all nodes participating in the iterative calculation are in an inactive state or the calculation is terminated when the maximum number of iterations N is reached; at this time, the actual controller of each node participating in the iterative calculation is used as the suspected actual controller of the active node, and the control path of the active node from the node of the actual controller of the current superstep calculation to the active node is used as the control path from the node of the suspected actual controller of the active node to the active node; The nodes participating in the iterative calculation refer to nodes that send at least one message or receive at least one message during the entire calculation process.

Images