Method and device for mining associated transactions among multiple objects based on graph risk control

Abstract

The invention discloses a method and device for mining associated transactions among multiple objects based on graph risk control, and relates to the technical field of artificial intelligence and the financial field, and the method comprises the steps: determining a first object and at least one second object having a transaction relationship and / or association relationship with the first object; establishing an entity relationship network according to the first object, the at least one second object and the corresponding association relationship; establishing a transaction relationship network according to the first object, the at least one second object and the corresponding transaction relationship; and performing multi-source network superposition on the entity relationship network and the transaction relationship network to generate an associated transaction knowledge graph between the first object and the at least one second object so as to mine associated transactions. According to the method and the system, the associated transaction path and the risk point can be clearly displayed through atlas analysis for an enterprise which transacts through a complex equity structure and an association relationship, so that a hidden associated party is identified, and the aim of mining associated transactions among multiple objects is fulfilled.

Description

Technical Field

[0001] This application belongs to the field of knowledge graph technology in artificial intelligence, specifically relating to a method and apparatus for mining multi-object related transactions based on graph risk control. Background Technology

[0002] Related-party transactions, a common business practice between related parties of an enterprise, have a dual nature: legitimate related-party transactions can help improve resource allocation efficiency, but without effective supervision, they may be overused by controlling shareholders to manipulate profits and embellish financial statements, thereby affecting the legitimate rights and interests of minority shareholders. The scale of such activities and the systemic risks they may pose have attracted widespread attention to the stability of the financial economy, thus urgently requiring the establishment of a more efficient and accurate identification and supervision mechanism.

[0003] In existing technologies, identifying the aforementioned illegal transactions mainly relies on self-reporting by enterprises and manual verification of business registration information, which has the following four major drawbacks:

[0004] 1. Information asymmetry: Transactions are highly concealed;

[0005] 2. Data silos: Information is not shared across systems;

[0006] 3. Limitations of human intervention: It is difficult to penetrate multiple layers of shareholding.

[0007] 4. Massive amounts of data: It is difficult to improve the efficiency of indicator calculation.

[0008] Existing identification algorithms are limited to two parties in a transaction, specifically identifying entities with both shareholding and sales relationships. They lack the technical means to identify multi-party related-party transactions. This is particularly true in multi-party related-party transaction scenarios. Because group-level enterprises often have entities that only perform management functions rather than being transactional entities, identifying related-party transactions requires first locating the corresponding transaction entities through shareholding relationships before further identifying and defining the transaction relationships. Traditional data mining methods cannot penetrate these multiple levels. Summary of the Invention

[0009] This invention can be used in the technical field of knowledge graph technology in finance, and can also be used in any field other than finance.

[0010] One objective of this invention is to provide a method for mining multi-object related transactions based on graph risk control, so as to solve at least a portion of the above-mentioned technical problems.

[0011] Another object of the present invention is to provide a device for mining multi-object related transactions based on graph risk control. A further object of the present invention is to provide an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor, when executing the computer program, implements the steps of the above-described method for mining multi-object related transactions based on graph risk control. A further object of the present invention is to provide a readable medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method for mining multi-object related transactions based on graph risk control.

[0012] To address the technical problems in the background section of this application, the present invention provides the following technical solutions:

[0013] In a first aspect, the present invention provides a method for mining multi-object related transactions based on graph risk control, comprising:

[0014] Identify a first object and at least one second object with which it has a transaction relationship and / or association relationship; wherein the transaction relationship includes information on fund outflows and inflows, and the frequency, amount, and type of transactions in derivative transaction relationships;

[0015] Establish an entity relationship network based on the first object, the at least one second object, and the corresponding association relationships;

[0016] A transaction relationship network is established based on the first object, the at least one second object, and the corresponding transaction relationship;

[0017] The entity relationship network and the transaction relationship network are overlaid with a multi-source network to generate a knowledge graph of related transactions between the first object and the at least one second object, and the related transactions between the first object and the at least one second object are mined through the knowledge graph of related transactions.

[0018] In some embodiments of the present invention, establishing a transaction relationship network based on the first object, the at least one second object, and the corresponding transaction relationship includes:

[0019] Find the connection path between the node corresponding to the first object and the node corresponding to the at least one second object in the pre-generated event graph;

[0020] Generate a spatial vector for the connection path; wherein the spatial vector is used to store the transaction identifier and transaction party information corresponding to the connection path;

[0021] The associated transaction path between the node corresponding to the first object and the node corresponding to the at least one second object is determined based on the spatial vector and the connection path.

[0022] The transaction relationship network is established based on the associated transaction path.

[0023] In some embodiments of the present invention, establishing the transaction relationship network based on the associated transaction path includes:

[0024] If the spatial vector does not contain the transaction identifier of the target transaction, then the multidimensional attribute features of the transaction corresponding to the spatial vector are extracted;

[0025] Based on the multidimensional attribute features, determine whether there is a related attribute between the first object and the at least one second object;

[0026] If there is an association attribute, the transaction relationship network is established based on the associated transaction path between the first object and the at least one second object.

[0027] In some embodiments of the present invention, the multidimensional attribute features include: behavioral features of transactions between the first object and the at least one second object, fund flow features, specific business features, and network features of the entity relationship network.

[0028] In some embodiments of the present invention, mining related transactions between the first object and the at least one second object through the related transaction knowledge graph includes:

[0029] Features describing the transaction between the first object and the at least one second object are extracted from the related transaction knowledge graph; the features include static structure, dynamic behavior, and graph embedding.

[0030] Based on the aforementioned features, a feature profile of the transaction between the first object and the at least one second object is abstracted;

[0031] The feature profiles are compared in the pre-generated related transaction model to uncover the related transactions.

[0032] In some embodiments of the present invention, mining related transactions between the first object and the at least one second object through the related transaction knowledge graph includes:

[0033] Extract transaction flow indicators between the first object and the at least one second object from the related transaction knowledge graph;

[0034] The pre-generated transaction flow indicator model is used to compare the transaction flow indicators and identify the related transactions.

[0035] Secondly, the present invention provides a multi-object inter-transaction correlation mining device based on graph risk control, the device comprising:

[0036] The transaction object determination module is used to determine a first object and at least one second object with which it has a transaction relationship and / or association relationship; wherein, the transaction relationship includes fund outflow and inflow information, transaction frequency, amount and type in derivative transaction relationships;

[0037] An entity relationship network establishment module is used to establish an entity relationship network based on the first object, the at least one second object, and the corresponding association relationships.

[0038] A transaction relationship network establishment module is used to establish a transaction relationship network based on the first object, the at least one second object, and the corresponding transaction relationship.

[0039] The related transaction mining module is used to overlay the entity relationship network and the transaction relationship network into a multi-source network to generate a related transaction knowledge graph between the first object and the at least one second object, and to mine related transactions between the first object and the at least one second object through the related transaction knowledge graph.

[0040] In some embodiments of the present invention, the transaction relationship network establishment module includes:

[0041] A connection path lookup unit is used to find a connection path between the node corresponding to the first object and the node corresponding to the at least one second object in a pre-generated event graph.

[0042] A spatial vector generation unit is used to generate a spatial vector of the connection path; wherein the spatial vector is used to store the transaction identifier and transaction party information corresponding to the connection path;

[0043] A related transaction path determination unit is used to determine the related transaction path between the node corresponding to the first object and the node corresponding to the at least one second object based on the spatial vector and the connection path.

[0044] A transaction relationship network establishment unit is used to establish the transaction relationship network based on the associated transaction path.

[0045] In some embodiments of the present invention, the transaction relationship network establishment unit includes:

[0046] A multidimensional attribute feature extraction unit is used to extract the multidimensional attribute features of the transaction corresponding to the spatial vector if the spatial vector does not contain the transaction identifier of the target transaction.

[0047] The association attribute determination unit is used to determine whether there is an association attribute between the first object and the at least one second object based on the multidimensional attribute features;

[0048] A transaction relationship network establishment subunit is used to establish the transaction relationship network based on the associated transaction path between the first object and the at least one second object, if the object has an association attribute.

[0049] In some embodiments of the present invention, the multidimensional attribute features include: behavioral features of transactions between the first object and the at least one second object, fund flow features, specific business features, and network features of the entity relationship network.

[0050] In some embodiments of the present invention, the related transaction mining module includes:

[0051] A transaction feature extraction unit is used to extract features from the associated transaction knowledge graph to describe the transaction between the first object and the at least one second object; the features include static structure, dynamic behavior, and graph embedding.

[0052] A feature profile extraction unit is used to abstract a feature profile of the transaction between the first object and the at least one second object based on the features.

[0053] The first unit for related-party transaction mining is used to compare the feature profiles in the pre-generated related-party transaction model to mine the related-party transactions.

[0054] In some embodiments of the present invention, the related transaction mining module includes:

[0055] A transaction flow indicator extraction unit is used to extract transaction flow indicators between the first object and the at least one second object from the related transaction knowledge graph.

[0056] The second unit for related transaction mining is used to compare the transaction indicators with the pre-generated transaction indicator model to mine the related transactions.

[0057] Thirdly, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of a method for mining multi-object related transactions based on graph risk control.

[0058] Fourthly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of a multi-object related transaction mining method based on graph risk control.

[0059] Fifthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for mining multi-object related transactions based on graph risk control.

[0060] As described above, embodiments of the present invention provide a method and apparatus for mining multi-object related transactions based on graph risk control. The method includes: first, determining a first object and at least one second object with which it has a transaction relationship and / or related relationship; wherein, the transaction relationship includes fund outflow and inflow information, transaction frequency, amount and type in derivative transaction relationships; next, establishing an entity relationship network based on the first object, at least one second object and the corresponding related relationship; establishing a transaction relationship network based on the first object, at least one second object and the corresponding transaction relationship; finally, superimposing the entity relationship network and the transaction relationship network into a multi-source network to generate a related transaction knowledge graph between the first object and at least one second object, and mining related transactions between the first object and at least one second object through the related transaction knowledge graph.

[0061] The present invention provides a method for mining multi-object related transactions based on graph risk control. For enterprises that conduct transactions through complex equity structures and related relationships, graph analysis can clearly show their related transaction paths and risk points, thereby identifying hidden related parties and achieving the purpose of mining multi-object related transactions. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1 A flowchart illustrating a method for mining multi-object related transactions based on graph risk control in an embodiment of the present invention;

[0064] Figure 2 This is a flowchart illustrating step 300 of a multi-object association transaction mining method based on graph risk control in an embodiment of the present invention.

[0065] Figure 3 This is a flowchart illustrating step 304 of a multi-object association transaction mining method based on graph risk control in an embodiment of the present invention.

[0066] Figure 4 This is a flowchart illustrating step 400 of a multi-object association transaction mining method based on graph risk control in an embodiment of the present invention. Figure 1 ;

[0067] Figure 5 This is a flowchart illustrating step 400 of a multi-object association transaction mining method based on graph risk control in an embodiment of the present invention. Figure 2;

[0068] Figure 6 This is a flowchart illustrating a method for mining multi-object related transactions based on graph risk control in a specific embodiment of the present invention.

[0069] Figure 7 This is a schematic diagram of multi-party related transactions of a group-level enterprise in a specific embodiment of the present invention;

[0070] Figure 8 This is a schematic diagram of a nested shareholding related-party transaction of a group-level enterprise in a specific embodiment of the present invention;

[0071] Figure 9 This is a block diagram of a multi-object association transaction mining device based on graph risk control in an embodiment of the present invention;

[0072] Figure 10 This is a block diagram of the transaction relationship network establishment module 30 in an embodiment of the present invention;

[0073] Figure 11 This is a block diagram of the transaction relationship network establishment unit 30d in an embodiment of the present invention;

[0074] Figure 12 This is a block of the related transaction mining module 40 in an embodiment of the present invention. Figure 1 ;

[0075] Figure 13 This is a block of the related transaction mining module 40 in an embodiment of the present invention. Figure 2 ;

[0076] Figure 14 This is a schematic diagram of the structure of an electronic device in an embodiment of the present invention. Detailed Implementation

[0077] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0078] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0079] It should be noted that the terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover a non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or apparatuses. Without conflict, the embodiments and features in the embodiments of this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0080] The information collected in the technical solution of this application is information and data authorized by the user or fully authorized by all parties. The collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.

[0081] Provide users with corresponding operation entry points, allowing them to choose to agree to or reject the automated decision results; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0082] The acquisition, storage, use, and processing of data in this application all comply with relevant laws and regulations. Specifically:

[0083] First, the information collected is information and data authorized by the user or fully authorized by all parties. The collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data shall comply with the relevant laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, not violate public order and good morals, and provide corresponding operation access points for users to choose to authorize or refuse.

[0084] Second, provide users with corresponding operation entry points for them to choose to agree to or reject the automated decision results; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0085] Patent application CN111444410A discloses a method for mining and identifying related transactions based on a knowledge graph, comprising the following steps: S1: acquiring target data and cleaning and filtering out reliable data; S2: extracting entities and relationships from the reliable data, importing the extracted entities and their corresponding relationships into a graph database to generate a preliminary knowledge graph; S3: searching for entity pairs that satisfy association rules from the preliminary knowledge graph, adding the association relationships between the entity pairs into the graph database obtained in S2, and obtaining a complete knowledge graph; S4: further increasing the association relationships through a graph representation learning algorithm to obtain a final knowledge graph; S5: using the final knowledge graph to identify related transaction information. This method has the following technical drawbacks:

[0086] 1. The previous plan did not involve multi-dimensional data, only shareholding and sales relationships. The new plan involves operational data with more comprehensive dimensions, and can set thresholds for related-party transactions to distinguish between benign and illegal related-party transactions.

[0087] 2. It does not involve massive amounts of data, does not have a big data ecosystem, and the calculation of indicators takes several hours to several days, which affects timeliness and hinders actual implementation.

[0088] Based on this, embodiments of the present invention provide a specific implementation of a method for mining multi-object related transactions based on graph risk control, see [link to relevant documentation]. Figure 1 The method specifically includes the following:

[0089] Step 100: Identify the first object and at least one second object with which it has a transaction relationship and / or association relationship; wherein, the transaction relationship includes information on fund outflows and inflows, and the frequency, amount, and type of transactions in derivative transaction relationships;

[0090] Step 200: Establish an entity relationship network based on the first object, the at least one second object, and the corresponding association relationships;

[0091] Step 300: Establish a transaction relationship network based on the first object, the at least one second object, and the corresponding transaction relationship;

[0092] Step 400: Overlay the entity relationship network and the transaction relationship network using a multi-source network to generate a knowledge graph of related transactions between the first object and the at least one second object, and mine the related transactions between the first object and the at least one second object through the knowledge graph of related transactions.

[0093] The present invention provides a method for mining multi-object related transactions based on graph risk control. For enterprises that conduct transactions through complex equity structures and related relationships, graph analysis can clearly show their related transaction paths and risk points, thereby identifying hidden related parties and achieving the purpose of mining multi-object related transactions.

[0094] Regarding step 100, it can be understood that a transaction relationship refers to the existence of historical transaction records between the first object and the second object, while an association relationship refers to the existence of social relationships between the two, such as relatives, friends, colleagues, and partners.

[0095] For step 200, the first step is to extract information related to "entities" and "relationships" from the multi-source, heterogeneous data. The multi-source, heterogeneous data mentioned above includes:

[0096] Individual users' ID card, name, communication ID, address, etc.; corporate clients' company name, business license number, legal representative, registered address, etc.

[0097] Transaction log data includes the accounts and names of both parties, transaction time, amount, purpose, etc., and is used to indirectly infer and verify the relationship between entities.

[0098] Business system data: loan application information, wealth management product purchase information, contract information, etc.

[0099] Basic company information: Shareholder information (shareholder name / company name, shareholding ratio), senior management information (directors, supervisors, general manager, etc.), branch office information.

[0100] Company change information: changes in historical shareholders, historical senior executives, and registered capital, etc.

[0101] Judicial data, such as litigation records and information on those subject to enforcement, can be used to identify risk associations.

[0102] Other data: intellectual property data, bidding data, etc.

[0103] Next, due to the significant ambiguity, duplication, and inconsistencies in the original data, cleaning and unification are necessary. First, it's crucial to determine whether records from different data sources point to the same entity. For example, techniques such as name standardization, fuzzy matching, and unified social credit code verification are used to determine whether "Tian Du Shi Ju Da Co., Ltd.", "Ju Da Co., Ltd.", and "Juda Group" refer to the same entity. This needs to be implemented. Next, a globally unique ID is assigned to each unique entity. For example, the unique ID for "Ju Da (China) Co., Ltd." is ultimately determined to be E_12345, and all related information is aggregated under this ID.

[0104] Define and extract relationships: Determine which types of edges (relationships) the network needs to include.

[0105] Business relationships include: shareholder-enterprise relationship (shareholding relationship), senior management-enterprise relationship (appointment relationship), enterprise-subsidiary / branch relationship (controlling relationship), and joint shareholding relationship (two entities jointly invest in the same enterprise).

[0106] Personal relationships: kinship (spouse, parents, children, etc., which can be inferred from certain business scenarios or external data), same address relationship (multiple entities registered or living at the same address, which may imply a connection), and guarantee relationship (person A provides a guarantee for person B's loan).

[0107] Next, relation extraction is performed by directly extracting shareholder and executive information from the business registration data table to generate "shareholding" and "appointment" relationship edges. Natural Language Processing (NLP) is then used to extract relations from news and announcement texts. For example, from a news article "Zhang was appointed CEO of XX Company," the triple (Zhang, Appointment, XX Company) is extracted.

[0108] Specifically, for indirect relationships, it can be inferred through transaction behavior. For example, if two different corporate entities make frequent and long-term transfers to the same personal account, it may suggest a potential control relationship that is not explicitly stated in the business registration.

[0109] Finally, the processed entities and relationships are stored as a graph structure for efficient association queries and network analysis.

[0110] For steps 300 and 400, knowledge graph technology can effectively address the aforementioned pain points due to its multi-dimensional relationship penetration capabilities:

[0111] 1. Realize a multi-dimensional relationship network between enterprises, shareholders, and transactions to improve the coverage of basic data.

[0112] 2. Identify hidden related paths (such as mining transaction entities with nested shareholdings).

[0113] 3. Monitor abnormal transaction patterns in real time.

[0114] 4. The big data ecosystem improves computing efficiency.

[0115] This will shift the supervision of related-party transactions from passive reporting to intelligent and proactive prevention and control.

[0116] For the multi-source network overlay in step 400, this application provides the following solution: the key to bridging overlay through a "shared entity" lies in finding the connection point between the two networks. This connection point is the shared entity that appears in both networks simultaneously.

[0117] Nodes in a transaction relationship network are typically accounts (corporate accounts, personal bank accounts), transaction IDs, etc. Nodes in an entity relationship network are typically individuals (ID card number) and enterprises (Unified Social Credit Code). Shared entities: An individual or an enterprise necessarily owns one or more accounts.

[0118] For example: Enterprise E (entity network node) owns account A (transaction network node). Individual P (entity network node) is the holder of account B (transaction network node). This "owning" or "being" relationship is the edge connecting the two networks.

[0119] In some embodiments of the present invention, see Figure 2 Step 300 includes:

[0120] Step 301: Find the connection path between the node corresponding to the first object and the node corresponding to the at least one second object in the pre-generated event graph;

[0121] The event graph in step 301 is a graph data structure where nodes represent entities (such as individuals, businesses, accounts, devices, etc.) or events (such as a transaction, a login), and edges represent relationships or behaviors between them (such as "transfer to", "belong to", "occurs at"). The system uses the nodes of the "first object" (such as suspicious person A) and the "second object" (such as shell company B) as the starting and ending points, and uses graph traversal algorithms (such as breadth-first search (BFS), depth-first search (DFS), or more advanced shortest path algorithms) to find all paths connecting these two nodes in the event graph. The purpose of this step is to discover all direct and indirect connections. A direct path (A->B) may be found, or multiple indirect paths (A->C->B, A->D->E->B) may be found. These paths reveal possible channels for the flow of funds or benefits.

[0122] Step 302: Generate a spatial vector for the connection path; wherein the spatial vector is used to store the transaction identifier and transaction party information corresponding to the connection path;

[0123] A spatial vector is a structured, computer-readable data format used to uniquely represent and describe a path. Preferably, the spatial vector is a sequence of feature codes.

[0124] For each path found, extract the key information along that path and package it into a data object (i.e., a "spatial vector"). This vector contains: Transaction identifiers: unique IDs for all transactions involved in the path. Transaction party information: information about all nodes (entities) traversed in the path. Other features: may also include derived features such as path length (number of hops), total transaction amount, and average transaction frequency.

[0125] Step 302 flattens the complex graph structure into a feature vector for subsequent automated calculations and pattern recognition (computers cannot directly "understand" graphs, but they can efficiently process and compare vectors).

[0126] Step 303: Determine the associated transaction path between the node corresponding to the first object and the node corresponding to the at least one second object based on the spatial vector and the connection path;

[0127] Understandably, not all found paths are important; rather, rules or models are needed to filter out truly meaningful "related transaction paths." Specifically, based on pre-defined risk rules or machine learning models, all "spatial vectors" generated in step 302 are analyzed. For example, a rule might be, "If the total amount of a path exceeds 1 million yuan and the number of hops is less than 4, then it is marked as a related transaction path." Alternatively, the vector can be input into a classification model, which will determine whether the path represented by the vector has suspicious related characteristics based on historical data.

[0128] Step 303 achieves the filtering from "massive potential paths" to "a few high-risk paths", which greatly improves the efficiency and quality of risk analysis and avoids the tedious work of manually reviewing all paths.

[0129] Step 304: Establish the transaction relationship network based on the associated transaction path.

[0130] Specifically, the "related transaction paths" identified as valuable in the previous step are fed back and enhanced into the existing transaction relationship network. These newly discovered, hidden related paths (e.g., A->C->B) and all their attributes (amount, time, node information) are added to the original transaction relationship network as new edges and nodes. Ultimately, the transaction relationship network can more accurately depict complex and hidden transaction relationships.

[0131] In some embodiments of the present invention, see Figure 3 Step 304 includes:

[0132] Step 3041: If the spatial vector does not contain the transaction identifier of the target transaction, then extract the multi-dimensional attribute features of the transaction corresponding to the spatial vector;

[0133] Step 3042: Determine whether there is an association attribute between the first object and the at least one second object based on the multidimensional attribute features;

[0134] Step 3043: If there is an association attribute, establish the transaction relationship network based on the association transaction path between the first object and the at least one second object.

[0135] In steps 3041 to 3043, if none of the spatial vectors contain the target transaction identifier, then the multi-dimensional attribute features corresponding to each spatial vector are extracted to identify whether there is a correlation attribute between the first user and the second user. If such an attribute exists, then the corresponding connection path is determined as its associated transaction path.

[0136] In some embodiments of the present invention, the multidimensional attribute features include: behavioral features of transactions between the first object and the at least one second object, fund flow features, specific business features, and network features of the entity relationship network.

[0137] Trading Behavior Characteristics: These characteristics focus on the historical trading patterns of an account or entity. The core idea is that abnormal behavior deviates from its own historical normal patterns or the general patterns of a group. Key indicators include:

[0138] Transaction frequency: The number of transactions per unit of time (e.g., daily / weekly / monthly transactions). Sudden spikes or drops are worth noting.

[0139] Transaction amount: Single transaction amount, daily / monthly average transaction amount, cumulative transaction amount. Pay close attention to the distribution of amounts (whether they are mostly integers, whether they are just below the reporting threshold), and the maximum / minimum amount.

[0140] Transaction time: The specific time when the transaction occurs (whether it is always late at night or outside of working hours), and the day of the week (whether it is always on the weekend).

[0141] Transaction speed: The speed at which multiple transactions occur in a very short period of time (e.g., 10 transactions in 1 minute).

[0142] Regularity of transactions: Do transactions have a fixed cycle (such as specific days of the month) or a fixed trading counterparty?

[0143] Transaction type preference: The proportion of different channels (online banking, ATM, counter) and different services (transfer, consumption, wealth management, loan repayment).

[0144] Fund Flow Characteristics: These characteristics focus on the movement and patterns of funds within a network of relationships, aiming to identify the source, destination, and circulation patterns of funds. They possess the following characteristics:

[0145] Fund Circulation: Does the fund circulate among multiple related accounts, returning to the starting point or related account after several stages, forming a "circular" flow? Net Fund Flow: Within a specific relationship (such as between A and B) or a period of time, is the fund experiencing net inflow, net outflow, or a basic balance?

[0146] Counterparty concentration: Is the flow of funds concentrated in / from a few counterparties, or dispersed in / from a large number of unrelated counterparties? Dispersed inflows and concentrated outflows (or vice versa) are all suspicious patterns.

[0147] Fund stratification: Whether a large amount of funds enters and is then quickly split into multiple smaller amounts and transferred to multiple downstream accounts.

[0148] Specific business characteristics: These characteristics are strongly correlated with the specific financial scenario and product. They have the following features (taking bond trading as an example):

[0149] Transaction direction: whether it is buying, selling, or repurchasing.

[0150] Bond type: Is the bond being traded a government bond, corporate bond, financial bond, or subordinated bond? Risk appetite differs.

[0151] Transaction Price / Yield: The deviation of the transaction price from the fair market value. Unusually discounted or premium transactions may indicate the transfer of benefits.

[0152] Trading Market: Whether a transaction takes place in the interbank market or the exchange market depends on the participants in each market.

[0153] Holding time: The length of time after purchase. Ultra-short-term trading may be used for wash trading or to create the illusion of liquidity.

[0154] Network characteristics of entity relationship networks: High-level metrics calculated from entity relationship networks (graph structures) using graph algorithms, used to quantify the position and importance of an entity in the entire network.

[0155] In some embodiments of the present invention, see Figure 4 Step 400 includes:

[0156] Step 401: Extract features from the related transaction knowledge graph to describe the transaction between the first object and the at least one second object; the features include static structure, dynamic behavior, and graph embedding;

[0157] From complex knowledge graphs, key indicators that describe the transactional relationship between "first object" and "second object" are quantitatively extracted. This is done across three dimensions, progressing from the surface to the core, from the concrete to the abstract:

[0158] Static structural features: Extracting the most direct and superficial "snapshot"-like attributes of the transaction relationship. Specifically, static structural features include the following:

[0159] Basic statistics: total transaction amount, number of transactions, average transaction amount, maximum / minimum single transaction amount.

[0160] Time span: The time interval between the first and last transactions.

[0161] Direct relationship: The frequency and amount of direct transactions between two objects.

[0162] Dynamic Behavioral Characteristics: This section analyzes the patterns, trends, and regularities of trading behavior over time. It focuses on "how" to trade, not just "how much." Specifically, dynamic behavioral characteristics include the following:

[0163] Trading timing patterns: Do the trades have a fixed cycle (e.g., every Friday), occur intensively at specific times (e.g., early morning), or suddenly change in trend (with the amount increasing continuously)?

[0164] Direction of fund flow: Is the fund flow unidirectional (always A pays B) or bidirectional (A and B transfer funds to each other)?

[0165] Behavioral sequence: The sequential pattern of transaction types, such as "large inflow -> rapid and dispersed outflow -> purchase of virtual goods".

[0166] Graph embedding features: Using algorithms such as graph neural networks, the global network structure information of each node (object) in a knowledge graph (including its indirect relationships beyond multi-hops) is compressed into a low-dimensional, dense numerical vector. This transforms complex, unstructured network relationships into "mathematical coordinates" that computers can efficiently compute and understand. In this vector space, nodes with similar network structures (e.g., both being "central violating accounts") are located close to each other, while nodes with significant structural differences are farther apart.

[0167] Step 402: Based on the features, abstract a feature profile of the transaction between the first object and the at least one second object;

[0168] The extracted massive amounts of multidimensional feature indicators are integrated and refined into a comprehensive and highly summarized "user profile" or "relationship profile". Through data modeling techniques (such as feature engineering, dimensionality reduction, clustering, etc.), the "essence" and "full picture" of the transaction relationship between the first object and the second object are outlined.

[0169] The output is a structured set of labels or a probability model. For example: suspected downstream accounts of an "ATM" structure: combining multiple features such as "small amount, high frequency, and dispersed transaction time". "Potential control" relationship: combining features such as "the actual beneficiary is the same person after equity penetration, highly synchronized transaction behavior, and extremely similar graph embedding vectors".

[0170] Step 403: Compare the feature profiles in the pre-generated related transaction model to mine the related transactions.

[0171] By inputting the feature profile into a pre-trained machine learning model and comparing it with known risk patterns, hidden risks can be automatically identified.

[0172] The related-party transaction model is trained on a large amount of historical data (including known normal and non-compliant cases). It has learned the characteristics of various risk patterns. The related-party transaction model calculates the similarity score between the current "feature profile" and various risk patterns. If the similarity exceeds a certain threshold, the model will "uncover" the potential risk type behind the related-party transaction and issue an alert.

[0173] Final output: The transaction path has a 92% similarity to the behavioral patterns of the fraudulent payment collection accounts.

[0174] In some embodiments of the present invention, see Figure 5 Step 400 also has another implementation method:

[0175] Step 404: Extract the transaction flow indicators between the first object and the at least one second object from the related transaction knowledge graph;

[0176] Step 405: Compare the transaction metrics in the pre-generated transaction metric model to identify the related transactions.

[0177] In steps 404 and 405, transaction flow indicators describing the transaction relationship between the first object and the second object are extracted from the related transaction knowledge graph, and it is verified whether there is a single indicator or combination of indicators that conforms to the preset abnormal transaction rules, so as to discover the related transactions.

[0178] Compared with existing technologies, the multi-object association transaction mining method based on graph risk control provided by this invention has the following beneficial effects:

[0179] 1. Establish a multi-dimensional relationship network among enterprises, shareholders, and transactions to improve the coverage of basic data;

[0180] 2. Identify hidden related-party transactions, such as multi-party related-party transactions and nested shareholding related-party transactions, and distinguish between legitimate related-party transactions and illegal related-party transactions;

[0181] 3. The application of the graph risk control community algorithm extends pattern recognition to the entire graph database for matching, which can discover and supplement related transaction patterns not covered by the abnormal transaction structure model, thereby improving the completeness of recognition;

[0182] 4. Incremental update capture, real-time monitoring of abnormal transaction patterns;

[0183] 5. The big data ecosystem improves computational efficiency by replacing full graph traversal with incremental computation.

[0184] For further explanation of the plan, see Figure 6 The present invention also provides a specific implementation method for mining multi-object related transactions based on graph risk control, which specifically includes the following contents.

[0185] In the specific embodiments provided by this invention, the storage structure of a graph database is applied to efficiently mine multi-level shareholding and sales relationships among main enterprises, extending these relationships to three levels above and below. Furthermore, to cope with the computational pressure brought by massive amounts of data to the graph database, incremental computation is used instead of full graph traversal, greatly saving computational resources and improving computational efficiency. The main graph technologies used include:

[0186] 1. Hypergraph Modeling: Integrates multi-dimensional relationships such as equity, guarantees, and fund transfers to construct complex relational networks. It realizes a multi-dimensional relationship network between enterprises, shareholders, and transactions, incorporating relevant information such as transaction proportion, transaction frequency, and registered address to improve the coverage of basic data;

[0187] 2. Graph structure modeling, considering complex transaction structures such as three-party and four-party transactions, and identifying hidden related paths (such as nested shareholding).

[0188] 3. Community detection algorithm: Extracts features of related-party transactions, distinguishes between benign and illegal related-party transactions, identifies enterprise clusters with dense related-party transactions, and locates potential groups involved in the transfer of benefits.

[0189] 4. Monitor abnormal transaction patterns in real time.

[0190] 5. The big data ecosystem improves computational efficiency by replacing full graph traversal with incremental computation.

[0191] S1: Governance and integration of existing data in the big data ecosystem.

[0192] The system integrates, cleans, and processes multi-source corporate tax data, tax-related business data, and judicial data. This data suffers from issues such as inconsistent formats (e.g., XML, JSON, CSV), inconsistent standards (e.g., differences in company name coding), and varying quality. Therefore, the system employs the following processing workflow:

[0193] Data standardization: Establish a unified enterprise entity identification system and link data through the enterprise social credit code. For historical data with missing codes, a composite matching algorithm of "enterprise name + legal representative + registered address" is used for association.

[0194] Data cleaning: This involves a combination of rule-based and machine learning methods. At the rule level, this includes deduplication and outlier detection; at the machine learning level, a classification model is trained to identify erroneous records in the data.

[0195] Data enhancement: Supplement enterprise-related information through external data sources, such as obtaining shareholder ownership relationships from business registration data sources and technology-related data from the intellectual property office. Establish a data quality assessment system, including dimensions such as completeness, accuracy, and timeliness, to ensure the reliability of downstream analysis.

[0196] S2: The big data ecosystem captures incremental financial business data.

[0197] Establish cleaning and updating mechanisms for different data sources to acquire the latest data and integrate it into the ecosystem; to maintain the timeliness of the knowledge graph, the system needs to establish a dynamic data update mechanism.

[0198] Unified Stream and Batch Architecture: Unifies streaming and batch processing within a single runtime framework for integrated processing. Real-time data streams and historical data batch processing can utilize the same set of data processing tools and technologies, the same set of data processing code, and be stored in the same data storage system, simplifying the complexity of data processing and improving its efficiency.

[0199] Change Data Capture (CDC): Deploy the Debezium connector on core business systems (such as bank transaction systems) to capture database binlogs in real time. For API data sources, design a backoff and retry mechanism to address interface rate limiting.

[0200] Incremental fusion strategy: Establish version management for enterprise data changes, adopting a SCD2-type slow-change dimension strategy to retain historical change trajectories. For key events such as equity changes, trigger real-time graph updates.

[0201] S3: Initial index calculation.

[0202] The big data ecosystem supports indicator processing and computation; the big data ecosystem needs to provide preprocessed indicators for graph construction.

[0203] Basic indicator calculation: Run indicator processing tasks on the Spark cluster, including: financial health indicators, such as debt-to-equity ratio and current ratio; operational risk indicators, such as number of lawsuits and amount of administrative penalties; and relationship complexity, such as number of holding companies and total amount of external investment.

[0204] Metrics as a Service: Processed metrics are managed uniformly through a feature platform, supporting low-latency real-time queries (Redis caching) and high-throughput batch retrieval (HBase).

[0205] S4: Graph construction and graph risk control index calculation.

[0206] The data and initial indicators are converted into a graph database storage structure. Based on the graph, graph risk control indicators are further calculated, and in-depth risk analysis is achieved at the graph database layer.

[0207] Graph Model Design: An attribute graph model is adopted, with node types including enterprises, individuals, and financial institutions, and edge types covering more than 20 relationships such as equity, guarantee, and transaction. Weight calculation rules are designed for each relationship, such as logarithmic standardization of transaction amounts.

[0208] Graph feature computation: Centrality index: PageRank identifies system-important enterprises; Community detection: Louvain algorithm detects potential related groups; Path analysis: A* algorithm finds abnormal funding links; Real-time graph computation: For large-scale graphs, GraphX ​​is used for distributed computation, and communication overhead is optimized through vertex cutting.

[0209] S5: Big data ecosystem and graph database in parallel.

[0210] By replacing full graph traversal with incremental computation, related transaction groups can be identified and differentiated, improving computational efficiency and enabling a highly efficient risk monitoring system.

[0211] Incremental graph processing: Based on the trigger-based incremental update mechanism, when a major change in the enterprise is detected (such as a change in equity exceeding 5%), the local subgraph is automatically recalculated to avoid full graph traversal.

[0212] Related-Party Transaction Identification: A multi-layered rule engine is constructed: shallow rules directly relate to transaction amount thresholds; deep analysis corresponds to the mining of counterparties in multi-level equity penetration, such as the algorithm for finding counterparties in third- and fourth-level shareholding relationships. The core objective of the algorithm for finding third- and fourth-level shareholding relationships is to accurately locate the indirect shareholding entities of a target company up to the fourth level through a systematic recursive traversal and optimization mechanism. This algorithm is based on a hierarchical recursive traversal framework, combined with shareholding path filtering and efficiency optimization strategies, constructing a search system that balances accuracy and performance. The specific method is described in the fifth step below.

[0213] Early warning feedback mechanism: False alarms are reduced through dynamic threshold adjustment (exponentially weighted moving average). Early warning information is automatically pushed through multiple channels such as WeChat and email, and the handling and feedback are recorded to form a closed loop.

[0214] S6: In the big data ecosystem, identify related transactions for updated financial knowledge graphs and issue real-time alerts to enterprise groups that exceed thresholds.

[0215] Step S5 establishes various related-party transaction models and thresholds to distinguish between legitimate and non-compliant transactions. The models include... Figure 7 as well as Figure 8As shown, it can identify multi-party related-party transactions and nested shareholding and related-party transactions within a group-level enterprise, and calculate and implement them in the graph risk control indicators of the big data ecosystem. Next, using the implementation of third- and fourth-level shareholding relationships, the specific steps for graph-based querying of counterparties in multi-level shareholding relationships are explained:

[0216] The core of finding third- and fourth-level shareholding relationships lies in traversing multi-layered shareholding structures using recursive functions, combined with a shareholding ratio filtering mechanism, to achieve accurate tracing and presentation of shareholding paths. The implementation process is explained in detail below with a specific example.

[0217] 1. Recursive search process:

[0218] (1) Starting with target company A (entity ID: C001), the system calls a recursive function to initiate the shareholding relationship search process, with the initial level set to 0. The specific steps are as follows:

[0219] (2) Level 1: Find the direct shareholders of A and identify entity B (entity ID: C002), which holds 30% of the shares. At this point, the level increases to 1.

[0220] (3) Level 2: Take B as the new target, find its shareholder C (entity ID: C003), with a shareholding ratio of 60%, and increase the level to 2;

[0221] (4) Level 3: Continue to trace the shareholder D (entity ID: C004) of C, with a shareholding ratio of 40%, reaching level 3 and forming a third-level shareholding relationship;

[0222] (5) Level 4: Find shareholder E (entity ID: C005) with D as the target, with a shareholding ratio of 20%, and increase the level to 4. Trigger the termination condition (the maximum level is preset to 4), and the recursive process terminates.

[0223] 2. Path filtering and result output.

[0224] After the recursion terminates, the system filters all tracing paths by shareholding ratio (default threshold is 5%), removing paths with shareholding ratios below the threshold. In this case, the shareholding ratio at each level is higher than 5%, so the complete path is retained. The final output is as follows:

[0225] (1) Third-level shareholding path: A←B←C←D (shareholding ratio of each level: B→A: 30%, C→B: 60%, D→C: 40%).

[0226] (2) Fourth-level shareholding path: A←B←C←D←E (shareholding ratio of each level: B→A: 30%, C→B: 60%, D→C: 40%, E→D: 20%).

[0227] Through the above process, the system achieves clear visualization of multi-level shareholding relationships, not only presenting the complete equity chain, but also including the specific shareholding ratio of each level, providing accurate data support for equity penetration analysis.

[0228] 3. Realization of the connection between shareholding relationship and transaction relationship.

[0229] The association between shareholding relationships and transaction relationships needs to be based on structured data from a knowledge graph, establishing a precise mapping between the two through multi-dimensional verification and quantitative calculation. Taking a specific scenario as an example: Suppose that a multi-level shareholding relationship has been used to locate the third-level shareholding company D of target company A (shareholding path A←B←C←D), it is necessary to further associate the transaction activities participated in by D. The specific implementation process is divided into the following steps.

[0230] First, targeted extraction of transaction data. Using the nGQL query language of the graph database, a pattern matching query "Enterprise D node - transaction --> Enterprise T node" can be executed to quickly obtain all transaction records T in which D is the transaction entity, realizing a preliminary connection from the shareholding entity to the transaction behavior.

[0231] Secondly, the validity of the time sequence needs to be verified. The correlation between the transaction and the shareholding relationship must be based on the time dimension. It is necessary to verify whether the "transaction time" of transaction T is within the "shareholding duration" of D to C, that is, satisfying the condition that D's "shareholding start time" ≤ T's "transaction time" ≤ D's "shareholding end time". This step can filter out non-related transactions caused by the failure or termination of the shareholding relationship.

[0232] Next, the correlation strength is quantitatively calculated. To achieve a quantitative assessment of the shareholding and trading relationship, a correlation strength index is introduced, and the calculation formula is as follows:

[0233] Association strength = (D's shareholding ratio in C × 0.5) + (Transaction amount of T / Annual revenue of target company A × 0.3) + (Time overlap × 0.2).

[0234] The temporal overlap is defined as the proportion of the transaction time within the holding period (if the transaction is instantaneous, the overlap is 1; if the transaction spans multiple time periods, the proportion is calculated based on the actual number of overlapping days). This formula comprehensively measures the degree of correlation through three dimensions: shareholding ratio (weight 0.5), relative value of transaction size (weight 0.3), and temporal matching degree (weight 0.2).

[0235] Key parameter descriptions:

[0236] (1) Shareholding ratio: reflects D’s control over C, with the highest weight (0.5), reflecting the core role of equity relationship;

[0237] (2) Transaction amount / annual revenue: Standardize the transaction size to eliminate the impact of differences in the size of different companies (weight 0.3);

[0238] (3) Timing overlap: Ensure the time consistency of transactions and shareholding behavior, and avoid non-synchronous correlation (weight 0.2).

[0239] Finally, effective association results are filtered. A correlation strength threshold (e.g., >0.6) is set to filter out transaction records that meet the criteria. The final output includes the association relationship between D and T, the specific strength score, and the parameter values ​​for each dimension, achieving accurate quantitative correlation between shareholding relationships and transaction relationships, providing data support for subsequent risk analysis or business decisions. This process, through the combination of the semantic association capabilities of knowledge graphs and quantitative models, effectively improves the accuracy and interpretability of transaction associations in multi-level shareholding scenarios.

[0240] Step S5 further includes the application of a graph-based risk control community detection algorithm. This algorithm extracts features of related-party transactions, distinguishes between benign and non-compliant related-party transaction features, identifies enterprise clusters with dense related-party transactions, locates potential groups involved in profit transfer, expands existing abnormal transaction structure models, and uses pattern recognition. It goes beyond the specific number of related parties and paths, abstracting the features of identified non-compliant related-party transactions, and matching potential related-party transaction groups that meet the features but differ in group size and transaction paths, thus achieving an early warning effect. This solution employs a multi-dimensional weighted community detection algorithm. This algorithm is no longer limited to simple topological structures but integrates structural, attribute, and temporal weights to construct a network that deeply reflects the real risk relationships between enterprises. Through the weighted Louvain clustering method, it can intelligently divide enterprise communities with close internal connections and similar risk attributes, laying a precise structural foundation for subsequent in-depth analysis.

[0241] After identifying potential communities, the system deeply extracts related transaction features from three levels: static structure, dynamic behavior, and graph embedding. The static level focuses on the community's size, density, and power concentration (such as the centrality Gini coefficient); the dynamic level analyzes the distribution patterns of transaction amounts, the fund closure rate, and temporal patterns; finally, graph embedding technology maps the entire community into a low-dimensional vector, thereby abstracting its overall characteristic profile. These multi-dimensional features collectively constitute the core judgment criteria for distinguishing between benign business cooperation and illegal transfer of benefits.

[0242] Risk assessment is accomplished through pattern recognition and machine learning models. The system trains a classifier using historical data, learning multi-dimensional feature patterns of known illegal and benign communities. Based on this, a general feature model of illegal transactions is abstracted, rather than being confined to specific connection paths and the number of companies. When the characteristics of a new community highly match the illegal pattern, even if its surface structure differs from known cases, the system can still provide accurate early warnings, thereby achieving intelligent and generalized discovery of mutated and new hidden related transaction behaviors.

[0243] As can be seen from the above description, the specific embodiments of the present invention provide a method for mining multi-object related transactions based on graph risk control, which has the following characteristics:

[0244] 1. Multi-dimensional massive data cleaning, processing and fusion mechanism;

[0245] 2. Data synchronization, used to synchronize the incremental financial business data to the big data ecosystem;

[0246] 3. An indicator calculation engine was designed to process data into indicators such as transaction relationships and transaction percentages;

[0247] 4. A graph risk control indicator calculation engine was designed to reorganize the cleaned data and the preliminarily processed indicators into a graph structure for storage and to perform graph risk control indicator modeling and calculation for models such as abnormal transaction structures.

[0248] 5. Incremental data capture, used to capture incremental financial business data;

[0249] 6. Incremental update, used in the big data ecosystem, responds to the incremental update request of the financial knowledge graph, and performs incremental update processing on the financial knowledge graph based on the incremental financial business data to obtain the updated indicator calculation results and knowledge graph;

[0250] 7. Related-party transaction identification and differentiation, used in the big data ecosystem for the updated financial knowledge graph.

[0251] 8. Identify and differentiate related-party transactions to determine the relevant corporate entities involved in non-compliant related-party transactions;

[0252] 9. An early warning module was designed to display and alert related transactions that exhibit irregularities.

[0253] As described above, this invention utilizes multi-source corporate tax data, tax-related business data, and judicial data, which, after integration, cleaning, and processing, construct a corporate relationship graph. This graph uses corporate equity structures, senior management information, and transaction relationships as nodes and edges, and employs graph algorithms to identify related-party transactions between companies. By analyzing the nature, scale, and frequency of related-party transactions, it determines whether there are risky behaviors such as improper transfer of benefits or financial fraud. For example, for companies that use complex equity structures and related-party relationships to transfer benefits, graph analysis can clearly display their related-party transaction paths and risk points, thereby identifying hidden related parties: penetrating equity, employment, and kinship relationships to discover the actual controller's transfer of benefits through multi-layered nested companies. Specifically, this invention has the following beneficial effects compared to existing technologies in the following four aspects:

[0254] 1. Scenario: Related-party transactions are divided into legitimate related-party transactions and illegal related-party transactions. This invention can identify related-party transactions involving nested shareholdings of group-level enterprises and multi-party related-party transactions;

[0255] 2. Technology: Graph databases are used to perform data mining on complex transaction structure models, enabling efficient identification and differentiation between legitimate and illegitimate transactions;

[0256] 3. Technical Architecture: It adopts a graph database based on the big data ecosystem and uses incremental update capture to reduce computational pressure and improve computational efficiency;

[0257] 4. Calculation method: Combining graph path algorithm and community discovery algorithm to discover related transaction patterns.

[0258] Based on the same inventive concept, this application also provides a graph-based risk control-based multi-object inter-relationship transaction mining device, which can be used to implement the method described in the above embodiments, as shown in the following embodiments. Since the principle of the graph-based risk control-based multi-object inter-relationship transaction mining device in solving the problem is similar to that of the graph-based risk control-based multi-object inter-relationship transaction mining method, the implementation of the graph-based risk control-based multi-object inter-relationship transaction mining device can refer to the implementation of the graph-based risk control-based multi-object inter-relationship transaction mining method, and repeated details will not be elaborated further. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0259] Embodiments of the present invention provide a specific implementation of a graph-based risk control-based multi-object inter-object association transaction mining device capable of implementing a graph-based risk control method for mining multi-object association transactions. See [link to specific implementation details]. Figure 9 A graph-based risk control-based multi-object inter-relationship transaction mining device specifically includes the following components:

[0260] The transaction object determination module 10 is used to determine a first object and at least one second object with which it has a transaction relationship and / or association relationship; wherein, the transaction relationship includes fund outflow and inflow information, transaction frequency, amount and type in derivative transaction relationships;

[0261] The entity relationship network establishment module 20 is used to establish an entity relationship network based on the first object, the at least one second object, and the corresponding association relationship.

[0262] The transaction relationship network establishment module 30 is used to establish a transaction relationship network based on the first object, the at least one second object, and the corresponding transaction relationship.

[0263] The related transaction mining module 40 is used to overlay the entity relationship network and the transaction relationship network into a multi-source network to generate a related transaction knowledge graph between the first object and the at least one second object, and to mine related transactions between the first object and the at least one second object through the related transaction knowledge graph.

[0264] In some embodiments of the present invention, see Figure 10 The transaction relationship network establishment module 30 includes:

[0265] The connection path lookup unit 30a is used to find the connection path between the node corresponding to the first object and the node corresponding to the at least one second object in the pre-generated event graph.

[0266] The spatial vector generation unit 30b is used to generate the spatial vector of the connection path; wherein the spatial vector is used to store the transaction identifier and transaction party information corresponding to the connection path;

[0267] The associated transaction path determination unit 30c is used to determine the associated transaction path between the node corresponding to the first object and the node corresponding to the at least one second object based on the spatial vector and the connection path.

[0268] The transaction relationship network establishment unit 30d is used to establish the transaction relationship network based on the associated transaction path. In some embodiments of the present invention, see... Figure 11 The transaction relationship network establishment unit 30d includes:

[0269] The multidimensional attribute feature extraction unit 30d1 is used to extract the multidimensional attribute features of the transaction corresponding to the spatial vector if the spatial vector does not contain the transaction identifier of the target transaction.

[0270] The association attribute judgment unit 30d2 is used to determine whether there is an association attribute between the first object and the at least one second object based on the multidimensional attribute features;

[0271] The transaction relationship network establishment subunit 30d3 is used to establish the transaction relationship network based on the associated transaction path between the first object and the at least one second object if there is an association attribute.

[0272] In some embodiments of the present invention, the multidimensional attribute features include: behavioral features of transactions between the first object and the at least one second object, fund flow features, specific business features, and network features of the entity relationship network.

[0273] In some embodiments of the present invention, see Figure 12 The related transaction mining module 40 includes:

[0274] The transaction feature extraction unit 40a is used to extract features from the related transaction knowledge graph to describe the transaction between the first object and the at least one second object; the features include static structure, dynamic behavior and graph embedding.

[0275] Feature profile extraction unit 40b is used to abstract a feature profile of the transaction between the first object and the at least one second object based on the features;

[0276] The first unit 40c for related transaction mining is used to compare the feature profile in the pre-generated related transaction model to mine the related transactions.

[0277] In some embodiments of the present invention, see Figure 13 The related transaction mining module 40 includes:

[0278] The transaction flow indicator extraction unit 40d is used to extract the transaction flow indicator between the first object and the at least one second object from the related transaction knowledge graph.

[0279] The second unit 40e for related transaction mining is used to compare the flow indicators in the pre-generated flow indicator model to mine the related transactions.

[0280] It should be noted that the multi-object related transaction mining method and apparatus based on graph risk control provided in this embodiment of the invention can be used in the financial field, or in any technical field other than the financial field. This embodiment of the invention does not limit the application field of the multi-object related transaction mining method and apparatus based on graph risk control.

[0281] The embodiments of this application also provide a specific implementation of an electronic device capable of implementing all steps in the graph-based risk control-based multi-object related transaction mining method described in the above embodiments. See [link to implementation details]. Figure 14 The electronic devices specifically include the following:

[0282] Processor 1201, memory 1202, communications interface 1203, and bus 1204;

[0283] The processor 1201, memory 1202, and communication interface 1203 communicate with each other via bus 1204; the communication interface 1203 is used to realize information transmission between server-side devices and user-side devices and other related devices.

[0284] The processor 1201 is used to call the computer program in the memory 1202. When the processor executes the computer program, it implements all the steps in the multi-object related transaction mining method based on graph risk control in the above embodiments. For example, when the processor executes the computer program, it implements the following steps:

[0285] Identify a first object and at least one second object with which it has a transaction relationship and / or association relationship; wherein the transaction relationship includes information on fund outflows and inflows, and the frequency, amount, and type of transactions in derivative transaction relationships;

[0286] Establish an entity relationship network based on the first object, the at least one second object, and the corresponding association relationships;

[0287] A transaction relationship network is established based on the first object, the at least one second object, and the corresponding transaction relationship;

[0288] The entity relationship network and the transaction relationship network are overlaid with a multi-source network to generate a knowledge graph of related transactions between the first object and the at least one second object, and the related transactions between the first object and the at least one second object are mined through the knowledge graph of related transactions.

[0289] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the graph-based risk control method for mining multi-object related transactions in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the graph-based risk control method for mining multi-object related transactions in the above embodiments. For example, when the processor executes the computer program, it implements the following steps:

[0290] Identify a first object and at least one second object with which it has a transaction relationship and / or association relationship; wherein the transaction relationship includes information on fund outflows and inflows, and the frequency, amount, and type of transactions in derivative transaction relationships;

[0291] Establish an entity relationship network based on the first object, the at least one second object, and the corresponding association relationships;

[0292] A transaction relationship network is established based on the first object, the at least one second object, and the corresponding transaction relationship;

[0293] The entity relationship network and the transaction relationship network are overlaid with a multi-source network to generate a knowledge graph of related transactions between the first object and the at least one second object, and the related transactions between the first object and the at least one second object are mined through the knowledge graph of related transactions.

[0294] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. In particular, hardware + program embodiments are relatively simple in description because they are fundamentally similar to method embodiments; relevant parts can be referred to the descriptions in the method embodiments.

[0295] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0296] While this application provides method operation steps as shown in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive labor. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only execution order. In actual device or user terminal product execution, the method can be executed in the order shown in the embodiments or drawings or in parallel (e.g., in a parallel processor or multi-threaded processing environment).

[0297] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing the embodiments of this specification, the functions of each module can be implemented in one or more software and / or hardware components, or a module that performs the same function can be implemented by a combination of multiple sub-modules or sub-units. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms.

[0298] Those skilled in the art will also know that, besides implementing the controller using purely computer-readable program code, the same functions can be achieved by logically programming the method steps, making the controller function as logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers (PLCs), and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the devices within it used to implement various functions can also be considered structures within that hardware component. Alternatively, the devices used to implement various functions can be considered as both software modules implementing the method and structures within a hardware component.

[0299] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0300] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0301] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, system embodiments are basically similar to method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. In the description of this specification, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the embodiments in this specification. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0302] The above description is merely an embodiment of the present specification and is not intended to limit the embodiments of the present specification. For those skilled in the art, various modifications and variations can be made to the embodiments of the present specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the embodiments of the present specification should be included within the scope of the claims of the embodiments of the present specification.

Claims

1. A method for mining multi-object related transactions based on graph risk control, characterized in that, include: Identify a first object and at least one second object with which it has a transaction relationship and / or association relationship; wherein the transaction relationship includes information on fund outflows and inflows, and the frequency, amount, and type of transactions in derivative transaction relationships; Establish an entity relationship network based on the first object, the at least one second object, and the corresponding association relationships; A transaction relationship network is established based on the first object, the at least one second object, and the corresponding transaction relationship; The entity relationship network and the transaction relationship network are overlaid with a multi-source network to generate a knowledge graph of related transactions between the first object and the at least one second object, and the related transactions between the first object and the at least one second object are mined through the knowledge graph of related transactions.

2. The method for mining inter-object related transactions according to claim 1, characterized in that, Establishing a transaction relationship network based on the first object, the at least one second object, and the corresponding transaction relationship includes: Find the connection path between the node corresponding to the first object and the node corresponding to the at least one second object in the pre-generated event graph; Generate a spatial vector for the connection path; wherein the spatial vector is used to store the transaction identifier and transaction party information corresponding to the connection path; The associated transaction path between the node corresponding to the first object and the node corresponding to the at least one second object is determined based on the spatial vector and the connection path. The transaction relationship network is established based on the associated transaction path.

3. The method for mining inter-object related transactions according to claim 2, characterized in that, Establishing the transaction relationship network based on the associated transaction path includes: If the spatial vector does not contain the transaction identifier of the target transaction, then the multidimensional attribute features of the transaction corresponding to the spatial vector are extracted; Based on the multidimensional attribute features, determine whether there is a related attribute between the first object and the at least one second object; If there is an association attribute, the transaction relationship network is established based on the associated transaction path between the first object and the at least one second object.

4. The method for mining inter-object related transactions according to claim 3, characterized in that, The multidimensional attribute features include: behavioral features of transactions between the first object and the at least one second object, fund flow features, specific business features, and network features of the entity relationship network.

5. The method for mining inter-object related transactions according to claim 1, characterized in that, Mining related transactions between the first object and the at least one second object using the related transaction knowledge graph includes: Features describing the transaction between the first object and the at least one second object are extracted from the related transaction knowledge graph; the features include static structure, dynamic behavior, and graph embedding. Based on the aforementioned features, a feature profile of the transaction between the first object and the at least one second object is abstracted; The feature profiles are compared in the pre-generated related transaction model to uncover the related transactions.

6. The method for mining multi-object related transactions according to any one of claims 1 to 4, characterized in that, Mining related transactions between the first object and the at least one second object using the related transaction knowledge graph includes: Extract transaction flow indicators between the first object and the at least one second object from the related transaction knowledge graph; The pre-generated transaction flow indicator model is used to compare the transaction flow indicators and identify the related transactions.

7. A multi-object association transaction mining device based on graph risk control, characterized in that, include: The transaction object determination module is used to determine a first object and at least one second object with which it has a transaction relationship and / or association relationship; wherein, the transaction relationship includes fund outflow and inflow information, transaction frequency, amount and type in derivative transaction relationships; An entity relationship network establishment module is used to establish an entity relationship network based on the first object, the at least one second object, and the corresponding association relationships. A transaction relationship network establishment module is used to establish a transaction relationship network based on the first object, the at least one second object, and the corresponding transaction relationship. The related transaction mining module is used to overlay the entity relationship network and the transaction relationship network into a multi-source network to generate a related transaction knowledge graph between the first object and the at least one second object, and to mine related transactions between the first object and the at least one second object through the related transaction knowledge graph.

8. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the steps of the multi-object association transaction mining method based on graph risk control as described in any one of claims 1 to 6.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the multi-object related transaction mining method based on graph risk control as described in any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the multi-object association transaction mining method based on graph risk control as described in any one of claims 1 to 6.

Images