Object recognition method, apparatus, electronic device, and storage medium

Abstract

The application relates to the computer technical field and provides an object identification method and device, electronic equipment and storage medium, which can be applied to a lending and borrowing scene. In view of a resource transfer process, the possibility of debt default of an inflow object of a multi-head type is relatively high, and resource inflow quantity and resource inflow frequency are important factors of the multi-head type. Therefore, at least one candidate inflow object of the multi-head type meeting preset inflow conditions can be screened out based on the resource inflow quantity and the resource inflow frequency, and the candidate inflow object is identified as a risk object when the candidate inflow object meets preset risk conditions. Through targeted analysis of the candidate inflow object of the multi-head type based on the preset risk conditions, the accuracy and efficiency of risk control identification are improved. Further, the risk object is taken as a seed object, more risk objects are mined from various business relationships, the breadth of the risk object is improved, the debt default risk is reduced, and the safety of an internet financial environment is improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to an object recognition method, apparatus, electronic device and storage medium. Background Technology

[0002] With the development of internet finance, more and more individuals or enterprises tend to borrow through online lending platforms. However, changes in the macroeconomic environment and industry policy adjustments in the financial market have a significant impact on borrowers. When borrowers have multiple loans at the same time, the probability of default increases if their repayment ability is insufficient. In particular, in unsecured credit scenarios, the default losses suffered by borrowers are even greater. Therefore, the group of people with multiple loans has become one of the important research subjects for identifying default risks in lending scenarios.

[0003] With the relevant technology, in the field of credit risk control identification, the identification of multiple borrowers can be carried out in, but is not limited to, the following two methods:

[0004] Method 1: Use query data provided by the central bank to identify groups with multiple borrowing.

[0005] When using method two, financial institutions, after obtaining authorization from borrowers, inquire about personal credit reports from the central bank, with the inquiry frequency limited to no more than once per quarter. This limitation by the central bank restricts the frequency and timing of inquiries, resulting in poor timeliness of the data. Financial institutions cannot promptly identify groups of borrowers with multiple debts who have defaulted, thus affecting the accuracy of risk identification for such groups. Furthermore, for unauthorized financial institutions or non-financial institutions (such as loan facilitation platforms), without access to data from the central bank, identifying groups of borrowers with multiple debts who have defaulted is even more difficult.

[0006] Method 2: Rely on the integrated data from multiple platforms to identify groups with multiple borrowing.

[0007] When using method one, credit data generally involves identity information and cannot be directly shared across domains, which increases the difficulty of identifying integrated data. Furthermore, the identification algorithm used is relatively simple, resulting in low accuracy and breadth in identifying multiple borrowers who have defaulted on their debts.

[0008] Therefore, risk identification for individuals with multiple borrowing has become an urgent issue to be addressed in lending scenarios. Summary of the Invention

[0009] This application provides an object recognition method, apparatus, electronic device, and storage medium to improve the accuracy of object recognition.

[0010] In a first aspect, embodiments of this application provide an object recognition method, the method comprising:

[0011] Obtain the first business dataset that matches the first resource transfer type;

[0012] Based on the resource flow of the first business dataset, at least one candidate inflow object is selected if the resource inflow amount and the number of resource inflows meet the preset inflow conditions.

[0013] For each of the candidate inflow objects, the following steps are performed: when a candidate inflow object meets the preset risk conditions, the candidate inflow object is used as a seed object, and at least one diffusion inflow object that is associated with the seed object and meets the preset inflow conditions and the preset risk conditions is selected from the various business relationships associated with the first resource transfer type in the pre-generated data.

[0014] Various sub-objects and each diffusion inflow object are considered as a risk object group.

[0015] Secondly, embodiments of this application provide an object recognition device, comprising:

[0016] The acquisition module is used to acquire the first business dataset that conforms to the first resource transfer type;

[0017] The filtering module is used to filter out at least one candidate inflow object whose resource inflow volume and resource inflow frequency meet preset inflow conditions based on the resource flow direction of the first business dataset.

[0018] The risk diffusion module is used to perform the following for each candidate inflow object: when a candidate inflow object meets the preset risk conditions, the candidate inflow object is used as a seed object, and at least one diffusion inflow object that is associated with the seed object and meets the preset inflow conditions and the preset risk conditions is selected from the multiple business relationships associated with the first resource transfer type in a pre-generated manner.

[0019] The identification module is used to classify various sub-objects and each diffusion inflow object as a risk object group.

[0020] Optionally, after obtaining the risk target group, the identification module is further used to:

[0021] For each risk object in the group of risk objects, the following is performed: Based on the first business dataset, a set of behavioral time sequences of a risk object is obtained, and each behavioral time sequence includes: multiple resource transfer operations performed by the risk object in a time sequence under a resource flow.

[0022] Based on the behavioral time sequence set of each risk object, at least one target risk object in the group of risk objects is obtained.

[0023] Optionally, the identification module is specifically used for:

[0024] Based on the first business dataset, target inflow behavior data for each risk object under the resource inflow direction are obtained, and the target inflow behavior data are arranged in chronological order to obtain an inflow behavior time series; and,

[0025] Based on the first business dataset, obtain the target outflow behavior data of each risk object under the resource outflow direction, and arrange the target outflow behavior data in chronological order to obtain the outflow behavior time sequence;

[0026] The inflow and outflow data of each target are mixed and arranged in chronological order to obtain a time sequence of the inflow and outflow behavior of a risk object under the resource inflow and outflow direction.

[0027] Optionally, the identification module is specifically used for:

[0028] Based on the first business dataset, obtain the initial inflow behavior data of a risk object under the resource inflow direction, and for each initial inflow behavior data, perform the following: extract the inflow information corresponding to each preset dimension from the initial inflow behavior data, format the inflow information, and obtain a target inflow behavior data.

[0029] Based on the first business dataset, obtain the initial outflow behavior data of a risk object under the resource outflow direction, and for each initial outflow behavior data, perform the following: extract the outflow information corresponding to each preset dimension from the initial outflow behavior data, format the outflow information, and obtain a target outflow behavior data.

[0030] Optionally, the filtering module is specifically used for:

[0031] Based on the resource flow direction of the first business dataset, a subset of resource inflow data in the resource inflow direction and a subset of resource outflow data in the resource outflow direction are obtained.

[0032] Based on the resource inflow data subset and the resource outflow data subset, multiple candidate outflow objects under the first resource transfer type are obtained;

[0033] Based on the first business dataset, obtain the attribute information of each of the multiple candidate outflow objects, and based on the attribute information, filter out at least one target outflow object from the multiple candidate outflow objects;

[0034] In the business relationship, at least one candidate inflow object is selected that meets the preset inflow conditions in terms of resource inflow amount and resource inflow frequency with the at least one target outflow object.

[0035] Optionally, the filtering module is specifically used for:

[0036] Obtain a second business dataset that conforms to the first resource transfer type; wherein the first business dataset is a subset of the second business dataset, and the time period corresponding to the second business dataset is earlier than the time period corresponding to the first business dataset;

[0037] Based on the second business dataset, various business relationships are established.

[0038] Optionally, the various business relationships include object relationships and object behavior relationships, and the filtering module is specifically used for:

[0039] Based on the second business dataset, for each inflow object whose resource inflow volume and resource inflow frequency meet the preset inflow conditions, a first object relationship is established between each inflow object and each outflow object;

[0040] Based on the multiple inflow objects that are simultaneously associated with the same thing among the aforementioned inflow objects, a second object relationship is established among the multiple inflow objects;

[0041] Based on the selection of at least one abnormal inflow object that meets the preset risk conditions from the inflow objects, a third object relationship is established between the at least one abnormal inflow object and the outflow objects.

[0042] Based on the resource entry and exit behaviors associated with each inflow object, an object behavior relationship is established between each inflow object and its corresponding resource entry and exit behaviors.

[0043] Optionally, the preset risk condition includes at least one of the following situations:

[0044] There is an interactive behavior between the candidate inflow object and the intermediate object, and the intermediate object is used to connect the candidate inflow object and the target outflow object.

[0045] For each of the aforementioned candidate inflow objects, there exists at least one negative evaluation;

[0046] The candidate inflow object exhibits at least one instance of abnormal behavior;

[0047] The risk level of a candidate inflow object reaches a set threshold.

[0048] Optionally, the acquisition module is specifically used for:

[0049] Acquire the pipeline data pool generated by resource transfer;

[0050] Based on the flow information of each flow data in the flow data pool, the flow data pool is classified to obtain the flow data sets corresponding to each resource transfer type; the flow information includes at least one of the flow method, flow direction, and flow remarks.

[0051] The pipeline dataset corresponding to the first resource transfer type is used as the first business dataset.

[0052] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the steps of any of the above-described object recognition methods.

[0053] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions thereon, which, when executed by an electronic device, implement the steps of any of the above-described object recognition methods.

[0054] Fifthly, embodiments of this application provide a computer program product comprising a computer program that, when executed by an electronic device, implements the steps of any of the above-described object recognition methods.

[0055] The beneficial effects of the object identification method, apparatus, electronic device, and storage medium provided in this application are as follows:

[0056] Resource transfer involves resource flow. The recipients of resources in the inflow direction are required to repay debts. However, multiple-entry recipients have a higher probability of debt default, necessitating risk control identification. Resource inflow volume and frequency are crucial factors for multiple-entry recipients. Therefore, preset inflow conditions can be set based on these factors. Using the resource flow direction of the first business dataset conforming to the first resource transfer type, at least one candidate inflow recipient meeting the preset inflow conditions can be initially screened. When a selected candidate inflow recipient meets preset risk conditions, it indicates a high risk of debt default, thus identifying it as a risky object. Targeted analysis based on the preset risk conditions for the selected candidate inflow recipient improves the accuracy and efficiency of risky object identification. Furthermore, the time period corresponding to the first business dataset can be selected as needed, ensuring the timeliness of the identification results.

[0057] On the other hand, since the first resource transfer type is associated with multiple business relationships, risk objects that meet the preset risk conditions can be used as seed objects. From the multiple business relationships, at least one diffusion inflow object that meets the preset inflow conditions and preset risk conditions can be mined, thereby identifying more risk objects that may default on debts. This reduces the debt default risk involved in the first resource transfer type and ensures the security of the Internet finance environment.

[0058] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 An application scenario architecture diagram provided for an embodiment of this application;

[0061] Figure 2 A system architecture diagram of the object recognition method provided in the embodiments of this application;

[0062] Figure 3 A flowchart illustrating an object recognition method provided in this application embodiment;

[0063] Figure 4 A flowchart illustrating a method for identifying candidate inflow objects provided in this application embodiment;

[0064] Figure 5 A schematic diagram illustrating various business relationships provided for embodiments of this application;

[0065] Figure 6 Examples of various business relationships provided in the embodiments of this application;

[0066] Figure 7 This is a schematic diagram illustrating the risk object diffusion process provided in an embodiment of this application.

[0067] Figure 8 A flowchart illustrating another object recognition method provided in this application embodiment;

[0068] Figure 9AA schematic diagram of a behavioral timing sequence set provided in an embodiment of this application;

[0069] Figure 9B A schematic diagram illustrating the object recognition process provided in this application embodiment;

[0070] Figure 10 A structural diagram of a network model provided in an embodiment of this application;

[0071] Figure 11 A schematic diagram illustrating the content of the basic image provided in the embodiments of this application;

[0072] Figure 12 A technical framework diagram for object recognition provided in the embodiments of this application;

[0073] Figure 13 This is a structural diagram of an object recognition device provided in an embodiment of this application;

[0074] Figure 14 This is a structural diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0075] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this application. Obviously, the described embodiments are only some embodiments of the technical solutions of this application, and not all embodiments. Based on the embodiments recorded in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the technical solutions of this application.

[0076] For ease of understanding, the terms used in the embodiments of this application are explained below.

[0077] Multiple borrowing: The same borrower has loan activities with multiple credit institutions at the same time, or the same borrower has multiple loan activities with one credit institution at the same time.

[0078] Multiple credit group: a collection of borrowers who simultaneously engage in lending activities with multiple credit institutions, or who simultaneously engage in multiple lending activities with a single credit institution.

[0079] Investee: The party receiving the resources, also known as the borrower in the credit scenario, is an enterprise or individual that obtains financial support from the lender by using its own credit and assets as collateral or by seeking third-party guarantees. As the recipient of the funds, it has the obligation to repay the principal and interest of the loan as agreed.

[0080] Outflowing party: The party from which resources flow out, also known as the lender in a credit scenario, is the person or lending institution that uses credit funds or its own funds to issue loans to borrowers. As the provider of funds, it has the rights and responsibilities of verifying the borrower's qualifications, setting loan conditions, issuing and recovering funds.

[0081] Resource flow: This includes the direction of resource inflow, the direction of resource outflow, and the direction of resource entry and exit. The direction of resource inflow is relative to the borrowing behavior, the direction of resource outflow is relative to the repayment behavior, and the direction of resource entry and exit is based on the borrowing behavior and the repayment behavior.

[0082] The design concept of the embodiments of this application is summarized below.

[0083] As a common type of resource transfer scenario, credit is prone to default when borrowers take on multiple loans simultaneously, which can lead to serious economic losses for lenders. Therefore, it is necessary to identify multiple borrowers in advance and take corresponding preventive measures to control the occurrence of debt defaults.

[0084] In the field of credit risk control identification, the scheme that uses query data provided by the central bank for identification requires the borrower's authorization and has poor effectiveness. At present, more of the identification relies on the integration of data from multiple parties or institutions, but such schemes have problems such as data usage standards, link security and high costs.

[0085] In view of this, embodiments of this application provide an object identification method, apparatus, electronic device, and storage medium, which can be applied to risk identification of multiple borrowing groups in credit scenarios. During resource transfer, the inflow object in the resource inflow direction, as the resource recipient, needs to repay debts. However, multiple borrowing inflow objects have a higher probability of debt default, thus requiring risk control identification of these objects. Resource inflow volume and frequency are important factors for multiple borrowing; therefore, preset inflow conditions can be set for multiple borrowing inflow objects based on these factors. Thus, based on the resource flow direction of the first business dataset conforming to the first resource transfer type, at least one candidate inflow object of the multiple borrowing group that meets the preset inflow conditions is initially screened. When the screened candidate inflow object meets the preset risk conditions, it indicates that the candidate inflow object has a high risk of debt default; therefore, the candidate inflow object is identified as a risk object. Targeted analysis is performed on the at least one candidate inflow object of the multiple borrowing group based on the preset risk conditions, improving the accuracy and efficiency of risk object identification. Furthermore, the first business dataset within the corresponding time period can be obtained at any time according to actual needs for risk object identification, resulting in stronger real-time performance.

[0086] On the other hand, since the first resource transfer type is associated with multiple business relationships, risk objects that meet the preset risk conditions can be used as seed objects. From the multiple business relationships, at least one diffusion inflow object that meets the preset inflow conditions and preset risk conditions can be mined, thereby identifying more risk objects that may default on debts. This reduces the debt default risk involved in the first resource transfer type and ensures the security of the Internet finance environment.

[0087] It is understood that, in the specific embodiments of this application, the transaction data and first business data involved have been authorized or agreed to by customers who have signed up on the platform of financial institutions when applied to the methods or products of the following embodiments of this application, and the collection, use and processing of the relevant data comply with the laws, regulations and standards of the relevant countries and regions.

[0088] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0089] It should be noted that the resources in this application embodiment include, but are not limited to, funds, stocks, funds, securities, gold bars, game coins, game equipment, etc.

[0090] It should be noted that the types of loans in this application embodiment are not restricted, including but not limited to consumer loans, cash loans, and mortgage loans.

[0091] It should be noted that the above lending scenarios are only shown for the purpose of understanding the spirit and principles of this application. The implementation methods of this application are not limited to specific application scenarios and can be applied to various resource transfer control scenarios. The object recognition method provided by the exemplary embodiments of this application is described below with reference to the accompanying drawings.

[0092] See Figure 1 The application scenario architecture diagram provided for the embodiments of this application includes at least a terminal device 100, a server 200, and a database 300.

[0093] The terminal device 100 is equipped with a resource trading application, including but not limited to resource transfer clients, mini-programs, browsers, etc. The object accesses the server 200 through the resource trading application to send resource requests to the server 200.

[0094] Server 200 is connected to database 300 on the Internet finance platform. Database 300 stores the transaction data of resource transfers of various objects on the Internet finance platform.

[0095] Server 200 retrieves transaction data from database 300 based on the resource request sent by terminal device 100, identifies whether an entity faces debt default risk based on the transaction data, and takes corresponding response measures based on the identification results. For example, it rejects the resource request of an identified risky entity and returns the requested resources to an identified non-risky entity.

[0096] Optionally, in some embodiments, after receiving a resource request from the terminal device 100, the server 200 may send an identification request to the server 400 on the Internet finance platform, so that the server 400 can identify the risk of the object and return the risk identification result to the server 200.

[0097] Optionally, in some embodiments, in order to more accurately and comprehensively identify more risky users in the Internet finance platform, the database 300 may also store transaction data obtained from other institutional platforms.

[0098] Optionally, terminal devices include, but are not limited to, mobile phones, computers, and financial devices. Servers can be cloud servers that provide basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms, or they can be server clusters.

[0099] The communication methods between terminal devices and servers, and between servers and databases, can be either wired communication technologies, such as connecting via network cables or serial cables, or wireless communication technologies, such as Bluetooth or Wi-Fi. There are no specific restrictions.

[0100] It is important to note that Figure 1 This is merely an example of an application scenario architecture and is not intended to be a limiting requirement of the embodiments of this application.

[0101] See Figure 2 The above is a system architecture diagram of the object recognition method provided in the embodiments of this application, which mainly includes four parts: multi-head customer group recognition, multi-dimensional risk recognition, abnormal behavior recognition, and behavior audit feedback optimization.

[0102] In the multi-headed customer group identification section, the resource inflow data and resource outflow data of multi-headed objects during the resource transfer process are mined. Through clustering and summarization, the outflowing object group that meets the first resource transfer type is mined. Then, based on the outflowing object group, the inflowing object group of multi-headed objects that have resource transfer relationship with it is inferred in reverse.

[0103] Taking the lending scenario as an example, in the identification of multiple borrowers, the inflow data can be loan data, the outflow data can be repayment data, the outflow target group can be the borrower group, including but not limited to lending institutions, lending service institutions, merchants, etc., and the inflow target group is the multiple borrower group, including but not limited to enterprises, individuals, etc.

[0104] In the multi-dimensional risk identification section, different risk conditions are set from multiple dimensions to screen out potential debt default risk groups from the multiple inflow target group. These risk conditions include, but are not limited to: interaction with intermediary entities, at least one negative evaluation, at least one instance of abnormal behavior, and risk levels reaching a set threshold.

[0105] Taking lending scenarios as an example, in the multi-dimensional risk identification section, interactive behaviors can be multiple credit inquiries, negative evaluations can be complaints about borrowing and repayment, abnormal behaviors can be behaviors that lead to debt default, and risk levels can be the probability of default.

[0106] In the abnormal behavior identification section, considering that the risk of debt default among the risky groups obtained from the multi-dimensional risk identification section is not very high, a deep time series model is used to make more accurate judgments on abnormal behavior by mining the time series of inflow behavior under the direction of resource inflow, the time series of outflow behavior under the direction of resource outflow, and the time series of inflow and outflow behavior under the direction of resource inflow and outflow. This is to accurately identify the low-quality groups of risky groups that are rapidly shifting and continuously hungry for resources.

[0107] Taking the lending scenario as an example, in the abnormal behavior identification part, the behavioral time sequence under multiple resource flows includes: borrowing behavior time sequence, repayment behavior time sequence, and borrowing and repayment behavior time sequence.

[0108] In the behavior audit feedback optimization section, the accuracy and reliability of the identification results are ensured by auditing the identification results of the objects.

[0109] Taking the lending scenario as an example, in the behavioral audit feedback optimization section, the audit content includes, but is not limited to: feedback on default performance, feedback on complaint data, and feedback on experimental comparison results.

[0110] Optionally, in some embodiments, the object recognition process may not necessarily include all four parts mentioned above. For example, in some scenarios where the recognition accuracy requirement is not high, the risk object group identified by the multi-dimensional risk recognition part can be directly output as the result without subsequent abnormal behavior recognition.

[0111] Optionally, in some embodiments, the above-mentioned behavior audit feedback optimization section is an optional part of the object recognition process, which can be flexibly added or removed according to the actual business scenario.

[0112] Optionally, in some embodiments, the three components of multi-headed customer group identification, multi-dimensional risk identification, and abnormal behavior identification can be implemented uniformly through a large model, or they can be implemented separately through multiple models. When implemented uniformly through a large model, the behavior audit feedback optimization part can iteratively optimize the large model based on the audit results; when implemented separately through multiple models, the audit results of the behavior audit feedback optimization part are applied to each model for iterative optimization.

[0113] See Figure 3 The present application provides a flowchart of an object recognition method, which mainly includes the following steps:

[0114] S301: The server obtains the first business dataset that matches the first resource transfer type.

[0115] In some embodiments, the massive amounts of transaction data generated by resource transfers within an internet finance platform can be stored in a database in the form of logs. The log files in the database may contain shared data provided by one or more outgoing objects.

[0116] Optionally, massive amounts of transaction data can be stored through a big data platform.

[0117] In some embodiments, the transaction data in the database may involve multiple types of resource transfers, among which there may be resource transfer types that do not involve debt defaults and do not require risk control identification.

[0118] Given the various types of resource transfers, those involving debt defaults are designated as the first resource transfer type. Therefore, it is necessary to sift through massive amounts of transaction data to identify the first business dataset that conforms to this first resource transfer type for object identification.

[0119] In some embodiments, the process of obtaining the first business dataset includes the following steps:

[0120] First, acquire the pipeline data pool generated by resource transfer.

[0121] The pipeline data pool includes pipeline data generated from various resource transfers. This pipeline data records object attributes and behavior information in log format.

[0122] For example, object attributes may include: account, name, remarks, customer concentration, etc., and object behaviors may include: borrowing, repayment, number of borrowings, number of repayments, amount, transaction time, etc.

[0123] Taking a lending scenario as an example, the transaction data is as follows: Credit card account X borrowed 30,000 yuan from account Y for the first time on December 4, 2024, and the transaction note is "loan"; another example is: Credit card account X repaid 30,000 yuan to account Y for the first time on January 1, 2025, and the name of account Y is a certain credit institution.

[0124] Then, based on the flow information of each flow data in the flow data pool, the flow data pool is classified to obtain the flow data sets corresponding to each resource transfer type.

[0125] The circulation information includes at least one of the following: circulation method, circulation direction, and circulation remarks.

[0126] Optional transfer methods include, but are not limited to, debit card transactions, savings card transactions, and credit card transactions; transfer directions include, but are not limited to, loans and repayments; transfer remarks include, shopping consumption, utility bill payments, interest balances, and credit.

[0127] Finally, the transaction dataset corresponding to the first resource transfer type is used as the first business dataset.

[0128] Taking credit as the first type of resource transfer as an example, the transaction data is as follows: Account Y borrowed 30,000 yuan from credit card account X for the first time on December 4, 2024, and the transaction note is "loan". Since the transfer method is credit card transaction, the transfer direction is credit card loan, and the transfer note is "credit", this transaction data is regarded as the first business data.

[0129] For example, the transaction data is as follows: Credit card account X made its first repayment of 30,000 yuan to account Y on January 1, 2025. The name corresponding to account Y is a certain credit institution. Since the transaction method is credit card transaction, the direction of the transaction is credit card repayment, and the name of the receiving institution is a certain credit institution, this transaction data is also regarded as the first business data.

[0130] Considering that not all resource transfer types in the massive flow data pool carry debt default risk, the first business dataset of the first resource transfer type with debt default risk is selected from the flow data pool based on the flow information. This allows for targeted risk control identification, reduces the workload of subsequent data processing, and improves the identification efficiency of multiple risk objects.

[0131] It should be noted that the length of the time period corresponding to the first business dataset can be flexibly set according to the business's real-time requirements, and this application embodiment does not impose any restrictive requirements.

[0132] For example, the first business dataset can be business data generated within one day that conforms to the first resource transfer type, or the first business dataset can be business data generated within one week that conforms to the first resource transfer type.

[0133] In some embodiments, the first business dataset may be acquired periodically or non-periodically.

[0134] It should be noted that the embodiments of this application do not impose any restrictive requirements on the period size of the first business dataset. It can be flexibly set according to actual business needs. For example, in order to ensure the real-time performance of recognition, the period of the first business dataset can be set to one day, one week, etc. For businesses with low real-time requirements, the period can be set to one month, one year, etc.

[0135] S302: Based on the resource flow of the first business dataset, the server filters out at least one candidate inflow object whose resource inflow volume and resource inflow frequency meet the preset inflow conditions.

[0136] In some embodiments, the resource flow direction includes the resource inflow direction, the resource outflow direction, and the resource entry and exit direction.

[0137] Taking the credit scenario as an example, the direction of resource inflow is the direction of borrowing, the direction of resource outflow is the direction of repayment, and the direction of resource use is the direction of borrowing and repayment.

[0138] In some embodiments, the borrower's direction can be determined through information such as the borrower's account and transaction notes in the business data.

[0139] For example, if the loan account in the business data is credit card X and the transaction notes are loan-related, it can be determined that the business data is related to a loan.

[0140] In some embodiments, the repayment direction can be determined through information such as the receiving account, the name of the receiving party, the concentration of the receiving party's customer base, and transaction remarks, which are aggregated from business data.

[0141] For example, if the receiving account in the business data is Y credit card and the recipient's name is Y lending institution, the business data can be identified as the repayment direction.

[0142] For example, if the receiving account in the business data is Y credit card, the recipient's name is Y credit institution, and the credit group accounts for 10% of its service population, then the business data can be identified as the repayment direction.

[0143] In some embodiments, the direction of borrowing and repayment can be determined through information such as the borrowing account, receiving account, recipient name, amount, and transaction remarks.

[0144] For example, if business data 1 shows a loan institution paying 20,000 yuan to credit card X, and business data 2 shows credit card X paying 21,000 yuan to credit institution Y's Y receiving account, then business data 1 and business data 2 are determined to be the direction of loan repayment.

[0145] Based on resource flow, the first business dataset can be divided into a resource inflow data subset corresponding to the resource inflow direction and a resource outflow data subset corresponding to the resource outflow direction. Combining the resource inflow data subset and the resource outflow data subset yields the resource inflow and outflow data subset corresponding to the resource inflow and outflow directions.

[0146] Taking the lending scenario as an example, resource inflow data is the business data corresponding to the borrowing direction, resource outflow data is the business data corresponding to the repayment direction, and resource inflow and outflow data is the business data corresponding to the borrowing and repayment direction.

[0147] After obtaining the resource flow direction, the inflow and outflow objects under the first resource transfer type can be obtained based on the business data under the corresponding flow direction.

[0148] Taking a lending scenario as an example, the inflow target is the borrower, and the outflow target is the lender.

[0149] Because entities with multiple resource inflows have a higher probability of defaulting on their debts, it is necessary to identify such inflow entities as candidate inflow entities for more efficient risk control. These multiple inflow entities are defined as those that simultaneously engage in transactions with multiple outflow entities, or have multiple transactions with a single outflow entity. Considering the higher risk of default for multiple inflow entities, and given that resource inflow volume and frequency are key factors in determining whether an entity is multiple, inflow conditions corresponding to multiple inflows can be pre-defined based on these factors.

[0150] Optionally, the preset inflow conditions are that the amount of resource inflow exceeds a preset value and the number of resource inflows exceeds a preset number. This is because a small number of multiple resource transfers make it less likely that the outflowing objects of the multi-head class will be unable to repay.

[0151] In some embodiments, the resource inflow can be the amount of a single resource inflow.

[0152] Taking a lending scenario as an example, suppose credit institution B borrows 30,000 yuan from borrower A, credit institution D borrows 50,000 yuan from borrower A, and credit institution E borrows 20,000 yuan from borrower A. If borrower A has taken out more than two loans, and borrower A has two single loans exceeding 20,000 yuan, then borrower A meets the preset inflow conditions, that is, borrower A is a multi-borrower.

[0153] In some embodiments, the resource inflow can be the sum of the amounts of multiple resource inflows.

[0154] Taking a lending scenario as an example, suppose credit institution B borrows 30,000 yuan from borrower A, credit institution D borrows 50,000 yuan from borrower A, and credit institution E borrows 20,000 yuan from borrower A. If borrower A has taken out more than two loans and the total amount of borrowing by borrower A exceeds 90,000 yuan, then borrower A is determined to meet the preset inflow conditions, that is, borrower A is a multi-borrower.

[0155] It should be noted that the preset value corresponding to the resource inflow and the preset number of resource inflows can be set according to actual needs, and this application embodiment does not make a limiting description.

[0156] During resource transfer, the resources of an outflowing object can flow to different inflowing objects, and the resources of different inflowing objects can come from the same outflowing object. Compared with inflowing objects, outflowing objects are fewer in number and their category information is more obvious.

[0157] Taking a credit scenario as an example, suppose credit institution B borrows money from object A and enterprise C respectively, and credit institution E borrows money from object F and enterprise G respectively. The outflow objects are all credit institutions and there are 2 of them. The outflow objects come from different enterprises and individuals and there are 4 of them.

[0158] In some embodiments, since outflow objects are more obvious in terms of quantity and category information, outflow objects can be mined first from the first business dataset, and then candidate inflow objects of the multi-head class can be obtained in reverse.

[0159] See Figure 4 The process of obtaining candidate inflow objects for multi-headed classes mainly includes the following steps:

[0160] S3021: Based on the resource flow direction of the first business dataset, obtain the resource inflow data subset under the resource inflow direction and the resource outflow data subset under the resource outflow direction.

[0161] The resource flow includes resource inflow direction, resource outflow direction, and resource entry / exit direction. The resource inflow data subset corresponding to the resource inflow direction and the resource outflow data subset corresponding to the resource outflow direction can be combined to obtain the resource entry / exit dataset corresponding to the resource outflow direction. Therefore, by using the resource outflow data subsets, all inflow objects and all outflow objects under the first resource transfer type can be obtained.

[0162] Each resource flow involves transactions between two parties, and the identity of the same object varies across different resource flow subsets.

[0163] Taking lending scenarios as an example, such as Figure 5As shown, the direction of resource inflow is the direction of borrowing, the direction of resource outflow is the direction of repayment, and the direction of resource transfer is the direction of borrowing and repayment. In the loan data subset under the borrowing direction, the lender is the payer, while in the repayment data subset under the repayment direction, the lender is the payee.

[0164] S3022: Based on the resource inflow data subset and the resource outflow data subset, obtain multiple candidate outflow objects under the first resource transfer type.

[0165] During resource transfer, the same object may appear in data from different flows. Therefore, by aggregating the same object with different identities in the resource inflow data subset and the resource outflow data subset, multiple candidate outflow objects under the first resource transfer type can be identified.

[0166] For example, in a lending scenario, the lender acts as the payer in the loan data subset under the borrowing direction and as the payee in the repayment data subset under the repayment direction. Therefore, by aggregating the payers and payees in the loan data subset and the repayment data subset, multiple candidate lenders for credit categories can be obtained.

[0167] S3023: Based on the first business dataset, obtain the attribute information of each of the multiple candidate outflow objects, and based on the attribute information, select at least one target outflow object from the multiple candidate outflow objects.

[0168] To improve the identification accuracy of candidate inflow objects in multi-headed categories, the accuracy of outflow objects under the first resource transfer type can be improved. Therefore, after obtaining multiple candidate outflow objects, the type of the candidate outflow objects can be further verified based on the attribute information of each candidate outflow object, thereby filtering out at least one target outflow object that conforms to the first resource transfer type.

[0169] Optionally, the attribute information includes, but is not limited to, the registration type, name, and service content of the outgoing object.

[0170] Taking the first transfer type as a loan as an example, suppose company C is a building materials company. Candidate outflow object 1 pays company C 10,000 yuan, while candidate outflow object 2 pays company C 20,000 yuan. The name of candidate outflow object 1 is X Credit Institution, and the name of candidate outflow object 2 is X Decoration Company. This indicates that the 10,000 yuan paid by candidate outflow object 1 to company C is a loan, while the 20,000 yuan paid by candidate outflow object 2 to company C is payment for goods. Therefore, candidate outflow object 1 is regarded as the target outflow object of the credit type.

[0171] S3024: In business relationships, filter out at least one candidate inflow object that meets the preset inflow conditions in terms of resource inflow amount and resource inflow frequency with at least one target outflow object.

[0172] In some embodiments, business relationships may be pre-generated based on a second business dataset that conforms to a first resource transfer type.

[0173] The first business dataset is a subset of the second business dataset. The second business dataset has a larger data volume than the first business dataset, and the time period corresponding to the second business dataset is earlier than that corresponding to the first business dataset. Therefore, the first business dataset is also called the real-time business dataset, and the second business dataset is also called the offline business dataset.

[0174] For example, the first business dataset consists of 10,000 business data entries generated by resource transfers that conform to the first resource transfer type within the past month, and the second business data subset consists of 100,000 business data entries generated by resource transfers that conform to the first resource transfer type within the past year.

[0175] Because the second business dataset has a large amount of data, it is used to establish multiple business relationships, which effectively improves the breadth and richness of business relationships.

[0176] In some embodiments, the multiple business relationships associated with the first resource transfer type include object relationships and object behavior relationships, such as... Figure 5 As shown.

[0177] Optionally, the object relationship includes the relationship between the inflow object and the outflow object, denoted as the first object relationship.

[0178] In specific implementation, the process of establishing the first object relationship includes: based on the second business data set, for each inflow object whose resource inflow volume and resource inflow frequency meet the preset inflow conditions, establishing the first object relationship between each inflow object and each outflow object. The outflow objects can be determined according to the resource flow direction.

[0179] Taking credit scenarios as an example, in Figure 5 If the inflow object of a multi-borrower class that meets the preset inflow conditions is a multi-borrower and the outflow object is a lender, then the first object relationship is the relationship between the multi-borrower and the lender.

[0180] Optionally, object relations include the relationships between inflowing objects, denoted as second object relations.

[0181] In practice, the process of establishing the second object relationship includes: establishing a second object relationship between multiple inflow objects that are simultaneously associated with the same thing.

[0182] Taking credit scenarios as an example, in Figure 5 If the inflow target is multiple borrowers, then the second target relationship is the relationship between multiple borrowers.

[0183] In lending scenarios, the same thing includes, but is not limited to, credit institutions, credit-related institutions (such as credit consulting agencies), and equipment of credit institutions.

[0184] For example, if object A and enterprise C are multiple inflow objects, and equipment 1 of credit institution B is used by object A and enterprise C for repayment operations, then the second object relationship includes the relationship between object A and enterprise C.

[0185] For example, if object A and company C are multiple inflow objects, and credit institution B simultaneously provides loans to object A and company C, then the second object relationship includes the relationship between object A and company C.

[0186] Optionally, object relationships include the relationship between abnormal inflow objects and outflow objects, denoted as the third object relationship.

[0187] In practice, the process of establishing the third object relationship includes: establishing a third object relationship between at least one abnormal inflow object that meets the preset risk conditions and selected from each inflow object.

[0188] When the amount of resources flowing into an object reaches a preset value, and the number of times resources flow into an object reaches a preset number, it indicates that the object is a multi-headed object. Furthermore, when a multi-headed object meets a preset risk condition, it can be further identified as an abnormal multi-headed object.

[0189] Taking credit scenarios as an example, in Figure 5 If the abnormal inflow target is an abnormally multiple borrower, then the third object relationship is the relationship between the abnormally multiple borrower and the lender.

[0190] The relevant descriptions of the preset risk conditions are provided in the following embodiments and will not be repeated here.

[0191] Optionally, object behavior relationships include the relationship between inflow objects and resource inflow / outflow behaviors.

[0192] In practice, the process of establishing object behavior relationships includes: establishing object behavior relationships between each inflow object and its corresponding resource inflow behavior based on the resource inflow volume and the number of resource inflows meeting the preset inflow conditions.

[0193] Taking credit scenarios as an example, in Figure 5 The inflow and outflow of resources in China includes borrowing and repayment activities. Therefore, the relationship between the objects of these activities is the relationship between multiple borrowers and borrowing activities, as well as the relationship between multiple borrowers and repayment activities.

[0194] For example, suppose credit institution B borrows 30,000 yuan from subject A, and subject A repays 32,000 yuan to credit institution B. Then the relationship between the subjects includes subject A borrowing 30,000 yuan and subject A repaying 32,000 yuan.

[0195] It should be noted that this application does not impose any restrictive requirements on the amount of loan repayment in a lending scenario. Generally, the repayment amount is greater than or equal to the loan amount because in a lending scenario, the borrower needs to repay not only the principal to the lender but also interest.

[0196] In the embodiments of this application, multiple business relationships are established between various business objects associated with the first resource transfer type, so that more risk objects that may default on debts under the first resource transfer type can be identified based on these multiple business relationships, thereby improving the breadth of risk identification.

[0197] In some embodiments, the business relationship may also be pre-generated based on a first business dataset that conforms to a first resource transfer type.

[0198] In practice, multiple real-time business relationships associated with the first resource transfer type are extracted from the first business dataset, and the corresponding historical business relationships obtained in the previous identification process are supplemented based on these multiple real-time business relationships, thereby obtaining a wider and richer range of business relationships to improve the ability to subsequently mine more multi-head risk objects based on these multiple business relationships.

[0199] In some embodiments, to clearly illustrate the various business relationships described above, these relationships can be represented in the form of a graph.

[0200] Optionally, multiple business relationships can be represented by a single graph or by multiple separate graphs.

[0201] In lending scenarios, taking a graph to represent multiple business relationships as an example, such as... Figure 6 As shown, the nodes in the graph represent objects, including multiple borrowers, abnormal multiple borrowers, and lenders. The edges represent object relationships, and the arrows indicate borrowing and repayment behaviors.

[0202] Since the first business dataset is a subset of the second business dataset, and the various business relationships established based on the second business dataset include the first object relationship between inflow objects and outflow objects of the multi-head class, as well as the object behavior relationship between inflow objects of the multi-head class and resource inflow and outflow behaviors, at least one candidate inflow object in the first business dataset that has resource exchanges with at least one target outflow object and meets the preset inflow conditions can be selected through the first business relationship and object behavior relationship.

[0203] During resource transfer, since the resources of an outflowing object can flow to different inflowing objects, and the resources of different inflowing objects can come from the same outflowing object, the outflowing objects are fewer in number and their category information is more obvious compared to the inflowing objects. Therefore, when mining candidate inflowing objects of multiple categories, the target outflowing object that conforms to the first resource transfer type is first determined from the first business dataset. Then, candidate inflowing objects of multiple categories that have resource exchanges with the target outflowing object are obtained in reverse. This reduces the difficulty of identifying candidate inflowing objects of multiple categories while improving the accuracy and efficiency of identifying candidate inflowing objects of multiple categories.

[0204] S303: For each candidate inflow object, the server executes the following: When a candidate inflow object meets the preset risk conditions, the candidate inflow object is used as a seed object, and at least one diffusion inflow object associated with the seed object and meeting the preset inflow conditions is selected from multiple business relationships.

[0205] Not every inflow of multiple-entry entities will default on its debt. To improve the accuracy of risk control identification, risky entities with abnormal transactions can be further identified from multiple candidate inflow entities based on preset risk conditions.

[0206] For at least one candidate inflow object obtained, targeted analysis is performed based on preset risk conditions, which improves the accuracy and efficiency of risk object identification. Moreover, the first business dataset within the corresponding time period can be obtained at any time according to actual needs to identify risk objects, making it more real-time.

[0207] In some embodiments, the preset risk conditions include at least one of the following:

[0208] (1) There is an interaction between a candidate inflow object and an intermediate object;

[0209] The intermediate object is used to connect a candidate inflow object and a target outflow object.

[0210] Taking a credit scenario as an example, the intermediary is a loan broker, which connects borrowers with multiple borrowing schemes with lenders. When a borrower with multiple borrowing schemes consults a loan broker about credit, it can be determined that the borrower is an abnormal borrower.

[0211] Optionally, the interaction between a candidate inflow object and an intermediate object can be a direct interaction or an indirect interaction.

[0212] For example, when a borrower with multiple loans in a credit scenario directly consults a loan intermediary for credit advice, this consultation behavior is a direct interaction behavior. When a borrower with multiple loans in a credit scenario consults a partner or third party of a loan intermediary for credit advice, this consultation behavior is an indirect interaction behavior.

[0213] (2) For a candidate inflow object, there is at least one negative evaluation.

[0214] Negative evaluations included complaints related to debt default.

[0215] Taking credit scenarios as an example, negative evaluations include, but are not limited to, overdue payments, failure to repay, underpayment of loans, and missed loan repayments.

[0216] (3) A candidate inflow object exhibits at least one instance of abnormal behavior.

[0217] The abnormal behavior can be detected by the target object itself or provided by a third party.

[0218] Taking credit scenarios as an example, abnormal behavior includes, but is not limited to, having a history of default with third parties, and engaging in multiple borrowing and cash-out activities with third parties.

[0219] (4) The risk level of a candidate inflow object reaches the set level threshold.

[0220] The risk level of a candidate inflow entity can be obtained by integrating data from multiple sources. The higher the risk level, the greater the likelihood of debt default.

[0221] Any of the above-mentioned preset risk conditions can be used as a long-term transfer risk label. When a candidate inflow object of the long-term category meets at least one preset risk condition, it is considered that the candidate inflow object hits at least one long-term transfer risk label, and therefore the candidate inflow object is determined to be a risk object.

[0222] In some embodiments, to improve the accuracy of risk object identification, when determining whether a candidate inflow object is a risk object, transaction information of other resource transfer types of the candidate object can also be considered. Optionally, transaction information of other resource transfer types includes, but is not limited to: daily consumption, salary information, and transfer information.

[0223] For example, if a candidate for inflow has complaints about overdue payments and its recent daily consumption has shifted from mid-to-high-end products to mid-to-low-end products, it indicates that its cash flow is in trouble, and therefore the candidate for inflow is identified as a risky entity.

[0224] In the embodiments of this application, risk objects in the candidate inflow objects of the multi-head class are analyzed in detail by using preset risk conditions corresponding to multiple dimensions, thereby improving the accuracy of risk object identification.

[0225] It should be noted that the aforementioned preset risk conditions can be extracted from the first business dataset or received from external sources.

[0226] In some embodiments, the pre-generated multiple business relationships associated with the first resource transfer type include a wider range of abnormal inflow objects of multiple types. In order to improve the security of the first resource transfer type, the risk objects identified in the first business data can be used as seed objects. In the multiple business relationships associated with the first resource transfer type, at least one abnormal inflow object associated with the seed object and satisfying the preset inflow conditions and preset risk conditions can be mined as a diffusion inflow object.

[0227] Taking lending scenarios as an example, such as Figure 7 As shown, object M has borrowing activities with credit institutions Q and P respectively, and the loan amount exceeds the preset threshold. At the same time, object M has complaints about overdue repayments. Therefore, object M is identified as a seed object. In multiple business relationships, object N has borrowing activities with credit institutions Q and R respectively, and the loan amount exceeds the preset threshold. The risk level of object N exceeds the set threshold. Therefore, object N is an abnormal inflow object in multiple business relationships. Since both abnormal inflow object N and seed object M have borrowing activities with credit institution Q, abnormal inflow object N is regarded as a diffusion inflow object associated with seed object M that has the risk of debt default.

[0228] S304: The server treats various sub-objects and each diffuse inflow object as a group of risky objects.

[0229] like Figure 7 As shown, both the seed object M and the abnormal inflow object N are risk objects.

[0230] Since the various business relationships associated with the first resource transfer type are pre-established based on a large second business dataset, the object relationships and object behavior relationships are relatively rich. Therefore, for risk objects that meet the preset risk conditions, they can be used as seed objects. From the various business relationships, at least one diffusion inflow object that meets the preset inflow conditions and multiple preset risk conditions can be mined. This will identify more risk objects that may default on debts, thereby reducing the debt default risk involved in the first resource transfer type and ensuring the security of the Internet finance environment.

[0231] In the process of resource transfer, the transfer behavior of a risky object has a contextual transaction relationship, and this contextual relationship can better reflect the resource status of the risky object, which is difficult to discover through simple condition judgment and diffusion.

[0232] Taking a credit scenario as an example, a risky individual borrows a large sum of money, repays the debt on the due date, and then immediately borrows or cashes out the repaid amount again. This behavior is a contextual transaction process that reflects a greater thirst for funds.

[0233] In some embodiments, in order to obtain the contextual transaction relationships of a risk object group, the temporal characteristics of the transaction behavior of the risk objects can be extracted, and based on the temporal characteristics of the transaction behavior, a target risk object group with a higher risk of debt default can be accurately and comprehensively screened from the risk object group, thereby improving the identification accuracy of the risk object group.

[0234] See Figure 8 The process of identifying target risk objects mainly includes the following steps:

[0235] S305: For each risk object in the risk object group, the server executes the following: Based on the first business dataset, obtain a set of behavioral time sequences for a risk object. Each behavioral time sequence contains: multiple resource transfer operations performed by a risk object in a time sequence under a resource flow.

[0236] When resource flow includes resource inflow, resource outflow, and resource entry / exit directions, the behavioral time sequence set includes inflow behavioral time sequence, outflow behavioral time sequence, and entry / exit behavioral time sequence. Each inflow behavioral time sequence contains multiple resource transfer operations executed chronologically by a risk object in the resource inflow direction; each outflow behavioral time sequence contains multiple resource transfer operations executed chronologically by a risk object in the resource outflow direction; and each entry / exit behavioral time sequence contains multiple resource transfer operations executed chronologically by a risk object in the resource entry / exit direction.

[0237] In practice, the process of obtaining the behavioral time series sequence set is as follows:

[0238] S3051: Based on the first business dataset, obtain the target inflow behavior data of each risk object in the direction of resource inflow, and arrange the target inflow behavior data in chronological order to obtain the inflow behavior time sequence.

[0239] In some embodiments, the process of obtaining each target inflow behavior data includes: based on the first business dataset, obtaining each initial inflow behavior data of the next risk object in the resource inflow direction, and for each initial inflow behavior data, performing the following respectively: extracting the inflow information corresponding to each preset dimension from the initial inflow behavior data, formatting each inflow information, and obtaining a target inflow behavior data.

[0240] Optionally, the inflow information for each preset dimension includes, but is not limited to, inflow time point, resource details, inflow type (such as credit card, other credit, etc.), and inflow scenario (such as online, offline). Among them, resource details include at least one of the following: inflow volume, inflow time interval, inflow location, source, and purpose.

[0241] Optionally, the formatting standard is {t, resource details, inflow type, inflow scenario}.

[0242] Taking the lending scenario as an example, for each risk object, the initial inflow behavior data is the initial borrowing data of that risk object. The inflow information of each preset dimension includes borrowing time, borrowing amount, borrowing time interval, borrowing location, borrowing source, borrowing purpose, borrowing type, borrowing scenario, etc.

[0243] For example, assuming the initial loan data is: on the Q credit institution's online platform, object M applied for a loan of 30,000 yuan through a credit card on December 3, 2024, then the formatted target loan data would be {December 3, 2024, (30,000 yuan, Q credit institution), credit card, online}.

[0244] For each risk object, after obtaining the formatted target inflow behavior data, the target inflow behavior data is arranged in chronological order to obtain the inflow behavior time sequence.

[0245] Taking credit scenarios as an example, such as Figure 9A As shown, the time sequence of inflow behavior is: target loan data 1, target loan data 2, target loan data 3 and target loan data 4, wherein the time point 1 of target loan data 1 is earlier than the time point 2 of target loan data 2, and the time point 2 of target loan data 2 is earlier than the time point 4 of target loan data 3 and target loan data 4.

[0246] S3052: Based on the first business dataset, obtain the outflow behavior data of each target of the next risk object in the direction of resource outflow, and arrange the outflow behavior data of each target in chronological order to obtain the outflow behavior time sequence.

[0247] In some embodiments, the process of obtaining each target outflow behavior data includes: based on the first business dataset, obtaining each initial outflow behavior data of the next risk object in the resource outflow direction, and for each initial outflow behavior data, performing the following respectively: extracting outflow information corresponding to each preset dimension from an initial outflow behavior data, formatting each outflow information to obtain a target outflow behavior data.

[0248] Optionally, the outflow information for each preset dimension includes, but is not limited to, outflow time, resource details, outflow type (such as credit card, other credit, etc.), and outflow scenario (such as online, offline). Among them, resource details include at least one of the following: outflow volume, outflow time interval, outflow location, destination, and purpose.

[0249] Optionally, the formatting standard for target outflow behavior data is consistent with the formatting standard for target inflow behavior data.

[0250] Taking the lending scenario as an example, for each risk object, the initial outflow behavior data is the initial repayment data of that risk object, and the inflow information of each preset dimension includes repayment time, repayment amount, repayment time interval, repayment location, repayment destination, repayment purpose, repayment type, repayment scenario, etc.

[0251] For example, assuming the initial repayment data is: on the Q credit institution's online platform, object M repays 30,000 yuan via credit card on January 3, 2025, then the formatted target repayment data is {January 3, 2025, (30,000 yuan, Q credit institution), credit card, online}.

[0252] For each risk object, after obtaining the formatted outflow behavior data of each target, the outflow behavior data of each target is arranged in chronological order to obtain the outflow behavior time sequence.

[0253] Taking credit scenarios as an example, such as Figure 9A As shown, the time sequence of outflow behavior is: Target Repayment Data 1, Target Repayment Data 2. Among them, the time point 3 of Target Repayment Data 1 is earlier than the time point 5 of Target Repayment Data 2, and so on.

[0254] In the embodiments of this application, by formatting the inflow behavior data and outflow behavior data, the data of different resource flows can be aligned, which improves the efficiency of data processing and thus improves the efficiency of risk object identification.

[0255] It should be noted that the embodiments of this application do not impose restrictive requirements on the execution order of data format standardization. For example, the inflow behavior data and outflow behavior data can also be formatted before the candidate inflow object is identified.

[0256] S3053: Mix and arrange the inflow and outflow behavior data of each target in chronological order to obtain the time sequence of the inflow and outflow behavior of the next risk object in the direction of resource inflow and outflow.

[0257] Since the target inflow and outflow behavior data are formatted using the same standard and contain time-point information, for each risk object, the target inflow and outflow behavior data are mixed and arranged in chronological order to obtain a time-series sequence of inflow and outflow behavior in the resource inflow and outflow directions.

[0258] Taking credit scenarios as an example, such as Figure 9AAs shown, the time sequence of entry and exit behaviors is as follows: Target Loan Data 1, Target Loan Data 2, Target Repayment Data 1, Target Loan Data 3, Target Loan Data 4, Target Repayment Data 2. Specifically, time point 1 of Target Loan Data 1 is earlier than time point 2 of Target Loan Data 2; time point 3 of Target Repayment Data 1 is later than time point 2 of Target Loan Data 2 but earlier than time points 4 of Target Loan Data 3 and Target Loan Data 4; and time point 5 of Target Repayment Data 2 is later than time point 4 of Target Loan Data 4.

[0259] Since the behavioral data in the inflow, outflow, and entry / exit behavior sequences are arranged in chronological order, these sequences can be used to obtain the contextual transaction relationships of a series of resource transfer behaviors for each risk object. By analyzing these contextual transaction relationships, it is possible to determine whether the transfer behavior of the risk object is normal, thereby further identifying high-risk groups and improving the security of the internet finance environment.

[0260] S306: The server obtains at least one target risk object from the risk object group based on the behavioral time sequence set of each risk object.

[0261] For each risk object, the contextual transaction relationship of the risk object's resource transfer behavior can be obtained through its inflow behavior time sequence, outflow behavior time sequence, and entry and exit behavior sequence. Based on the contextual transaction relationship, the target risk object group with high debt default risk can be obtained.

[0262] Taking credit scenarios as an example, the entire process of identifying the target risk group is as follows: Figure 9B As shown, firstly, after formatting loan and repayment data generated at different time points, the payers in the loan data and the payees in the repayment data are aggregated. Combining the attribute information of the payers and payees, credit institutions are identified. Then, based on the loan data, repayment data, and borrowing and repayment data, a group of credit entities with resource exchanges with credit institutions is obtained. Next, from the group of credit entities, a group of borrowers with multiple borrowing behaviors whose resource inflow volume and frequency meet preset inflow conditions are identified. Finally, from the group of borrowers, seed entities that meet preset risk conditions are identified. Based on the seed entities, a broader group of risky borrowers is mined, and based on the behavioral time sequence set obtained from the loan and repayment data, a group of low-quality target risky borrowers is selected from the group of risky borrowers.

[0263] In some embodiments, to ensure the real-time nature of the identification results, when the time period corresponding to the first business dataset is short, the resource transfer behavior in a short period of time may not accurately express the contextual transaction relationship. Offline data can be introduced to increase the temporal characteristics of the risk object's behavior, thereby improving the accuracy of the target risk object identification.

[0264] The process of extracting behavioral time-series sequence sets from offline data is the same as the process of extracting behavioral time-series sequence sets from the first business dataset, and will not be described again here.

[0265] In some embodiments, the target risk object can be obtained through a behavior time-series recognition model built based on deep learning.

[0266] Taking the behavioral temporal recognition model as an example, which uses a Long Short-Term Memory (LSTM) network employing an attention mechanism, such as... Figure 10 As shown, the identification process for the target risk object group is as follows: After the behavioral time series sequence set of each risk object is input into the model, it is encoded by the embedding layer to obtain the feature vector of the behavioral time series sequence set. The LATM network captures the contextual transaction relationship in the behavioral time series sequence set based on these feature vectors to predict the debt default situation of the risk object. The attention network weights the prediction vector of the LSTM network based on the object features in the behavioral time series sequence set, so that the model can focus on the most important part of the current prediction. After the feature dimensionality reduction of the prediction vector of the LSTM network by the pooling layer (e.g., dimensionality reduction to 512), the Softmax layer performs a multi-classification task based on the dimensionality reduction prediction vector to obtain the target risk object group with high default risk and other multiple inflow object groups (e.g., multiple inflow objects with low default risk, multiple inflow objects that do not default, etc.).

[0267] Optionally, the behavior time-series recognition model can be trained using a multi-class cross-entropy loss function, where the cross-entropy loss function is expressed as follows:

[0268]

[0269] Where x represents the training sample, y represents the true label, y′ represents the predicted label, and n represents the total number of samples.

[0270] It should be noted that the network structure of the behavior time sequence recognition model in this application is not limited to a single deep learning algorithm. Different learning algorithms can be selected according to the actual business effect and resource cost. For example, a deep learning network with large parameters, such as Transformer or Timenet, can be used to replace the LSTM network.

[0271] Throughout the risk control identification process, the target outflow object group, candidate inflow object group, risk object group, and target risk object group can be output using different network models, or they can be output using a large model.

[0272] Considering that the predictions and inferences of network models are not always 100% accurate, and that the recognition performance of network models will decline over time due to changes in transaction habits and resource forms, using the inference results of network models as feedback to optimize network models is a very important part of the practical application of network models.

[0273] Regarding the network model optimization process, this application embodiment uses various behavior auditing methods to provide feedback on the prediction problems of the network model and iteratively updates the network model in a timely manner to improve the accuracy of the network model in subsequent use.

[0274] Taking credit scenarios as an example, various behavioral auditing methods include, but are not limited to, the debt default behavior of the subject in real life, complaints generated when using the identification results for risk control management, and test results of setting up different experimental control groups.

[0275] During the resource transfer process, the business data of the first resource transfer type reported by the terminal can be divided into two parts: real-time and offline. The time period corresponding to the real-time part is shorter than that corresponding to the offline part, and the data volume of the real-time part is smaller than that of the offline part. Therefore, the real-time part can be stored in the Redis database of the big data platform, while the offline part can be stored in the Hadoop Distributed File System (HDFS).

[0276] In some embodiments, streaming computing frameworks such as Flink can be used to perform data analysis on the real-time part to build a real-time basic profile, and big data frameworks such as HDFS+Spark can be used to perform data analysis on the offline part to build an offline basic profile.

[0277] It should be noted that this application does not impose any restrictive requirements on the big data framework used for building the basic profile. For example, frameworks such as MapReduce, SparkSQL, and PyTorch can also be used.

[0278] In some embodiments, when identifying target risk groups, analysis can be performed based on both real-time and offline profiles.

[0279] In some embodiments, the basic profile includes at least one of the following: resource flow, various business relationships between business objects, behavioral time sequence set, and multi-head transfer risk label.

[0280] like Figure 11As shown, the basic profile includes resource flow direction (resource inflow direction, resource outflow direction, and resource entry / exit direction), multiple business relationships (first object relationship between inflow and outflow objects, second object relationship between inflow objects, third object relationship between abnormal inflow and outflow objects, and object behavior relationship between inflow objects and entry / exit behaviors), behavior sequence set (inflow behavior sequence, outflow behavior sequence, and entry / exit behavior sequence), and multi-head transfer risk label (interaction with intermediate objects, at least one negative evaluation, at least one abnormal behavior, and risk level reaching a set threshold).

[0281] Taking credit scenarios as an example, the entire technical process of risk control identification is as follows: Figure 12 As shown, the technology includes basic profile construction, multi-borrowing behavior identification, multi-borrowing risk target group identification, target risk target group identification, and model optimization. The basic profile construction technology establishes rich and interconnected basic profiles by mining information such as transaction behavior and target identity. The multi-borrowing behavior identification technology divides transaction behavior into borrowing and repayment behavior. The multi-borrowing risk target group identification technology uses preset inflow conditions to determine multiple borrowers and screen multi-borrower groups, and combines multi-borrowing transfer risk labels to determine risk, obtaining seed borrowers for target diffusion and uncovering more risky borrower groups. The target risk target group identification technology obtains contextual transaction relationships through behavioral time sequence sets, thereby more accurately identifying potentially defaulting low-quality borrower groups. The model optimization technology audits the identification results for anomalies such as transaction loops and aggregations and returns the results to the model for iterative optimization.

[0282] In the field of credit risk control, this application's embodiments utilize big data and deep learning technologies to systematically construct a full-process system for identifying multi-risk customer groups across multiple credit scenarios, based on transaction characteristics and behavioral time-series sequences. This system improves the effectiveness of default prediction by several times, enabling more accurate, comprehensive, and timely identification of potential defaulting borrower groups. It effectively prevents high-risk customer groups from engaging in lending activities, prevents lenders from incurring greater financial losses, and thus protects the security of the internet finance environment in the face of significant economic fluctuations.

[0283] Based on the same technical concept, this application provides an object recognition device that can implement the above-described object recognition method and achieve the same technical effect.

[0284] See Figure 13 The object recognition device includes an acquisition module 1301, a screening module 1302, a risk diffusion module 1304, and an identification module 1304, wherein:

[0285] The acquisition module 1301 is used to acquire the first business dataset that conforms to the first resource transfer type;

[0286] The filtering module 1302 is used to filter out at least one candidate inflow object whose resource inflow volume and resource inflow frequency meet the preset inflow conditions based on the resource flow direction of the first business dataset.

[0287] The risk diffusion module 1303 is used to perform the following for each candidate inflow object: when a candidate inflow object meets the preset risk conditions, the candidate inflow object is used as a seed object, and at least one diffusion inflow object that is associated with the seed object and meets the preset inflow conditions and preset risk conditions is selected from the multiple business relationships associated with the first resource transfer type in the pre-generated first resource transfer type.

[0288] The identification module 1304 is used to identify various sub-objects and various diffusion inflow objects as a risk object group.

[0289] Optionally, after obtaining the risk target group, the identification module 1304 is also used for:

[0290] For each risk object in the risk object group, perform the following: Based on the first business dataset, obtain a set of behavioral time sequences for a risk object. Each behavioral time sequence contains: multiple resource transfer operations performed by a risk object in a time sequence under a resource flow.

[0291] Based on the behavioral time sequence set of each risk object, at least one target risk object in the risk object group is obtained.

[0292] Optionally, the identification module 1304 is specifically used for:

[0293] Based on the first business dataset, obtain the target inflow behavior data of each risk object in the resource inflow direction, and arrange the target inflow behavior data in chronological order to obtain the inflow behavior time series; and,

[0294] Based on the first business dataset, obtain the outflow behavior data of each target of the next risk object in the direction of resource outflow, and arrange the outflow behavior data of each target in chronological order to obtain the outflow behavior time sequence.

[0295] By mixing and arranging the inflow and outflow behavior data of each target in chronological order, a time sequence of the inflow and outflow behavior of the next risk object in the direction of resource inflow and outflow is obtained.

[0296] Optionally, the identification module 1304 is specifically used for:

[0297] Based on the first business dataset, obtain the initial inflow behavior data of each risk object in the resource inflow direction, and for each initial inflow behavior data, perform the following: extract the inflow information corresponding to each preset dimension from the initial inflow behavior data, format the inflow information, and obtain a target inflow behavior data;

[0298] Based on the first business dataset, obtain the initial outflow behavior data of each risk object in the resource outflow direction, and for each initial outflow behavior data, perform the following: extract the outflow information corresponding to each preset dimension from the initial outflow behavior data, format the outflow information, and obtain a target outflow behavior data.

[0299] Optionally, the filtering module 1302 is specifically used for:

[0300] Based on the resource flow of the first business dataset, obtain the resource inflow data subset under the resource inflow direction and the resource outflow data subset under the resource outflow direction;

[0301] Based on the resource inflow data subset and the resource outflow data subset, multiple candidate outflow objects under the first resource transfer type are obtained;

[0302] Based on the first business dataset, obtain the attribute information of each of the multiple candidate outflow objects, and based on the attribute information, select at least one target outflow object from the multiple candidate outflow objects;

[0303] In business relationships, at least one candidate inflow object is selected that meets the preset inflow conditions in terms of resource inflow amount and resource inflow frequency with at least one target outflow object.

[0304] Optionally, the filtering module 1302 is specifically used for:

[0305] Obtain a second business dataset that conforms to the first resource transfer type; wherein the first business dataset is a subset of the second business dataset, and the time period corresponding to the second business dataset is earlier than the time period corresponding to the first business dataset;

[0306] Based on the second business dataset, establish multiple business relationships.

[0307] Optionally, multiple business relationships include object relationships and object behavior relationships. The filtering module 1302 is specifically used for:

[0308] Based on the second business dataset, for each inflow object whose resource inflow volume and resource inflow frequency meet the preset inflow conditions, a first object relationship is established between each inflow object and each outflow object.

[0309] Based on multiple inflow objects that are simultaneously associated with the same thing, establish a second object relationship between the multiple inflow objects;

[0310] Based on screening at least one abnormal inflow object that meets the preset risk conditions from each inflow object, a third object relationship is established between the at least one abnormal inflow object and each outflow object.

[0311] Based on the resource entry and exit behaviors associated with each inflow object, an object behavior relationship is established between each inflow object and its corresponding resource entry and exit behaviors.

[0312] Optionally, the preset risk conditions include at least one of the following:

[0313] There is an interaction between a candidate inflow object and an intermediate object, which is used to connect a candidate inflow object and a target outflow object.

[0314] For each candidate applicant, there is at least one negative review.

[0315] A candidate inflow object exhibits at least one instance of anomalous behavior;

[0316] A candidate for inflow has reached a risk level threshold.

[0317] Optionally, module 1301 is specifically used for:

[0318] Acquire the pipeline data pool generated by resource transfer;

[0319] Based on the flow information of each flow data in the flow data pool, the flow data pool is classified to obtain the flow data sets corresponding to each resource transfer type; the flow information includes at least one of the following: flow method, flow direction, and flow remarks.

[0320] Use the transaction dataset corresponding to the first resource transfer type as the first business dataset.

[0321] The object identification device provided in this application addresses the issue of resource transfer involving resource flow. Since the inflowing object in the resource inflow direction is the resource recipient and needs to repay debts, and multiple inflowing objects have a higher probability of debt default, risk control identification of these objects is necessary. Resource inflow volume and frequency are important factors for multiple inflows. Therefore, preset inflow conditions can be set for multiple inflowing objects based on these factors. Based on the resource flow of the first business dataset conforming to the first resource transfer type, at least one candidate inflowing object meeting the preset inflow conditions is initially screened. When a selected candidate inflowing object meets preset risk conditions, it indicates a high risk of debt default, and thus, the candidate inflowing object is identified as a risk object. Targeted analysis based on preset risk conditions is performed on the selected candidate inflowing objects of the multiple inflow type, improving the accuracy and efficiency of risk object identification. Furthermore, the device can acquire the first business dataset within the corresponding time period for risk object identification as needed, providing stronger real-time performance.

[0322] On the other hand, since the first resource transfer type is associated with multiple business relationships, risk objects that meet the preset risk conditions can be used as seed objects. From the multiple business relationships, at least one diffusion inflow object that meets the preset inflow conditions and preset risk conditions can be mined, thereby identifying more risk objects that may default on debts. This reduces the debt default risk involved in the first resource transfer type and ensures the security of the Internet finance environment.

[0323] For ease of description, the above sections are divided into modules (or units) according to their functions and described separately. Of course, in implementing this application, the functions of each module (or unit) can be implemented in one or more software or hardware components.

[0324] Those skilled in the art will understand that various aspects of this application can be implemented as a system, method, or program product. Therefore, various aspects of this application can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, collectively referred to herein as a "circuit," "module," or "system."

[0325] Having described the object recognition method and apparatus according to exemplary embodiments of this application, we will now describe an electronic device according to another exemplary embodiment of this application.

[0326] In one embodiment, the electronic device may be Figure 1 The server in the middle. For example... Figure 14As shown, the structure of the electronic device may include a memory 1401, a communication module 1403, and one or more processors 1402.

[0327] The memory 1401 is used to store computer programs executed by the processor 1402. The memory 1401 may mainly include a program storage area and a data storage area, wherein the program storage area may store the operating system and the operating instruction set, etc.

[0328] Memory 1401 may be volatile memory, such as random-access memory (RAM); memory 1401 may also be non-volatile memory, such as read-only memory, flash memory, hard disk drive (HDD), or solid-state drive (SSD); or memory 1401 may be any other medium capable of carrying or storing a desired computer program having the form of instructions or data structures and accessible by a computer, but is not limited thereto. Memory 1401 may be a combination of the above-described memories.

[0329] Processor 1402 may include one or more central processing units (CPUs) or digital processing units, etc. Processor 1402 is used to implement the steps of the object recognition method described above when calling the computer program stored in memory 1401.

[0330] The communication module 1403 is used to communicate with other servers and terminal devices.

[0331] This application embodiment does not limit the specific connection medium between the memory 1401, communication module 1403, and processor 1402. This application embodiment... Figure 14 The memory 1401 and the processor 1402 are connected via a bus 1404, and the bus 1404 is in Figure 14 The diagram uses thick lines to describe the connections between other components; these are for illustrative purposes only and should not be considered limiting. The 1404 bus can be divided into address bus, data bus, control bus, etc. For ease of description, Figure 14 It is described using only a thick line, but does not indicate that there is only one bus or one type of bus.

[0332] The memory 1401 stores a computer storage medium, which stores computer-executable instructions for implementing the object recognition method of this application embodiment. The processor 1402 is used to execute the steps of the above-described object recognition method.

[0333] In some possible implementations, various aspects of the object recognition method provided in this application can also be implemented as a program product, which includes a computer program. When the program product is run on an electronic device, the computer program causes the electronic device to perform the steps of the object recognition method according to the various exemplary embodiments of this application described above. For example, the electronic device can perform actions such as... Figure 3 , Figure 8 The steps are shown in the figure.

[0334] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical fiber, a portable compact disk read-only memory, an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0335] The program product of the embodiments of this application may employ a portable compact disk read-only memory and include a computer program, and may run on an electronic device. However, the program product of this application is not limited thereto. In this document, the readable storage medium may be any tangible medium that contains or stores a program that may be used by or in conjunction with a command execution system, apparatus, or device.

[0336] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying a readable computer program. This propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting a program for use by or in conjunction with a command execution system, apparatus, or device.

[0337] Computer programs contained on readable media may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0338] Computer programs for performing the operations of this application can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The computer program can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device.

[0339] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application 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 a computer-usable computer program.

[0340] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0341] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. An object recognition method, characterized in that, The method includes: Obtain the first business dataset that matches the first resource transfer type; Based on the resource flow of the first business dataset, at least one candidate inflow object is selected if the resource inflow amount and the number of resource inflows meet the preset inflow conditions. For each of the candidate inflow objects, the following steps are performed: when a candidate inflow object meets the preset risk conditions, the candidate inflow object is used as a seed object, and at least one diffusion inflow object that is associated with the seed object and meets the preset inflow conditions and the preset risk conditions is selected from the various business relationships associated with the first resource transfer type in the pre-generated data. Various sub-objects and each diffusion inflow object are considered as a risk object group.

2. The method as described in claim 1, characterized in that, After obtaining the risk target group, the method further includes: For each risk object in the group of risk objects, the following is performed: Based on the first business dataset, a set of behavioral time sequences of a risk object is obtained, and each behavioral time sequence includes: multiple resource transfer operations performed by the risk object in a time sequence under a resource flow. Based on the behavioral time sequence set of each risk object, at least one target risk object in the group of risk objects is obtained.

3. The method as described in claim 2, characterized in that, The step of obtaining a behavioral time-series sequence set of a risk object based on the first business dataset includes: Based on the first business dataset, target inflow behavior data for each risk object under the resource inflow direction are obtained, and the target inflow behavior data are arranged in chronological order to obtain an inflow behavior time series; and, Based on the first business dataset, obtain the target outflow behavior data of each risk object under the resource outflow direction, and arrange the target outflow behavior data in chronological order to obtain the outflow behavior time sequence; The inflow and outflow data of each target are mixed and arranged in chronological order to obtain a time sequence of the inflow and outflow behavior of a risk object under the resource inflow and outflow direction.

4. The method as described in claim 3, characterized in that, The step of obtaining the inflow behavior data of a risk object under the resource inflow direction based on the first business dataset includes: Based on the first business dataset, obtain the initial inflow behavior data of a risk object under the resource inflow direction, and for each initial inflow behavior data, perform the following: extract the inflow information corresponding to each preset dimension from the initial inflow behavior data, format the inflow information, and obtain a target inflow behavior data. The step of obtaining target outflow behavior data for a risk object under the resource outflow direction based on the first business dataset includes: Based on the first business dataset, obtain the initial outflow behavior data of a risk object under the resource outflow direction, and for each initial outflow behavior data, perform the following: extract the outflow information corresponding to each preset dimension from the initial outflow behavior data, format the outflow information, and obtain a target outflow behavior data.

5. The method according to any one of claims 1-4, characterized in that, The step of filtering at least one candidate inflow object based on the resource flow direction of the first business dataset, where the resource inflow volume and the number of resource inflows meet preset inflow conditions, includes: Based on the resource flow direction of the first business dataset, a subset of resource inflow data in the resource inflow direction and a subset of resource outflow data in the resource outflow direction are obtained. Based on the resource inflow data subset and the resource outflow data subset, multiple candidate outflow objects under the first resource transfer type are obtained; Based on the first business dataset, obtain the attribute information of each of the multiple candidate outflow objects, and based on the attribute information, filter out at least one target outflow object from the multiple candidate outflow objects; In the business relationship, at least one candidate inflow object is selected that meets the preset inflow conditions in terms of resource inflow amount and resource inflow frequency with the at least one target outflow object.

6. The method according to any one of claims 1-4, characterized in that, The various business relationships were obtained in the following ways: Obtain a second business dataset that conforms to the first resource transfer type; wherein the first business dataset is a subset of the second business dataset, and the time period corresponding to the second business dataset is earlier than the time period corresponding to the first business dataset; Based on the second business dataset, various business relationships are established.

7. The method as described in claim 6, characterized in that, The various business relationships include object relationships and object behavior relationships, and each of the business relationships is obtained using any of the following methods: Based on the second business dataset, for each inflow object whose resource inflow volume and resource inflow frequency meet the preset inflow conditions, a first object relationship is established between each inflow object and each outflow object; Based on the multiple inflow objects that are simultaneously associated with the same thing among the aforementioned inflow objects, a second object relationship is established among the multiple inflow objects; Based on the selection of at least one abnormal inflow object that meets the preset risk conditions from the inflow objects, a third object relationship is established between the at least one abnormal inflow object and the outflow objects. Based on the resource entry and exit behaviors associated with each inflow object, an object behavior relationship is established between each inflow object and its corresponding resource entry and exit behaviors.

8. The method according to any one of claims 1-4, characterized in that, The preset risk conditions include at least one of the following situations: There is an interactive behavior between the candidate inflow object and the intermediate object, and the intermediate object is used to connect the candidate inflow object and the target outflow object. For each of the aforementioned candidate inflow objects, there exists at least one negative evaluation; The candidate inflow object exhibits at least one instance of abnormal behavior; The risk level of a candidate inflow object reaches a set threshold.

9. The method according to any one of claims 1-4, characterized in that, The step of obtaining the first business dataset that conforms to the first resource transfer type includes: Acquire the pipeline data pool generated by resource transfer; Based on the flow information of each flow data in the flow data pool, the flow data pool is classified to obtain the flow data sets corresponding to each resource transfer type; the flow information includes at least one of the flow method, flow direction, and flow remarks. The pipeline dataset corresponding to the first resource transfer type is used as the first business dataset.

10. An object recognition device, characterized in that, include: The acquisition module is used to acquire the first business dataset that conforms to the first resource transfer type; The filtering module is used to filter out at least one candidate inflow object whose resource inflow volume and resource inflow frequency meet preset inflow conditions based on the resource flow direction of the first business dataset. The risk diffusion module is used to perform the following for each candidate inflow object: when a candidate inflow object meets the preset risk conditions, the candidate inflow object is used as a seed object, and at least one diffusion inflow object that is associated with the seed object and meets the preset inflow conditions and the preset risk conditions is selected from the multiple business relationships associated with the first resource transfer type in a pre-generated manner. The identification module is used to classify various sub-objects and each diffusion inflow object as a risk object group.

11. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of any of the methods described in claims 1-9.

12. A computer-readable storage medium, characterized in that, It includes a computer program that, when run on an electronic device, causes the electronic device to perform the steps of any of the methods described in claims 1-9.

13. A computer program product, characterized in that, The method includes a computer program stored in a computer-readable storage medium; when a processor of an electronic device reads the computer program from the computer-readable storage medium, the processor executes the computer program, causing the electronic device to perform the steps of any one of the methods of claims 1-9.

Images