Multi-source data joint risk identification method, device, equipment, medium and program product

Abstract

The application provides a multi-source data joint risk identification method and device, equipment, medium and program product, which can be applied to the field of big data technology, and in particular relates to the application of big data in the financial technology scene, including: obtaining customer behavior data of a target object in each data source; extracting a sparse feature variable from the customer behavior data, the sparse feature variable being a feature variable whose proportion of zero values in the customer behavior data exceeds a preset threshold within a preset statistical period; inputting the sparse feature variable into a pre-trained risk identification model to output a risk quantification result, the risk identification model being obtained based on a joint objective function, the joint objective function including a loss function term on each data source and a non-convex group-level penalty term, the non-convex group-level penalty term acting on a parameter group composed of regression coefficients of the same sparse feature variable in each data source.

Description

Technical Field

[0001] This application relates to the field of big data technology, to the application of big data in financial technology scenarios, and more specifically to a method, apparatus, device, medium and program product for joint risk identification of multi-source data. Background Technology

[0002] As commercial banks continue their digital transformation, banking systems have accumulated massive amounts of customer behavior data. In actual business scenarios, many customers do not trigger risk events within the statistical period, therefore many characteristic variables in the behavioral data cannot be utilized.

[0003] Most risk identification methods are trained on a single data source, assuming the data follows an approximately normal distribution. In such data scenarios, they are susceptible to noise and exhibit poor model stability. Furthermore, directly concatenating multi-source data can easily lead to feature redundancy and overfitting. While Lasso-type penalty methods can achieve variable selection, they impose a continuous shrinkage on large-coefficient variables, resulting in biased estimations of important risk features and low accuracy in risk identification. Summary of the Invention

[0004] In view of the above problems, this application provides a method, apparatus, device, medium and program product for joint risk identification of multi-source data.

[0005] According to a first aspect of this application, a multi-source data joint risk identification method is provided, comprising: acquiring customer behavior data of a target object from various data sources; extracting sparse feature variables from the customer behavior data, wherein the sparse feature variables are feature variables in which the proportion of zero values ​​in the customer behavior data exceeds a preset threshold within a preset statistical period; inputting the sparse feature variables into a pre-trained risk identification model, and outputting risk quantification results, wherein the risk identification model is trained based on a joint objective function, the joint objective function including loss function terms on each data source and a non-convex group-level penalty term, wherein the non-convex group-level penalty term acts on a parameter group consisting of the regression coefficients of the same sparse feature variable in each data source.

[0006] According to an embodiment of this application, the non-convex group-level penalty term is an L2 Group MCP (Type II Group Minimax Concavity) penalty term. The penalty function of the L2 Group MCP penalty term is as follows: for any sparse feature variable in the parameter group consisting of the regression coefficients in each data source, the value of the penalty function is determined by the Euclidean norm of the parameter group and the minimax concavity adjustment parameter; when the Euclidean norm is less than the product of the minimax concavity adjustment parameter and the regularization parameter, the marginal growth rate of the penalty function decreases linearly as the Euclidean norm increases; when the Euclidean norm is greater than or equal to the product, the marginal growth rate of the penalty function is zero.

[0007] According to an embodiment of this application, the training process of the risk identification model includes: constructing a joint objective function based on each data source, the joint objective function including the sum of squared losses on each data source and the non-convex group-level penalty term; iteratively optimizing the joint objective function using a group coordinate descent algorithm until the joint objective function converges to obtain the trained risk identification model; wherein, in each iteration, the parameter group corresponding to each sparse feature variable is selected sequentially for updating, while the remaining parameter groups in the joint objective function, except for the corresponding parameter groups, remain unchanged.

[0008] According to an embodiment of this application, the sparse feature variable is a zero-inflated variable; the step of extracting sparse feature variables from the customer behavior data includes: determining the proportion of zero values ​​for each feature variable based on the number of times each feature variable in the customer behavior data has zero values ​​within a preset statistical period; and identifying feature variables whose proportion of zero values ​​exceeds a preset threshold as zero-inflated variables.

[0009] According to an embodiment of this application, obtaining customer behavior data of target objects from various data sources includes: associating original customer behavior data belonging to the same target object with a unified customer identifier to obtain associated customer behavior data; and performing time alignment on the associated customer behavior data according to a preset time window to obtain customer behavior data of the target object.

[0010] According to an embodiment of this application, the risk quantification result includes at least one of customer risk score, abnormal behavior probability, and risk warning level; the method further includes: triggering a risk warning signal based on whether the customer risk score or the abnormal behavior probability exceeds the corresponding warning threshold.

[0011] According to embodiments of this application, the risk quantification result includes a customer risk score; the method further includes: outputting a credit recommendation or approval decision based on a comparison between the customer risk score and a preset credit approval threshold range; and / or, the risk quantification result includes an abnormal behavior probability; the method further includes: marking the target object and related transaction records as pending review when the abnormal behavior probability exceeds a preset abnormal transaction warning threshold; and / or, the risk quantification result includes a customer risk score; the method further includes: acquiring customer behavior data of existing customers according to a preset period and inputting it into the risk identification model, and updating the risk level of existing customers according to the changing trend of the output customer risk score.

[0012] A second aspect of this application provides a multi-source data joint risk identification device, comprising: a data acquisition module for acquiring customer behavior data of a target object from various data sources; a feature extraction module for extracting sparse feature variables from the customer behavior data, wherein the sparse feature variables are feature variables in which the proportion of zero values ​​in the customer behavior data exceeds a preset threshold within a preset statistical period; and a risk identification module for inputting the sparse feature variables into a pre-trained risk identification model and outputting risk quantification results, wherein the risk identification model is trained based on a joint objective function, the joint objective function including loss function terms from each data source and a non-convex group-level penalty term, wherein the non-convex group-level penalty term acts on a parameter group composed of the regression coefficients of the same sparse feature variable in each data source.

[0013] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0014] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0015] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0016] According to embodiments of this application, by acquiring customer behavior data of target objects from various data sources, sparse feature variables with a zero-value ratio exceeding a preset threshold within a preset statistical period are extracted from the customer behavior data. These sparse feature variables are then input into a risk identification model trained based on a joint objective function. This effectively addresses the high sparsity problem of zero-inflation behavioral data in banks. By constraining the parameter set composed of regression coefficients of the same sparse feature variable in various data sources through a non-convex group-level penalty term, the feature redundancy and overfitting problems caused by simply splicing multi-source data in traditional methods are avoided. This achieves cross-data source information sharing and joint screening of risk features, improving the prediction accuracy and stability of the risk identification model in zero-inflation high-dimensional sparse data scenarios. Attached Figure Description

[0017] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0018] Figure 1 The illustration schematically depicts an application scenario of a multi-source data joint risk identification method, apparatus, device, medium, and program product according to embodiments of this disclosure;

[0019] Figure 2 A flowchart illustrating a multi-source data joint risk identification method according to an embodiment of this application is shown in the schematic diagram.

[0020] Figure 3 This illustration schematically shows the overall flowchart of joint learning of multi-source zero-inflation behavioral data of banks according to an embodiment of this application;

[0021] Figure 4 This illustration schematically shows a flowchart of L2 Group MCP model training and parameter optimization according to an embodiment of this application;

[0022] Figure 5 This illustration shows a flowchart of a bank risk identification business implementation according to an embodiment of this application;

[0023] Figure 6 This schematically illustrates a structural block diagram of a multi-source data joint risk identification device according to an embodiment of this application; and

[0024] Figure 7 A block diagram of an electronic device for a multi-source data joint risk identification method according to an embodiment of this application is shown schematically. Detailed Implementation

[0025] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0026] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0027] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0028] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0029] In the technical solution of this application, the user information (including but not limited to user personal information, user cross-border remittance information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, application, and application of the relevant data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0030] In scenarios involving automated decision-making using personal information, the methods, devices, and systems provided in this application all offer users corresponding entry points for choosing to agree to or reject the automated decision-making results. If the user chooses to reject, the process proceeds to the expert decision-making stage. Here, "automated decision-making" refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests, or economic, health, and credit status through computer programs, and then making a decision. Here, "expert decision-making" refers to the activity of making decisions by personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0031] Figure 1 The illustration schematically depicts an application scenario of a multi-source data joint risk identification method, apparatus, device, medium, and program product according to embodiments of the present disclosure.

[0032] like Figure 1 As shown, the application scenario according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105.

[0033] Network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0034] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to initiate risk identification tasks or obtain risk assessment results. Various bank risk control management platforms, risk monitoring systems, credit approval systems, or other risk control-related applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103.

[0035] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be requesting nodes used to initiate risk identification requests. The specific forms of the first terminal device 101, the second terminal device 102, and the third terminal device 103 include, but are not limited to, personal computers, laptops, workstations, physical servers, virtual machines, container instances, etc. The first terminal device is typically deployed in a dynamic network environment, and its network address is dynamic.

[0036] Server 105 can be a service provider node, used to receive and process risk identification requests from requesting nodes. Server 105 is typically deployed in a data center or cloud service environment, exposing an application programming interface (API). Server 105 can integrate a multi-source data joint risk identification service module to output risk identification results.

[0037] It should be noted that the multi-source data joint risk identification method provided in this disclosure embodiment can generally be executed by server 105. Correspondingly, the multi-source data joint risk identification device provided in this disclosure embodiment can generally be located in server 105. The multi-source data joint risk identification method provided in this disclosure embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the multi-source data joint risk identification device provided in this disclosure embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0038] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0039] Figure 2 A flowchart illustrating a multi-source data joint risk identification method according to an embodiment of this application is shown.

[0040] like Figure 2 As shown, the multi-source data joint risk identification method in this embodiment includes operations S210 to S230.

[0041] In operation S210, customer behavior data of the target object is obtained from various data sources.

[0042] According to embodiments of this application, customer behavior data refers to raw data collected from multiple business systems that can reflect customer behavior characteristics.

[0043] According to the embodiments of this application, each data source refers to a different business system or data platform. The data structure, data dimensions and sample size of different data sources may be different from each other, exhibiting heterogeneous characteristics.

[0044] According to embodiments of this application, the target object refers to a bank customer or business entity, and customer behavior data from various data sources are associated through a unified customer identifier to identify behavior records belonging to the same target object.

[0045] In operation S220, sparse feature variables are extracted from customer behavior data. Sparse feature variables are feature variables in customer behavior data where the proportion of zero values ​​exceeds a preset threshold within a preset statistical period.

[0046] According to embodiments of this application, sparse feature variables refer to feature variables in customer behavior data where the proportion of zero values ​​exceeds a preset threshold within a preset statistical period. In banking scenarios, since most customers do not trigger risk events within the statistical period, a large number of behavioral variables remain at zero values ​​for a long time, such as customers not engaging in abnormal transactions, not having overdue behaviors, or not triggering certain risk rules.

[0047] According to embodiments of this application, by setting a threshold for the proportion of zero values, feature variables with sparse characteristics can be identified from high-dimensional customer behavior data. The preset threshold can be flexibly configured according to business scenarios and data characteristics.

[0048] In operation S230, sparse feature variables are input into a pre-trained risk identification model, and risk quantification results are output.

[0049] According to embodiments of this application, the risk identification model is a machine learning model pre-trained based on a joint objective function. This model takes sparse feature variables as input and risk quantification results as output. The core of the model lies in the joint objective function used during its training process, which consists of two parts: the first part is a loss function term for each data source, used to measure the model's fitting accuracy on each data source; the second part is a non-convex group-level penalty term, used to impose constraints on each parameter group.

[0050] According to embodiments of this application, regression coefficients refer to the model parameters of each sparse feature variable in the risk identification model corresponding to each data source. For the same sparse feature variable, it corresponds to different regression coefficients in different data sources. The regression coefficients of the same sparse feature variable in each data source constitute a parameter set. That is, each sparse feature variable uniquely corresponds to a parameter set, which contains the regression coefficients of that feature variable in all data sources. Different sparse feature variables correspond to different parameter sets, and each parameter set is independent of the others.

[0051] According to embodiments of this application, the non-convex group-level penalty term refers to a non-convex penalty function applied to each parameter group. This penalty term applies penalty on a per-parameter-group basis, rather than on a single regression coefficient. "Group-level" can mean that all regression coefficients within the same parameter group (i.e., regression coefficients of the same sparse feature variable across various data sources) are considered holistically and subjected to a uniform constraint, thereby achieving a unified assessment of the overall contribution of that feature variable across all data sources. "Non-convex" can mean that the penalty function is mathematically non-convex, distinguishing it from the convex penalty function used by Lasso. The advantage of non-convex penalty is that its constraint on large-coefficient parameter groups gradually weakens as the amplitude of the parameter group increases, thus avoiding excessive constraint on important risk features with large coefficients.

[0052] According to embodiments of this application, a balance can be achieved between fitting accuracy within each data source and joint feature selection across data sources by adjusting the regularization parameter of the penalty term in the joint objective function.

[0053] According to embodiments of this application, the risk quantification result is an output value calculated by the risk identification model based on the input sparse feature variables, used to quantify and reflect the customer's risk status. The risk identification model uses the regression coefficient estimation results obtained during training to calculate the input sparse feature variables and output a quantified risk indicator.

[0054] According to embodiments of this application, by acquiring customer behavior data of target objects from various data sources, sparse feature variables with a zero-value ratio exceeding a preset threshold within a preset statistical period are extracted from the customer behavior data. These sparse feature variables are then input into a risk identification model trained based on a joint objective function. This effectively addresses the high sparsity problem of zero-inflation behavioral data in banks. By constraining the parameter set composed of regression coefficients of the same sparse feature variable in various data sources through a non-convex group-level penalty term, the feature redundancy and overfitting problems caused by simply splicing multi-source data in traditional methods are avoided. This achieves cross-data source information sharing and joint screening of risk features, improving the prediction accuracy and stability of the risk identification model in zero-inflation high-dimensional sparse data scenarios.

[0055] Figure 3 The diagram illustrates the overall flowchart of joint learning of multi-source zero-inflation behavioral data of banks according to an embodiment of this application.

[0056] like Figure 3 As shown in the figure, the multi-source data joint risk identification method of this application first obtains customer behavior data from multiple banking business systems. These banking business systems may include core transaction systems, credit business systems, mobile banking access log systems, payment and settlement systems, risk warning systems, customer profiling systems, etc. The data structures, data dimensions, and sample sizes of each data source may differ, exhibiting heterogeneous characteristics.

[0057] According to embodiments of this application, the data sources include banking business systems. A banking business system refers to an information system within a commercial bank used to support various business operations. For example, customer behavior data includes any combination of account transaction records, credit application records, mobile banking access logs, and risk warning data. These data originate from different business systems within the bank and exhibit typical characteristics such as heterogeneity, high dimensionality, and sparsity.

[0058] According to the embodiments of this application, after obtaining the original customer behavior data, data preprocessing operations can be performed. Data preprocessing includes: associating the original customer behavior data belonging to the same target object with a unified customer identifier to obtain associated customer behavior data; and aligning the associated customer behavior data according to a preset time window to obtain the target object's customer behavior data.

[0059] In some embodiments, data cleaning operations can be performed before time alignment of the associated customer behavior data according to a preset time window. For example, missing data handling, anomaly handling, etc.

[0060] Since different business systems typically use different customer identification systems—for example, transaction log systems use customer ID numbers, credit record systems use ID card numbers, and access log systems use usernames—unified identification mapping is used to associate original customer behavior data belonging to the same target object with a unified customer identification, resulting in associated customer behavior data.

[0061] The length of the preset time window can be set in advance according to the business scenario. For example, it can be set to the most recent 12 months in the credit approval scenario, the most recent 3 months in the abnormal transaction scenario, and the most recent 6 months in the post-loan monitoring scenario.

[0062] According to the embodiments of this application, by aligning a unified time window, the behavioral data of the same customer in each data source is ensured to be comparable in the time dimension.

[0063] According to the embodiments of this application, after data preprocessing, zero-inflated variable identification can be performed. Based on the number of times each feature variable in the customer behavior data has zero values ​​within a preset statistical period, the proportion of zero values ​​for each feature variable is determined; feature variables with a proportion of zero values ​​exceeding a preset threshold are identified as zero-inflated variables, thereby forming a zero-inflated feature matrix.

[0064] Specifically, the number of times each feature variable in customer behavior data appears as a zero value in each data source is counted, and the zero value percentage of each feature variable is calculated. The zero value percentage refers to the proportion of samples with a zero value for a particular feature variable in the total number of samples in a single data source. For example, for the feature variable "historical overdue payment count," if 9,500 customers in a credit application record data source have a zero value of 0 for this feature, then the zero value percentage for this feature in that data source is 95%. Feature variables with a zero value percentage exceeding a preset threshold are identified as zero-inflated variables. The preset threshold can be dynamically set based on the total number of samples in each data source and the risk indication strength of the feature variable, or it can be preset as a fixed threshold, for example, 80%. By identifying zero-inflated variables, modeling resources can be concentrated on feature variables with sparse distribution characteristics, avoiding interference from a large number of dense features in model training and improving modeling efficiency.

[0065] According to the embodiments of this application, after zero-inflation variables are identified, feature encoding transformation and numerical standardization can be performed on customer behavior data to eliminate noise and inconsistencies in the original data, providing clean and standardized input data for subsequent modeling.

[0066] Subsequently, a risk identification model is constructed, which can be a joint learning model based on L2 Group MCP. The regression coefficients of the same sparse feature variable (i.e., the identified zero-inflated variable) in different data sources are grouped into a parameter set. Specifically, the regression coefficients of the same sparse feature variable in different data sources are combined into a vector form, and this vector form is used as the parameter set for that sparse feature variable. For example, assuming there are 3 data sources (K=3), the regression coefficients of the j-th sparse feature variable in the 1st, 2nd, and 3rd data sources are respectively... , , Then the parameter set of the j-th sparse feature variable can be expressed as: The dimension of this parameter set is equal to the number of data sources, K, meaning that each sparse feature variable corresponds to a parameter set in the form of a K-dimensional vector.

[0067] According to the embodiments of this application, a joint objective function is further determined. This joint objective function includes the sum of the loss functions from each data source and a non-convex group-level penalty term. The loss function is used to measure the degree of fit of the regression coefficients to the relationship between customer behavior data and risk labels. The loss function for each data source can be a squared loss function or a logistic loss function, and the sum of the loss functions from each data source is the direct sum of the loss functions from all data sources. The non-convex group-level penalty term applies constraints on a per-parameter-group basis. All regression coefficients within the same parameter group are considered holistically and receive a uniform penalty. Risk features are jointly screened through the non-convex group-level penalty mechanism, retaining feature variables that significantly contribute to risk identification and eliminating redundant or noisy variables. The model parameters can be optimized using group coordinate descent to obtain the trained risk identification model.

[0068] After the model is trained, the customer data to be evaluated is input into the trained risk identification model. The output risk identification results are used to support business scenarios such as commercial bank credit approval, abnormal transaction identification, and post-loan risk monitoring. Specifically, it can output risk scores, risk levels, and risk warning results. For example, the risk score can be a comprehensive quantitative score calculated by weighting customer behavior data from various data sources using the optimal regression coefficients, reflecting the overall risk level of customers. Risk warning results can be mapped to a preset risk level range based on the optimal regression coefficient estimation results.

[0069] Figure 4 The flowchart illustrating the training and parameter optimization of the L2 Group MCP joint learning model according to an embodiment of this application is shown.

[0070] like Figure 4 As shown, the model training and parameter optimization process includes: first, inputting a multi-source feature matrix and risk labels. The multi-source feature matrix refers to structured data generated after preprocessing, containing feature variables from various data sources, such as a wide feature table with customers as rows and features from each data source as different field groups. Risk labels are identifiers of known risk outcomes corresponding to each data source. For example, in a credit application record data source, risk labels can be represented as "whether overdue has occurred" (binary label) or "number of overdue days" (continuous label).

[0071] After inputting the data, a joint objective function is constructed. This joint objective function includes the sum of the loss functions from each data source and a non-convex group-level penalty term.

[0072] According to an embodiment of this application, a non-convex group-level penalty term is used as the L2 Group MCP penalty term. The penalty function of the L2 Group MCP penalty term is as follows: for any sparse feature variable in the parameter set composed of regression coefficients in each data source, the value of the penalty function is determined by the Euclidean norm of the parameter set and the minimax concavity adjustment parameter; when the Euclidean norm is less than the product of the minimax concavity adjustment parameter and the regularization parameter, the marginal growth rate of the penalty function decreases linearly with the increase of the Euclidean norm; when the Euclidean norm is greater than or equal to the product, the marginal growth rate of the penalty function is zero. This piecewise form makes the contraction of the large coefficient parameter set by the L2 Group MCP penalty term gradually weaken and eventually disappear as the amplitude of the parameter set increases, thereby avoiding excessive contraction of important risk features with large coefficients.

[0073] According to an embodiment of this application, the specific mathematical form of the joint objective function is as follows. The input for model training includes: a multi-source feature matrix X. (k) Risk label Y (k) The preset regularization parameter λ, the preset MCP adjustment parameter γ, and the initialized regression coefficient β (k) .

[0074] Based on the above inputs, the joint objective function Q is determined. This joint objective function consists of two parts: the first part is the sum of the loss functions from each data source, and the second part is the non-convex group level penalty term. The mathematical expression of the joint objective function is:

[0075]

[0076] Among them, the first term on the right side of the equal sign ;β (k) The sum of loss functions across all data sources measures the accuracy of the regression coefficients in fitting the relationship between customer behavior data and risk labels across each data source; the second term... This is a non-convex group level penalty term used to apply sparsity constraints to each parameter group.

[0077] In the joint objective function above, K represents the total number of data sources, k is the index of the data source, k=1,2,…,K; X(k) represents the feature matrix of the k-th data source, Y(k) represents the risk label corresponding to the k-th data source, and β (k) βj represents the regression coefficient vector corresponding to the k-th data source; P represents the total number of sparse feature variables, j is the index of the sparse feature variable, j=1,2,…,P; βj represents the parameter set of the j-th sparse feature variable, that is, the vector formed by the regression coefficients of the sparse feature variable in each data source. ρ(⋅) represents the L2 norm of parameter group βj, which is used to measure the overall magnitude of the regression coefficients of this sparse feature variable in various data sources; ρ(⋅) is the MCP penalty function; λ is the regularization parameter, which is used to control the overall strength of the penalty term.

[0078] By constructing the joint objective function described above, the fitting accuracy requirements for multi-source data and the joint feature selection requirements across data sources are unified into a single optimization objective. (First term) To ensure the model has good fitting ability across various data sources, the second term... This ensures that only sparse feature variables that significantly contribute to risk identification are retained. The balance between the two is adjusted by the regularization parameter λ: the larger λ is, the more the joint objective function tends to retain fewer sparse feature variables (stronger sparsity); the smaller λ is, the more the joint objective function tends to minimize the fitting error of each data source (higher fitting accuracy).

[0079] According to embodiments of this application, by employing an L2 Group MCP penalty term as a non-convex group-level penalty term, the marginal growth rate of its penalty function decreases linearly with increasing norm when the Euclidean norm of the parameter group is less than the product of the minimax concave adjustment parameter and the regularization parameter, and is zero when the norm is greater than or equal to the product. This piecewise form causes the contraction force of the penalty term on large-coefficient parameter groups to gradually weaken and eventually disappear as the amplitude of the parameter group increases, which is significantly different from the convex penalty function used by Lasso (which applies continuous contraction to large coefficients). Therefore, it effectively avoids the estimation bias caused by excessive contraction of important risk features with large coefficients in traditional Lasso-type penalty methods, ensuring that important features that play a key indicative role in risk identification are accurately preserved, further improving the model's ability to identify sparse important features and the accuracy of parameter estimation in zero-inflation data scenarios.

[0080] According to an embodiment of this application, the training process of the risk identification model includes: constructing a joint objective function based on each data source, the joint objective function including the sum of squared losses on each data source and a non-convex group-level penalty term; iteratively optimizing the joint objective function using a group coordinate descent algorithm until the joint objective function converges, thereby obtaining the trained risk identification model; wherein, in each iteration, the parameter group corresponding to each sparse feature variable is selected sequentially for updating, while the remaining parameter groups in the joint objective function, except for the corresponding parameter groups, remain unchanged.

[0081] Specifically, all other parameter sets in the joint objective function Q are fixed except for the current parameter set. All regression coefficients outside the current parameter set are treated as constants in this update and do not participate in the current optimization. This decomposes the originally high-dimensional joint optimization problem into multiple low-dimensional subproblems. Only the current parameter set is updated, and the optimal value of the current parameter set that minimizes the joint objective function Q is found. The process iterates through all parameter sets sequentially; one iteration means that all parameter sets are updated once. After each iteration, the joint objective function is checked for convergence, for example, by calculating the relative change in the joint objective function |Q| between two adjacent iterations. new -Q| and preset tolerance The comparison is performed. If convergence is not achieved, the next parameter set is used as the current parameter set, and the above fixing and updating operations are performed again until the joint objective function converges.

[0082] In this context, after the current parameter set is updated, the regression coefficients of the same sparse feature variable change synchronously across all data sources. That is, when the j-th parameter set is updated... At this time, all K regression coefficients in the parameter group are updated to new values ​​simultaneously, rather than updating each coefficient individually. This maintains the integrity of the parameter group structure and ensures that group-level penalties are executed at the optimization solution level.

[0083] When the joint objective function converges, the iteration stops, and the optimal parameter estimation results are output. At this point, the risk identification model training is complete. The optimal parameter estimation results can reflect the optimal contribution of each sparse feature variable to risk prediction in each data source—the larger the absolute value of the coefficient, the stronger the indicative role of that feature variable in the corresponding data source for risk events.

[0084] According to embodiments of this application, by employing a group coordinate descent algorithm to iteratively optimize the joint objective function during training, in each iteration, the parameter sets corresponding to each sparse feature variable are sequentially selected for updating while the remaining parameter sets remain unchanged. This decomposes the originally high-dimensional joint optimization problem into multiple low-dimensional sub-problems, significantly reducing the computational complexity of model training. Simultaneously, since all regression coefficients within the same parameter set are updated synchronously, the unified execution of group-level constraints at the optimization solution level is ensured, enabling reliable joint feature selection across data sources at the algorithm level. When the joint objective function converges, the trained risk identification model is obtained, ensuring the optimality of model parameters and the stability of training.

[0085] Figure 5 The diagram illustrates a flowchart of a bank risk identification process according to an embodiment of this application.

[0086] like Figure 5 As shown in the embodiments of this application, the multi-source data joint risk identification method can be applied to the business scenario of commercial bank customer risk identification. The business implementation process includes customer behavior data collection, risk feature extraction, joint learning model analysis, and risk identification result output.

[0087] During the customer behavior data collection phase, customer behavior data is collected from multiple banking business systems. These data sources include banking business systems, and the customer behavior data includes any combination of account transaction data, credit behavior data, account behavior data, and channel behavior data.

[0088] Account transaction data originates from the bank's core transaction system or payment clearing system, recording every customer's transaction and reflecting their transaction activity and abnormal transaction risks. Credit behavior data comes from the bank's credit management system or credit approval system, recording the customer's credit application history and repayment behavior, reflecting their creditworthiness and repayment ability. Account behavior data comes from the bank's mobile channel system, recording customer actions on mobile channels, reflecting their mobile channel usage habits and account security risks. Channel behavior data comes from the bank's risk monitoring system or anti-money laundering system, recording various risk rules or external risk events triggered by the customer, directly reflecting their external risk status and compliance risk level. The data from each data source is heterogeneous, high-dimensional, and sparse.

[0089] In the joint learning model analysis phase, sparse feature variables are input into the pre-trained risk identification model, and the risk quantification results are output. The risk identification model is trained based on the joint objective function, which includes loss function terms from each data source and a non-convex group-level penalty term. The non-convex group-level penalty term acts on the parameter group composed of the regression coefficients of the same sparse feature variable in each data source.

[0090] According to embodiments of this application, the risk quantification result includes at least one of a customer risk score, abnormal behavior probability, and risk warning level. A risk warning signal can be triggered based on whether the customer risk score or abnormal behavior probability exceeds the corresponding warning threshold.

[0091] The customer risk score is a comprehensive quantitative score calculated by weighting customer behavior data from various data sources using the optimal regression coefficients. It reflects the overall risk level of customers, ranking them from highest to lowest; a higher score indicates greater risk. The probability of abnormal behavior is the probability of an abnormal event occurring, determined by a single data source or through joint analysis across data sources, such as the probability of abnormal transactions or overdue defaults. The risk warning level is mapped to a preset risk level range based on the optimal regression coefficient estimation results, such as low-risk, medium-risk, and high-risk ranges. Different risk level ranges correspond to different business handling strategies. A risk warning signal is triggered based on whether the customer risk score or the probability of abnormal behavior exceeds the corresponding warning threshold.

[0092] According to embodiments of this application, by outputting at least one of a customer risk score, abnormal behavior probability, and risk warning level as the risk quantification result, the different needs of various business scenarios for risk quantification methods are met. The score is suitable for customer risk ranking and credit limit allocation, the probability is suitable for determining the probability of abnormal events, and the warning level is suitable for triggering different levels of business handling strategies. By triggering risk warning signals based on whether the customer risk score or abnormal behavior probability exceeds the corresponding warning threshold, automated identification and warning of risk events are achieved, shortening risk response time and improving the real-time and automation level of commercial bank risk management.

[0093] According to embodiments of this application, in credit approval business scenarios, risk quantification results include customer risk scores. Credit recommendations or approval decisions can be output based on a comparison of the customer's risk score with a preset credit approval threshold range. For example, when a customer applies for credit card credit, the model outputs a comprehensive risk score of 752 (out of 1000). The preset credit approval threshold ranges are: high-risk range (0-600 points), medium-risk range (601-800 points), and low-risk range (801-1000 points). Since the customer falls within the medium-risk range, the system automatically outputs a suggested credit limit of 100,000 yuan and marks the credit application as requiring secondary approval from a reviewer.

[0094] According to the embodiments of this application, in the credit approval scenario, credit suggestions or approval decisions are output based on the comparison results between the customer's risk score and the preset credit approval threshold range. This provides an objective and quantitative basis for credit approval, avoids the subjective bias of manual approval, improves the standardization and efficiency of credit approval, and realizes differentiated credit strategies through the fine division of the threshold range.

[0095] According to embodiments of this application, in the abnormal transaction identification business scenario, the risk quantification result includes the probability of abnormal behavior. When the probability of abnormal behavior exceeds a preset abnormal transaction warning threshold, the target object and related transaction records can be marked as pending review. For example, if the abnormal behavior probability output by the model is 12%, which does not exceed the preset abnormal transaction warning threshold (30%), the abnormal transaction review process is not triggered; if the probability of abnormal behavior for a certain transaction reaches 45%, it exceeds the threshold, and the system automatically marks the target object and the transaction record as pending review and pushes them to the abnormal transaction review workbench.

[0096] According to the embodiments of this application, when the probability of abnormal behavior exceeds a preset abnormal transaction warning threshold, the target object and related transaction records are marked as pending review, thereby realizing the automated identification and marking of abnormal transactions. This concentrates limited risk control review resources on transactions and customers that are truly high-risk, improving the targeting and resource utilization efficiency of abnormal transaction review work.

[0097] According to embodiments of this application, in post-loan risk monitoring scenarios, risk quantification results include customer risk scores. Customer behavior data of existing customers can also be acquired at preset intervals and input into the risk identification model. Based on the trend of changes in the output customer risk scores, the risk level of existing customers is updated. For example, for existing customers with disbursed loans, their latest transaction records, repayment records, and access logs are collected monthly and input into the risk identification model to recalculate the risk score. If the score rises from 680 points in the previous period to 830 points, and the risk level moves from the medium-risk range to the high-risk range, the system automatically updates the customer's risk level and triggers the post-loan key monitoring process.

[0098] According to embodiments of this application, by acquiring customer behavior data of existing customers at preset intervals and inputting it into a risk identification model, the risk level of existing customers is updated based on the changing trend of customer risk scores, thus achieving continuous and dynamic monitoring of the risk of existing customers. Compared with traditional static risk rating, this method can promptly detect dynamic changes in customer risk, providing timely risk warning signals for post-loan management, and effectively supporting the dynamic adjustment of the risk level of existing customers and the formulation of differentiated post-loan management strategies.

[0099] Figure 6 A schematic block diagram of a multi-source data joint risk identification device according to an embodiment of this application is shown.

[0100] like Figure 6 As shown, the multi-source data joint risk identification device 600 of this embodiment includes: a data acquisition module 610, a feature extraction module 620, and a risk identification module 630.

[0101] The data acquisition module 610 is used to acquire customer behavior data of the target object from various data sources.

[0102] The feature extraction module 620 is used to extract sparse feature variables from customer behavior data. Sparse feature variables are feature variables in customer behavior data where the proportion of zero values ​​exceeds a preset threshold within a preset statistical period.

[0103] The risk identification module 630 is used to input sparse feature variables into a pre-trained risk identification model and output risk quantification results. The risk identification model is trained based on a joint objective function, which includes loss function terms from each data source and a non-convex group-level penalty term. The non-convex group-level penalty term applies to the parameter group consisting of the regression coefficients of the same sparse feature variable in each data source.

[0104] According to the embodiments of this application, the non-convex group level penalty term is an L2 Group MCP penalty term. The penalty function of the L2 Group MCP penalty term is as follows: for any sparse feature variable in the parameter group consisting of the regression coefficients in each data source, the value of the penalty function is determined by the Euclidean norm of the parameter group and the minimax concavity adjustment parameter; when the Euclidean norm is less than the product of the minimax concavity adjustment parameter and the regularization parameter, the marginal growth rate of the penalty function decreases linearly as the Euclidean norm increases; when the Euclidean norm is greater than or equal to the product, the marginal growth rate of the penalty function is zero.

[0105] According to an embodiment of this application, the training process of the risk identification model includes: constructing a joint objective function based on each data source, the joint objective function including the sum of squared losses on each data source and a non-convex group-level penalty term; iteratively optimizing the joint objective function using a group coordinate descent algorithm until the joint objective function converges, thereby obtaining the trained risk identification model; wherein, in each iteration, the parameter group corresponding to each sparse feature variable is selected sequentially for updating, while the remaining parameter groups in the joint objective function, except for the corresponding parameter groups, remain unchanged.

[0106] According to an embodiment of this application, sparse feature variables are zero-inflated variables; the feature extraction module 620 includes: a zero-value proportion determination unit, used to determine the zero-value proportion of each feature variable based on the number of times each feature variable in the customer behavior data has zero values ​​within a preset statistical period; and a zero-inflated variable identification unit, used to identify feature variables whose zero-value proportion exceeds a preset threshold as zero-inflated variables.

[0107] According to an embodiment of this application, the data acquisition module includes: an identifier mapping unit, used to associate original customer behavior data belonging to the same target object with a unified customer identifier to obtain associated customer behavior data; and a time alignment unit, used to perform time alignment on the associated customer behavior data according to a preset time window to obtain customer behavior data of the target object in each data source.

[0108] According to the embodiments of this application, the risk quantification result includes at least one of customer risk score, abnormal behavior probability, and risk warning level; the device 600 further includes: a warning triggering module, used to trigger a risk warning signal based on whether the customer risk score or abnormal behavior probability exceeds the corresponding warning threshold.

[0109] According to an embodiment of this application, the device 600 further includes at least one of the following modules: a credit approval module, used to output credit recommendations or approval decisions based on a comparison between the customer risk score and a preset credit approval threshold range when the risk quantification result includes a customer risk score; an abnormal transaction identification module, used to mark the target object and related transaction records as pending review when the abnormal behavior probability exceeds a preset abnormal transaction warning threshold when the risk quantification result includes an abnormal behavior probability; and a post-loan monitoring module, used to acquire customer behavior data of existing customers according to a preset period and input it into the risk identification model when the risk quantification result includes a customer risk score, and update the risk level of existing customers based on the output trend of customer risk score changes.

[0110] According to embodiments of this application, any multiple modules among the data acquisition module 610, feature extraction module 620, and risk identification module 630 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some functions of one or more functional modules among these modules can be combined with at least some functions of other functional modules and implemented in one module. According to embodiments of this application, at least one of the data acquisition module 610, feature extraction module 620, and risk identification module 630 can be at least partially implemented as a hardware circuit, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip (SoC), or an integrated circuit on a substrate, or at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0111] Figure 7 A block diagram schematically illustrates an electronic device suitable for implementing a multi-source data joint risk identification method according to an embodiment of this application.

[0112] like Figure 7 As shown, an electronic device 700 according to an embodiment of this application includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage portion 708 into a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a dedicated microprocessor. The processor 701 may also include onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for executing different steps of the method flow according to an embodiment of this application.

[0113] RAM 703 stores various programs and data required for the operation of electronic device 700. Processor 701, ROM 702, and RAM 703 are interconnected via bus 704. Processor 701 executes various steps of the method flow according to embodiments of this application by executing programs in ROM 702 and / or RAM 703. It should be noted that the programs may also be stored in one or more memories other than ROM 702 and RAM 703. Processor 701 may also execute various steps of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0114] According to embodiments of this application, the electronic device 700 may further include an input / output (I / O) interface 705, which is also connected to a bus 704. The electronic device 700 may also include one or more of the following components connected to the I / O interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card, such as a local area network card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 710 as needed so that computer programs read from it can be installed into the storage section 708 as needed.

[0115] Embodiments of this application also provide a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the multi-source data joint risk identification method according to embodiments of this application.

[0116] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In embodiments of this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 702 and / or RAM 703 and / or one or more memories other than ROM 702 and RAM 703 described above.

[0117] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the multi-source data joint risk identification method provided in the embodiments of this application.

[0118] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 709, and / or installed from a removable medium 711. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0119] In embodiments of this application, the computer program can be downloaded and installed from a network via communication section 709, and / or installed from removable medium 711. When the computer program is executed by processor 701, it performs the functions defined in the system of embodiments of this application. According to embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0120] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. The program code can be executed entirely on the user's computing device, partially on the user's device, 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 (e.g., via the Internet using an Internet service provider).

[0121] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0122] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A method for joint risk identification using multi-source data, characterized in that, include: Obtain customer behavior data of the target object from various data sources; Sparse feature variables are extracted from the customer behavior data. The sparse feature variables are feature variables in the customer behavior data in which the proportion of zero values ​​exceeds a preset threshold within a preset statistical period. The sparse feature variables are input into a pre-trained risk identification model, and the risk quantification result is output. The risk identification model is trained based on a joint objective function, which includes loss function terms from each data source and a non-convex group-level penalty term. The non-convex group-level penalty term acts on the parameter group consisting of the regression coefficients of the same sparse feature variable in each data source.

2. The method according to claim 1, characterized in that, The non-convex group-level penalty term is a type-two group minimax concave penalty term, and the penalty function of the type-two group minimax concave penalty term is: For any of the sparse feature variables in the parameter set consisting of the regression coefficients in each of the data sources, the value of the penalty function is determined by the Euclidean norm of the parameter set and the minimum-maximum-concave adjustment parameter. When the Euclidean norm is less than the product of the minimax concavity adjustment parameter and the regularization parameter, the marginal growth rate of the penalty function decreases linearly as the Euclidean norm increases. When the Euclidean norm is greater than or equal to the product, the marginal growth rate of the penalty function is zero.

3. The method according to claim 1, characterized in that, The training process of the risk identification model includes: Based on the data sources, a joint objective function is constructed, which includes the sum of squared losses on each data source and the non-convex group level penalty term; The joint objective function is iteratively optimized using a group coordinate descent algorithm until it converges, thus obtaining the trained risk identification model. In each iteration, the parameter set corresponding to each sparse feature variable is selected and updated sequentially, while the remaining parameter sets in the joint objective function, except for the corresponding parameter sets, remain unchanged.

4. The method according to claim 1, characterized in that, The sparse feature variables are zero-inflated variables; the extraction of sparse feature variables from the customer behavior data includes: Based on the number of times each feature variable in the customer behavior data has zero values ​​within a preset statistical period, the percentage of zero values ​​for each feature variable is determined. The feature variable whose proportion of zero values ​​exceeds a preset threshold is identified as a zero-inflation variable.

5. The method according to claim 1, characterized in that, The acquisition of customer behavior data of the target object from various data sources includes: The original customer behavior data belonging to the same target object is associated with a unified customer identifier to obtain the associated customer behavior data. The associated customer behavior data is time-aligned according to a preset time window to obtain the customer behavior data of the target object.

6. The method according to claim 1, characterized in that, The risk quantification result includes at least one of customer risk score, probability of abnormal behavior, and risk warning level; the method further includes: A risk warning signal is triggered based on whether the customer's risk score or the probability of abnormal behavior exceeds the corresponding warning threshold.

7. The method according to claim 6, characterized in that, The risk quantification result includes a customer risk score; the method further includes: outputting a credit recommendation or approval decision based on a comparison between the customer risk score and a preset credit approval threshold range; and / or, The risk quantification result includes the probability of abnormal behavior; the method further includes: when the probability of abnormal behavior exceeds a preset abnormal transaction warning threshold, marking the target object and related transaction records as pending review; and / or, The risk quantification result includes a customer risk score; the method further includes: acquiring customer behavior data of existing customers according to a preset period and inputting it into the risk identification model, and updating the risk level of existing customers according to the trend of changes in the output customer risk score.

8. A multi-source data joint risk identification device, characterized in that, include: The data acquisition module is used to acquire customer behavior data of the target object from various data sources; The feature extraction module is used to extract sparse feature variables from the customer behavior data. The sparse feature variables are feature variables in the customer behavior data where the proportion of zero values ​​exceeds a preset threshold within a preset statistical period. The risk identification module is used to input the sparse feature variables into a pre-trained risk identification model and output the risk quantification result. The risk identification model is trained based on a joint objective function, which includes loss function terms from each data source and a non-convex group-level penalty term. The non-convex group-level penalty term applies to the parameter group consisting of the regression coefficients of the same sparse feature variable in each data source.

9. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

Images