Client credit evaluation model construction method and device

Abstract

The invention provides a method and a device for constructing a customer credit evaluation model, which are applied to the technical field of computers, and comprise the following steps: processing initial data of each data source by using a truth value discovery algorithm to obtain a training data set; training each initial sub-model by using the training data set to obtain a credit evaluation sub-model corresponding to each initial sub-model; performing fusion processing on the model parameters of the credit evaluation sub-models by using a truth value discovery algorithm to obtain fused model parameters; and constructing a customer credit evaluation model by using the fused model parameters. A truth value discovery algorithm is used for processing the data, the processed data is used for training a plurality of models, the trained models are fused to obtain a customer credit evaluation model, the customer credit evaluation model can process the data of a plurality of data sources, and evaluation results of the models are fused to obtain a customer credit evaluation result. The error of credit evaluation by a single model is reduced, and the accuracy of credit evaluation is improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and device for constructing a customer credit assessment model. Background Art

[0002] In the financial field, financial demand is increasing day by day, and financial risks are also increasing. In order to help financial institutions better manage risks and improve credit quality, credit assessment models have emerged. Credit assessment models are important tools for assessing personal or corporate credit risks.

[0003] The inventors have found that the credit assessment model currently used adopts a single algorithm for risk assessment. This singleness may lead to limitations in the model when processing complex and changeable financial data, and it is unable to fully and accurately capture the credit risk characteristics of users, which directly affects the accuracy of the credit assessment model output. Summary of the invention

[0004] In view of this, an embodiment of the present application provides a method and device for constructing a customer credit assessment model. The present application is applied to process data from multiple data sources using a truth discovery algorithm, and multiple sub-models are trained using the processed data to improve the robustness of the sub-models. The trained sub-models are then integrated into a customer credit assessment model. The customer credit assessment model can process data from multiple data sources and integrate the assessment results of multiple models, thereby reducing the error of credit assessment performed by a single model and improving the accuracy of customer credit assessment.

[0005] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0006] A method for constructing a customer credit assessment model, comprising:

[0007] Get the initial data of each data source;

[0008] Using a truth value discovery algorithm to process the initial data of each of the data sources to obtain credibility information;

[0009] Using the credibility information to process each of the initial data to obtain a training data set;

[0010] Using the training data set to train each preset initial sub-model to obtain a credit assessment sub-model corresponding to each of the initial sub-models;

[0011] Using a truth value discovery algorithm to fuse the model parameters of each of the credit assessment sub-models to obtain fused model parameters;

[0012] Use the fused model parameters to build a customer credit assessment model.

[0013] In the above method, optionally, the using of a truth value discovery algorithm to process the initial data of each data source to obtain credibility information includes:

[0014] Based on the initial data of each of the data sources, obtaining the data of each of the preset data items of each of the data sources;

[0015] Setting the data credibility of each of the data sources;

[0016] For each of the data items, using the data credibility of each of the data sources and the data of each of the data sources in the data item, determine the data true value of the data item;

[0017] For each of the data sources, determining a data error of the data source in each of the data items, and updating the data credibility of the data source based on each of the data errors of the data source;

[0018] Acquire first convergence data, and determine whether the first convergence data meets a preset first convergence condition;

[0019] If the first converged data satisfies the first convergence condition, generating credibility information using the data true value of each data item of each data source and the data credibility of each data source;

[0020] If the first converged data does not satisfy the convergence condition, the process returns to execute the step of determining the data truth value of each data item using the data credibility of each data source and the data of each data source in the data item.

[0021] In the above method, optionally, the obtaining of first convergence data and determining whether the first convergence data satisfies a preset first convergence condition includes:

[0022] Based on the true values ​​of the data items, determining the true value change value of each of the data items;

[0023] Updating the first iteration number, and generating first convergence data based on the updated first iteration number and the true value change value of each of the data items;

[0024] When the first iteration number in the first convergence data reaches a preset number, and / or the true value change value of each data item in the first convergence data is within the corresponding change interval, it is determined that the first convergence data satisfies the first convergence condition; otherwise, it is determined that the first convergence data does not satisfy the first convergence condition.

[0025] In the above method, optionally, the use of a truth value discovery algorithm to fuse the model parameters of each credit assessment sub-model to obtain the fused model parameters includes:

[0026] Setting the parameter credibility of each credit assessment sub-model;

[0027] Obtaining model parameters of each of the credit assessment sub-models;

[0028] Determine the true value of the parameter using the parameter credibility and model parameters of each credit assessment sub-model;

[0029] Determining the distance value between the model parameter of each credit assessment sub-model and the true value of the parameter;

[0030] Updating the parameter credibility of each credit assessment sub-model based on each of the distance values;

[0031] Acquire second convergence data, and determine whether the second convergence data meets a preset second convergence condition;

[0032] If the second convergence data does not satisfy the second convergence condition, returning to the step of determining the true value of the parameter using the parameter credibility and model parameters of each credit assessment sub-model;

[0033] If the second convergence data satisfies the second convergence condition, the parameter credibility and model parameters of each credit assessment sub-model are processed according to a preset algorithm to obtain fused model parameters.

[0034] The above method may optionally further include:

[0035] Obtain the evaluation data of the customer to be evaluated;

[0036] The customer credit evaluation model is called to process the data to be evaluated, and the credit evaluation information of the customer is output.

[0037] A device for constructing a customer credit assessment model, comprising:

[0038] A first acquisition unit, used to acquire initial data from each data source;

[0039] A first processing unit, configured to process the initial data of each data source using a truth value discovery algorithm to obtain credibility information;

[0040] A second processing unit, configured to process each of the initial data using the credibility information to obtain a training data set;

[0041] A training unit, used to train each preset initial sub-model using the training data set to obtain a credit assessment sub-model corresponding to each of the initial sub-models;

[0042] A third processing unit is used to use a truth value discovery algorithm to perform fusion processing on the model parameters of each credit assessment sub-model to obtain fused model parameters;

[0043] The building unit is used to build a customer credit assessment model using the fused model parameters.

[0044] In the above device, optionally, the first processing unit executes a process of using a truth value discovery algorithm to process the initial data of each data source to obtain credibility information, including:

[0045] Based on the initial data of each of the data sources, obtaining the data of each of the preset data items of each of the data sources;

[0046] Setting the data credibility of each of the data sources;

[0047] For each of the data items, using the data credibility of each of the data sources and the data of each of the data sources in the data item, determine the data true value of the data item;

[0048] For each of the data sources, determining a data error of the data source in each of the data items, and updating the data credibility of the data source based on each of the data errors of the data source;

[0049] Acquire first convergence data, and determine whether the first convergence data meets a preset first convergence condition;

[0050] If the first converged data satisfies the first convergence condition, generating credibility information using the data true value of each data item of each data source and the data credibility of each data source;

[0051] If the first converged data does not satisfy the convergence condition, the process returns to execute the step of determining the data truth value of each data item using the data credibility of each data source and the data of each data source in the data item.

[0052] In the above device, optionally, the first processing unit executes a process of acquiring first convergence data and determining whether the first convergence data satisfies a preset first convergence condition, including:

[0053] Based on the true values ​​of the data items, determining the true value change value of each of the data items;

[0054] Updating the first iteration number, and generating first convergence data based on the updated first iteration number and the true value change value of each of the data items;

[0055] When the first iteration number in the first convergence data reaches a preset number, and / or the true value change value of each data item in the first convergence data is within the corresponding change interval, it is determined that the first convergence data satisfies the first convergence condition; otherwise, it is determined that the first convergence data does not satisfy the first convergence condition.

[0056] In the above device, optionally, the third processing unit performs a process of fusing the model parameters of each credit assessment sub-model using a truth value discovery algorithm to obtain the fused model parameters, including:

[0057] Setting the parameter credibility of each credit assessment sub-model;

[0058] Obtaining model parameters of each of the credit assessment sub-models;

[0059] Determine the true value of the parameter using the parameter credibility and model parameters of each credit assessment sub-model;

[0060] Determining the distance value between the model parameter of each credit assessment sub-model and the true value of the parameter;

[0061] Updating the parameter credibility of each credit assessment sub-model based on each of the distance values;

[0062] Acquire second convergence data, and determine whether the second convergence data meets a preset second convergence condition;

[0063] If the second convergence data does not satisfy the second convergence condition, returning to the step of determining the true value of the parameter using the parameter credibility and model parameters of each credit assessment sub-model;

[0064] If the second convergence data satisfies the second convergence condition, the parameter credibility and model parameters of each credit assessment sub-model are processed according to a preset algorithm to obtain fused model parameters.

[0065] The above device may optionally further include:

[0066] A second acquisition unit, used for acquiring the to-be-evaluated data of the to-be-evaluated customer;

[0067] The calling unit is used to call the customer credit evaluation model to process the data to be evaluated and output the credit evaluation information of the user.

[0068] Compared with the prior art, this application has the following advantages:

[0069] The present application provides a method and device for constructing a customer credit assessment model, including: obtaining initial data from each data source; using a truth value discovery algorithm to process the initial data from each data source to obtain credibility information; using the credibility information to process each initial data to obtain a training data set; using the training data set to train each preset initial sub-model to obtain a credit assessment sub-model corresponding to each initial sub-model; using a truth value discovery algorithm to fuse the model parameters of each credit assessment sub-model to obtain fused model parameters; using the fused model parameters to construct a customer credit assessment model. The present application uses a truth value discovery algorithm to process data to obtain highly reliable training data, uses the obtained training data to train a model, so that the accuracy of the trained model is higher, and then uses a truth value discovery algorithm to fuse the trained models to obtain a final customer credit assessment model, which can use data from multiple sources and can fuse the assessment results of multiple models to avoid the assessment error of a single model and improve the accuracy of the output assessment results. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0071] Figure 1 A flowchart of a method for constructing a customer credit assessment model provided in an embodiment of the present application;

[0072] Figure 2 A flow chart of a method for processing initial data of each data source using a truth value discovery algorithm to obtain credibility information provided in an embodiment of the present application;

[0073] Figure 3 A flowchart of using a truth value discovery algorithm to fuse model parameters of each credit assessment sub-model to obtain fused model parameters provided in an embodiment of the present application;

[0074] Figure 4 An example flow chart of building a customer credit assessment model provided in an embodiment of the present application;

[0075] Figure 5 A flowchart of using a truth value discovery algorithm to perform data preprocessing on an original data set provided in an embodiment of the present application;

[0076] Figure 6 A schematic diagram of the structure of a device for constructing a customer credit assessment model provided in an embodiment of the present application;

[0077] Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0078] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0079] In this application, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.

[0080] Terminology explanation:

[0081] Truth value discovery: refers to the process of finding the most accurate observation value of the same object from multiple contradictory or conflicting data sources. Truth value discovery technology is widely used in large-scale data processing and data analysis. In the data cleaning and preprocessing stage, the truth value refers to the correctness and completeness of the data record. The goal of truth value discovery is to identify and correct erroneous or inconsistent data to improve data quality; in the prediction model, the truth value refers to the consistency between the model prediction results and the actual results. The goal of truth value discovery is to improve the prediction accuracy of the model.

[0082] Robustness: In the field of artificial intelligence, robustness refers to the ability of machine learning algorithms to handle abnormal data such as noise and missing values. The more a system, algorithm or model can maintain stable performance in the face of abnormal inputs, uncertain environments or parameter changes, the more robust it is.

[0083] Credit assessment model: a tool used to identify, analyze and predict credit risk. It predicts the credit risk of customers by analyzing data such as customer historical behavior and financial status. It is often used in financial institutions' credit decision-making and other links. By establishing a customer credit assessment model, financial institutions can improve risk control capabilities, better understand customer needs and preferences, promote business development, and improve decision-making efficiency.

[0084] Ensemble learning: It is a machine learning method that improves the overall prediction performance by combining the predictions of multiple learners (usually called "base learners"), making up for the shortcomings of a single model in some cases. Ensemble learning does not refer to a specific algorithm, but a training method. The advantage of ensemble learning is that it significantly improves model performance and reduces the bias and variance of the model by reasonably combining multiple models, thereby improving the generalization ability of the model.

[0085] As the background technology shows, the traditional credit assessment model uses a single algorithm for assessment, which has problems of overfitting and weak scalability, resulting in reduced model accuracy on new data sets; in addition to these problems, when faced with data provided by different data sources, the traditional credit assessment model has difficulty in judging conflicting data in different data sources, and is also unable to judge the authenticity of the data provided by the data source, resulting in low accuracy of the credit assessment model output.

[0086] In order to solve the above-mentioned problems, the present application provides a solution for constructing a customer credit assessment model. Using the solution provided by the present application, a customer credit assessment model can be constructed. When constructing the customer credit assessment model, the initial data of each data source is processed using a truth discovery algorithm to obtain a training data set, and then the training data set is used to train each initial sub-model to obtain an evaluation sub-model, and then the model parameters of each evaluation sub-model are fused, and the fused model parameters are used to construct a customer credit assessment model, thereby improving the reliability and quality of the data, and using the model parameters after fusing the model parameters of multiple evaluation sub-models to construct a model, effectively avoiding the limitations of single model prediction, and the model can be applied to a variety of scenarios, improving the generalization and prediction accuracy of the model.

[0087] The present application can be used in many general or special computing device environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multi-processor devices, distributed computing environments including any of the above devices or equipment, etc.

[0088] This application can be applied to various financial institutions in the financial field, such as banks. Banks can use this solution to build a model, and then use the model to assess the credit risk of customers, and then provide corresponding services to customers based on the assessment results to avoid putting themselves at risk.

[0089] Reference Figure 1 , which is a flow chart of a method for constructing a customer credit assessment model provided in an embodiment of the present application, and is specifically described as follows:

[0090] S101: Obtain initial data from each data source.

[0091] The various data sources cover internal and external data sources, and the various data sources include but are not limited to credit reporting agencies, banks, other financial institutions, etc. The various data sources may also be databases of different financial institutions.

[0092] The initial data includes but is not limited to customer transaction history, credit record, assets and other financial data.

[0093] S102: Use a truth-value discovery algorithm to process the initial data of each data source to obtain credibility information.

[0094] Reference Figure 2 , which is a flow chart of a method for processing the initial data of each data source using a truth value discovery algorithm to obtain credibility information provided by an embodiment of the present application, and is specifically described as follows:

[0095] S201. Based on the initial data of each data source, obtain the data of each preset data item of each data source.

[0096] When obtaining the data of each preset data item of each data source, each initial data is first cleaned to deduplicate the initial data and process the missing values ​​in the initial data to obtain the cleaned data of each initial data; for each data source, the data of each data item is extracted from the cleaned data of the data source, and further, the extracted data can be standardized, for example, the category features are encoded, so that the data of each data item of each data source is standardized data.

[0097] Preferably, the data items can be regarded as feature items, and the extracted data can be regarded as feature data. Exemplarily, each data item includes but is not limited to key feature items such as income, credit card usage frequency, and debt ratio.

[0098] S202: Set the data credibility of each data source.

[0099] Preferably, when setting the data credibility of each data source, it can be set according to actual needs, and the same value can be used as the data credibility of each data source.

[0100] S203: For each data item, determine the data truth value of the data item using the data credibility of each data source and the data of each data source in the data item.

[0101] Preferably, the data true value of the data item can be expressed as: in, Represents the data truth value of data item i; Represents the data of data source k in data item i; Indicates the data provided by data source k in data item i The true value of data item i further, the smaller the distance is, the closer the data provided by the data source for the data item is to the true value, and the higher the credibility of the data source is; conversely, the farther the data provided by the data source for the data item is from the true value, the lower the credibility of the data source is.

[0102] The data credibility of each data source and the data of each data source in a certain data item can be brought into the data truth value expression of the above data item, so as to obtain the data truth value of the data item.

[0103] Preferably, the derivation formula of the weight value of the data source (the weight value of the data source is the data credibility of the data source) and the true value of the data source can be:

[0104]

[0105] Among them, Min f() represents the minimization loss function; V i * represents the true value of data item i; W i represents the numerical weight of data item i; W represents the data source weight set; k represents the kth data source; K represents the total number of data sources; m represents the attribute value dimension of data item i; δ(W) represents the weight normalization function; S represents that W must meet the specified threshold; std() represents the calculated standard deviation;

[0106] in, Represents the data truth value of data item i; Represents the data of data source k in data item i; Indicates the data provided by data source k in data item i The true value of data item i The distance between is the loss function, which is used to calculate the true value of data item i The deviation between the data from data source k on data item i.

[0107] S204: For each data source, determine the data error of the data source in each data item, and update the data credibility of the data source based on each data error of the data source.

[0108] The above loss function can be used to determine the data error of each data source in each data item Calculation: For each data source, when calculating the data error of the data source in a certain data item, the data of the data source in the data item and the true value of the data item are substituted into the above loss function. In this way, the data error of the data source in the data item is obtained.

[0109] For each data source, the data error of each data item of the data source is obtained, and then the data reliability of the data source is adjusted using these data errors. Adjusting the data reliability of the data source is to update the data reliability of the data source.

[0110] After the data credibility of the data source is updated, step S205 is executed.

[0111] S205, obtaining first convergence data, and determining whether the first convergence data meets a preset first convergence condition; if the first convergence data meets the first convergence condition, executing S206; if the first convergence data does not meet the convergence condition, returning to executing step S203.

[0112] The process of obtaining the first convergence data is as follows:

[0113] Based on the true values ​​of the data items, determining the true value change value of each data item;

[0114] The first iteration number is updated, and first convergence data is generated based on the updated first iteration number and the true value change value of each data item.

[0115] When determining the true value change of each data item, the current data true value and the previous data true value of the data item can be used for calculation to obtain the true change value of the data item; preferably, the true value change value of the data item can be used to represent the credibility of the current data true value of the data item. The smaller the true value change value of the data item, the higher the credibility of the current data true value of the data item, and vice versa.

[0116] In the present application, any method that can calculate the true value change of a data item can be applied.

[0117] It should be noted that the initial value of the first iteration number is 0. When updating the first iteration number, it can be updated by adding one. The updated first iteration number can indicate the number of iterations. For example, after the first iteration, the updated first iteration number is 1. After the second iteration, the updated first iteration number is 2. And so on. It will not be explained here.

[0118] The first convergence data includes the updated first iteration number and the true value change value of each data item.

[0119] The process of determining whether the first convergence data meets the preset first convergence condition is as follows:

[0120] When the first iteration number in the first convergence data reaches a preset number, and / or the true value change value of each data item in the first convergence data is within the corresponding change interval, it is determined that the first convergence data satisfies the first convergence condition; otherwise, it is determined that the first convergence data does not satisfy the first convergence condition.

[0121] The preset number of times can be set according to actual needs. The change intervals of different data items are different, and the change intervals of different data items can be set according to actual needs.

[0122] Further, when the first convergence data does not satisfy the first convergence condition, the process returns to step S203 and the process is iterated continuously until the first convergence data satisfies the first convergence condition, and then step S206 is executed.

[0123] After returning to execute step S203, the data credibility of the data source applied in this step is the updated data credibility.

[0124] S206: Generate credibility information using the data truth value of each data item of each data source and the data credibility of each data source.

[0125] Furthermore, the credibility information includes the data credibility of each data source and the data true value of each data source in each data item.

[0126] Furthermore, Figure 2 The content shown can be regarded as using the truth value discovery algorithm to process the initial data, iteratively derive the multi-source data, perform feature selection and feature importance evaluation, solve the inconsistency problem between the data, and avoid the conflict between the data provided by different data sources, thereby improving the data reliability of the training data set determined by the subsequent application credibility information.

[0127] S103: Use the credibility information to process each initial data to obtain a training data set.

[0128] When using the credibility information to process each initial data, the credibility information can be used to screen each initial data, thereby obtaining reliable and high-quality data, and the obtained data can be used to form a training data set. Preferably, the training data set can be constructed using the true value of each data in the credibility information.

[0129] Preferably, the credibility information obtained by the true value algorithm is used to screen and identify the data, and by calculating the contribution of each feature to the target variable, it is determined which features have a greater impact on the target variable, and then the true value of the data feature is obtained by weighted averaging, thereby obtaining a training data set.

[0130] Exemplarily, when the credibility information is used to process each initial data, data with high credibility is preferentially selected as the training data set, thereby improving the data quality and data reliability of the training data set.

[0131] S104: Use the training data set to train each preset initial sub-model to obtain a credit assessment sub-model corresponding to each initial sub-model.

[0132] Preferably, the initial sub-model can be regarded as a base learner, and the algorithms of different initial sub-models are different, such as decision tree, random forest, gradient boosting tree, support vector machine, neural network, etc.

[0133] Each initial sub-model is trained using the training data set provided in S103. After each initial sub-model is trained, a credit assessment sub-module of each initial sub-model is obtained.

[0134] S105. Use a truth value discovery algorithm to fuse the model parameters of each credit assessment sub-model to obtain fused model parameters.

[0135] Reference Figure 3 , which is a flowchart of using the truth value discovery algorithm provided in the embodiment of the present application to fuse the model parameters of each credit assessment sub-model to obtain the fused model parameters, and the specific description is as follows:

[0136] S301. Set the parameter credibility of each credit assessment sub-model.

[0137] A numerical value can be set as the parameter credibility of all credit assessment sub-models.

[0138] Preferably, the parameter credibility of the credit assessment sub-model can be used as the parameter weight of the credit assessment sub-model.

[0139] S302: Obtain model parameters of each credit assessment sub-model.

[0140] The model parameters of the credit assessment sub-model are the parameters obtained after the credit assessment sub-model training is completed.

[0141] S303: Use the parameter credibility and model parameters of each credit assessment sub-model to determine the true value of the parameter.

[0142] Preferably, the total number of credit assessment sub-models is represented by n, and the parameter credibility of the credit assessment sub-model is represented by u, where u j Represents the parameter credibility of credit assessment sub-model j.

[0143] The model parameter of the credit assessment sub-model is expressed as N, as follows: j represents the model parameter of the credit assessment sub-model j, j∈n; the true value of the parameter is represented by N * .

[0144] The parameter credibility and model parameters of each credit assessment sub-model are calculated to obtain the true value of the parameter, and the algorithm is continuously iterated until it reaches stability. For example, The description of the relevant parameters in the formula will not be repeated here.

[0145] S304: Determine the distance value between the model parameter of each credit assessment sub-model and the true value of the parameter.

[0146] A preset distance function may be used to determine the distance value between the model parameter and the true value of the parameter of each credit assessment sub-model.

[0147] The distance function can be expressed as: Among them, N * is the true value of the parameter, N j represents the model parameters of the credit assessment sub-model j, d(N j ,N * ) represents the distance value between the credit assessment sub-model j and the true value of the parameter.

[0148] S305: Update the parameter credibility of each credit assessment sub-model based on each distance value.

[0149] Preferably, the parameter credibility of the credit assessment sub-model can be expressed as: This formula can be used to update the parameter credibility of each credit assessment sub-model. The description of each parameter in the formula can refer to the relevant description above and will not be repeated here.

[0150] S306, obtaining second convergence data, and determining whether the second convergence data meets a preset second convergence condition; if the second convergence data does not meet the second convergence condition, returning to step S303; if the second convergence data meets the second convergence condition, then S307.

[0151] When obtaining the second convergence data, the second iteration number is updated, and the loss value is obtained according to the true value discovery algorithm, and the second convergence data is obtained based on the updated second iteration number and the loss value.

[0152] The method for updating the second number of iterations may refer to the method for updating the first number of iterations described above, and will not be described in detail here.

[0153] When the second iteration number in the second convergence data reaches the preset iteration number, and / or the convergence value in the second convergence data is less than or equal to the preset convergence value, it is determined that the second convergence data satisfies the second convergence condition; otherwise, it is determined that the second convergence data does not satisfy the second convergence condition.

[0154] The preset number of iterations and the preset convergence value can be set according to actual needs.

[0155] After determining that the second convergence data meets the second convergence condition, step S307 is executed; when determining that the second convergence data does not meet the second convergence condition, the process returns to step S303 until the second convergence condition is met.

[0156] S307: Process the parameter credibility and model parameters of each credit assessment sub-model according to a preset algorithm to obtain fused model parameters.

[0157] Preferably, the parameter credibility of the credit assessment sub-model here is the parameter credibility of the credit assessment sub-model when the second convergence condition is satisfied.

[0158] The parameter credibility and model parameters of each credit assessment sub-model are brought into the preset formula for fusion processing to obtain the fused model parameters.

[0159] The fused model parameters can be expressed as: Among them, N g is the fused model parameter. The meanings of the remaining parameters refer to the above description and will not be repeated here.

[0160] S106. Use the fused model parameters to build a customer credit assessment model.

[0161] After obtaining the fused model parameters, the fused model parameters are used to build a customer credit assessment model. The customer credit assessment model can utilize information from multiple sources when assessing the customer's credit. The customer credit assessment model will combine the prediction results of multiple models, which can reduce the prediction error of a single model, improve the generalization of the model, and improve the accuracy of the credit assessment results.

[0162] Preferably, after obtaining the customer credit assessment model, the model can be deployed in a production environment, and the model can be retrained and updated iteratively based on the latest test data, so as to continuously optimize the model and improve the accuracy of the data output by the model.

[0163] Preferably, after obtaining the customer credit assessment model, the customer credit assessment model is used to perform credit assessment, and the specific process is as follows:

[0164] Obtain the data to be evaluated of the customer to be evaluated; call the customer credit evaluation model to process the data to be evaluated, and output the customer's credit evaluation information.

[0165] In the method provided in the embodiment of the present application, the initial data of each data source is obtained; the initial data of each data source is processed using a truth value discovery algorithm to obtain credibility information; each initial data is processed using the credibility information to obtain a training data set; each preset initial sub-model is trained using the training data set to obtain a credit assessment sub-model corresponding to each initial sub-model; the model parameters of each credit assessment sub-model are fused using a truth value discovery algorithm to obtain fused model parameters; and a customer credit assessment model is constructed using the fused model parameters. The present application uses a truth value discovery algorithm to process data to obtain highly reliable training data, uses the obtained training data to train a model, so that the trained model has higher accuracy, and then uses a truth value discovery algorithm to fuse the trained models to obtain a final customer credit assessment model, which can use data from multiple sources and can fuse the assessment results of multiple models to avoid the assessment error of a single model and improve the accuracy of the output assessment results.

[0166] Reference Figure 4 , is an example flow chart of building a customer credit assessment model provided in an embodiment of the present application, and is specifically described as follows:

[0167] First, obtain the original data set from multiple data sources. The original data set contains the initial data of each data source. Use the truth value discovery algorithm to preprocess the original data set. Figure 5 , which is a flow chart of using a truth value discovery algorithm to perform data preprocessing on an original data set provided in an embodiment of the present application. After the original data set is preprocessed using the truth value discovery algorithm, a training data set is obtained; each sub-model (equivalent to the initial sub-model mentioned above) is trained using the training data set, and after each sub-model is successfully trained, the truth value discovery algorithm is used to perform model fusion processing on each successfully trained sub-model, thereby obtaining a customer credit assessment model.

[0168] right Figure 5 The process shown is described as follows:

[0169] Set the weight value of each data source (equivalent to the data credibility mentioned above), determine the data of each data source in each data item; use the data of each data source in each data item and the weight value of each data source to determine the true value of each data item; for each data source, use the data of the data source in each data item and the true value of each data item to determine the loss value of the data source in each data item (equivalent to the data error mentioned above); update the weight value of each data source based on the respective loss values ​​of each data source; determine whether convergence occurs; if not, return to execute the step of determining the true value of each data item using the data of each data source in each data item and the weight value of each data source, and if convergence occurs, output the true value data; the output true value data includes the true value of each data source in each data item.

[0170] The existing credit assessment model has different credibility of data sources during data preprocessing. The data provided by different data sources are likely to conflict, making it difficult to determine which data sources provide true information; a single credit assessment model often has problems of overfitting and weak scalability, resulting in reduced model accuracy on new data sets; in the existing integrated learning credit assessment model, an average or voting strategy is usually used to combine the prediction results of the model, but these methods cannot effectively deal with noise or outliers in the data, and cannot aggregate models according to the different contributions of different models, so the robustness and fairness are poor. This application adopts an integrated learning model to obtain a customer credit assessment module by constructing, training and fusing multiple models. The customer credit assessment module makes up for the limitations of single model prediction, thereby improving the generalization ability of the model and the accuracy of the overall prediction of the model; in view of the problem of multiple data sources and uneven data quality, this solution introduces a truth value discovery algorithm in the data preprocessing stage to perform feature selection and feature importance evaluation, solves the problem of inconsistency between data, and improves the reliability of training data; in view of the different contribution levels of sub-models of integrated learning, this solution designs a credibility evaluator based on the truth value discovery algorithm in the model integration stage. By introducing the truth discovery algorithm twice, before and after model training, a comprehensive credit assessment model is constructed, which maximizes the robustness and accuracy of the prediction model.

[0171] In the process of constructing a customer credit assessment model in the present application, before model training, the present application proposes to use a truth value discovery algorithm to iteratively derive the credibility and true value of multi-source data, thereby improving the credibility of the training data and further improving the robustness of the credit assessment model; in the model training stage, the present application optimizes the model integration steps by dynamically adjusting the weights of different models by introducing a truth value discovery algorithm, thereby improving the fairness and accuracy of the credit assessment model; compared with the traditional use of ordinary machine learning models to train and evaluate models, the integrated learning model adopted in the present application combines the prediction results of multiple models, which can reduce the prediction error of a single model and improve the generalization of the model; using the solution provided by the present application, a customer credit assessment integrated learning model is constructed that can fully utilize multi-source information while maintaining accuracy and robustness, thereby improving the accuracy of the credit assessment results for customers.

[0172] Although the present invention depicts operations in a particular order, this should not be understood as requiring that the operations be performed in the particular order shown or in a sequential order.Multitasking and parallel processing may be advantageous under certain circumstances.

[0173] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0174] The present application also provides a device for constructing a customer credit assessment model, which is arranged on a processor or a server and is used to support Figure 1 Implementation of the method shown.

[0175] Reference Figure 6 , is a schematic diagram of a structure of a device for constructing a customer credit assessment model provided in an embodiment of the present application, and is specifically described as follows:

[0176] A first acquisition unit 401 is used to acquire initial data from each data source;

[0177] A first processing unit 402 is used to process the initial data of each data source using a truth value discovery algorithm to obtain credibility information;

[0178] A second processing unit 403 is used to process each of the initial data using the credibility information to obtain a training data set;

[0179] A training unit 404 is used to train each preset initial sub-model using the training data set to obtain a credit assessment sub-model corresponding to each of the initial sub-models;

[0180] The third processing unit 405 is used to use a truth value discovery algorithm to perform fusion processing on the model parameters of each credit assessment sub-model to obtain fused model parameters;

[0181] The construction unit 406 is used to construct a customer credit assessment model using the fused model parameters.

[0182] In the device provided by the embodiment of the present application, the initial data of each data source is obtained; the initial data of each data source is processed using a truth value discovery algorithm to obtain credibility information; each initial data is processed using the credibility information to obtain a training data set; each preset initial sub-model is trained using the training data set to obtain a credit assessment sub-model corresponding to each initial sub-model; the model parameters of each credit assessment sub-model are fused using a truth value discovery algorithm to obtain fused model parameters; and a customer credit assessment model is constructed using the fused model parameters. The present application uses a truth value discovery algorithm to process data to obtain highly reliable training data, uses the obtained training data to train a model, so that the trained model has higher accuracy, and then uses a truth value discovery algorithm to fuse the trained models to obtain a final customer credit assessment model, which can use data from multiple sources and can fuse the assessment results of multiple models to avoid the assessment error of a single model and improve the accuracy of the output assessment results.

[0183] In another embodiment provided by the present application, the first processing unit of the device executes a process of using a truth value discovery algorithm to process the initial data of each data source to obtain credibility information, including:

[0184] Based on the initial data of each of the data sources, obtaining the data of each of the preset data items of each of the data sources;

[0185] Setting the data credibility of each of the data sources;

[0186] For each of the data items, using the data credibility of each of the data sources and the data of each of the data sources in the data item, determine the data true value of the data item;

[0187] For each of the data sources, determining a data error of the data source in each of the data items, and updating the data credibility of the data source based on each of the data errors of the data source;

[0188] Acquire first convergence data, and determine whether the first convergence data meets a preset first convergence condition;

[0189] If the first converged data satisfies the first convergence condition, generating credibility information using the data true value of each data item of each data source and the data credibility of each data source;

[0190] If the first converged data does not satisfy the convergence condition, the process returns to execute the step of determining the data truth value of each data item using the data credibility of each data source and the data of each data source in the data item.

[0191] In another embodiment provided by the present application, the first processing unit of the device executes a process of acquiring first convergence data and determining whether the first convergence data satisfies a preset first convergence condition, including:

[0192] Based on the true values ​​of the data items, determining the true value change value of each of the data items;

[0193] Updating the first iteration number, and generating first convergence data based on the updated first iteration number and the true value change value of each of the data items;

[0194] When the first iteration number in the first convergence data reaches a preset number, and / or the true value change value of each data item in the first convergence data is within the corresponding change interval, it is determined that the first convergence data satisfies the first convergence condition; otherwise, it is determined that the first convergence data does not satisfy the first convergence condition.

[0195] In another embodiment provided by the present application, the third processing unit of the device performs a process of fusing the model parameters of each credit assessment sub-model using a truth value discovery algorithm to obtain the fused model parameters, including:

[0196] Setting the parameter credibility of each credit assessment sub-model;

[0197] Obtaining model parameters of each of the credit assessment sub-models;

[0198] Determine the true value of the parameter using the parameter credibility and model parameters of each credit assessment sub-model;

[0199] Determining the distance value between the model parameter of each credit assessment sub-model and the true value of the parameter;

[0200] Updating the parameter credibility of each credit assessment sub-model based on each of the distance values;

[0201] Acquire second convergence data, and determine whether the second convergence data meets a preset second convergence condition;

[0202] If the second convergence data does not satisfy the second convergence condition, returning to the step of determining the true value of the parameter using the parameter credibility and model parameters of each credit assessment sub-model;

[0203] If the second convergence data satisfies the second convergence condition, the parameter credibility and model parameters of each credit assessment sub-model are processed according to a preset algorithm to obtain fused model parameters.

[0204] In another embodiment provided by the present application, the device further includes:

[0205] A second acquisition unit, used for acquiring the to-be-evaluated data of the to-be-evaluated customer;

[0206] The calling unit is used to call the customer credit evaluation model to process the data to be evaluated and output the credit evaluation information of the user.

[0207] An embodiment of the present application further provides a storage medium, which includes stored instructions, wherein when the instructions are executed, the device where the storage medium is located is controlled to execute the method for constructing the customer credit assessment model.

[0208] The embodiment of the present invention further provides an electronic device, the structural diagram of which is shown in FIG. Figure 7 As shown, it specifically includes a memory 601 and one or more instructions 602, wherein the one or more instructions 602 are stored in the memory 601 and are configured to be executed by one or more processors 603 to perform the following operations:

[0209] Get the initial data of each data source;

[0210] Using a truth value discovery algorithm to process the initial data of each of the data sources to obtain credibility information;

[0211] Using the credibility information to process each of the initial data to obtain a training data set;

[0212] Using the training data set to train each preset initial sub-model to obtain a credit assessment sub-model corresponding to each of the initial sub-models;

[0213] Using a truth value discovery algorithm to fuse the model parameters of each of the credit assessment sub-models to obtain fused model parameters;

[0214] Use the fused model parameters to build a customer credit assessment model.

[0215] It should be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards in the relevant regions.

[0216] The specific implementation processes and derivative methods of the above-mentioned embodiments are all within the protection scope of the present invention.

[0217] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.

[0218] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0219] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for constructing a customer credit assessment model, characterized in that: include: Get the initial data of each data source; Using a truth value discovery algorithm to process the initial data of each of the data sources to obtain credibility information; Using the credibility information to process each of the initial data to obtain a training data set; Using the training data set to train each preset initial sub-model to obtain a credit assessment sub-model corresponding to each of the initial sub-models; Using a truth value discovery algorithm to fuse the model parameters of each of the credit assessment sub-models to obtain fused model parameters; Use the fused model parameters to build a customer credit assessment model.

2. The method according to claim 1, characterized in that The using of the truth value discovery algorithm to process the initial data of each data source to obtain the credibility information includes: Based on the initial data of each of the data sources, obtaining the data of each of the preset data items of each of the data sources; Setting the data credibility of each of the data sources; For each of the data items, using the data credibility of each of the data sources and the data of each of the data sources in the data item, determine the data true value of the data item; For each of the data sources, determining a data error of the data source in each of the data items, and updating the data credibility of the data source based on each of the data errors of the data source; Acquire first convergence data, and determine whether the first convergence data meets a preset first convergence condition; If the first converged data satisfies the first convergence condition, generating credibility information using the data true value of each data item of each data source and the data credibility of each data source; If the first converged data does not satisfy the convergence condition, the process returns to execute the step of determining the data truth value of each data item using the data credibility of each data source and the data of each data source in the data item.

3. The method according to claim 2, characterized in that The obtaining of first convergence data and determining whether the first convergence data satisfies a preset first convergence condition includes: Based on the true values ​​of the data items, determining the true value change value of each of the data items; Updating the first iteration number, and generating first convergence data based on the updated first iteration number and the true value change value of each of the data items; When the first iteration number in the first convergence data reaches a preset number, and / or the true value change value of each data item in the first convergence data is within the corresponding change interval, it is determined that the first convergence data satisfies the first convergence condition; otherwise, it is determined that the first convergence data does not satisfy the first convergence condition.

4. The method according to claim 1, characterized in that The method of using a truth value discovery algorithm to fuse the model parameters of each credit assessment sub-model to obtain fused model parameters includes: Setting the parameter credibility of each credit assessment sub-model; Obtaining model parameters of each of the credit assessment sub-models; Determine the true value of the parameter using the parameter credibility and model parameters of each credit assessment sub-model; Determining the distance value between the model parameter of each credit assessment sub-model and the true value of the parameter; Updating the parameter credibility of each credit assessment sub-model based on each of the distance values; Acquire second convergence data, and determine whether the second convergence data meets a preset second convergence condition; If the second convergence data does not satisfy the second convergence condition, returning to the step of determining the true value of the parameter using the parameter credibility and model parameters of each credit assessment sub-model; If the second convergence data satisfies the second convergence condition, the parameter credibility and model parameters of each credit assessment sub-model are processed according to a preset algorithm to obtain fused model parameters.

5. The method according to claim 1, characterized in that Also includes: Obtain the evaluation data of the customer to be evaluated; The customer credit evaluation model is called to process the data to be evaluated, and the credit evaluation information of the customer is output.

6. A device for constructing a customer credit assessment model, characterized in that: include: A first acquisition unit, used to acquire initial data from each data source; A first processing unit, configured to process the initial data of each data source using a truth value discovery algorithm to obtain credibility information; A second processing unit, configured to process each of the initial data using the credibility information to obtain a training data set; A training unit, used to train each preset initial sub-model using the training data set to obtain a credit assessment sub-model corresponding to each of the initial sub-models; A third processing unit is used to use a truth value discovery algorithm to perform fusion processing on the model parameters of each credit assessment sub-model to obtain fused model parameters; The building unit is used to build a customer credit assessment model using the fused model parameters.

7. The device according to claim 6, characterized in that The first processing unit executes a process of using a truth value discovery algorithm to process the initial data of each data source to obtain credibility information, including: Based on the initial data of each of the data sources, obtaining the data of each of the preset data items of each of the data sources; Setting the data credibility of each of the data sources; For each of the data items, using the data credibility of each of the data sources and the data of each of the data sources in the data item, determine the data true value of the data item; For each of the data sources, determining a data error of the data source in each of the data items, and updating the data credibility of the data source based on each of the data errors of the data source; Acquire first convergence data, and determine whether the first convergence data meets a preset first convergence condition; If the first converged data satisfies the first convergence condition, generating credibility information using the data true value of each data item of each data source and the data credibility of each data source; If the first converged data does not satisfy the convergence condition, the process returns to execute the step of determining the data truth value of each data item using the data credibility of each data source and the data of each data source in the data item.

8. The device according to claim 7, characterized in that The first processing unit executes a process of acquiring first convergence data and determining whether the first convergence data satisfies a preset first convergence condition, including: Based on the true values ​​of the data items, determining the true value change value of each of the data items; Updating the first iteration number, and generating first convergence data based on the updated first iteration number and the true value change value of each of the data items; When the first iteration number in the first convergence data reaches a preset number, and / or the true value change value of each data item in the first convergence data is within the corresponding change interval, it is determined that the first convergence data satisfies the first convergence condition; otherwise, it is determined that the first convergence data does not satisfy the first convergence condition.

9. The device according to claim 6, characterized in that The third processing unit performs a process of fusing the model parameters of each credit assessment sub-model using a truth value discovery algorithm to obtain the fused model parameters, including: Setting the parameter credibility of each credit assessment sub-model; Obtaining model parameters of each of the credit assessment sub-models; Determine the true value of the parameter using the parameter credibility and model parameters of each credit assessment sub-model; Determining the distance value between the model parameter of each credit assessment sub-model and the true value of the parameter; Updating the parameter credibility of each credit assessment sub-model based on each of the distance values; Acquire second convergence data, and determine whether the second convergence data meets a preset second convergence condition; If the second convergence data does not satisfy the second convergence condition, returning to the step of determining the true value of the parameter using the parameter credibility and model parameters of each credit assessment sub-model; If the second convergence data satisfies the second convergence condition, the parameter credibility and model parameters of each credit assessment sub-model are processed according to a preset algorithm to obtain fused model parameters.

10. The device according to claim 6, characterized in that Also includes: A second acquisition unit is used to acquire the to-be-evaluated data of the to-be-evaluated customer; The calling unit is used to call the customer credit evaluation model to process the data to be evaluated and output the credit evaluation information of the user.

Images