Credit subject data risk assessment system and method

Abstract

The invention discloses a credit subject data risk assessment system and method, and the system comprises a data collection module which is used for collecting credit investigation associated data of a credit subject from a plurality of data channels; the data preprocessing module is used for denoising and de-duplicating the credit investigation associated data, and classifying the credit investigation associated data according to a preset association type to obtain a plurality of to-be-evaluated data group classes; the data evaluation module is used for processing the to-be-evaluated data group classes based on an input layer, a hidden layer and an output layer of a preset neural network model to obtain evaluation results corresponding to the to-be-evaluated data group classes; and the data weighting module is used for acquiring a weight value corresponding to each to-be-evaluated data group class, and performing weighting processing on each evaluation result based on each weight value to obtain a data risk score of the credit subject. According to the invention, data related to credit investigation can be comprehensively and accurately analyzed and processed, so that enterprise credit investigation evaluation is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of big data analysis and processing, and in particular to a credit subject data risk assessment system and method. Background Art

[0002] In modern society, the importance of corporate credit is self-evident. It has a significant impact on identifying and assessing credit risks, investment decisions, and business cooperation. Therefore, accurate processing of credit-related data is an effective way to ensure accurate assessment of corporate credit. However, the current assessment of corporate credit usually relies on data directly related to credit, resulting in incomplete data used to assess corporate credit. At the same time, the analysis and processing of data directly related to credit is not in-depth enough, resulting in inaccurate corporate credit assessment. Therefore, how to comprehensively and accurately analyze and process credit-related data to make corporate credit assessment more accurate is one of the technical issues that need to be solved at present. Summary of the invention

[0003] The main purpose of the present invention is to provide a credit subject data risk assessment system and method, aiming to solve the technical problem of how to comprehensively and accurately analyze and process credit-related data in the prior art to make corporate credit assessment more accurate.

[0004] To achieve the above object, the present invention provides a credit subject data risk assessment system, the credit subject data risk assessment system comprising:

[0005] A data collection module, used to collect credit-related data of a credit subject from multiple data channels, wherein the credit-related data at least includes the credit subject's loan data, news media data, tax data, invoice data, financial statements, and transaction data;

[0006] A data preprocessing module, used to remove noise and duplication from the credit-related data, and classify the credit-related data according to preset association types to obtain multiple data groups to be evaluated;

[0007] A data evaluation module, used for processing each of the data groups to be evaluated based on the input layer, hidden layer and output layer of the preset neural network model to obtain evaluation results corresponding to each of the data groups to be evaluated;

[0008] The data weighting module is used to obtain the weight values ​​corresponding to each of the data groups to be evaluated, and to perform weighted processing on each of the evaluation results based on the weight values ​​to obtain the data risk score of the credit subject.

[0009] Preferably, the data evaluation module comprises:

[0010] A transmission unit, used to transmit each of the data groups to be evaluated to the input layer of the preset neural network model, and generate evaluation results corresponding to each of the data groups to be evaluated through analysis and calculation of the hidden layer and the output layer in the preset neural network model;

[0011] Among them, the formula for hidden layer analysis calculation is:

[0012]

[0013] The formula for the output layer analysis and calculation is:

[0014]

[0015] Where Yj represents the output vector of the jth neuron in the hidden layer, f1(·) represents the activation function of the hidden layer, n represents the number of neurons in the input layer, wji represents the weight value between the jth neuron in the hidden layer and the ith neuron in the input layer, Ii represents the class of the data group to be evaluated input by the i-th neuron in the input layer, and bj represents the bias value of the jth neuron in the hidden layer; Pk represents the evaluation result of the kth neural output in the output layer, f2(·) represents the activation function of the output layer, m represents the number of neurons in the hidden layer, wkj represents the weight value between the kth neuron in the output layer and the jth neuron in the hidden layer, and bk represents the bias value of the kth neuron in the output layer;

[0016] An acquisition unit is used to obtain each of the evaluation results output by the output layer of the preset neural network model.

[0017] Preferably, the data preprocessing module includes:

[0018] A grouping unit, used to classify the credit-related data according to a preset association type to obtain a plurality of data group classes;

[0019] A verification unit, configured to verify each data element in each data group class according to the association relationship between each data element in each data group class, and generate each data group class as a verification data group class based on the verification;

[0020] An identification unit is used to identify the unit magnitude of each data element in each of the verification data group classes, and convert the data elements in the verification data group classes according to the conversion relationship between the unit magnitudes, and generate a data group class to be evaluated from the verification data group class based on the conversion.

[0021] Preferably, the credit subject data risk assessment system further includes:

[0022] A generation module, used to add the evaluation results corresponding to each of the data groups to be evaluated and the data risk score of the credit subject to a preset data report for encryption processing to generate a risk report;

[0023] The verification module is used to verify the access permission corresponding to the access request when receiving the access request for accessing the risk report, and send the risk report to the request account corresponding to the access request after the access permission is verified.

[0024] Preferably, the credit subject data risk assessment system further includes:

[0025] An acquisition module, used for acquiring a large number of data group samples, and dividing each of the data group samples into a training data group and a verification data group;

[0026] A training module, used to train a preset initial model based on the training data group, and to verify the preset initial model based on the verification data group when the statistical training times reach a preset times, to obtain a verification result;

[0027] The calculation module is used to calculate the error value of the preset initial model based on the verification result, and the calculation formula is:

[0028]

[0029] Wherein, E represents the error value, T represents the number of verification data sets, qt represents the verification result corresponding to the t-th verification data set, and Qt represents the reference result corresponding to the t-th verification data set;

[0030] A judgment module, used to judge whether the error value is less than a preset threshold value, and if it is less than the preset threshold value, stop training the preset initial model and generate the preset initial model as the preset neural network model;

[0031] The updating module is used to update the model parameters of the preset initial model when the error value is not less than the preset threshold, and the training module iteratively trains the updated preset initial model until the error value is less than the preset threshold.

[0032] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a credit subject data risk assessment method, the credit subject data risk assessment method comprising:

[0033] Collecting credit-related data of credit subjects from multiple data channels, wherein the credit-related data at least includes the credit subject's loan data, news media data, tax data, invoice data, financial statements, and transaction data;

[0034] De-noising and de-duplication of the credit-related data, and classifying the credit-related data according to preset association types to obtain multiple data group classes to be evaluated;

[0035] Processing each of the data groups to be evaluated based on the input layer, hidden layer and output layer of the preset neural network model to obtain evaluation results corresponding to each of the data groups to be evaluated;

[0036] Obtain weight values ​​corresponding to each of the data groups to be evaluated, and perform weighted processing on each of the evaluation results based on the weight values ​​to obtain the data risk score of the credit subject.

[0037] Preferably, the step of processing each of the data groups to be evaluated based on the input layer, hidden layer and output layer of the preset neural network model to obtain evaluation results corresponding to each of the data groups to be evaluated comprises:

[0038] Transmitting each of the data groups to be evaluated to the input layer of a preset neural network model, and generating evaluation results corresponding to each of the data groups to be evaluated through analysis and calculation by a hidden layer and an output layer in the preset neural network model;

[0039] Among them, the formula for hidden layer analysis calculation is:

[0040]

[0041] The formula for the output layer analysis and calculation is:

[0042]

[0043] Where Yj represents the output vector of the jth neuron in the hidden layer, f1(·) represents the activation function of the hidden layer, n represents the number of neurons in the input layer, wji represents the weight value between the jth neuron in the hidden layer and the ith neuron in the input layer, Ii represents the class of the data group to be evaluated input by the i-th neuron in the input layer, and bj represents the bias value of the jth neuron in the hidden layer; Pk represents the evaluation result of the kth neural output in the output layer, f2(·) represents the activation function of the output layer, m represents the number of neurons in the hidden layer, wkj represents the weight value between the kth neuron in the output layer and the jth neuron in the hidden layer, and bk represents the bias value of the kth neuron in the output layer;

[0044] Obtain each of the evaluation results output by the output layer of the preset neural network model.

[0045] Preferably, the step of classifying the credit-related data according to the preset association type to obtain a plurality of data groups to be evaluated includes:

[0046] Classifying the credit-related data according to the preset association type to obtain multiple data group categories;

[0047] Verifying each data element in each data group class according to the association relationship between each data element in each data group class, and generating each data group class as a verification data group class based on the verification;

[0048] The unit magnitude of each data element in each of the verification data group classes is identified, and the data elements in the verification data group class are converted according to the conversion relationship between the unit magnitudes, and the verification data group class is generated into a data group class to be evaluated based on the conversion.

[0049] Preferably, after the step of weighting the evaluation results based on the weight values ​​to obtain the data risk score of the credit subject, the step further includes:

[0050] Add the evaluation results corresponding to each of the data groups to be evaluated and the data risk score of the credit subject to a preset data report for encryption processing to generate a risk report;

[0051] When an access request for accessing the risk report is received, the access permission corresponding to the access request is verified, and after the access permission is verified, the risk report is sent to the request account corresponding to the access request.

[0052] Preferably, before the step of processing each of the data groups to be evaluated based on the input layer, hidden layer and output layer of the preset neural network model, the step further includes:

[0053] Acquire a large number of data set samples, and divide each of the data set samples into a training data set and a validation data set;

[0054] Training a preset initial model based on the training data set, and when the statistical number of training times reaches a preset number of times, verifying the preset initial model based on the verification data set to obtain a verification result;

[0055] The calculation module is used to calculate the error value of the preset initial model based on the verification result, and the calculation formula is:

[0056]

[0057] Wherein, E represents the error value, T represents the number of verification data sets, qt represents the verification result corresponding to the t-th verification data set, and Qt represents the reference result corresponding to the t-th verification data set;

[0058] Determine whether the error value is less than a preset threshold value, if so, stop training the preset initial model, and generate the preset initial model as the preset neural network model;

[0059] When the error value is not less than a preset threshold, the model parameters of the preset initial model are updated, and the training module iteratively trains the updated preset initial model until the error value is less than the preset threshold.

[0060] The credit subject data risk assessment system and method of the present invention include a data acquisition module, a data preprocessing module, a data evaluation module and a data weighting module. The data acquisition module first collects the credit-related data of the credit subject from multiple data channels, and the credit-related data at least includes the credit subject's loan data, news media data, tax data, invoice data, financial statements and transaction data; the data preprocessing module then removes noise and duplication on each collected credit-related data, and classifies them according to preset association types to obtain multiple data groups to be evaluated; the data evaluation module then processes each data group to be evaluated according to the input layer, hidden layer and output layer of a preset neural network model to obtain evaluation results corresponding to each data group to be evaluated; and the data weighting module obtains weight values ​​corresponding to each data group to be evaluated, and performs weighted processing on each evaluation result based on each weight value to obtain a data risk score for the credit subject. In this way, various credit-related data of credit subjects are collected from multiple data channels with the help of crawler technology. These various credit-related data include data directly related to the credit of the credit subject and data indirectly related to the credit of the credit subject, making the data used for credit assessment more comprehensive. After preprocessing and classification to obtain multiple data groups to be evaluated, each credit-related data is analyzed by a preset neural network model to obtain multiple evaluation results, and weighted by the weight values ​​corresponding to each data group to be evaluated to obtain the final data risk score of the credit subject. Among them, the preset neural network model is formed by training a large number of sample data related to corporate credit, and the evaluation results obtained by its analysis and processing are more accurate. The weight value of each data to be evaluated reflects the degree of influence of each data group to be evaluated on credit. The weighted processing of each evaluation result by each weight value further improves the accuracy of the generated data risk score, realizes comprehensive and accurate analysis and processing of credit-related data, and makes corporate credit assessment more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a module diagram of the first embodiment of the credit subject data risk assessment system of the present invention;

[0062] Figure 2 This is a module diagram of the second embodiment of the credit subject data risk assessment system of the present invention;

[0063] Figure 3 This is a flow chart of the first embodiment of the credit subject data risk assessment method of the present invention;

[0064] Figure 4 It is a flow chart of the second embodiment of the credit subject data risk assessment method of the present invention.

[0065] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0066] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0067] The present invention provides a credit subject data risk assessment system, please refer to Figure 1 , Figure 1 The module diagram of the first embodiment of the credit subject data risk assessment system of the present invention is shown in FIG. Specifically, the credit subject data risk assessment system in this embodiment at least includes a data acquisition module 10, a data preprocessing module 20, a data assessment module 30 and a data weighting module.

[0068] The data collection module 10 is used to collect credit-related data of the credit subject from multiple data channels, and the credit-related data at least includes the credit subject's loan data, news media data, tax data, invoice data, financial statements, and transaction data;

[0069] In this embodiment, the credit subject can be a financial enterprise or other non-financial enterprise. The data risk assessment of the credit subject can be to assess the credit risk of the credit subject through various data related to the credit subject and credit investigation. Specifically, when the credit subject has an assessment demand, an assessment request is initiated to the credit subject data risk assessment system of this embodiment. After receiving the assessment request, the system reads the enterprise ID (Identification) carried therein, and determines the specific enterprise that needs to be assessed by the enterprise ID. Then, the data risk assessment is initiated for the credit subject, and the data collection module 10 collects various credit-related data of the credit subject from multiple data channels, such as at least official websites, news websites, social platforms, government affairs platforms, etc., by calling APIs, web crawlers and other technologies. The various credit-related data collected include at least the credit subject's loan data, news media data, tax data, invoice data, financial statements and transaction data. It includes both data directly related to credit investigation and data indirectly related to credit investigation. For example, lending data includes data directly related to the credit report of the credit subject, such as the loan amount and loan term, as well as data not directly related to the credit report, such as the lender's business information, negative information, public opinion information, senior management information, etc., which are used to reflect the degree of financing tension and social credit status of the credit subject.

[0070] The data preprocessing module 20 is used to remove noise and duplication from the credit-related data, and classify the credit-related data according to preset association types to obtain multiple data groups to be evaluated;

[0071] Understandably, the various types of credit-related data collected have varying degrees of accuracy, and even contain data that is completely irrelevant to the credit subject, so it is necessary to perform denoising and deduplication preprocessing on the collected credit-related data through the data preprocessing module 20. At the same time, each type of credit-related data may reflect the credit of the credit subject from multiple dimensions, and different dimensions have different degrees of influence on the credit, so in order to conduct credit assessment on the credit subject according to the dimensions, it is necessary to classify the credit-related data after denoising and deduplication. A preset association type representing each dimension is pre-set, and each credit-related data is classified according to the preset association type to obtain multiple data group classes to be evaluated. Among them, the preset association type at least includes enterprise basic information class, industry information class, financing information class, business information class, financial information class and credit information class. The correspondingly divided data group classes to be evaluated include enterprise basic information group class, industry information group class, financing information group class, business information group class, financial information group class and credit information group class. The basic information group of enterprises mainly includes basic information such as the nature of the enterprise, senior management information, registered capital, paid-in capital, etc.; the industry information group mainly includes information such as industry characteristics, development trends, policy support, credit policy, etc.; the financing information group mainly includes information such as the number of credit lines approved, credit limit approved, guarantee status, debt repayment risk, credit overdue, restructuring investment, etc.; the operating information group mainly includes information such as main products, proportion of main business, annual plan and completion status, profit, market share, etc.; the financial information group mainly includes information such as capital structure, cash status, profitability, operating ability, growth ability, etc.; the credit information group mainly includes credit risk rating and credit score, etc.

[0072] Specifically, in order to classify each credit-related data according to the preset association type, the data preprocessing module 20 includes:

[0073] A grouping unit 21, used to classify the credit-related data according to a preset association type to obtain a plurality of data group categories;

[0074] A verification unit 22, configured to verify each data element in each data group class according to the association relationship between each data element in each data group class, and generate each data group class as a verification data group class based on the verification;

[0075] The identification unit 23 is used to identify the unit magnitude of each data element in each of the verification data group classes, and convert the data elements in the verification data group classes according to the conversion relationship between the unit magnitudes, and generate the data group class to be evaluated from the verification data group class based on the conversion.

[0076] Further, the grouping unit 21 classifies and processes each credit-related data according to the preset association type to obtain multiple data groups. Each data group contains multiple data elements, each of which comes from the same or different data channels, and there may be contradictory data elements. For example, a certain news report claims that the profit of the credit subject is greater than a, and the official website of the credit subject claims that the profit is less than a. In this regard, the verification unit 22 needs to verify each data element through the association relationship between each data element. A relatively accurate data element is determined from each contradictory data element, and other contradictory data is eliminated. Among them, the association relationship between each data element refers to the data element that indicates that the same indicator has an association, such as the data element representing profit as mentioned above. The verification can be based on the source of the data element on the one hand, and on the amount of data on the other hand. For the source, the data of the government website and the official website shall prevail. For the amount of data, the data of the amount of data shall be referred to first. After each data group is verified, each data group is formed into a verification data group.

[0077] Understandably, the data elements in each verification data group class exist in various forms, for example, data representing the same indicator may not have the same unit magnitude. In this regard, the identification unit 23 needs to identify the unit magnitude of each data element, and convert each data element according to the conversion relationship between the unit magnitudes, and realize the normalization between the data elements representing the same indicator of the verification data group class through conversion verification. After the data elements of the verification data group class are normalized, the verification data group class forms the data group class to be evaluated.

[0078] The data evaluation module 30 is used to process each of the data groups to be evaluated based on the input layer, hidden layer and output layer of the preset neural network model to obtain evaluation results corresponding to each of the data groups to be evaluated.

[0079] Furthermore, a preset neural network model is formed in the credit subject data risk assessment system by training a large amount of sample data related to credit investigation. The preset neural network model can be set to include an input layer, a hidden layer and an output layer. The data assessment module 30 evaluates each data group to be evaluated based on the input layer, hidden layer and output layer of the preset neural network model to obtain the evaluation results corresponding to each data group to be evaluated. Specifically, the data assessment module 30 includes:

[0080] A transmission unit 31 is used to transmit each of the data groups to be evaluated to the input layer of the preset neural network model, and generate an evaluation result corresponding to each of the data groups to be evaluated through analysis and calculation of the hidden layer and the output layer in the preset neural network model;

[0081] The acquisition unit 32 is used to obtain each of the evaluation results output by the output layer of the preset neural network model.

[0082] Furthermore, the transmission module 31 transmits each data group to be evaluated to the input layer of the preset neural network model, and the hidden layer in the preset neural network model performs analysis and calculation according to its model parameters, and obtains the output result of the hidden layer as the input of the output layer of the preset neural network model, and then the output layer performs analysis and calculation on the input according to its model parameters to generate evaluation results corresponding to each data group to be evaluated. The formulas for the analysis and calculation of the hidden layer and the output layer can be specifically referred to as the following formulas (1) and (2), respectively.

[0083]

[0084] Where Yj represents the output vector of the jth neuron in the hidden layer, f1(·) represents the activation function of the hidden layer, n represents the number of neurons in the input layer, wji represents the weight value between the jth neuron in the hidden layer and the ith neuron in the input layer, Ii represents the class of the data group to be evaluated input by the i-th neuron in the input layer, and bj represents the bias value of the j-th neuron in the hidden layer; Pk represents the evaluation result of the k-th neural output in the output layer, f2(·) represents the activation function of the output layer, m represents the number of neurons in the hidden layer, wkj represents the weight value between the k-th neuron in the output layer and the j-th neuron in the hidden layer, and bk represents the bias value of the k-th neuron in the output layer.

[0085] Furthermore, the output layer of the preset neural network model calculates and outputs each evaluation result, and the receiving unit 32 receives each evaluation result of the output, so that the data evaluation unit 30 obtains each evaluation result.

[0086] The data weighting module 40 is used to obtain the weight values ​​corresponding to each of the data groups to be evaluated, and to perform weighted processing on each of the evaluation results based on the weight values ​​to obtain the data risk score of the credit subject.

[0087] Furthermore, the degree of influence of each data group to be evaluated on the credit can be reflected by a pre-set weight value. The data weighting model 40 obtains the weight value corresponding to each data group to be evaluated, and performs weighted processing on each evaluation result by each weight value, multiplies the weight value corresponding to the data group to be evaluated by the evaluation result corresponding to the data group to be evaluated, and obtains the operation result. After each data group to be evaluated obtains the operation result, each operation result is added up, and the operation result obtained is the data risk score indicating the credit risk level of the credit subject.

[0088] Furthermore, in order to enable the credit subject to more conveniently view the data risk score and the various assessment results of the data risk assessment, this embodiment also generates a risk report. Specifically, the credit subject data risk assessment system also includes:

[0089] Generating module a1, used for adding the evaluation results corresponding to each of the data groups to be evaluated and the data risk score of the credit subject to a preset data report for encryption processing to generate a risk report;

[0090] The verification module a2 is used to verify the access permission corresponding to the access request when receiving the access request for accessing the risk report, and send the risk report to the request account corresponding to the access request after the access permission is verified.

[0091] Furthermore, a preset data report for generating a risk report is pre-set, and the preset data report is called by the generation module a1, and the evaluation results corresponding to each data group to be evaluated and the data risk score of the credit subject are added to the data area corresponding to the preset data report to form an initial risk report. In addition, in order to prevent the privacy leakage of such risk reports, the generation module a1 encrypts the generated initial risk report according to a pre-set encryption algorithm, and generates a final risk report through the encryption process.

[0092] In order to further ensure the security of the risk report, if the system receives an access request to access the risk report, the access permission corresponding to the access request is verified through the verification module a2. The verification method can be a whitelist verification, that is, setting an account whitelist that can access the risk report, and judging whether the request account corresponding to the access request is in the account whitelist. If it is in the account whitelist, it means that the request account corresponding to the access request has the permission to access the risk report, and the access permission verification is judged to be passed, and the risk report is sent to the request account corresponding to the access request. On the contrary, if the request account corresponding to the access request is not in the account whitelist, it means that the request account corresponding to the access request does not have the permission to access the risk report, so the feedback information of no access permission is returned to the request account. In this way, the security of the risk report is ensured through encryption and access permission control.

[0093] The credit subject data risk assessment system implemented in this embodiment includes a data collection module, a data preprocessing module, a data evaluation module and a data weighting module. The data collection module first collects the credit-related data of the credit subject from multiple data channels, and the credit-related data at least includes the credit subject's loan data, news media data, tax data, invoice data, financial statements and transaction data; the data preprocessing module then removes noise and duplications from the collected credit-related data, and classifies them according to preset association types to obtain multiple data groups to be evaluated; the data evaluation module then processes each data group to be evaluated according to the input layer, hidden layer and output layer of the preset neural network model to obtain evaluation results corresponding to each data group to be evaluated; and the data weighting module obtains weight values ​​corresponding to each data group to be evaluated, and performs weighted processing on each evaluation result based on each weight value to obtain a data risk score for the credit subject. In this way, various credit-related data of credit subjects are collected from multiple data channels with the help of crawler technology. These various credit-related data include data directly related to the credit of the credit subject and data indirectly related to the credit of the credit subject, making the data used for credit assessment more comprehensive. After preprocessing and classification to obtain multiple data groups to be evaluated, each credit-related data is analyzed by a preset neural network model to obtain multiple evaluation results, and weighted by the weight values ​​corresponding to each data group to be evaluated to obtain the final data risk score of the credit subject. Among them, the preset neural network model is formed by training a large number of sample data related to corporate credit, and the evaluation results obtained by its analysis and processing are more accurate. The weight value of each data to be evaluated reflects the degree of influence of each data group to be evaluated on credit. The weighted processing of each evaluation result by each weight value further improves the accuracy of the generated data risk score, realizes comprehensive and accurate analysis and processing of credit-related data, and makes corporate credit assessment more accurate.

[0094] For further information, please refer to Figure 2 Based on the first embodiment of the credit subject data risk assessment system of the present invention, a second embodiment of the credit subject data risk assessment system of the present invention is proposed.

[0095] The difference between the second embodiment of the credit subject data risk assessment system and the first embodiment of the credit subject data risk assessment system is that the credit subject data risk assessment system further includes:

[0096] An acquisition module 50 is used to acquire a large number of data group samples and divide each of the data group samples into a training data group and a verification data group;

[0097] A training module 60 is used to train a preset initial model based on the training data group, and when the statistical training times reach a preset times, verify the preset initial model based on the verification data group to obtain a verification result;

[0098] A calculation module 70, configured to calculate an error value of the preset initial model based on the verification result;

[0099] A judgment module 80 is used to judge whether the error value is less than a preset threshold value. If the error value is less than the preset threshold value, the training of the preset initial model is stopped, and the preset initial model is generated as the preset neural network model;

[0100] The updating module 90 is used to update the model parameters of the preset initial model when the error value is not less than the preset threshold, and the training module iteratively trains the updated preset initial model until the error value is less than the preset threshold.

[0101] Furthermore, the credit subject data risk assessment system is also provided with a module for training to form a preset neural network model, specifically including an acquisition module 50, a training module 60, a calculation module 70, a judgment module 80 and an update module 90. First, a large number of data group samples related to credit investigation are acquired through the acquisition module 50, and each data group sample is divided into a training data group and a verification data group according to a preset division ratio, such as 8 to 2, 7 to 3, etc. Then, the training module 60 performs training based on the divided training data group using the preset initial model, and pre-sets the preset number of times. When the statistical training number reaches the preset number of times, the verification is performed through the preset initial model of the verification data group, that is, the verification data group is analyzed and processed by the trained preset initial model to generate a verification result corresponding to the verification data group.

[0102] Furthermore, in order to reflect the effect of the preset initial model on the analysis and processing of the verification data set, the calculation module 70 calculates the error value of the preset initial model according to the verification result. The specific calculation formula can be seen in the following formula (3).

[0103]

[0104] Wherein, E represents the error value, T represents the number of verification data groups, qt represents the verification result corresponding to the t-th verification data group, and Qt represents the reference result corresponding to the t-th verification data group.

[0105] Further, in order to indicate the size of the error value, a preset threshold is pre-set. The calculated error value is compared with the preset threshold to determine whether the error value is less than the preset threshold. If it is less than the preset threshold, it means that the preset initial model processes the verification data group, and the obtained verification result is not much different from the reference result of the verification data group, and the processing effect of the preset initial model is good. At this time, the training of the preset initial model is stopped, and the preset initial model is generated as a preset neural network model. On the contrary, if the error value is determined to be not less than the preset threshold by comparison, it means that the difference between the verification result and the reference result is large, and the preset initial model needs to be further trained. At this time, the model parameters of the preset initial model are updated. The model parameters include at least the weight value between each neuron in the hidden layer and each neuron in the data layer, the weight value between each neuron in the output layer and each neuron in the hidden layer, and the bias value of each neuron in the hidden layer and the output layer. The updating method can pre-set the update formula of the weight value and the update formula of the bias value, and update each weight value and the bias value respectively through their respective update formulas. In addition, it can also be set to partially update, while the other part is not updated during this training process.

[0106] Furthermore, after the model parameters of the preset initial model are updated, the training module 60 iteratively trains the updated preset initial model based on the training data group, and regenerates the verification result through the verification data group when the number of training times reaches the preset number again, and calculates the error value based on the verification result, until the calculated error value is less than the preset threshold, the training of the preset initial model is stopped to generate the preset neural network model.

[0107] This embodiment uses a large number of credit-related data group samples to train and verify the preset initial model. When the verification result generated by the preset initial model after processing the verification data group is not much different from the reference result and the preset initial model has good data processing performance, the training of the preset initial model is stopped and a preset neural network model is generated, thereby ensuring the accuracy of the preset neural network model in processing each data group related to credit reporting, and further ensuring the accuracy of the data risk assessment of the credit subject.

[0108] In addition, the present invention provides a credit subject data risk assessment method, please refer to Figure 3 , Figure 3 This is a flowchart of the first embodiment of the credit subject data risk assessment method of the present invention. It should be noted that although the logical order is shown in the flowchart, in some cases, the steps shown or described can be performed in a different order than here. Specifically, the credit subject data risk assessment method in this embodiment includes:

[0109] Step S10, collecting credit-related data of the credit subject from multiple data channels, wherein the credit-related data at least includes the credit subject's loan data, news media data, tax data, invoice data, financial statements, and transaction data;

[0110] In this embodiment, the credit subject can be a financial enterprise or other non-financial enterprise. The data risk assessment of the credit subject can be to assess the credit risk of the credit subject through various data related to the credit subject and credit investigation. Specifically, when the credit subject has an assessment demand, an assessment request is initiated to the credit subject data risk assessment system of this embodiment. After receiving the assessment request, the system reads the enterprise ID (Identification) carried therein, and determines the specific enterprise that needs to be assessed by the enterprise ID. Then, the data risk assessment is initiated for the credit subject, and various credit-related data of the credit subject are collected from multiple data channels, such as at least official websites, news websites, social platforms, government affairs platforms, etc., by calling APIs, web crawlers and other technologies. The various credit-related data collected include at least the credit subject's loan data, news media data, tax data, invoice data, financial statements and transaction data. It includes both data directly related to credit investigation and data indirectly related to credit investigation. For example, lending data includes data directly related to the credit report of the credit subject, such as the loan amount and loan term, as well as data not directly related to the credit report, such as the lender's business information, negative information, public opinion information, senior management information, etc., which are used to reflect the degree of financing tension and social credit status of the credit subject.

[0111] Step S20, denoising and deduplicating the credit-related data, and classifying the credit-related data according to preset association types to obtain a plurality of data groups to be evaluated;

[0112] Understandably, the accuracy of various types of credit-related data collected varies, and even contains data that is completely irrelevant to the credit subject, so it is necessary to pre-process the collected credit-related data by denoising and de-duplication. At the same time, each type of credit-related data may reflect the credit of the credit subject from multiple dimensions, and different dimensions have different degrees of influence on credit. Therefore, in order to conduct credit assessment on the credit subject according to the dimensions, it is necessary to classify the credit-related data after denoising and de-duplication. A preset association type representing each dimension is pre-set, and each credit-related data is classified according to the preset association type to obtain multiple data groups to be evaluated. Among them, the preset association type at least includes enterprise basic information class, industry information class, financing information class, business information class, financial information class and credit information class. The corresponding data groups to be evaluated include enterprise basic information group class, industry information group class, financing information group class, business information group class, financial information group class and credit information group class. The basic information group of enterprises mainly includes basic information such as the nature of the enterprise, senior management information, registered capital, paid-in capital, etc.; the industry information group mainly includes information such as industry characteristics, development trends, policy support, credit policy, etc.; the financing information group mainly includes information such as the number of credit lines approved, credit limit approved, guarantee status, debt repayment risk, credit overdue, restructuring investment, etc.; the operating information group mainly includes information such as main products, proportion of main business, annual plan and completion status, profit, market share, etc.; the financial information group mainly includes information such as capital structure, cash status, profitability, operating ability, growth ability, etc.; the credit information group mainly includes credit risk rating and credit score, etc.

[0113] Specifically, in order to classify each credit-related data according to a preset association type, the step of classifying the credit-related data according to the preset association type to obtain a plurality of data groups to be evaluated includes:

[0114] Step S21, classifying the credit-related data according to the preset association type to obtain a plurality of data group categories;

[0115] Step S22, verifying each data element in each data group class according to the association relationship between each data element in each data group class, and generating each data group class as a verification data group class based on the verification;

[0116] Step S23, identifying the unit magnitude of each data element in each verification data group class, and converting the data elements in the verification data group class according to the conversion relationship between the unit magnitudes, and generating a data group class to be evaluated from the verification data group class based on the conversion.

[0117] Further, according to the preset association type, each credit investigation related data is classified and processed to obtain multiple data group classes. Each data group class contains multiple data elements, each data element comes from the same or different data channels, and there may be contradictory data elements. For example, a certain news promotes that the profit of the credit subject is greater than a, and the official website of the credit subject claims that the profit is less than a. In this regard, it is necessary to verify each data element through the association relationship between each data element. From each contradictory data element, a relatively accurate data element is determined, and other contradictory data are eliminated. Among them, the association relationship between each data element refers to the data element that indicates the same indicator has a correlation, such as the data element representing profit mentioned above. The verification can be based on the source of the data element on the one hand, and on the amount of data on the other hand. For the source, the data of the government website and the official website shall prevail. For the amount of data, the data of the amount of data shall be referred to first. After each data group class is verified, each data group class is formed into a verification data group class.

[0118] Understandably, the data elements in each verification data group class exist in various forms. For example, the data representing the same indicator may not have the same unit magnitude. To this end, it is necessary to identify the unit magnitude of each data element, and convert each data element according to the conversion relationship between the unit magnitudes. The normalization between the data elements representing the same indicator in the verification data group class is achieved through conversion verification. After the data elements of the verification data group class are normalized, the verification data group class forms the data group class to be evaluated.

[0119] Step S30, processing each of the data groups to be evaluated based on the input layer, hidden layer and output layer of the preset neural network model to obtain evaluation results corresponding to each of the data groups to be evaluated.

[0120] Furthermore, a preset neural network model is formed in the credit subject data risk assessment system by training a large amount of sample data related to credit investigation in advance. The preset neural network model can be set to include an input layer, a hidden layer, and an output layer. According to the input layer, hidden layer, and output layer of the preset neural network model, each data group to be evaluated is evaluated and processed to obtain the evaluation results corresponding to each data group to be evaluated. Specifically, the steps of processing each data group to be evaluated based on the input layer, hidden layer, and output layer of the preset neural network model to obtain the evaluation results corresponding to each data group to be evaluated include:

[0121] Step S31, transmitting each of the data groups to be evaluated to the input layer of the preset neural network model, and generating evaluation results corresponding to each of the data groups to be evaluated through analysis and calculation of the hidden layer and the output layer in the preset neural network model;

[0122] Step S32, obtaining each of the evaluation results output by the output layer of the preset neural network model.

[0123] Furthermore, each data group to be evaluated is transmitted to the input layer of the preset neural network model, and the hidden layer in the preset neural network model performs analysis and calculation according to its model parameters, and obtains the output result of the hidden layer as the input of the output layer of the preset neural network model, and then the output layer performs analysis and calculation on the input according to its model parameters to generate evaluation results corresponding to each data group to be evaluated. Among them, the formulas for the analysis and calculation of the hidden layer and the output layer can be specifically referred to as the above formula (1) and formula (2), which will not be repeated here.

[0124] Furthermore, the output layer of the preset neural network model will calculate and output each evaluation result, and obtain each evaluation result of the output to obtain each evaluation result.

[0125] Step S40, obtaining weight values ​​corresponding to each of the data groups to be evaluated, and performing weighted processing on each of the evaluation results based on the weight values ​​to obtain the data risk score of the credit subject.

[0126] Furthermore, the degree of influence of each data group to be evaluated on credit reporting can be reflected by a pre-set weight value. The weight values ​​corresponding to each data group to be evaluated are obtained, and each evaluation result is weighted by each weight value. The weight value corresponding to the data group to be evaluated is multiplied by the evaluation result corresponding to the data group to be evaluated to obtain the operation result. After each data group to be evaluated has obtained the operation result, each operation result is added up, and the operation result obtained is the data risk score indicating the credit risk level of the credit subject.

[0127] Furthermore, in order to enable the credit subject to more conveniently view the data risk score and the evaluation results of the data risk evaluation, this embodiment also generates a risk report. Specifically, after the step of weighting the evaluation results based on the weight values ​​to obtain the data risk score of the credit subject, the step also includes:

[0128] Step b1, adding the evaluation results corresponding to each of the data groups to be evaluated and the data risk score of the credit subject to a preset data report for encryption processing to generate a risk report;

[0129] Step b2: upon receiving an access request for accessing the risk report, verifying the access permission corresponding to the access request, and after the access permission is verified, sending the risk report to the requesting account corresponding to the access request.

[0130] Furthermore, a preset data report for generating a risk report is pre-set, the preset data report is called, and the evaluation results corresponding to each data group to be evaluated and the data risk score of the credit subject are added to the data area corresponding to the preset data report to form an initial risk report. In addition, in order to prevent the privacy leakage of such risk reports, the generated initial risk report is encrypted according to a pre-set encryption algorithm, and the final risk report is generated through the encryption process.

[0131] In order to further ensure the security of risk reports, if the system receives an access request for access to risk reports, it verifies the access rights corresponding to the access request. The verification method can be whitelist verification, that is, setting a whitelist of accounts that can access risk reports, and determining whether the request account corresponding to the access request is in the account whitelist. If it is in the account whitelist, it means that the request account corresponding to the access request has the authority to access the risk report, and the access right verification is determined to be passed, and the risk report is sent to the request account corresponding to the access request. On the contrary, if the request account corresponding to the access request is not in the account whitelist, it means that the request account corresponding to the access request does not have the authority to access the risk report, so feedback information of no access rights is returned to the request account. In this way, the security of risk reports is ensured through encryption and access right control.

[0132] The credit subject data risk assessment method implemented in this embodiment first collects the credit subject's credit-related data from multiple data channels, and the credit-related data at least includes the credit subject's loan data, news media data, tax data, invoice data, financial statements and transaction data; then denoises and de-duplicates the collected credit-related data, and classifies them according to the preset association type to obtain multiple data groups to be evaluated; then processes each data group to be evaluated according to the input layer, hidden layer and output layer of the preset neural network model to obtain the assessment results corresponding to each data group to be evaluated; and obtains the weight values ​​corresponding to each data group to be evaluated, and performs weighted processing on each assessment result based on each weight value to obtain the data risk score of the credit subject. In this way, various types of credit-related data of the credit subject are collected from multiple data channels with the help of crawler technology, and the various types of credit-related data include both data directly related to the credit of the credit subject and data indirectly related to the credit of the credit subject, so that the data used for credit assessment is more comprehensive. For each credit-related data, after preprocessing and classification to obtain multiple data groups to be evaluated, the preset neural network model is used for analysis to obtain multiple evaluation results, which are weighted by the weight values ​​corresponding to each data group to be evaluated to obtain the final data risk score of the credit subject. Among them, the preset neural network model is trained by a large number of sample data related to corporate credit, and the evaluation results obtained by its analysis and processing are more accurate. The weight value of each data to be evaluated reflects the degree of influence of each data group to be evaluated on credit. The weighted processing of each evaluation result through each weight value further improves the accuracy of the generated data risk score, realizes comprehensive and accurate analysis and processing of credit-related data, and makes corporate credit evaluation more accurate.

[0133] For further information, please refer to Figure 4 Based on the first embodiment of the credit subject data risk assessment method of the present invention, a second embodiment of the credit subject data risk assessment method of the present invention is proposed.

[0134] The difference between the second embodiment of the credit subject data risk assessment method and the first embodiment of the credit subject data risk assessment method is that after the step of performing cluster analysis on each of the enterprise characteristic data to obtain a plurality of cluster data groups, the step further includes:

[0135] Step S50, obtaining a large number of data group samples, and dividing each of the data group samples into a training data group and a verification data group;

[0136] Step S60, training a preset initial model based on the training data set, and when the statistical number of training times reaches a preset number, verifying the preset initial model based on the verification data set to obtain a verification result;

[0137] Step S70, calculating the error value of the preset initial model based on the verification result;

[0138] Step S80, determining whether the error value is less than a preset threshold value, if so, stopping the training of the preset initial model, and generating the preset initial model as the preset neural network model;

[0139] Step S90, when the error value is not less than a preset threshold, the model parameters of the preset initial model are updated, and the training module iteratively trains the updated preset initial model until the error value is less than the preset threshold.

[0140] Furthermore, in order to train and form a preset neural network model, this embodiment acquires a large number of data group samples related to credit investigation, and divides each data group sample into a training data group and a verification data group according to a preset division ratio, such as 8 to 2, 7 to 3, etc. Then, the preset initial model is trained based on the divided training data group, and a preset number of times is preset. When the statistical training number reaches the preset number of times, the preset initial model is verified by the verification data group, that is, the verification data group is analyzed and processed by the trained preset initial model to generate a verification result corresponding to the verification data group.

[0141] Furthermore, in order to reflect the effect of the preset initial model on the analysis and processing of the verification data set, the error value of the preset initial model is calculated according to the verification result. The specific calculation formula can be found in the above formula (3), which will not be repeated here.

[0142] Further, in order to indicate the size of the error value, a preset threshold is pre-set. The calculated error value is compared with the preset threshold to determine whether the error value is less than the preset threshold. If it is less than the preset threshold, it means that the preset initial model processes the verification data group, and the obtained verification result is not much different from the reference result of the verification data group, and the processing effect of the preset initial model is good. At this time, the training of the preset initial model is stopped, and the preset initial model is generated as a preset neural network model. On the contrary, if the error value is determined to be not less than the preset threshold by comparison, it means that the difference between the verification result and the reference result is large, and the preset initial model needs to be further trained. At this time, the model parameters of the preset initial model are updated. The model parameters include at least the weight value between each neuron in the hidden layer and each neuron in the data layer, the weight value between each neuron in the output layer and each neuron in the hidden layer, and the bias value of each neuron in the hidden layer and the output layer. The updating method can pre-set the update formula of the weight value and the update formula of the bias value, and update each weight value and the bias value respectively through their respective update formulas. In addition, it can also be set to partially update, while the other part is not updated during this training process.

[0143] Furthermore, after the model parameters of the preset initial model are updated, the updated preset initial model is iteratively trained based on the training data group, and the verification result is regenerated through the verification data group when the number of training times reaches the preset number again, and the error value is calculated based on the verification result, until the calculated error value is less than the preset threshold, the training of the preset initial model is stopped to generate the preset neural network model.

[0144] This embodiment uses a large number of credit-related data group samples to train and verify the preset initial model. When the verification result generated by the preset initial model after processing the verification data group is not much different from the reference result and the preset initial model has good data processing performance, the training of the preset initial model is stopped and a preset neural network model is generated, thereby ensuring the accuracy of the preset neural network model in processing each data group related to credit reporting, and further ensuring the accuracy of the data risk assessment of the credit subject.

[0145] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims. All equivalent structures or equivalent process changes made using the contents of the specification and drawings of the present invention, or directly or indirectly used in other related technical fields, are protected by the present invention.

Claims

1. A credit subject data risk assessment system, characterized in that: The credit subject data risk assessment system includes: A data collection module, used to collect credit-related data of a credit subject from multiple data channels, wherein the credit-related data at least includes the credit subject's loan data, news media data, tax data, invoice data, financial statements, and transaction data; A data preprocessing module, used to remove noise and duplication from the credit-related data, and classify the credit-related data according to preset association types to obtain multiple data groups to be evaluated; A data evaluation module, used for processing each of the data groups to be evaluated based on the input layer, hidden layer and output layer of the preset neural network model to obtain evaluation results corresponding to each of the data groups to be evaluated; The data weighting module is used to obtain the weight values ​​corresponding to each of the data groups to be evaluated, and to perform weighted processing on each of the evaluation results based on the weight values ​​to obtain the data risk score of the credit subject.

2. The credit subject data risk assessment system according to claim 1, characterized in that: The data evaluation module includes: A transmission unit, used to transmit each of the data groups to be evaluated to the input layer of the preset neural network model, and generate evaluation results corresponding to each of the data groups to be evaluated through analysis and calculation of the hidden layer and the output layer in the preset neural network model; Among them, the formula for hidden layer analysis calculation is: The formula for the output layer analysis and calculation is: Where Yj represents the output vector of the jth neuron in the hidden layer, f1(·) represents the activation function of the hidden layer, n represents the number of neurons in the input layer, wji represents the weight value between the jth neuron in the hidden layer and the ith neuron in the input layer, Ii represents the class of the data group to be evaluated input by the i-th neuron in the input layer, and bj represents the bias value of the jth neuron in the hidden layer; Pk represents the evaluation result of the kth neural output in the output layer, f2(·) represents the activation function of the output layer, m represents the number of neurons in the hidden layer, wkj represents the weight value between the kth neuron in the output layer and the jth neuron in the hidden layer, and bk represents the bias value of the kth neuron in the output layer; An acquisition unit is used to obtain each of the evaluation results output by the output layer of the preset neural network model.

3. The credit subject data risk assessment system according to claim 1, characterized in that: The data preprocessing module comprises: A grouping unit, used to classify the credit-related data according to a preset association type to obtain a plurality of data group classes; A verification unit, configured to verify each data element in each data group class according to the association relationship between each data element in each data group class, and generate each data group class as a verification data group class based on the verification; An identification unit is used to identify the unit magnitude of each data element in each of the verification data group classes, and convert the data elements in the verification data group classes according to the conversion relationship between the unit magnitudes, and generate a data group class to be evaluated from the verification data group class based on the conversion.

4. The credit subject data risk assessment system according to claim 1, characterized in that: The credit subject data risk assessment system further includes: A generation module, used to add the evaluation results corresponding to each of the data groups to be evaluated and the data risk score of the credit subject to a preset data report for encryption processing to generate a risk report; The verification module is used to verify the access permission corresponding to the access request when receiving the access request for accessing the risk report, and send the risk report to the request account corresponding to the access request after the access permission is verified.

5. The credit subject data risk assessment system according to any one of claims 1 to 4, characterized in that: The credit subject data risk assessment system further includes: An acquisition module, used for acquiring a large number of data group samples, and dividing each of the data group samples into a training data group and a verification data group; A training module, used to train a preset initial model based on the training data group, and to verify the preset initial model based on the verification data group when the statistical training times reach a preset times, to obtain a verification result; The calculation module is used to calculate the error value of the preset initial model based on the verification result, and the calculation formula is: Wherein, E represents the error value, T represents the number of verification data sets, qt represents the verification result corresponding to the t-th verification data set, and Qt represents the reference result corresponding to the t-th verification data set; A judgment module, used to judge whether the error value is less than a preset threshold value, and if it is less than the preset threshold value, stop training the preset initial model and generate the preset initial model as the preset neural network model; The updating module is used to update the model parameters of the preset initial model when the error value is not less than the preset threshold, and the training module iteratively trains the updated preset initial model until the error value is less than the preset threshold.

6. A credit subject data risk assessment method, characterized in that: The credit subject data risk assessment method includes: Collecting credit-related data of credit subjects from multiple data channels, wherein the credit-related data at least includes the credit subject's loan data, news media data, tax data, invoice data, financial statements, and transaction data; De-noising and de-duplication of the credit-related data, and classifying the credit-related data according to preset association types to obtain multiple data group classes to be evaluated; Processing each of the data groups to be evaluated based on the input layer, hidden layer and output layer of the preset neural network model to obtain evaluation results corresponding to each of the data groups to be evaluated; Obtain weight values ​​corresponding to each of the data groups to be evaluated, and perform weighted processing on each of the evaluation results based on the weight values ​​to obtain the data risk score of the credit subject.

7. The credit subject data risk assessment method according to claim 6, characterized in that: The step of processing each of the data groups to be evaluated based on the input layer, hidden layer and output layer of the preset neural network model to obtain evaluation results corresponding to each of the data groups to be evaluated comprises: Transmitting each of the data groups to be evaluated to the input layer of a preset neural network model, and generating evaluation results corresponding to each of the data groups to be evaluated through analysis and calculation by a hidden layer and an output layer in the preset neural network model; Among them, the formula for hidden layer analysis calculation is: The formula for the output layer analysis and calculation is: Where Yj represents the output vector of the jth neuron in the hidden layer, f1(·) represents the activation function of the hidden layer, n represents the number of neurons in the input layer, wji represents the weight value between the jth neuron in the hidden layer and the ith neuron in the input layer, Ii represents the class of the data group to be evaluated input by the i-th neuron in the input layer, and bj represents the bias value of the jth neuron in the hidden layer; Pk represents the evaluation result of the kth neural output in the output layer, f2(·) represents the activation function of the output layer, m represents the number of neurons in the hidden layer, wkj represents the weight value between the kth neuron in the output layer and the jth neuron in the hidden layer, and bk represents the bias value of the kth neuron in the output layer; Obtain each of the evaluation results output by the output layer of the preset neural network model.

8. The credit subject data risk assessment method according to claim 6, characterized in that: The step of classifying the credit-related data according to the preset association type to obtain a plurality of data groups to be evaluated includes: Classifying the credit-related data according to the preset association type to obtain multiple data group categories; Verifying each data element in each data group class according to the association relationship between each data element in each data group class, and generating each data group class as a verification data group class based on the verification; The unit magnitude of each data element in each of the verification data group classes is identified, and the data elements in the verification data group class are converted according to the conversion relationship between the unit magnitudes, and the verification data group class is generated into a data group class to be evaluated based on the conversion.

9. The credit subject data risk assessment method according to claim 6, characterized in that: After the step of weighting the evaluation results based on the weight values ​​to obtain the data risk score of the credit subject, the step further includes: Add the evaluation results corresponding to each of the data groups to be evaluated and the data risk score of the credit subject to a preset data report for encryption processing to generate a risk report; When an access request for accessing the risk report is received, the access permission corresponding to the access request is verified, and after the access permission is verified, the risk report is sent to the request account corresponding to the access request.

10. The credit subject data risk assessment method according to any one of claims 6 to 9, characterized in that: Before the step of processing each of the data groups to be evaluated based on the input layer, hidden layer and output layer of the preset neural network model, the step further includes: Acquire a large number of data set samples, and divide each of the data set samples into a training data set and a validation data set; Training a preset initial model based on the training data group, and when the statistical number of training times reaches a preset number of times, verifying the preset initial model based on the verification data group to obtain a verification result; The calculation module is used to calculate the error value of the preset initial model based on the verification result, and the calculation formula is: Wherein, E represents the error value, T represents the number of verification data sets, qt represents the verification result corresponding to the t-th verification data set, and Qt represents the reference result corresponding to the t-th verification data set; Determine whether the error value is less than a preset threshold value, if so, stop training the preset initial model, and generate the preset initial model as the preset neural network model; When the error value is not less than a preset threshold, the model parameters of the preset initial model are updated, and the training module iteratively trains the updated preset initial model until the error value is less than the preset threshold.

Images