Customer portrait label configuration method and system for financing decision

Abstract

The invention discloses a customer portrait label configuration method and system for financing decision, and belongs to the field of data processing, and the method comprises the steps: collecting a basic information set of a target customer; performing data risk coefficient identification based on K data acquisition channels corresponding to the K background data sets; performing same-factor integration identification on the K background data sets, performing data feature analysis in combination with the K data risk coefficients, and generating a data feature set of the target customer; and configuring a preset tolerance bandwidth based on the M data feature risk coefficients, and performing portrait tag analysis on the preset tolerance bandwidth and the M data features by using a tag configuration network layer. According to the method and the device, the technical problems of inaccurate customer portrait and insufficient data integration and risk assessment caused by data acquisition limitation and data credibility due to privacy protection in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a method and system for configuring customer portrait labels for financing decisions. Background Art

[0002] With the digital transformation of the financial industry, financing decisions are increasingly dependent on the accurate analysis and judgment of customer data. Customer portrait tags can effectively extract key information about customers, helping financial institutions quickly understand key indicators such as customers' credit status, repayment ability, and risk preferences, so as to make more scientific and reasonable financing decisions.

[0003] Traditional customer data collection methods often rely on information proactively provided by customers, which is not comprehensive and reliable enough. The rapid development of big data and artificial intelligence technologies has provided new possibilities for customer profiling and risk assessment. However, existing technologies are still limited in data integration and risk assessment. In order to improve the accuracy of financing decisions, there is an urgent need for a system and method that can integrate multi-channel data collection, effectively conduct data risk assessment, and realize data integration and label configuration. Summary of the invention

[0004] This application provides a method and system for configuring customer portrait labels for financing decisions, aiming to solve the technical problems in the prior art of inaccurate customer portraits, insufficient data integration and risk assessment, caused by data acquisition limitations and data credibility due to privacy protection.

[0005] In view of the above problems, the present application provides a method and system for configuring customer portrait labels for financing decisions.

[0006] The first aspect disclosed in the application provides a method for configuring customer portrait labels for financing decisions, the method comprising collecting a basic information set of a target customer, wherein the basic information set comprises K background data sets; identifying data risk coefficients based on K data collection channels corresponding to the K background data sets to generate K data risk coefficients; integrating and identifying the K background data sets with the same factors, performing data feature analysis in combination with the K data risk coefficients to generate a data feature set of the target customer, wherein the data feature set comprises M data features and M data feature risk coefficients; configuring a preset tolerance bandwidth based on the M data feature risk coefficients, and performing portrait label analysis on the preset tolerance bandwidth and the M data features using a label configuration network layer to obtain a portrait label set of the target customer.

[0007] Another aspect disclosed in the present application provides a customer portrait label configuration system for financing decision-making, the system including a basic information collection module: used to collect a basic information set of a target customer, wherein the basic information set includes K background data sets; a risk coefficient identification module: used to identify data risk coefficients based on K data collection channels corresponding to the K background data sets, and generate K data risk coefficients; a same-factor integration identification module: used to perform same-factor integration identification on the K background data sets, perform data feature analysis in combination with the K data risk coefficients, and generate a data feature set of the target customer, wherein the data feature set includes M data features and M data feature risk coefficients; a portrait label analysis module: used to configure a preset tolerance bandwidth based on the M data feature risk coefficients, and use a label configuration network layer to perform a portrait label analysis on the preset tolerance bandwidth and the M data features to obtain a portrait label set of the target customer.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, by collecting K background data sets of target customers, the comprehensiveness and diversity of information are ensured. Then, based on the risk assessment of different data collection channels, K data risk coefficients are generated to provide a quantitative basis for data quality. Then, through the integration and identification of the same factors and the analysis of data features, a data feature set containing M data features and corresponding risk coefficients is formed, which is both a deep processing of the data and a refinement of the value of information. Finally, according to these data feature risk coefficients, the preset tolerance bandwidth is configured, and the network layer is used to perform the portrait label analysis through the label configuration, so as to accurately generate the portrait label set of the target customer, which solves the technical problems of data acquisition restrictions and data credibility caused by privacy protection in the existing technology, resulting in inaccurate customer portraits, insufficient data integration and risk assessment, and realizes the construction of a comprehensive and accurate portrait of the target customer. It effectively improves the accuracy and reliability of customer portraits, provides strong data support for financing decisions, and improves decision-making efficiency and risk management level.

[0009] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A flow chart of a method for configuring customer portrait tags for financing decision-making is provided for the embodiment of the present application; Figure 2A structural diagram of a customer portrait label configuration system for financing decision-making is provided for an embodiment of the present application.

[0011] Explanation of the reference numerals: basic information collection module 11, risk factor identification module 12, same factor integration identification module 13, portrait label analysis module. DETAILED DESCRIPTION

[0012] The overall idea of ​​the technical solution provided by this application is as follows: by collecting a set of basic information of the target customers, including multiple background data sets, to fully understand the background situation of the customers. Then, based on the risk of different data collection channels, a corresponding data risk coefficient is generated for each background data set to improve the reliability of the data. By integrating the identification of the same factors of the background data set and analyzing the data features, combined with the previously generated data risk coefficient, a data feature set containing multiple data features and their corresponding risk coefficients is formed. This step not only integrates various information about the customer, but also conducts a risk assessment on it. Based on the data feature risk coefficient, a preset tolerant bandwidth is configured, and the data features are analyzed by using the label configuration network layer to generate a portrait label set of the target customer. This step aims to provide intuitive and accurate customer portrait information for financing decisions through data analysis and model processing.

[0013] The embodiment of the present application provides a method and system for configuring customer portrait labels for financing decisions. It aims to improve the reliability and accuracy of financing decisions through systematic data collection, risk assessment, data integration and label configuration. The specific steps are as follows: collecting basic information sets, identifying data risk coefficients, integrating and identifying the same factors, and presetting tolerance bandwidth configuration. Through the above steps, the system realizes comprehensive analysis of customer data and accurate label configuration, providing a reliable data basis for financing decisions.

[0014] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically described below in conjunction with the drawings in the specification. Example

[0015] like Figure 1 As shown, the embodiment of the present application provides a method for configuring customer portrait tags for financing decision-making, the method comprising: Step S100: Collect a basic information set of a target customer, wherein the basic information set includes K background data sets.

[0016] Specifically: The target customer refers to an individual or enterprise that needs to be evaluated for financing decisions. The basic information set refers to a collection of all basic data about the target customer, which is used for further analysis and processing. Various basic information about the target customer is collected from multiple channels. The construction of the basic information set is the starting point and foundation of the entire customer portrait tag configuration method.

[0017] The basic information set includes K background data sets, where "K" represents the number of background data sets. The background data set may contain information of different types and sources, including but not limited to: Personal information: such as age, gender, occupation, educational background, etc. Financial information: such as income, expenditure, bank account information, credit score, etc. Social credit data: such as credit report, loan history, overdue record, etc. Consumer behavior data: such as shopping records, transaction history, consumer preferences, etc. Social media data: such as social network activities, interests and hobbies, social circles, etc. Other related data: such as geographic location data, vehicle information, real estate information, etc.

[0018] Background data sets are collected from different channels and sources to provide a comprehensive view of target customers. Each background data set can be regarded as a dimension of data, through which we can have a more comprehensive understanding of all aspects of customers.

[0019] This step collects data through multiple channels and sources to ensure the comprehensiveness of basic information and avoid missing important information. Multi-channel data collection can verify each other and improve the accuracy and reliability of data. Data collection from different backgrounds provides multi-dimensional information, which helps to conduct comprehensive customer profiling and risk assessment.

[0020] Step S200: Identify data risk coefficients based on the K data collection channels corresponding to the K background data sets to generate K data risk coefficients.

[0021] Specifically, the K data collection channels refer to one data collection channel for each background data set. For example, personal information may come from a government database, financial information may come from a bank, social credit data may come from a credit agency, and consumer behavior data may come from an e-commerce platform.

[0022] Among them, the identification of data risk coefficients includes: data collection channel analysis, that is, for each background data set, analyzing the data collection channel of its source. For example, verifying the credibility of financial information provided by banks and evaluating the reliability of credit data provided by credit institutions. Data accuracy assessment, that is, checking the accuracy of data. For example, by cross-validating data from different channels, check whether there are inconsistent or contradictory information. Data integrity assessment, used to ensure the integrity of data. Check whether there are missing data or incomplete information, and evaluate the impact of these missing data on the overall risk assessment. Data relevance assessment, that is, evaluating the relevance of data. For example, check whether the information in the background data set is consistent with the actual situation of the target customer, and whether these data can effectively reflect the credit and risk status of the customer. Then, a data risk coefficient is generated for each background data set. The data risk coefficient is a quantitative indicator that indicates the risk level of the data set. For example, a data set with high data accuracy, strong integrity and good relevance has a low risk coefficient; conversely, the risk coefficient is high.

[0023] In this step, by conducting risk assessment on the data, high-quality data can be screened out and the reliability of the overall data can be improved. In the subsequent data integration and feature analysis process, the risk coefficient can be used to weigh the impact of different data to ensure that the results of data integration and analysis are more accurate. By identifying and quantifying data risks, the risk status of target customers can be better understood, providing an important reference for financing decisions. The risk coefficients of different background data sets can be effectively identified and quantified, providing an important foundation and basis for the subsequent configuration of customer portrait labels.

[0024] Step S300: Integrate and identify the same factors on the K background data sets, perform data feature analysis in combination with the K data risk coefficients, and generate a data feature set for the target customer, wherein the data feature set includes M data features and M data feature risk coefficients.

[0025] Specifically, the same factor integration identification is used to integrate the same or similar factors in different background data sets to form a unified view. The same factor integration identification method includes: factor extraction, data cleaning, consistency check, data integration and other steps.

[0026] Then, data feature analysis is performed in combination with the data risk coefficient to extract valuable data features. The main methods include: Risk weighting: Use the risk coefficient of each background data set to weight the corresponding data. Data with a higher risk coefficient has a lower weight, and data with a lower risk coefficient has a higher weight. Feature extraction: Extract representative data features from the integrated data through statistical analysis, machine learning and other methods. For example, cluster analysis is used to extract customer consumption pattern features, and regression analysis is used to extract customer credit change trends. Risk assessment: Perform risk assessment on the extracted data features and generate corresponding feature risk coefficients. For example, historical data analysis is used to assess the stability and risk of customer income.

[0027] Finally, name and classify each data feature for subsequent analysis and application. For example, income features, credit features, consumption features, etc. Record the risk factor of each data feature as an important parameter for subsequent analysis. Thus, a data feature set is obtained, which includes M data features and M data feature risk factors.

[0028] In this step, the data feature set combines factors from multiple data sources and fully reflects all aspects of the target customer. By integrating and identifying the same factors of K background data sets and analyzing data features, the data feature set of the target customer is generated, providing a solid foundation for subsequent portrait label configuration and financing decisions.

[0029] Step S400: configuring a preset tolerance bandwidth based on the risk coefficients of the M data features, and performing a portrait label analysis on the preset tolerance bandwidth and the M data features using a label configuration network layer to obtain a portrait label set of the target customer.

[0030] Specifically: The preset tolerance bandwidth is a reasonable fluctuation range set for each data feature. This range is determined based on the risk factor of the data feature, aiming to take into account the normal fluctuation of the data feature while avoiding misjudgment due to outliers or noise data. For data features with a lower risk factor, a wider tolerance bandwidth can be set to accommodate greater data fluctuations. For data features with a higher risk factor, a narrower tolerance bandwidth needs to be set to more accurately control the fluctuation range of the data.

[0031] The label configuration network layer is a model based on machine learning or rule engine, which is used to perform image label analysis based on data features and preset tolerance bandwidth. Its design method includes: network layer design, model training, label generation, etc.

[0032] Specifically, a label configuration network layer is designed, including an input layer, a hidden layer, and an output layer. The input layer receives M data features and a preset tolerance bandwidth, the hidden layer extracts and fuses features, and the output layer generates a portrait label. The network layer is trained using sample data. The sample data includes the tolerance bandwidth, data features, and portrait labels of multiple customers. Through supervised learning, the model learns how to convert input data into label output. The M data features and preset tolerance bandwidth of the target customer are input, and the corresponding portrait label set is generated through the trained network layer.

[0033] Among them, the portrait tag analysis is used to use the generated portrait tag set in financing decisions to help evaluate the credit and risk of target customers. That is, according to different data characteristics, the portrait tags are divided into different categories, such as credit rating tags, risk level tags, behavior pattern tags, etc. The generated portrait tags are explained and illustrated to ensure their usability and explainability in financing decisions.

[0034] The portrait label set finally generated through this step provides an accurate and reliable customer portrait for financing decisions, helping financial institutions to better assess the credit and risk of customers.

[0035] Furthermore, the method described in the present application also includes: Taking the same factor as the aggregation target, the K background data sets are integrated and identified to generate M background factor data sets, wherein each background factor data set is multiple data of a background factor in the K background data sets, and each background factor data has a data risk coefficient identifier; the M background factor data sets are consistency identified to obtain M first risk coefficients; the M background factor data sets are respectively subjected to feature intensive search to obtain M data features; the M first risk coefficients are corrected using the K data risk coefficients to obtain M data feature risk coefficients.

[0036] Specifically: extract the same or similar factors from each background data set. Integrate the extracted same or similar factors to form M background factor data sets. Each background factor data set includes multiple data from K background data sets. Add a corresponding data risk coefficient mark to each background factor data to indicate the risk level of the data.

[0037] Further, the M background factor data sets are subjected to consistency identification to evaluate the consistency of each background factor data set and generate a first risk coefficient. The consistency of the data is ensured by performing consistency checks on the M background factor data sets. M first risk coefficients are obtained to represent the consistency risk of each background factor data set.

[0038] The feature-intensive search is used to extract features from the background factor data set and generate data features. By performing a feature-intensive search on each background factor data set, the feature with the highest feature density is found and M data features are generated. Combined with the K data risk coefficients, the M first risk coefficients are corrected to generate M data feature risk coefficients. For example, the first risk coefficient of the income data set is combined with the risk coefficient of the corresponding background data set to generate the final income data feature risk coefficient.

[0039] This step can integrate the same or similar factors in different background data sets, identify their consistency, and extract data features through feature-intensive search. Using the data feature risk coefficient corrected by the risk coefficient, the credit and risk status of the target customer can be more accurately assessed, providing strong support for financing decisions.

[0040] Furthermore, the method described in the present application also includes: Each background factor data in the M background factor data sets is taken as a feature point to construct M background factor box plots; M interquartile ranges in the M background factor box plots are extracted, and the M interquartile ranges are taken as M first risk coefficients.

[0041] Specifically: take each background factor data in each background factor data set as a feature point to construct a background factor box plot. Specifically, each background factor data set contains multiple data points, for example, the income background factor data set contains multiple income data points. Take each background factor data as a feature point. For example, each income data point in the income background factor data set is a feature point. Use these feature points to construct a background factor box plot. The box plot is a graphical method to display data dispersion, which displays data distribution through five statistics (minimum, first quartile, median, third quartile and maximum value).

[0042] Then, the interquartile range is extracted from each background factor box plot and used as the first risk coefficient. The interquartile range is a key statistic in the box plot, which represents the range of the middle 50% of the data. It is calculated by the difference between the third quartile (Q3) and the first quartile (Q1), IQR = Q3 - Q1. Among them, the larger the interquartile range, the higher the data dispersion and the greater the risk coefficient; the smaller the interquartile range, the lower the data dispersion and the smaller the risk coefficient.

[0043] For example, collect multiple income data points of target customers, such as 5,000 yuan, 6,000 yuan, 7,000 yuan, etc. Use these income data points as feature points to construct a box plot of the income background factor. The box plot shows the distribution of these data, including the minimum value, Q1, Q2 (median), Q3, and maximum value. From the box plot of the income background factor, find Q1 and Q3. For example, Q1 is 5,500 yuan and Q3 is 7,500 yuan. Calculate the interquartile range IQR: IQR = Q3 - Q1 = 7,500 yuan - 5,500 yuan = 2,000 yuan. Use the interquartile range of 2,000 yuan as the first risk coefficient of the income background factor.

[0044] In this step, by constructing a box plot for each data point in the background factor data set as a feature point, and extracting the interquartile range from the box plot as the first risk coefficient, the dispersion and risk of the data can be effectively evaluated. This method is intuitive and easy to implement, and provides an important basis for subsequent data feature analysis and risk assessment.

[0045] Furthermore, the method described in the present application also includes: Randomly extract M first background factor data from the M background factor data set; traverse and calculate the M first densities of the M first background factor data in the M background factor data set; randomly extract M second background factor data from the M background factor data set again, wherein the M second background factor data have M second densities; respectively determine whether the M second densities are greater than the M first densities, and if so, use the M second background factor data as M stage data features; randomly extract M third background factor data from the M background factor data set again; update and iterate the M stage data features according to the M third background factor data until the preset number of iterations is met, and use the M stage data features corresponding to the maximum values ​​of the densities during the iteration process as the M data features.

[0046] Specifically: for each background factor data set (such as income, credit score, etc.), a data point is randomly selected to form a first background factor data set. For example, an income data point is randomly selected from the income data set, and a credit score data point is randomly selected from the credit score data set.

[0047] Then, the first density of the first background factor data in the background factor data set is traversed and calculated. The density refers to the number or similarity of neighboring data points of a data point in the data set in which it is located. Specifically, the density reflects the degree of crowding of the position of a data point in the set, that is, the distribution density of the data points around the data point.

[0048] Further, a data point is randomly selected again for each background factor data set to form a second background factor data set. The density of each second background factor data is calculated, which is called the second density.

[0049] Determine whether each second density is greater than the corresponding first density. If so, use the second background factor data as the stage data feature. Otherwise, retain the first background factor data as the stage data feature.

[0050] A data point is randomly selected from each background factor data set again to form a third background factor data set. According to the third background factor data, the stage data features are updated and iterated. The density is continued to be compared. If the density of the third background factor data is higher than the density of the current stage data features, the stage data features are updated. Repeat the above process until the preset number of iterations is reached or other termination conditions are met.

[0051] The stage data feature corresponding to the maximum density in the iteration process is used as the final data feature. The change of density in each iteration is recorded, and the stage data feature corresponding to the maximum density is selected as the M data features.

[0052] Based on this step, the most densely populated data features can be extracted from the M background factor data sets, and multiple random extractions and iterative updates are used to ensure that the extracted features are representative and reliable, providing a solid foundation for subsequent portrait label analysis.

[0053] Furthermore, the method described in the present application also includes: The similarities between the M first background factor data and the multiple background factor data in the M background factor data sets are respectively calculated to obtain M first similarity sets; the number of the M first similarity sets that meet the preset similarity threshold is counted, and the ratio of the M first statistics to the M data amounts of the M background factor data sets is used as the M first density.

[0054] Specifically: for each first background factor data, calculate its similarity with each data point in the background factor data set. The similarity calculation can be based on Euclidean distance, cosine similarity or other similarity measurement methods. For example, for an income data point, the Euclidean distance between it and all other income data points can be calculated; for a credit score data point, the cosine similarity between it and all other credit score data points can be calculated.

[0055] The number of data points that meet the preset similarity threshold in each first similarity set is counted, that is, a similarity threshold is set, and whether each similarity meets the threshold is determined. The number of data points that meet the preset similarity threshold in each first similarity set is counted. For example, if the set similarity threshold is 0.8, the number of data points with a similarity greater than or equal to 0.8 is counted.

[0056] The number of data in each first similarity set that meets a preset similarity threshold is recorded as a first statistic. The ratio of the first statistic to the total data volume of the background factor data set is calculated as a first density.

[0057] For example: for a randomly selected income data point, calculate the Euclidean distance between it and all other income data points to obtain a similarity set. Assume that the income data set has 100 data points, and the calculated similarity set is [0.85, 0.78, 0.92, ...]. Set the similarity threshold to 0.8, and count the number of similarity points greater than or equal to 0.8 in the similarity set. For example, there are 40 data points in the similarity set that meet the similarity threshold of 0.8. The total amount of data in the background factor data set is 100. The first statistic is 40, and the first density calculation is 40 / 100, which is 0.4.

[0058] This step can effectively evaluate the density of data points in the set by calculating the similarity between each first background factor data and other data points in the background factor data set, counting the number of data points that meet the preset similarity threshold, and calculating the corresponding density. This method provides a quantitative indicator for selecting and updating the most dense data features in multiple random extractions and iterations, providing a reliable basis for subsequent portrait label analysis.

[0059] Furthermore, the method described in the present application also includes: Constructing a tolerant bandwidth identifier; using the tolerant bandwidth identifier to identify the risk coefficients of the M data features to obtain the preset tolerant bandwidth; inputting the preset tolerant bandwidth and the M data features into the label configuration network layer to generate the portrait label set.

[0060] Specifically: Construct a tolerance bandwidth identifier to evaluate the risk factor of data features. This can be constructed using machine learning models or statistical methods, such as regression analysis or decision tree-based methods. Use the feature risk coefficients and corresponding tolerance bandwidths in historical data as training sets to train the tolerance bandwidth identifier.

[0061] The specific construction process includes: Data preparation: Collect historical data, including characteristic risk factors and known tolerance bandwidths of multiple customers. For example, characteristic risk factors include customer income risk factors, credit score risk factors, etc., and tolerance bandwidth may be set based on specific financial models or business rules. Model selection: Select a suitable machine learning algorithm to build a tolerance bandwidth identifier. For example, a regression analysis model is used to predict tolerance bandwidth, with the input being the characteristic risk factor and the output being the corresponding tolerance bandwidth. Model training: Use training data to train the model and adjust the model parameters by minimizing the prediction error. For example, use the gradient descent method to optimize the regression model parameters so that the predicted tolerance bandwidth is as close to the actual value as possible. Model validation: Use a validation set to validate the model to ensure the generalization ability and accuracy of the model. For example, calculate the mean square error (MSE) or root mean square error (RMSE) of the model on the validation set to evaluate the model performance.

[0062] Then, the risk coefficients of the M data features are input into the tolerance bandwidth identifier, and the identifier outputs the corresponding preset tolerance bandwidth according to the input risk coefficient.

[0063] Design the label configuration network layer, including input layer, hidden layer and output layer. The input layer receives M data features and preset tolerance bandwidth. The hidden layer extracts and fuses the input data.

[0064] The output layer generates a set of image labels.

[0065] This step builds a tolerance bandwidth identifier, uses the identifier to conduct risk assessment on the data feature risk factor, determines the tolerance bandwidth of each feature, and then inputs these preset tolerance bandwidths and data features into the label configuration network layer to generate a portrait label set. This improves the accuracy and reliability of customer portraits and provides strong support for financing decisions.

[0066] Furthermore, the method described in the present application also includes: Acquire multiple sample tolerance bandwidths and multiple sample data features, as well as multiple sample portrait label sets as construction data; use the construction data to perform supervised learning on the encoder and decoder until the model output reaches convergence, and obtain the trained label configuration network layer.

[0067] Specifically: collect sample data for training the label configuration network layer, including sample tolerance bandwidth, sample data features and sample portrait label set. Among them, the sample tolerance bandwidth can collect the tolerance bandwidth data of historical customers. These data can be determined based on historical decisions and actual risk conditions. The sample data features can collect various data features of historical customers, including income, credit score, consumption behavior, etc. The sample portrait label set can collect a portrait label set of historical customers, which reflects the customer's credit rating, risk level, consumption habits, etc.

[0068] The sample tolerance bandwidth, sample data features and sample portrait label sets are integrated together to form a constructed data set. The data is preprocessed, including data cleaning, standardization, and normalization, to improve the training effect.

[0069] Then, the encoder and decoder models are supervised by the constructed dataset to train the label configuration network layer. Specifically, the input tolerance bandwidth and data features are encoded into an intermediate representation. The encoder can use models such as multi-layer perceptron (MLP), convolutional neural network (CNN) or long short-term memory network (LSTM). The intermediate representation is decoded into a set of portrait labels. The decoder can use a model structure similar to the encoder.

[0070] Then, the model parameters are optimized through training to make the model output converge, that is, the error between the model's prediction result and the actual image label is minimized. Specifically, a loss function is defined, such as mean square error (MSE), cross entropy loss, etc., to measure the difference between the model output and the actual label. Use optimization algorithms such as gradient descent (SGD) and Adam optimizer to adjust the model parameters by minimizing the loss function. Iterate the model on the training set until the model output converges, that is, the loss function no longer decreases significantly.

[0071] By obtaining sample data and building a data set, using supervised learning to train the encoder and decoder models, and optimizing the label configuration network layer, efficient customer portrait label generation can be achieved. This method ensures the accuracy and generalization ability of the model, providing reliable data support for financing decisions.

[0072] In summary, the customer portrait tag configuration method for financing decision-making provided in the embodiment of the present application has the following technical effects: 1. Generate accurate customer profile labels through systematic basic information collection, risk factor identification, data integration and label configuration, improve the comprehensiveness and credibility of data, and provide reliable support for financing decisions.

[0073] 2. Taking the same factors as the aggregation target, we integrate and identify data from multiple channels, generate background factor data sets, and conduct feature analysis in combination with risk factors. This achieves efficient integration and feature extraction of data from multiple sources, and enhances the comprehensiveness and reliability of the data.

[0074] 3. Use multiple random extractions and density calculation methods to iteratively extract data features to ensure that the extracted features are representative and reliable. This improves the accuracy and robustness of feature extraction and enhances the effect of data analysis.

[0075] 4. Build a tolerance bandwidth identifier, use it to identify the risk factor of data characteristics, configure the preset tolerance bandwidth, input the label configuration network layer, and generate a portrait label set. This improves the flexibility and reliability of data processing and enhances the accuracy of portrait labels. Example

[0076] Based on the same inventive concept as the customer portrait tag configuration method for financing decision-making in the aforementioned embodiment, Figure 2 As shown, the embodiment of the present application provides a customer portrait label configuration system for financing decision-making, the system comprising: Basic information collection module 11: used to collect a basic information set of a target customer, wherein the basic information set includes K background data sets; The risk coefficient identification module 12 is used to identify the data risk coefficient based on the K data collection channels corresponding to the K background data sets, and generate K data risk coefficients; The same-factor integration and identification module 13 is used to perform same-factor integration and identification on the K background data sets, perform data feature analysis in combination with the K data risk coefficients, and generate a data feature set of the target customer, wherein the data feature set includes M data features and M data feature risk coefficients; The portrait label analysis module 14 is used to configure a preset tolerance bandwidth based on the risk coefficients of the M data features, and use the label configuration network layer to perform a portrait label analysis on the preset tolerance bandwidth and the M data features to obtain a portrait label set of the target customer.

[0077] Furthermore, the system is also used for: Taking the same factor as the aggregation target, the K background data sets are integrated and identified to generate M background factor data sets, wherein each background factor data set is a plurality of data of a background factor in the K background data sets, and each background factor data has a data risk coefficient identifier; Performing consistency identification on the M background factor data sets to obtain M first risk coefficients; Performing feature-intensive search on the M background factor data sets respectively to obtain M data features; The M first risk coefficients are corrected using the K data risk coefficients to obtain M data feature risk coefficients.

[0078] Furthermore, the system is also used for: Taking each background factor data in the M background factor data sets as a feature point, constructing M background factor box plots; M interquartile ranges are extracted from the box plots of the M background factors, and the M interquartile ranges are used as M first risk coefficients.

[0079] Furthermore, the system is also used for: Randomly extracting M first background factor data from the M background factor data sets; Traversing and calculating M first densities of the M first background factor data in the M background factor data sets; Randomly extracting M second background factor data from the M background factor data sets again, wherein the M second background factor data have M second densities; respectively determining whether the M second densities are greater than the M first densities, and if so, taking the M second background factor data as M stage data features; Randomly extracting M third background factor data from the M background factor data sets again; The M stage data features are updated and iterated according to the M third background factor data until a preset number of iterations is met, and the M stage data features corresponding to the maximum density in the iteration process are used as M data features.

[0080] Furthermore, the system is also used for: Calculating similarities between the M first background factor data and a plurality of background factor data in the M background factor data sets respectively, to obtain M first similarity sets; The number of the M first similarity sets that meet a preset similarity threshold is counted, and the ratio of the M first statistics to the M data amounts of the M background factor data sets is used as the M first densities.

[0081] Furthermore, the system is also used for: Constructing a tolerant bandwidth identifier; Using the tolerance bandwidth identifier to perform risk identification on the M data feature risk coefficients to obtain the preset tolerance bandwidth; The preset tolerance bandwidth and the M data features are input into the label configuration network layer to generate the portrait label set.

[0082] Furthermore, the system is also used for: Obtain multiple sample tolerance bandwidths, multiple sample data features, and multiple sample portrait label sets as construction data; The encoder and decoder are supervised by using the constructed data until the model output reaches convergence, thereby obtaining the trained label configuration network layer.

[0083] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, without any unnecessary restrictions.

[0084] Furthermore, the first or second mentioned above may not only represent an order relationship, but may also represent a specific concept, and / or refer to multiple elements that can be selected individually or in full. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A customer portrait tag configuration method for financing decision-making, characterized in that: The method comprises: Collecting a basic information set of target customers, wherein the basic information set includes K background data sets; Identify data risk coefficients based on K data collection channels corresponding to the K background data sets to generate K data risk coefficients; Integrate and identify the same factors on the K background data sets, perform data feature analysis in combination with the K data risk coefficients, and generate a data feature set of the target customer, wherein the data feature set includes M data features and M data feature risk coefficients; A preset tolerance bandwidth is configured based on the risk coefficients of the M data features, and a label configuration network layer is used to perform a portrait label analysis on the preset tolerance bandwidth and the M data features to obtain a portrait label set of the target customer.

2. The method according to claim 1, characterized in that The method comprises: Taking the same factor as the aggregation target, the K background data sets are integrated and identified to generate M background factor data sets, wherein each background factor data set is a plurality of data of a background factor in the K background data sets, and each background factor data has a data risk coefficient identifier; Performing consistency identification on the M background factor data sets to obtain M first risk coefficients; Performing feature-intensive search on the M background factor data sets respectively to obtain M data features; The M first risk coefficients are corrected using the K data risk coefficients to obtain M data feature risk coefficients.

3. The method according to claim 2, characterized in that Performing consistency identification on the M background factor data sets to obtain M first risk coefficients, the method comprising: Taking each background factor data in the M background factor data sets as a feature point, constructing M background factor box plots; M interquartile ranges are extracted from the box plots of the M background factors, and the M interquartile ranges are used as M first risk coefficients.

4. The method according to claim 2, characterized in that Performing feature-intensive search on the M background factor data sets respectively to obtain M data features, the method comprising: Randomly extracting M first background factor data from the M background factor data sets; Traversing and calculating M first densities of the M first background factor data in the M background factor data sets; Randomly extracting M second background factor data from the M background factor data sets again, wherein the M second background factor data have M second densities; respectively determining whether the M second densities are greater than the M first densities, and if so, taking the M second background factor data as M stage data features; Randomly extracting M third background factor data from the M background factor data sets again; The M stage data features are updated and iterated according to the M third background factor data until a preset number of iterations is met, and the M stage data features corresponding to the maximum density in the iteration process are used as M data features.

5. The method according to claim 4, characterized in that The method comprises: Calculating similarities between the M first background factor data and a plurality of background factor data in the M background factor data sets respectively, to obtain M first similarity sets; The number of the M first similarity sets that meet a preset similarity threshold is counted, and the ratio of the M first statistics to the M data amounts of the M background factor data sets is used as the M first densities.

6. The method according to claim 1, characterized in that The method comprises: Constructing a tolerant bandwidth identifier; Using the tolerance bandwidth identifier to perform risk identification on the M data feature risk coefficients to obtain the preset tolerance bandwidth; The preset tolerance bandwidth and the M data features are input into the label configuration network layer to generate the portrait label set.

7. The method according to claim 6, characterized in that The method comprises: Obtain multiple sample tolerance bandwidths, multiple sample data features, and multiple sample portrait label sets as construction data; The encoder and decoder are supervised by using the constructed data until the model output reaches convergence, thereby obtaining the trained label configuration network layer.

8. A customer profile tag configuration system for financing decision making, characterized in that: For executing the method according to any one of claims 1 to 7, the system comprises: Basic information collection module: used to collect a basic information set of target customers, wherein the basic information set includes K background data sets; Risk coefficient identification module: used to identify data risk coefficients based on K data collection channels corresponding to the K background data sets, and generate K data risk coefficients; The same-factor integration and identification module is used to perform same-factor integration and identification on the K background data sets, perform data feature analysis in combination with the K data risk coefficients, and generate a data feature set of the target customer, wherein the data feature set includes M data features and M data feature risk coefficients; Portrait label analysis module: used to configure a preset tolerance bandwidth based on the risk coefficients of the M data features, and use the label configuration network layer to perform portrait label analysis on the preset tolerance bandwidth and the M data features to obtain a portrait label set of the target customer.

Images