Sample generation method and device, credit evaluation method and system and electronic equipment

Abstract

The invention discloses a sample generation method and device, a credit evaluation method and system and electronic equipment. The sample generation method comprises the steps that key feature dimensions are determined from feature dimensions related to a sample set, the sample set comprises positive samples and negative samples, the positive samples are samples obtained based on historical users with risk transaction behaviors, and the negative samples are samples obtained based on historical users with non-risk transaction behaviors; the sample set is used for training a user classification model, the user classification model is used for evaluating a credit score of a to-be-evaluated user, and the credit score is used for representing a credit risk degree of the to-be-evaluated user; feature distribution of key feature dimensions of at least two target samples is obtained, the target samples are samples in a sample set, and the at least two target samples are both positive samples or negative samples; and generating a new sample of the sample set based on the feature distribution and the target sample, the new sample and the target sample being both positive samples or negative samples. Through the scheme of the invention, the generated sample can be more representative, and the accuracy and generalization ability of the user classification model can be improved.

Description

Technical Field

[0001] This application belongs to the field of data processing technology, and in particular relates to a sample generation method, apparatus, credit assessment method, system, electronic device and storage medium. Background Technology

[0002] A sample set is fundamental for training machine learning models. In the fintech field, the collected sample sets often face the problem of insufficient sample quantity, which affects the accuracy and generalization ability of machine learning models. Currently, traditional sample generation methods do not take into account the inherent relationships between sample data, making it possible for the generated synthetic samples to not always accurately reflect the distribution of real data, leading to the introduction of noise or unnecessary data assumptions into the sample set. In other words, even sample sets with sufficient sample quantity obtained based on traditional sample generation methods still cannot avoid impacting the accuracy and generalization ability of machine learning models. Summary of the Invention

[0003] This application provides a sample generation method, apparatus, credit assessment method, system, electronic device, and storage medium, which can make the generated samples more representative and help improve the accuracy and generalization ability of the model.

[0004] Firstly, this application provides a sample generation method, including:

[0005] Key feature dimensions are selected from the feature dimensions related to the sample set. The sample set contains positive samples and negative samples. Positive samples are samples obtained based on historical users with risky transaction behavior, and negative samples are samples obtained based on historical users with non-risky transaction behavior. The sample set is used to train the user classification model, the user classification model is used to evaluate the credit score of the user to be evaluated, and the credit score is used to characterize the degree of credit risk of the user to be evaluated.

[0006] Obtain the feature distribution of key feature dimensions of at least two target samples, wherein the target samples are samples in the sample set, and at least two target samples are either both positive samples or both negative samples;

[0007] Based on the feature distribution and the target sample, a new sample set is generated, wherein the new sample and the target sample are either both positive samples or both negative samples.

[0008] In one specific implementation of the first aspect, before obtaining the feature distributions of the key feature dimensions of at least two target samples, the sample generation method may further include:

[0009] Compare the number of positive and negative samples in the sample set;

[0010] The class with fewer samples is identified as the target sample.

[0011] In one specific implementation of the first aspect, generating new samples for the sample set based on the feature distribution and the target samples may include:

[0012] Based on the feature distribution, determine the range of feature values ​​to be generated for the key feature dimensions;

[0013] Based on the target sample and the feature value range of the key feature dimensions, generate new samples for the sample set.

[0014] In one specific implementation of the first aspect, generating new samples for the sample set based on the target sample and the feature value generation range of the key feature dimension may include:

[0015] Extract feature values ​​from the non-critical feature dimensions of the target sample;

[0016] Generate feature values ​​for the key feature dimensions within the range of feature value generation for the key feature dimensions;

[0017] New samples are generated from the sample set based on the feature values ​​of the non-key feature dimensions and the feature values ​​of the key feature dimensions.

[0018] In one specific implementation of the first aspect, after extracting the feature values ​​of the non-critical feature dimensions of the target sample, the sample generation method may further include:

[0019] Random noise is added to the feature values ​​of non-critical feature dimensions, and the random noise is less than a preset noise threshold.

[0020] Accordingly, based on the feature values ​​of non-critical feature dimensions and the feature values ​​of critical feature dimensions, new samples of the sample set are generated, including:

[0021] New samples are generated from the sample set based on the feature values ​​of the key feature dimensions and the feature values ​​of the non-key feature dimensions after adding random noise.

[0022] In one specific implementation of the first aspect, determining the key feature dimensions from the feature dimensions associated with the sample set may include:

[0023] Based on the pre-defined importance assessment model, importance scores for each feature dimension related to the sample set are obtained;

[0024] Key feature dimensions are determined based on the importance scores of each feature dimension and the preset first feature dimension selection criteria.

[0025] In one specific implementation of the first aspect, the sample generation method may further include:

[0026] If the number of feature dimensions related to the sample set exceeds the preset threshold for the number of feature dimensions, the feature dimensions to be removed are determined based on the importance score of each feature dimension and the preset second feature dimension screening conditions.

[0027] Remove the feature dimensions to be removed from the feature dimensions associated with the sample set.

[0028] Secondly, this application provides a sample generation apparatus, comprising:

[0029] The filtering module is used to filter out key feature dimensions from the feature dimensions related to the sample set. The sample set contains positive samples and negative samples. Positive samples are samples obtained based on historical users with risky transaction behavior, and negative samples are samples obtained based on historical users with non-risky transaction behavior. The sample set is used to train the user classification model. The user classification model is used to evaluate the credit score of the user to be evaluated. The credit score is used to characterize the degree of credit risk of the user to be evaluated.

[0030] The acquisition module is used to acquire the feature distribution of key feature dimensions of at least two target samples, wherein the target samples are samples in the sample set, and at least two target samples are either both positive samples or both negative samples.

[0031] The generation module is used to generate new samples for the sample set based on the feature distribution and the target sample, wherein the new sample and the target sample are either both positive samples or both negative samples.

[0032] In one specific implementation of the second aspect, the sample generation device may further include:

[0033] The comparison module is used to compare the number of positive and negative samples in the sample set.

[0034] The first determination module is used to identify the class of samples with a smaller sample size as the target sample.

[0035] In one specific implementation of the second aspect, the generation module may include:

[0036] The range determination unit is used to determine the range of feature values ​​generated for key feature dimensions based on the feature distribution.

[0037] The sample generation unit is used to generate new samples for the sample set based on the target sample and the feature value generation range of the key feature dimensions.

[0038] In one specific implementation of the second aspect, the sample generation unit may include:

[0039] Extraction sub-units are used to extract feature values ​​of non-critical feature dimensions of the target sample;

[0040] The first generation subunit is used to generate feature values ​​for the key feature dimension within the range of feature value generation for the key feature dimension.

[0041] The second generation subunit is used to generate new samples for the sample set based on the feature values ​​of non-key feature dimensions and the feature values ​​of key feature dimensions.

[0042] In one specific implementation of the second aspect, the sample generation unit may further include:

[0043] The noise-adding subunit is used to add random noise to the feature values ​​of non-critical feature dimensions, and the random noise is less than a preset noise threshold.

[0044] Correspondingly, the second generation subunit is specifically used to generate new samples for the sample set based on the feature values ​​of the key feature dimensions and the feature values ​​of the non-key feature dimensions after adding random noise.

[0045] In one specific implementation of the second aspect, the filtering module may include:

[0046] The scoring acquisition unit is used to acquire importance scores for each feature dimension related to the sample set based on a preset importance assessment model.

[0047] The dimension determination unit is used to determine the key feature dimensions based on the importance scores of each feature dimension and the preset first feature dimension screening conditions.

[0048] In one specific implementation of the second aspect, the sample generation device may further include:

[0049] The second determining module is used to determine the feature dimensions to be deleted when the number of feature dimensions related to the sample set exceeds a preset threshold for the number of feature dimensions, based on the importance score of each feature dimension and the preset second feature dimension filtering conditions.

[0050] The elimination module is used to remove the feature dimensions to be eliminated from the feature dimensions associated with the sample set.

[0051] Thirdly, this application provides a credit assessment method, including:

[0052] Obtain user data for the user to be evaluated;

[0053] User data is input into a user classification model, an RFM (Recency, Frequency, Monetary) model, and a user behavior model, respectively, to obtain the first credit score of the user to be evaluated output by the user classification model, the second credit score of the user to be evaluated output by the RFM model, and the third credit score of the user to be evaluated output by the user behavior model. Among them, at least one sample in the sample set used by the user classification model during training is obtained based on the sample generation method provided in the first aspect.

[0054] Based on the first credit score, second credit score, and third credit score, a comprehensive credit score is determined for the user to be evaluated.

[0055] Fourthly, this application provides a credit assessment system, including:

[0056] The acquisition module is used to acquire user data of the user to be evaluated.

[0057] The processing module is used to input user data into the user classification model, the RFM model and the user behavior model respectively, and obtain the first credit score of the user to be evaluated output by the user classification model, the second credit score of the user to be evaluated output by the RFM model and the third credit score of the user to be evaluated output by the user behavior model. Among them, at least one sample in the sample set used by the user classification model during training is obtained based on the sample generation method provided in the first aspect.

[0058] The determination module is used to determine the comprehensive credit score of the user to be evaluated based on the first credit score, the second credit score, and the third credit score.

[0059] Fifthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the above-described sample generation methods and / or credit assessment methods.

[0060] Sixthly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the above-described sample generation methods and / or credit assessment methods.

[0061] Seventhly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above-described sample generation methods and / or credit assessment methods.

[0062] The advantages of this application compared to existing technologies are as follows: First, this application selects key feature dimensions from the feature dimensions associated with the sample set. The sample set includes positive and negative samples. Positive samples are obtained from historical users with risky transaction behavior, and negative samples are obtained from historical users with non-risky transaction behavior. The sample set is used to train a user classification model, which is used to evaluate the credit score of the user to be evaluated. The credit score characterizes the degree of credit risk of the user to be evaluated. Then, the feature distributions of the key feature dimensions of at least two target samples are obtained. The target samples are samples in the sample set, and both target samples are either positive or both are negative. Finally, based on the feature distributions and the target samples, new samples are generated for the sample set. These new samples are either both positive or both are negative, just like the target samples. This application's solution specifically considers the key feature dimensions of a class of samples (e.g., positive or negative samples), specifically by considering the feature distributions under the key feature dimensions. By leveraging key feature dimensions, the representativeness and distinguishability of samples can be improved. By analyzing feature distribution, the inherent data relationships between samples of the same type can be understood, thereby making the generated samples more representative and effectively expanding the sample set. This helps to improve the accuracy and generalization ability of user classification models.

[0063] It is understood that the beneficial effects of the second to seventh aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0064] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 This is a flowchart illustrating the sample generation method provided in the embodiments of this application;

[0066] Figure 2 This is a flowchart illustrating the credit assessment method provided in the embodiments of this application;

[0067] Figure 3 This is an example diagram illustrating the implementation of the credit assessment method provided in the embodiments of this application;

[0068] Figure 4 This is a structural block diagram of the sample generation device provided in the embodiments of this application;

[0069] Figure 5 This is a structural block diagram of the credit assessment system provided in the embodiments of this application;

[0070] Figure 6 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0071] To make the inventive objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0072] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0073] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0074] It should also be further understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0075] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [the described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [the described condition or event] is detected," or "in response to detection of [the described condition or event]."

[0076] Furthermore, in the description of this application, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0077] In the field of machine learning, the sample set is the foundation for model training, validation, and testing, playing a crucial role throughout the entire machine learning process. It directly affects the model's performance, generalization ability, and final application results. Generally speaking, a larger sample set is considered to allow the model to better capture patterns in the data, while with a small sample set, the model may struggle to generalize to new data. Furthermore, imbalanced data in the sample set can also lead to significant deviations in the model's predictions.

[0078] Currently, a common upsampling method to address the problem of imbalanced data sets is SMOTE (Synthetic Minority Oversampling Technique). The basic idea is to find neighboring samples among minority class samples in the feature space, then randomly select a point and synthesize a new minority class sample along the line segment between that point and its neighbors, thereby improving the balance of the data set. However, SMOTE does not consider the inherent relationships between sample data, meaning that the generated new samples may not always accurately reflect the distribution of the real data, leading to the introduction of noise or unnecessary data assumptions into the data set. Therefore, even with optimization of the data set using SMOTE, it is still difficult to avoid impacting the accuracy and generalization ability of machine learning models.

[0079] In view of this, embodiments of this application provide a sample generation method that makes the generated samples more representative, thereby helping to improve the accuracy and generalization ability of the model. To illustrate the technical solution proposed in these embodiments, specific examples are provided below.

[0080] The following describes a sample generation method provided by an embodiment of this application. The subject executing this sample generation method can be an electronic device, including but not limited to desktop computers, laptops, PDAs, mobile phones, and servers, etc., and is not limited thereto. Please refer to... Figure 1 The sample generation method in this application embodiment includes:

[0081] Step 101: Select key feature dimensions from the feature dimensions related to the sample set.

[0082] To train the model, the electronic device can first acquire a sample set; alternatively, the sample set can be pre-collected by other devices, and the electronic device only needs to receive and import the pre-collected sample set. This application does not limit the method by which the electronic device acquires the sample set. Based on the model's prediction purpose, each sample in the sample set can be pre-classified as positive or negative samples.

[0083] Specifically, in the fintech field, sample sets can be used to train user classification models. These models can then be used to assess the credit scores of users, which characterize the degree of credit risk. For example, a higher credit score indicates a lower perceived credit risk, meaning the user is considered relatively trustworthy; conversely, a lower credit score indicates a higher perceived credit risk, meaning the user is considered relatively untrustworthy.

[0084] Based on the prediction objective of the above user classification model, positive samples can be defined as: samples obtained from historical users with risky transaction behavior, where risky transaction behavior includes but is not limited to refund behavior and / or order dispute behavior, etc., which are not limited here; correspondingly, negative samples can be defined as: samples obtained from historical users with non-risky transaction behavior, where non-risky transaction behavior includes but is not limited to recharge behavior when no risky transaction behavior has occurred, etc., which are not limited here.

[0085] Each sample in a sample set can be viewed as a multi-dimensional vector, where each dimension represents a different feature; that is, there are usually multiple feature dimensions associated with the sample set. Electronic devices can select feature dimensions with strong representativeness and discriminative power from these feature dimensions. These feature dimensions have a significant impact on the model's performance and accurately reflect the characteristics of the samples. For ease of differentiation, these selected feature dimensions can be referred to as key feature dimensions. In some examples, electronic devices can use statistical methods to select key feature dimensions, such as evaluating the importance of each feature using statistical methods like information gain, chi-square test, or Pearson correlation coefficient, and selecting the feature with the highest score. In other examples, electronic devices can also use model-based methods, such as using feature importance scoring in tree models (including but not limited to random forest models or XGBoost models) to select the most critical feature dimensions. In still other examples, electronic devices can also use principal component analysis (PCA) to select key feature dimensions, such as using dimensionality reduction techniques like PCA to compress high-dimensional features into a few principal components, extracting the most representative features.

[0086] In some embodiments, to achieve a comprehensive consideration of feature dimensions and ensure that key feature dimensions can indeed have a significant impact on model performance, the specific process of selecting key feature dimensions based on the model of an electronic device can be as follows:

[0087] First, based on a pre-defined importance assessment model, the electronic device can obtain importance scores for each feature dimension related to the sample set. As described above, importance assessment models include, but are not limited to, XGBoost and Random Forest models, etc., and are not limited here. These importance assessment models can output importance scores for each feature dimension for use by the electronic device.

[0088] Then, based on the importance scores of each feature dimension and the preset first feature dimension screening criteria, the electronic device can determine the key feature dimensions. The first feature dimension screening criteria can be: feature dimensions with an importance score greater than a preset first importance score threshold are key feature dimensions; or, the first feature dimension screening criteria can also be: the top N feature dimensions ranked from highest to lowest importance score are key feature dimensions. That is, the electronic device can determine the top N feature dimensions with the highest importance scores as key feature dimensions, or it can determine feature dimensions with an importance score higher than the first importance score threshold as key feature dimensions. Of course, the electronic device can also set the first feature dimension screening criteria based on other standards, and this embodiment does not limit this.

[0089] In some embodiments, after initially obtaining the key feature dimensions through the methods proposed above, the electronic device may further consider the correlation between these key feature dimensions. By assessing the correlation between the key feature dimensions, the electronic device can eliminate highly redundant or linearly dependent key feature dimensions, thereby ensuring that the retained key feature dimensions have strong discriminative power and independence.

[0090] Step 102: Obtain the feature distribution of key feature dimensions of at least two target samples.

[0091] Electronic devices can identify a specific type of sample (e.g., positive or negative samples) as a target sample. Generally, there are two or more target samples.

[0092] In one application scenario, an electronic device can first detect whether the total number of samples in the sample set has reached a preset sample number threshold. If the total number of samples fails to reach the threshold, it is considered that the current sample set is at risk of insufficient sample quantity. In this case, the target sample can be a positive sample in the sample set to increase the number of positive samples; or, the target sample can be a negative sample in the sample set to increase the number of negative samples, which is not limited here.

[0093] In another application scenario, electronic devices can first compare the number of positive and negative samples in the sample set to determine whether there is a significant imbalance between the positive and negative samples. If there is a significant imbalance, the sample with fewer samples is identified as the target sample. As an example, the process of comparing positive and negative samples can be briefly described as follows: Determine whether the ratio of positive to negative samples is greater than a preset first sample ratio threshold; and whether the ratio is less than a preset second sample ratio threshold. The first sample ratio threshold is greater than 1, and the second sample ratio threshold is greater than 0 and less than 1. For example, the first sample ratio threshold could be 2, and the second sample ratio threshold could be 0.5; this is not limited here. If the ratio is greater than the first sample ratio threshold, a significant imbalance between positive and negative samples is considered to exist, specifically: the number of positive samples in the sample set is significantly greater than the number of negative samples, meaning the number of negative samples is relatively smaller. Therefore, the electronic device can identify the negative sample as the target sample. If the proportion of the sample is less than the second sample proportion threshold, it is considered that there is a significant imbalance between positive and negative samples. Specifically, the number of negative samples in the sample set is significantly greater than the number of positive samples, that is, the number of positive samples is relatively smaller. Therefore, the electronic device can identify the positive samples as the target samples.

[0094] Based on the currently identified target samples, the electronic device can analyze the distribution patterns of at least two target samples across key feature dimensions to understand the performance of the target samples in those dimensions, thereby obtaining the feature distribution. In some examples, the feature distribution can be represented by parameters such as the mean and standard deviation, which are not limited here. It should be noted that when there are more than two key feature dimensions, the electronic device should obtain the feature distribution of at least two target samples in each key feature dimension.

[0095] Step 103: Generate new samples for the sample set based on the feature distribution and the target samples.

[0096] Electronic devices can use any target sample as the basis for sample generation, upsampling the target sample through feature distribution to generate new samples in the sample set, thereby expanding the sample set. It is important to note that the generated new sample belongs to the same class as the target sample; that is, if the target sample is a positive sample, the new sample is also a positive sample; if the target sample is a negative sample, the new sample is also a negative sample.

[0097] In some embodiments, electronic devices may generate new samples in a manner that makes the new samples more representative of that type of sample:

[0098] First, based on the feature distribution, determine the range of feature values ​​for the key feature dimensions. When the feature distribution includes the mean and standard deviation (i.e., the feature distribution is expressed using these two parameters), the standard deviation represents the dispersion of feature values ​​for at least two target samples under the key feature dimension. Therefore, the range of feature values ​​for the key feature dimension can be set to [mean - a*std, mean + a*std]. Here, mean represents the average value; std represents the standard deviation; and a represents the range coefficient, usually a constant, such as 3. Of course, if the feature distribution includes other parameters, this range of feature values ​​can be adjusted accordingly; this is not limited here.

[0099] Then, based on the target sample and the feature value generation range of the key feature dimension, a new sample set is generated. Specifically, the feature value generation range of the key feature dimension can be used as the basis for upsampling. For any target sample, the electronic device can perform the following processing: first, extract the feature values ​​of the non-key feature dimension (i.e., feature dimensions other than the key feature dimension) of the target sample; then, randomly generate the feature value of the key feature dimension within its feature value generation range, realizing a targeted and restricted upsampling operation; at this point, the feature values ​​of both the non-key feature dimension and the key feature dimension have been obtained, and the electronic device can generate a new sample based on this.

[0100] Building upon this, to ensure the generated new samples better reflect the true data distribution, the electronic device can slightly perturb the feature values ​​of the non-critical feature dimensions after extracting them from the target sample. Specifically, random noise is added to these feature values. The random noise is less than a preset noise threshold to prevent excessive noise from causing inaccurate data. In this way, the electronic device can generate new samples based on the feature values ​​of the critical feature dimension and the feature values ​​of the non-critical feature dimension after adding random noise.

[0101] In some embodiments, to improve the efficiency of subsequent model training, the electronic device may also consider simplifying the feature dimensions related to the sample set. Therefore, the sample generation method proposed in this application may further include the following steps: if the number of feature dimensions related to the sample set exceeds a preset threshold, it is considered that there are too many feature dimensions and potentially redundant features. Based on this, the electronic device can determine the feature dimensions to be removed based on the importance score of each feature dimension and a preset second feature dimension filtering condition. The second feature dimension filtering condition may be: feature dimensions with an importance score less than a preset second importance score threshold are the feature dimensions to be removed, where the second importance score threshold is lower than the first importance score threshold mentioned above; or, the second feature dimension filtering condition may also be: after sorting based on importance scores from low to high, the top M feature dimensions are the feature dimensions to be removed. That is, the electronic device can determine the top M feature dimensions with the lowest importance scores as the feature dimensions to be removed, or it can determine feature dimensions with importance scores lower than the second importance score threshold as the feature dimensions to be removed. Of course, the electronic device can also set the second feature dimension filtering condition based on other criteria, and this application embodiment does not limit this. After identifying the feature dimension to be removed, the electronic device can remove that feature dimension from the feature dimensions associated with the sample set; in this way, the amount of data for each sample in the sample set can be reduced, thereby improving the efficiency of subsequent model training.

[0102] The following provides a specific example of the sample generation method proposed in the embodiments of this application to facilitate understanding of the sample generation method:

[0103] To help the user classification model learn the credit scoring patterns exhibited by users based on different transaction behaviors, feature dimensions related to the sample set can be set based on performance related to transaction behaviors. These include, but are not limited to: user level, user experience value, number of refund orders within a specified time, amount of refund orders within a specified time, number of recharge orders within a specified time, and amount of recharge orders within a specified time, etc., which will not be elaborated here. For ease of explanation, it is assumed that the feature dimensions related to the sample set include the following six: A, B, C, D, E, and F. That is, each sample in the sample set includes six feature values, corresponding to the above six feature dimensions. After the electronic device analyzes the sample set using an importance assessment model (such as the XGBoost model), the feature dimensions are sorted in descending order of importance score as ECDABF. Therefore, the electronic device can determine the top three feature dimensions with the highest scores, namely E, C, and D, as the key feature dimensions.

[0104] Assuming the sample set suffers from an imbalance between positive and negative samples, specifically that positive samples are significantly fewer than negative samples, the electronic device can identify positive samples as target samples. Through statistical analysis of the key feature dimensions of all positive samples in the sample set, the electronic device can determine the feature distribution of each key feature dimension, expressed as: E_mean (mean of positive samples in feature dimension E), E_std (standard deviation of positive samples in feature dimension E), C_mean (mean of positive samples in feature dimension C), C_std (standard deviation of positive samples in feature dimension C), D_mean (mean of positive samples in feature dimension D), and D_std (standard deviation of positive samples in feature dimension D).

[0105] Based on E_mean and E_std, the eigenvalue generation range of feature dimension E can be determined as [E_mean-3*E_std, E_mean+3*E_std]; based on C_mean and C_std, the eigenvalue generation range of feature dimension C can be determined as [C_mean-3*C_std, C_mean+3*C_std]; based on D_mean and D_std, the eigenvalue generation range of feature dimension D can be determined as [D_mean-3*D_std, D_mean+3*D_std].

[0106] For any positive sample, assuming its original feature values ​​are: E = 100, C = 50, D = 30, A = 10, B = 20, F = 30, then:

[0107] For the three key feature dimensions E, C and D, feature values ​​can be generated within the determined feature value generation range, resulting in feature value E' within [E_mean-3*E_std,E_mean+3*E_std], feature value C' within [C_mean-3*C_std,C_mean+3*C_std], and feature value D' within [D_mean-3*D_std,D_mean+3*D_std];

[0108] For the three non-critical feature dimensions A, B, and F, the electronic device can be slightly perturbed to obtain feature value A' (obtained by adding random noise 1 to the original feature value 10), feature value B' (obtained by adding random noise 2 to the original feature value 20), and feature value F' (obtained by adding random noise 3 to the original feature value 30).

[0109] At this point, the electronic device has generated a new sample based on a positive sample. This new sample includes feature values ​​A', B', C', D', E', and F'. The process of generating new samples based on other positive samples can be deduced similarly, and will not be elaborated here.

[0110] As can be seen from the above, in this embodiment, key feature dimensions of a class of samples (e.g., positive or negative samples) are specifically considered, specifically the feature distribution under the key feature dimensions. Key feature dimensions can improve the representativeness and distinguishability of samples, while feature distribution can reveal the inherent data relationships between samples of the same type, thereby making the generated samples more representative and effectively expanding the sample set. This helps improve the accuracy and generalization ability of the user's classification model.

[0111] After applying the sample generation method proposed above to the field of fintech, this application also proposes a credit assessment method, which can be used to assess a user's creditworthiness and achieve risk monitoring. Similar to the sample generation method, the subject executing this credit assessment method can be an electronic device, including but not limited to desktop computers, laptops, PDAs, mobile phones, and servers, etc., without limitation here. Please refer to... Figure 2 The credit assessment method in this application embodiment includes:

[0112] Step 201: Obtain user data of the user to be evaluated.

[0113] The user to be evaluated can be any user of a financial application, and this application embodiment does not limit this. After the user to be evaluated logs into the financial application, the electronic device can obtain the user data of the user to be evaluated reported by the financial application. The user data includes, but is not limited to: user transaction data, user social data, and user behavior data, etc., which can be set according to the input requirements of the various models proposed later, and this application embodiment does not limit this.

[0114] Step 202: Input the user data into the user classification model, RFM model and user behavior model respectively to obtain the first credit score of the user to be evaluated output by the user classification model, the second credit score of the user to be evaluated output by the RFM model and the third credit score of the user to be evaluated output by the user behavior model.

[0115] To assess user credit, this application proposes three types of models: a user classification model, an RFM model, and a user behavior model. These three models are described below:

[0116] User classification models can be built based on the XGBoost algorithm. As an example only, the positive samples in the user classification model's sample set can follow the following positive sample rules:

[0117] 1. Users who have processed X1 or more refunded orders within the past six months. In other words, users with at least X1 refunded orders within the past six months can be selected from historical order data as the positive sample source.

[0118] 2. Users whose single refund amount exceeds X2 yuan within the past six months. That is, users who have made refunds within the past six months and whose single refund amount exceeds X2 yuan can be selected from historical order data as positive sample sources.

[0119] 3. Users with 4 or more disputed orders within the most recent X3 months. In other words, users with 4 or more disputed orders can be selected from the order data of the most recent X3 months as a positive sample source. Disputed orders include, but are not limited to, orders that have resulted in disputes or complaints.

[0120] As an example only, the negative samples in the sample set of a user classification model can follow the following positive sample rules:

[0121] After removing users who meet the positive sample rules, users who have been active in the last X5 days and have made in-app purchases are selected. In other words, users with in-app purchase behavior can be selected from the active users in the last X5 days, and then users who meet the positive sample rules are removed from them. The remaining users can be used as a source of negative samples.

[0122] By using the positive and negative sample rules defined above, users with abnormal business behavior characteristics can be effectively identified.

[0123] As an example only, the feature dimensions associated with the sample set of a user classification model may include, but are not limited to, the following:

[0124] The number of other users a user follows can be used to reflect a user's social activity level.

[0125] The number of followers a user has can be used to reflect the user's influence or popularity.

[0126] User levels can be calculated based on user behavior and points, and can be used to reflect a user's level in a system (such as a financial system).

[0127] User experience points can be used to reflect a user's level of activity and engagement.

[0128] The number of days a user has been registered can be used to reflect the duration of a user's usage.

[0129] The number of successful recharge orders placed by a user in the past 180 days can be obtained through statistics and can be used to reflect the user's recharge activity level.

[0130] The number of devices a user has linked (in the last 180 days) can be obtained through statistics and can be used to reflect the user's activity level on different devices;

[0131] The number of times a user device is associated with a UID twice (in the last 180 days) can be obtained through statistics and can be used to reflect the degree of association between users;

[0132] The number of refund orders placed by users associated with a device in the past 30 days over the past 180 days can be obtained through statistics and can be used to reflect the associated behavior between users.

[0133] The refund order amount of users associated with the device in the past 30 days and the past 180 days can be obtained through statistics and can be used to reflect the economic impact of the associated behavior between users.

[0134] The amount a user has recharged in the past 180 days can be obtained through statistics and can be used to reflect the user's recharge behavior.

[0135] By setting the various feature dimensions as described above, the sample set can provide more comprehensive input information for the user classification model, thereby improving the predictive ability and performance of the user classification model.

[0136] At this point, the sample set for the user classification model can be constructed. However, the sample set constructed in this way often suffers from an imbalance between positive and negative samples. Therefore, the electronic device can generate new samples belonging to the minority class from the constructed sample set based on the sample generation method proposed above, thereby increasing the number of minority class samples in the sample set until the imbalance between positive and negative samples is basically eliminated. During the training process, the electronic device can train the user classification model based on this sample set (that is, the sample set generated and supplemented with new samples using the sample generation method proposed above), thus obtaining the trained user classification model.

[0137] It is understandable that, based on the user data of the user to be evaluated, the user classification model can obtain two probability values ​​for the user to be evaluated: a positive sample probability value and a negative sample probability value. According to the positive sample rules and negative sample rules given above, negative samples originate from relatively high-quality customers; therefore, the user classification model can output the first credit score of the user to be evaluated based on the negative sample probability value.

[0138] The Recency Factor (RFM) model is a commonly used model for user analysis and management. It assesses a user's value and importance by analyzing metrics such as the time of their most recent purchase (Recency), purchase frequency, and purchase amount (Monetary). Specifically, it begins by obtaining each user's purchase records, including but not limited to transaction dates and amounts. Then, based on these records, the RFM metric for each user is calculated. Next, an RFM threshold is calculated using the average of these RFM metrics, and users are stratified based on this threshold, thus establishing the RFM model. Of course, this RFM model can be fine-tuned after its establishment. For example, user numbers can be counted based on the user stratification defined in the RFM model to calculate the percentage of users in each stratum. This percentage is then compared with reference data provided by operations personnel. If significant discrepancies exist, the RFM threshold can be adjusted, and the user stratification can be redefining.

[0139] It's understandable that, for the RFM model, different user tiers correspond to different credit score ranges. Based on the user data of the user to be evaluated, the RFM model processes the data to determine the user tier to which the user belongs. Finally, the RFM model outputs the user's second credit score based on the maximum value of the credit score range corresponding to that user tier.

[0140] User behavior models are models that assess user credit based on user behavior. Specifically, user behavior models can calculate historical profile scores, daily activity scores, and daily risk scores based on user data of the user to be assessed.

[0141] As an example only, the historical profile score can be calculated based on the user's historical behavioral data. This historical behavioral data includes, but is not limited to, consumption behavior, transaction frequency, transaction amount, and refund status, etc., which will not be elaborated here. It can be understood that this historical profile score reflects the user's credit performance and stability over a past period.

[0142] As an example only, the daily activity score can be calculated based on the user's daily activity data. This daily activity data includes, but is not limited to, login frequency, number of transactions, and number of pages viewed, etc., which will not be elaborated here. It can be understood that the daily activity score reflects the user's current behavioral status and level of engagement.

[0143] As an example only, the daily risk score can be calculated based on the user's risk data for that day. This daily risk data includes, but is not limited to, abnormal transactions, abnormal logins, and abnormal access. It can be understood that the daily risk score reflects the user's current risk status and stability.

[0144] It is understandable that the user behavior model can calculate a weighted average of the user's historical profile score, daily activity score, and daily risk score, and output a third credit score based on the weighted calculation result. The weights in the weighted calculation can be adjusted according to actual circumstances to ensure that the impact of different factors on the credit score is reasonably reflected; this is not limited here.

[0145] Step 203: Determine the comprehensive credit score of the user to be evaluated based on the first credit score, the second credit score, and the third credit score.

[0146] Using the three proposed models (user classification model, RFM model, and user behavior model), the electronic device can obtain three credit scores (first credit score, second credit score, and third credit score). These three credit scores represent different evaluation perspectives. To ensure the comprehensiveness and accuracy of the evaluation, the electronic device can comprehensively consider the evaluation results of each model and weight the three credit scores to obtain the overall credit score of the user being evaluated. The weights used in the weighting calculation can be adjusted according to actual circumstances and are not limited here.

[0147] Please see Figure 3 , Figure 3 An example diagram illustrating the implementation of a credit assessment method is provided. (For example...) Figure 3 As shown, to address the imbalance between positive and negative samples in the user classification model's sample set, the previously proposed sample generation method can be used to increase the number of minority class samples, thereby achieving a basic balance between positive and negative samples and optimizing the sample set. Training the user classification model based on this optimized sample set can improve its generalization ability and accuracy. Furthermore, the user classification model, RFM model, and user behavior model can evaluate the user from different perspectives, and the combined evaluation results yield a final comprehensive credit score, indicating whether the user is trustworthy.

[0148] In this embodiment, based on the comprehensive credit score of the user to be evaluated, the electronic device can perform targeted automated decision-making for that user. For example, in a financial loan approval application scenario, the electronic device can quickly decide whether to approve the user's loan application based on the user's comprehensive credit score; in a consumer service application scenario, the electronic device can intelligently decide whether the user is suitable for installment services based on the user's comprehensive credit score, helping to reduce the merchant's credit risk.

[0149] Corresponding to the sample generation method in the above embodiment, Figure 4This diagram illustrates a structural block diagram of a sample generation apparatus according to an embodiment of this application. The sample generation apparatus 4 may include:

[0150] The filtering module 401 is used to filter out key feature dimensions from the feature dimensions related to the sample set. The sample set contains positive samples and negative samples. Positive samples are samples obtained based on historical users with risky transaction behavior, and negative samples are samples obtained based on historical users with non-risky transaction behavior. The sample set is used to train the user classification model. The user classification model is used to evaluate the credit score of the user to be evaluated. The credit score is used to characterize the degree of credit risk of the user to be evaluated.

[0151] The acquisition module 402 is used to acquire the feature distribution of key feature dimensions of at least two target samples, wherein the target samples are samples in the sample set, and at least two target samples are either both positive samples or both negative samples.

[0152] The generation module 403 is used to generate new samples for the sample set based on the feature distribution and the target sample, wherein the new sample and the target sample are either both positive samples or both negative samples.

[0153] In one specific implementation of this application embodiment, the sample generation device 4 may further include:

[0154] The comparison module is used to compare the number of positive and negative samples in the sample set.

[0155] The first determination module is used to identify the class of samples with a smaller sample size as the target sample.

[0156] In one specific implementation of this application embodiment, the generation module 403 may include:

[0157] The range determination unit is used to determine the range of feature values ​​generated for key feature dimensions based on the feature distribution.

[0158] The sample generation unit is used to generate new samples for the sample set based on the target sample and the feature value generation range of the key feature dimensions.

[0159] In one specific implementation of this application embodiment, the sample generation unit may include:

[0160] Extraction sub-units are used to extract feature values ​​of non-critical feature dimensions of the target sample;

[0161] The first generation subunit is used to generate feature values ​​for the key feature dimension within the range of feature value generation for the key feature dimension.

[0162] The second generation subunit is used to generate new samples for the sample set based on the feature values ​​of non-key feature dimensions and the feature values ​​of key feature dimensions.

[0163] In one specific implementation of this application embodiment, the sample generation unit may further include:

[0164] The noise-adding subunit is used to add random noise to the feature values ​​of non-critical feature dimensions, and the random noise is less than a preset noise threshold.

[0165] Correspondingly, the second generation subunit is specifically used to generate new samples for the sample set based on the feature values ​​of the key feature dimensions and the feature values ​​of the non-key feature dimensions after adding random noise.

[0166] In one specific implementation of this application embodiment, the filtering module 401 may include:

[0167] The scoring acquisition unit is used to acquire importance scores for each feature dimension related to the sample set based on a preset importance assessment model.

[0168] The dimension determination unit is used to determine the key feature dimensions based on the importance scores of each feature dimension and the preset first feature dimension screening conditions.

[0169] In one specific implementation of this application embodiment, the sample generation device 4 may further include:

[0170] The second determining module is used to determine the feature dimensions to be deleted when the number of feature dimensions related to the sample set exceeds a preset threshold for the number of feature dimensions, based on the importance score of each feature dimension and the preset second feature dimension filtering conditions.

[0171] The elimination module is used to remove the feature dimensions to be eliminated from the feature dimensions associated with the sample set.

[0172] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0173] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0174] A credit assessment method corresponding to the above embodiment, Figure 5 This diagram illustrates a structural block diagram of a credit assessment system provided in an embodiment of this application. The credit assessment system 5 may include:

[0175] Module 501 is used to acquire user data of the user to be evaluated.

[0176] The processing module 502 is used to input user data into the user classification model, the RFM model and the user behavior model respectively, and obtain the first credit score of the user to be evaluated output by the user classification model, the second credit score of the user to be evaluated output by the RFM model and the third credit score of the user to be evaluated output by the user behavior model. Among them, at least one sample in the sample set used by the user classification model during training is obtained based on the sample generation method proposed above.

[0177] The determination module 503 is used to determine the comprehensive credit score of the user to be evaluated based on the first credit score, the second credit score, and the third credit score.

[0178] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0179] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0180] Figure 6 A schematic block diagram of an electronic device provided in an embodiment of this application is shown. For ease of explanation, only the parts related to the embodiment of this application are shown.

[0181] like Figure 6 As shown, the electronic device 6 of this embodiment includes: a processor 602, a memory 601, and a computer program stored in the memory 601 and executable on the processor 602. When the processor 602 executes the computer program, it implements the steps in the various method embodiments described above, for example... Figure 1 Steps 101 to 103 shown, and / or, Figure 2 Steps 201 to 203 are shown. Alternatively, when the processor 602 executes the computer program, it implements the functions of each module / unit in the above-described device / system embodiments, for example... Figure 4 The functions of modules 401 to 403 shown, and / or, Figure 5 The functions of modules 501 to 503 are shown.

[0182] For example, a computer program may be divided into one or more modules / units, one or more of which are stored in memory 601 and executed by processor 602 to complete this application. One or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in electronic device 6.

[0183] Electronic device 6 may include, but is not limited to, desktop computers, laptops, PDAs, mobile phones, and servers. Those skilled in the art will understand that... Figure 6 This is merely an example of electronic device 6 and does not constitute a limitation on electronic device 6. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 6 may also include input / output devices, network access devices, buses, etc.

[0184] The processor 602 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0185] The memory 601 can be an internal storage unit of the electronic device 6, such as a hard disk or RAM. The memory 601 can also be an external storage device of the electronic device 6, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory 601 can include both internal and external storage units of the electronic device 6. The memory 601 is used to store computer programs and other programs and data required by the electronic device 6. The memory 601 can also be used to temporarily store data that has been output or will be output.

[0186] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0187] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0188] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0189] In the embodiments provided in this application, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0190] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0191] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0192] If an integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable storage medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable storage medium does not include electrical carrier signals and telecommunication signals.

[0193] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A sample generation method, characterized in that, include: Key feature dimensions are selected from the feature dimensions associated with the sample set, which includes positive and negative samples. The positive samples are samples obtained based on historical users with risky transaction behavior, and the negative samples are samples obtained based on historical users with non-risky transaction behavior. The sample set is used to train a user classification model, which is used to evaluate the credit score of the user to be evaluated. The credit score is used to characterize the degree of credit risk of the user to be evaluated. Obtain the feature distribution of key feature dimensions of at least two target samples, wherein the target samples are samples in the sample set, and the at least two target samples are either both positive samples or both negative samples; Based on the feature distribution and the target sample, a new sample is generated for the sample set, wherein the new sample and the target sample are either both positive samples or both negative samples.

2. The sample generation method as described in claim 1, characterized in that, Before obtaining the feature distribution of key feature dimensions of at least two target samples, the sample generation method further includes: Compare the number of positive samples and negative samples in the sample set; The sample with the smaller sample size is identified as the target sample.

3. The sample generation method as described in claim 1 or 2, characterized in that, The step of generating new samples for the sample set based on the feature distribution and the target samples includes: Based on the feature distribution, determine the feature value generation range of the key feature dimension; Based on the target sample and the feature value generation range of the key feature dimension, a new sample of the sample set is generated.

4. The sample generation method as described in claim 3, characterized in that, The step of generating new samples for the sample set based on the target sample and the feature value range of the key feature dimension includes: Extract the feature values ​​of the non-critical feature dimensions of the target sample; Within the range of feature values ​​generated for the key feature dimension, generate the feature values ​​for the key feature dimension; Based on the feature values ​​of the non-critical feature dimensions and the feature values ​​of the critical feature dimensions, new samples of the sample set are generated.

5. The sample generation method as described in claim 4, characterized in that, After extracting the feature values ​​of the non-critical feature dimensions of the target sample, the sample generation method further includes: Random noise is added to the feature values ​​of the non-critical feature dimensions, and the random noise is less than a preset noise threshold. Accordingly, generating new samples for the sample set based on the feature values ​​of the non-critical feature dimensions and the feature values ​​of the critical feature dimensions includes: Based on the feature values ​​of the key feature dimensions and the feature values ​​of the non-key feature dimensions after adding the random noise, a new sample of the sample set is generated.

6. The sample generation method according to any one of claims 1 to 5, characterized in that, The process of determining key feature dimensions from feature dimensions associated with the sample set includes: Based on a preset importance assessment model, importance scores for each feature dimension related to the sample set are obtained; The key feature dimensions are determined based on the importance scores of each feature dimension and the preset first feature dimension screening conditions.

7. The sample generation method as described in claim 6, characterized in that, The sample generation method further includes: If the number of feature dimensions related to the sample set exceeds a preset threshold for the number of feature dimensions, the feature dimensions to be removed are determined based on the importance score of each feature dimension and the preset second feature dimension filtering conditions. The feature dimension to be removed is removed from the feature dimensions associated with the sample set.

8. A credit assessment method, characterized in that, include: Obtain user data for the user to be evaluated; The user data is input into a user classification model, an RFM model, and a user behavior model, respectively, to obtain the first credit score of the user to be evaluated output by the user classification model, the second credit score of the user to be evaluated output by the RFM model, and the third credit score of the user to be evaluated output by the user behavior model. The user classification model uses at least one sample in the sample set during training, which is obtained based on the sample generation method according to any one of claims 1 to 7. Based on the first credit score, the second credit score, and the third credit score, a comprehensive credit score for the user to be evaluated is determined.

9. A sample generation device, characterized in that, include: A filtering module is used to filter out key feature dimensions from the feature dimensions related to the sample set. The sample set includes positive samples and negative samples. The positive samples are samples obtained based on historical users with risky transaction behavior, and the negative samples are samples obtained based on historical users with non-risky transaction behavior. The sample set is used to train a user classification model. The user classification model is used to evaluate the credit score of the user to be evaluated. The credit score is used to characterize the degree of credit risk of the user to be evaluated. The acquisition module is used to acquire the feature distribution of key feature dimensions of at least two target samples, wherein the target samples are samples in the sample set, and the at least two target samples are either both positive samples or both negative samples; A generation module is used to generate new samples for the sample set based on the feature distribution and the target sample, wherein the new samples and the target samples are either both positive samples or both negative samples.

10. A credit assessment system, characterized in that, include: The acquisition module is used to acquire user data of the user to be evaluated. The processing module is used to input the user data into a user classification model, an RFM model, and a user behavior model respectively, to obtain the first credit score of the user to be evaluated output by the user classification model, the second credit score of the user to be evaluated output by the RFM model, and the third credit score of the user to be evaluated output by the user behavior model, wherein at least one sample in the sample set used by the user classification model during training is obtained based on the sample generation method according to any one of claims 1 to 7. The determination module is used to determine the comprehensive credit score of the user to be evaluated based on the first credit score, the second credit score, and the third credit score.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 7; and / or, when the processor executes the computer program, it implements the method as described in claim 8.

12. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7; and / or, when the computer program is executed by a processor, it implements the method as described in claim 8.

Images