Entity card issuing determination method and device, equipment, medium and product

Abstract

The invention provides an entity card issuing determination method and device, equipment, a medium and a product, relates to artificial intelligence, and can be used in the field of financial science and technology. Comprising the steps of issuing a virtual card to a to-be-decided user, and collecting user behavior data and user feedback information of the virtual card; an SVM model is adopted to analyze the user behavior data, credit risk level information of the to-be-decided user is obtained, and the credit risk level information comprises a low credit risk level and a high credit risk level; the user feedback information is analyzed according to the NLP technology and a logistic regression model, emotion information of the to-be-decided user is obtained, and the emotion information comprises forward emotion and reverse emotion; and inputting the credit risk level information and the emotion information into the XGBoost model to obtain an analysis result, thereby improving the processing efficiency and security of credit card application.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of financial technology, in particular to a method and device for determining whether to issue a physical card, equipment, medium and product. BACKGROUND

[0002] In recent years, the application and use of credit cards have become the focus of more and more families and enterprises. With the development of big data, artificial intelligence and blockchain technology, how to improve the efficiency, security and user experience of credit card application has become an important research direction.

[0003] In the prior art, the review and issuance of a credit card is mainly through manual comprehensive evaluation of the applicant's data to determine the applicant's qualifications, credit status and repayment ability to determine whether to issue a card to the applicant.

[0004] However, the above-mentioned application process is complicated and the review period is long, which reduces the efficiency and security of credit card issuance. SUMMARY

[0005] The present application provides a method, device, equipment, medium and product for determining whether to issue a physical card to solve the technical problems of the existing credit card application process being complicated, low in security and low in efficiency.

[0006] In a first aspect, the present application provides a method for determining whether to issue a physical card, comprising:

[0007] issuing a virtual card to a user to be decided and collecting user behavior data and user feedback information of the virtual card, the virtual card being used for real consumption by the user to be decided, the user behavior data including the number of consumption, the amount of consumption and the frequency of consumption, and the user feedback information including function feedback, service quality feedback and security feedback;

[0008] analyzing the user behavior data using an SVM model to obtain credit risk level information of the user to be decided, the credit risk level information including low credit risk level and high credit risk level;

[0009] analyzing the user feedback information according to NLP technology and a logistic regression model to obtain sentiment information of the user to be decided, the sentiment information including positive sentiment and negative sentiment;

[0010] inputting the credit risk level information and the sentiment information into an XGBoost model to obtain an analysis result, the analysis result being used to determine whether to issue a physical card to the user to be decided.

[0011] In a second aspect, the present application provides a device for determining whether to issue a physical card, comprising:

[0012] The information collection module is used to issue virtual cards to users who are to make decisions, and to collect user behavior data and user feedback information of the virtual cards. The virtual cards are used by the users to make real consumption. The user behavior data includes the number of consumptions, the amount of consumption, and the frequency of consumption. The user feedback information includes functional feedback, service quality feedback, and security feedback.

[0013] The credit risk level information acquisition module is used to analyze the user behavior data using an SVM model to obtain the credit risk level information of the user to be decided, wherein the credit risk level information includes low credit risk level and high credit risk level.

[0014] The emotional information acquisition module for users to be decided is used to analyze the user feedback information based on NLP technology and logistic regression model to acquire the emotional information of the users to be decided, including positive and negative emotions.

[0015] The analysis results module is used to input the credit risk level information and the sentiment information into the XGBoost model to obtain analysis results, which are used to determine whether to issue a physical card to the user to be decided.

[0016] Thirdly, this application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0017] The memory stores computer-executed instructions;

[0018] The processor executes computer execution instructions stored in the memory to implement the method as described in the first aspect.

[0019] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method described in the first aspect.

[0020] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0021] The method, apparatus, equipment, medium, and product for determining the issuance of physical cards provided in this application issue virtual cards to users awaiting decision-making and collect user behavior data and user feedback information for these virtual cards. The virtual cards are used by the users to make real purchases. User behavior data includes the number of transactions, transaction amount, and transaction frequency. User feedback information includes functional feedback, service quality feedback, and security feedback. An SVM model is used to analyze the user behavior data to obtain the credit risk level information of the users awaiting decision-making, including low and high credit risk levels. NLP technology and a logistic regression model are used to analyze the user feedback information to obtain the sentiment information of the users awaiting decision-making, including positive and negative sentiment. The credit risk level information and sentiment information are input into an XGBoost model to obtain the analysis results, which are used to determine whether to issue a physical card to the users awaiting decision-making, thus improving the processing efficiency and security of credit card applications. Attached Figure Description

[0022] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0023] Figure 1 An overall architecture diagram of the method for determining the issuance of physical cards provided in this application;

[0024] Figure 2 A flowchart illustrating an embodiment of the method for determining the issuance of physical cards provided in this application;

[0025] Figure 3 A flowchart illustrating Embodiment 2 of the method for determining the issuance of physical cards provided in this application;

[0026] Figure 4 A flowchart illustrating Embodiment 3 of the method for determining the issuance of physical cards provided in this application;

[0027] Figure 5 A flowchart illustrating Embodiment 4 of the method for determining the issuance of physical cards provided in this application;

[0028] Figure 6 A flowchart illustrating Embodiment 5 of the method for determining the issuance of physical cards provided in this application;

[0029] Figure 7 A flowchart illustrating Embodiment Six of the method for determining the issuance of physical cards provided in this application;

[0030] Figure 8 A flowchart illustrating Embodiment Seven of the method for determining the issuance of physical cards provided in this application;

[0031] Figure 9A schematic diagram of the device for determining the issuance of physical cards provided in this application;

[0032] Figure 10 A schematic diagram of the resulting electronic device provided in this application.

[0033] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0034] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0035] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with the relevant laws, regulations, and standards of the relevant countries and regions, have taken necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation access points for users to choose to authorize or refuse.

[0036] Furthermore, the technical solution involved in this application, which involves big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.) and the use of artificial intelligence technology for automated decision-making, and makes decisions that have a significant impact on personal rights based on the results of automated decision-making, provides users with corresponding operation entry points for users to choose to agree to or reject the results of automated decision-making; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0037] It should be noted that the methods, devices, equipment, media and products for determining the issuance of physical cards provided in this application can be used in the field of fintech, or in any field other than fintech. The application fields of the methods, devices, equipment, storage and products for determining the issuance of physical cards in this application are not limited.

[0038] In existing credit card or virtual card issuance processes, user credit assessment and decision-making primarily rely on a single data dimension (such as a user's transaction history or historical credit score), lacking comprehensive integration of user behavior and subjective feedback. This approach not only fails to adequately reflect a user's actual credit risk and service satisfaction but also easily leads to biased assessments and decision-making errors. Furthermore, traditional assessment methods largely depend on manual review or rule engines, which are inefficient and struggle to adapt to the non-linear distribution and complex patterns of user behavior data, further limiting the accuracy and automation level of issuance analysis.

[0039] To address the aforementioned issues, this technical solution proposes a method, device, equipment, medium, and product for determining physical card issuance based on multi-dimensional data collection and intelligent analysis. It utilizes an SVM model for credit risk assessment and employs NLP techniques and a logistic regression model for sentiment analysis, comprehensively integrating users' objective behavior and subjective experience. Simultaneously, it leverages the XGBoost model to achieve intelligent decision-making based on multi-dimensional data, accurately determining whether to issue a physical card. This automates and automates credit assessment and sentiment analysis, improving the comprehensiveness and accuracy of decision-making, significantly increasing process efficiency, and optimizing the user experience.

[0040] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0041] Figure 1 A schematic diagram illustrating the overall architecture of the method for determining the issuance of physical cards provided in this application. (See diagram below.) Figure 1As shown, the overall architecture 10 of the method for determining physical card issuance includes an information collection section 101, a credit risk assessment section 102, a sentiment analysis section 103, a comprehensive analysis section 104, and a feedback and adjustment section 105. The information collection section 101 collects user feedback, including functional feedback, service quality feedback, and security feedback, providing a comprehensive user data foundation for subsequent credit assessment and sentiment analysis. The credit risk assessment section 102 processes complex or linearly inseparable user behavior data using methods such as linear classification, slack variables, and kernel functions, outputting the user's credit risk level (high or low credit risk), providing core credit indicators for issuance analysis. The sentiment analysis section 103 uses a logistic regression model to classify the extracted text features for sentiment, outputting the user's sentiment information (positive or negative sentiment), providing a quantitative analysis of the user's subjective satisfaction with the service, supplementing the deficiencies in behavioral data, and improving the comprehensiveness of the analysis. The comprehensive analysis section 104 combines multi-dimensional features to intelligently generate the final analysis results, improving the accuracy and scientific rigor of the analysis results. The feedback and adjustment section 105 dynamically adjusts the model parameters to adapt to changes in user behavior patterns and feedback information, ensuring the robustness and scalability of the method in practical applications.

[0042] The overall architecture of the physical card issuance determination method provided in this application, through multi-part collaboration, comprehensively improves the efficiency, accuracy, and user experience of physical card issuance determination, while reducing potential financial risks.

[0043] Figure 2 This is a flowchart illustrating an embodiment of the method for determining the issuance of physical cards provided in this application. Figure 2 As shown, it includes:

[0044] S201. Issue virtual cards to users awaiting decision-making and collect user behavior data and user feedback information from the virtual cards.

[0045] Virtual cards are payment tools that do not require a physical medium. They are typically used for online or offline consumption and have a corresponding virtual card number. They are used by users who are about to make a decision to make real purchases. Their functions are similar to physical cards, but they are easier to distribute and manage.

[0046] User behavior data describes a user's specific actions when using a virtual card, including the number of transactions, transaction amount, and transaction frequency.

[0047] User feedback refers to user feedback on functionality, service quality, and security after using the virtual card. Functional feedback refers to user satisfaction with the virtual card usage process and functions, service quality feedback refers to user evaluation of customer service or platform services, and security feedback refers to user trust in the privacy protection and transaction security of the virtual card.

[0048] Specifically, virtual cards will be issued to potential users, allowing them to use them for everyday transactions such as online shopping and money transfers. During the trial period, user behavior data will be recorded in the backend. For example, if a user makes five transactions totaling 1000 yuan within a week, averaging once a day, the system will also record user feedback on the virtual card's functionality, such as feedback on the convenience of the payment process and the speed of transfers.

[0049] S202. Use the SVM model to analyze user behavior data and obtain credit risk level information of users to be decided.

[0050] Support Vector Machine (SVM) is a machine learning classification model suitable for binary classification tasks with high-dimensional data. By finding the optimal classification hyperplane, the SVM model can distinguish the credit risk level information of users to be decided.

[0051] Credit risk rating information is a quantitative assessment of a user's creditworthiness, including low credit risk and high credit risk. A low credit risk rating indicates that the user has good behavior and a good credit record, while a high credit risk rating indicates that the user has a higher risk of defaulting on their credit.

[0052] Specifically, the collected user behavior data is input into the SVM model, which calculates the credit risk level of each user based on their spending patterns. For example, a user with stable monthly spending and no history of overspending is classified as having low credit risk, while a user with large spending amounts or unusual frequency (such as a large number of transactions within a short period) is classified as having high credit risk. Finally, the output credit risk level information is used to assess the user's financial health, providing a basis for subsequent analysis.

[0053] S203. Analyze user feedback information using NLP techniques and logistic regression models to obtain the emotional information of users who need to make decisions.

[0054] Natural Language Processing (NLP) technology is used to analyze and process text data from user feedback for decision-making, including text cleaning, word segmentation, and word vector representation.

[0055] Logistic regression is a binary classification algorithm used to predict the sentiment information of users' feedback based on feature data. The sentiment information includes positive sentiment and negative sentiment. Positive sentiment indicates that users are satisfied with the virtual card usage experience, while negative sentiment indicates that users are dissatisfied with the virtual card usage experience.

[0056] Specifically, NLP techniques are used to preprocess feedback from users awaiting decision-making. This includes removing irrelevant characters and stop words, converting the text into word vectors, and calculating key words in the feedback. The preprocessed feedback data is then input into a logistic regression model to categorize the sentiment information of the users' feedback. For example, a user's comment, "Virtual card payment is convenient, and the service is timely," is categorized as positive sentiment; conversely, a comment, "Payment failed several times, and customer service was not prompt," is categorized as negative sentiment. In short, the output sentiment information is used for further comprehensive analysis of user satisfaction with the trial.

[0057] S204. Input credit risk level information and sentiment information into the XGBoost model to obtain the analysis results.

[0058] Extreme Gradient Boosting (XGBoost) is an ensemble learning algorithm based on the gradient boosting framework. It can handle nonlinear relationships and generate high-precision prediction results on multidimensional data.

[0059] The analysis results are a comprehensive assessment generated by combining the user's credit risk level and sentiment information, which directly determines whether a physical card is issued to the user who is waiting to make a decision.

[0060] Specifically, the credit risk level and sentiment information of the user to be decided are used as input features and fed into the XGBoost model. The XGBoost model combines these two types of data to make a comprehensive decision: if the user's credit risk level is low and the sentiment feedback is positive, the model outputs "issue a physical card"; if the user's credit risk level is high or the sentiment feedback is negative, the model outputs "do not issue a physical card".

[0061] Additionally, based on the model results, if a physical card is issued, a success notification will be sent; if no physical card is issued, a rejection notification will be sent, briefly informing the user that they failed the trial evaluation.

[0062] The method for determining the issuance of physical cards provided in this embodiment collects user data by issuing virtual cards, and combines SVM model, NLP technology, logistic regression model and XGBoost model to comprehensively analyze the behavior and feedback of users to be decided, and finally accurately determines the decision on issuing physical cards, thereby improving the comprehensiveness, accuracy and intelligence of the decision.

[0063] Figure 3 This is a flowchart illustrating Embodiment Two of the method for determining the issuance of physical cards provided in this application. Figure 3 As shown, in Figure 2Based on the previous embodiment, before issuing virtual cards to users to be decided and collecting user behavior data and user feedback information of the virtual cards, the method further includes:

[0064] S301. Collect user data information from user log information, and train the initial GBDT model based on the user data information to obtain the target GBDT model.

[0065] User log information is used to record historical data of user behavior, including transaction amount, transaction frequency, payment delay days, and consumption category.

[0066] Gradient Boosting Decision Tree (GBDT) is an ensemble learning model that optimizes prediction performance by iteratively fitting residuals.

[0067] Specifically, features are extracted from user log information, such as transaction amount (X1), transaction frequency (X2), payment delay days (X3), and consumption category (X4). These data features (X1, X2, X3, X4) are then standardized using the following formula:

[0068]

[0069] Where σ represents the standard deviation and μ represents the mean.

[0070] Secondly, the GBDT model needs to be trained. The model is initialized as follows:

[0071]

[0072] Then iterative training is performed. For example, in the m-th iteration, the residual is first calculated, which is the difference between the current model prediction and the true value:

[0073]

[0074] Then use the residual To train decision trees The updated model is obtained:

[0075]

[0076] Therefore, after M iterations, the target GBDT model can be obtained:

[0077]

[0078] in, Let represent the decision tree for the m-th lesson, η represent the learning efficiency, and M represent the number of iterations.

[0079] S302. Based on the target GBDT model, perform credit prediction for the user to be decided, and obtain the user's predicted credit rating.

[0080] Credit prediction, in this context, refers to using a user's feature vector x as input to the target GBDT model to calculate a credit risk score, thereby completing the credit prediction. The model outputs a credit risk score. :

[0081]

[0082] in, A value greater than 0.5 indicates high credit risk. A value less than 0.5 indicates low credit risk.

[0083] Specifically, the feature X of the user to be decided is input into the target GBDT model, and the target GBDT model then performs weighted predictions sequentially through each decision tree:

[0084]

[0085] Suppose the output of the first tree is The output of the second tree is Learning rate The overall prediction is:

[0086]

[0087] Then, a credit rating is determined. Since 0.07 > 0.5, the user to be decided is a low-credit-risk user.

[0088] S303. If the user predicts that the credit rating indicates that the user to be decided meets the credit requirements, then a virtual card will be issued to the user to be decided.

[0089] In this step, based on the prediction results of S302, if the user meets the low-risk conditions, a virtual card account is generated for the user to be decided, and the payment function is bound to it. After the virtual card is issued to the user, the user can make consumption.

[0090] The method for determining the issuance of physical cards proposed in this application improves the accuracy of credit assessment by extracting data from user logs and optimizing model parameters through iterative training of the GBDT model. Then, the trained GBDT model is used to predict the credit of users, generate a quantitative credit risk level, and determine whether the user meets the credit requirements based on the prediction results. If the conditions are met, a virtual card is issued to provide users with trial payment functions, thus ensuring the intelligence of credit assessment and the reliability of virtual card issuance.

[0091] Figure 4 This is a flowchart illustrating Embodiment 3 of the method for determining the issuance of physical cards provided in this application.Figure 4 As shown, in Figure 2 Based on the previous example, an SVM model is used to analyze user behavior data to obtain credit risk level information for users to be decided, including:

[0092] S401. Use the SVM model to analyze user behavior data and obtain the feature vector and bias term corresponding to the user behavior data.

[0093] The formula for the SVM model is as follows:

[0094]

[0095] Where w is the feature vector, i.e., the normal vector, representing the direction of the hyperplane. X is the input feature vector, such as user behavior data, and b is the bias term, representing the position of the hyperplane.

[0096] For example, suppose the user's behavioral data is ,

[0097] Then the feature data (such as X1, X2, X3) are standardized:

[0098]

[0099] Where σ represents the standard deviation and μ represents the mean.

[0100] One-hot encoding is performed on categorical features (such as X4), for example, .

[0101] Finally, the feature vector and bias term are calculated, which are the preprocessed features. When input into the SVM model, the feature vectors w=[0.3,0.2,0.1,0.4,-0.1] and the bias term b=-0.5 are obtained.

[0102] S402. Determine the target hyperplane of the SVM model based on the eigenvectors and bias terms.

[0103] The target hyperplane is defined as follows: , is the classification boundary used by SVM to distinguish between high-risk and low-risk users. The feature vector w determines the direction of the hyperplane, and the bias term b determines the position of the hyperplane.

[0104] For example, the calculation of the target hyperplane can be divided into the following steps:

[0105] Calculate user behavior characteristics The inner product of the eigenvector w:

[0106]

[0107] The calculation result is as follows: .

[0108] Therefore, the hyperplane formula is:

[0109] S403. Based on the target hyperplane, obtain the credit risk level information of the user to be decided.

[0110] The credit risk level information is determined based on the target hyperplane data, which specifically classifies users into high credit risk and low credit risk levels. If the target hyperplane data is greater than a preset value, the credit risk level of the user to be decided is determined to be high credit risk; if the target hyperplane data is less than the preset value, the credit risk level of the user to be decided is determined to be low credit risk.

[0111] In the embodiments of this application, the preset value is usually set to 0, but in actual applications it can be adjusted according to the needs of different services.

[0112] For example, assuming the preset value is 0, the target hyperplane data Then determine the risk level. Therefore, the credit risk level of the user to be decided is high.

[0113] The method for determining the issuance of physical cards provided in this application calculates the feature vector and bias term of user behavior data using an SVM model to determine the relative position of user data and classification boundary; then, combining the feature vector and bias term, a target hyperplane is defined to distinguish between high-risk and low-risk users; finally, based on the target hyperplane data, the user's credit risk level is determined, providing intelligent classification results to support subsequent decision-making.

[0114] Figure 5 This is a flowchart illustrating Embodiment 4 of the method for determining the issuance of physical cards provided in this application. Figure 5 As shown, in Figure 4 Based on the embodiments, the method further includes:

[0115] S501. If user behavior data cannot be linearly separated, then slack variables and kernel functions are used to optimize the SVM model.

[0116] Slack variables allow some sample points to violate classification boundary conditions, thus handling noise in the data or outliers on the boundaries. Kernel functions are mapping techniques used to map raw data from a low-dimensional space to a high-dimensional space, making linearly inseparable data linearly separable in the high-dimensional space. Therefore, by using slack variables and kernel functions, the SVM model can be optimized to handle linearly inseparable user behavior data.

[0117] For example, when user behavior data is linearly inseparable, meaning the data is complexly distributed in two-dimensional or three-dimensional space and cannot be effectively segmented using a single linear hyperplane, slack variables are introduced to optimize the classification boundary, and kernel functions are used to map the data to a higher-dimensional space. Specifically:

[0118] The first step is to introduce slack variables to optimize the classification boundary:

[0119] The original SVM classification criteria are:

[0120]

[0121] Add slack variables when the data cannot be linearly separated. Therefore, the SVM model formula becomes:

[0122]

[0123] in, Let be the slack variable for the i-th sample, representing the degree to which this sample deviates from the normal classification margin. Then, the optimization objective is to minimize the classification error and the hyperplane complexity:

[0124]

[0125] Here, C is the penalty parameter, which controls the weight of the slack variables. It's important to note that a larger C reduces classification errors but may also lead to overfitting. Therefore, the value of C needs to be carefully considered when introducing slack variables. Specifically, by adjusting the penalty parameter C, a larger classification tolerance is given to high-credit-risk points, making the SVM model more flexible.

[0126] The second step is to use kernel functions to map the data to a higher-dimensional space:

[0127]

[0128] Specifically, using appropriate kernel functions, such as radial basis function kernels, can capture complex nonlinear modes:

[0129]

[0130] Among them, Y is used to control the influence range of the kernel function. In the embodiment of this application, Y is the optimal value determined by cross-validation.

[0131] After the above two steps, the user behavior data becomes linearly separable, and then data classification can be performed. Specifically, the user behavior features X=[1000,10,5] are input, and the radial basis function is used to map the data to a high-dimensional space. Then, SVM solves for the weights w and bias term b through Lagrange dual optimization, and then performs classification to obtain the classification function:

[0132]

[0133] Among them, if If so, the user to be decided is identified as a high-credit-risk user. If so, the user to be decided is identified as a low-credit-risk user.

[0134] The method for determining the issuance of physical cards provided in this application optimizes the SVM model by introducing slack variables and kernel functions. The optimized SVM model can not only adapt to complex scenarios, but also enhance the accuracy and stability of classification, and realize the processing of linearly inseparable user behavior data, thereby accurately obtaining the credit risk level information of the user to be decided.

[0135] Figure 6 A flowchart illustrating Embodiment 5 of the method for determining the issuance of physical cards provided in this application. Figure 6 As shown, in Figure 2 Based on the implementation examples, user feedback information is analyzed using NLP techniques and logistic regression models to obtain the emotional information of users who need to make decisions, including:

[0136] S601. Based on NLP technology, perform text processing and feature extraction on user feedback information to obtain text features.

[0137] The purpose of NLP technology is to transform unstructured data (such as text data) into structured features (such as numerical vectors).

[0138] Text processing refers to the standardization of user feedback information, such as removing irrelevant characters, word segmentation, and part-of-speech tagging. For example, "The virtual card is very convenient to use and payment is fast" can be transformed into ["virtual card", "use", "convenient", "payment", "fast"].

[0139] Feature extraction refers to representing processed text as quantifiable features. A common method is word vectorization, which maps words to a high-dimensional vector space to capture semantic relationships.

[0140] Exemplarily, process the user feedback "Virtual card payment is fast, but the binding operation is a bit complex", for example, remove stop words (of, but, a bit), extract word segments (virtual card, payment, fast, binding, operation, complex), and then perform feature extraction to convert the text into a vector representation: X = [0.3, 0.2, 0.4, 0.6, 0.1, 0.3, 0.5], where each value corresponds to the importance of a word.

[0141] S602. Classify and predict the text features according to the logistic regression model to obtain the sentiment information of the user to be determined.

[0142] Among them, the logistic regression model used in the embodiments of this application is a linear classification model for binary classification tasks, and can output sentiment classification probabilities through the S-shaped (English: Sigmoid Function, abbreviated: Sigmoid) function:

[0143]

[0144] Among them, is the Sigmoid function; w is the weight vector, indicating the influence degree of each feature on classification, X is the text feature vector, and b is the bias term. In addition, the classification basis: if , then it is classified as positive sentiment, that is, it means the user is satisfied with the service. If , then it is classified as negative sentiment, that is, it means the user is not satisfied with the service.

[0145] Exemplarily, the steps of logistic regression prediction can be: use the previously extracted text feature X = [0.3, 0.2, 0.4, 0.6, 0.1, 0.3, 0.5], assume the logistic regression weight vector w = [0.5, 0.4, 0.3, 0.2, 0.1], and the bias term b = -0.1, then calculate the predicted value:

[0146]

[0147] Then, use the Sigmoid function to convert z into a probability:

[0148]

[0149] Furthermore, perform classification judgment. Assume the preset threshold is 0.5. Since , it is determined that the user has positive sentiment.

[0150] The method for determining the issuance of physical cards provided by the embodiments of this application can perform text processing and feature extraction on user feedback through NLP technology, and achieve sentiment classification in combination with the logistic regression model, which can accurately obtain the sentiment information of users and provide strong support for user decision-making and service optimization.

[0151] Figure 7 A flowchart illustrating Embodiment Six of the method for determining the issuance of physical cards provided in this application. Figure 7 As shown, in Figure 2 Based on the example, credit risk level information and sentiment information are input into the XGBoost model to obtain analysis results, including:

[0152] S701. Input credit risk level information and sentiment information into the XGBoost model to analyze the users who need to make decisions and obtain the analysis results.

[0153] The analysis results indicate whether a physical card is issued. Possible values ​​are a first preset value, which means the physical card is issued to the user awaiting decision, or a second preset value, which means the physical card is issued to the user awaiting decision, and the second preset value is different from the first preset value.

[0154] For example, credit risk level information X1 and sentiment information X2 are used as input features, such as X1=0 (low credit risk) and X2=1 (positive sentiment). Then, the XGBoost model is used for comprehensive analysis. The input feature vector is X=[X1,X2]=[0,1]. The model generates prediction results for each decision tree: decision tree 1 outputs h1(X)=0.4, decision tree 2 outputs ℎ2(X)=0.5, and decision tree 3 outputs ℎ3(X)=0.2h3.

[0155] Finally, the overall decision result is as follows:

[0156]

[0157] Assume weights:

[0158]

[0159] but:

[0160]

[0161] Furthermore, in this embodiment of the application, threshold conditions for a first preset value and a second preset value are set, if... If the analysis result is the first preset value, it indicates that a physical card will be issued. If the result is the second preset value, then no physical card will be issued. In the above calculation, Therefore, the analysis result for seeding is the second preset value, that is, no physical cards will be issued.

[0162] The method for determining the issuance of physical cards provided in this application reduces human intervention by inputting credit risk level information and sentiment information into the XGBoost model, comprehensively assesses the user's credit status and satisfaction, and improves risk control capabilities and user experience, thereby increasing the efficiency of the physical card issuance process.

[0163] Figure 8 This is a flowchart illustrating Embodiment Seven of the method for determining the issuance of physical cards provided in this application. Figure 8 As shown, in Figure 2 Based on the embodiments, the method further includes:

[0164] S801. Obtain behavioral information and feedback information from user log information to obtain behavioral characteristics, feedback characteristics, and user text with sentiment tags.

[0165] Among them, behavioral features are numerical features extracted from user behavior information, such as transaction amount, transaction frequency, and payment delay days. Feedback features are structured features obtained by analyzing user feedback information, including sentiment information and service satisfaction ratings. User text with sentiment tags is tagged data formed after user text data has been labeled with sentiment analysis, including Tag 1: positive sentiment, and Tag 2: negative sentiment.

[0166] Specifically, behavioral information is extracted from user logs: transaction amount X1, transaction frequency X2, and delay days X3. Feedback information includes user evaluation text, such as "fast payment speed, excellent service attitude." Then, feature extraction is performed. For behavioral features, the behavioral information is normalized to obtain a feature vector [X1, X2, X3]. For feedback features, NLP techniques are used to extract text features and label them with sentiment tags. The text "fast payment speed" represents positive sentiment (label 1), and the text "payment failed" represents negative sentiment (label 0).

[0167] S802. Obtain the original XGBoost model and the original logistic regression model.

[0168] The original XGBoost model refers to the XGBoost model in its initial state, which has not been trained and does not contain decision trees. The model formula is:

[0169]

[0170] Where T represents the number of decision trees, which is initially 0.

[0171] The original logistic regression model does not yet include any weight parameters w and bias terms b. The model formula is:

[0172]

[0173] Therefore, the weight w=0 and the bias term b=0.

[0174] S803. Based on the behavioral and feedback features, train the original XGBoost model to obtain the XGBoost model.

[0175] In this step, behavioral features and feedback features are used as inputs to build a decision tree through multiple iterations. The model formula is as follows:

[0176]

[0177] The residual r of each training iteration:

[0178]

[0179] For example, the user's behavioral features X=[1000,12,2] and feedback features X4=1 are input, and the label is the user's credit risk level y; then the model is trained: the fitting residual of the first tree is r=y-f0(X), assuming r=[0.2,0.8,-0.01], and the decision tree is trained using r. The second tree update prediction is:

[0180]

[0181] Then continue fitting the updated residuals. .

[0182] After completing T iterations, the trained XGBoost model can be output.

[0183] S804. Train the original logistic regression model based on user text with sentiment tags to obtain the logistic regression model.

[0184] In this step, the numerical features X of the user text and the sentiment classification label y are input into the original logistic regression model for model training. The initial parameters w = [0,0,0] and the bias term b = 0. Then, the loss function is calculated using the features X and the label y. w and b are updated according to the gradient, and finally, the optimized logistic regression model P(Y=1|X) = σ(w∙X+b) is obtained.

[0185] The method for determining physical card issuance provided in this application obtains behavioral and feedback features from user logs to provide training data for XGBoost and logistic regression models. Then, based on the behavioral and feedback features, a comprehensive model for credit risk assessment is constructed. Furthermore, based on user text with sentiment tags, the model parameters are optimized to accurately classify user sentiment information, thereby improving the XGBoost model's comprehensive processing capability for user behavior and sentiment feedback and enhancing the comprehensiveness and accuracy of physical card issuance analysis.

[0186] Figure 9 A schematic diagram illustrating the result of the device for determining the issuance of the physical card provided in this application. (See diagram below.) Figure 9 As shown, the physical card issuance determining device 90 includes:

[0187] The information collection module 901 is used to issue virtual cards to users who are about to make decisions, and to collect user behavior data and user feedback information of the virtual cards. The virtual cards are used by users who are about to make decisions to make real consumption. The user behavior data includes the number of consumptions, consumption amount and consumption frequency. The user feedback information includes functional feedback, service quality feedback and security feedback.

[0188] The credit risk level information acquisition module 902 is used to analyze user behavior data using an SVM model to obtain the credit risk level information of the user to be decided. The credit risk level information includes low credit risk level and high credit risk level.

[0189] The emotional information acquisition module 903 for users to be decided is used to analyze user feedback information based on NLP technology and logistic regression model to acquire emotional information of users to be decided, including positive and negative emotions.

[0190] The analysis results yield module 904, which is used to input credit risk level information and sentiment information into the XGBoost model to obtain analysis results. These results are used to determine whether to issue physical cards to users who are pending decisions.

[0191] Furthermore, the information acquisition module 901 is also specifically used for:

[0192] User data is collected from user log information, and the initial GBDT model is trained based on the user data to obtain the target GBDT model;

[0193] Based on the target GBDT model, credit prediction is performed on the users to be decided, and the predicted credit rating of the users is obtained.

[0194] If a user predicts that the credit rating indicates the user to be decided meets the credit requirements, then a virtual card will be issued to the user to be decided.

[0195] Furthermore, the credit risk rating information acquisition module 902 is also specifically used for:

[0196] The SVM model is used to analyze user behavior data to obtain feature vectors and bias terms corresponding to the user behavior data;

[0197] The target hyperplane of the SVM model is determined based on the feature vectors and bias terms.

[0198] Based on the target hyperplane, obtain the credit risk level information of the user to be decided;

[0199] If the target hyperplane data is greater than the preset value, the credit risk level information of the user to be decided is determined to be a high credit risk level.

[0200] If the target hyperplane data is less than the preset value, the credit risk level information of the user to be decided is determined to be low credit risk level.

[0201] Furthermore, the credit risk rating information acquisition module 902 is also specifically used for:

[0202] If user behavior data cannot be linearly separated, slack variables and kernel functions are used to optimize the SVM model;

[0203] Among them, slack variables and kernel functions are used to optimize the SVM model into a model that can handle linearly inseparable user behavior data.

[0204] Furthermore, the emotional information acquisition module 903 for users awaiting decision-making is also specifically used for:

[0205] Based on NLP technology, text processing and feature extraction are performed on user feedback information to obtain text features;

[0206] Based on the logistic regression model, text features are classified and predicted to obtain the sentiment information of users to be decided.

[0207] If the classification prediction result is greater than the preset threshold, then the emotional information of the user to be decided is determined to be positive.

[0208] If the classification prediction result is less than or equal to the preset threshold, then the emotional information of the user to be decided is determined to be reverse sentiment.

[0209] Furthermore, the analysis results show that module 904 is also specifically used for:

[0210] Credit risk rating information and sentiment information are input into the XGBoost model to analyze the users who need to make decisions and obtain the analysis results;

[0211] If the analysis result is the first preset value, then the physical card will be issued to the user who is waiting to make a decision.

[0212] If the analysis result is the second preset value, then the physical card will not be issued to the user awaiting decision. The second preset value is different from the first preset value.

[0213] Furthermore, the physical card issuance determination device 90 is also specifically used for:

[0214] Behavioral and feedback information is obtained from user logs to obtain behavioral characteristics, feedback characteristics, and user text with sentiment tags.

[0215] Obtain the original XGBoost model and the original logistic regression model;

[0216] The original XGBoost model is trained based on behavioral and feedback features to obtain the XGBoost model.

[0217] The original logistic regression model is trained based on user text with sentiment tags to obtain the logistic regression model.

[0218] The physical card issuance determination device provided in this application embodiment combines four major functions: information collection, credit assessment, sentiment analysis, and intelligent analysis. It realizes a comprehensive, accurate, and efficient issuance process, improves the accuracy of credit assessment, optimizes user experience, and reduces financial risks, providing an intelligent and automated solution for physical card issuance.

[0219] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 10 As shown, the electronic device 100 includes:

[0220] The electronic device 100 may include a processor 1001 with one or more processing cores, a memory 1002 with one or more computer-readable storage media, a communication component 1003, and other components. The processor 1001, memory 1002, and communication component 1003 are connected via a bus 1004.

[0221] In the specific implementation process, at least one processor 1001 executes computer execution instructions stored in memory 1002, causing at least one processor 1001 to execute the above-mentioned big data-based abnormal number tracing method.

[0222] The specific implementation process of processor 1001 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0223] In the above Figure 10In the illustrated embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0224] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0225] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0226] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0227] According to one aspect of this application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium.

[0228] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0229] It should be understood that the above-described device embodiments are merely illustrative, and the device of this application can also be implemented in other ways. For example, the division of units / modules in the above embodiments is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units, modules, or components may be combined, or integrated into another system, or some features may be ignored or not executed.

[0230] Furthermore, unless otherwise specified, the functional units / modules in the various embodiments of this application can be integrated into one unit / module, or each unit / module can exist physically separately, or two or more units / modules can be integrated together. The integrated units / modules described above can be implemented in hardware or as software program modules.

[0231] In the above embodiments, the descriptions of each embodiment have their own emphasis. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0232] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0233] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for determining the issuance of physical cards, characterized in that, include: Virtual cards are issued to users awaiting decision-making, and user behavior data and user feedback information of the virtual cards are collected. The virtual cards are used by the users awaiting decision-making to make real consumption. The user behavior data includes the number of consumptions, consumption amount and consumption frequency. The user feedback information includes functional feedback, service quality feedback and security feedback. The user behavior data is analyzed using an SVM model to obtain the credit risk level information of the user to be decided, including low credit risk level and high credit risk level. The user feedback information is analyzed using NLP techniques and logistic regression models to obtain the emotional information of the user to be decided, including positive and negative emotions. The credit risk level information and the sentiment information are input into the XGBoost model to obtain the analysis results, which are used to determine whether to issue a physical card to the user to be decided.

2. The method according to claim 1, characterized in that, Before issuing virtual cards to users awaiting decision-making and collecting user behavior data and user feedback information for the virtual cards, the method further includes: User data is collected from user log information, and the initial GBDT model is trained based on the user data to obtain the target GBDT model; Based on the target GBDT model, a credit prediction is performed on the user to be decided, and the user's predicted credit rating is obtained. If the user's predicted credit rating indicates that the user to be decided meets the credit requirements, then a virtual card is issued to the user to be decided.

3. The method according to claim 1, characterized in that, The user behavior data is analyzed using an SVM model to obtain the credit risk level information of the user to be decided, including: The user behavior data is analyzed using an SVM model to obtain the feature vector and bias term corresponding to the user behavior data; The target hyperplane of the SVM model is determined based on the feature vector and the bias term. Based on the target hyperplane, obtain the credit risk level information of the user to be decided; If the target hyperplane data is greater than a preset value, then the credit risk level information of the user to be decided is determined to be a high credit risk level. If the target hyperplane data is less than a preset value, then the credit risk level information of the user to be decided is determined to be a low credit risk level.

4. The method according to claim 3, characterized in that, The method further includes: If the user behavior data cannot be linearly separated, then slack variables and kernel functions are used to optimize the SVM model; The slack variables and the kernel function are used to optimize the SVM model into a model that can handle linearly inseparable user behavior data.

5. The method according to claim 1, characterized in that, The step of analyzing the user feedback information using NLP techniques and logistic regression models to obtain the emotional information of the user to be decided includes: Based on NLP technology, the user feedback information is processed and features are extracted to obtain text features; Based on the logistic regression model, the text features are classified and predicted to obtain the sentiment information of the user to be decided. If the classification prediction result is greater than a preset threshold, then the emotional information of the user to be decided is determined to be positive. If the classification prediction result is less than or equal to the preset threshold, then the emotional information of the user to be decided is determined to be reverse sentiment.

6. The method according to claim 1, characterized in that, The credit risk rating information and the sentiment information are input into the XGBoost model to obtain the analysis results, including: The credit risk level information and the sentiment information are input into the XGBoost model to analyze the user to be decided, and the analysis results are obtained. If the analysis result is a first preset value, then the physical card will be issued to the user to be decided. If the analysis result is the second preset value, then the physical card will not be issued to the user to be decided. The second preset value is different from the first preset value.

7. The method according to claim 1 or 2, characterized in that, The method further includes: Behavioral and feedback information is obtained from user logs to obtain behavioral characteristics, feedback characteristics, and user text with sentiment tags. Obtain the original XGBoost model and the original logistic regression model; The original XGBoost model is trained based on the behavioral features and the feedback features to obtain the XGBoost model. The original logistic regression model is trained based on the user text with sentiment tags to obtain the logistic regression model.

8. A device for determining the issuance of physical cards, characterized in that, include: The information collection module is used to issue virtual cards to users who are to make decisions, and to collect user behavior data and user feedback information of the virtual cards. The virtual cards are used by the users to make real consumption. The user behavior data includes the number of consumptions, the amount of consumption, and the frequency of consumption. The user feedback information includes functional feedback, service quality feedback, and security feedback. The credit risk level information acquisition module is used to analyze the user behavior data using an SVM model to obtain the credit risk level information of the user to be decided, wherein the credit risk level information includes low credit risk level and high credit risk level. The emotional information acquisition module for users to be decided is used to analyze the user feedback information based on NLP technology and logistic regression model to acquire the emotional information of the users to be decided, including positive and negative emotions. The analysis results module is used to input the credit risk level information and the sentiment information into the XGBoost model to obtain analysis results, which are used to determine whether to issue a physical card to the user to be decided.

9. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 7.

11. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method of any one of claims 1 to 7.

Images