Method and device for judging credit risk of user and electronic equipment

Abstract

The invention provides a user credit risk judgment method and device and electronic equipment, and the method comprises the steps: collecting first network operator information corresponding to a mobile phone number provided by a target user in an express entry link after obtaining authorization information of the target user for operator DPI data; obtaining target user bottom layer data from a corresponding operator side based on the first network operator information; performing feature extraction and feature engineering processing on the target user underlying data to obtain target feature data; and inputting the target feature data into a pre-trained credit risk judgment model, processing the target feature data through the credit risk judgment model, and outputting a credit risk result corresponding to the target user. According to the method, the credit risk level of the user can be described based on the unstructured data of the operator besides the credit investigation scope, and the credit risk of the user is supplemented and insighted outside the original credit investigation system.

Description

Technical Field

[0001] The present application relates to the field of financial technology, and in particular to a method, device and electronic device for judging user credit risk. Background Art

[0002] In the credit sector, common methods for assessing user credit risk include credit scoring models, machine learning models, behavioral analysis, social network analysis, rule engines, blacklist systems, real-time monitoring and early warning, multi-dimensional data verification, manual review, and credit risk pattern recognition. These methods are typically used selectively when identifying credit risk. This leads to the following four deficiencies in existing technologies:

[0003] 1. Low automation rate: Traditional credit risk control processes, especially those at consumer finance companies or banks, require face-to-face interviews for most processes. The inclusion of these steps requires significant manual review.

[0004] 2. High server costs: Existing credit risk processes require the construction of complex systems, such as blacklist systems and behavioral analysis modules. Implementing a complete set of engine rules requires the investment of a series of server resources, and the layers of nested decision-making result in high server costs.

[0005] 3. Serious time consumption: Each system's individual decision-making link, whether connected in series or in parallel, requires time consideration. Relying solely on a stacked application approach cannot ensure an efficient credit decision-making process. If decisions are made according to traditional system processes, the final response time will not reach an acceptable level (for example, several hundred milliseconds), ultimately resulting in a poor user experience.

[0006] 4. Low utilization rate of non-credit data: Traditional risk control relies more on the credit reporting system of the central bank, but does not make much use of unstructured data. Data based on operator DPI is an important supplement to the credit reporting system of the central bank. Summary of the Invention

[0007] The purpose of this application is to provide a method, device and electronic device for judging user credit risk. It is a method for judging user credit risk based on the operator's DPI and parameter server. It can identify user credit risk in a low-cost, low-time and relatively simple framework. It can ensure that the user's credit risk level is portrayed based on the operator's unstructured data outside the scope of credit reporting, and supplement and gain insight into the user's credit risk outside the original credit reporting system.

[0008] In a first aspect, the present application provides a method for judging user credit risk, the method comprising: after obtaining the target user's authorization information for the operator's DPI data, collecting the first network operator information corresponding to the mobile phone number provided by the target user in the application process; based on the first network operator information, obtaining the target user's underlying data from the corresponding operator side; the target user's underlying data includes: user's website access information, user's peak traffic usage period, and user's location trajectory information; performing feature extraction and feature engineering on the target user's underlying data to obtain target feature data; inputting the target feature data into a pre-trained credit risk judgment model, processing the target feature data through the credit risk judgment model, and outputting the credit risk result corresponding to the target user; the risk judgment model is obtained after training in a parameter server deployed with a machine learning model using a training sample set.

[0009] Furthermore, the above-mentioned step of extracting features and performing feature engineering on the underlying data of the target user to obtain target feature data includes: extracting and preprocessing key features of the underlying data of the target user to obtain target key features related to credit; and performing feature engineering on the target key features to obtain target feature data for prediction.

[0010] Furthermore, the training process of the above-mentioned credit risk judgment model is as follows: obtaining a training sample set; the samples in the training sample set include: feature data in the user underlying data of the network operator obtained based on the user authorization operation, and the corresponding user credit rating label; multiple samples are input into the parameter server where the machine learning model is deployed to train the model parameters of the machine learning model; during the training process, the model parameters of the machine learning model are updated through the multiple model parameters stored in the parameter server until the training target reaches the preset value, thereby obtaining a credit risk judgment model; the model parameters include: tree depth, number of leaf nodes, learning rate and L1 regularization term.

[0011] Furthermore, the above-mentioned training goal reaches a preset value including: the final KS or AUC of the model reaches a specified level.

[0012] Furthermore, the above-mentioned step of obtaining a training sample set includes: for each historical user, the following operations are performed: after obtaining the historical user's authorization information for the operator's DPI data, collecting the second network operator information corresponding to the mobile phone number provided by the historical user in the submission link; based on the second network operator information, obtaining the historical user's underlying data from the corresponding operator side; performing feature extraction and feature engineering processing on the historical user's underlying data to obtain historical feature data; labeling the user's credit rating based on the historical feature data to obtain a training sample corresponding to the historical user; and the training samples corresponding to multiple historical users constitute a training sample set.

[0013] Furthermore, the above-mentioned step of processing the target feature data through the credit risk judgment model and outputting the credit risk result corresponding to the target user includes: predicting the target feature data through the credit risk judgment model to obtain the credit risk probability corresponding to the target user; performing logical conversion based on the credit risk probability to obtain the credit risk result represented by the credit risk score.

[0014] Furthermore, the above-mentioned step of performing logical conversion based on the credit risk probability to obtain a credit risk result represented by a credit risk score includes: determining whether the credit risk probability is greater than a preset threshold; if so, setting the credit risk score to 0 points to represent a low-risk user; if not, setting the credit risk score to 100 points to represent a high-risk user.

[0015] On the second aspect, the present application also provides a device for judging user credit risk, which includes: an information acquisition module, which is used to collect the first network operator information corresponding to the mobile phone number provided by the target user in the application process after obtaining the target user's authorization information for the operator's DPI data; a data acquisition module, which is used to obtain the target user's underlying data from the corresponding operator side based on the first network operator information; the target user's underlying data includes: user's website access information, user's peak traffic usage period, and user's location trajectory information; a feature extraction module, which is used to perform feature extraction and feature engineering on the target user's underlying data to obtain target feature data; a model prediction module, which is used to input the target feature data into a pre-trained credit risk judgment model, process the target feature data through the credit risk judgment model, and output the credit risk result corresponding to the target user; the risk judgment model is obtained after training with a training sample set in a parameter server deployed with a machine learning model.

[0016] In a third aspect, the present application further provides an electronic device comprising a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the method described in the first aspect above.

[0017] In a fourth aspect, the present application also provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the method described in the first aspect above.

[0018] The present application provides a method, device, and electronic device for determining user credit risk. First, after obtaining authorization information for operator DPI data from the target user, the first network operator information corresponding to the mobile phone number provided by the target user during the application submission process is collected. Then, based on the first network operator information, the target user's underlying data is obtained from the corresponding operator. The target user's underlying data includes information about the websites visited by the user, peak traffic usage periods, and location trajectory information. Feature extraction and feature engineering are performed on the target user's underlying data to obtain target feature data. Finally, the target feature data is input into a pre-trained credit risk determination model, which processes the target feature data and outputs a credit risk result corresponding to the target user. The credit risk determination model is trained using a training sample set on a parameter server deployed with a machine learning model. The present application provides a method for determining user credit risk based on operator DPI and a parameter server. This method can identify user credit risk within a low-cost, low-time, and relatively simple framework. It can ensure that the user's credit risk level is characterized based on the operator's unstructured data, beyond the scope of credit reporting, and supplement and provide insight into the user's credit risk beyond the existing credit reporting system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the specific implementation methods of the present application or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the specific implementation methods or the description of the prior art. Obviously, the drawings described below are some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0020] Figure 1 A flowchart of a method for determining user credit risk provided in an embodiment of the present application;

[0021] Figure 2 A flowchart of a model training process provided in an embodiment of the present application;

[0022] Figure 3 A schematic diagram of a framework of a model training process provided in an embodiment of the present application;

[0023] Figure 4 A structural block diagram of a device for determining user credit risk provided in an embodiment of the present application;

[0024] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions of this application in conjunction with the embodiments. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0026] In the credit field, common methods for judging user credit risk include:

[0027] 1. Credit scoring model: A credit score is generated using data such as the user's credit history, repayment record, and asset status. The lower the score, the higher the credit risk.

[0028] 2. Machine Learning Model: Use algorithms such as logistic regression, random forest, gradient boosted tree (GBM), neural network, etc. to train models based on historical data to predict the probability of default.

[0029] 3. Behavioral analysis: Analyze user behavior patterns, such as unusual shopping preferences and large expenditures. Abnormal behavior may indicate credit risk.

[0030] 4. Social network analysis: Identify intermediary groups or associated risks by analyzing users’ social relationships.

[0031] 5. Rule engine: Set rules (such as applying for loans multiple times in a short period of time), and users who trigger the rules are considered high-risk.

[0032] 6. Blacklist system: put known risk users on the blacklist and reject their loan applications.

[0033] 7. Real-time monitoring and early warning: Monitor transactions and behaviors in real time and issue early warnings when anomalies are detected.

[0034] 8. Multi-dimensional data verification: Verify user information through third-party data (such as credit reporting agencies, operators, etc.), and indicate risks when there are inconsistencies.

[0035] 9. Manual review: Manual review of high-risk users to further confirm the possibility of credit risk.

[0036] 10. Credit risk pattern identification: By analyzing historical overdue cases, common patterns are identified and applied to the risk assessment of new users.

[0037] The above methods are usually used selectively when identifying credit risks, which can easily lead to the following three defects: The following four defects: 1. Low automation rate: In traditional credit risk control links, especially in consumer finance companies or banks, most processes require face-to-face interviews. The addition of these links requires more manual review, which does not conform to the existing credit 310 process (i.e., 3 minutes to fill out the form, 1 second to review, and 0 manual intervention); 2. High server costs: The existing credit risk process requires the construction of a complex system, such as a blacklist system and a behavioral analysis module. If a complete set of engine rules is to be implemented, a series of server costs must be invested. Server resources, nested layers will make this part of the decision-making require higher server costs; 3. The time consumption problem is serious: the individual decision-making links of each system, whether in series or parallel, need to consider the time consumption problem. Relying solely on the stacking application method cannot meet the requirements of an efficient credit decision-making process. If decisions are made according to the traditional system process, the final response time cannot reach a reasonable level (such as a few hundred milliseconds), which ultimately results in a poor user experience; 4. The utilization rate of non-credit data is low: Traditional risk control relies more on the central bank's credit system, but does not make much use of unstructured data. Data based on operator DPI is an important supplement to the central bank's credit system.

[0038] Based on this, the embodiments of the present application provide a method, device and electronic device for judging user credit risk. It is a method for judging user credit risk based on the operator's DPI and parameter server. It can identify user credit risk in a low-cost, low-time and relatively simple framework. It can ensure that the user's credit risk level is portrayed based on the operator's unstructured data outside the scope of credit reporting, and supplement and gain insight into the user's credit risk outside the original credit reporting system.

[0039] To facilitate understanding of this embodiment, a method for determining user credit risk disclosed in an embodiment of the present application is first introduced in detail.

[0040] Figure 1 This is a flowchart of a method for determining user credit risk provided in an embodiment of the present application, which specifically includes the following steps:

[0041] Step S102: After obtaining the target user's authorization information for the operator's DPI data, collect the first network operator information corresponding to the mobile phone number provided by the target user during the application submission process;

[0042] Carrier DPI data: This refers to refined data obtained through in-depth analysis of network traffic using Deep Packet Inspection (DPI) technology. This data not only contains basic user network behavior information, but also identifies application layer protocols, content types, and user behavior characteristics, providing carriers with multi-dimensional network management and service optimization support.

[0043] The first network operator information includes at least one of the three major operators.

[0044] Step S104: Based on the first network operator information, obtain the target user's underlying data from the corresponding operator. The target user's underlying data includes: information about websites visited by the user, peak traffic usage time periods, and location information of the user.

[0045] For example, the underlying data of target users include the websites visited by users (such as browsing China Merchants Bank Consumer Finance, Alipay, China Merchants Bank, etc.), the time periods when users use peak traffic (for example, the time periods when users browse are mainly between 9:30-11:30 in the morning and between 0:00 and 4:00 in the morning, etc.), and the user's location trajectory (the location of the user signal is in Shanghai, Zhejiang Province, etc.).

[0046] Step S106: performing feature extraction and feature engineering on the underlying data of the target user to obtain target feature data;

[0047] This step can be specifically achieved in the following ways: extracting and preprocessing the underlying data of the target user to obtain target key features related to credit; performing feature engineering on the target key features to obtain target feature data for prediction.

[0048] The feature extraction in this step is actually the extraction of key features, which mainly aims to gain insights into features related to the user's credit preferences. For example, around the theme of long positions, there is the number of loan apps browsed by users, and factors related to user risk preferences, such as whether the user's traffic usage is during high-risk periods such as early morning. In addition to obtaining key features related to credit risk from the original underlying data, this step can also complete data preprocessing. For example, from the URLs visited by users, only loan-related ones are filtered out, and URLs that are not related to credit risk are identified and eliminated.

[0049] Feature engineering processing, for example, deriving credit-related features from variables such as users' visits to lending websites, such as whether the user spends more than 2 hours on the lending website, the number of lending apps the user has browsed in the past week, and the number of days the user visited the lending app in the early morning in the past 30 days.

[0050] Step S108: Input the target feature data into a pre-trained credit risk judgment model, process the target feature data through the credit risk judgment model, and output the credit risk result corresponding to the target user; the credit risk judgment model is obtained by training the training sample set in a parameter server deployed with a machine learning model.

[0051] A parameter server is a programming framework primarily used to support the distributed storage and coordination of large-scale parameters. In machine learning, model training often generates a large number of model parameters, which need to be stored and shared to facilitate parallel computation and optimization across multiple computing nodes. The parameter server is used to manage and distribute these model parameters. In this embodiment, the parameter server has two main functions: first, it is a server that can perform model training; second, the parameter server can store a large number of model parameters, providing parameter traversal for each model training to optimize the training score.

[0052] Machine learning models can include either the XGB model or the LGB model. The training sample set includes characteristic data from the network operator's underlying user data, obtained based on user authorization, along with the corresponding user credit rating label. After training on a parameter server hosting the machine learning model using the samples in this training sample set, a credit risk assessment model can be generated that meets pre-defined training objectives. For example, the model's final KS or AUC reaches a specified threshold.

[0053] The above-mentioned step of processing the target feature data through the credit risk judgment model and outputting the credit risk result corresponding to the target user may include: predicting the target feature data through the credit risk judgment model to obtain the credit risk probability corresponding to the target user; performing logical conversion based on the credit risk probability to obtain the credit risk result represented by the credit risk score.

[0054] Furthermore, the above-mentioned step of performing logical conversion based on the credit risk probability to obtain a credit risk result represented by a credit risk score includes: determining whether the credit risk probability is greater than a preset threshold; if so, setting the credit risk score to 0 points to represent a low-risk user; if not, setting the credit risk score to 100 points to represent a high-risk user.

[0055] The method for determining user credit risk provided in the embodiment of the present application is a method for determining user credit risk based on the operator's DPI and parameter server. It can identify user credit risk in a low-cost, low-time, and relatively simple framework. It can ensure that the user's credit risk level is portrayed based on the operator's unstructured data outside the scope of credit reporting, supplementing and providing insight into the user's credit risk outside the original credit reporting system.

[0056] The following is a detailed explanation of the training process of the above credit risk judgment model:

[0057] See also Figure 2 As shown in Figure 2, the training process of the above credit risk judgment model is as follows:

[0058] Step S202: obtaining a training sample set; the samples in the training sample set include: feature data in the user underlying data of the network operator obtained based on the user authorization operation, and the corresponding user credit rating label;

[0059] During specific implementation, the following operations are performed for each historical user:

[0060] After obtaining the historical user's authorization information for the operator's DPI data, collect the second network operator information corresponding to the mobile phone number provided by the historical user in the submission process; based on the second network operator information, obtain the historical user's underlying data from the corresponding operator side; perform feature extraction and feature engineering on the historical user's underlying data to obtain historical feature data; annotate the user's credit rating label based on the historical feature data to obtain the training sample corresponding to the historical user; and the training samples corresponding to multiple historical users constitute a training sample set.

[0061] The process of obtaining the training data set is similar to the process for processing the target user described above, and will not be repeated here.

[0062] In step S204, multiple samples are input into a parameter server hosting a machine learning model to train the model parameters of the machine learning model. During the training process, the model parameters of the machine learning model are updated using the various model parameters stored in the parameter server until the training target reaches a preset value, thereby obtaining a credit risk judgment model. The model parameters include: tree depth, number of leaf nodes, learning rate, and L1 regularization term. Achieving the preset training target includes: the model's final KS or AUC reaching a specified level.

[0063] In this embodiment, the parameter server includes a server, a client, and a scheduler. The server is used to store model parameters, receive gradients uploaded by the client, and update local parameters. It is responsible for storing and updating parameters to ensure the consistency of model parameters. The client is used to obtain the latest parameters from the server, calculate gradients using local data, and upload gradients to the server. It is responsible for calculating gradients and synchronizing model parameters. The scheduler is used to manage server and client nodes, and complete functions such as data synchronization between nodes and adding / deleting nodes.

[0064] See also Figure 3The architecture diagram of the model training process shown in the figure specifically includes: data authorization module, data acquisition layer, input layer, preprocessing module, feature engineering, parameter server import module, parameter server application module and decision module. The following is a detailed description of each module:

[0065] The first part is the data authorization module. In the credit application process, the first step is to obtain the user's authorization for the operator's DPI data. Usually, in the application process, the user needs to allow the financial institution to obtain the viewing permission of the operator information corresponding to the provided mobile phone number, and assist it in risk judgment based on the DPI-related data. It is clearly stated on the operation interface that the mobile phone number is allowed to obtain relevant information. Only with the user's authorization can the credit application process be carried out, otherwise the process ends.

[0066] In financial services, "submission" refers to the process of submitting a client's application materials to a financial institution for review. Specifically, this involves submitting a client's loan application, fund application, or other financial service application materials to the relevant approval department for review. This information typically includes the client's personal information, financial status, credit history, and other information to facilitate the financial institution's assessment and approval of the application.

[0067] The second part is the data collection layer. Since the current information acquisition is mainly based on the three major operators rather than other data sources, this module needs to collect the relevant data of the network operator corresponding to the mobile phone number provided by the user;

[0068] The third part is the input layer. This framework mainly obtains the user's underlying data from the operator. Due to operator control, this data is currently only retained on the operator's side. However, financial institutions can select important dimensions, such as the websites visited by users (such as browsing CMB Consumer Finance, Alipay, and China Merchants Bank), the user's peak traffic usage time (for example, the user's browsing time is mainly between 9:30-11:30 in the morning and 12:00-4:00 in the morning), and the user's location trajectory (the user signal is located in Shanghai and Zhejiang Province) for subsequent processing and application.

[0069] The fourth part is the preprocessing module, which extracts key features for the feature engineering module in the fifth part. These key features mainly provide insights into users' credit preferences, such as the number of loan apps viewed by users, and factors related to their risk preferences, such as whether users use data during high-risk periods such as early morning. Since the input layer of the third part inputs raw data, the fourth part mainly extracts key features related to credit risk from the raw data and completes data preprocessing. For example, it filters out loan-related websites from the websites visited by users and eliminates websites that are not related to credit risk.

[0070] The fifth part is the feature engineering module, which performs feature engineering on the variables preprocessed in the fourth part. This module processes them into variables that can be subsequently fed into the model. For example, from variables such as a user's visits to loan websites, it can derive credit-related features such as whether the user spends more than two hours on a loan website, the number of loan apps the user has browsed in the past week, and the number of days the user visited loan apps in the early morning in the past 30 days.

[0071] The sixth part is to import the module based on the parameter server, which is divided into two parts:

[0072] 1. This module imports the variables after feature engineering into the parameter server. The parameter server has two main functions. First, it is a server that can perform model training. Second, the parameter server can store a large number of model parameters for each training model to provide parameter traversal to optimize the training score.

[0073] 2. Set and input the user's credit risk y(mob3_30+), which is the training target (generally speaking, the goal of each training is to make the model's final KS or AUC reach a certain level. For example, for small loan customers, the KS should be as high as possible to exceed 0.2). This is equivalent to inputting a series of samples with y into the server and training the model based on the credit variables obtained in the fifth part.

[0074] The seventh part is the parameter server application module, which stores model training parameters, including tree depth (depth), number of leaf nodes (leaves), learning rate (learning_rate), L1 regularization term (rambda_l1), etc. Through the user y introduced in the sixth part, combined with the above parameters, the user's credit risk score is obtained. For example, this may include various tree depths (depth = 2, 3, 4...), number of leaf nodes (leaves = 4, 8, 16...), and learning rates (learning_rate = 0.1, 0.2, 0.3...). By polling these parameters, an optimal parameter combination is obtained, and then the user's risk probability prob is trained using machine learning.

[0075] The eighth part is the decision module. This module uses a logistic transformation to convert the risk probability (prob) generated in the seventh part into a score using the logistic transformation formula: log odds = log(1-prob / prob). This transformation formula follows the standard logistic transformation rule. Because the prob trained in the seventh part is only the probability of default, it needs to be converted into a relatively interpretable value on a scale of 0-100.

[0076] Linearly map log odds to a scale of 0-100: The most basic approach is to directly assign a score of 0 to high-risk users (prob>0.8) and 100 to low-risk users (prob<0.2). Consumer finance companies can directly reject high-risk users, while allowing low-risk users to go through the normal credit application process.

[0077] The above modules form an organic whole, providing a highly automated device with a clear framework and complementary gains to the existing system.

[0078] The method for determining user credit risk provided in the embodiment of the present application is a high-performance, well-structured, and user-friendly solution that can identify user credit risk in a low-cost, low-time, and relatively simple framework without the need for additional manual intervention, and can quickly complete credit risk classification for users and make the final decision.

[0079] Based on the above method embodiment, the present application embodiment also provides a device for judging user credit risk, see Figure 4 As shown, the device includes: an information acquisition module 42, which is used to collect the first network operator information corresponding to the mobile phone number provided by the target user in the application process after obtaining the target user's authorization information for the operator's DPI data; a data acquisition module 44, which is used to obtain the target user's underlying data from the corresponding operator side based on the first network operator information; the target user's underlying data includes: user's website access information, user's traffic peak period, and user's location trajectory information; a feature extraction module 46, which is used to extract features and perform feature engineering on the target user's underlying data to obtain target feature data; a model prediction module 48, which is used to input the target feature data into a pre-trained credit risk judgment model, process the target feature data through the credit risk judgment model, and output the credit risk result corresponding to the target user; the risk judgment model is obtained after training in a parameter server deployed with a machine learning model using a training sample set.

[0080] Furthermore, the feature extraction module 46 is used to extract and preprocess the target user's underlying data to obtain target key features related to credit; and perform feature engineering on the target key features to obtain target feature data for prediction.

[0081] Furthermore, the above-mentioned device also includes: a module training module, which is used to execute the following training process of the credit risk judgment model: obtaining a training sample set; the samples in the training sample set include: feature data in the user underlying data of the network operator obtained based on the user authorization operation, and the corresponding user credit rating label; inputting multiple samples into a parameter server deployed with a machine learning model to train the model parameters of the machine learning model; during the training process, updating the model parameters of the machine learning model through multiple model parameters stored in the parameter server until the training target reaches the preset value, thereby obtaining a credit risk judgment model; the model parameters include: tree depth, number of leaf nodes, learning rate and L1 regularization term.

[0082] Furthermore, the above-mentioned training goal reaches a preset value including: the final KS or AUC of the model reaches a specified level.

[0083] Furthermore, the above-mentioned module training module is used to: perform the following operations for each historical user: after obtaining the historical user's authorization information for the operator's DPI data, collect the second network operator information corresponding to the mobile phone number provided by the historical user in the submission link; based on the second network operator information, obtain the historical user's underlying data from the corresponding operator side; perform feature extraction and feature engineering on the historical user's underlying data to obtain historical feature data; perform user credit rating labeling based on the historical feature data to obtain training samples corresponding to the historical user; and constitute a training sample set by the training samples corresponding to multiple historical users.

[0084] Furthermore, the above-mentioned model prediction module 48 is used to predict the target feature data through the credit risk judgment model to obtain the credit risk probability corresponding to the target user; and perform logical conversion based on the credit risk probability to obtain a credit risk result represented by a credit risk score.

[0085] Furthermore, the above-mentioned model prediction module 48 is used to determine whether the credit risk probability is greater than a preset threshold; if so, the credit risk score is set to 0 points to characterize a low-risk user; if not, the credit risk score is set to 100 points to characterize a high-risk user.

[0086] The device provided in the embodiment of the present application has the same implementation principle and technical effects as those in the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the embodiment of the device, reference can be made to the corresponding content in the aforementioned method embodiment.

[0087] The present application also provides an electronic device, such as Figure 5As shown, it is a structural diagram of the electronic device, wherein the electronic device includes a processor 51 and a memory 50, the memory 50 stores computer executable instructions that can be executed by the processor 51, and the processor 51 executes the computer executable instructions to implement the above method.

[0088] exist Figure 5 In the illustrated embodiment, the electronic device further includes a bus 52 and a communication interface 53 , wherein the processor 51 , the communication interface 53 and the memory 50 are connected via the bus 52 .

[0089] Among them, the memory 50 may include a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 53 (which can be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. can be used. The bus 52 can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 52 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0090] The processor 51 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by hardware integrated logic circuits in the processor 51 or by software instructions. The above processor 51 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of the present application can be directly implemented as being executed by a hardware decoding processor, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor 51 reads the information in the memory and completes the steps of the method of the above embodiment in combination with its hardware.

[0091] An embodiment of the present application also provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are called and executed by the processor, the computer-executable instructions prompt the processor to implement the above-mentioned method. The specific implementation can be found in the above-mentioned method embodiment, which will not be repeated here.

[0092] The computer program products of the methods, devices, and electronic devices provided in the embodiments of the present application include a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the previous method embodiments. For specific implementation, please refer to the method embodiments and will not be repeated here.

[0093] Unless otherwise specifically stated, the relative steps, numerical expressions and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present application.

[0094] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0095] In the description of this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate the description of this application and simplify the description. They do not indicate or imply that the devices or components referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on this application. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0096] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The scope of protection of the present application is not limited thereto. Although the present application has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed in the present application, or perform equivalent replacements for some of the technical features thereof. These modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application shall be subject to the scope of protection of the claims.

Claims

1. A method for determining user credit risk, characterized in that: The method comprises: After obtaining the target user's authorization information for the operator's DPI data, collect the first network operator information corresponding to the mobile phone number provided by the target user during the application process; Based on the first network operator information, obtain the target user's underlying data from the corresponding operator; the target user's underlying data includes: user's visited website information, user's peak traffic usage period, and user's location trajectory information; Perform feature extraction and feature engineering on the underlying data of the target user to obtain target feature data; The target feature data is input into a pre-trained credit risk judgment model, the target feature data is processed by the credit risk judgment model, and the credit risk result corresponding to the target user is output; the credit risk judgment model is obtained after training with a training sample set in a parameter server deployed with a machine learning model.

2. The method according to claim 1, characterized in that The step of performing feature extraction and feature engineering on the underlying data of the target user to obtain target feature data includes: Extracting and preprocessing the underlying data of the target user to obtain target key features related to credit; The target key features are subjected to feature engineering processing to obtain target feature data for prediction.

3. The method according to claim 1, characterized in that The training process of the credit risk judgment model is as follows: Acquire a training sample set; the samples in the training sample set include: feature data in the user underlying data of the network operator obtained based on the user authorization operation, and the corresponding user credit rating label; Multiple samples are input into a parameter server deployed with a machine learning model to train the model parameters of the machine learning model; during the training process, the model parameters of the machine learning model are updated using the multiple model parameters stored in the parameter server until the training target reaches a preset value, thereby obtaining a credit risk judgment model; the model parameters include: tree depth, number of leaf nodes, learning rate and L1 regularization term.

4. The method according to claim 3, characterized in that The training goal reaches the preset value, which includes: the final KS or AUC of the model reaches a specified level.

5. The method according to claim 1, wherein The steps to obtain the training sample set include: For each historical user, perform the following operations: After obtaining the authorization information of the historical user for the operator DPI data, collect the second network operator information corresponding to the mobile phone number provided by the historical user during the application process; Based on the second network operator information, obtain historical user underlying data from the corresponding operator side; Performing feature extraction and feature engineering on the historical user underlying data to obtain historical feature data; Labeling user credit ratings based on the historical feature data to obtain training samples corresponding to the historical users; The training sample set is composed of training samples corresponding to multiple historical users.

6. The method according to claim 1, characterized in that The step of processing the target feature data by using the credit risk judgment model and outputting the credit risk result corresponding to the target user includes: Predicting the target feature data using the credit risk judgment model to obtain the credit risk probability corresponding to the target user; A logical transformation is performed based on the credit risk probability to obtain a credit risk result represented by a credit risk score.

7. The method according to claim 6, characterized in that The step of performing a logical transformation based on the credit risk probability to obtain a credit risk result represented by a credit risk score includes: Determining whether the credit risk probability is greater than a preset threshold; If yes, the credit risk score is set to 0 to indicate a low-risk user; If not, the credit risk score is set to 100 points to characterize a high-risk user.

8. A device for determining user credit risk, characterized in that: The device comprises: An information acquisition module is configured to, after obtaining authorization information of the target user for the operator DPI data, collect the first network operator information corresponding to the mobile phone number provided by the target user during the application submission process; A data acquisition module is configured to acquire underlying data of a target user from a corresponding operator based on the first network operator information; the underlying data of the target user includes: information on websites visited by the user, peak periods of user traffic usage, and location trajectory information of the user; A feature extraction module is used to perform feature extraction and feature engineering on the underlying data of the target user to obtain target feature data; The model prediction module is used to input the target feature data into a pre-trained credit risk judgment model, process the target feature data through the credit risk judgment model, and output the credit risk result corresponding to the target user; the credit risk judgment model is obtained after training with a training sample set in a parameter server deployed with a machine learning model.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the method according to 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. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the method according to any one of claims 1 to 7.

Images