Credit risk assessment method, system, device, storage medium and program product

Abstract

The disclosure provides a credit risk assessment method, system, device, storage medium and program product, relating to the technical field of data processing. The method comprises: in response to a credit service request of a user to be evaluated, a financial client acquires financial behavior data of the user to be evaluated; the financial client processes the financial behavior data of the user to be evaluated through a risk assessment model to obtain a credit risk assessment result of the user to be evaluated; the risk assessment model is obtained through federated learning training of historical financial behavior data and corresponding evaluation result labels of historical users by the financial client, historical spatio-temporal behavior data of historical users by an operator client, and historical consumption behavior data of historical users by an Internet client, and the disclosure can improve the credit risk assessment accuracy.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to a credit risk assessment method, a credit risk assessment system, an electronic device, a computer-readable storage medium, and a computer program product. Background Technology

[0002] With the rapid development of financial technology, credit risk control has become an important task for banks and other financial institutions.

[0003] In related technologies, analysis and prediction are typically based on data from the People's Bank of China and credit data accumulated by financial institutions. However, the aforementioned credit risk control methods suffer from insufficient coverage of the population covered by proprietary credit data, reducing the accuracy and effectiveness of credit risk assessment.

[0004] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] This disclosure provides a credit risk assessment method, system, device, storage medium, and program product, which at least to some extent overcomes the problems of low accuracy and poor effectiveness of credit risk assessment results in related technologies.

[0006] Other features and advantages of this disclosure will become apparent from the following detailed description, or may be learned in part from practice of this disclosure.

[0007] According to one aspect of this disclosure, a credit risk assessment method is provided, wherein in response to a credit business request from a user to be assessed, a financial client obtains the financial behavior data of the user to be assessed;

[0008] The financial client processes the financial behavior data of the user to be assessed through a risk assessment model to obtain the credit risk assessment result of the user to be assessed.

[0009] The risk assessment model is obtained by federated learning of the historical financial behavior data of the historical user and the corresponding first assessment result label of the financial client, the historical spatiotemporal behavior data of the historical user and the corresponding second assessment result label of the operator client, and the historical consumption behavior data of the historical user and the corresponding third assessment result label of the internet client.

[0010] In one embodiment of this disclosure, the risk assessment model is trained in the following manner:

[0011] The financial client obtains the first gradient of the initial risk assessment model based on the historical financial behavior data of the historical user, the corresponding assessment result tags, and the model parameters of the financial client. It then interacts with the operator client through homomorphic encryption to obtain the second gradient, and interacts with the internet client through homomorphic encryption to obtain the third gradient. The first gradient, the second gradient, and the third gradient are aggregated to obtain the total gradient of the financial client.

[0012] The operator client interacts with the financial client and the internet client through homomorphic encryption to obtain the total gradient of the operator client;

[0013] The internet client interacts with the financial client and the operator client through homomorphic encryption to obtain the total gradient of the internet client;

[0014] The model parameters are iteratively updated based on gradient descent until the preset conditions are met, thus obtaining the risk assessment model.

[0015] The second gradient is determined by the historical spatiotemporal behavior data of the historical user, the model parameters of the operator's client, and the historical financial behavior data; the third gradient is determined by the historical consumption behavior data of the historical user, the model parameters of the internet client, and the historical financial behavior data.

[0016] In one embodiment of this disclosure, the financial client trains an initial risk assessment model based on the historical financial behavior data of the historical user and the corresponding assessment result labels to obtain the first gradient of the initial risk assessment model, including:

[0017] The financial client extracts features from the historical financial behavior data to obtain a first feature vector.

[0018] The financial client processes the first feature vector using the initial risk assessment model to obtain a first assessment result;

[0019] The financial client performs local gradient updates based on the first evaluation result and the first evaluation result label to determine the first gradient.

[0020] In one embodiment of this disclosure, the financial client interacts with the operator client via homomorphic encryption to obtain a second gradient, including:

[0021] The operator client extracts spatial and temporal features from the historical spatiotemporal behavior data to obtain a second feature vector. Based on the model parameters of the operator client and the second feature vector, a second evaluation result is obtained. The second evaluation result is homomorphically encrypted to obtain the ciphertext of the second evaluation result and sent to the financial client.

[0022] The financial client processes the ciphertext of the second evaluation result based on the first feature vector and the first random number to obtain the ciphertext of the second gradient information.

[0023] The operator client receives the ciphertext of the second gradient information, decrypts the ciphertext of the second gradient information, and sends the decrypted second gradient information.

[0024] The financial client receives the decrypted second gradient information and obtains the second gradient based on the first random number and the decrypted second gradient information.

[0025] In one embodiment of this disclosure, the financial client interacts with the internet client via homomorphic encryption to obtain a third gradient, including:

[0026] The internet client extracts features from the historical consumption behavior data to obtain a third feature vector. Based on the model parameters of the internet client and the third feature vector, a third evaluation result is obtained. The third evaluation result is homomorphically encrypted to obtain the ciphertext of the third evaluation result and sent to the financial client.

[0027] The financial client processes the ciphertext of the third evaluation result based on the first feature vector and the second random number to obtain the ciphertext of the third gradient information.

[0028] The operator client receives the ciphertext of the third information and decrypts the ciphertext of the second gradient information to obtain the decrypted third gradient information.

[0029] The financial client receives the decrypted third gradient information and, based on the second random number and the decrypted third gradient information, determines how many third gradients there are.

[0030] In one embodiment of this disclosure, the step of iteratively updating the model parameters based on gradient descent until a preset condition is met to obtain the risk assessment model includes:

[0031] If the total gradient reaches the minimum, or the number of iterations reaches a preset threshold, or the iteration duration reaches a preset duration threshold, then the preset conditions are satisfied, and the risk assessment model is constructed based on the model parameters that satisfy the preset conditions.

[0032] In one embodiment of this disclosure, before the financial client processes the financial behavior data of the user to be assessed through a risk assessment model to obtain the credit risk assessment result of the user to be assessed, the method further includes: the financial client determines the risk assessment model according to the scenario type of the credit business request, with one risk assessment model corresponding to one scenario type.

[0033] In one embodiment of this disclosure, the method further includes:

[0034] Based on multiple application scenarios, we construct learning tasks corresponding to meta-learning, with one learning task corresponding to one application scenario.

[0035] The learning tasks are processed by the multi-task risk assessment model to be trained, and the model gradients of each learning task are obtained.

[0036] Gradient aggregation is performed on the model gradients of each learning task to minimize the gradient of the multi-task risk assessment model. Iterative training is then performed to obtain the multi-task assessment model.

[0037] According to another aspect of this disclosure, a credit risk assessment system is also provided, including a financial client, a telecom operator client, and an internet client, wherein: the financial client is used to respond to a credit business request from a user to be assessed and obtain the financial behavior data of the user to be assessed; the financial behavior data of the user to be assessed is processed by a risk assessment model to obtain a credit risk assessment result for the user to be assessed; wherein the risk assessment model is obtained by federated learning training of historical financial behavior data of historical users and corresponding first assessment result labels by the financial client, historical spatiotemporal behavior data of historical users and corresponding second assessment result labels by the telecom operator client, and historical consumption behavior data of historical users and corresponding third assessment result labels by the internet client.

[0038] According to another aspect of this disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the credit risk assessment method described above by executing the executable instructions.

[0039] According to another aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the credit risk assessment method described above.

[0040] According to another aspect of this disclosure, a computer program product is provided, the computer program product comprising a computer program or computer instructions, the computer program or computer instructions being loaded and executed by a processor to enable a computer to implement the credit risk assessment method described above.

[0041] In this embodiment, in response to a credit business request from a user to be assessed, the financial client obtains the user's financial behavior data. The financial client processes the user's financial behavior data using a risk assessment model to obtain the user's credit risk assessment result. The risk assessment model is trained through federated learning using historical financial behavior data and corresponding assessment result labels from the financial client, historical spatiotemporal behavior data from the operator client, and historical consumption behavior data from the internet client. This disclosure combines user's financial behavior data, spatiotemporal behavior data, and consumption behavior data to extract user risk characteristics from multi-source data, trains the risk assessment model, and achieves accurate identification and early warning of user credit risk through the risk assessment model, thereby improving credit.

[0042] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0043] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0044] Figure 1 A flowchart of a credit risk assessment method provided by an embodiment of this disclosure is shown.

[0045] Figure 2 This diagram illustrates a flowchart of a risk assessment model construction method provided in an embodiment of the present disclosure.

[0046] Figure 3 The diagram illustrates the principle of a risk assessment model construction method provided in an embodiment of this disclosure.

[0047] Figure 4 This diagram illustrates another risk assessment model construction method provided by an embodiment of the present disclosure.

[0048] Figure 5 A flowchart of another credit risk assessment method provided by an embodiment of this disclosure is shown.

[0049] Figure 6This diagram illustrates the principle of training a credit risk assessment model according to an embodiment of the present disclosure.

[0050] Figure 7 This diagram illustrates a credit risk assessment system provided in an embodiment of the present disclosure.

[0051] Figure 8 A structural block diagram of an electronic device provided in an embodiment of this disclosure is shown. Detailed Implementation

[0052] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that this disclosure will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0053] Furthermore, the accompanying drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor systems and / or microcontroller systems.

[0054] To enable those skilled in the art to better understand the technical solutions of this disclosure, the following detailed description is provided in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure.

[0055] In related technologies, analysis and prediction are typically based on data from the People's Bank of China (PBOC) and financial behavior data accumulated by financial institutions. PBOC data refers to various financial data collected and managed by banks, including money supply, financial behavior data, and payment system data. However, because the PBOC data and the financial behavior data accumulated by financial institutions do not cover the full population, effective risk assessment cannot be conducted on the uncovered populations, resulting in low accuracy and poor effectiveness in credit risk assessment.

[0056] Financial institutions typically combine accumulated financial behavior data with data from the People's Bank of China, and use statistical or machine learning methods to build risk assessment models to improve credit risk control.

[0057] Descriptive statistical analysis based on data statistics methods can be performed on static structured data, such as calculating the mean, standard deviation, and correlation coefficient, to understand the user's basic static characteristics. However, this approach typically only provides simple statistical indicators and cannot delve into user behavior patterns and regularities, resulting in low accuracy of the evaluation results.

[0058] Machine learning-based risk assessment methods can train risk assessment models based on historical data to predict a user's credit status. However, this method cannot address the issue of data that cannot be aggregated due to privacy and regulatory constraints, thus limiting its application scope.

[0059] In related technologies, credit risk control feature mining only uses credit data from financial institutions, which is a single perspective and cannot take into account the features of users in other scenarios that can measure creditworthiness. Therefore, it cannot accurately identify the user's credit risk level.

[0060] The method of evaluating user credit risk using consumption data and spatiotemporal behavioral data has obvious time-series characteristics and spatiotemporal behavioral data has spatial attributes. The related technologies simply perform frequency statistics on the out-of-control information of unilateral data, which cannot extract the spatiotemporal characteristics of multi-party data, resulting in insufficient utilization of data information.

[0061] To at least partially solve the aforementioned technical problems, this disclosure responds to a credit business request from a user to be evaluated by having a financial client obtain the user's financial behavior data; the financial client processes the user's financial behavior data using a risk assessment model to obtain the user's credit risk assessment result; the risk assessment model is trained through federated learning using historical financial behavior data and corresponding assessment result labels from the financial client, historical spatiotemporal behavior data from the operator client, and historical consumption behavior data from the internet client. This disclosure combines user's financial behavior data, spatiotemporal behavior data, and consumption behavior data to extract user risk characteristics from multi-source data, trains the risk assessment model, and achieves accurate identification and early warning of user credit risk through the risk assessment model, thereby improving credit.

[0062] It should be noted that in the technical solutions disclosed herein, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to the aforementioned financial behavior data, spatiotemporal behavior data, and consumer behavior data) involved 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 respect relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0063] It should be noted that, unless otherwise specified, the embodiments of the present invention and the technical features thereof can be combined with each other.

[0064] The following detailed description of this exemplary implementation method is provided in conjunction with the accompanying drawings and embodiments.

[0065] First, this disclosure provides a credit risk assessment method, which can be executed by any electronic device with computing capabilities. The credit risk assessment method provided in this disclosure can be executed by a financial client or by an aggregation server. In one embodiment, the aggregation server, also known as a central node, can be a server providing various services, such as a back-end management server supporting the system operations performed by users using terminal devices. The back-end management server can analyze and process received data such as credit business requests and feed the processing results back to the terminal device.

[0066] An aggregation server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks, and big data and artificial intelligence platforms.

[0067] In one embodiment, the financial client can also act as an aggregation server to perform gradient aggregation and build a risk assessment model.

[0068] Figure 1 This document illustrates a flowchart of a credit risk assessment method provided in an embodiment of the present disclosure, as shown below. Figure 1 As shown in the embodiments of this disclosure, the credit risk assessment method includes the following steps:

[0069] S102. In response to a credit business request from a user to be evaluated, the financial client obtains the financial behavior data of the user to be evaluated.

[0070] In one embodiment, a financial client is an application software that provides financial services to users. The financial client can be integrated into the smartphones, tablets, and other mobile devices of financial institutions, enabling users to access various online financial services.

[0071] The user to be assessed is the user for whom credit risk assessment is to be conducted. The user to be assessed can send credit requests to the financial client through the financial application installed on the user's mobile device.

[0072] Depending on the application scenario of the user's requested business, the types of credit business requests may include consumer loan credit requests, cash loan income requests, etc., and this disclosure does not make specific limitations on them.

[0073] The financial behavior data of the user to be evaluated can include the user's daily borrowing data, online activity data at different times of the day, and credit data such as whether there are any overdue loans. Daily borrowing data can include loan amount, loan term, loan interest rate, loan approval date, and loan type; online activity data at different times of the day includes login, browsing, application, and withdrawal behaviors. Whether there are any overdue loans includes the number of historical on-time repayment periods and the number of historical overdue repayment periods.

[0074] A user identifier is used to uniquely identify each user to be evaluated, distinguishing them from other users. The user identifier can be a user's ID number, mobile phone number, etc., with no restrictions.

[0075] It should be noted that the user identifier of the user to be evaluated can also be encrypted identity information. For example, the string obtained by processing the user's name and ID number using a hash algorithm can be used as the user identifier, thereby avoiding information leakage and improving security.

[0076] S104. The financial client processes the financial behavior data of the user to be assessed through the risk assessment model to obtain the credit risk assessment result of the user to be assessed. The risk assessment model is obtained by federated learning training of the historical financial behavior data of the historical user and the corresponding assessment result label of the financial client, the historical spatiotemporal behavior data of the historical user of the operator client, and the historical consumption behavior data of the historical user of the internet client.

[0077] The mobile devices used by the users to be evaluated may have operator clients or internet clients installed. Operator clients can be used to obtain the user's spatiotemporal behavioral data, while internet clients can be used to obtain the user's consumption behavior data. The system receives spatiotemporal behavioral data of the users to be evaluated from the operator clients.

[0078] Mobile devices can be various electronic devices, including but not limited to smartphones, tablets, laptops, smartwatches, wearable devices, augmented reality devices, virtual reality devices, etc.

[0079] The client for an application installed on different mobile devices may be the same, or it may be the client for the same type of application based on different operating systems. Depending on the terminal platform, the specific form of the application's client can also differ; for example, it can be a mobile client, a PC client, etc.

[0080] In one embodiment, an operator client refers to a client for communication services provided by a telecommunications operator, through which users can conduct communication services such as voice calls, internet access, video calls, and MMS.

[0081] The operator's client can collect spatiotemporal behavioral data of the user corresponding to the user identifier.

[0082] Internet clients refer to shopping platforms installed by users on their mobile devices. Internet clients can obtain the corresponding user's consumption behavior data based on the user's identifier.

[0083] The risk assessment model is used to learn the mapping relationship between users' financial behavior data, spatiotemporal behavior data, consumption behavior data and credit risk assessment results.

[0084] It should be noted that during the training of the risk assessment model, financial clients hold historical financial behavior data and corresponding assessment result labels, telecom clients hold historical spatiotemporal behavior data of historical users, and internet clients hold historical consumption behavior data of historical users.

[0085] Historical spatiotemporal behavior data for historical users can include communication data such as calls, text messages, and data usage in different regions and time periods. It can collect spatiotemporal behavior data for historical users across four time periods within a single day, each lasting six hours. For example, each six-hour period could be divided into four time periods: 0-6 (early morning), 6-12 (morning), 12-18 (afternoon), and 18-24 (evening).

[0086] Historical user consumption behavior data can include social data such as the amount spent on purchases, number of purchases, items purchased, consumption cycle and frequency of daily necessities, usage within a normal cycle, and purchase data of non-daily necessities over the past three months.

[0087] In one embodiment, the risk assessment model is iteratively trained based on the user's historical financial behavior data, historical spatiotemporal behavior data, historical consumption behavior data, and corresponding risk assessment result labels. The risk assessment model takes financial behavior data as input and credit risk assessment results as output, and can perform risk assessment on the user to be assessed.

[0088] In one embodiment, the risk identification model may employ machine learning models such as logistic regression, decision trees, distributed gradient boosting library XGBoost, and learning organization LightGBM, without making specific limitations in this disclosure.

[0089] The credit risk assessment results mentioned above can be at least one of risk assessment score and risk assessment level.

[0090] When the credit risk assessment result is a risk assessment score, if the risk assessment score is greater than or equal to a preset risk assessment score threshold, the user to be assessed is determined to be a high-risk user; if the risk assessment score is less than the preset risk assessment score threshold, the user to be assessed is determined to be a low-risk user. It should be noted that the preset risk assessment score threshold can be pre-configured in the financial client, and the value of the preset risk assessment score threshold is not specifically limited.

[0091] When the credit risk assessment result is a risk assessment level, corresponding processing strategies can be adopted according to the value of the risk assessment level. For example, when the risk assessment level is high, the user's credit business request can be rejected; when the risk assessment level is medium, the credit risk assessment time for the user to be assessed can be extended; when the risk assessment level is low, the user's credit business request can be accepted and the corresponding service can be provided to the user.

[0092] In this embodiment, in response to a credit business request from a user to be assessed, the financial client obtains the user's financial behavior data. The financial client processes the user's financial behavior data using a risk assessment model to obtain the user's credit risk assessment result. The risk assessment model is trained through federated learning using historical financial behavior data and corresponding assessment result labels from the financial client, historical spatiotemporal behavior data from the operator client, and historical consumption behavior data from the internet client. This disclosure combines user's financial behavior data, spatiotemporal behavior data, and consumption behavior data to extract user risk characteristics from multi-source data, trains the risk assessment model, and achieves accurate identification and early warning of user credit risk through the risk assessment model, thereby improving credit.

[0093] Figure 2 This diagram illustrates a flowchart of a risk assessment model construction method provided by an embodiment of this disclosure. Figure 2 As shown, in one embodiment, the risk assessment model is trained in the following manner:

[0094] S202. The financial client obtains the first gradient of the initial risk assessment model based on the historical financial behavior data of historical users, the corresponding assessment result labels, and the model parameters of the financial client; it obtains the second gradient by interacting with the operator client through homomorphic encryption, and obtains the third gradient by interacting with the Internet client through homomorphic encryption.

[0095] S204. The operator client interacts with the financial client and the internet client through homomorphic encryption to obtain the total gradient of the operator client;

[0096] S206. The Internet client interacts with the financial client and the operator client through homomorphic encryption to obtain the total gradient of the Internet client;

[0097] S208. Update the model parameters iteratively based on gradient descent until the preset conditions are met to obtain the risk assessment model;

[0098] The second gradient is determined by historical spatiotemporal behavior data of historical users, model parameters of operator clients, and historical financial behavior data, while the third gradient is determined by historical consumption behavior data of historical users, model parameters of internet clients, and historical financial behavior data.

[0099] In S202, the initial risk assessment model can be determined through the model parameters of the financial client. Historical financial behavior data can be input into the initial risk assessment model, and the corresponding assessment results can be output. The difference between the assessment results and the assessment result labels is calculated to obtain the first gradient, which is then used to update the model parameters through gradient descent.

[0100] In one embodiment, homomorphic encryption can process encrypted data without decryption, and the processed result is consistent with the original expected result after decryption. This allows historical consumption behavior data, model parameters of Internet clients, historical spatiotemporal behavior data, and model parameters of operator clients to be calculated by third parties during transmission or storage without leaking the data itself, thus ensuring data privacy and security.

[0101] Homomorphic encryption can employ semi-homomorphic encryption, partial homomorphic encryption, full homomorphic encryption, etc. This disclosure determines the type of homomorphic encryption based on actual needs and does not limit the type of homomorphic encryption.

[0102] In S202, the obtained first gradient, second gradient and third gradient can be aggregated to obtain the total gradient of the initial risk assessment model for the financial client.

[0103] The initial risk assessment model is the risk assessment model to be trained. The initial risk assessment model can adopt machine learning models such as logistic regression model, decision tree, Xgboost, and LightGBM.

[0104] During model training, financial client A, telecom operator client B, and internet client C each possess partial features of the samples, while only the financial client holds the evaluation result label. If the number of features for the financial client is d... A The number of features of the operator's client is d B The number of features of an internet client is d. C Then the overall sample dimension is D = d A +d B +d CSuppose we split the entire feature matrix into three parts of the same dimension and store them on the financial client, the telecom operator client, and the internet client, respectively. Then the gradient of the risk assessment model... It can be represented as:

[0105]

[0106] Where, x A x B x C The characteristics of financial clients, carrier clients, and internet clients are respectively θ A θ B θ C The model parameters are y for financial clients, carrier clients, and internet clients, respectively. i Label the evaluation results.

[0107] x a x B x C The dimension is n×D, θ A θ B θ C All dimensions are 1×D.

[0108] The parameter update methods for financial clients, carrier clients, and internet clients are expressed as follows:

[0109]

[0110] Where α is the learning rate, which takes a value between 0 and 1.

[0111] Removing the constant term from Equation 1, the gradient of financial client A can be further decomposed as follows:

[0112]

[0113] in, It can be calculated locally by financial client A. and This can be obtained through homomorphic encryption between financial client A and operator client and internet client.

[0114] In S204, correspondingly, removing the constant term in Equation 1, the gradient of the operator's client B is further decomposed and expressed as:

[0115]

[0116] in, It can be calculated locally by operator client B. This can be obtained through homomorphic encryption between operator client B and financial client A. It can be obtained through homomorphic encryption between operator client B and internet client C.

[0117] In S206, correspondingly, removing the constant term in Equation 1, the gradient of the operator client C is further decomposed and expressed as:

[0118]

[0119] in, It can be calculated locally by Internet client C. This can be obtained through homomorphic encryption between internet client C and financial client A. It can be obtained through homomorphic encryption between operator client B and internet client C.

[0120] In S208, the goal is to minimize the total gradient, and iterative training is performed until the preset conditions are met, resulting in a trained risk assessment model.

[0121] The preset conditions can be determined according to actual needs, and this disclosure does not impose specific limitations on them.

[0122] Minimum total gradient means that the total gradient of financial clients, total gradient of carrier clients, and total gradient of internet clients are all minimized.

[0123] In this disclosed embodiment, federated learning can collaboratively train the risk assessment model using data from clients distributed across different data owners, thereby improving the accuracy of credit risk assessment results and ensuring security and reliability.

[0124] Figure 3 This diagram illustrates the principle of feature vector extraction by three clients according to an embodiment of this disclosure. Figure 3 As shown, in one embodiment, the S202 financial client trains the initial risk assessment model based on the historical financial behavior data of historical users and the corresponding assessment result labels to obtain the first gradient of the initial risk assessment model, including:

[0125] S202A: The financial client extracts features from historical financial behavior data to obtain the first feature vector x. A ;

[0126] S202B, the financial client uses the initial risk assessment model to analyze the first feature vector x. A The process was carried out to obtain the first assessment result.

[0127] S202C and financial clients, based on the first assessment results And the first assessment result label y i Perform local gradient updates and determine the first gradient.

[0128] In one embodiment, the financial client can employ a transformer model to extract features from historical financial behavior data, obtaining a first feature vector. The transformer model, also known as the Transformer model, is a deep learning model architecture used for natural language processing and other sequence-to-sequence methods. It introduces a self-attention mechanism, enabling the model to consider all positions in the input sequence simultaneously. This allows the model to assign different attention weights based on different parts of the input sequence, thereby better capturing semantic relationships. The Transformer model can extract features from historical users' historical financial behavior data to obtain a first feature vector. This first feature vector not only includes basic statistics on daily user credit but also correlation trend features of user network activities at different times within the day, thus extracting more feature vectors and improving the accuracy of credit risk assessment.

[0129] The first assessment result is the prediction result output by the initial risk assessment model. Based on the first assessment result and its label, the loss function value can be calculated using the mean squared error function, and a local gradient update can be performed to determine the first gradient. This is denoted as the determination in Formula 5. The process.

[0130] In this embodiment, the financial client extracts features from historical financial behavior data through a converter model and calculates the first gradient for the initial risk assessment model, which can reduce the computational load of the financial client and improve the training efficiency of the risk assessment model.

[0131] Continue to refer to Figure 3 In one embodiment, the financial client in S202 above interacts with the operator client via homomorphic encryption to obtain the second gradient, including:

[0132] S202A: The operator client extracts spatial and temporal features from historical spatiotemporal behavior data to obtain a second feature vector. Based on the model parameters and the second feature vector of the operator client, a second evaluation result is obtained. The second evaluation result is homomorphically encrypted to obtain the ciphertext of the second evaluation result and sent to the financial client.

[0133] S202B: The financial client processes the ciphertext of the second evaluation result based on the first feature vector and the first random number to obtain the ciphertext of the second gradient information.

[0134] S202C: The operator client receives the ciphertext of the second gradient information and decrypts it, then sends the decrypted second gradient information to the client.

[0135] S202D and the financial client receive the decrypted second gradient information and obtain the second gradient based on the first random number and the decrypted second gradient information.

[0136] In one embodiment, the first neural network model can be a convolutional neural network (CNN) model; the second neural model can be a recurrent neural network (RNN) model, or it can be based on the RNN model and integrate gating mechanisms such as long short-term memory (LSTM) or gated recurrent unit (GRU) to improve the model's expressive and memory capabilities and better process long sequence data.

[0137] Spatiotemporal behavioral data from different regions and time periods have both temporal and spatial attributes. CNNs can be used to extract spatial features from spatiotemporal behavioral data for each time period within a 24-hour period each day to obtain initial features. The initial features are then input into RNNs for temporal feature extraction to extract the relationships between time periods, resulting in a second feature vector. The second feature vector represents the communication status and credit status of the user to be evaluated in different regions. Typically, users who frequently and rapidly switch between different regions have a higher risk of fraud.

[0138] It should be noted that the user's temporal features can also be extracted first through RNN, and then spatial features can be extracted through CNN model to obtain the second feature vector. This disclosure does not make specific limitations on this.

[0139] In one embodiment, spatial features can be extracted from historical spatiotemporal behavior data using a first neural network model to obtain spatial feature vectors, and temporal features can be extracted from historical spatiotemporal behavior data using a second neural network model to obtain temporal feature vectors. The spatial feature vectors and temporal feature vectors can be fused to construct cross-domain fusion features for historical users, and the initial risk assessment model can be iteratively trained based on the cross-domain fusion features.

[0140] The aforementioned feature fusion methods include, but are not limited to, feature concatenation, feature summation, and feature dot product, etc., and this disclosure does not specifically limit them.

[0141] In one embodiment, the operator client generates a homomorphic encryption public key and a homomorphic encryption private key, sends the homomorphic encryption public key to the financial client, and the financial client receives the homomorphic encryption public key.

[0142] The operator's client extracts spatial and temporal features from historical spatiotemporal behavior data to obtain a second feature vector x. B Based on the model parameters θ of the operator's client B Second eigenvector x B The second evaluation result was obtained. Using homomorphic encryption private key to evaluate the second result Homomorphic encryption is performed to obtain the ciphertext of the second evaluation result. And send it to the financial client;

[0143] The financial client uses the first feature vector x A and the first random number R A Confidential text of the second assessment results The ciphertext of the second gradient information is obtained through processing. The first random number is a random number generated by the financial client.

[0144] The ciphertext of the second-gradient information received by the operator's client And use the homomorphic encryption public key to ciphertext the second gradient information Decryption is performed, and the decrypted second-gradient information is...

[0145] The financial client receives the decrypted second-gradient information. Based on the first random number R A and the decrypted second gradient information Obtain the second gradient

[0146] In this embodiment, the operator client extracts time and space features from historical spatiotemporal behavioral data, effectively extracting risk features from multi-party cross-domain data, improving the accuracy of credit risk identification, and enhancing data transmission security through homomorphic encryption technology.

[0147] Continue to refer to Figure 3 In another embodiment, the financial client in S202 above interacts with the internet client via homomorphic encryption to obtain the third gradient, including:

[0148] S202E and the Internet client extract features from historical consumption behavior data to obtain a third feature vector. Based on the model parameters of the Internet client and the third feature vector, a third evaluation result is obtained. The third evaluation result is homomorphically encrypted to obtain the ciphertext of the third evaluation result and sent to the financial client.

[0149] S202F and the financial client process the ciphertext of the third evaluation result based on the first feature vector and the second random number to obtain the ciphertext of the third gradient information.

[0150] S202G and the operator's client receive the ciphertext of the third information and decrypt the ciphertext of the second gradient information to obtain the decrypted third gradient information.

[0151] S202H and the financial client receive the decrypted third gradient information and obtain the third gradient based on the second random number and the decrypted third gradient information.

[0152] In one embodiment, the Internet client extracts features from historical consumption behavior data using a graph convolutional neural network model to obtain a third feature vector X3; the Internet client processes the third feature vector X3 using an initial risk assessment model to obtain a third assessment result, and transmits the third assessment result to the financial client using homomorphic encryption technology.

[0153] In one embodiment, the Graph Convolutional Network (GCN) model is a convolutional neural network that, for each node, obtains the node's feature information from all its neighboring nodes.

[0154] The GCN model extracts features from social network data, including analyzing and transforming data such as the consumption cycle, frequency, and usage within a normal cycle of daily necessities. This constructs a feature vector describing the daily necessities purchases of the user to be evaluated. It also analyzes the purchase data of non-daily necessities to obtain a feature vector of non-daily necessities at the given location. The feature vectors of daily necessities and non-daily necessities can be combined to form a third feature vector X3. The third feature vector X3 can describe the consumption information of the user to be evaluated, characterizing the user's consumption style, consumption habits, and income, thereby providing a comprehensive assessment of the credit risk of the user and improving the accuracy of the assessment results.

[0155] It should be noted that the process by which the financial client and the internet client determine the third gradient using homomorphic encryption technology is similar to the process by which the financial client and the operator client determine the second gradient using homomorphic encryption technology in the aforementioned embodiment. It will not be repeated here, and both can improve the security of data transmission.

[0156] Figure 4 This diagram illustrates another risk assessment model construction method provided by an embodiment of the present disclosure. Figure 4 As shown, in one embodiment, the above-mentioned S208 iteratively updates the model parameters based on gradient descent until a preset condition is met, thereby obtaining a risk assessment model, including:

[0157] S2082. If the total gradient reaches the minimum, the number of iterations reaches the preset number threshold, or the iteration duration reaches the preset duration threshold, then the preset conditions are met, and a risk assessment model is constructed based on the model parameters that meet the preset conditions.

[0158] In S2082, the model parameters of the risk assessment model can be calculated according to Equation 1 by using the global gradient descent method.

[0159] The loss function can be the mean square error function, the root mean square error function, etc.

[0160] The preset number of times threshold and the preset duration threshold can be determined according to actual needs, and this disclosure does not impose specific limitations on them.

[0161] In this disclosed embodiment, during federated learning and model inference, each participating party does not need to send private or sensitive data to the financial client. Instead, the model is trained by transmitting gradients, thereby improving the security of credit risk assessment.

[0162] Accordingly, removing the constant term in Formula 1, the gradient of operator client B is further decomposed and expressed using Formula 6:

[0163]

[0164] in, It can be calculated locally by the operator's client B.

[0165] This can be obtained through homomorphic encryption between operator client B and financial client A. It can be obtained through homomorphic encryption between operator client B and internet client C.

[0166] calculate The process is as follows:

[0167] Financial client A generates a homomorphic encryption public key and a homomorphic encryption private key, and sends the homomorphic encryption public key to operator client B;

[0168] Financial client A uses homomorphic encrypted private key to calculate The results will be sent to operator client B;

[0169] Operator Client B Calculation The result is then sent to the financial client, where R2 is the second random number generated by the operator client B;

[0170] Financial client A uses homomorphic encryption private key pairs Decryption, obtain Will Send to operator client B;

[0171] Carrier client computing get Used for subsequent calculation of the total gradient of operator client B.

[0172] calculate The process is as follows:

[0173] Internet client C generates a homomorphic encryption public key and a homomorphic encryption private key, and sends the homomorphic encryption public key to operator client B;

[0174] Internet client C uses its own key to calculate The results will be sent to operator client B;

[0175] Operator Client B Calculation The result is then sent to Internet client C, where R3 is the third random number generated by operator client B;

[0176] Internet client C uses homomorphic encryption and private key decryption. get Send the results to operator client B;

[0177] Operator Client B Calculation get

[0178] It should be noted that the training process for Internet client C is similar to that for operator client B, and will not be described in detail here.

[0179] Figure 5 A flowchart illustrating another credit risk assessment method provided in an embodiment of this disclosure is shown. Figure 1 Based on the previous example, S103 is added before S104 to limit the method of determining the corresponding risk assessment model for credit risk assessment according to different scenarios. For example... Figure 5 As shown, in one embodiment, the credit risk assessment method provided by this disclosure includes steps S102 to S104. Specifically, before step S104, where the financial client processes the financial behavior data of the user to be assessed using a risk assessment model to obtain the credit risk assessment result, the method further includes:

[0180] S103. The financial client determines the risk assessment model based on the scenario type of the credit business request, with one risk assessment model corresponding to one scenario type.

[0181] In one embodiment, the scenario types for credit business requests include, but are not limited to, consumer loan credit assessment and cash loan income assessment.

[0182] Depending on the type of scenario, corresponding risk assessment model training tasks can be constructed and the models can be trained.

[0183] After completing all training tasks in sequence, a risk assessment model corresponding to each scenario type is obtained, thereby accurately assessing the credit risk of the user to be assessed in different credit scenarios based on the risk assessment model.

[0184] In one embodiment, the credit risk assessment method provided in this disclosure further includes: constructing learning tasks corresponding to meta-learning based on multiple application scenarios, with one learning task corresponding to one application scenario; processing each learning task through the multi-task risk assessment model to be trained to obtain the model gradient of each learning task; performing gradient aggregation on the model gradients of each learning task, with the goal of minimizing the gradient of the multi-task risk assessment model, and iteratively training to obtain the multi-task assessment model.

[0185] In one embodiment, each participant (financial client, operator client, internet client) can extract common features for each scenario and train a local model for the corresponding scenario based on the extracted common features to obtain the model parameters and gradients of the local model. The model parameters and gradients of the local model are then sent to the aggregation server so that the aggregation server can construct the above-mentioned multi-task risk assessment model based on each local model and its corresponding model parameters and gradients.

[0186] Figure 6 This diagram illustrates the principle of training a credit risk assessment model according to an embodiment of this disclosure. Figure 6 As shown, the multi-task risk assessment model is built based on features from Scenario 1 and Scenario 2. For Scenario 1 features, a support dataset and a query dataset can be constructed. The financial client uses the support dataset to train a first local model and performs gradient updates to obtain the first model parameters. Based on the query dataset, the loss function is calculated, and the fourth gradient is calculated to obtain the updated first model parameters. The fourth gradient is then transmitted to the aggregation server. The operator client uses the support dataset to train a second local model and performs gradient updates to obtain the second model parameters. Based on the query dataset, the loss function is calculated, and the fifth gradient is calculated to obtain the updated second model parameters. The fifth gradient is then transmitted to the aggregation server. The internet client uses the support dataset to train a third local model and performs gradient updates to obtain the third model parameters. Based on the query dataset, the loss function is calculated, and the sixth gradient is calculated to obtain the updated third model parameters. The sixth gradient is then transmitted to the aggregation server. For Scenario 2 features, the model training method is similar to that in Scenario 1, and will not be elaborated here.

[0187] The aggregation server performs global gradient updates based on the model parameters trained on the features of each scenario, resulting in a multi-task risk assessment model.

[0188] In this embodiment, a meta-learning technology architecture is adopted to extract common features from different credit scenarios and construct a multi-task risk assessment model, thereby improving the model's generalization ability.

[0189] Based on the same inventive concept, this disclosure also provides a credit risk assessment system, as shown in the following embodiments. Since the principle by which this system embodiment solves the problem is similar to that of the above method embodiments, the implementation of this system embodiment can refer to the implementation of the above method embodiments, and repeated details will not be described again.

[0190] Figure 7 This diagram illustrates a credit risk assessment system provided by an embodiment of the present disclosure. Figure 7 As shown, the credit risk assessment system in this embodiment includes a financial client 710, a telecom operator client 720, and an internet client 730.

[0191] Among them, the financial client 710 is used to respond to the credit business request of the user to be evaluated and obtain the financial behavior data of the user to be evaluated; the financial behavior data of the user to be evaluated is processed through the risk assessment model to obtain the credit risk assessment result of the user to be evaluated; the risk assessment model is obtained by federated learning training of the historical financial behavior data of historical users and corresponding assessment result labels of the financial client 710, the historical spatiotemporal behavior data of historical users of the operator client 720, and the historical consumption behavior data of historical users of the internet client 830.

[0192] In one embodiment, the financial client 710 is used to obtain the first gradient of the initial risk assessment model based on the historical financial behavior data of historical users, the corresponding assessment result labels, and the model parameters of the financial client; to obtain the second gradient by interacting with the operator client through homomorphic encryption; to obtain the third gradient by interacting with the Internet client through homomorphic encryption; and to aggregate the first, second, and third gradients to obtain the total gradient of the financial client.

[0193] The operator client 720 is used to interact with the financial client 710 and the internet client 730 through homomorphic encryption to obtain the total gradient of the operator client;

[0194] Internet client 730 is used to interact with financial client 710 and operator client 720 through homomorphic encryption to obtain the total gradient of the Internet client;

[0195] The model parameters are iteratively updated based on gradient descent until the preset conditions are met, thus obtaining the risk assessment model.

[0196] The second gradient is determined by historical spatiotemporal behavior data of historical users, model parameters of operator clients, and historical financial behavior data; the third gradient is determined by historical consumption behavior data of historical users, model parameters of internet clients, and historical financial behavior data.

[0197] In one embodiment, the financial client 710 is used to extract features from historical financial behavior data to obtain a first feature vector; process the first feature vector through an initial risk assessment model to obtain a first assessment result; and perform local gradient updates based on the first assessment result and the first assessment result label to determine the first gradient.

[0198] In one embodiment, the operator client 720 is used to extract spatial and temporal features from historical spatiotemporal behavior data to obtain a second feature vector. Based on the model parameters and the second feature vector of the operator client, a second evaluation result is obtained. The second evaluation result is homomorphically encrypted to obtain the ciphertext of the second evaluation result and sent to the financial client 710.

[0199] Financial client 710 is used to process the ciphertext of the second evaluation result based on the first feature vector and the first random number to obtain the ciphertext of the second gradient information;

[0200] The operator client 720 is used to receive the ciphertext of the second gradient information, decrypt the ciphertext of the second gradient information, and send the decrypted second gradient information.

[0201] Financial client 710 is used to receive the decrypted second gradient information and obtain the second gradient based on the first random number and the decrypted second gradient information.

[0202] In one embodiment, the Internet client 730 is used to extract features from historical consumption behavior data to obtain a third feature vector. Based on the model parameters of the Internet client and the third feature vector, a third evaluation result is obtained. The third evaluation result is homomorphically encrypted to obtain the ciphertext of the third evaluation result and sent to the financial client 710.

[0203] Financial client 710 is used to process the ciphertext of the third evaluation result based on the first feature vector and the second random number to obtain the ciphertext of the third gradient information.

[0204] The operator client 730 is used to receive the ciphertext of the third information and decrypt the ciphertext of the second gradient information to obtain the decrypted third gradient information.

[0205] Financial client 710 is used to receive the decrypted third gradient information and obtain the third gradient based on the second random number and the decrypted third gradient information.

[0206] In one embodiment, the financial client 710 is configured to determine that a preset condition is met if the total gradient reaches the minimum, the number of iterations reaches a preset threshold, or the iteration duration reaches a preset duration threshold, and then construct a risk assessment model based on the model parameters that meet the preset condition.

[0207] In one embodiment, the financial client 710 is used to determine the risk assessment model based on the scenario type of the credit business request before the financial client processes the financial behavior data of the user to be assessed through the risk assessment model to obtain the credit risk assessment result of the user to be assessed. One scenario type corresponds to one risk assessment model.

[0208] In one embodiment, the financial client 710 is used to construct learning tasks corresponding to meta-learning based on multiple application scenarios, with one learning task corresponding to one application scenario; to process each learning task through the multi-task risk assessment model to be trained, and to obtain the model gradient of each learning task; to perform gradient aggregation on the model gradients of each learning task, with the goal of minimizing the gradient of the multi-task risk assessment model, and to perform iterative training to obtain the multi-task assessment model.

[0209] In this embodiment, in response to a credit business request from a user to be evaluated, the financial client 710 acquires the user's financial behavior data. The financial client 710 processes the user's financial behavior data using a risk assessment model to obtain the user's credit risk assessment result. The risk assessment model is trained through federated learning by the financial client 710 using historical financial behavior data of historical users and corresponding assessment result labels, the operator client 720 using historical spatiotemporal behavior data of historical users, and the internet client 730 using historical consumption behavior data of historical users. This disclosure combines user's financial behavior data, spatiotemporal behavior data, and consumption behavior data to extract user risk characteristics from multi-source data, trains the risk assessment model, and achieves accurate identification and early warning of user credit risk through the risk assessment model, thereby improving credit.

[0210] Those skilled in the art will understand that various aspects of the present invention can be implemented as systems, methods, or program products. Therefore, various aspects of the present invention can be specifically implemented in the following forms: entirely hardware implementations, entirely software implementations (including firmware, microcode, etc.), or implementations combining hardware and software aspects, collectively referred to herein as “circuits,” “modules,” or “systems.”

[0211] The following reference Figure 8To describe an electronic device 800 according to this embodiment of the present invention. The electronic device 800 includes a processor; and a memory for storing executable instructions of the processor, wherein the processor is configured to perform the steps of various exemplary embodiments of the present disclosure described in the "Exemplary Methods" section above by executing the executable instructions. Figure 8 The electronic device 800 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0212] like Figure 8 As shown, the electronic device 800 is manifested in the form of a general-purpose computing device. The components of the electronic device 800 may include, but are not limited to: at least one processing unit 810, at least one storage unit 820, and a bus 830 connecting different system components (including storage unit 820 and processing unit 810).

[0213] The storage unit stores program code, which can be executed by the processing unit 810 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of the present invention. For example, the processing unit 810 can perform actions such as... Figure 1 As shown, in response to a credit business request from a user to be evaluated, the financial client obtains the user's financial behavior data. The financial client processes the user's financial behavior data through a risk assessment model to obtain the user's credit risk assessment result. The risk assessment model is trained by federated learning using the financial client's historical financial behavior data and corresponding assessment result labels, the operator's historical spatiotemporal behavior data of historical users, and the internet client's historical consumption behavior data of historical users.

[0214] Storage unit 820 may include a readable medium in the form of a volatile storage unit, such as random access memory (RAM) 8201 and / or cache memory 8202, and may further include a read-only memory (ROM) 8203.

[0215] The storage unit 820 may also include a program / utility 8204 having a set (at least one) of program modules 8205, such program modules 8205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0216] Bus 830 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0217] Electronic device 800 can also communicate with one or more external devices 840 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with the system, and / or with any device that enables the electronic device 800 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed through input / output (I / O) interface 860. Furthermore, the system can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) through network adapter 860. Figure 8 As shown, network adapter 860 communicates with other modules of electronic device 800 via bus 830. It should be understood that, although... Figure 8 As not shown in the diagram, other hardware and / or software modules may be used in conjunction with the electronic device 800, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0218] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal system, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0219] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. This computer-readable storage medium may be a readable signal medium or a readable storage medium.

[0220] In some possible implementations, various aspects of this disclosure may also be implemented as a program product comprising a computer program or computer instructions which, when run on a terminal device, are loaded and executed by a processor to cause the computer to perform the steps described in the “Exemplary Methods” section of this specification according to various exemplary embodiments of this disclosure.

[0221] More specific examples of computer-readable storage media in this disclosure may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0222] In this disclosure, a computer-readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of transmitting, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, system, or device.

[0223] Optionally, the program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0224] In practical implementation, program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0225] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0226] Furthermore, although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0227] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, mobile terminal, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0228] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure 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 disclosure are indicated by the appended claims.

Claims

1. A credit risk assessment method, characterized in that, The method includes: In response to a credit business request from a user to be evaluated, the financial client obtains the financial behavior data of the user to be evaluated; The financial client processes the financial behavior data of the user to be assessed through a risk assessment model to obtain the credit risk assessment result of the user to be assessed. The risk assessment model is obtained by federated learning of the historical financial behavior data and corresponding assessment result labels of the historical users by the financial client, the historical spatiotemporal behavior data of the historical users by the operator client, and the historical consumption behavior data of the historical users by the internet client.

2. The credit risk assessment method according to claim 1, characterized in that, The risk assessment model was trained in the following manner: The financial client obtains the first gradient of the initial risk assessment model based on the historical financial behavior data of the historical user, the corresponding assessment result tags, and the model parameters of the financial client. It then interacts with the operator client through homomorphic encryption to obtain the second gradient, and interacts with the internet client through homomorphic encryption to obtain the third gradient. The first gradient, the second gradient, and the third gradient are aggregated to obtain the total gradient of the financial client. The operator client interacts with the financial client and the internet client through homomorphic encryption to obtain the total gradient of the operator client; The internet client interacts with the financial client and the operator client through homomorphic encryption to obtain the total gradient of the internet client; The model parameters are iteratively updated based on gradient descent until the preset conditions are met, thus obtaining the risk assessment model. The second gradient is determined by the historical spatiotemporal behavior data of the historical user, the model parameters of the operator's client, and the historical financial behavior data; the third gradient is determined by the historical consumption behavior data of the historical user, the model parameters of the internet client, and the historical financial behavior data.

3. The credit risk assessment method according to claim 2, characterized in that, The financial client trains the initial risk assessment model based on the historical financial behavior data of the historical users and the corresponding assessment result labels, obtaining the first gradient of the initial risk assessment model, including: The financial client extracts features from the historical financial behavior data to obtain a first feature vector. The financial client processes the first feature vector using the initial risk assessment model to obtain a first assessment result; The financial client performs local gradient updates based on the first evaluation result and the first evaluation result label to determine the first gradient.

4. The credit risk assessment method according to claim 3, characterized in that, The financial client interacts with the operator client via homomorphic encryption to obtain the second gradient, which includes: The operator client extracts spatial and temporal features from the historical spatiotemporal behavior data to obtain a second feature vector. Based on the model parameters of the operator client and the second feature vector, a second evaluation result is obtained. The second evaluation result is homomorphically encrypted to obtain the ciphertext of the second evaluation result and sent to the financial client. The financial client processes the ciphertext of the second evaluation result based on the first feature vector and the first random number to obtain the ciphertext of the second gradient information. The operator client receives the ciphertext of the second gradient information, decrypts the ciphertext of the second gradient information, and sends the decrypted second gradient information. The financial client receives the decrypted second gradient information and obtains the second gradient based on the first random number and the decrypted second gradient information.

5. The credit risk assessment method according to claim 3, characterized in that, The financial client interacts with the internet client via homomorphic encryption to obtain the third gradient, which includes: The internet client extracts features from the historical consumption behavior data to obtain a third feature vector. Based on the model parameters of the internet client and the third feature vector, a third evaluation result is obtained. The third evaluation result is homomorphically encrypted to obtain the ciphertext of the third evaluation result and sent to the financial client. The financial client processes the ciphertext of the third evaluation result based on the first feature vector and the second random number to obtain the ciphertext of the third gradient information. The operator client receives the ciphertext of the third information and decrypts the ciphertext of the second gradient information to obtain the decrypted third gradient information. The financial client receives the decrypted third gradient information and obtains the third gradient based on the second random number and the decrypted third gradient information.

6. The credit risk assessment method according to claim 2, characterized in that, The step of iteratively updating the model parameters based on gradient descent until a preset condition is met to obtain the risk assessment model includes: If the total gradient reaches the minimum, or the number of iterations reaches a preset threshold, or the iteration duration reaches a preset duration threshold, then the preset conditions are satisfied, and the risk assessment model is constructed based on the model parameters that satisfy the preset conditions.

7. The credit risk assessment method according to claim 1, characterized in that, Before the financial client processes the financial behavior data of the user to be assessed through a risk assessment model to obtain the credit risk assessment result of the user to be assessed, the method further includes: The financial client determines the risk assessment model based on the scenario type of the credit business request, with one risk assessment model corresponding to one scenario type.

8. The credit risk assessment method according to claim 7, characterized in that, The method further includes: Based on multiple application scenarios, we construct learning tasks corresponding to meta-learning, with one learning task corresponding to one application scenario. The learning tasks are processed by the multi-task risk assessment model to be trained, and the model gradients of each learning task are obtained. Gradient aggregation is performed on the model gradients of each learning task to minimize the gradient of the multi-task risk assessment model. Iterative training is then performed to obtain the multi-task assessment model.

9. A credit risk assessment system, characterized in that, This includes financial clients, carrier clients, and internet clients, among which: The financial client is used to respond to the credit business request of the user to be evaluated, obtain the financial behavior data of the user to be evaluated, and process the financial behavior data of the user to be evaluated through a risk assessment model to obtain the credit risk assessment result of the user to be evaluated. The risk assessment model is obtained by federated learning of the historical financial behavior data and corresponding assessment result labels of the historical users by the financial client, the historical spatiotemporal behavior data of the historical users by the operator client, and the historical consumption behavior data of the historical users by the internet client.

10. An electronic device, characterized in that, include: processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the credit risk assessment method as described in any one of claims 1-8 by executing the executable instructions.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the credit risk assessment method as described in any one of claims 1-8.

12. A computer program product, characterized in that, The computer program product includes a computer program or computer instructions, which are loaded and executed by a processor to enable the computer to implement the credit risk assessment method as described in any one of claims 1-8.

Images