Anti-fraud method, device, system, equipment and storage medium

Abstract

The present application discloses an anti-fraud method, device, system, equipment and storage medium. The method further includes: obtaining data to be processed and inputting the data to be processed into a preset federated anti-fraud model; wherein the preset federated anti-fraud model is a target model obtained by performing iterative training on a preset prediction model to be trained based on preset fraud-related data with preset labels through executing a preset federated process; performing fraud recognition processing on the data to be processed based on the preset federated anti-fraud model to obtain a prediction result. The present application solves the technical problem in the prior art that fraud data is held in different devices such as different enterprises or government agencies, reducing the preventability of fraud.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology in financial technology (Fintech), and particularly to an anti-fraud method, device, system, equipment and storage medium. Background Art

[0002] With the continuous development of financial technology, especially Internet technology finance, more and more technologies are applied in the financial field. However, the financial industry also poses higher requirements on technologies. For example, the financial industry has higher requirements for anti-fraud.

[0003] With the advent of the Internet era, there are various fraud means, such as telephone fraud and network fraud, which are repeatedly prohibited and cause property losses to enterprises, people, etc. Among them, national public security organs, telecommunications companies, various firewall companies, etc. are all working hard to prevent people from being deceived by various means, such as marking fraud information and marking fraud calls. Although certain results have been achieved, there are still many fish slipping through the net. One of the reasons is that fraud data is held by different enterprises or government agencies. Due to user data privacy issues, this data generally cannot and will not be shared, which reduces the preventability of fraud. Summary of the Invention

[0004] The main purpose of the present application is to provide an anti-fraud method, device, system, equipment and storage medium, aiming to solve the technical problem in the prior art that fraud data is held by different devices, such as different enterprises or government agencies, which reduces the preventability of fraud.

[0005] To achieve the above purpose, the present application provides an anti-fraud method, and the anti-fraud method includes:

[0006] The anti-fraud method is applied to a first device, and the anti-fraud method includes:

[0007] Obtain data to be processed, and input the data to be processed into a preset federated anti-fraud model;

[0008] Wherein, the preset federated anti-fraud model is a target model obtained by performing iterative training on a preset prediction model to be trained through executing a preset federated process based on preset fraud-related data with preset labels;

[0009] Perform fraud identification processing on the data to be processed based on the preset federated anti-fraud model to obtain a prediction result.

[0010] Optionally, the data to be processed includes data of a loan to be processed;

[0011] After the step of performing fraud recognition processing on the to-be-processed data based on the preset federated anti-fraud model to obtain a prediction result, the method further includes:

[0012] If the prediction result is that the to-be-processed loan data is not of a fraud type, input the to-be-processed loan data into a preset risk control model;

[0013] Based on the preset risk control model, score the to-be-processed loan data to obtain a scoring result;

[0014] Based on the scoring result, output the target loanable numerical range of the to-be-processed loan data.

[0015] Optionally, before the step of inputting the to-be-processed data into the preset federated anti-fraud model, the method further includes:

[0016] Obtain preset fraud-related data with preset labels, and perform iterative training on the preset to-be-trained prediction model to train and update the model variables of the preset to-be-trained prediction model;

[0017] Determine whether the to-be-predicted model after iterative training reaches a preset replacement and update condition;

[0018] If the to-be-trained prediction model after iterative training reaches the preset replacement and update condition, then by executing the preset federated process, replace and update the model variables of the preset to-be-trained prediction model that has been trained and updated, and obtain the preset to-be-trained prediction model that has been replaced and updated;

[0019] Continuously perform iterative training and replacement and update on the preset to-be-trained prediction model that has been replaced and updated until the preset to-be-trained model meets the preset training completion condition, and obtain the preset federated anti-fraud model.

[0020] Optionally, the first device performs preset communication with the third device respectively;

[0021] The step of, if the to-be-trained prediction model after iterative training reaches the preset replacement and update condition, then by executing the preset federated process, replace and update the model variables of the preset to-be-trained prediction model that has been trained and updated, and obtain the preset to-be-trained prediction model that has been replaced and updated, includes:

[0022] If the to-be-trained prediction model after iterative training reaches the preset replacement and update condition, then obtain the first target model parameters of the to-be-trained prediction model that reaches the preset replacement and update condition;

[0023] Send the first target model parameters to the third device, so that the third device combines the second target model parameters sent by each second device and the first target model parameters to calculate the third target model parameters, where there is a preset vertical and / or horizontal federated association relationship between the first device and each second device;

[0024] Receive the third target model parameters sent by the third device, replace and update the first target model parameters with the third target model parameters, and obtain the preset prediction model to be trained after replacement and update.

[0025] Optionally, after the step of performing fraud identification processing on the data to be processed based on the preset federated anti-fraud model to obtain a prediction result, the method further includes:

[0026] Receive the target contribution degree of the first device calculated by the third device during the process of obtaining the prediction result, where the target contribution degree of the first device is recorded by a preset blockchain.

[0027] Optionally, after the step of receiving the target contribution degree of the first device calculated by the third device during the process of obtaining the prediction result, the method further includes:

[0028] When the target contribution degree is greater than the preset contribution degree, send the target contribution degree to the client associated with the first device.

[0029] The present application also provides an anti-fraud method, which is applied to a third device. The third device communicates with a first device and a second device respectively through preset communication. The anti-fraud method includes:

[0030] Receive the first target model parameters sent by the first device and receive the second target model parameters sent by the second device;

[0031] Combine the first target model parameters and the second target model parameters to calculate the third target model parameters, where there is a preset vertical and / or horizontal federated association relationship between the first device and each second device;

[0032] Send the third target model parameters to the first device for the first device to change the preset prediction model to be trained based on the third target model parameters.

[0033] Optionally, the first device is communicatively connected to a client. After the step of sending the third target model parameters to the first device for the first device to change the preset prediction model to be trained based on the third target model parameters, the method further includes:

[0034] When it is detected that the prediction of the data to be processed is successful, determine the target contribution degree of the first device and the contribution degrees of the second devices;

[0035] Send the target contribution degree of the first device and the contribution degrees of the second devices to a block of a preset blockchain, so that the block of the preset blockchain pushes the first device with a contribution degree greater than a preset value to the client.

[0036] This application also provides an anti-fraud device, which is applied to the first device. The anti-fraud device includes:

[0037] A first acquisition module, configured to acquire data to be processed and input the data to be processed into a preset federated anti-fraud model;

[0038] An identification module, configured to perform fraud identification processing on the data to be processed based on the preset federated anti-fraud model to obtain a prediction result;

[0039] Wherein, the preset federated anti-fraud model is a target model obtained by performing iterative training on a preset prediction model to be trained through executing a preset federated process based on preset fraud association data with preset labels.

[0040] Optionally, the anti-fraud device includes:

[0041] An input module, configured to input the data to be processed loan data into a preset risk control model when the prediction result is that the data to be processed loan data is not of a fraud type;

[0042] A scoring module, configured to score the data to be processed loan data based on the preset risk control model to obtain a scoring result;

[0043] An output module, configured to output a target loanable numerical range of the data to be processed loan data based on the scoring result.

[0044] Optionally, the anti-fraud device includes:

[0045] A second acquisition module, configured to acquire preset fraud association data with preset labels and perform iterative training on the preset prediction model to be trained to update the model variables of the preset prediction model to be trained;

[0046] A judgment module, configured to judge whether the prediction model to be predicted after iterative training reaches a preset replacement and update condition;

[0047] A third acquisition module, configured to, if the to-be-trained prediction model after iterative training meets a preset replacement and update condition, replace and update the model variables of the preset to-be-trained prediction model that has been trained and updated by executing the preset federated process, so as to obtain the preset to-be-trained prediction model after replacement and update;

[0048] A fourth acquisition module, configured to continuously perform iterative training and replacement and update on the preset to-be-trained prediction model after replacement and update until the preset to-be-trained model meets a preset training completion condition, so as to obtain the preset federated anti-fraud model.

[0049] Optionally, the first device performs preset communication with the third device respectively;

[0050] The third acquisition module includes:

[0051] A first acquisition unit, configured to, if the to-be-trained prediction model after iterative training meets a preset replacement and update condition, acquire first target model parameters of the to-be-trained prediction model that meets the preset replacement and update condition;

[0052] A sending unit, configured to send the first target model parameters to the third device, so that the third device combines the second target model parameters sent by each second device and the first target model parameters to calculate third target model parameters, where there is a preset vertical and / or horizontal federated association relationship between the first device and each second device;

[0053] A receiving unit, configured to receive the third target model parameters sent by the third device, and replace and update the first target model parameters with the third target model parameters, so as to obtain the preset to-be-trained prediction model after replacement and update.

[0054] Optionally, the anti-fraud device further includes:

[0055] A contribution degree acquisition module, configured to receive the target contribution degree of the first device calculated by the third device during the process of obtaining the prediction result, where the target contribution degree of the first device is recorded by a preset blockchain.

[0056] Optionally, the anti-fraud device further includes:

[0057] A contribution degree sending module, configured to send the target contribution degree to the client associated with the first device when the target contribution degree is greater than a preset contribution degree.

[0058] The present application further provides an anti-fraud device, which is applied to a third device. The third device performs preset communication with a first device and a second device respectively. The anti-fraud device further includes:

[0059] A receiving module, configured to receive the first target model parameters sent by the first device and receive the second target model parameters of the second device;

[0060] A calculation module, configured to jointly calculate the first target model parameters and the second target model parameters to obtain third target model parameters, where there is a preset vertical and / or horizontal federated association relationship between the first device and each second device;

[0061] An alteration module, configured to send the third target model parameters to the first device for the first device to alter a preset prediction model to be trained based on the third target model parameters.

[0062] Optionally, the anti-fraud device further includes:

[0063] A determination module, configured to determine the target contribution degree of the first device and the contribution degrees of each second device when it is detected that the prediction of the data to be processed is successful;

[0064] A pushing module, configured to send the target contribution degree of the first device and the contribution degrees of each second device to a block of a preset blockchain for the block of the preset blockchain to push the first device with a contribution degree greater than a preset value to a client.

[0065] This application further provides an anti-fraud system, where the anti-fraud system includes the anti-fraud device for the first device described above and a third device. The third device communicates with the first device and the second device respectively through preset communication, and the third device is configured to:

[0066] Receive the first target model parameters sent by the first device and receive the second target model parameters of the second device;

[0067] Jointly calculate the first target model parameters and the second target model parameters to obtain third target model parameters, where there is a preset vertical and / or horizontal federated association relationship between the first device and each second device;

[0068] Send the third target model parameters to the first device for the first device to alter a preset prediction model to be trained based on the third target model parameters.

[0069] This application further provides an anti-fraud device, where the anti-fraud device is a physical device, and the anti-fraud device includes: a memory, a processor, and a program of the anti-fraud method stored on the memory and executable on the processor. When the program of the anti-fraud method is executed by the processor, the steps of the anti-fraud method as described above can be implemented.

[0070] The present application also provides a storage medium, on which a program for implementing the above anti-fraud method is stored. When the program of the anti-fraud method is executed by a processor, the steps of the anti-fraud method as described above are implemented.

[0071] The present application obtains data to be processed and inputs the data to be processed into a preset federated anti-fraud model. Among them, the preset federated anti-fraud model is a target model obtained by iteratively training a preset prediction model to be trained based on preset fraud-related data with preset labels through executing a preset federated process. Based on the preset federated anti-fraud model, fraud identification processing is performed on the data to be processed to obtain a prediction result. In the present application, after obtaining the data to be processed, fraud identification processing is performed on the data to be processed based on the preset federated anti-fraud model to obtain a prediction result. The preset federated anti-fraud model is a target model obtained by iteratively training a preset prediction model to be trained based on preset fraud-related data with preset labels through executing a preset federated process (data does not move while the model moves). That is, while protecting data privacy, data does not move while the model moves is realized through the federated process, small data island analysis is changed into big data analysis, and the preventability of fraud is improved, solving the technical problem in the prior art that fraud data is held by different enterprises or government agencies, reducing the preventability of fraud. Description of the Drawings

[0072] The drawings here are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0073] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0074] Figure 1 It is a schematic flowchart of the first embodiment of the anti-fraud method of the present application;

[0075] Figure 2 It is a schematic flowchart of the refined steps before the step of inputting the data to be processed into the preset federated anti-fraud model in the first embodiment of the anti-fraud method of the present application;

[0076] Figure 3 It is a schematic diagram of the device structure of the hardware operating environment involved in the solution of the embodiment of the present application;

[0077] Figure 4 It is a schematic diagram of the scenario of the anti-fraud method of the present application.

[0078] The realization of the purpose, functional features and advantages of this application will be further described in conjunction with embodiments with reference to the accompanying drawings. Detailed implementation manners

[0079] It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0080] An embodiment of this application provides an anti-fraud method. The anti-fraud method is applied to a first device. In the first embodiment of the anti-fraud method of this application, with reference to Figure 1 , the anti-fraud method includes:

[0081] Step S10, obtain data to be processed, and input the data to be processed into a preset federated anti-fraud model;

[0082] Among them, the preset federated anti-fraud model is a target model obtained by iteratively training a preset prediction model to be trained based on preset fraud-related data with preset labels by executing a preset federated process;

[0083] Step S20, perform fraud identification processing on the data to be processed based on the preset federated anti-fraud model to obtain a prediction result.

[0084] The specific steps are as follows:

[0085] Step S10, obtain data to be processed, and input the data to be processed into a preset federated anti-fraud model;

[0086] Among them, the preset federated anti-fraud model is a target model obtained by iteratively training a preset prediction model to be trained based on preset fraud-related data with preset labels by executing a preset federated process;

[0087] In this embodiment, it should be noted that the anti-fraud method is applied to the first device, which belongs to the anti-fraud system, and the anti-fraud system is subordinate to the anti-fraud device. For the anti-fraud system, in addition to the first device, it also includes a second device. Among them, the second device has a preset federated association relationship with the first device, and the preset federated association relationship includes a preset vertical federated association relationship or a preset horizontal federated association relationship. Among them, the data features overlap less between devices with a preset vertical federated association relationship (such as the first device and the second device), while the user overlap is more. The part of users with the same users but different user data features and the corresponding data can be taken out between devices for joint machine learning training. For example, there are two devices A and B in the same region. Device A is a bank, and device B is an e-commerce platform. Devices A and B have many same users in the same region, but the businesses of A and B are different, and the recorded user data features are different. In particular, the user data features recorded by devices A and B may be complementary. Vertical federated learning can be used to jointly build a prediction model for A and B to achieve big data analysis and then provide better services to customers.

[0088] Different from the situation where the data features overlap less and the user overlap is more between devices with a preset vertical federated association relationship, the data features overlap more and the user overlap is less between devices with a preset horizontal federated association relationship. Therefore, the part of data with the same data features but not completely the same users is taken out for horizontal joint machine learning. Specifically, for example, there are two banks in different regions. Their user groups come from their respective regions, and the intersection is very small. However, their businesses are very similar, and the recorded user data features are the same. Horizontal federated learning can be used to jointly build a prediction model for the two banks.

[0089] Overall, the first device and each second device form a federated anti-fraud model. In this federated anti-fraud model, devices (enterprises or companies) in the same industry use horizontal federated learning modeling technology to establish a horizontal federated anti-fraud model for this industry, and devices (enterprises or companies) in different industries use vertical federated learning modeling technology to establish a vertical federated anti-fraud model for this region. Among them, each device (enterprise or company) is an endpoint in the federation (including horizontal federation and vertical federation).

[0090] In addition, it should be noted that the anti-fraud system may also include a third device, which can assist the first device and the second device in federation as a third party. For example, the national public security organ, as a public credibility institution, can act as the third device or the central service node.

[0091] Obtain the data to be processed, and input the data to be processed into a preset federated anti-fraud model. Among them, the preset federated anti-fraud model is a target model obtained by performing iterative training on a preset prediction model to be trained through a preset federated process based on preset fraud-related data with preset labels; since the preset federated anti-fraud model has been iteratively trained and is obtained as a target model after performing the preset federated process and federating with other devices for iterative training, it can accurately predict the data to be processed.

[0092] It should be noted that after obtaining the data to be processed and inputting the data to be processed into the preset federated anti-fraud model, since the preset federated anti-fraud model is a target model obtained by performing iterative training on a preset prediction model to be trained through a preset federated process based on preset fraud-related data with preset labels, that is, the preset federated anti-fraud model is trained with labeled data. If the data to be processed is a certain type of data in the labeled data, the preset federated anti-fraud model can match and predict the data to be processed. If it cannot be matched and predicted, the preset federated anti-fraud model is updated to achieve accurate prediction.

[0093] Such as Figure 2 As shown, before the step of inputting the data to be processed into the preset federated anti-fraud model, the method further includes:

[0094] Step S01, obtain preset fraud-related data with preset labels, and perform iterative training on the preset prediction model to be trained to train and update the model variables of the preset prediction model to be trained;

[0095] The first device obtains preset fraud-related data with preset labels, such as telecommunications data (such as SMS sub-data / phone sub-data, etc.), Internet data (such as social sub-data / shopping sub-data, etc.) or financial data (such as transfer sub-data / financial product sub-data, etc.), and performs iterative training on the preset prediction model to be trained to train and update the model variables of the preset prediction model to be trained. For example, perform iterative training on the preset prediction model to be trained through the preset telecommunications data with preset labels to train and update the model variables of the preset prediction model to be trained, or perform iterative training on the preset prediction model to be trained through the preset Internet data with preset labels to train and update the model variables of the preset prediction model to be trained. In this embodiment, the method for performing iterative training on the preset prediction model to be trained includes, but is not limited to, the gradient descent method.

[0096] Step S02, determine whether the prediction model to be predicted after iterative training reaches the preset replacement and update condition;

[0097] Step S03. If the to-be-trained prediction model after iterative training meets the preset replacement and update condition, then by executing the preset federated process, replace and update the model variables of the preset to-be-trained prediction model after training and update, and obtain the preset to-be-trained prediction model after replacement and update;

[0098] It should be noted that the preset replacement and update conditions include reaching the first iteration number threshold, reaching the first training round threshold, etc. In this embodiment, if the preset to-be-trained prediction model after iterative training meets the preset replacement and update condition, then by executing the preset federated process, replace and update the model variables of the preset to-be-trained prediction model after training and update, and obtain the preset to-be-trained prediction model after replacement and update. Specifically, replacing and updating the model variables of the preset to-be-trained prediction model after training and update to obtain the preset to-be-trained prediction model after replacement and update includes: directly receiving the corresponding model variables of the second device (including multiple), and then based on the corresponding model variables of the second device, obtaining the aggregated model variables with the model variables of the first device. After obtaining the aggregated model variables, then based on the aggregated model variables, replace and update the model variables of the preset to-be-trained prediction model to obtain the preset to-be-trained prediction model after replacement and update.

[0099] Step S04. Continuously perform iterative training and replacement and update on the preset to-be-trained prediction model after replacement and update until the preset to-be-trained model meets the preset training completion condition, and obtain the preset federated anti-fraud model.

[0100] In this embodiment, based on the model variables after replacement and update, re-perform iterative training on the preset to-be-trained prediction model after replacement and update and judge whether the preset replacement and update condition is met. Among them, the preset training completion conditions include reaching the second iteration number threshold, reaching the second training round threshold, etc. Continuously perform iterative training and replacement and update on the preset to-be-trained prediction model after replacement and update until the preset to-be-trained model meets the preset training completion condition, and obtain the preset federated anti-fraud model.

[0101] Further, the first device performs preset communication with the third device;

[0102] The step of if the to-be-trained prediction model after iterative training meets the preset replacement and update condition, then by executing the preset federated process, replace and update the model variables of the preset to-be-trained prediction model after training and update, and obtain the preset to-be-trained prediction model after replacement and update includes:

[0103] Step S031. If the to-be-trained prediction model after iterative training meets the preset replacement and update condition, then obtain the first target model parameters of the to-be-trained prediction model that meets the preset replacement and update condition;

[0104] Step S032: Send the first target model parameters to the third device, so that the third device combines the multiple second target model parameters sent by each second device and the first target model parameters to calculate third target model parameters, where there is a preset vertical and / or horizontal federated association relationship between the first device and each second device;

[0105] In this embodiment, the third device can assist the first device and the second device in federation as a third party. Specifically, the first device sends the first target model parameters (encrypted) to the third device, that is, the third device receives the (encrypted) first target model parameters of the first device and receives the (encrypted) second target model parameters of the second device. In particular, it receives the (encrypted) second target model parameters of multiple second devices.

[0106] The third device combines the multiple second target model parameters sent by each second device and the first target model parameters to calculate third target model parameters, such as performing mean processing on the multiple second target model parameters and the first target model parameters to obtain third target model parameters.

[0107] It should be noted that there is a preset vertical and / or horizontal federated association relationship between the first device and each second device. When there is a preset vertical federated association relationship between the first device and each second device, overall, the first device and the second devices in different industries form a federated anti-fraud model. When there is a preset horizontal federated association relationship between the first device and each second device, overall, the first device and each second device in the same region form a federated anti-fraud model. Among them, each device (enterprise or company) is an endpoint in the federation (including horizontal federation and vertical federation), such as Figure 4 shown, the endpoints can be telecommunications enterprises, Internet enterprises, etc. That is, in this embodiment, a large federated anti-fraud model is formed.

[0108] Step S033: Receive the third target model parameters sent by the third device, replace and update the first target model parameters with the third target model parameters, and obtain the preset to-be-trained prediction model after replacement and update.

[0109] For the first device, receive the third target model parameters sent by the third device, replace and update the first target model parameters with the third target model parameters, and obtain the preset to-be-trained prediction model after replacement and update until a preset federated anti-fraud model is obtained.

[0110] It should be noted that after the preset federal anti-fraud model is established, the first device distributes the preset federal anti-fraud model to the user side for fraud identification.

[0111] Step S20: Perform fraud identification processing on the to-be-processed data based on the preset federal anti-fraud model to obtain a prediction result.

[0112] In this embodiment, the preset federal anti-fraud model can also be updated according to the to-be-processed data and the prediction result to achieve closed-loop training.

[0113] In this embodiment, after the to-be-processed data is input into the preset federal anti-fraud model, fraud identification processing is performed on the to-be-processed data based on the preset federal anti-fraud model to obtain a prediction result. The prediction result includes that the corresponding result of the to-be-processed data is fraud or non-fraud.

[0114] This application obtains to-be-processed data and inputs the to-be-processed data into a preset federal anti-fraud model; wherein, the preset federal anti-fraud model is a target model obtained by performing iterative training on a preset to-be-trained prediction model through executing a preset federal process based on preset fraud-associated data with preset labels. Fraud identification processing is performed on the to-be-processed data based on the preset federal anti-fraud model to obtain a prediction result. In this application, after obtaining the to-be-processed data, fraud identification processing is performed on the to-be-processed data based on the preset federal anti-fraud model to obtain a prediction result, and the preset federal anti-fraud model is a target model obtained by performing iterative training on a preset to-be-trained prediction model through executing a preset federal process (data does not move while the model moves) based on preset fraud-associated data with preset labels. That is, while protecting data privacy, data does not move while the model moves through the federal process, changing small data island analysis into big data analysis, improving the preventability of fraud, and solving the technical problem in the prior art that fraud data is held by different enterprises or government agencies, reducing the preventability of fraud.

[0115] Further, based on the first embodiment in this application, in another embodiment of this application, after the step of performing fraud identification processing on the to-be-processed data based on the preset federal anti-fraud model to obtain a prediction result, the method further includes:

[0116] Step A1: Receive the target contribution degree of the first device calculated by the third device during the process of obtaining the prediction result, where the target contribution degree of the first device is recorded by a preset blockchain.

[0117] Receive the target contribution of the first device calculated by the third device during the process of obtaining the prediction result, where the target contribution of the first device is recorded by a preset blockchain. Specifically, after each prediction result is obtained by the third device, the prediction result is sent to the blockchain for recording, and based on the blockchain, the contribution of each device in the case where the prediction result is obtained is recorded. The contribution of each device in the case where the prediction result is obtained is calculated by the third device. Specifically, after the third target model parameters corresponding to the preset federated anti-fraud model are set in the case where the prediction result is obtained, the third device determines the model change magnitude or parameter change ratio between the third target model parameters corresponding to the preset federated anti-fraud model and the first target parameters corresponding to the case of the prediction result, and determines the target contribution based on the model change magnitude. It should be noted that during the process of obtaining the preset federated anti-fraud model, each time the third target model parameters are obtained, the third target model parameters are sent to the blockchain block for recording. Since the preset federated anti-fraud model is trained and updated regularly, the contribution recorded by the blockchain will also be updated each time it is trained.

[0118] In this embodiment, the third device also determines the contribution of each second device based on the corresponding third target model parameters in the case where the prediction result is obtained. Specifically, the contribution of each second device is determined by the parameter change ratio between the corresponding third target model parameters and the corresponding second target parameters in the case where the prediction result is obtained. After obtaining the target contribution of the first device and the contributions of each second device, the target contribution of the first device and the contributions of each second device are sent to the blockchain for the block record of the blockchain. Since the contributions of each device are recorded by the block record of the blockchain, the contribution ranking cannot be tampered with, and the implementation process is open and transparent. It should be noted that if it is detected in the blockchain record that the contribution of any device is greater than the preset value, the device with the contribution greater than the preset value can be pushed to the client associated with the preset federated anti-fraud model to enhance the credibility of the device with the contribution greater than the preset value.

[0119] After the step of receiving the target contribution of the first device calculated by the third device during the process of obtaining the prediction result, the method further includes:

[0120] Step A2, when the target contribution is greater than the preset contribution, send the target contribution to the client associated with the first device.

[0121] In this embodiment, when the target contribution is greater than the preset contribution, the target contribution is sent to the client associated with the first device to enhance the credibility of the first device.

[0122] This embodiment receives the target contribution degree of the first device in the process of obtaining the prediction result calculated by the third device, where the target contribution degree of the first device is recorded by a preset blockchain. In this embodiment, since the contribution degrees of each device are recorded by the blockchain, the contribution degree calculation process is made public and transparent, which is convenient for the device to conduct image promotion.

[0123] Further, based on the first embodiment and the second embodiment in this application, in another embodiment of this application, the data to be processed includes data of a loan to be processed;

[0124] After the step of performing fraud identification processing on the data to be processed based on the preset federated anti-fraud model to obtain a prediction result, the method further includes:

[0125] Step S40, if the prediction result is that the data of the loan to be processed is not of a fraud type, input the data of the loan to be processed into a preset risk control model;

[0126] In this embodiment, the prediction result is: the data to be processed is of a fraud type (the probability of being of a fraud type is greater than or equal to a preset value), or the data to be processed is not of a fraud type (the probability of being of a fraud type is less than the preset value). If the prediction result is that the data of the loan to be processed is not of a fraud type, input the data of the loan to be processed into a preset risk control model, where the preset risk control model is a model that has been trained.

[0127] Step S50, based on the preset risk control model, score the data of the loan to be processed to obtain a scoring result;

[0128] Step S60, based on the scoring result, output the target loanable numerical range of the data of the loan to be processed.

[0129] Based on the preset risk control model, score the data of the loan to be processed to obtain a scoring result, and based on the scoring result, output the target loanable numerical range of the data of the loan to be processed. Specifically, according to the level or range of the scoring result, and the association relationship between the levels or ranges of each preset scoring result and each loanable numerical range, obtain the target loanable numerical range of the data of the loan to be processed and output it.

[0130] In this embodiment, when the prediction result indicates that the loan data to be processed is not of the fraud type, the loan data to be processed is input into a preset risk control model; based on the preset risk control model, the loan data to be processed is scored to obtain a scoring result; based on the scoring result, the target loanable value range of the loan data to be processed is output. In this embodiment, after performing fraud identification processing on the loan data to be processed based on a preset federal anti-fraud model, the target loanable value range is accurately determined.

[0131] Further, an embodiment of the present application further provides an anti-fraud method, which is applied to a third device. The third device performs preset communication with a first device and a second device respectively. The anti-fraud method includes:

[0132] Step B1, receiving the first target model parameter sent by the first device and receiving the second target model parameter sent by the second device;

[0133] Step B2, combining the first target model parameter and the second target model parameter to calculate a third target model parameter. There is a preset vertical and / or horizontal federal association relationship between the first device and each second device;

[0134] In this embodiment, the anti-fraud method is applied to a third device. The third device performs preset communication with a first device and a second device respectively. In this embodiment, the first target model parameter sent by the first device is received, and the second target model parameter sent by the second device is received to combine the first target model parameter and the second target model parameter to calculate a third target model parameter. Specifically, the first target model parameter and the second target model parameter (one or more) are subjected to mean processing to obtain the third target model parameter. There is a preset vertical and / or horizontal federal association relationship between the first device and each second device. Therefore, overall, through the third device, the first device and the second devices in different industries form a federal anti-fraud model, and through the third device, the first device and the second devices in the same region form a federal anti-fraud model. Each device (enterprise or company) is an endpoint in the federation (including horizontal federation and vertical federation). That is, in this embodiment, a large federal anti-fraud model is formed.

[0135] Step B3, sending the third target model parameter to the first device for the first device to change a preset prediction model to be trained based on the third target model parameter.

[0136] In this embodiment, the third target model parameter is sent to the first device, so that the first device can change the preset prediction model to be trained based on the third target model parameter, so as to accurately obtain the preset federated anti-fraud model of the first device.

[0137] After the first device is communicatively connected to the client, and after the step of sending the third target model parameter to the first device so that the first device can change the preset prediction model to be trained based on the third target model parameter, the method further includes:

[0138] Step C1, when it is detected that the prediction of the data to be processed is successful, determine the target contribution degree of the first device and the contribution degrees of the respective second devices;

[0139] In this embodiment, when it is detected that the prediction of the data to be processed is successful, the target contribution degree of the first device and the contribution degrees of the respective second devices are determined through the third target model parameter. Specifically, the target contribution degree of the first device is determined through the third target model parameter and the first target model parameter, and the contribution degrees of the respective second devices are determined through the third target model parameter and the second target model parameter.

[0140] Step C2, send the target contribution degree of the first device and the contribution degrees of the respective second devices to a block of a preset blockchain, so that the block of the preset blockchain pushes the first device with a contribution degree greater than a preset value to the client.

[0141] In this embodiment, the third device is used to send the target contribution degree of the first device and the contribution degrees of the respective second devices to a block of a preset blockchain, so that the block of the preset blockchain pushes the first device with a contribution degree greater than a preset value to the client, so as to promote the first device with a contribution degree greater than a preset value.

[0142] In this embodiment, the first target model parameter sent by the first device is received, and the second target model parameter of the second device is received; the first target model parameter and the second target model parameter are combined to calculate the third target model parameter, where there is a preset vertical and / or horizontal federated association relationship between the first device and each second device; the third target model parameter is sent to the first device, so that the first device can change the preset prediction model to be trained based on the third target model parameter. This lays a foundation for accurately performing quality inspection voice.

[0143] Refer to Figure 3 , Figure 3 is a schematic diagram of the device structure of the hardware operating environment involved in the solution of the embodiment of the present application.

[0144] Such as Figure 3As shown in the figure, the anti-fraud device may include: a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002. Among them, the communication bus 1002 is used to implement the connection and communication between the processor 1001 and the memory 1005. The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0145] Optionally, the anti-fraud device may further include a rectangular user interface, a network interface, a camera, an RF (Radio Frequency) circuit, sensors, an audio circuit, a WiFi module, etc. The rectangular user interface may include a display screen and an input sub-module such as a keyboard. Optionally, the rectangular user interface may further include a standard wired interface and a wireless interface. The network interface may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0146] Those skilled in the art can understand that Figure 3 the anti-fraud device structure shown in the figure does not constitute a limitation on the anti-fraud device, and may include more or fewer components than shown, or combine some components, or have different component arrangements.

[0147] As Figure 3 shown, the memory 1005 as a storage medium may include an operating system, a network communication module, and an anti-fraud program. The operating system is a program for managing and controlling the hardware and software resources of the anti-fraud device, and supports the operation of the anti-fraud program and other software and / or programs. The network communication module is used to implement the communication between the components inside the memory 1005, as well as the communication with other hardware and software in the anti-fraud system.

[0148] In Figure 3 the anti-fraud device shown, the processor 1001 is used to execute the anti-fraud program stored in the memory 1005 to implement the steps of the anti-fraud method described in any one of the above.

[0149] The specific implementation manner of the anti-fraud device in this application is basically the same as that of each embodiment of the above anti-fraud method, and will not be elaborated here.

[0150] This application also provides an anti-fraud device, a first acquisition module, which is used to acquire data to be processed and input the data to be processed into a preset federated anti-fraud model;

[0151] an identification module, which is used to perform fraud identification processing on the data to be processed based on the preset federated anti-fraud model to obtain a prediction result;

[0152] Among them, the preset federated anti-fraud model is a target model obtained by performing iterative training on a preset prediction model to be trained through a preset federated process based on preset fraud-related data with preset labels.

[0153] Optionally, the anti-fraud device includes:

[0154] An input module, configured to input the to-be-processed loan data into a preset risk control model when the prediction result indicates that the to-be-processed loan data is not of a fraud type;

[0155] A scoring module, configured to score the to-be-processed loan data based on the preset risk control model to obtain a scoring result;

[0156] An output module, configured to output a target loanable numerical range of the to-be-processed loan data based on the scoring result.

[0157] Optionally, the anti-fraud device includes:

[0158] A second acquisition module, configured to acquire preset fraud-related data with preset labels and perform iterative training on the preset prediction model to be trained to train and update the model variables of the preset prediction model to be trained;

[0159] A judgment module, configured to judge whether the to-be-predicted model after iterative training reaches a preset replacement and update condition;

[0160] A third acquisition module, configured to, if the to-be-trained prediction model after iterative training reaches the preset replacement and update condition, perform replacement and update on the model variables of the preset prediction model to be trained that have been trained and updated by executing the preset federated process to obtain the preset prediction model to be trained that has been replaced and updated;

[0161] A fourth acquisition module, configured to continuously perform iterative training and replacement and update on the preset prediction model to be trained that has been replaced and updated until the preset prediction model meets the preset training completion condition to obtain the preset federated anti-fraud model.

[0162] Optionally, the first device performs preset communication with the third device respectively;

[0163] The third acquisition module includes:

[0164] A first acquisition unit, configured to, if the to-be-trained prediction model after iterative training reaches the preset replacement and update condition, acquire first target model parameters of the to-be-trained prediction model that reaches the preset replacement and update condition;

[0165] A sending unit, configured to send the first target model parameters to the third device, so that the third device combines the second target model parameters sent by each second device with the first target model parameters to calculate third target model parameters, where there is a preset vertical and / or horizontal federated association relationship between the first device and each second device;

[0166] A receiving unit, configured to receive the third target model parameters sent by the third device, and replace and update the first target model parameters with the third target model parameters to obtain the preset prediction model to be trained after replacement and update.

[0167] Optionally, the anti-fraud device further includes:

[0168] A contribution degree acquisition module, configured to receive the target contribution degree of the first device in the process of obtaining the prediction result calculated by the third device, where the target contribution degree of the first device is recorded by a preset blockchain.

[0169] Optionally, the anti-fraud device further includes:

[0170] A contribution degree sending module, configured to send the target contribution degree to the client associated with the first device when the target contribution degree is greater than a preset contribution degree.

[0171] The specific implementation manners of the anti-fraud device in this application are basically the same as those of the above anti-fraud method embodiments, and will not be elaborated here.

[0172] This application further provides an anti-fraud device, which is applied to a third device. The third device communicates with a first device and a second device respectively through preset communication. The anti-fraud device further includes:

[0173] A receiving module, configured to receive the first target model parameters sent by the first device and receive the second target model parameters sent by the second device;

[0174] A calculation module, configured to combine the first target model parameters and the second target model parameters to calculate third target model parameters, where there is a preset vertical and / or horizontal federated association relationship between the first device and each second device;

[0175] An update module, configured to send the third target model parameters to the first device for the first device to update the preset prediction model to be trained based on the third target model parameters.

[0176] Optionally, the anti-fraud device further includes:

[0177] A determination module, when it is detected that the prediction of the data to be processed is successful, determines the target contribution degree of the first device and the contribution degrees of the respective second devices;

[0178] A push module, configured to send the target contribution degree of the first device and the contribution degrees of the respective second devices to a block of a preset blockchain, so that the block of the preset blockchain pushes the first device with a contribution degree greater than a preset value to a client.

[0179] The specific implementation manner of the anti-fraud device in this application is basically the same as that of the various embodiments of the above anti-fraud method, and will not be elaborated here.

[0180] This application also provides an anti-fraud system, which includes the anti-fraud device for the first device described above, and a third device. The third device performs preset communication with the first device and the second device respectively, and the third device is used to implement:

[0181] Receiving the first target model parameter sent by the first device, and receiving the second target model parameter of the second device;

[0182] Combining the first target model parameter and the second target model parameter to calculate a third target model parameter, where there is a preset vertical and / or horizontal federated association relationship between the first device and the respective second devices;

[0183] Sending the third target model parameter to the first device for the first device to change a preset model to be trained and predicted based on the third target model parameter.

[0184] The specific implementation manner of the anti-fraud system in this application is basically the same as that of the various embodiments of the above anti-fraud method, and will not be elaborated here.

[0185] An embodiment of this application provides a storage medium, and the storage medium stores one or more programs, and the one or more programs can also be executed by one or more processors to implement the steps of the anti-fraud method described in any one of the above.

[0186] The specific implementation manner of the storage medium in this application is basically the same as that of the various embodiments of the above anti-fraud method, and will not be elaborated here.

[0187] The above are only the preferred embodiments of this application, and do not limit the patent scope of this application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of this application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent scope of this application.

Claims

1. A fraud prevention method, characterized in that, the fraud prevention method is applied to a first device, and the fraud prevention method includes: obtaining data to be processed, and inputting the data to be processed into a preset federated fraud prevention model; wherein, the preset federated fraud prevention model is a target model obtained by performing iterative training on a preset prediction model to be trained based on preset fraud-related data with preset labels through executing a preset federated process; performing fraud identification processing on the data to be processed based on the preset federated fraud prevention model to obtain a prediction result; after the step of performing fraud identification processing on the data to be processed based on the preset federated fraud prevention model to obtain a prediction result, the method further includes: receiving the target contribution degree of the first device calculated by a third device in the process of obtaining the prediction result, wherein the target contribution degree of the first device is recorded by a preset blockchain, and wherein, after the third device obtains the corresponding third target model parameters of the preset federated fraud prevention model in the case of obtaining the prediction result, the third device determines the model change magnitude or parameter change ratio between the corresponding third target model parameters of the preset federated fraud prevention model and the corresponding first target parameters in the case of the prediction result, and determines the target contribution degree based on the model change magnitude. The third device determines the contribution degrees of each second device by the parameter change ratio between the corresponding third target model parameters and the corresponding second target parameters in the case of obtaining the prediction result. After obtaining the target contribution degree of the first device and the contribution degrees of each second device, the third device sends the target contribution degree of the first device and the contribution degrees of each second device to the blockchain for the block of the blockchain to record the contribution degrees of each device due to recording the contribution degrees of each device by the block of the blockchain.

2. The fraud prevention method according to claim 1, characterized in that, the data to be processed includes data to be processed for a loan; after the step of performing fraud identification processing on the data to be processed based on the preset federated fraud prevention model to obtain a prediction result, the method further includes: if the prediction result is that the data to be processed for a loan is not of a fraud type, inputting the data to be processed for a loan into a preset risk control model; scoring the data to be processed for a loan based on the preset risk control model to obtain a scoring result; outputting a target loanable numerical range of the data to be processed for a loan based on the scoring result.

3. The fraud prevention method according to claim 1, characterized in that, before the step of inputting the data to be processed into the preset federated fraud prevention model, the method further includes: obtaining preset fraud-related data with preset labels, and performing iterative training on the preset prediction model to be trained to train and update the model variables of the preset prediction model to be trained; judging whether the preset prediction model to be predicted after iterative training reaches a preset replacement and update condition; if the preset prediction model to be trained after iterative training reaches the preset replacement and update condition, then by executing the preset federated process, replacing and updating the model variables of the preset prediction model to be trained that has been trained and updated to obtain the replaced and updated preset prediction model to be trained. Continuously perform iterative training and replacement update on the preset to-be-trained prediction model that is replaced and updated until the preset to-be-trained model meets the preset training completion condition, and obtain the preset federated anti-fraud model.

4. The anti-fraud method according to claim 3, wherein, the first device performs preset communication with the third device respectively; the step of, if the to-be-trained prediction model after the iterative training reaches the preset replacement update condition, then replacing and updating the model variables of the preset to-be-trained prediction model that is trained and updated by executing the preset federated process, and obtaining the preset to-be-trained prediction model that is replaced and updated, includes: if the to-be-trained prediction model after the iterative training reaches the preset replacement update condition, then obtain the first target model parameters of the to-be-trained prediction model that reaches the preset replacement update condition; send the first target model parameters to the third device for the third device to calculate the third target model parameters by combining the second target model parameters sent by each second device and the first target model parameters, wherein there is a preset vertical and / or horizontal federated association relationship between the first device and each second device; receive the third target model parameters sent by the third device, and replace and update the first target model parameters with the third target model parameters to obtain the preset to-be-trained prediction model that is replaced and updated.

5. The anti-fraud method according to claim 1, wherein, after the step of receiving the target contribution degree of the first device calculated by the third device in the process of obtaining the prediction result, the method further includes: when the target contribution degree is greater than the preset contribution degree, send the target contribution degree to the client associated with the first device.

6. An anti-fraud method, wherein, the anti-fraud method is applied to the third device, the third device performs preset communication with the first device and the second device respectively, and through the third device, the first device and the second devices in different industries form a federated anti-fraud model, wherein each device is an endpoint in the federation, and the anti-fraud method includes: receive the first target model parameters sent by the first device and receive the second target model parameters of the second device; combine the first target model parameters and the second target model parameters to calculate the third target model parameters, wherein there is a preset vertical and / or horizontal federated association relationship between the first device and each second device; send the third target model parameters to the first device for the first device to change the preset to-be-trained prediction model based on the third target model parameters, and the changed preset to-be-trained prediction model is used to form a preset federated anti-fraud model for the first device to process the to-be-processed data based on the preset federated anti-fraud model.

7. The anti-fraud method according to claim 6, wherein, After the step of communicating and connecting the first device with the client and sending the third target model parameters to the first device for the first device to change a preset prediction model to be trained based on the third target model parameters, the method further includes: When it is detected that the prediction of the data to be processed is successful, determining the target contribution degree of the first device and the contribution degrees of the second devices; Sending the target contribution degree of the first device and the contribution degrees of the second devices to a block of a preset blockchain, so that the block of the preset blockchain pushes the first device with a contribution degree greater than a preset value to the client.

8. A fraud prevention device Characterized in that The fraud prevention device is applied to a first device, and the fraud prevention device includes: A first acquisition module, configured to acquire data to be processed and input the data to be processed into a preset federated anti-fraud model; An identification module, configured to perform fraud identification processing on the data to be processed based on the preset federated anti-fraud model to obtain a prediction result; Wherein, the preset federated anti-fraud model is a target model obtained by performing iterative training on a preset prediction model to be trained by executing a preset federated process based on preset fraud-related data with preset labels; The fraud prevention device is used to implement: Receiving the target contribution degree of the first device in the process of obtaining the prediction result calculated by a third device, where the target contribution degree of the first device is recorded by a preset blockchain. After the third device determines the third target model parameters corresponding to the preset federated anti-fraud model in the case of obtaining the prediction result, the third device determines the model change magnitude or parameter change ratio between the third target model parameters corresponding to the preset federated anti-fraud model and the first target parameters corresponding to the case of the prediction result, and determines the target contribution degree based on the model change magnitude. The third device determines the contribution degrees of the second devices by the parameter change ratio between the third target model parameters and the corresponding second target parameters corresponding to the case of obtaining the prediction result. After obtaining the target contribution degree of the first device and the contribution degrees of the second devices, the third device sends the target contribution degree of the first device and the contribution degrees of the second devices to the blockchain for the block of the blockchain to record the contribution degrees of each device.

9. A fraud prevention system Characterized in that The fraud prevention system includes the fraud prevention device according to claim 8, and a third device. The third device is respectively in preset communication with the first device and the second device, and the third device is used to implement: A receiving module, configured to receive the first target model parameters sent by the first device and receive the second target model parameters of the second device; A calculation module, configured to jointly calculate the first target model parameters and the second target model parameters to obtain third target model parameters, where there is a preset vertical and / or horizontal federated association relationship between the first device and the second devices; A change module, configured to send the third target model parameter to the first device, so that the first device changes a preset prediction model to be trained based on the third target model parameter, and the changed preset prediction model to be trained is used to form a preset federated anti-fraud model, so that the first device processes data to be processed based on the preset federated anti-fraud model.

10. An anti-fraud device, characterized in that the anti-fraud device includes: a memory, a processor, and a program stored on the memory for implementing the anti-fraud method, the memory is used for storing the program for implementing the anti-fraud method; the processor is used for executing the program for implementing the anti-fraud method to implement the steps of the anti-fraud method according to any one of claims 1 to 7.

11. A storage medium, characterized in that a program for implementing the anti-fraud method is stored on the storage medium, and the program for implementing the anti-fraud method is executed by a processor to implement the steps of the anti-fraud method according to any one of claims 1 to 7.

Images