Training methods and apparatus for data recognition models, and data processing methods and apparatus

Abstract

The present invention discloses a training method and device for a data recognition model, and a data processing method and device. The training method for a data recognition model includes: obtaining sample data of a user, the sample data pointing to features with financial fraud risks; based on the sample data, using a first machine learning model to perform data disguise training to obtain disguised data corresponding to the sample data; based on the disguised data, using a second machine learning model to perform data recognition training to obtain an objective function; adjusting a first model parameter of the first machine learning model and a second model parameter of the second machine learning model to optimize the value of the objective function to a preset value; using the second machine learning model corresponding to the current second model parameter as a data recognition model. When using the data recognition model for data processing, the recognition and adaptability of disguised features with financial fraud risks can be improved, thereby greatly improving the reliability and effectiveness of anti-financial fraud.

Description

Technical Field

[0001] The present invention relates to the field of Internet communication technology, and in particular to a training method and device for a data recognition model, and a data processing method and device. Background Art

[0002] With the rapid development of computer and Internet technology, traditional industries are also developing rapidly in combination with them. Internet financial products are increasingly used in people's daily lives. While lowering the threshold for use and bringing convenience to users, they also face the risk of financial fraud.

[0003] Fraudsters, for the purpose of illegal possession, fabricate facts or conceal the truth to defraud banks or other financial institutions of loans, insurance money, etc. Once financial fraud occurs, it is difficult to recover the property. Fraudsters often conduct corresponding research on anti-fraud methods. Therefore, it is necessary to provide more reliable or more effective solutions to accurately identify the characteristics of financial fraud risks. Summary of the invention

[0004] In order to solve the problem that the existing technology has low accuracy in identifying features with financial fraud risks, the present invention provides a data recognition model training method and device, and a data processing method and device:

[0005] In one aspect, the present invention provides a method for training a data recognition model, the method comprising:

[0006] Acquire sample data of the user, wherein the sample data indicates characteristics of financial fraud risk;

[0007] Based on the sample data, use a first machine learning model to perform data camouflage training to obtain camouflaged data corresponding to the sample data;

[0008] Based on the disguised data, use a second machine learning model to perform data recognition training to obtain a target function;

[0009] Adjusting a first model parameter of the first machine learning model and a second model parameter of the second machine learning model to optimize the value of the objective function to a preset value;

[0010] Using the second machine learning model corresponding to the current second model parameter as the data recognition model;

[0011] Among them, the first machine learning model and the second machine learning model are established based on a generative adversarial network.

[0012] Another aspect provides a data processing method, the method comprising:

[0013] Obtain the user's data to be identified;

[0014] Inputting the data to be identified into a data identification model for data identification processing, and identifying the characteristics of the data to be identified that have financial fraud risks;

[0015] The data recognition model is determined by machine learning training based on the user's sample data and disguised data, and the data to be recognized, the sample data, and the disguised data belong to the same type of data.

[0016] Another aspect provides a training device for a data recognition model, the device comprising:

[0017] Sample data acquisition module: used to acquire sample data of users, where the sample data points to features with financial fraud risks;

[0018] A data disguise training module: used to perform data disguise training based on the sample data using a first machine learning model to obtain disguised data corresponding to the sample data;

[0019] A data recognition training module: used to perform data recognition training based on the disguised data using a second machine learning model to obtain a target function;

[0020] Model parameter adjustment module: used to adjust the first model parameters of the first machine learning model and the second model parameters of the second machine learning model, optimize the value of the objective function to a preset value; use the second machine learning model corresponding to the current second model parameter as the data recognition model;

[0021] Among them, the first machine learning model and the second machine learning model are established based on a generative adversarial network.

[0022] Another aspect provides a data processing device, the device comprising:

[0023] To-be-identified data acquisition module: used to acquire the user's to-be-identified data;

[0024] Data identification processing module: used for inputting the data to be identified into the data identification model for data identification processing, and identifying the characteristics of the data to be identified that have financial fraud risks;

[0025] The data recognition model is determined by machine learning training based on the user's sample data and disguised data, and the data to be recognized, the sample data, and the disguised data belong to the same type of data.

[0026] On the other hand, a training server is provided, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the training method of the data recognition model as described above.

[0027] On the other hand, a computer-readable storage medium is provided, in which at least one instruction, at least one program, a code set or an instruction set is stored, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the training method of the data recognition model as described above.

[0028] On the other hand, a data processing server is provided, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the data processing method as described above.

[0029] On the other hand, a computer-readable storage medium is provided, in which at least one instruction, at least one program, a code set or an instruction set is stored, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the data processing method as described above.

[0030] The data recognition model training method and device, and data processing method and device provided by the present invention have the following technical effects:

[0031] The present invention utilizes a first machine learning model to disguise sample data, and a second machine learning model to identify the disguised data. Through adversarial learning of the two machine learning models, a data recognition model with high generalization ability is obtained. When the data recognition model is used for data processing, the recognition and adaptability of disguised features with financial fraud risks can be improved, thereby greatly improving the reliability and effectiveness of anti-financial fraud. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 It is a flowchart of a method for training a data recognition model provided by an embodiment of the present invention;

[0034] Figure 2 It is a flow chart of obtaining a target function by performing data recognition training using a second machine learning model based on the disguised data provided by an embodiment of the present invention;

[0035] Figure 3 It is a flow chart of adjusting the first model parameter and the second model parameter to optimize the objective function to a preset value provided by an embodiment of the present invention;

[0036] Figure 4 is a schematic diagram of an application scenario of a data recognition model provided by an embodiment of the present invention;

[0037] Figure 5 is a schematic diagram of a machine learning model provided by an embodiment of the present invention;

[0038] Figure 6 It is a flowchart of a data processing method provided by an embodiment of the present invention;

[0039] Figure 7 It is a block diagram of a data recognition model training device provided by an embodiment of the present invention;

[0040] Figure 8 is a block diagram of a data processing device provided by an embodiment of the present invention;

[0041] Fig. 9 It is a structural diagram of a server provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0043] It should be noted that the terms "including" and "having" and any variations thereof in the specification and claims of the present invention and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0044] At present, in some application scenarios, some users may disguise the situation of the insurance subject in order to successfully purchase insurance or buy insurance at a low price. For example, when purchasing car insurance, they disguise the data of the number of accidents and violations in the previous year, and when purchasing disease insurance, they disguise the data of family disease history. In order to successfully obtain a loan, some users may disguise the relevant credit data. There are many ways for fraudsters to commit financial fraud. The existing financial fraud cases as modeling samples are not enough to support complex models. A single financial fraud case (such as overdue loans or malicious insurance fraud) often brings tens of thousands or even hundreds of thousands of losses. The labeling cost of features with financial fraud risks is high and there are errors. Accumulating samples through online experiments is inefficient and costly. Fraudsters often conduct corresponding research on anti-fraud methods. Fraudsters disguise features with financial fraud risks faster. Therefore, the existing data recognition model cannot accurately identify the features with financial fraud risks disguised by fraudsters, and has poor adaptability to the recognition of disguised features with financial fraud risks, resulting in the inability to reliably and effectively reduce the occurrence of financial fraud in practical applications.

[0045] In response to the problems existing in the above-mentioned method of identifying features that are disguised and carry financial fraud risks, a machine learning training data recognition model can be used. Machine learning training can be used to obtain a data recognition model with high generalization ability. When the data recognition model is used for data processing, the recognition and adaptability of disguised features that carry financial fraud risks can be improved, thereby greatly improving the reliability and effectiveness of anti-financial fraud.

[0046] The following is a detailed introduction to the implementation process of the neural network machine learning training data recognition model:

[0047] Figure 1 It is a flowchart of a method for training a data recognition model provided by an embodiment of the present invention. This specification provides method operation steps as described in the embodiment or flowchart, but may include more or fewer operation steps based on conventional or non-creative labor. The order of steps listed in the embodiment is only one way of executing the steps among many orders, and does not represent the only order of execution. When the actual system or server product is executed, it can be executed in sequence or in parallel (for example, in a parallel processor or multi-threaded processing environment) according to the method shown in the embodiment or the accompanying drawings. Specifically, Figure 1 As shown, the method may include:

[0048] S101: Acquire sample data of a user, where the sample data indicates features with financial fraud risk;

[0049] In an embodiment of the present invention, before performing neural network machine learning, training data may be determined first, and sample data with features indicating financial fraud risks may be obtained as training data.

[0050] In actual applications, users can be randomly selected from a database (such as WeChat's database) to obtain sample data of users at the current time t. Sample data includes but is not limited to basic information such as age, gender, living area, occupation, and WeChat platform data such as public account reading and social networking. For example, if the sample data corresponding to user i is x i , x i Can be a vector of length K.

[0051] S102: Based on the sample data, use a first machine learning model to perform data disguise training to obtain disguised data corresponding to the sample data;

[0052] In an embodiment of the present invention, the sample data is input into the first machine learning model; the sample data is combined with random data using the first machine learning model to generate the disguised data. The first machine learning model can disguise the sample data without relying on the collection, experimentation and accumulation of financial fraud cases. The first machine learning model is used as a neural network A(x i , W A ), by disguising the sample data in order to deceive the second machine learning model, W A is the first model parameter of the first machine learning model.

[0053] Specifically, according to x i ′=A(x i , W A ), the sample data x is transformed into i =(x i，0 , x i，1 , ..., x i，K ) and random data u i =(u i，0 ,u i，1 , ..., u i，M ) to generate the corresponding disguised data x i ′=(x i，0 ′,x i，1 ′,...,x i，K ′). When random data is introduced, the random data corresponding to the sample data of each user is different, and the random data corresponding to the sample data of the same user at different times is also different, so as to avoid the similarity of the generated disguised data.

[0054] In actual applications, the first machine learning model disguises features that have financial fraud risks. For example, the number of extreme sports articles read by a user is an average of two per day, or x. i，j =2, the number of readings of the two articles is much higher than the average level. The data recognition model may therefore regard the user as a high-risk sports enthusiast and refuse to provide insurance or loan. The first machine learning model can generate random data u between -0.25 and 0.25. i，M x i，j =2 disguised as x i，j ′=0.75+u i，M , so disguised data x i，j ′ corresponds to a number between 0.5 and 1, thereby causing the data recognition model to misjudge.

[0055] S103: Based on the disguised data, use a second machine learning model to perform data recognition training to obtain a target function;

[0056] like Figure 2 As shown, the step of performing data recognition training using a second machine learning model based on the disguised data to obtain an objective function includes:

[0057] S201: Inputting the disguised data into the second machine learning model;

[0058] In an embodiment of the present invention, the first machine learning model and the second machine learning model are established based on a generative adversarial network (GAN). The second machine learning model is a neural network D(x i , W D ), identify the disguised data, W D is the second model parameter of the second machine learning model. The second machine learning model can identify that the disguised data has the risk of financial fraud by restoring the relevant features in the sample data (for example, Class A auto insurance requires that the number of accidents in the previous year is less than or equal to 2, and the first machine learning model disguises the real number of accidents 3 as 2. At this time, the second machine learning model restores the disguised number of accidents 2 to 3, thereby achieving recognition), or it can be based on feature B in the disguised data or other features of the user to determine that feature C in the disguised data has a certain degree of financial fraud risk, and of course it can also be based on other methods.

[0059] Specifically, the second machine learning model performs data recognition training based on a binary classification network algorithm. The corresponding neural network can use DNN (Deep Neural Network), RNN (Recurrent Neural Network), Wide & Deep learning, etc. The second machine learning model performs D(x i ′,W D )=0 or 1.

[0060] S202: When the second machine learning model identifies that the disguised data has characteristics of financial fraud risk, a first identification result is obtained;

[0061] At this time, the first recognition result is obtained

[0062] S203: When the second machine learning model cannot identify that the disguised data has the characteristics of financial fraud risk, a second identification result is obtained;

[0063] At this time, the second recognition result is obtained

[0064] S204: Generate the objective function according to the first recognition result and the second recognition result.

[0065] In the embodiment of the present invention, starting from time t, B sample data are extracted from the current time each time. Then, a first loss function is constructed based on the first recognition result: l(0, D(x i ′,W D )); Based on the second recognition result, a second loss function is constructed: l(1, D(A(x i , W A ), W D )); through the formula: Get the objective function; where x i is the i-th sample data, x i ′ is the disguised data corresponding to the i-th sample data, B is the number of sample data within time t, D(x i ′,W D ) indicates that the second machine learning model cannot identify the characteristics of the disguised data as having financial fraud risk, D(A(x i , W A ), W D ) indicates that the second machine learning model identifies that the disguised data has the characteristics of financial fraud risk. According to the objective function, the value of the objective function increases or decreases depending on whether the second machine learning model identifies that the disguised data has the characteristics of financial fraud risk.

[0066] S104: Adjust a first model parameter of the first machine learning model and a second model parameter of the second machine learning model to optimize the value of the objective function to a preset value;

[0067] In the embodiment of the present invention, when the machine learning model is initialized, the value of the objective function Q is 0, and the first model parameter W A The second model parameter W is randomly generated in the form of normal distribution, uniform distribution or any artificially specified distribution; D Randomly generated in the form of normal distribution, uniform distribution or any artificial distribution. Figure 5 As shown, the first machine learning model and the second machine learning model tune their respective models through reinforcement adversarial learning.

[0068] Specifically, Figure 3 As shown, the step of adjusting the first model parameter of the first machine learning model and the second model parameter of the second machine learning model to optimize the value of the objective function to a preset value includes:

[0069] S301: updating the value of the second model parameter based on the value of the first model parameter updated in the previous iteration and the objective function;

[0070] Calculate the value of the objective function Q within time t, keeping the first model parameter W A No change, by using Q to W D Take partial derivative and update the current second model parameter W D = Previous second model parameter ε D As the step size, its value can be manually set according to the actual situation, because if the step size is too small, it will slow down the execution speed of the iteration, while if the step size is too large, it may take a detour when descending or accidentally skip the optimal solution.

[0071] S302: updating the value of the first model parameter based on the value of the second model parameter updated in the previous iteration and the objective function;

[0072] Using the second model parameter W updated in S301 D , and then calculate the value of the objective function Q within time t, keeping the second model parameter W D No change, by using Q to W A Take partial derivative and update the current first model parameter W A = Previous second model parameter ε A As the step size, its value can be manually set according to the actual situation, because if the step size is too small, it will slow down the execution speed of the iteration, while if the step size is too large, it may take a detour when descending or accidentally skip the optimal solution.

[0073] S303: Iterate along the gradient descent direction of the objective function, and repeat the above two steps until the value of the objective function reaches a preset value.

[0074] Repeat the above two steps S301 and S302, updating W each time D and W A When the value of the objective function reaches the preset value, the value of the objective function tends to be stable, W D The value of W A The value of also tends to be stable, and the adversarial capabilities of the first machine learning model and the second machine learning model are stable (converging).

[0075] For the first model parameter W A and the second model parameter W D The hyperparameters of the model involved in the adjustment can be determined by cross-validation, train-test-split, etc. Hyperparameters include the learning rate of gradient descent, the number of iterations, and the learning step of Q-Learning. The learning rate of gradient descent can be set to 0.0005 to 0.001, and the number of iterations can be set to 50 to 100.

[0076] S105: Use the second machine learning model corresponding to the current second model parameters as the data recognition model.

[0077] like Figure 4 As shown, Figure 4 It is a schematic diagram of an application scenario of a data recognition model provided by an embodiment of the present invention. Figure 4 The training data is sample data that points to features with financial fraud risks. Correspondingly, the data recognition model trained subsequently can identify features with financial fraud risks in the data to be identified (the data to be identified and the sample data and the disguised data are of the same type, for example, they are all vectors of a certain length).

[0078] In addition, it should be noted that the machine learning model described in the embodiment of the present invention is not limited to the above-mentioned neural network machine learning model, and may also include a BP (back propagation) neural network machine learning model, a Bayesian neural network machine learning model, etc. In practical applications, other machine learning models may also be included, such as a degenerate regression machine learning model, a decision tree machine learning model, etc. The embodiment of the present invention is not limited to the above.

[0079] It can be seen from the technical solutions provided in the above embodiments of this specification that the first machine learning model and the second machine learning model in the embodiments of this specification optimize their respective models through adversarial learning similar to playing chess and playing games. The first machine learning model can break through the defense of the second machine learning model for identification to the greatest extent through camouflage, and the first machine learning model generates nearly perfect camouflage data. At the same time, the second machine learning model can identify the difference between sample data and camouflage data to the greatest extent. The embodiments of this specification do not require supervised learning, and the camouflage data can be generated by the first machine learning model, and no additional data collection is required. At the same time, the two machine learning models learn online, and the models and model parameters can be updated at any time, and they can be used in real time for anti-financial fraud.

[0080] Based on the above data recognition model, a specific embodiment of a data processing method of this specification is introduced below. Figure 6 It is a flowchart of a data processing method provided by an embodiment of the present invention. The present invention provides method operation steps as described in the embodiment or flowchart, but may include more or fewer operation steps based on conventional or non-creative labor. The order of steps listed in the embodiment is only one way of executing the steps among many orders, and does not represent the only order of execution. When the actual system or client product is executed, it can be executed in sequence or in parallel according to the method shown in the embodiment or the accompanying drawings (for example, in a parallel processor or multi-threaded processing environment). Specifically, Figure 6 As shown, the method may include:

[0081] S601: Obtaining the user's data to be identified;

[0082] In actual applications, the data to be identified may come from users who are currently applying for insurance or loans, and the data to be identified, the sample data, and the disguised data belong to the same type of data. The data to be identified includes but is not limited to basic information such as age, gender, living area, occupation, and WeChat platform data such as public account reading and social networking. The data to be identified corresponding to user i can be x i and x i ′,x i and x i ′ can be a vector of length K.

[0083] S602: Inputting the data to be identified into a data identification model for data identification processing to identify the features of the data to be identified that have financial fraud risks;

[0084] In an embodiment of the present invention, the data recognition model is determined by machine learning training based on the user's sample data and disguised data. The data recognition model identifies that the characteristics of the data to be identified as having financial fraud risks can be the restoration of relevant characteristics in the data to be identified, or it can be based on feature D in the data to be identified or other characteristics of the user to determine that feature E in the data to be identified has a certain degree of financial fraud risk. Of course, it can also be based on other methods.

[0085] In practical applications, taking insurance risk control as an example, generally speaking, men, older people, and people with low education levels are more likely to engage in high-risk occupations (such as machine operators and high-altitude workers), and there is also a higher probability of accidents in accident insurance. In order to buy insurance smoothly or at a low price, some of these people may modify their basic information to be unemployed or farmers. The data to be identified obtained by the data recognition model = (male, 55 years old, low education, farmer, no regular work wifi, agricultural products circle of friends reading = 0, ...), the data recognition model determines that the data to be identified is significantly different from the usual "farmers", and the data recognition model identifies the characteristics of the data to be identified that have financial fraud risks, and then the insurance company refuses to insure.

[0086] Taking the credit investigation business as an example, if a loan customer claims that he has many friends, and disguises the feature of the number of friends, the data recognition model obtains the data to be identified, in which the average payment frequency of the loan customer's friends is less than once a week. If the average payment frequency of friends of normal users' valid friends is 3 times a week, the data recognition model determines that the "number of friends" in the data to be identified is significantly different from the usual "number of friends". The data recognition model identifies that the data to be identified points to features with financial fraud risks, and the bank then refuses to lend.

[0087] It can be seen from the technical solutions provided in the above embodiments of this specification that in the embodiments of this specification, the relevant data of each user requesting financial services is identified through a data recognition model, and it is determined whether the user is a fraudster, and if so, the service is denied.

[0088] The embodiment of the present invention also provides a training device for a data recognition model, such as Figure 7 As shown, the device comprises:

[0089] Sample data acquisition module 71: used to acquire sample data of a user, where the sample data points to features with financial fraud risks;

[0090] A data disguise training module 72 is configured to perform data disguise training based on the sample data using a first machine learning model to obtain disguised data corresponding to the sample data;

[0091] Data recognition training module 73: used to perform data recognition training based on the disguised data using a second machine learning model to obtain a target function;

[0092] Model parameter adjustment module 74: used to adjust the first model parameters of the first machine learning model and the second model parameters of the second machine learning model, optimize the value of the objective function to a preset value; and use the second machine learning model corresponding to the current second model parameters as the data recognition model.

[0093] It should be noted that the device in the device embodiment and the method embodiment are based on the same inventive concept.

[0094] The embodiment of the present invention also provides a data processing device, such as Figure 8 As shown, the device comprises:

[0095] The to-be-identified data acquisition module 81 is used to acquire the to-be-identified data of the user;

[0096] Data identification processing module 82: used for inputting the data to be identified into the data identification model for data identification processing, and identifying the characteristics of the data to be identified that have financial fraud risks;

[0097] Among them, the data recognition model is determined by machine learning training based on the user's sample data and disguised data, the sample data points to features with financial fraud risks, and the data to be identified and the disguised data belong to the same type of data.

[0098] It should be noted that the device in the device embodiment and the method embodiment are based on the same inventive concept.

[0099] An embodiment of the present invention provides a data processing server, which includes a processor and a memory, in which at least one instruction, at least one program, a code set or an instruction set is stored. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the data processing method provided in the above method embodiment.

[0100] The memory can be used to store software programs and modules. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, application programs required for functions, etc.; the data storage area can store data created according to the use of the device, etc. In addition, the memory can include a high-speed random access memory and can also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices. Accordingly, the memory can also include a memory controller to provide the processor with access to the memory.

[0101] The embodiment of the present invention also provides a schematic diagram of the structure of a server, see Fig. 9 , the server 900 is used to implement the data processing method provided in the above embodiment. Specifically, the server structure may include the above device. The server 900 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 910 (for example, one or more processors) and a memory 930, and one or more storage media 920 (for example, one or more mass storage devices) storing application programs 923 or data 922. Among them, the memory 930 and the storage medium 920 can be short-term storage or permanent storage. The program stored in the storage medium 920 may include one or more modules, each of which may include a series of instruction operations in the server. Furthermore, the central processing unit 910 may be configured to communicate with the storage medium 920 and execute a series of instruction operations in the storage medium 920 on the server 900. The server 900 may also include one or more power supplies 960, one or more wired or wireless network interfaces 950, one or more input and output interfaces 940, and / or one or more operating systems 921, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.

[0102] An embodiment of the present invention also provides a storage medium, which can be set in a server to store at least one instruction, at least one program, a code set or an instruction set related to a data processing method in a method embodiment. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the data processing method provided by the above method embodiment.

[0103] Optionally, in this embodiment, the storage medium may be located in at least one of the multiple network servers of the computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disk.

[0104] An embodiment of the present invention provides a training server, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the training method of the data recognition model provided by the above method embodiment.

[0105] An embodiment of the present invention also provides a storage medium, in which at least one instruction, at least one program, a code set or an instruction set is stored. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the training method of the data recognition model provided by the above method embodiment.

[0106] It should be noted that the sequence of the embodiments of the present invention described above is for description only and does not represent the advantages and disadvantages of the embodiments. The above describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0107] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system and server embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0108] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0109] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for training a data recognition model, characterized in that: The method comprises: Acquire sample data of the user, wherein the sample data has characteristics of financial fraud risk; Based on the sample data, use a first machine learning model to perform data camouflage training to obtain camouflaged data corresponding to the sample data; Based on the disguised data, using a second machine learning model to perform data recognition training to obtain an objective function, wherein the first machine learning model and the second machine learning model are established based on a generative adversarial network; Adjusting a first model parameter of the first machine learning model and a second model parameter of the second machine learning model to optimize the value of the objective function to a preset value; Using the second machine learning model corresponding to the current second model parameter as the data recognition model; The step of performing data disguise training based on the sample data using a first machine learning model to obtain disguised data corresponding to the sample data includes: inputting the sample data into the first machine learning model; combining the sample data with random data using the first machine learning model to generate the disguised data; The method of using a second machine learning model to perform data recognition training based on the disguised data to obtain a target function includes: inputting the disguised data into the second machine learning model; obtaining a first recognition result when the second machine learning model recognizes that the disguised data has the characteristics of financial fraud risk; obtaining a second recognition result when the second machine learning model cannot recognize that the disguised data has the characteristics of financial fraud risk; and generating the target function according to the first recognition result and the second recognition result.

2. The method according to claim 1, characterized in that The adjusting the model parameters of the first machine learning model and the model parameters of the second machine learning model to optimize the objective function to a preset value includes: Based on the values ​​of the first model parameters updated in the previous iteration and the objective function, updating the values ​​of the second model parameters; Based on the values ​​of the second model parameters updated in the previous iteration and the objective function, updating the values ​​of the first model parameters; Iterate along the gradient descent direction of the objective function, and repeat the above two steps until the value of the objective function reaches a preset value.

3. The method according to claim 1, characterized in that The generating the objective function according to the first recognition result and the second recognition result includes: Based on the first recognition result, a first loss function is constructed: l(0, D(x′ i , W D )); A second loss function is constructed based on the second recognition result: l(1, D(A(x i , W A ), W D )); By formula: Obtaining the objective function; Among them, x i is the i-th sample data, x′ i is the disguised data corresponding to the i-th sample data, B is the number of sample data within time t, W A is the first model parameter, W D is the second model parameter, D(x′ i , W D ) indicates that the second machine learning model cannot identify the characteristics of the disguised data as having financial fraud risk, D(A(x i , W A ), W D ) indicates that the second machine learning model recognizes that the disguised data has characteristics of financial fraud risk.

4. The method according to claim 1, characterized in that: Set the learning rate of gradient descent to 0.0005 to 0.001 and the number of iterations to 50 to 100.

5. The method according to claim 1, characterized in that: The second machine learning model performs data recognition training based on a binary classification discriminant network algorithm.

6. A data processing method, characterized in that: The method comprises: Obtain the user's data to be identified; The data to be identified is input into a data identification model for data identification processing to identify whether the data to be identified has characteristics of financial fraud risk, and the data identification model is obtained based on the training method of the data identification model as described in any one of claims 1 to 5.

7. A training device for a data recognition model, characterized in that: The device comprises: Sample data acquisition module: used to acquire sample data of users, where the sample data points to features with financial fraud risks; A data disguise training module: used to perform data disguise training based on the sample data using a first machine learning model to obtain disguised data corresponding to the sample data; A data recognition training module: used to perform data recognition training based on the disguised data using a second machine learning model to obtain a target function, wherein the first machine learning model and the second machine learning model are established based on a generative adversarial network; Model parameter adjustment module: used to adjust the first model parameters of the first machine learning model and the second model parameters of the second machine learning model, optimize the value of the objective function to a preset value; use the second machine learning model corresponding to the current second model parameter as the data recognition model; The step of performing data disguise training based on the sample data using a first machine learning model to obtain disguised data corresponding to the sample data includes: inputting the sample data into the first machine learning model; combining the sample data with random data using the first machine learning model to generate the disguised data; The method of using a second machine learning model to perform data recognition training based on the disguised data to obtain a target function includes: inputting the disguised data into the second machine learning model; obtaining a first recognition result when the second machine learning model recognizes that the disguised data has the characteristics of financial fraud risk; obtaining a second recognition result when the second machine learning model cannot recognize that the disguised data has the characteristics of financial fraud risk; and generating the target function according to the first recognition result and the second recognition result.

8. The device according to claim 7, characterized in that The step of adjusting the model parameters of the first machine learning model and the model parameters of the second machine learning model to optimize the objective function to a preset value includes: updating the value of the second model parameter based on the value of the first model parameter updated in the previous iteration and the objective function; updating the value of the first model parameter based on the value of the second model parameter updated in the previous iteration and the objective function; iterating along the gradient descent direction of the objective function, and repeating the above two steps until the value of the objective function reaches a preset value.

9. The device according to claim 7, characterized in that The generating the objective function according to the first recognition result and the second recognition result includes: Based on the first recognition result, a first loss function is constructed: l(0, D(x′ i , W D )); A second loss function is constructed based on the second recognition result: l(1, D(A(x i , W A ), W D )); By formula: Obtaining the objective function; Among them, x i is the i-th sample data, x′ i is the disguised data corresponding to the i-th sample data, B is the number of sample data within time t, W A is the first model parameter, W D is the second model parameter, D(x′ i , W D ) indicates that the second machine learning model cannot identify the characteristics of the disguised data as having financial fraud risk, D(A(x i , W A ), W D ) indicates that the second machine learning model recognizes that the disguised data has characteristics of financial fraud risk.

10. The device according to claim 7, characterized in that Set the learning rate of gradient descent to 0.0005 to 0.001 and the number of iterations to 50 to 100.

11. The device according to claim 7, characterized in that The second machine learning model performs data recognition training based on a binary classification discriminant network algorithm.

12. A data processing device, characterized in that: The device comprises: To-be-identified data acquisition module: used to acquire the user's to-be-identified data; Data recognition and processing module: used to input the data to be recognized into a data recognition model for data recognition processing, and to identify whether the data to be recognized has the characteristics of financial fraud risk, wherein the data recognition model is obtained based on the training method of the data recognition model as described in any one of claims 1 to 5.

13. A server, characterized in that: The server includes a processor and a memory, wherein the memory stores at least one instruction, at least one program or a set of codes, and the at least one instruction, at least one program or a set of codes are loaded and executed by the processor to implement the training method of the data recognition model as described in any one of claims 1 to 5, or the data processing method as described in claim 6.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one instruction, at least one program or a set of codes, and the at least one instruction, at least one program or a set of codes are loaded and executed by a processor to implement the training method of the data recognition model as described in any one of claims 1 to 5, or the data processing method as described in claim 6.

Images