Model structure, model training method, transaction processing method, equipment, medium

Abstract

The present disclosure provides a model structure, including multiple layers of one-dimensional convolutional layers, weighting modules, splicing modules, and fully connected layers connected in sequence, wherein the multiple layers of one-dimensional convolutional layers are used to extract features from input data to obtain multiple feature vectors, the weighting modules are used to weight the multiple feature vectors, the splicing modules are used to sequentially splice the weighted multiple feature vectors to obtain a one-dimensional feature vector, and the fully connected layers are used to determine the risk level of the input data according to the one-dimensional feature vector. The present disclosure also provides a model training method, a transaction processing method, a training device, an execution device, and a readable storage medium.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and more specifically, to a model structure, a model training method, a transaction processing method, a device, and a medium. Background Art

[0002] In recent years, with the continuous development of media channels, communications, and mobile payment technologies, the popularization of mobile phones, televisions, promotional methods, and their equipment, online information has been integrated into our lives and has penetrated into people of all ages in different ways. Along with this, fraud gangs have continued to "upgrade" and use high-tech "hiding".

[0003] In the prior art, a two-dimensional convolutional neural network model is used to effectively perceive data changes within a time period. However, as a receptive field of a small window, it cannot perceive data changes over a long period of time. If it is a chronic fraud and the abnormal changes are not significant, this method will limit the versatility and accuracy of the diagnosis. Summary of the invention

[0004] In view of this, the present disclosure provides a model structure, a model training method, a transaction processing method and a device.

[0005] A first aspect of the present disclosure provides a model structure, comprising a plurality of one-dimensional convolutional layers, a weighting module, a concatenation module, and a fully connected layer connected in sequence;

[0006] The multi-layer one-dimensional convolutional layer is used to extract features from the input data to obtain multiple feature vectors;

[0007] The weighting module is used to weight the multiple feature vectors;

[0008] The concatenation module is used to sequentially concatenate the weighted multiple feature vectors to obtain a one-dimensional feature vector;

[0009] The fully connected layer is used to determine the risk level of the input data according to the one-dimensional feature vector.

[0010] In one embodiment of the present disclosure, the input data is target data to be processed, and the model structure further includes a construction module, and the construction module is used to:

[0011] When a transaction of a trading account is detected, obtaining historical transaction data of the trading account;

[0012] Target data to be processed is constructed based on the historical transaction data of the transaction account, and the target data is one-dimensional data.

[0013] In one of the embodiments of the present disclosure, the type of input data includes any one or more of transaction account information, transaction counterparty account information, changes in income and expenditure amounts, types of counter services, changes in income and expenditure amounts for counter services, changes in income and expenditure amounts for self-service machines, changes in income and expenditure amounts for mobile payments, transaction outlet information, number of card operations, transaction location, and maximum income and expenditure amounts.

[0014] In one of the embodiments of the present disclosure, the size of the convolution kernel of the one-dimensional convolution layer is related to the type of the input data.

[0015] In one embodiment of the present disclosure, the weighting module is specifically used for:

[0016] The multiple feature vectors are weighted in a 1×1 manner.

[0017] In one embodiment of the present disclosure, the input data is a training sample in a training set, and the model structure further includes a sample processing module, and the sample processing module is used to:

[0018] Get historical transaction data of sample accounts;

[0019] Building a training sample based on the historical transaction data of the sample account, wherein the training sample is one-dimensional data;

[0020] The probability of each training sample being at each preset risk level is marked.

[0021] A second aspect of the present disclosure provides a model training method, wherein the model includes a plurality of one-dimensional convolutional layers, a weighting module, a splicing module, and a fully connected layer connected in sequence, and the method includes:

[0022] Inputting training samples into the model, wherein the training samples are marked with risk levels;

[0023] Extracting features from the training samples through the multi-layer one-dimensional convolutional layers to obtain multiple feature vectors;

[0024] weighting the multiple feature vectors by the weighting module;

[0025] The weighted multiple feature vectors are sequentially spliced ​​by the splicing module to obtain a one-dimensional feature vector;

[0026] Determining the risk level of the training sample according to the one-dimensional feature vector through the fully connected layer;

[0027] Training is performed according to the risk level determined by the model, the marked risk level and the training error to obtain a trained model.

[0028] In one embodiment of the present disclosure, before extracting features from the training samples through the multi-layer one-dimensional convolutional layers to obtain multiple feature vectors, the method further includes:

[0029] Get historical transaction data of sample accounts;

[0030] Building a training sample based on the historical transaction data of the sample account, wherein the training sample is one-dimensional data;

[0031] The probability of each training sample being at each preset risk level is marked.

[0032] In one embodiment of the present disclosure, the method further comprises:

[0033] The size of the convolution kernel of the one-dimensional convolution layer is determined according to the type of historical transaction data in the training sample.

[0034] In one embodiment of the present disclosure, weighting the multiple feature vectors by the weighting module includes:

[0035] The weighting module performs 1×1 weighting on the multiple feature vectors.

[0036] In one of the embodiments of the present disclosure, the types of data in the training samples include any one or more of transaction account information, transaction counterparty account information, changes in income and expenditure amounts, types of counter services, changes in income and expenditure amounts for counter services, changes in income and expenditure amounts for self-service machines, changes in income and expenditure amounts for mobile payments, transaction outlet information, number of card operations, transaction location, and maximum income and expenditure amounts.

[0037] In one embodiment of the present disclosure, the method further includes:

[0038] Deploy the trained model on the target device.

[0039] A third aspect of the present disclosure provides a transaction processing method, comprising:

[0040] When a transaction is detected in the transaction account, constructing target data to be processed based on historical transaction data of the transaction account;

[0041] Inputting the target data to be processed into the trained model, wherein the trained model includes multiple one-dimensional convolutional layers, a weighted module, a splicing module, and a fully connected layer;

[0042] Extracting features of the target data through the multi-layer one-dimensional convolutional layers to obtain multiple feature vectors;

[0043] weighting the multiple feature vectors by the weighting module;

[0044] The weighted multiple feature vectors are sequentially spliced ​​by the splicing module to obtain a one-dimensional feature vector;

[0045] The risk level of the transaction is determined according to the one-dimensional feature vector through the fully connected layer.

[0046] In one of the embodiments of the present disclosure, the types of historical transaction data include transaction account information, transaction counterparty account information, changes in income and expenditure amounts, types of counter services, changes in income and expenditure amounts for counter services, changes in income and expenditure amounts for self-service machines, changes in income and expenditure amounts for mobile payments, transaction outlet information, number of card operations, transaction location, and any one or more of the maximum income and expenditure amounts.

[0047] In one embodiment of the present disclosure, weighting the multiple feature vectors by the weighting module includes:

[0048] The weighting module performs 1×1 weighting on the multiple feature vectors.

[0049] The fourth aspect of the present disclosure provides a training device, which has the function of implementing the method described in the second aspect, and the function is implemented by hardware or by hardware executing corresponding software, and the hardware or the software includes one or more modules corresponding to the function.

[0050] The fifth aspect of the present disclosure provides an execution device, which has the function of implementing the method described in the third aspect, and the function is implemented by hardware or by hardware executing corresponding software. The hardware or the software includes one or more modules corresponding to the function.

[0051] A sixth aspect of the present disclosure provides a training device, including a processor and a memory, wherein the processor and the memory are coupled.

[0052] The memory is used to store programs;

[0053] The processor is used to execute the program in the memory, so that the training device executes the method as described in the second aspect.

[0054] A seventh aspect of the present disclosure provides a training device, including a processor and a memory, wherein the processor and the memory are coupled.

[0055] The memory is used to store programs;

[0056] The processor is used to execute the program in the memory, so that the training device executes the method described in the third aspect.

[0057] An eighth aspect of the present disclosure provides a computer-readable storage medium storing computer-executable instructions, which are used to implement the method described above when executed.

[0058] According to the embodiment of the present disclosure, the comprehensive structure of the model structure, which includes multiple layers of one-dimensional convolutional layers, weighted modules, splicing modules, and fully connected layers connected in sequence, can effectively ensure and improve monitoring accuracy while reducing the number of neural network parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The above and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0060] Figure 1 A schematic diagram schematically shows a model structure according to an embodiment of the present disclosure;

[0061] Figure 2 A schematic diagram of a multi-layer one-dimensional convolutional layer according to an embodiment of the present disclosure is schematically shown;

[0062] Figure 3 A schematic diagram of a process flow of a model training method according to an embodiment of the present disclosure is schematically shown;

[0063] Figure 4 A schematic diagram of a transaction processing method according to an embodiment of the present disclosure is shown schematically;

[0064] Figure 5 The structure diagram of the training device according to the embodiment of the present disclosure is schematically shown;

[0065] Figure 6 The schematic diagram of the structure of the execution device according to the embodiment of the present disclosure is shown;

[0066] Figure 7 A block diagram of a training device or an execution device suitable for implementing the method described above according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0067] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0068] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0069] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0070] In the case of using expressions such as "at least one of A, B, and C, etc.", it should generally be interpreted in accordance with the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.). In the case of using expressions such as "at least one of A, B, or C, etc.", it should generally be interpreted in accordance with the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, or C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0071] The present disclosure provides a model structure, a model training method, a transaction processing method, a device, and a medium. First, specific customer card usage behavior data from different channels are collected and stored in a database. Then, data classification, sample collection and processing are performed to obtain training samples to make them suitable for the model structure. Next, feature extraction is performed on the input data through multiple one-dimensional convolutional layers to obtain multiple feature vectors. The multiple feature vectors are weighted through a weighting module. The multiple weighted feature vectors are sequentially spliced ​​through a splicing module to obtain a one-dimensional feature vector. The risk level of the input data is determined according to the one-dimensional feature vector through a fully connected layer.

[0072] Figure 1 A schematic diagram schematically shows a model structure according to an embodiment of the present disclosure.

[0073] like Figure 1 As shown, the model structure includes a multi-layer one-dimensional convolutional layer 110, a weighting module 120, a splicing module 130, and a fully connected layer 140 which are connected in sequence.

[0074] The multi-layer one-dimensional convolutional layer 110 is used to extract features from the input data to obtain multiple feature vectors.

[0075] The weighting module 120 is used to weight the multiple feature vectors.

[0076] The concatenation module 130 is used to sequentially concatenate the weighted multiple feature vectors to obtain a one-dimensional feature vector.

[0077] The fully connected layer 140 is used to determine the risk level of the input data according to the one-dimensional feature vector.

[0078] In the disclosed embodiments, the model refers to a model for determining the risk level task, which is essentially a neural network or a part of its structure is a neural network. The application of the model includes two stages: training and reasoning. In the training stage, the input data is a training set, which is used to train the model according to the training set to obtain a trained model. In the reasoning stage, the input data is target data, which is used to use the trained model to determine the risk level of unlabeled instances (i.e., target data of trading accounts where transactions occur) to monitor whether the account is at risk of being defrauded.

[0079] In the embodiment of the present disclosure, the number of the multi-layer one-dimensional convolutional layers 110 can be two, three or four, etc. When the number of the multi-layer one-dimensional convolutional layers 110 is two, it includes one input layer and one convolutional layer. When the number of the multi-layer one-dimensional convolutional layers 110 is three, it includes one input layer and two convolutional layers. For example, when the number of the multi-layer one-dimensional convolutional layers 110 is three, please refer to Figure 2 (The input layer is not shown), the convolution kernel of the input layer is 1×n×16, the convolution kernel of the first convolution layer is 32@1×3×1, and the convolution kernel of the second convolution layer is 64@1×3×32, and finally 30×64 1×t output feature vectors are obtained.

[0080] Among them, the use of statistical characteristic values ​​can effectively extract information with more prominent features. If the data change pattern is relatively flat and not prominent, the characteristic information cannot be effectively extracted. Therefore, we can make plans based on the type of data collected. We can use data with specific meanings. For example, to prevent telecommunications fraud, we can use numbers or call records as input data. Data with specific meanings can be more specific for extracting feature information for different scenarios. Improve the training and processing accuracy of the diagnostic network. In the disclosed embodiment, for determining the transaction risk level in banking business, the type of input data can be any one or more of changes in income and expenditure amounts, types of counter business, changes in income and expenditure amounts for counter business, changes in income and expenditure amounts for self-service machines, changes in income and expenditure amounts for mobile payments, transaction outlet information, number of card operations, transaction province, transaction city, maximum income and expenditure amounts, etc.

[0081] In the disclosed embodiment, the size of the convolution kernel of the one-dimensional convolution layer 110 is related to the type of input data. Wherein, n is the convolution kernel size value set according to the type of input data, and n can also be adjusted according to actual conditions. For input of different types of data, multi-adaptive convolution kernels of different scales are used to effectively extract information from different types of data, improve information extraction capabilities, and enhance network accuracy.

[0082] In the disclosed embodiment, the weighting module 120 is specifically used to perform 1×1 weighting on the multiple feature vectors, perform 1×1 weighting on the 30×64 feature vectors, and sequentially splice them using the splicing module 130 to obtain an m-element one-dimensional feature vector. Where m=30×64×1×t. Then, the obtained m-element one-dimensional feature vector is input into the fully connected layer 140 to determine the risk level of the input data. It is understandable that the weights can be self-learned with the network, effectively learning to filter the proportion of different information and the degree of influence on the results, making adaptive changes, reducing redundant information, and improving monitoring accuracy.

[0083] According to the embodiment of the present disclosure, the comprehensive structure of the model structure, which includes multiple layers of one-dimensional convolutional layers 110, weighting modules 120, splicing modules 130, and fully connected layers 140 connected in sequence, can effectively ensure and improve monitoring accuracy while reducing the number of neural network parameters.

[0084] In an embodiment of the present disclosure, when the input data is a training sample in a training set, the model structure also includes a sample processing module, which is used to obtain historical transaction data of a sample account and construct a training sample based on the historical transaction data of the sample account. The training sample is one-dimensional data and marks the probability of each training sample being at a preset risk level.

[0085] In an embodiment of the present disclosure, when the input data is target data to be processed, the model structure also includes a construction module, which is used to obtain historical transaction data of the trading account when a transaction is detected in the trading account, and to construct the target data to be processed based on the historical transaction data of the trading account, and the target data is one-dimensional data.

[0086] Specifically, the above-mentioned historical transaction data is obtained, for example: daily card usage data of user accounts, including daily card usage data, weekly card usage data, monthly card usage data, etc. It can be statistically summarized and stored according to the time scale and recorded as one-dimensional data. Exemplarily, taking the daily time scale as an example, and 30 days as the sample time period, the training samples obtained by the sample processing module or the target data obtained by the construction module are one-dimensional data containing data of 1×30 at a scale of 30. In the present disclosure, by using historical transaction data, broadening the time scale of feature extraction, and then using a convolutional neural network, it is possible to increase the time scale range of the input without causing a large increase in the amount of calculation, and can effectively balance and control the indicators between the parameter quantity and accuracy, while achieving considerable accuracy, reducing the computational cost.

[0087] Furthermore, historical transaction data can be stored separately in a database. Historical transaction data within a certain period is stored in the database, and data older than a certain period can be completely deleted. For example, data older than four years can be considered useless information and deleted from the database.

[0088] Specifically, for the probability of marking each training sample at each preset risk level, the risk level can be set according to business needs, for example, set to 3 levels, namely high, medium, and low, and for another example, set to 5 levels, namely high, medium-high, medium, medium-low, and low. It can be understood that no matter how many levels are divided, for a training sample or target data, the sum of the overall probabilities of the training sample or target data being at these levels is 1.

[0089] Among them, when the input data is a training sample in a training set, the training cycle can be set for the number of training samples in the training set. For example, the training cycle is 3 months (90 days), that is, each time training, the number of samples corresponding to each account input into the neural network is 61, and the sample composition step is set to 1 day. For example, the first sample is 1-30, the second sample is 2-31, and so on.

[0090] Among them, the training samples in the training set include two categories. One is historical sample-label, which is the accurate sample label. The other is the risk level determined by model structure diagnosis. We can record the real-time diagnosis results, judge the accuracy of the current diagnosis results, and conduct "error correction" training.

[0091] In one embodiment of the present disclosure, the size of the convolution kernel of the one-dimensional convolution layer 110 is related to the type of the input data. The training sample or target data is a one-dimensional vector of transaction data of 30 input scales, and the convolution kernel is set according to the characteristics of different quantity types, for example:

[0092] Trading account income amount, 1×5

[0093] Trading account expenditure amount, 1×5

[0094] Types of transaction account business, 1×3,

[0095] The amount of income from the other party's account, 1×3

[0096] The amount of money spent by the counterparty to the transaction account, 1×3

[0097] Type of business handling for the counterparty account of the transaction account, 1×3,

[0098] The convolution kernel is adjusted and set according to specific circumstances, such as different regions, provinces, cities, data collection time scales (daily data, weekly data, monthly data), etc.

[0099] Figure 3 The flowchart of the model training method according to the embodiment of the present disclosure is schematically shown. The model includes a plurality of one-dimensional convolutional layers, a weighting module, a splicing module, and a fully connected layer connected in sequence. Specifically, it can be Figure 1 The model structure shown.

[0100] like Figure 3 As shown, the method includes operations S301 to S304.

[0101] In operation S301 , a training device inputs a training sample into a model, where the training sample is marked with a risk level.

[0102] In operation S302, features are extracted from the training samples through multiple layers of one-dimensional convolutional layers in the model to obtain multiple feature vectors.

[0103] In operation S303, the plurality of feature vectors are weighted by a weighting module in the model.

[0104] In operation S304, the weighted multiple feature vectors are sequentially spliced ​​through a splicing module in the model to obtain a one-dimensional feature vector.

[0105] In operation S305, the risk level of the training sample is determined according to the one-dimensional feature vector through the fully connected layer in the model.

[0106] In operation S306, the training device performs training according to the risk level determined by the model, the marked risk level and the training error to obtain a trained model.

[0107] In the implementation of the present disclosure, the loss function, error and number of iterations can be set before training. After each risk level is obtained through the model, it is determined whether the number of iterations has been reached. If so, it is determined whether the error accuracy has been achieved or whether the training has been repeated a preset number of times. If not, the loss function, error and number of iterations are set again. If so, the training is completed.

[0108] In one embodiment of the present disclosure, before operation S301, it also includes: obtaining historical transaction data of a sample account, constructing a training sample based on the historical transaction data of the sample account, the training sample being one-dimensional data, and marking the probability of each training sample being at a preset risk level.

[0109] In the disclosed embodiments, the type of transaction data may be any one or more of changes in the amount of income and expenditure, types of counter services, changes in the amount of income and expenditure for counter services, changes in the amount of income and expenditure for self-service machines, changes in the amount of income and expenditure for mobile payments, transaction point information, number of card operations, transaction province, transaction city, maximum amount of income and expenditure, etc.

[0110] In an embodiment of the present disclosure, constructing a training sample based on the historical transaction data of the sample account can be to obtain daily card usage data of the user account, where the daily card usage data can be daily card usage data, weekly card usage data, monthly card usage data, etc., and then multiple pieces of daily card usage data are combined into a training sample, which is one-dimensional data.

[0111] It can be understood that the training samples in the training set include normal (risk-free) samples and samples of various risk levels.

[0112] In the embodiment of the present disclosure, for marking the probability of each training sample being at a preset risk level, the risk level can be set according to business needs. It can be understood that no matter how many levels it is divided into, for a training sample or target data, the sum of the overall probabilities of the training sample or target data being at these levels is 1.

[0113] In one embodiment of the present disclosure, the size of the convolution kernel of the one-dimensional convolution layer is determined according to the type of historical transaction data in the training sample. For example, if the transaction data is the amount of transaction account income, the size of the convolution kernel of the one-dimensional convolution layer is 1×5, and if the transaction data is the type of transaction account business, the size of the convolution kernel of the one-dimensional convolution layer is 1×3. It is understandable that the above is only an illustrative example, and the above can also be set to other sizes, and the present disclosure does not limit this.

[0114] In one embodiment of the present disclosure, the weighting module performs 1×1 weighting on the multiple feature vectors. It is understandable that the weights can be self-learned with the network, effectively learning to filter the weight of different information and the influence on the results, making adaptive changes, reducing redundant information, and improving monitoring accuracy.

[0115] In one of the embodiments of the present disclosure, the trained model is deployed on a target device, for example, on an edge device or a terminal device, such as a mobile phone, a tablet, a laptop computer, a monitoring system, etc.

[0116] Figure 4 The flowchart of the transaction processing method according to the embodiment of the present disclosure is schematically shown.

[0117] like Figure 4 As shown, the method includes operations S401 to S406.

[0118] In operation S401 , when it is detected that a transaction occurs in the trading account, the execution device constructs target data to be processed based on historical transaction data of the trading account.

[0119] In operation S402, the execution device inputs the target data to be processed into the trained model, where the trained model includes multiple one-dimensional convolutional layers, weighted modules, concatenation modules, and fully connected layers.

[0120] In operation S403, feature extraction is performed on the target data through the multi-layer one-dimensional convolution layer to obtain multiple feature vectors.

[0121] In operation S404, the plurality of feature vectors are weighted by the weighting module.

[0122] In operation S405, the plurality of weighted feature vectors are sequentially concatenated by the concatenation module to obtain a one-dimensional feature vector.

[0123] In operation S406, the risk level of the transaction is determined according to the one-dimensional feature vector through the fully connected layer.

[0124] In the embodiment of the present disclosure, the structure of the trained model can be Figure 1 The structures of the models shown are identical.

[0125] In the disclosed embodiment, the number of multi-layer one-dimensional convolutional layers can be two, three or four layers, etc. For determining the transaction risk level in banking business, the type of target data can be any one or more of the changes in the amount of income and expenditure, the type of counter business, the changes in the amount of income and expenditure of counter business, the changes in the amount of income and expenditure of self-service machines, the changes in the amount of income and expenditure of mobile payments, transaction outlet information, the number of card operations, the transaction province, the transaction city, the maximum amount of income and expenditure, etc.

[0126] In the disclosed embodiment, the size of the convolution kernel of the one-dimensional convolution layer is related to the type of input data. For example, if the transaction data is the amount of transaction account income, the size of the convolution kernel of the one-dimensional convolution layer is 1×5, and if the transaction data is the type of transaction account business, the size of the convolution kernel of the one-dimensional convolution layer is 1×3.

[0127] In the disclosed embodiment, the weighting module is specifically used to perform 1×1 weighting on the multiple feature vectors. It is understandable that the weights can be self-learned with the network, effectively learning to filter the weight of different information and the influence on the results, making adaptive changes, reducing redundant information, and improving monitoring accuracy.

[0128] Figure 5 The structural diagram of the training device according to the embodiment of the present disclosure is schematically shown.

[0129] like Figure 5 As shown, the training device 500 includes a first input module 510 and a training module 520 .

[0130] The first input module 510 is used to input training samples into the model.

[0131] The training module 520 is used to perform training according to the risk level determined by the model, the marked risk level and the training error to obtain a trained model.

[0132] It should be noted that in the embodiment of the present application, the model structure deployed on the training device 500 can be found in Figure 1 For the detailed description of the embodiment shown in the figure, please refer to the above Figure 3 Description of the related embodiments shown.

[0133] Furthermore, the training device 500 may also include a sample acquisition module, which is used to obtain historical transaction data of a sample account, and construct a training sample based on the historical transaction data of the sample account. The training sample is one-dimensional data, and marks the probability of each training sample being at a preset risk level.

[0134] Figure 6 The schematic diagram shows the structure of an execution device according to an embodiment of the present disclosure.

[0135] like Figure 6 As shown, the execution device 600 includes a construction module 610 and a second input module 620 .

[0136] The construction module 610 is used to construct the target data to be processed based on the historical transaction data of the transaction account when it is detected that a transaction occurs in the transaction account.

[0137] The second input module 620 is used to input the target data to be processed into the trained model.

[0138] It should be noted that in the embodiment of the present application, the model structure deployed on the execution device 600 can be referred to Figure 1 For the detailed description of the embodiment shown in the figure, please refer to the above Figure 4 Description of the related embodiments shown.

[0139] The training device, execution device, etc. provided in the embodiments of the present disclosure may be specifically a chip, and the chip includes: a processing unit and a communication unit, the processing unit may be, for example, a processor, and the communication unit may be, for example, an input / output interface, a pin, or a circuit lamp. The processing unit may execute the computer execution instructions stored in the storage unit, so that the chip in the training device executes the steps executed by the training device described in the above-mentioned embodiments, or the chip in the execution device executes the operations executed by the execution device described in the above-mentioned embodiments.

[0140] Optionally, the storage unit is a storage unit within the chip, such as a register, a cache lamp, and the storage unit can also be a bit within the wireless access device end and a storage unit outside the chip, such as a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM), etc.

[0141] According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits, or at least part of the functions of any one of them can be implemented in one module. According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits can be split into multiple modules for implementation. According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits can be at least partially implemented as hardware circuits, such as field programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems on chips, systems on substrates, systems on packages, application specific integrated circuits (ASICs), or can be implemented by hardware or firmware in any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in any appropriate combination of any of them. Alternatively, according to the embodiments of the present invention, one or more of the modules, submodules, units, and subunits can be at least partially implemented as computer program modules, and when the computer program modules are run, the corresponding functions can be performed.

[0142] For example, any multiple of the first input module 510 and the training module 520 can be combined in one module / unit / sub-unit for implementation, or any one of the modules / units / sub-units can be split into multiple modules / units / sub-units. Alternatively, at least part of the functions of one or more of these modules / units / sub-units can be combined with at least part of the functions of other modules / units / sub-units and implemented in one module / unit / sub-unit. According to an embodiment of the present disclosure, at least one of the first input module 510 and the training module 520 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in any appropriate combination of any of them. Alternatively, at least one of the first input module 510 and the training module 520 can be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function can be performed.

[0143] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0144] Figure 7 A block diagram of a training device or an execution device suitable for implementing the method described above according to an embodiment of the present disclosure is schematically shown. Figure 7 The training device or execution device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0145] like Figure 7 As shown, the training device or execution device 700 according to an embodiment of the present disclosure includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage part 708 to a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include an onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0146] In RAM 703, various programs and data required for the operation of system 700 are stored. Processor 701, ROM 702 and RAM 703 are connected to each other through bus 704. Processor 701 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 702 and / or RAM 703. It should be noted that the program can also be stored in one or more memories other than ROM 702 and RAM 703. Processor 701 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.

[0147] According to an embodiment of the present disclosure, the device 700 may further include an input / output (I / O) interface 705, which is also connected to the bus 704. The system 700 may further include one or more of the following components connected to the I / O interface 705: an input section 706 including a keyboard, a mouse, etc.; an output section 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, a modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 710 as needed, so that a computer program read therefrom is installed into the storage section 708 as needed.

[0148] According to an embodiment of the present disclosure, the method flow according to an embodiment of the present disclosure can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 709, and / or installed from the removable medium 711. When the computer program is executed by the processor 701, the above-mentioned functions defined in the system of the embodiment of the present disclosure are executed. According to an embodiment of the present disclosure, the system, equipment, device, module, unit, etc. described above can be implemented by a computer program module.

[0149] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0150] According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium. For example, it may include, but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, apparatus, or device.

[0151] For example, according to an embodiment of the present disclosure, the computer-readable storage medium may include the ROM 702 and / or the RAM 703 described above and / or one or more memories other than the ROM 702 and the RAM 703 .

[0152] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0153] It will be appreciated by those skilled in the art that the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations and / or combinations are not explicitly described in the present disclosure. In particular, the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways without departing from the spirit and teachings of the present disclosure. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0154] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is defined by the attached claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A model structure, including multiple layers of one-dimensional convolutional layers, weighted modules, splicing modules, and fully connected layers connected in sequence; The multi-layer one-dimensional convolutional layer is used to extract features from the input data to obtain multiple feature vectors; The weighting module is used to weight the multiple feature vectors; The concatenation module is used to sequentially concatenate the weighted multiple feature vectors to obtain a one-dimensional feature vector; The fully connected layer is used to determine the risk level of the input data according to the one-dimensional feature vector, in, The input data is target data to be processed, and the model structure further includes a construction module, which is used to: When a transaction of a trading account is detected, obtaining historical transaction data of the trading account; Target data to be processed is constructed based on the historical transaction data of the transaction account, and the target data is one-dimensional data.

2. According to the model structure of claim 1, the types of input data include transaction account information, transaction counterparty account information, changes in income and expenditure amounts, types of counter services, changes in income and expenditure amounts for counter services, changes in income and expenditure amounts for self-service machines, changes in income and expenditure amounts for mobile payments, transaction outlet information, number of card operations, transaction location, and any one or more of the maximum income and expenditure amounts.

3. The model structure according to any one of claims 1 to 2, wherein: The size of the convolution kernel of the one-dimensional convolution layer is related to the type of the input data.

4. According to the model structure of claim 1, the weighting module is specifically used for: The multiple feature vectors are weighted in a 1×1 manner.

5. A model structure, comprising a plurality of one-dimensional convolutional layers, a weighted module, a splicing module, and a fully connected layer connected in sequence; The multi-layer one-dimensional convolutional layer is used to extract features from the input data to obtain multiple feature vectors; The weighting module is used to weight the multiple feature vectors; The concatenation module is used to sequentially concatenate the weighted multiple feature vectors to obtain a one-dimensional feature vector; The fully connected layer is used to determine the risk level of the input data according to the one-dimensional feature vector, in, The input data is a training sample in a training set, and the model structure further includes a sample processing module, which is used to: Get historical transaction data of sample accounts; Building a training sample based on the historical transaction data of the sample account, wherein the training sample is one-dimensional data; The probability of each training sample being at each preset risk level is marked.

6. A model training method, the model comprising a plurality of one-dimensional convolutional layers, a weighting module, a splicing module, and a fully connected layer connected in sequence, the method comprising: Inputting training samples into the model, wherein the training samples are marked with risk levels; Extracting features from the training samples through the multi-layer one-dimensional convolutional layers to obtain multiple feature vectors; weighting the multiple feature vectors by the weighting module; The weighted multiple feature vectors are sequentially spliced ​​by the splicing module to obtain a one-dimensional feature vector; Determining the risk level of the training sample according to the one-dimensional feature vector through the fully connected layer; Training is performed according to the risk level determined by the model, the marked risk level and the training error to obtain a trained model, Before extracting features from the training samples through the multi-layer one-dimensional convolutional layers to obtain multiple feature vectors, the method further includes: Get historical transaction data of sample accounts; Building a training sample based on the historical transaction data of the sample account, wherein the training sample is one-dimensional data; The probability of each training sample being at each preset risk level is marked.

7. The method according to claim 6, wherein: The method further comprises: The size of the convolution kernel of the one-dimensional convolution layer is determined according to the type of historical transaction data in the training sample.

8. The method according to claim 6, wherein weighting the plurality of feature vectors by the weighting module comprises: The weighting module performs 1×1 weighting on the multiple feature vectors.

9. According to the method described in claim 6, the types of data in the training samples include any one or more of transaction account information, transaction counterparty account information, changes in income and expenditure amounts, types of counter services, changes in income and expenditure amounts for counter services, changes in income and expenditure amounts for self-service machines, changes in income and expenditure amounts for mobile payments, transaction outlet information, number of card operations, transaction location, and maximum income and expenditure amounts.

10. The method according to any one of claims 6 to 9, further comprising: Deploy the trained model on the target device.

11. A transaction processing method, comprising: When a transaction is detected in a trading account, target data to be processed is constructed based on historical transaction data of the trading account; Inputting the target data to be processed into the trained model, wherein the trained model includes multiple one-dimensional convolutional layers, a weighted module, a splicing module, and a fully connected layer; Extracting features of the target data through the multi-layer one-dimensional convolutional layers to obtain multiple feature vectors; weighting the multiple feature vectors by the weighting module; The weighted multiple feature vectors are sequentially spliced ​​by the splicing module to obtain a one-dimensional feature vector; The risk level of the transaction is determined according to the one-dimensional feature vector through the fully connected layer.

12. According to the method described in claim 11, the types of historical transaction data include any one or more of transaction account information, transaction counterparty account information, changes in income and expenditure amounts, types of counter services, changes in income and expenditure amounts for counter services, changes in income and expenditure amounts for self-service machines, changes in income and expenditure amounts for mobile payments, transaction outlet information, number of card operations, transaction location, and maximum income and expenditure amounts.

13. The method according to claim 11, wherein weighting the plurality of feature vectors by the weighting module comprises: The weighting module performs 1×1 weighting on the multiple feature vectors.

14. A training device, which has the function of implementing the method described in any one of claims 6 to 10, wherein the function is implemented by hardware or by hardware executing corresponding software, and the hardware or the software includes one or more modules corresponding to the function.

15. An execution device, wherein the device has the function of implementing the method described in any one of claims 11 to 13, wherein the function is implemented by hardware or by hardware executing corresponding software, and the hardware or the software includes one or more modules corresponding to the function.

16. A training device comprising a processor and a memory, wherein the processor and the memory are coupled, The memory is used to store programs; The processor is used to execute the program in the memory, so that the training device executes the method as described in any one of claims 6 to 10.

17. An execution device, comprising a processor and a memory, wherein the processor and the memory are coupled, The memory is used to store programs; The processor is used to execute the program in the memory, so that the execution device executes the method according to any one of claims 11 to 13.

18. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enables the processor to implement the method of any one of claims 6 to 10, or the method of any one of claims 11 to 13.

Images