Transaction data processing method and device, storage medium and program product

Abstract

The invention discloses a transaction data processing method and device, a storage medium and a program product. Relates to the field of artificial intelligence, and the method comprises the steps: obtaining target transaction data of a target financial account, the target financial account comprising a financial account having a loan behavior; converting the target transaction data into a matrix to obtain a target matrix; the target matrix is input into a target evaluation model, the default risk of the target financial account is scored through the target evaluation model, a risk score value is obtained, the target evaluation model comprises a first extraction layer, a second extraction layer and an attention layer, a convolutional neural network is deployed in the first extraction layer, a long-short-term memory network is deployed in the second extraction layer, and a long-short-term memory network is deployed in the attention layer; the attention layer is deployed with an attention mechanism; and determining a default risk level of the target financial account based on the risk score value. The problem that the processing efficiency is low when the transaction data after the financial account loan is analyzed and processed in a manual analysis mode in the related technology is solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and more specifically, to a method, device, storage medium, and program product for processing transaction data. Background Art

[0002] In related technologies, manual analysis or pre-established screening rules are used to analyze transaction data after a financial account loan (for example, a credit card loan) to determine whether there are any defaults or overdue repayments, so as to determine the default risk level of the financial account. However, these technologies suffer from problems such as low accuracy, poor real-time performance, and inability to effectively deal with complex transaction data.

[0003] Currently, no effective solution has been proposed to address the problem of low efficiency in analyzing and processing post-loan transaction data of financial accounts using manual analysis methods in related technologies. Summary of the Invention

[0004] The main purpose of this application is to provide a transaction data processing method, device, storage medium and program product to solve the problem of low processing efficiency in the related art of using manual analysis to analyze and process post-loan transaction data of financial accounts.

[0005] To achieve the above-mentioned objectives, according to one aspect of the present application, a method for processing transaction data is provided. The method comprises: obtaining target transaction data of a target financial account, wherein the target financial account comprises a financial account with loan transactions, and the target transaction data comprises transaction data of the target financial account at t different time points, where t is a positive integer; converting the target transaction data into a matrix to obtain a target matrix; inputting the target matrix into a target assessment model, and scoring the default risk of the target financial account using the target assessment model to obtain a risk score value, wherein the target assessment model comprises: a first extraction layer, a second extraction layer, and an attention layer, wherein the first extraction layer is deployed with a convolutional neural network, the second extraction layer is deployed with a long short-term memory network, and the attention layer is deployed with an attention mechanism; and determining the default risk level of the target financial account based on the risk score value.

[0006] Furthermore, the target assessment model also includes: a fully connected layer, inputting the target matrix into the target assessment model, scoring the default risk of the target financial account through the target assessment model, and obtaining a risk score value, including: inputting the target matrix into the first extraction layer for feature extraction to obtain first feature data; inputting the first feature data into the second extraction layer to extract the time series features of the first feature data to obtain second feature data; inputting the second feature data into the attention layer to perform weighted calculation on the second feature data to obtain target feature data; inputting the target feature data into the fully connected layer to obtain the risk score value.

[0007] Furthermore, the first extraction layer includes: a convolution layer and a pooling layer. Inputting the target matrix into the first extraction layer for feature extraction to obtain first feature data includes: inputting the target matrix into the convolution layer for feature extraction to obtain an extraction result; inputting the extraction result into the pooling layer for dimensionality reduction processing to obtain the first feature data.

[0008] Furthermore, the second extraction layer includes: an input gate, a forget gate, an output gate and a memory unit, wherein the input gate is used to input the first feature data and store the first feature data in the memory unit, the forget gate is used to discard the feature data in the memory unit, the output gate is used to determine the second feature data based on the memory unit, and the memory unit is used to retain and transmit the feature data.

[0009] Furthermore, after determining the default risk level of the target financial account based on the risk score value, the method further includes: determining product recommendation data based on the risk level of the target financial account, wherein the product recommendation data includes: data of financial products to be recommended to the target financial account; obtaining transaction records of the target financial account on N trading platforms, wherein N is a positive integer; screening the N trading platforms based on the transaction records to obtain screened trading platforms; and sending the product recommendation data to the screened trading platforms.

[0010] Furthermore, the target evaluation model is obtained in the following manner: acquiring transaction data of M financial accounts within a historical period to obtain historical transaction data; converting the historical transaction data into a three-dimensional matrix to obtain a historical transaction matrix; performing model training on an initial evaluation model based on the historical transaction data, and when the model training reaches a predetermined convergence condition, determining the trained initial evaluation model as the target evaluation model, wherein the initial evaluation model is a target evaluation model that has not been trained.

[0011] Furthermore, after acquiring transaction data of M financial accounts within a historical period and obtaining the historical transaction data, the method further includes: determining, based on the historical transaction data, the duration of overdue repayment for each of the M financial accounts and the number of transactions for each of the M financial accounts; dividing the M financial accounts based on the duration of overdue repayment for each of the financial accounts and the number of transactions for each of the financial accounts to obtain a division result; and marking the default risk level of each of the financial accounts based on the division result.

[0012] To achieve the above-mentioned purpose, according to another aspect of the present application, a transaction data processing device is provided, the device comprising: a first acquisition unit, configured to acquire target transaction data of a target financial account, wherein the target financial account comprises: a financial account with loan activity, and the target transaction data comprises: transaction data of the target financial account at t different time points, where t is a positive integer; a conversion unit, configured to convert the target transaction data into a matrix to obtain a target matrix; a scoring unit, configured to input the target matrix into a target assessment model, and score the default risk of the target financial account using the target assessment model to obtain a risk score value, wherein the target assessment model comprises: a first extraction layer, a second extraction layer, and an attention layer, wherein the first extraction layer is deployed with a convolutional neural network, the second extraction layer is deployed with a long short-term memory network, and the attention layer is deployed with an attention mechanism; and a first determination unit, configured to determine the default risk level of the target financial account based on the risk score value.

[0013] Furthermore, the target assessment model also includes: a fully connected layer, and the scoring unit includes: a first extraction subunit, used to input the target matrix into the first extraction layer for feature extraction to obtain first feature data; a second extraction subunit, used to input the first feature data into the second extraction layer to extract the time series features of the first feature data to obtain second feature data; a calculation subunit, used to input the second feature data into the attention layer to perform weighted calculation on the second feature data to obtain target feature data; a processing subunit, used to input the target feature data into the fully connected layer to obtain the risk score value.

[0014] Furthermore, the first extraction layer includes: a convolution layer and a pooling layer, and the first extraction sub-unit includes: an extraction module, used to input the target matrix into the convolution layer for feature extraction to obtain an extraction result; a dimensionality reduction module, used to input the extraction result into the pooling layer for dimensionality reduction processing to obtain the first feature data.

[0015] Furthermore, the second extraction layer includes: an input gate, a forget gate, an output gate and a memory unit, wherein the input gate is used to input the first feature data and store the first feature data in the memory unit, the forget gate is used to discard the feature data in the memory unit, the output gate is used to determine the second feature data based on the memory unit, and the memory unit is used to retain and transmit the feature data.

[0016] Furthermore, the transaction data processing device also includes: a second determination unit, used to determine product recommendation data based on the risk level of the target financial account after determining the default risk level of the target financial account based on the risk score value, wherein the product recommendation data includes: data of financial products to be recommended to the target financial account; a second acquisition unit, used to obtain transaction records of the target financial account on N trading platforms, wherein N is a positive integer; a screening unit, used to screen the N trading platforms based on the transaction records to obtain screened trading platforms; and a sending unit, used to send the product recommendation data to the screened trading platforms.

[0017] Furthermore, the target evaluation model is obtained through the following sub-units: an acquisition sub-unit, used to acquire transaction data of M financial accounts within a historical period to obtain historical transaction data; a conversion sub-unit, used to convert the historical transaction data into a three-dimensional matrix to obtain a historical transaction matrix; a training sub-unit, used to perform model training on the initial evaluation model based on the historical transaction data, and when the model training reaches a predetermined convergence condition, the trained initial evaluation model is determined as the target evaluation model, wherein the initial evaluation model is a target evaluation model that has not been model trained.

[0018] Furthermore, the transaction data processing device also includes: a determination subunit, configured to obtain transaction data of M financial accounts within a historical period, and after obtaining the historical transaction data, determine the overdue repayment period of each of the M financial accounts and the number of transactions of each of the financial accounts based on the historical transaction data; a division subunit, configured to divide the M financial accounts based on the overdue repayment period of each of the financial accounts and the number of transactions of each of the financial accounts, to obtain a division result; and a marking subunit, configured to mark the default risk level of each of the financial accounts based on the division result.

[0019] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored executable program, wherein when the executable program is run, the device where the computer-readable storage medium is located is controlled to execute the transaction data processing method.

[0020] According to another aspect of the present application, an electronic device is provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the method for processing transaction data is executed when the program is running.

[0021] According to another aspect of the present application, a computer program product is provided, comprising computer instructions, which implement the steps of the transaction data processing method when executed by a processor.

[0022] In an embodiment of the present application, target transaction data of a target financial account is obtained, wherein the target financial account includes: a financial account with a loan behavior, and the target transaction data includes: transaction data of the target financial account at t different time points, where t is a positive integer; the target transaction data is converted into a matrix to obtain a target matrix; the target matrix is input into a target evaluation model, and the default risk of the target financial account is scored by the target evaluation model to obtain a risk score value, wherein the target evaluation model includes: a first extraction layer, a second extraction layer, and an attention layer, wherein a convolutional neural network is deployed in the first extraction layer, a long short-term memory network is deployed in the second extraction layer, and an attention mechanism is deployed in the attention layer; based on the risk score value, the default risk level of the target financial account is determined, thereby solving the technical problem of low processing efficiency in the related art of analyzing and processing the post-loan transaction data of the financial account by manual analysis.

[0023] In this application, the post-loan transaction data of a financial account is processed by a target evaluation model composed of a convolutional neural network, a long short-term memory network, and an attention mechanism to evaluate the default risk level of the target financial account, thereby avoiding the low efficiency of manually analyzing the post-loan transaction data of the financial account to evaluate the default risk level of the target financial account, thereby achieving the technical effect of improving the processing efficiency of the post-loan transaction data of the financial account. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0025] Figure 1 A hardware structure block diagram of a computer terminal for implementing a method for processing transaction data is shown;

[0026] Figure 2 is a flowchart of a method for processing transaction data provided in an embodiment of the present application;

[0027] Figure 3 This is an architecture diagram of a target evaluation model provided according to an embodiment of the present application;

[0028] Figure 4 This is a flowchart of the model training provided in accordance with an embodiment of the present application;

[0029] Figure 5 is a schematic diagram of a transaction data processing device provided in accordance with an embodiment of the present application;

[0030] Figure 6 This is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0031] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings 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 in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0032] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising 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.

[0033] It should be noted that the collected information (including but not limited to user device information, user personal information, transaction information, etc.) and data (including but not limited to data for display, analysis, transaction data, etc.) involved in this application are information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse. For example, an interface is set up between this system and relevant users or institutions to provide users with corresponding operation entrances for users to choose to agree or refuse the automated decision-making results; if the user chooses to refuse, the expert decision-making process will be entered.

[0034] Example 1

[0035] According to an embodiment of the present application, a method embodiment of a method for processing transaction data is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0036] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal (or mobile device) for implementing a method for processing transaction data is shown. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more (illustrated as 102a, 102b, ..., 102n in the figure) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0037] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuitry". The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuitry may be a single independent processing module, or may be incorporated in whole or in part into any of the other components of the computer terminal 10 (or mobile device). As described in the embodiments of the present application, the data processing circuitry serves as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0038] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the transaction data processing method in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, implementing the above-mentioned transaction data processing method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0039] The transmission device 106 is configured to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by the communications provider of the computer terminal 10. In one embodiment, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, the transmission device 106 may be a radio frequency (RF) module, which is configured to communicate with the Internet wirelessly.

[0040] The display may be, for example, a touch screen liquid crystal display (LCD), which enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0041] Under the above operating environment, this application provides Figure 2 The transaction data processing method shown. Figure 2 This is a flowchart of a method for processing transaction data according to Example 1 of the present application.

[0042] Step S201 : acquiring target transaction data of a target financial account, wherein the target financial account includes: a financial account with loan transactions, and the target transaction data includes: transaction data of the target financial account at t different time points, where t is a positive integer.

[0043] The target financial account mentioned above may refer to an account where a loan has already occurred, including but not limited to: a personal loan account, a mortgage account, a car loan account, and a credit card loan account.

[0044] The target transaction data mentioned above refers to the transaction information collected from the target financial account, which may include transaction data generated at t different time points, including but not limited to: (1) transaction amount: the amount of the transaction occurring at each time point; (2) transaction type: distinguishing whether it is a purchase, repayment or other type of transaction; (3) transaction frequency: the number of transactions within a certain time interval; (4) transaction channel: whether the transaction is completed through an online platform, an automated teller machine (ATM) or other channels; (5) account balance: the balance of the account at each time point. Overdue status: whether there are any bills or loans that have not been repaid on time.

[0045] The above target transaction data may cover t different time points, which means that the target transaction data is arranged in chronological order, forming a time series.

[0046] Step S202: Convert the target transaction data into a matrix to obtain a target matrix.

[0047] In this embodiment, the target transaction data can be converted into a matrix to obtain a target matrix, which aims to convert irregular transaction records into a format suitable for processing by a machine learning model. This can include data structuring, standardization, and feature engineering, specifically including: (1) Data structuring: Since the target transaction data may exist in the form of irregular tables or records, including information such as date, transaction amount, transaction type (such as deposit, withdrawal, consumption, repayment, etc.), and account balance, before conversion to a matrix, the target transaction data can be organized into a regular, time-ordered structure, for example, each row represents a time point, and each column represents a specific transaction feature.

[0048] To ensure data is available at all time points, if there are no transactions at a certain point in time, missing values can be filled. This can be done by filling in 0, the value at the previous time point, or the average. The data can then be standardized, such as by normalization or standardization, to ensure comparability between different feature values and prevent certain features from having a significant impact on the model due to their wide range of values.

[0049] The constructed target matrix can have t rows, each corresponding to a time point; the number of columns can depend on the number of features selected. Each element of the matrix can represent the value of a specific transaction feature at a specific time point. For example, the first column might represent the average monthly transaction amount, the second column the number of transactions per month, and so on. In this way, the target matrix can clearly demonstrate the transaction behavior and patterns of the target financial account at t different time points, providing structured and standardized data for input into the machine learning model (target evaluation model).

[0050] Step S203: Input the target matrix into the target assessment model, score the default risk of the target financial account through the target assessment model, and obtain a risk score value. The target assessment model includes: a first extraction layer, a second extraction layer, and an attention layer. The first extraction layer is deployed with a convolutional neural network, the second extraction layer is deployed with a long short-term memory network, and the attention layer is deployed with an attention mechanism.

[0051] The target assessment model described above can be a composite model consisting of multiple sub-layers, used to extract features from transaction data and perform risk assessment. The target assessment model can include a first extraction layer, a second extraction layer, and an attention layer. The first extraction layer (convolutional neural network, CNN) employs a convolutional neural network (CNN) for preliminary feature extraction, specifically to extract local patterns or behavioral features from the target matrix. For example, it can identify patterns such as unusually high spending or frequent small withdrawals in a financial account. The CNN captures these local features by sliding convolution kernels over the data and enhances the feature representation through nonlinear activation functions. The second extraction layer (long short-term memory network, LSTM) processes the CNN output and the time series characteristics of the target matrix itself through the LSTM. This allows for capturing patterns in account holders' transaction behavior over time, such as changes in repayment patterns or sudden increases in spending habits. The attention layer (attention mechanism) employs an attention mechanism, enabling the model to focus on and weight important features or time periods. In risk assessment of financial accounts, transaction data at different points in time may have varying importance; for example, recent transaction activity may better reflect current risk than activity from a long time ago. The attention mechanism allows the model to independently decide which parts of the data are more valuable for default risk prediction and assign higher weights to them, thereby improving the model's predictive ability and interpretability.

[0052] In this example, the target matrix (containing transaction data for financial accounts at t different time points) can be input into the first layer of the target assessment model (a convolutional neural network (CNN)), which extracts local characteristic patterns from the time series data. The extracted features (along with the original time series data) can be input into the second layer (a long short-term memory network (LSTM)) to extract the overall dynamic characteristics of the time series. Finally, the attention layer can weight the features extracted by the first two layers to ensure that the most critical signals for risk assessment are given priority consideration, thereby improving the overall performance of the model.

[0053] By integrating the strengths of CNN, LSTM, and the attention mechanism, the target assessment model can learn deeper and more complex patterns from complex financial transaction data, thereby providing a comprehensive and accurate score for the default risk of the target financial account, called a risk score. This score serves as a tool for banks or financial institutions to determine the risk level of an account.

[0054] Step S204: Determine the default risk level of the target financial account based on the risk score.

[0055] For example, different default risk levels may correspond to different preset score intervals, and the default risk level of the target financial account may be determined based on the preset score interval to which the risk score value belongs.

[0056] After determining the default risk level of the target financial account, corresponding risk control measures can be formulated based on the level. For example, for low-risk accounts, higher credit limits, preferential interest rates, or reward points can be offered. For high-risk accounts, stricter monitoring measures can be implemented, such as reducing credit limits, increasing the frequency of regular reviews, requiring customers to provide additional security or supporting documents, or even freezing accounts or prematurely recovering loans.

[0057] After determining the default risk level of the target financial account, financial products can also be pushed to the target financial account based on the risk level of the target financial account. Since high-risk accounts have lower purchasing power and low-risk accounts have higher purchasing power, financial products can be pushed based on the risk level of the financial account to avoid pushing financial products to all accounts, wasting system resources and low push accuracy.

[0058] Through the above steps, in this embodiment, post-loan transaction data of financial accounts is processed using a target assessment model composed of a convolutional neural network, a long short-term memory network, and an attention mechanism to assess the default risk level of the target financial account. This avoids the inefficient manual analysis of post-loan transaction data to assess the default risk level of the target financial account, thereby achieving the technical effect of improving the processing efficiency of post-loan transaction data of financial accounts. This further solves the technical problem of low processing efficiency of manual analysis of post-loan transaction data of financial accounts in related technologies.

[0059] Optionally, in the transaction data processing method provided in the embodiment of the present application, the target assessment model also includes: a fully connected layer, inputting the target matrix into the target assessment model, scoring the default risk of the target financial account through the target assessment model, and obtaining a risk score value, including: inputting the target matrix into the first extraction layer for feature extraction to obtain first feature data; inputting the first feature data into the second extraction layer to extract the time series features of the first feature data to obtain second feature data; inputting the second feature data into the attention layer to perform weighted calculation on the second feature data to obtain target feature data; inputting the target feature data into the fully connected layer to obtain a risk score value.

[0060] In this embodiment, the target matrix can be input into the first extraction layer (feature extraction) to extract local features from the target matrix. For example, to find common consumption patterns in transaction data, transaction peaks in specific time periods, etc., the data can be scanned by a moving filter (convolution kernel) to capture these features and convert them into first feature data. The features extracted by the first layer can then be further processed using a long short-term memory network (LSTM), and the time series features of the first feature data (for example, the changing trend of repayments over a period of time) can be extracted to obtain second feature data. The second feature data can then be weighted using an attention mechanism, and the output is the target feature data.

[0061] Figure 3 This is an architecture diagram of the target evaluation model provided in the embodiment of the present application, such as Figure 3 As shown in Figure 1, the fully connected layer, the final layer of the target assessment model, integrates the target feature data output by the previous layers. By connecting each neuron to all neurons in the next layer, this information is finally synthesized. At this layer, the model performs intensive mathematical operations to convert the target feature data into a single risk score, which directly reflects the account's default risk.

[0062] For example, the processed data (target matrix) can be input into a convolutional neural network (CNN), where the convolutional layers extract local features from the data. CNNs can capture underlying behavioral patterns from customer transaction records, such as frequent unusual spending and credit limit usage, helping to determine a customer's default risk. The feature sequence extracted by the CNN is then input into a long-short-term memory (LSTM) network. The LSTM can model the temporal relationships of a customer's historical behavior and capture trends in their creditworthiness. For example, recent delays in customer payments and credit limit usage can reflect a customer's likelihood of default. An attention mechanism is incorporated into the LSTM output, allowing the model to assign different weights to behavioral features within different time periods, thereby focusing on the most valuable information for risk prediction. This attention mechanism helps the model automatically identify the most critical features for predicting customer default risk.

[0063] Finally, the features obtained by combining CNN, LSTM and Attention are input into the fully connected layer for final risk prediction to obtain the risk score value.

[0064] For example, the features obtained by combining CNN, LSTM and Attention can be input into the fully connected layer for final risk prediction.

[0065] Level=tanh(WY t+1 +b)

[0066] Among them, Y t+1 is the output of the attention layer, b represents the bias value, W represents the weight, and h represents the hidden state

[0067] The output Level is a risk score value in the range [-1, 1]. Values closer to -1 indicate a higher risk of default, and risk control strategies should be implemented. Values closer to 1 indicate better credit and a lower risk level, and credit limits or preferential treatment may be appropriate.

[0068] This multi-layer deep learning model can capture complex patterns and nuances in the data, thereby improving the accuracy and reliability of default risk predictions.

[0069] Optionally, in the transaction data processing method provided in the embodiment of the present application, the first extraction layer includes: a convolution layer and a pooling layer, and the target matrix is input into the first extraction layer for feature extraction to obtain first feature data, including: inputting the target matrix into the convolution layer for feature extraction to obtain an extraction result; inputting the extraction result into the pooling layer for dimensionality reduction processing to obtain the first feature data.

[0070] For example, processed data can be fed into a convolutional neural network (CNN), where local features are extracted through the convolutional layer. CNNs can capture underlying behavioral patterns from customer transaction records, such as frequent abnormal spending and credit limit usage. These features can help determine a customer's default risk.

[0071] The calculation formula of the convolutional layer is:

[0072] I t =ReLU(X t *W t +b t )

[0073] Among them, I t is the output value after convolution, X t is the input vector, W t is the convolution kernel, b tis the bias term, and * represents the convolution operation.

[0074] In addition, in order to prevent overfitting and increase the generalization ability of the model, a pooling layer is added to reduce the dimensionality of the data to obtain the first feature data.

[0075] Through the combination of convolutional layers and pooling layers, the first extraction layer can effectively screen out features that have a significant impact on default risk assessment from the original transaction data of financial accounts, and pass them to the next layer in the form of first feature data after dimensionality reduction processing. This enables the deep learning model to learn the key to effective patterns from high-dimensional and complex input data, helps to improve the accuracy and efficiency of the model, and provides a basis for subsequent time series feature modeling and risk scoring.

[0076] Optionally, in the transaction data processing method provided in the embodiment of the present application, the second extraction layer includes: an input gate, a forget gate, an output gate and a memory unit, wherein the input gate is used to input the first feature data and store the first feature data in the memory unit, the forget gate is used to discard the feature data in the memory unit, the output gate is used to determine the second feature data based on the memory unit, and the memory unit is used to retain and transmit the feature data.

[0077] The input gate described above determines which information is written to the LSTM's memory cells. When processing the first feature data, the input gate generates a weight for each feature, which determines how heavily that feature is stored in the memory cell. If a feature is crucial for the current prediction, the input gate assigns it a higher weight; if it is less crucial, it assigns it a lower weight. This allows the LSTM to selectively store information useful for subsequent predictions in its memory cells.

[0078] The forget gate described above can be used to determine which information in a memory cell should be retained and which should be discarded. Since historical data may gradually lose its relevance over time, the forget gate can dynamically adjust the degree of retention of this information, ensuring that the model only focuses on the most relevant historical data, thereby avoiding information overload and wasted computing resources.

[0079] The output gate controls which information is output from the memory cell for use by subsequent layers (such as attention layers or fully connected layers). At each time step, the output gate generates an output weight based on the current memory cell state and input information. This weight determines which time series features have the greatest impact on the output at that time point. Through the output gate, the LSTM generates secondary feature data, which contains the critical time series information required for prediction at the current time point.

[0080] The memory cells described above are responsible for long-term information retention and transmission, maintaining data relevance and continuity even across long intervals in the sequence. When processing the first feature data, the memory cells store the information selected by the input gate and discard outdated or irrelevant information according to the instructions of the forget gate. This allows the memory cells to dynamically update their stored content, ensuring that each time step reflects the latest state and long-term trends of the sequence data.

[0081] In LSTM, the forget gate is represented as:

[0082] f t =σ(W f ·[h t-1 , l t ]+b f )

[0083] Among them, f t It is the forget gate, W f is the weight of the forget gate, h t-1 is the output value at the previous time, I t is the input value of the current time, b f is the bias term.

[0084] The input gate is represented as:

[0085] i t =σ(W i ·[h t-1 ,I t ]+b i )

[0086] Among them, i t is the input gate, W i is the weight of the input gate, b i is the bias term.

[0087] The above-mentioned memory units may include candidate memory units and update memory units.

[0088] The candidate memory unit is represented as:

[0089]

[0090] in, is a candidate memory unit, W c is the weight of the candidate memory unit, b c is the bias term.

[0091] The output gate is represented as:

[0092] o t =σ(W o ·[h t-1 , I t]+b o )

[0093] Among them, t is the output gate, W o is the weight of the input gate, b o is the bias term.

[0094] Updating the memory unit is expressed as:

[0095]

[0096] Updating the hidden state is expressed as:

[0097]

[0098] Through the synergistic effect of these components, LSTM is able to extract and retain the time series features critical for predicting default risk in financial accounts from the first feature data, generating the second feature data. This not only considers information at the current point in time but also incorporates the influence of historical data, enabling the model to more comprehensively and accurately assess default risk. This enables the model to effectively handle long-term dependencies, avoiding the vanishing or exploding gradient problems that can occur with traditional recurrent neural networks when processing long sequences of data, thereby improving the model's stability and predictive capabilities.

[0099] It should be noted that in the attention mechanism (Attention layer), the formula for calculating the attention score is as follows:

[0100]

[0101] Then, we can weight and sum the results to get the output of the attention layer:

[0102]

[0103] Optionally, in the transaction data processing method provided in the embodiment of the present application, after determining the default risk level of the target financial account based on the risk score value, it also includes: determining product recommendation data based on the risk level of the target financial account, wherein the product recommendation data includes: data of financial products to be recommended to the target financial account; obtaining transaction records of the target financial account on N trading platforms, where N is a positive integer; screening the N trading platforms based on the transaction records to obtain the screened trading platforms; and sending the product recommendation data to the screened trading platforms.

[0104] After determining the risk level of a target financial account based on the risk score, financial institutions can further utilize the risk level of the financial account to personalize financial product recommendations to customers. This not only strengthens risk management but also promotes the effective integration of customer relationship management and product recommendations. Specifically, financial institutions can match appropriate products to target financial accounts based on their risk level. For example, for customers with low risk levels, more financial products or those with higher transaction amounts can be recommended. For customers with high risk levels, fewer financial products or those with lower transaction amounts can be recommended. This avoids recommending financial products to all financial accounts, which would waste system resources, and ultimately achieves the goal of recommending appropriate financial products based on the purchasing power of financial accounts.

[0105] In an optional example, if the target financial account is at the first risk level (high risk level), it may be prohibited to recommend financial products to the target financial account. If the target financial account is at the second risk level (low risk level), financial products may be recommended to the target financial account based on the historical transaction records of the target financial account. The default risk of the first risk level is higher than the default risk of the second risk level.

[0106] To more accurately recommend financial products, it's also possible to analyze a financial account's trading habits and preferences across multiple platforms (N trading platforms, where N is a positive integer). This includes transaction records from multiple channels, such as online banking, third-party payment platforms, and e-commerce websites. These records provide insights into customer spending behavior and provide data support for customized product recommendations. For example, based on financial account transaction records, it's possible to analyze which platforms are most active or important to customers. This helps determine the optimal channel for product recommendations and ensures that recommendations effectively reach customers. The selected trading platforms may be those that customers frequently use, have high preferences for, or have large transaction volumes, thereby increasing the success rate of product recommendations and customer acceptance.

[0107] Finally, the product recommendation data prepared for the target financial account can be sent to the selected trading platforms via email, SMS notifications, software platform push, etc., ensuring that customers receive personalized financial product recommendations on the platforms they use most frequently. This improves the customer experience and increases the likelihood of product conversion.

[0108] Optionally, in the transaction data processing method provided in the embodiment of the present application, the target evaluation model is obtained in the following manner: obtaining transaction data of M financial accounts within a historical period to obtain historical transaction data; converting the historical transaction data into a three-dimensional matrix to obtain a historical transaction matrix; performing model training on the initial evaluation model based on the historical transaction data, and when the model training reaches a predetermined convergence condition, determining the trained initial evaluation model as the target evaluation model, wherein the initial evaluation model is a target evaluation model that has not been model trained.

[0109] In this embodiment, transaction data of M financial accounts can be collected, for example, data such as post-loan usage records, repayment history, and asset holdings of financial accounts. The data is then cleaned and standardized to ensure data quality and applicability. Afterwards, the historical transaction data can be converted into a matrix to obtain a historical transaction matrix.

[0110] An optional three-dimensional matrix, where C represents the number of customers, P represents the customer's characteristic dimension (for example, monthly consumption, monthly repayment, average monthly deposit, etc.), and T represents time (calculated on a monthly basis). Therefore, the three-dimensional matrix can be expressed as:

[0111] X=R C×P×T

[0112] Then, for each time t, the corresponding two-dimensional matrix is:

[0113] X t =R C×P

[0114] Customer characteristics P = {P1, P2, P3, ..., P 18}Includes the following information: P1 is the bank's contribution star rating for the month, P2 is the bank's service star rating for the month, P3 is the total assets for the month, P4 is the total deposits for the month, P5 is the total liabilities for the month, P6 is the average daily assets for the month, P7 is the average daily deposits for the month, P8 is the average daily liabilities for the month, P9 is the number of credit card transactions for the month, P 10 is the credit card transaction amount for the month, P 11 is the number of debit card transactions in the month, P 12 is the debit card transaction amount for the month, P 13 is the number of online transactions in that month, P 14 is the amount of online transactions in that month, P 15 is the number of offline transactions in the month, P 16 is the offline transaction amount of the month, P 17 is the number of overseas transactions in that month, P 18 The amount of overseas transactions in that month.

[0115] Afterwards, the historical transaction matrix can be fed into the initial evaluation model as a two-dimensional matrix at time t for model training. The model parameters can be adjusted using a backpropagation algorithm to bring the model's predictions closer to the actual labels (e.g., account default risk level). Model training can be repeated multiple times, with parameters adjusted based on the prediction error each iteration, gradually reducing the prediction error on the training data.

[0116] The above-mentioned predetermined convergence conditions refer to the stopping criteria during the training process. For example, when the model's loss function no longer decreases significantly in several consecutive iterations, or when the preset maximum number of iterations is reached, the model training will stop. When the model training reaches the predetermined convergence conditions, it can be determined that the model has learned enough patterns in historical trading data and the parameter adjustments have stabilized. At this point, the trained initial evaluation model can be regarded as the target evaluation model.

[0117] Figure 4 This is a flow chart of the model training provided in the embodiment of the present application. Figure 4 As shown, the data can be input and calculated through the convolution layer, pooling layer, LSTM layer, attention mechanism layer (abbreviated as attention layer), and fully connected layer to determine whether the training results converge.

[0118] After the target assessment model is trained, it can be used to predict the default risk of new accounts in real time or in batches, providing financial institutions with an important risk management tool.

[0119] Optionally, in the transaction data processing method provided in the embodiment of the present application, after obtaining the transaction data of M financial accounts within a historical period and obtaining the historical transaction data, it also includes: determining the overdue repayment period of each financial account in the M financial accounts and the number of transactions of each financial account based on the historical transaction data; dividing the M financial accounts based on the overdue repayment period of each financial account and the number of transactions of each financial account to obtain a division result; and marking the default risk level of each financial account based on the division result.

[0120] In this embodiment, financial accounts can be labeled according to their default risk level, with the default risk levels being: non-performing, average, and high-quality. Non-performing customers are selected from customers who have been overdue for more than one month within six months. Average customers can be randomly selected from regular customers with at least five monthly active accounts. High-quality customers are selected from a list of high-quality customers, with non-performing customers labeled -1, average customers 0, and high-quality customers 1. It should be noted that historical transaction data can also be divided into a training set (80%) and a test set (20%).

[0121] This application can realize risk prediction in post-loan management of financial accounts by combining three neural network technologies: CNN, LSTM and Attention, and achieve the following effects: 1. Improve prediction accuracy: Through multi-level and multi-dimensional feature learning, the model can more accurately identify the customer's default risk, avoiding the defect that traditional models cannot capture complex customer behavior patterns. 2. Improve work efficiency: Because the model can automatically predict risks, banks can reduce the need for manual review and intervention, and improve the automation and work efficiency of post-loan management. Especially for the processing of large amounts of customer data, this automated method can significantly save time and labor costs. 3. Comprehensively improve risk control capabilities: By integrating multiple data (such as customer transaction records, financial account usage, repayment history, etc.) for modeling, it can comprehensively assess customer risks from multiple angles and dimensions, and has stronger differentiation capabilities than traditional single models.

[0122] The transaction data processing method provided in the embodiment of the present application has the following advantages: 1. Time series feature modeling: LSTM can make full use of the customer's historical behavior data and capture the long-term dependencies in the time series, thereby effectively predicting the customer's future default risk. 2. Strong adaptability: This method is highly adaptable and can process different types of customer data and adjust according to actual business needs. It has high versatility and practicality. 3. Strong interpretability: The introduction of the attention mechanism enhances the interpretability of the model. It can show which key behaviors or time periods the model focuses on when making predictions, thereby helping banks understand the prediction results and facilitating further decision-making and customer communication. 4. Automated feature selection: By introducing the attention mechanism, the model can automatically focus on the features most relevant to risk prediction, avoiding manual intervention and the limitations of traditional rule models. 5. Real-time warning function: Combined with the time series characteristics of LSTM, it can provide real-time risk warnings to help banks promptly identify potential high-risk customers in post-loan management and reduce default rates.

[0123] In summary, this application uses a convolutional neural network (CNN) in the target assessment model to effectively process customer transaction data by extracting features. The long short-term memory (LSTM) network can sequentially model customer historical behavior and capture temporal correlations. The attention mechanism can weightedly select different pieces of information, focusing on the most valuable components for prediction. This can significantly improve the efficiency of risk assessment and risk management capabilities of financial institutions in post-loan financial account lending.

[0124] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0125] Example 2

[0126] The present application also provides a transaction data processing device. It should be noted that the transaction data processing device of the present application can be used to execute the transaction data processing method provided in the present application. The transaction data processing device provided in the present application is described below.

[0127] According to an embodiment of the present application, a device for implementing the above-mentioned transaction data processing method is also provided. Figure 5 As shown, the device includes: a first acquisition unit 51, a conversion unit 52, a scoring unit 53 and a first determination unit 54.

[0128] The first acquisition unit 51 is configured to acquire target transaction data of a target financial account, wherein the target financial account includes a financial account with a loan transaction, and the target transaction data includes transaction data of the target financial account at t different time points, where t is a positive integer.

[0129] A conversion unit 52, configured to convert the target transaction data into a matrix to obtain a target matrix;

[0130] Scoring unit 53 is configured to input the target matrix into a target assessment model, score the default risk of the target financial account using the target assessment model, and obtain a risk score value. The target assessment model includes: a first extraction layer, a second extraction layer, and an attention layer. The first extraction layer is deployed with a convolutional neural network, the second extraction layer is deployed with a long short-term memory network, and the attention layer is deployed with an attention mechanism.

[0131] The first determining unit 54 is configured to determine the default risk level of the target financial account based on the risk score value.

[0132] In the transaction data processing device provided in the embodiment of the present application, a first acquisition unit 51 can be used to obtain target transaction data of a target financial account, wherein the target financial account includes: a financial account with loan behavior, and the target transaction data includes: transaction data of the target financial account at t different time points, where t is a positive integer; a conversion unit 52 is used to convert the target transaction data into a matrix to obtain a target matrix; a scoring unit 53 is used to input the target matrix into a target assessment model, and score the default risk of the target financial account through the target assessment model to obtain a risk score value, wherein the target assessment model includes: a first extraction layer, a second extraction layer, and an attention layer, wherein the first extraction layer is deployed with a convolutional neural network, the second extraction layer is deployed with a long short-term memory network, and the attention layer is deployed with an attention mechanism; a first determination unit 54 is used to determine the default risk level of the target financial account based on the risk score value. This solves the technical problem of low processing efficiency in the related art of using manual analysis to analyze and process the transaction data of financial accounts after lending. In this embodiment, the transaction data of financial accounts after lending are processed by a target evaluation model composed of a convolutional neural network, a long short-term memory network, and an attention mechanism to evaluate the default risk level of the target financial account, thereby avoiding the purpose of manually analyzing the transaction data of financial accounts after lending to evaluate the default risk level of the target financial account, thereby achieving the technical effect of improving the processing efficiency of the transaction data of financial accounts after lending.

[0133] Optionally, in the transaction data processing device provided in the embodiment of the present application, the target evaluation model also includes: a fully connected layer, and the scoring unit includes: a first extraction subunit, used to input the target matrix into the first extraction layer for feature extraction to obtain first feature data; a second extraction subunit, used to input the first feature data into the second extraction layer to extract the time series features of the first feature data to obtain second feature data; a calculation subunit, used to input the second feature data into the attention layer to perform weighted calculation on the second feature data to obtain target feature data; a processing subunit, used to input the target feature data into the fully connected layer to obtain a risk score value.

[0134] Optionally, in the transaction data processing device provided in the embodiment of the present application, the first extraction layer includes: a convolution layer and a pooling layer, and the first extraction sub-unit includes: an extraction module, used to input the target matrix into the convolution layer for feature extraction to obtain an extraction result; a dimensionality reduction module, used to input the extraction result into the pooling layer for dimensionality reduction processing to obtain first feature data.

[0135] Optionally, in the transaction data processing device provided in the embodiment of the present application, the second extraction layer includes: an input gate, a forget gate, an output gate and a memory unit, wherein the input gate is used to input the first feature data and store the first feature data in the memory unit, the forget gate is used to discard the feature data in the memory unit, the output gate is used to determine the second feature data based on the memory unit, and the memory unit is used to retain and transmit the feature data.

[0136] Optionally, in the transaction data processing device provided in the embodiment of the present application, the transaction data processing device further includes: a second determination unit, for determining the default risk level of the target financial account based on the risk score value, and then determining product recommendation data based on the risk level of the target financial account, wherein the product recommendation data includes: data of financial products to be recommended to the target financial account; a second acquisition unit, for obtaining transaction records of the target financial account on N trading platforms, wherein N is a positive integer; a screening unit, for screening the N trading platforms based on the transaction records to obtain the screened trading platforms; and a sending unit, for sending the product recommendation data to the screened trading platforms.

[0137] Optionally, in the transaction data processing device provided in the embodiment of the present application, the target evaluation model is obtained through the following sub-units: an acquisition sub-unit, used to acquire the transaction data of M financial accounts within a historical time period to obtain historical transaction data; a conversion sub-unit, used to convert the historical transaction data into a three-dimensional matrix to obtain a historical transaction matrix; a training sub-unit, used to perform model training on the initial evaluation model based on the historical transaction data, and when the model training reaches a predetermined convergence condition, the trained initial evaluation model is determined as the target evaluation model, wherein the initial evaluation model is a target evaluation model that has not been model trained.

[0138] Optionally, in the transaction data processing device provided in the embodiment of the present application, the transaction data processing device further includes: a determination subunit, used to obtain transaction data of M financial accounts within a historical period, and after obtaining the historical transaction data, determine the overdue repayment period of each financial account in the M financial accounts and the number of transactions of each financial account based on the historical transaction data; a division subunit, used to divide the M financial accounts based on the overdue repayment period of each financial account and the number of transactions of each financial account to obtain a division result; and a marking subunit, used to mark the default risk level of each financial account based on the division result.

[0139] It should be noted that the above-mentioned first acquisition unit 51, conversion unit 52, scoring unit 53, and first determination unit 54 correspond to steps S201 to S204 in the first embodiment. The examples and application scenarios implemented by the units and corresponding steps are the same, but are not limited to the contents disclosed in the first embodiment. It should be noted that the above-mentioned modules or units can be hardware components or software components stored in a memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above-mentioned modules can also be run as part of the device in the computer terminal 10 provided in the first embodiment.

[0140] Example 3

[0141] An embodiment of the present application may provide an electronic device, Figure 6 This is a structural block diagram of an electronic device according to an embodiment of the present application. Figure 6 As shown, the electronic device may include: one or more ( Figure 6 Only one is shown) processor 602, memory 604, storage controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0142] Among them, the memory can be used to store software programs and modules, such as program instructions / modules corresponding to the methods and devices in the embodiments of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, implementing the above-mentioned method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0143] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: obtain target transaction data of the target financial account, wherein the target financial account includes: a financial account with loan behavior, and the target transaction data includes: transaction data of the target financial account at t different time points, t is a positive integer; convert the target transaction data into a matrix to obtain a target matrix; input the target matrix into a target assessment model, and score the default risk of the financial account through the target assessment model to obtain a risk score value, wherein the target assessment model includes: a first extraction layer, a second extraction layer, and an attention layer, the first extraction layer is deployed with a convolutional neural network, the second extraction layer is deployed with a long short-term memory network, and the attention layer is deployed with an attention mechanism; based on the risk score value, determine the default risk level of the target financial account.

[0144] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: The target assessment model also includes: a fully connected layer, inputting the target matrix into the target assessment model, scoring the default risk of the target financial account through the target assessment model, and obtaining a risk score value, including: inputting the target matrix into the first extraction layer for feature extraction to obtain first feature data; inputting the first feature data into the second extraction layer to extract the time series features of the first feature data to obtain second feature data; inputting the second feature data into the attention layer to perform weighted calculation on the second feature data to obtain target feature data; inputting the target feature data into the fully connected layer to obtain a risk score value.

[0145] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: the first extraction layer includes: a convolution layer and a pooling layer, and the target matrix is input into the first extraction layer for feature extraction to obtain the first feature data, including: inputting the target matrix into the convolution layer for feature extraction to obtain the extraction result; inputting the extraction result into the pooling layer for dimensionality reduction processing to obtain the first feature data.

[0146] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: The second extraction layer includes: an input gate, a forget gate, an output gate and a memory unit, wherein the input gate is used to input the first feature data and store the first feature data in the memory unit, the forget gate is used to discard the feature data in the memory unit, the output gate is used to determine the second feature data based on the memory unit, and the memory unit is used to retain and transmit the feature data.

[0147] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: after determining the default risk level of the target financial account based on the risk score value, it also includes: determining product recommendation data based on the risk level of the target financial account, wherein the product recommendation data includes: data of financial products to be recommended to the target financial account; obtaining transaction records of the target financial account on N trading platforms, where N is a positive integer; screening the N trading platforms based on the transaction records to obtain the screened trading platforms; and sending the product recommendation data to the screened trading platforms.

[0148] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: the target evaluation model is obtained by: obtaining the transaction data of M financial accounts within the historical time to obtain historical transaction data; converting the historical transaction data into a three-dimensional matrix to obtain a historical transaction matrix; performing model training on the initial evaluation model based on the historical transaction data, and when the model training reaches the predetermined convergence condition, determining the trained initial evaluation model as the target evaluation model, wherein the initial evaluation model is the target evaluation model that has not been model trained.

[0149] The processor can also call information and applications stored in the memory through the transmission device to perform the following steps: after obtaining the transaction data of M financial accounts within the historical time, the historical transaction data is also included: based on the historical transaction data, determining the overdue repayment period of each financial account in the M financial accounts and the number of transactions of each financial account; based on the overdue repayment period of each financial account and the number of transactions of each financial account, the M financial accounts are divided to obtain a division result; based on the division result, the default risk level of each financial account is marked.

[0150] By adopting the embodiment of the present application, the post-loan transaction data of the financial account is processed by a target evaluation model composed of a convolutional neural network, a long short-term memory network, and an attention mechanism to evaluate the default risk level of the target financial account, thereby avoiding the purpose of manually analyzing the post-loan transaction data of the financial account to evaluate the default risk level of the target financial account, thereby achieving the technical effect of improving the processing efficiency of the post-loan transaction data of the financial account.

[0151] It can be understood by those skilled in the art that Figure 6 The structure shown is for illustration only, and the electronic device may also be a terminal device such as a smart phone, a tablet computer, a PDA, a mobile Internet device (MID), or a PAD. Figure 6 It does not limit the structure of the above electronic device. For example, the electronic device may also include Figure 6More or fewer components (such as network interfaces, display devices, etc.) shown in, or with Figure 6 Different configurations shown.

[0152] A person skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0153] Example 4

[0154] The embodiment of the present application further provides a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the transaction data processing method provided in the first embodiment.

[0155] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0156] The present application also provides a computer program product which, when executed on a data processing device, is suitable for executing the steps of the method for processing transaction data.

[0157] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0158] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0159] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0160] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0161] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0162] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0163] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for processing transaction data, characterized in that: include: Obtaining target transaction data of a target financial account, wherein the target financial account includes: a financial account with which a loan has been made, and the target transaction data includes: transaction data of the target financial account at t different time points, where t is a positive integer; Converting the target transaction data into a matrix to obtain a target matrix; Inputting the target matrix into a target assessment model, scoring the default risk of the target financial account using the target assessment model to obtain a risk score value, wherein the target assessment model includes: a first extraction layer, a second extraction layer, and an attention layer, wherein the first extraction layer is deployed with a convolutional neural network, the second extraction layer is deployed with a long short-term memory network, and the attention layer is deployed with an attention mechanism; Based on the risk score, a default risk level of the target financial account is determined.

2. The processing method according to claim 1, characterized in that The target assessment model further includes: a fully connected layer, which inputs the target matrix into the target assessment model, scores the default risk of the target financial account using the target assessment model, and obtains a risk score value, including: Inputting the target matrix into the first extraction layer to perform feature extraction to obtain first feature data; Inputting the first feature data into the second extraction layer to extract the time series features of the first feature data to obtain second feature data; Inputting the second feature data into the attention layer to perform weighted calculation on the second feature data to obtain target feature data; The target feature data is input into the fully connected layer to obtain the risk score value.

3. The processing method according to claim 2, characterized in that The first extraction layer includes: a convolution layer and a pooling layer. The target matrix is input into the first extraction layer for feature extraction to obtain first feature data, including: Inputting the target matrix into the convolutional layer for feature extraction to obtain an extraction result; The extraction result is input into the pooling layer for dimensionality reduction processing to obtain the first feature data.

4. The processing method according to claim 2, characterized in that The second extraction layer includes: an input gate, a forget gate, an output gate and a memory unit, wherein the input gate is used to input the first feature data and store the first feature data in the memory unit, the forget gate is used to discard the feature data in the memory unit, the output gate is used to determine the second feature data based on the memory unit, and the memory unit is used to retain and transmit the feature data.

5. The processing method according to claim 1, characterized in that After determining the default risk level of the target financial account based on the risk score, the method further includes: Determining product recommendation data based on the risk level of the target financial account, wherein the product recommendation data includes: data of financial products to be recommended to the target financial account; Obtaining transaction records of the target financial account on N trading platforms, where N is a positive integer; Filtering the N trading platforms based on the transaction records to obtain filtered trading platforms; The product recommendation data is sent to the filtered trading platform.

6. The processing method according to claim 1, characterized in that The target evaluation model is obtained in the following way: Obtain transaction data of M financial accounts within a historical period to obtain historical transaction data; Converting the historical transaction data into a three-dimensional matrix to obtain a historical transaction matrix; The initial evaluation model is trained based on the historical transaction data. When the model training reaches a predetermined convergence condition, the trained initial evaluation model is determined as the target evaluation model, wherein the initial evaluation model is a target evaluation model that has not been trained.

7. The processing method according to claim 6, characterized in that After obtaining the transaction data of M financial accounts within a historical period, the following steps are also included: Determining, based on the historical transaction data, the duration of overdue repayment for each of the M financial accounts and the number of transactions for each of the financial accounts; Dividing the M financial accounts based on the duration of overdue repayment of each financial account and the number of transactions of each financial account to obtain a division result; Based on the classification result, the default risk level of each financial account is marked.

8. A transaction data processing device, characterized in that: include: A first acquiring unit is configured to acquire target transaction data of a target financial account, wherein the target financial account includes a financial account with a loan transaction, and the target transaction data includes transaction data of the target financial account at t different time points, where t is a positive integer; a conversion unit, configured to convert the target transaction data into a matrix to obtain a target matrix; a scoring unit, configured to input the target matrix into a target assessment model, score the default risk of the target financial account using the target assessment model, and obtain a risk score value, wherein the target assessment model includes: a first extraction layer, a second extraction layer, and an attention layer, wherein the first extraction layer is deployed with a convolutional neural network, the second extraction layer is deployed with a long short-term memory network, and the attention layer is deployed with an attention mechanism; The first determining unit is configured to determine a default risk level of the target financial account based on the risk score value.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored executable program, wherein when the executable program is run, the device where the computer-readable storage medium is located is controlled to execute the transaction data processing method according to any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that When the computer instructions are executed by a processor, the steps of the transaction data processing method according to any one of claims 1 to 7 are implemented.

Images