Risk assessment method and device, model training method and device, equipment and storage medium

Abstract

The invention provides a risk assessment method and device, a model training method and device, equipment and a storage medium. The risk assessment method is used for performing risk assessment on target software installed in intelligent terminal equipment of a financial institution. Specifically, when risk assessment is carried out, firstly, business transaction information of a first business of a second business processed by target software can be obtained; and then, prediction can be performed according to the business transaction information of the first business, and the business volume of the target software at the target moment is determined. Based on the business volume of the target software at the target moment, the risk condition of the target software at the target moment can be predicted. Through the method, before the target software is online, the business volume and the risk of the target software can be predicted according to the related data of the first business similar to the second business, and normal operation of the business after the target software is online is ensured.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a risk assessment method, a model training method, an apparatus, a device and a storage medium. Background Art

[0002] Financial institutions such as banks have the ability to handle a variety of businesses. For traditional financial institutions, many businesses require sales staff to manually handle them for users. In the process of handling business, sales staff can show multiple forms to users and ask users to fill them out. Obviously, this method wastes human resources and is inefficient.

[0003] With the advancement of digitalization, many businesses can be handled by devices instead of humans. The devices used by financial institutions to handle business for users instead of humans can be called smart terminal devices. Users can complete the handling of multiple services on a single smart terminal device, which improves the efficiency of service handling and saves the human resources of branch salesmen.

[0004] The functions of the smart terminal devices are realized by the software installed in the smart terminal devices. With the launch of different services, the smart terminal devices are required to have new functions, which also requires the installation of new software in the smart terminal devices. Summary of the invention

[0005] In view of this, the present application provides a risk assessment method, a model training method, an apparatus, a device and a storage medium, which aim to predict the risks of software installed in smart terminal devices.

[0006] In a first aspect, the present application provides a risk assessment method, the method being used to perform risk assessment on target software installed in a smart terminal device of a financial institution, the method comprising:

[0007] Acquiring business transaction information of a first business, where a similarity between the first business and a second business is higher than a threshold, and the target software is used to process the second business;

[0008] Predicting the business volume of the target software at a target time based on the business transaction information of the first business;

[0009] Based on the business volume of the target software at the target time, the risk situation of the target software at the target time is predicted.

[0010] In some possible implementations, after the target software is launched, the method further includes:

[0011] Obtaining the actual business volume and actual risk situation of the target software at the target time;

[0012] Based on the actual business volume and the actual risk situation, fine-tuning the prediction model, the prediction model is used to predict the risk situation of the target software, and the prediction model is a long short-term memory network LSTM model;

[0013] The fine-tuned prediction model is used to predict the risk of the target software.

[0014] In a second aspect, the present application provides a model training method, the method is used to train a risk prediction model, the risk prediction model is used to perform risk assessment on target software installed in a smart terminal device of a financial institution, the method comprising:

[0015] Acquire multiple risk features of a first business, the similarity between the first business and a second business is higher than a threshold, and the target software is used to process the second business;

[0016] preprocessing the multiple risk features;

[0017] Based on the risk characteristics, a long short-term memory network LSTM model is pre-trained to obtain a source model, and the source model is used to predict the risk of the target software before the target software goes online;

[0018] After the target software is launched, obtaining actual risk information of the target software;

[0019] Based on the actual risk information, the source model is fine-tuned to obtain the risk prediction model.

[0020] In some possible implementations, the acquiring multiple risk characteristics of the first service includes:

[0021] Obtain multiple original features;

[0022] Calculate the correlation between each of the original features and the risk respectively;

[0023] Based on the correlation, the risk feature is determined from the plurality of raw features.

[0024] In a third aspect, the present application provides a risk assessment device, the device being used to perform risk assessment on target software installed in a smart terminal device of a financial institution, the device comprising:

[0025] an acquisition unit, configured to acquire business transaction information of a first business, wherein a similarity between the first business and a second business is higher than a threshold, and the target software is configured to process the second business;

[0026] A business volume prediction unit, configured to predict the business volume of the target software at a target time based on the business transaction information of the first business;

[0027] The risk prediction unit is used to predict the risk situation of the target software at the target time based on the business volume of the target software at the target time.

[0028] In some possible implementations, the device further includes a fine-tuning unit;

[0029] The acquisition unit is further used to acquire the actual business volume and actual risk situation of the target software at the target time;

[0030] The fine-tuning unit is used to fine-tune the prediction model based on the actual business volume and the actual risk situation. The prediction model is used to predict the risk situation of the target software. The prediction model is a long short-term memory network LSTM model; the risk situation of the target software is predicted using the fine-tuned prediction model.

[0031] In a fourth aspect, the present application provides a model training device, the device is used to train a risk prediction model, the risk prediction model is used to perform risk assessment on target software installed in a smart terminal device of a financial institution, the device comprising:

[0032] An acquisition unit, used for acquiring a plurality of risk features of a first business, wherein a similarity between the first business and a second business is higher than a threshold, and the target software is used for processing the second business;

[0033] A preprocessing unit, configured to preprocess the plurality of risk features;

[0034] A training unit, used for pre-training a long short-term memory network (LSTM) model based on the risk characteristics to obtain a source model, wherein the source model is used for performing risk prediction on the target software before the target software goes online;

[0035] The acquisition unit is further used to acquire actual risk information of the target software after the target software is launched;

[0036] A fine-tuning unit is used to fine-tune the source model based on the actual risk information to obtain the risk prediction model.

[0037] In some possible implementations, the acquisition unit is specifically configured to acquire a plurality of original features; respectively calculate the correlation between each of the original features and the risk; and determine the risk feature from the plurality of original features based on the correlation.

[0038] In a fifth aspect, the present application provides a device comprising a memory and a processor, wherein the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the risk assessment method described in any one of the first aspects, or executes the model training method described in any one of the second aspects.

[0039] In a sixth aspect, the present application provides a computer storage medium storing a code. When the code is executed, the device executing the code implements the risk assessment method described in any one of the first aspects, or executes the model training method described in any one of the second aspects.

[0040] In the seventh aspect, a computer program product comprising instructions is provided, which, when executed on a computer, enables the computer to execute the risk assessment method described in any one of the first aspects, or execute the model training method described in any one of the second aspects.

[0041] The present application provides a risk assessment method, a model training method, an apparatus, a device and a storage medium. Among them, the risk assessment method is used to perform risk assessment on the target software installed in the intelligent terminal device of the financial institution. Specifically, when performing risk assessment, the business transaction information of the first business and the second business processed by the target software can be obtained first. Then, the business volume of the target software at the target time can be predicted based on the business transaction information of the first business. Based on the business volume of the target software at the target time, the risk situation of the target software at the target time can be predicted. Through the above method, before the target software goes online, the business volume and risk of the target software can be predicted based on the relevant data of the first business similar to the second business. In this way, before the target software goes online, the business situation and risk situation of the target software after going online can be predicted based on the situation of similar businesses, and it can be judged whether the target software will have risks after going online, so as to adjust the target software in time to ensure the normal operation of the business after the target software goes online. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 A schematic diagram of a process flow of a risk assessment method provided in an embodiment of the present application;

[0044] Figure 2A schematic diagram of a flow chart of a model training method provided in an embodiment of the present application;

[0045] Figure 3 A schematic diagram of the structure of a risk assessment device provided in an embodiment of the present application;

[0046] Figure 4 A structural schematic diagram of a model training device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0047] For financial enterprises, traditional business processing usually involves a lot of form filling, signatures and complicated processes, which not only consumes a lot of customers' time, but also increases the labor costs of enterprises. In this era, in order to better meet customer needs and improve service efficiency, financial enterprises are constantly exploring and innovating channels to directly connect with customers.

[0048] In order to solve the problem of complicated business handling procedures, many financial enterprises use smart terminal devices to handle business. Smart terminal devices refer to terminal devices developed specifically for the business of financial enterprises. One or more software can be run on smart terminal devices to provide business handling services to users through the functions of one or more software. In addition, the hardware of smart terminal devices is specially designed for application scenarios, which can better adapt to application scenarios.

[0049] If a new service is required on a smart terminal device, the software corresponding to the new service needs to be installed on the smart terminal device. However, the newly installed software may have risks and needs to be subject to risk assessment. If the newly installed software fails during operation, a production safety accident may occur. For financial enterprises such as banks, production safety accidents need to be avoided as much as possible.

[0050] On the one hand, the software itself can be shared and evaluated from the perspective of software design. On the other hand, it is also necessary to predict the working conditions of the software after installation. However, risk assessment needs to be based on historical data. Before the software is launched, it is impossible to obtain the historical data of the software, and it is also impossible to predict the risks of the software.

[0051] In order to solve the above problems, the embodiments of the present application provide a risk assessment method and a model training method. Among them, the risk assessment method is used to predict the risk of the target software before the target software goes online. The model training method is used to train the model used in the risk assessment process. The above risk assessment method can be applied to a risk assessment device, and the model training method can be applied to a model training device. Optionally, the risk assessment device can run on a server, and can be specifically applied to a test server or a server cluster.

[0052] The risk assessment method provided by the embodiment of the present application is described below from the perspective of the risk assessment device. Obviously, the described embodiment is only a part of the embodiment of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0053] See also Figure 1 , Figure 1 A method flow chart of a risk assessment method provided in an embodiment of the present application includes:

[0054] S101: Acquire business transaction information of a first business.

[0055] Before the first service is launched, in order to predict the risk of the target software, it is necessary to first predict the service carried by the target software. To this end, the business transaction information of the first service can be obtained first. The first service refers to a service similar to the service carried by the target software. That is, if the target software is used to handle the second service, then the first service with the highest similarity to the second service can be found from multiple existing services, and the business transaction information of the first service can be obtained.

[0056] The first business is similar to the second business and has been processed through software. Since the first business has been processed through software, the background system of the financial institution records relevant information of the first business. Therefore, the risk assessment device can obtain business transaction information of the first business.

[0057] The business transaction information of the first business is relevant information indicating the transaction status and risk status of the first business. Based on the business transaction information of the first business, the business volume of the second business and the risk status after the second business is processed through the target software can be predicted. Optionally, the business transaction information of the first business may include information such as transaction volume, transaction success rate, failed transaction volume, and transaction response time.

[0058] S102: Based on the business transaction information of the first business, predict the business volume of the target software at the target time.

[0059] After obtaining the business transaction information of the first business, the business transaction volume of the target software at the target time can be predicted. The target time can be any time after the target software is launched. The target time can be obtained based on the obtained business transaction information of the first business. For example, if the business transaction information of the first business in the past week is obtained, the business volume of the target software at any time in the week can be predicted.

[0060] The business volume of the target software at the target time refers to the business volume that the target software needs to process at the target time, corresponding to the business volume of the second business at the target time. It is understandable that as the amount of business processed increases, the probability of failure or abnormality of the target software will also increase. Therefore, by predicting the business volume that the target software needs to process at the target time based on the business transaction information of the first business, the load of the target software can be determined, thereby determining whether the target software will fail or be abnormal under the corresponding load.

[0061] Furthermore, when fine-tuning the model later, the predicted business volume and the actual business volume can be compared to determine whether the business transaction information of the first business can be used for risk prediction of the second business. For an introduction to this part, please refer to the following, which will not be repeated here.

[0062] In some possible implementations, the business volume of the target software at the target time can be predicted by the model. Optionally, the business volume of the target software at the target time can be predicted by a long short-term memory network (LSTM) model. In some possible implementations, the LSTM model can include two tasks, one of which corresponds to step S102 and is used to predict the business volume of the target software at the target time, and the other task corresponds to step S103 and is used to predict the risk situation of the target software at the target time.

[0063] S103: Predicting the risk status of the target software at the target time based on the business volume of the target software at the target time.

[0064] After obtaining the business volume of the target software at the target time, the risk situation of the target software at the target time can be predicted. Specifically, the business transaction information of the first business may include the response time and success rate of the first business. Then, when predicting the risk situation of the target software, the response time and success rate of the first business similar to the second business can be used to determine whether the response time and success rate of the second business at the target time are normal, thereby determining whether the target software will have risks at the target time.

[0065] Optionally, the risk situation of the target software at the target time can be predicted by the model. Optionally, the risk situation of the target software at the target time can be predicted by the LSTM model.

[0066] The present application provides a risk assessment method. Among them, the risk assessment method is used to perform risk assessment on the target software installed in the intelligent terminal device of the financial institution. Specifically, when performing risk assessment, the business transaction information of the first business and the second business processed by the target software can be obtained first. Then, the business volume of the target software at the target time can be predicted based on the business transaction information of the first business. Based on the business volume of the target software at the target time, the risk situation of the target software at the target time can be predicted. Through the above method, before the target software goes online, the business volume and risk of the target software can be predicted based on the relevant data of the first business similar to the second business. In this way, before the target software goes online, the business situation and risk situation of the target software after going online can be predicted based on the situation of similar businesses, and it can be judged whether the target software will have risks after going online, so as to adjust the target software in time to ensure the normal operation of the business after the target software goes online.

[0067] In the implementation method described above, for the target software that carries the second business, the relevant data of the first business with a high degree of similarity to the second business can be used to make predictions before the target software goes online. In this way, although the risk of the target software can be predicted without the data of the second business, it is not predicted using the actual data of the second business after all, and there is still a difference between the predicted result and the actual result. For this reason, after the target software goes online, fine-tuning is performed using the actual working conditions of the target software.

[0068] Specifically, after the target software is launched, the actual business volume and actual risk situation of the target software at the target time can be obtained. Based on the actual business volume and actual risk situation of the target software, the model can be fine-tuned. Specifically, the predicted business volume and the actual business volume can be subtracted, and / or the predicted risk situation and the actual risk situation can be subtracted, and the result of the subtraction can be used as a loss function (loss), and the model can be fine-tuned through the loss function to obtain a model that can accurately predict the risk situation of the target software. Then, the fine-tuned model prediction can be used to continue predicting the risk situation of the target software. In this way, by fine-tuning the model, the model can be matched with the actual working conditions of the target software, thereby accurately predicting the risk situation of the target software.

[0069] In the implementation described above, the business volume of the target software at the target time and the risk situation of the target software at the target time can be obtained through model prediction. The embodiment of the present application also provides a model training method, which can be used to train the above-mentioned model for predicting business volume and risk situation. The following is a detailed introduction.

[0070] See also Figure 2 , Figure 2A method flow chart of a model training method provided in an embodiment of the present application includes:

[0071] S201: Acquire multiple risk characteristics of a first business.

[0072] In order to predict the risk of the target software carrying the second business, the first business similar to the second business can be determined first, and multiple risk features of the first business can be obtained. Among them, similar to the second business means that the similarity between the first business and the second business is higher than a threshold. The similarity between the first business and the second business can be determined based on information such as the business types of the first business and the second business, the handling information of the first business and the second business, and the names of the first business and the second business.

[0073] The multiple risk characteristics of the first business refer to the characteristics of the first business that can reflect the risk situation of the first business. Optionally, the risk characteristics of the first business include transaction volume, transaction success rate, transaction response time, and failed transaction volume of the first business.

[0074] In some possible implementations, the first business may have multiple features. Then the model training device is required to select features related to the wind direction of the first business from the multiple features of the first business. Specifically, the model training device can first obtain multiple original features of the first business. Then, the correlation between each original feature and the risk can be calculated separately. For example, the Pearson correlation coefficient method can be used. The correlation between each original feature and the risk situation of the first business is calculated using the Pearson correlation coefficient method. Then, all features can be sorted according to the size of the correlation, and the risk features can be determined based on the sorting results. For example, the k original features with the highest correlation can be selected as target features.

[0075] S202: Preprocess multiple risk features.

[0076] After multiple risk features are determined, the multiple risk features can be preprocessed. For example, the risk feature data can be cleaned and normalized to remove control and outliers in the risk feature data. If the model to be trained is a time-related model, such as an LSTM model, the risk features can also be mapped to a time-related vector or matrix. For example, multiple risk features at the same time can be organized into a vector. Alternatively, multiple risk features at n adjacent moments can be organized into an n-column matrix.

[0077] S203: Based on the risk characteristics, pre-train the LSTM model to obtain a source model.

[0078] After preprocessing the risk features, the LSTM model can be trained based on the risk features to obtain a source model. The source model refers to a model trained using the risk features of the first business.

[0079] Optionally, the hidden state dimension of the optimized input layer LSTM cell is 96, the hidden state dimensions of the first hidden layer and the second hidden layer LSTM cell are 128 and 1 respectively, and the output layer is a fully connected layer containing one neuron. The output layer, which is a fully connected layer, can further increase the nonlinearity of the LSTM neural network, thereby improving the accuracy of monitoring. Stochastic gradient descent is used as the optimizer, and the mean square error MSE is selected as the loss function for model training.

[0080] The activation function used in the output layer is ReLU, and the activation function used in the remaining layers is Leaky ReLU. Compared with the ReLU activation function, the Leaky ReLU activation function will produce a small slope for negative inputs, avoiding the problem that the derivative of the ReLU activation function for negative inputs is zero, thereby reducing the possibility of silent nerve cells in the LSTM neural network. Finally, the learning rate during the optimization of LSTM neural network training is 0.01. This method will use the F1 score evaluation indicator for model evaluation. The higher the F1 score, the better the model performs in both accuracy and recall. When processing unbalanced data sets such as real-time data of smart terminals, the F1 score is usually more reliable than accuracy.

[0081] Optionally, after the initial training of the source model, the model can be migrated through deep transfer learning technology to obtain a source model that can be used to perform risk assessment on the target software. Through deep transfer learning technology, knowledge can be transferred from the learned task (i.e., the model corresponding to the first business) to improve the performance of the new task (i.e., risk assessment of the target software). Through deep transfer learning, the source model trained with the relevant data of the first business can be migrated to the application scenario corresponding to the second business, thereby predicting the risk of the target software.

[0082] S204: After the target software is launched, actual risk information of the target software is obtained.

[0083] S205: Based on the actual risk information, fine-tune the source model to obtain a risk prediction model.

[0084] After the target software is launched, the source model can be fine-tuned. Specifically, the actual risk information of the target software can be obtained. The source model can be fine-tuned using fine-tuning technology to optimize the source model and obtain the corresponding risk prediction model.

[0085] In this way, the model is trained based on the risk characteristics of the first business and fine-tuned using the actual risk information of the second business. The resulting risk prediction model can accurately predict the risk situation of the target software.

[0086] The above are some specific implementations of the risk assessment method and model training method provided in the embodiments of the present application. Based on this, the present application also provides corresponding risk assessment devices and model training devices. The risk assessment device and model training device provided in the embodiments of the present application will be introduced from the perspective of functional modularization.

[0087] See also Figure 3 , Figure 3 A schematic diagram of a risk assessment device provided in an embodiment of the present application. Specifically, Figure 3 The risk assessment device 300 shown includes:

[0088] An acquisition unit 310 is used to acquire business transaction information of a first business, wherein the similarity between the first business and a second business is higher than a threshold, and the target software is used to process the second business;

[0089] A business volume prediction unit 320, configured to predict the business volume of the target software at a target time based on the business transaction information of the first business;

[0090] The risk prediction unit 330 is used to predict the risk situation of the target software at the target time based on the business volume of the target software at the target time.

[0091] In some possible implementations, the device further includes a fine-tuning unit;

[0092] The acquisition unit 310 is further used to acquire the actual business volume and actual risk situation of the target software at the target time;

[0093] The fine-tuning unit is used to fine-tune the prediction model based on the actual business volume and the actual risk situation. The prediction model is used to predict the risk situation of the target software. The prediction model is a long short-term memory network LSTM model; the risk situation of the target software is predicted using the fine-tuned prediction model.

[0094] See also Figure 4 , Figure 4 A schematic diagram of the structure of the model training device provided in the embodiment of the present application. Specifically, Figure 4 The model training device 400 shown includes:

[0095] The acquisition unit 410 is used to acquire multiple risk features of a first business, the similarity between the first business and a second business is higher than a threshold, and the target software is used to process the second business;

[0096] A preprocessing unit 420, configured to preprocess the plurality of risk features;

[0097] A training unit 430 is used to pre-train a long short-term memory network (LSTM) model based on the risk characteristics to obtain a source model, wherein the source model is used to perform risk prediction on the target software before the target software goes online;

[0098] The acquisition unit 410 is further used to acquire actual risk information of the target software after the target software is launched;

[0099] The fine-tuning unit 440 is used to fine-tune the source model based on the actual risk information to obtain the risk prediction model.

[0100] In some possible implementations, the acquisition unit 410 is specifically configured to acquire a plurality of original features; calculate the correlation between each of the original features and the risk; and determine the risk feature from the plurality of original features based on the correlation.

[0101] The embodiments of the present application also provide corresponding devices, computer storage media and computer program products for implementing the technical solutions provided by the embodiments of the present application.

[0102] The device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the risk assessment method or model training method described in any embodiment of the present application.

[0103] The computer storage medium stores code, and when the code is executed, the device executing the code implements the risk assessment method or model training method described in any embodiment of the present application.

[0104] The computer program product includes instructions. When the instructions are executed on a computer, the computer is enabled to execute the risk assessment method or model training method described in any embodiment of the present application.

[0105] The "first" and "second" in the names such as "first" and "second" (if any) mentioned in the embodiments of the present application are only used as name identifiers and do not represent the first or second in order.

[0106] Through the description of the above implementation methods, it can be known that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a read-only memory (ROM) / RAM, a magnetic disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0107] Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0108] The above description is merely an exemplary embodiment of the present application and is not intended to limit the protection scope of the present application.

Claims

1. A risk assessment method, characterized in that: The method is used to perform risk assessment on target software installed in a smart terminal device of a financial institution, and the method comprises: Acquiring business transaction information of a first business, where a similarity between the first business and a second business is higher than a threshold, and the target software is used to process the second business; Predicting the business volume of the target software at a target time based on the business transaction information of the first business; Based on the business volume of the target software at the target time, the risk situation of the target software at the target time is predicted.

2. The method according to claim 1, characterized in that After the target software is launched, the method further includes: Obtaining the actual business volume and actual risk situation of the target software at the target time; Based on the actual business volume and the actual risk situation, fine-tuning the prediction model, the prediction model is used to predict the risk situation of the target software, and the prediction model is a long short-term memory network LSTM model; The fine-tuned prediction model is used to predict the risk of the target software.

3. A model training method, characterized in that: The method is used to train a risk prediction model, and the risk prediction model is used to perform risk assessment on target software installed in a smart terminal device of a financial institution. The method includes: Acquire multiple risk features of a first business, the similarity between the first business and a second business is higher than a threshold, and the target software is used to process the second business; preprocessing the multiple risk features; Based on the risk characteristics, a long short-term memory network LSTM model is pre-trained to obtain a source model, and the source model is used to predict the risk of the target software before the target software goes online; After the target software is launched, obtaining actual risk information of the target software; Based on the actual risk information, the source model is fine-tuned to obtain the risk prediction model.

4. The method according to claim 3, characterized in that The obtaining of the multiple risk characteristics of the first business includes: Obtain multiple original features; Calculate the correlation between each of the original features and the risk respectively; Based on the correlation, the risk feature is determined from the plurality of raw features.

5. A risk assessment device, characterized in that: The device is used to perform risk assessment on target software installed in a smart terminal device of a financial institution, and the device includes: an acquisition unit, configured to acquire business transaction information of a first business, wherein a similarity between the first business and a second business is higher than a threshold, and the target software is configured to process the second business; A business volume prediction unit, configured to predict the business volume of the target software at a target time based on the business transaction information of the first business; The risk prediction unit is used to predict the risk situation of the target software at the target time based on the business volume of the target software at the target time.

6. The device according to claim 5, characterized in that The device also includes a fine-tuning unit; The acquisition unit is further used to acquire the actual business volume and actual risk situation of the target software at the target time; The fine-tuning unit is used to fine-tune the prediction model based on the actual business volume and the actual risk situation. The prediction model is used to predict the risk situation of the target software. The prediction model is a long short-term memory network LSTM model; the risk situation of the target software is predicted using the fine-tuned prediction model.

7. A model training device, characterized in that: The device is used to train a risk prediction model, and the risk prediction model is used to perform risk assessment on target software installed in a smart terminal device of a financial institution. The device includes: An acquisition unit, used for acquiring a plurality of risk features of a first business, wherein a similarity between the first business and a second business is higher than a threshold, and the target software is used for processing the second business; A preprocessing unit, configured to preprocess the plurality of risk features; A training unit, used for pre-training a long short-term memory network (LSTM) model based on the risk characteristics to obtain a source model, wherein the source model is used for performing risk prediction on the target software before the target software goes online; The acquisition unit is further used to acquire actual risk information of the target software after the target software is launched; A fine-tuning unit is used to fine-tune the source model based on the actual risk information to obtain the risk prediction model.

8. The device according to claim 7, characterized in that The acquisition unit is specifically used to acquire multiple original features; calculate the correlation between each of the original features and the risk respectively; and determine the risk feature from the multiple original features based on the correlation.

9. A device, characterized in that: The device includes a memory and a controller, the memory is used to store instructions or codes, and the controller is used to execute the instructions or codes stored in the memory to implement the method according to any one of claims 1 to 4.

10. A computer storage medium, characterized in that: The computer storage medium stores codes, and when the codes are executed, a device executing the codes implements the method according to any one of claims 1 to 4.

Images