Data determination methods, apparatus, equipment, storage media, and program products

Abstract

This application provides a data determination method applicable to the field of artificial intelligence technology. The method includes: acquiring transaction feedback data sent by a distributed stream processing node, and reference data affecting user transaction behavior within a first predetermined time period after performing parameter adjustment operations on a predetermined account for a target user; invoking a locally stored second model, and outputting feedback information corresponding to a first operable range of data based on the transaction feedback data and the reference data, wherein the feedback information characterizes the degree of fit between the first operable range of data and the user transaction behavior and the reference data; updating the model parameters of the first model based on the feedback information, and sending the updated model parameters to a first computing node, so that the first computing node invokes the first model and outputs target operable range data based on the updated model parameters. This application also provides a data determination apparatus, device, storage medium, and program product.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and more specifically to a data determination method, apparatus, device, storage medium, and program product. Background Technology

[0002] Related methods typically adjust parameters of a user's pre-defined account based on their transaction behavior, such as adjusting the user's account limit or credit card limit based on their historical transaction activity. However, user transactions usually generate a large amount of transaction data, leading to slow computation, high hardware resource consumption, and the account limit adjustment methods suffer from response lag and difficulty in accurately depicting the dynamic state of individual users. Summary of the Invention

[0003] In view of the above problems, this application provides a data determination method, apparatus, device, storage medium, and program product.

[0004] According to a first aspect of this application, a data determination method is provided, comprising: acquiring transaction feedback data sent by a distributed stream processing node, and reference data affecting user transaction behavior within a first predetermined time period after performing parameter adjustment operations on a predetermined account of a target user, wherein the transaction feedback data includes transaction data generated by the target user based on first adjustable range data within the first predetermined time period, the first adjustable range data being output by multiple first computing nodes calling a locally stored first model based on pre-acquired historical transaction data of the user within a second predetermined time period and historical reference data affecting user transaction behavior; calling a locally stored second model, and outputting feedback information corresponding to the first adjustable range data based on the transaction feedback data and the reference data, wherein the feedback information is used to characterize the degree of adaptation between the first adjustable range data and user transaction behavior and the reference data; updating the model parameters of the first model based on the feedback information, and sending the updated model parameters to the first computing nodes, so that the first computing nodes call the first model and output target adjustable range data based on the updated model parameters.

[0005] According to an embodiment of this application, the data determination method further includes: acquiring predicted transaction behavior data for the target user within a third predetermined time period after the parameter adjustment operation is performed, wherein the duration of the third predetermined time period is longer than the duration of the first predetermined time period; and adjusting the weight of the reward function based on the predicted transaction behavior data, so as to update the model parameters of the first model based on the adjusted weight of the reward function.

[0006] According to an embodiment of this application, obtaining predicted transaction behavior data for a user within a third predetermined time period after performing parameter adjustment operations on a target user's predetermined account includes: obtaining actual transaction behavior data for the target user within a fourth predetermined time period after performing parameter adjustment operations, wherein the duration of the fourth predetermined time period is longer than the duration of the first predetermined time period and shorter than the duration of the third predetermined time period; and calling a prediction model to output predicted transaction behavior data based on the actual transaction behavior data.

[0007] According to an embodiment of this application, the reward function includes: a transaction feedback reward indicator, a risk penalty indicator, and a user feedback reward indicator, as well as the weights corresponding to each of the transaction feedback reward indicator, the risk penalty indicator, and the user feedback reward indicator.

[0008] According to an embodiment of this application, adjusting the weight of the reward function based on predicted transaction behavior data includes: increasing the weight of the risk penalty indicator in the reward function when the predicted transaction behavior data characterizes the target user's transaction risk as meeting preset conditions.

[0009] According to an embodiment of this application, the data determination method further includes: acquiring historical transaction data of multiple users; determining the transaction risk level of multiple users based on the historical transaction data, and sorting the transaction risk levels; learning the historical operable range data of users corresponding to the transaction risk levels in sequence according to the order of the transaction risk levels; and determining model parameters based on the learning results.

[0010] According to an embodiment of this application, the data determination method further includes: collecting transaction behavior data of the target user in real time; and automatically triggering parameter adjustment operations for the target user's pre-defined account in response to the target user's transaction behavior triggering preset risk conditions.

[0011] A second aspect of this application provides a data determination apparatus, comprising: an acquisition module, configured to acquire transaction feedback data sent by a distributed stream processing node, and reference data affecting user transaction behavior within a first predetermined time period after performing parameter adjustment operations on a predetermined account of a target user, wherein the transaction feedback data includes transaction data generated by the target user based on first adjustable range data within the first predetermined time period, the first operable range data being output by multiple first computing nodes calling a locally stored first model based on pre-acquired historical transaction data of the user within a second predetermined time period and historical reference data affecting user transaction behavior; an invocation module, configured to call a locally stored second model, and output feedback information corresponding to the first operable range data based on the transaction feedback data and the reference data, wherein the feedback information is used to characterize the degree of adaptation between the first operable range data and user transaction behavior and the reference data; and an update module, configured to update the model parameters of the first model based on the feedback information, and send the updated model parameters to the first computing nodes, so that the first computing nodes call the first model and output the target operable range data based on the updated model parameters.

[0012] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0013] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0014] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0015] According to the embodiments of this application, by having a first computing node call a local first model to output a first operable range, a distributed stream processing node to read and send transaction feedback data, and a second computing node to call a locally stored second model to output feedback information and update the model parameters of the first model based on the feedback information, the computing tasks can be distributed to multiple computing nodes for parallel execution through distributed processing. Compared with related centralized computing, this solves the problems of slow computing and exhaustion of hardware resources when the data volume is large. Furthermore, the first model first determines the first operable range data, and then the second model determines the feedback data after the first operable range data based on the transaction feedback data and reference data generated by the user for the first operable range data, and updates the parameters of the first model based on the feedback data to output more accurate operable range data that is more matched with the user and the transaction environment. This can achieve a "personalized" effect and realize personalized adjustments for users. Attached Figure Description

[0016] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0017] Figure 1 The illustrations schematically depict data determination methods, apparatus, devices, storage media, and program products according to embodiments of this application;

[0018] Figure 2 A flowchart illustrating a data determination method according to an embodiment of this application is shown schematically;

[0019] Figure 3 A schematic diagram illustrating a data determination method according to an embodiment of this application is shown.

[0020] Figure 4 This schematically illustrates a structural block diagram of a data determination apparatus according to an embodiment of the present application; and

[0021] Figure 5 A block diagram schematically illustrates an electronic device suitable for implementing a data determination method according to an embodiment of this application. Detailed Implementation

[0022] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0023] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0024] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0025] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0026] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0027] Related methods typically perform parameter adjustments to a user's pre-defined account, such as adjusting credit card limits. However, these account parameter adjustments are usually based on pre-defined static rules (e.g., adjustments at predetermined intervals) or solely on historical data-driven scoring models, leading to the following problems: 1. These methods often require processing large amounts of user data, resulting in slow computation and exhaustion of hardware resources. 2. The reliance on pre-defined static rules causes lag in the adjustment process. For example, changes in user spending behavior (e.g., sudden large purchases, fluctuations in repayment ability) or changes in external environmental factors (e.g., market interest rate fluctuations) can cause credit limit adjustments to lag behind actual needs. 3. These methods lack personalization. For instance, they often adjust parameters based on group characteristics, making it difficult to accurately depict the dynamic state of individual users and achieve differentiated management. Furthermore, these methods suffer from a risk-reward imbalance; for example, pursuing returns may overlook potential default risks, or strictly controlling risks may lead to missed opportunities with high-value customers.

[0028] Furthermore, the applications of reinforcement learning are mostly concentrated in the fields of games or robot control. In financial scenarios, it faces the following challenges: 1. Sparse reward problem: The feedback cycle of user credit behavior is long (such as defaults may appear after several months), which makes it difficult for model training to converge; 2. Conflict of multi-objective optimization: It is necessary to optimize bank revenue (interest income, handling fees), risk indicators (default rate, bad debt rate) and customer experience (credit limit satisfaction) at the same time, which makes the design of objective function complex.

[0029] In view of this, embodiments of this application provide a data determination method, comprising: acquiring transaction feedback data sent by a distributed stream processing node, and reference data affecting user transaction behavior within a first predetermined time period after performing parameter adjustment operations on a predetermined account of a target user, wherein the transaction feedback data includes transaction data generated by the target user based on first adjustable range data within the first predetermined time period, the first operable range data being output by multiple first computing nodes calling a locally stored first model based on pre-acquired user historical transaction data within a second predetermined time period and historical reference data affecting user transaction behavior; calling a locally stored second model, based on the transaction feedback data and reference data, outputting feedback information corresponding to the first operable range data, wherein the feedback information is used to characterize the degree of adaptation between the first operable range data and user transaction behavior and reference data; updating the model parameters of the first model based on the feedback information, and sending the updated model parameters to the first computing nodes, so that the first computing nodes call the first model and output target operable range data based on the updated model parameters.

[0030] Figure 1 The illustrations illustratively depict data determination methods, apparatus, devices, storage media, and program products according to embodiments of this application.

[0031] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0032] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0033] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0034] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0035] It should be noted that the data determination method provided in this application embodiment can generally be executed by server 105. Correspondingly, the data determination device provided in this application embodiment can generally be located in server 105. The data determination method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the data determination device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0036] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0037] The following will be based on Figure 1 The described scene, through Figures 2-3 The data determination method according to the embodiments of this application will be described in detail.

[0038] Figure 2 A flowchart illustrating a data determination method according to an embodiment of this application is shown schematically.

[0039] like Figure 2 As shown, the data determination method of this embodiment includes operations S210 to S230, and the data determination method can be executed by the second computing node.

[0040] In operation S210, transaction feedback data sent by the distributed stream processing node is obtained, as well as reference data affecting user transaction behavior during the first predetermined time period after the parameter adjustment operation for the target user's predetermined account is performed.

[0041] According to an embodiment of this application, the transaction feedback data includes transaction data generated by the target user based on the first adjustable range data within a first predetermined time period. The first adjustable range data is output by multiple first computing nodes calling a locally stored first model, based on the user's historical transaction data within a pre-acquired second predetermined time period and historical reference data that affects the user's transaction behavior.

[0042] For example, the second predetermined time period may include a preset historical time period, during which transaction data generated by the target user can be collected to obtain the user's historical transaction data. This historical transaction data may include, for example, transaction history data, repayment data, and credit score data. Historical reference data that influences user transaction behavior may include environmental data corresponding to the second predetermined time period, such as transaction environment data, such as interest rates.

[0043] The first model may include an actor network from an actor-critic algorithm, which can be used, for example, to generate credit limit adjustment strategies. The first operable range data may include credit limit adjustment data for the target account output by the first model, such as credit card limit adjustment data. Specifically, user historical transaction data and historical reference data can be input into the first model, which can output first operable range data, such as credit limit adjustment data for the target user's target account.

[0044] For example, the second pre-scheduled period may include the past three months. The first computing node can obtain transaction data for the pre-scheduled account of the target user for the past three months, as well as transaction environment data for the past three months. The above transaction data and transaction environment data can be input into the first model and the credit limit adjustment data can be output. For example, for a customer with stable monthly consumption and continuously improving credit score, the credit limit adjustment data output by the first model may include increasing the credit limit of the pre-scheduled account by 15%.

[0045] In embodiments of this application, user consent or authorization can be obtained before acquiring user information. For example, a request to acquire user information can be sent to the user before operation S210. If the user consents or authorizes the acquisition of user information, operation S210 is executed.

[0046] According to an embodiment of this application, after the first model outputs the first operable range data, a parameter adjustment operation for the predetermined account can be performed based on the first operable range data. The parameter adjustment operation may include adjusting the limit of the predetermined account to the first operable range. For example, if the first operable range data includes increasing the limit of the predetermined account by 15%, the parameter adjustment operation may include performing the operation of increasing the limit of the predetermined account by 15%.

[0047] After performing the parameter adjustment operation, the subsequent behavior of the target user within a first predetermined time period can be obtained in real time, yielding transaction feedback data. For example, this could include real-time acquisition of the user's repayment behavior (e.g., whether the user has experienced repayment delays) or transaction behavior (e.g., whether consumption has increased) within two months after the parameter adjustment operation. In this case, the first predetermined time period includes two months after the parameter adjustment operation. Specifically, since the transaction feedback data may be large, it can be read from a message queue using a distributed stream processing node. Furthermore, the distributed stream processing node can also read reference data within the first predetermined time period, such as environmental reference data, for example, interest rate adjustments within two months after the parameter adjustment operation. The distributed stream processing node can then send the read reference data and transaction feedback data to a second computing node, allowing the second computing node to invoke a second model and output feedback information.

[0048] In operation S220, the second model stored locally is invoked, and based on transaction feedback data and reference data, feedback information corresponding to the first operable range data is output. The feedback information is used to characterize the degree of adaptation between the first operable range data and the user's transaction behavior and reference data.

[0049] For example, the first model may include an evaluator network in an actor-evaluator algorithm, which can, for example, assess the merits of actions chosen by the actor network. The second model can evaluate the rationality of the first operable range data based on transaction feedback data and reference data, that is, assess the degree of fit between the first operable range data and user transaction behavior and reference data.

[0050] For example, for a customer with stable monthly spending and a continuously improving credit score, the first operational range output by the first model includes increasing the credit limit by 15%. After this adjustment is executed, the user's subsequent behavior within two months can be obtained in real time. For example, if the user's spending increases under the new credit limit and there are no repayment delays, the second model can determine that the first operational range of "increasing the credit limit by 15%" is more reasonable and consistent with the user's transaction behavior and the current transaction environment, based on transaction feedback data such as actual benefits (e.g., increased interest income) and reference data such as risk indicators (e.g., no increase in the probability of default).

[0051] In operation S230, the model parameters of the first model are updated based on the feedback information, and the updated model parameters are sent to the first computing node so that the first computing node can call the first model to output the target operable range data based on the updated model parameters.

[0052] Optionally, by updating the model parameters of the first model based on feedback information, the parameters of the first model can be optimized in reverse based on the evaluation results of the second model, so that the operable range data output by the first model is more adapted to the user's trading behavior and trading environment. Updating the model parameters of the first model based on feedback information may include: when it is determined that the first operable range data is adapted to the user's trading behavior and reference data, calling the second model to calculate a positive reward based on the feedback information, so as to strengthen the generation weight of the first model for such actions that output the first operable range data through positive rewards. After adjusting the model parameters of the first model, the first model can use the updated and more adapted model parameters to output more reasonable target operable range data.

[0053] For example, for a user with stable monthly spending and a continuously improving credit score, the first model might suggest increasing the credit limit by 15%. After this increase, the user's subsequent behavior can be tracked in real time. If the user's spending increases under the new credit limit and there are no repayment delays, the second model will calculate a positive reward based on actual benefits (such as increased interest income) and risk indicators (such as no increase in the probability of default), thus reinforcing the generation weight of this type of action.

[0054] According to embodiments of this application, by storing the first model and the second model locally, the problem of massive network requests and latency spikes caused by the need for computing nodes to remotely call the central model for each decision can be avoided, thereby reducing cross-node communication overhead.

[0055] Figure 3 A schematic diagram illustrating a data determination method according to an embodiment of this application is shown.

[0056] like Figure 3 As shown, the system utilizes a first computing node to call the first model to output the first adjustable range data and update the model parameters. The second computing node calls the second model to output the feedback information corresponding to the first operable range data and output the model parameters. The distributed stream processing node acquires and sends transaction feedback data and reference data. The first computing node can focus on generating the quota adjustment data, the second computing node on evaluating the rationality of the quota adjustment data, and the distributed stream processing node on real-time data acquisition. This achieves precise allocation and isolation of computing resources, improving the system's large-scale parallel processing capability. For example, when the number of users increases, the number of first computing nodes can be independently expanded; or when the amount of data increases, the number of distributed stream processing nodes can be independently expanded. Parallel decision-making among different nodes can be achieved, avoiding the problems of slow computation and hardware resource exhaustion caused by large data volumes.

[0057] According to embodiments of this application, a first operational range is output by a first computing node calling a local first model, transaction feedback data is read and sent by a distributed stream processing node, and feedback information is output by a second computing node calling a locally stored second model. The model parameters of the first model are updated based on the feedback information. Distributed processing can allocate computational tasks to multiple computing nodes for parallel execution. Compared to centralized computing, this solves the problems of slow computation and hardware resource exhaustion when dealing with large amounts of data. Furthermore, the first model first determines the first operational range data, and the second model determines the feedback data following the first operational range data based on the transaction feedback data and reference data generated by the user for the first operational range data. The first model parameters are updated based on the feedback data to output more accurate operational range data that better matches the user and the transaction environment, achieving a "personalized" effect and improving user satisfaction. Moreover, by simulating the interaction between user credit behavior and the external economic environment in real time, credit limit adjustment decisions can be optimized, achieving the dual goals of maximizing returns and controlling default risk, avoiding resource waste caused by excessive credit granting.

[0058] Optionally, the transaction feedback data and reference data within the first predetermined time period in operation S210 can be: short-term transaction feedback data and reference data after performing parameter adjustment operations on the target user's predetermined account, to make a short-term value judgment on the parameter adjustment operation. The amount of data for short-term value judgment is low, thus reducing the computational burden. Furthermore, user transaction behavior is often sudden (such as sudden large-amount spending), requiring real-time response; by acquiring a small amount of short-term data, real-time response to parameter adjustments can be achieved.

[0059] Furthermore, to improve the accuracy of parameter adjustment operations and enhance their suitability for target users, long-term value judgments can be made on the parameter adjustment operations. For example, the data determination method also includes the following operations: acquiring predicted transaction behavior data for the target user within a third predetermined time period after the parameter adjustment operation is performed, wherein the duration of the third predetermined time period is longer than the duration of the first predetermined time period; adjusting the weights of the reward function based on the predicted transaction behavior data, so as to update the model parameters of the first model based on the adjusted weights of the reward function.

[0060] Optionally, the predicted transaction data of the target user for a longer period than the first predetermined period (such as a third predetermined period) after the parameter adjustment operation can be obtained to assess the long-term value of the parameter adjustment operation. For example, after the first model performs the credit limit increase action, the second model can assess the long-term value of the credit limit increase action by analyzing the user's consumption and repayment data for the next three months.

[0061] For example, the second model can consider both short-term gains and long-term risks when assessing state value. For instance, when the transaction environment might increase repayment risk, the second model can lower its long-term value prediction for users in high-risk industries, even if their current spending performance is excellent. If the second model detects that a credit limit increase would lead to a significant increase in long-term risk (e.g., the user's credit limit utilization rate is approaching a critical value), it will generate a negative TD error signal, notifying the first model to reduce the probability of generating similar strategies.

[0062] Furthermore, based on feedback from the second model, the first model can incorporate a "risk buffer mechanism" into subsequent decisions. For example, for users with highly volatile credit scores, even if their short-term spending is active, the first model will limit the credit limit increase to prioritize ensuring controllable risk. It can also allow the second model to appropriately relax risk weights during evaluation to improve customer satisfaction.

[0063] For example, if the transaction environment is determined to potentially lead to a sharp increase in funding costs, the second model can reassess the long-term value of all users within a predetermined timeframe (e.g., one hour), lowering the predicted returns for high-risk users by 30%. Correspondingly, the first model can immediately suspend automatic credit limit increases for users in high-risk industries and initiate a "credit limit review" process for existing highly indebted users, placing 10% of users in a manual review queue. Thus, through real-time environmental awareness, the model can proactively tighten credit limits for high-risk customers during periods of rising interest rates and moderately loosen credit limits for high-quality customers during peak consumption seasons, maximizing the potential for increased returns while decreasing default rates, thereby achieving a dynamic balance between returns and risks.

[0064] According to an embodiment of this application, obtaining predicted transaction behavior data for a user within a third predetermined time period after performing parameter adjustment operations on a target user's predetermined account includes: obtaining actual transaction behavior data for the target user within a fourth predetermined time period after performing parameter adjustment operations, wherein the duration of the fourth predetermined time period is longer than the duration of the first predetermined time period and shorter than the duration of the third predetermined time period; and calling a prediction model to output predicted transaction behavior data based on the actual transaction behavior data.

[0065] Directly acquiring transaction data over a longer period (such as within the third predetermined period) could lead to a large computational burden due to the large time span. Therefore, it's advisable to first acquire transaction data within a shorter period (such as within the fourth predetermined period), and then use a prediction model to forecast long-term transaction data based on this shorter-term data. This reduces computational load and conserves hardware resources. For example, the prediction model could include a Long Short-Term Memory (LSTM) subnetwork. Actual transaction behavior data for the target user within the fourth predetermined period can be input into the LTM subnetwork. The LTM subnetwork can capture long-term dependencies in the time series, such as trends and periodicity in short-term data, thereby inferring future long-term patterns of change.

[0066] For example, if the input is a user's "consumption data for the most recent month", the Long Short-Term Memory subnetwork can learn the patterns in the data, such as consumption frequency, amount fluctuations, and the impact of holidays, to predict consumption trends for the next three months.

[0067] By calling the predictive model, we can predict data for a longer period based on shorter-term data, which can reduce computational overhead, reduce reliance on long-term data, and improve predictive flexibility.

[0068] Optionally, the architecture of the first model and the second model may include the following modules:

[0069] State space: includes user dimensions (such as user spending frequency in the past 3 months, credit limit utilization rate, repayment delay days, and credit score change trends) and environmental dimensions (benchmark interest rate, industry default rate, etc.).

[0070] Action space: includes continuous action outputs, representing the credit limit adjustment ratio (e.g., ±20%), which can support gradient strategy optimization.

[0071] Reward function R t It can include transaction feedback reward indicators, risk penalty indicators, and user feedback reward indicators, as well as the weights corresponding to each of the transaction feedback reward indicators, risk penalty indicators, and user feedback reward indicators. The reward function can be, for example, as shown in equation (1).

[0072] (1)

[0073] Among them, R t For the reward function, , , The weights for transaction feedback rewards, risk penalties, and user feedback rewards are as follows: Transaction feedback rewards can be determined based on the interest and transaction fee income generated by the user's current consumption; risk penalties can be dynamically weighted according to risk level by predicting the user's default probability over a longer period (such as the next 90 days); and user feedback rewards can be determined by combining user's historical credit limit satisfaction survey data and credit limit utilization thresholds (such as triggering positive rewards when it exceeds 80%).

[0074] By collecting user transaction records, repayment history, credit scores, and economic data (such as interest rates) in real time, a multi-dimensional state space can be constructed, which can dynamically perceive changes in the environment.

[0075] Optionally, the weights of the reward function can be adjusted based on the predicted transaction behavior data, so that the model parameters of the first model can be updated based on the adjusted weights of the reward function.

[0076] For example, adjusting the weights of the reward function based on predicted transaction behavior data includes increasing the weight of the risk penalty indicator in the reward function when the predicted transaction behavior data indicates that the target user's transaction risk meets preset conditions.

[0077] For example, meeting preset conditions for transaction risk can include: the target user's transaction risk exceeding a preset threshold; meeting preset conditions can also include: the target user's transaction risk falling within a preset numerical range. The values ​​of the preset threshold and preset numerical range can be set according to actual circumstances and are not limited here. When the target user's transaction risk meets preset conditions, it indicates that after performing parameter adjustments for the target user's pre-defined account, the user's transaction risk is high. To prevent the first model from generating similar operational range data that increases transaction risk again during subsequent decision-making, and to encourage subsequent strategies to be more conservative in similar scenarios, the weight of the risk penalty indicator in the reward function can be increased.

[0078] For example, after increasing the credit limit by 15%, the second model analyzes the user's spending and repayment data over the next three months to assess the long-term value of this action. If the assessment results show that the user increased their spending due to the increased credit limit, but the industry default rate suddenly rises, leading to an increased risk factor, the second model will dynamically adjust the risk penalty weight in the reward function, prompting subsequent strategies to become more conservative in similar scenarios.

[0079] By setting a reward function, complex business objectives (such as "increasing revenue while controlling risk") can be transformed into calculable numerical signals. When the actions of the first model bring positive results (such as increased user spending, increased interest income, and no increase in default rate), the reward function outputs a positive reward (e.g., +10 points), thereby reinforcing the first model's tendency to generate such actions. Conversely, when the actions of the first model lead to negative results (such as repayment delays after a user's credit limit is increased), the reward function outputs a negative penalty (e.g., -5 points), weakening the weight of such actions and prompting the first model to reduce such actions in similar scenarios in the future (e.g., adjusting the credit limit more conservatively). This allows for flexible and accurate adjustment of the first model's actions to output more accurate credit limit adjustment strategies that are tailored to the user's situation.

[0080] By designing a composite reward function, we can balance short-term gains (such as improved credit limit utilization), long-term risks (predicting default probability), and customer lifetime value, avoiding overfitting to a single objective and achieving multi-objective synergistic optimization. This solves the problem of complex objective function design caused by simultaneously optimizing financial institution revenues (such as interest income and fees), risk indicators (default rate and bad debt rate), and customer experience (credit limit satisfaction).

[0081] According to embodiments of this application, the data determination method may further include: acquiring historical transaction data of multiple users; determining the transaction risk level of the multiple users based on the historical transaction data, and sorting the transaction risk levels; sequentially learning the historical operable range data of users corresponding to the transaction risk levels according to the order of the transaction risk levels; and determining model parameters based on the learning results. For example, this may involve acquiring model parameters of a first model and / or a second model.

[0082] Optionally, when training the first or second model, after obtaining the historical transaction data of multiple users, users can be directly classified into specific risk levels based on preset rules. For example, users who simultaneously meet the following criteria can be classified as low-risk users: zero overdue payments in the past 12 months and a repayment rate of 100%; users with one overdue payment in the past 12 months and a repayment rate of 90%-100%; and users with two or more overdue payments in the past 12 months and a repayment rate of less than 90%. The above preset rules are for illustrative purposes only and can be adjusted according to actual needs, without limitation.

[0083] The trading risk levels can be sorted, for example, from low to high. Based on the order of the trading risk levels, the historical operational range data of users corresponding to each risk level can be learned sequentially. This can include: learning the historical operational range data of corresponding users in order of trading risk level from low to high, thereby gradually expanding the training sample, prioritizing the learning of simple strategies from low-risk users, and gradually transitioning to the learning of complex decisions from high-risk users.

[0084] Furthermore, model parameters can be determined based on the learning results. For example, learning objectives and evaluation metrics corresponding to different training stages can be defined. For instance, the learning objective in the low-risk user training stage could include mastering basic transaction patterns (such as stable repayments and regular consumption), and evaluation metrics could include loss values ​​(such as MSE for regression tasks and cross-entropy for classification tasks) and goodness of fit. The learning objective in the medium-risk user training stage could include adapting to moderately volatile scenarios (such as occasional overdue payments and credit limit fluctuations), and evaluation metrics could include risk coverage (such as the proportion of correctly identified medium-risk users) and robustness metrics (the coefficient of variation of the strategy's error on volatile samples). The learning objective in the high-risk user training stage could include handling complex decisions (such as high-frequency overdue payments), and evaluation metrics could include risk identification recall (such as the proportion of high-risk users who are actually in default). The direction of model parameter adjustments can be determined through these learning objectives, and the degree of objective achievement can be quantified through evaluation metrics to guide the specific adjustment range of parameters. For example, in the low-risk user training stage, where the learning objective is to master basic transaction patterns, the basic parameters can be adjusted accordingly.

[0085] Furthermore, inverse reinforcement learning pre-training can be introduced, for example, by using historical high-quality credit limit adjustment decisions (such as high-value user cases) to infer the implicit reward function, thereby initializing the network parameters of the second model.

[0086] By introducing inverse reinforcement learning and a course-based learning measure that progressively expands the training samples according to user risk levels from low to high, implicit reward signals can be extracted from historical transaction data, allowing for phased model training to accelerate convergence. This addresses the problem of long user credit behavior feedback cycles (e.g., defaults may manifest months later), which hinders model training convergence, alleviates the sparse reward problem, and enables the model to output reasonable decisions even in scenarios with missing data (such as with new users), thereby enhancing model robustness.

[0087] According to an embodiment of this application, the data determination method further includes: collecting transaction behavior data of the target user in real time; and automatically triggering parameter adjustment operations for the target user's pre-defined account in response to the target user's transaction behavior triggering preset risk conditions.

[0088] Target user transaction behavior data may include the number of days the user's repayment is late, and / or the credit limit utilization rate. Preset risk conditions may include the number of days the repayment is late exceeding a preset threshold, and / or the credit limit utilization rate being less than a preset threshold. Parameter adjustment operations may include, for example, increasing the priority of the first model for the "reduce credit limit" action and decreasing the priority of the first model for the "increase credit limit" action. For instance, when a user's repayment is late by more than 3 days for two consecutive months, the state space will trigger a risk flag, and the first model will automatically increase the priority of the "reduce credit limit" action to the highest level while suppressing the generation of the "increase credit limit" action. Thus, the action generation of the first model can be adjusted from the user perspective.

[0089] Furthermore, the assessment of user value or behavioral benefits can be revised based on reference data that influences user transaction behavior (such as transaction environment data, like interest rate adjustment data). For example, when assessing state value, the second model typically focuses on monitoring reference data that influences user transaction behavior. For instance, if an interest rate increase leads to higher funding costs, the second model can globally lower its long-term return forecast for all users. The first model then adjusts its strategy accordingly, prioritizing credit limit optimization for high-profit customers (such as users with high installment payment frequency) rather than expanding credit across the board, thus allowing for more cautious control of credit limits.

[0090] Furthermore, by adjusting the weights in the reward function in real time, the priorities between the goals of "pursuing returns" and "controlling risks" can be flexibly switched, achieving a dynamic balance between returns and risks. For example, the reward function adjusts the priority of objectives in real time through weighted coefficients. During a preset promotional period, the weight of the transaction feedback reward indicator can be temporarily increased, allowing the first model to increase the credit limit for customers with low transaction risk. Conversely, during periods of interest rate reduction, the weight of the risk penalty indicator can be automatically increased, making the second model more cautious in suggesting credit limit adjustments for users with higher transaction risk. Through this multi-objective reward synergy mechanism, a dynamic balance between returns and risks can be achieved.

[0091] Furthermore, for customers who have previously reported insufficient credit limits, the second model can add a "customer experience compensation factor" to their subsequent status assessment. For example, when such users' spending frequency rebounds, even if their credit score only improves slightly, the first model will generate a higher-than-usual credit limit increase recommendation to repair customer relationships and extend lifetime value.

[0092] For example, if a user has repeatedly requested a temporary credit limit increase due to credit limit restrictions, and compared to related methods that often struggle to respond promptly due to risk control rules, the embodiments of this application can perform the following operations: First, status recognition can be performed. For example, if the user's monthly spending continues to increase, repayment on-time rate is 100%, and credit score is stable above 780, the attention mechanism can identify "credit limit utilization rate (95%)" as a key limiting factor, with a weighting of 60%. Further, action generation and optimization can be performed. For example, the first model might initially suggest a 20% increase in credit limit, but the second model might predict, based on industry data, that the default rate in the user's retail industry may rise, generating a negative TD error. In this case, the model can quickly correct its strategy through a parameter sharing layer, ultimately executing a tiered credit limit increase, such as an initial 10% increase, followed by a second 8% increase if the user's subsequent spending remains stable and the risk does not increase. You can obtain user feedback data after performing the above-mentioned tiered credit limit increase operation. User feedback data includes, for example, a 25% increase in user spending, a 12% increase in annual interest income, and a customer satisfaction score that rises from 4.3 to 4.8, without triggering any additional risks.

[0093] Optionally, a multi-head attention mechanism can be embedded in the first model to output the contribution of each state feature to the final decision and generate a readable report, such as a decision report. The decision report can indicate that "interest rate changes" and "industry risks" are the main factors in determining the credit limit adjustment strategy. The decision can include "a 0.5% increase in the interest rate will lead to a 10% tightening of the credit limit", thereby increasing the interpretability of the business.

[0094] For example, an attention mechanism could be used to automatically assign weights to different status characteristics during the credit limit decision-making process. For instance, for new users, the model would prioritize "initial spending patterns" and "occupational stability," while for existing customers, it would focus on "historical repayment records" and "credit limit usage trends." A readable report could be generated based on these weights of different status characteristics, highlighting key decision-making factors. For example, a business report might include: "The main reasons for an 8% increase in credit limit: a 30% increase in spending frequency in the past three months (weight 45%), and a 20-point increase in credit score (weight 30%)." The risk control department can quickly verify the model's logic based on this business report.

[0095] For example, if users complain about the lack of transparency in credit limit adjustments, a decision report can be generated. This report can clearly indicate to users that "interest rate changes" and "industry risks" are the main influencing factors, and explain the specific adjustment logic to users through customer service channels, resulting in a 40% decrease in the complaint rate.

[0096] Based on the above data determination method, this application also provides a data determination apparatus. The following will be combined with... Figure 4 The device is described in detail.

[0097] Figure 4 A schematic block diagram of a data determination apparatus according to an embodiment of this application is shown.

[0098] like Figure 4 As shown, the data determination device 400 of this embodiment includes an acquisition module 410, a calling module 420, and an update module 430.

[0099] The acquisition module 410 is used to acquire transaction feedback data sent by the distributed stream processing node, and reference data affecting user transaction behavior within a first predetermined time period after performing parameter adjustment operations on the target user's predetermined account. The transaction feedback data includes transaction data generated by the target user based on first adjustable range data within the first predetermined time period. The first adjustable range data is output by multiple first computing nodes calling a locally stored first model, based on pre-acquired historical user transaction data within a second predetermined time period and historical reference data affecting user transaction behavior. In one embodiment, the acquisition module 410 can be used to perform the operation S210 described above, which will not be repeated here.

[0100] Module 420 is used to invoke the locally stored second model. Based on transaction feedback data and reference data, it outputs feedback information corresponding to the first operable range data. The feedback information characterizes the degree of compatibility between the first operable range data and the user's transaction behavior and the reference data. Module 420 can be used to execute the operation S220 described above, which will not be repeated here.

[0101] The update module 430 is used to update the model parameters of the first model based on the feedback information and send the updated model parameters to the first computing node, so that the first computing node can call the first model to output the target operable range data based on the updated model parameters. In one embodiment, the update module 430 can be used to perform the operation S230 described above, which will not be repeated here.

[0102] According to embodiments of this application, the data determination device further includes a predictive transaction behavior data acquisition module and an adjustment module.

[0103] The predictive transaction behavior data acquisition module is used to acquire predictive transaction behavior data for the target user within a third predetermined time period after the parameter adjustment operation is performed, wherein the duration of the third predetermined time period is longer than the duration of the first predetermined time period; the adjustment module is used to adjust the weight of the reward function based on the predictive transaction behavior data, so as to update the model parameters of the first model based on the adjusted weight of the reward function.

[0104] According to an embodiment of this application, the predictive transaction behavior data acquisition module includes an acquisition unit and a calling unit.

[0105] The acquisition unit is used to acquire actual transaction behavior data of the target user within the fourth predetermined time period after the parameter adjustment operation is performed, wherein the duration of the fourth predetermined time period is longer than the duration of the first predetermined time period and shorter than the duration of the third predetermined time period; the invocation unit is used to invoke the prediction model to output predicted transaction behavior data based on the actual transaction behavior data.

[0106] According to embodiments of this application, the adjustment module includes adding sub-modules.

[0107] A submodule is added to increase the weight of the risk penalty indicator in the reward function when the predicted transaction behavior data characterizes the target user's transaction risk as meeting preset conditions.

[0108] According to embodiments of this application, the data determination device further includes a historical transaction data acquisition module, a sorting module, a learning module, and a determination module.

[0109] The historical transaction data acquisition module is used to acquire the historical transaction data of multiple users; the sorting module is used to determine the transaction risk level of multiple users based on the historical transaction data and sort the transaction risk levels; the learning module is used to learn the historical operable range data of users corresponding to the transaction risk level in sequence according to the order of transaction risk levels; and the determination module is used to determine the model parameters based on the learning results.

[0110] According to embodiments of this application, the data determination device further includes a real-time acquisition module and an automatic triggering module.

[0111] The real-time acquisition module is used to collect transaction behavior data of target users in real time; the automatic triggering module is used to respond to the transaction behavior of target users to trigger preset risk conditions and automatically trigger parameter adjustment operations for the target user's reserved account.

[0112] According to embodiments of this application, any multiple modules among the acquisition module 410, the calling module 420, and the update module 430 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the acquisition module 410, the calling module 420, and the update module 430 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in any one of software, hardware, and firmware methods, or in a suitable combination of any of these. Alternatively, at least one of the acquisition module 410, the calling module 420, and the update module 430 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0113] Figure 5 A block diagram schematically illustrates an electronic device suitable for implementing a data determination method according to an embodiment of this application.

[0114] like Figure 5 As shown, an electronic device 500 according to an embodiment of this application includes a processor 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage portion 508 into a random access memory (RAM) 503. The processor 501 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 501 may also include onboard memory for caching purposes. The processor 501 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0115] RAM 503 stores various programs and data required for the operation of electronic device 500. Processor 501, ROM 502, and RAM 503 are interconnected via bus 504. Processor 501 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 502 and / or RAM 503. It should be noted that programs may also be stored in one or more memories other than ROM 502 and RAM 503. Processor 501 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in one or more memories.

[0116] According to embodiments of this application, the electronic device 500 may further include an input / output (I / O) interface 505, which is also connected to a bus 504. The electronic device 500 may also include one or more of the following components connected to the input / output (I / O) interface 505: an input section 506 including a keyboard, mouse, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the input / output (I / O) interface 505 as needed. A removable medium 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 510 as needed so that computer programs read from it can be installed into the storage section 508 as needed.

[0117] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0118] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 502 and / or RAM 503 and / or one or more memories other than ROM 502 and RAM 503 described above.

[0119] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to enable the computer system to implement the data determination method provided in the embodiments of this application.

[0120] When the computer program is executed by the processor 501, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0121] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 509, and / or installed from a removable medium 511. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0122] In such an embodiment, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is executed by processor 501, it performs the functions defined in the system of this application embodiment. According to embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0123] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0124] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0125] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A data determination method, characterized in that, The method includes: The system acquires transaction feedback data sent by distributed stream processing nodes, as well as reference data affecting user transaction behavior within a first predetermined time period after performing parameter adjustment operations on the target user's pre-defined account. The transaction feedback data includes transaction data generated by the target user based on a first adjustable range of data within the first predetermined time period. The first adjustable range of data is output by multiple first computing nodes calling a locally stored first model based on pre-acquired user historical transaction data within a second predetermined time period and historical reference data affecting user transaction behavior. The second model stored locally is invoked, and based on the transaction feedback data and the reference data, feedback information corresponding to the first operable range data is output, wherein the feedback information is used to characterize the degree of adaptation between the first operable range data and the user's transaction behavior and the reference data; The model parameters of the first model are updated based on the feedback information, and the updated model parameters are sent to the first computing node so that the first computing node can call the first model to output target operable range data based on the updated model parameters.

2. The method according to claim 1, characterized in that, The method further includes: Acquire predicted transaction behavior data for the target user within a third predetermined time period after the parameter adjustment operation is performed, wherein the duration of the third predetermined time period is longer than the duration of the first predetermined time period; Based on the predicted transaction behavior data, the weights of the reward function are adjusted so that the model parameters of the first model are updated based on the adjusted weights of the reward function.

3. The method according to claim 2, characterized in that, The acquisition of predicted transaction behavior data for the user during the third predetermined time period after the parameter adjustment operation for the target user's pre-defined account is performed includes: Acquire actual transaction behavior data for the target user within a fourth predetermined time period after the parameter adjustment operation is performed, wherein the duration of the fourth predetermined time period is longer than the duration of the first predetermined time period and shorter than the duration of the third predetermined time period; The prediction model is invoked to output the predicted transaction behavior data based on the actual transaction behavior data.

4. The method according to claim 2, wherein the reward function comprises: The transaction feedback reward metric, risk penalty metric, and user feedback reward metric, and the weights corresponding to each of the transaction feedback reward metric, risk penalty metric, and user feedback reward metric.

5. The method according to claim 4, wherein adjusting the weights of the reward function based on the predicted transaction behavior data includes: If the predicted transaction behavior data indicates that the target user's transaction risk meets preset conditions, the weight of the risk penalty indicator in the reward function is increased.

6. The method according to claim 1, further comprising: Obtain historical transaction data for multiple users; The transaction risk levels of the multiple users are determined based on the historical transaction data, and the transaction risk levels are sorted. Based on the order of the transaction risk levels, the historical operable range data of users corresponding to the transaction risk levels are learned sequentially. Model parameters are determined based on the learning results.

7. The method according to claim 1, further comprising: Real-time collection of transaction behavior data of the target users; In response to the target user's transaction behavior triggering preset risk conditions, the system automatically triggers parameter adjustment operations for the target user's pre-defined account.

8. A data determination device, characterized in that, The method includes: The acquisition module is used to acquire transaction feedback data sent by the distributed stream processing node, and reference data that affects the user's transaction behavior within a first predetermined period after performing parameter adjustment operations on the target user's predetermined account. The transaction feedback data includes transaction data generated by the target user based on a first adjustable range of data within the first predetermined period. The first adjustable range of data is output by multiple first computing nodes calling a locally stored first model based on the user's historical transaction data and historical reference data that affect the user's transaction behavior within a pre-acquired second predetermined period. The calling module is used to call the second model stored locally, and output feedback information corresponding to the first operable range data based on the transaction feedback data and the reference data. The feedback information is used to characterize the degree of adaptation between the first operable range data and the user's transaction behavior and the reference data. The update module is used to update the model parameters of the first model based on the feedback information and send the updated model parameters to the first computing node, so that the first computing node calls the first model to output target operable range data based on the updated model parameters.

9. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

Images