A method, apparatus, equipment and medium for credit approval processing

Abstract

This invention discloses a credit approval processing method, apparatus, device, and medium, applicable to the field of fintech. The method includes: obtaining at least one state information belonging to the state space of a decision model based on the state space of the target credit approval task; inputting the at least one state information into the decision model to output at least one target action information belonging to the action space of the decision model; and generating a target approval result corresponding to the target credit approval task based on the at least one target action information. Embodiments of this invention can improve the accuracy of the generated credit approval result.

Description

Technical Field

[0001] This invention relates to the field of financial technology, and in particular to a credit approval processing method, apparatus, equipment, and medium. Background Technology

[0002] With the continuous development of internet finance and intelligent risk control technologies, the types of business data in credit approval scenarios are gradually increasing. In addition to basic user information, these include various heterogeneous business data such as credit data, financial data, transaction data, and behavioral data. To improve credit approval efficiency, relevant systems typically need to analyze the user's approval status based on the business data corresponding to the target credit approval task and generate the corresponding approval results.

[0003] Currently, existing technologies typically use Boolean logic rules to process credit approval tasks and generate corresponding approval results.

[0004] However, the approval results generated by existing technologies differ significantly from the actual approval results, resulting in poor accuracy of the generated approval results. Summary of the Invention

[0005] This invention provides a credit approval processing method, apparatus, equipment, and medium. The embodiments of this invention can improve the accuracy of the generated credit approval results.

[0006] In a first aspect, embodiments of the present invention provide a credit approval processing method, the method comprising:

[0007] Based on the state space of the decision model, obtain at least one state information belonging to the state space of the target credit approval task; the state space includes at least one of the following states: identity verification state, risk detection state, financial analysis state, compliance review state, and tool execution state;

[0008] Input at least one state information into the decision model to output at least one target action information belonging to the action space of the decision model; the action space includes at least one of the following actions: approval action, approval rejection action, and supplementary material action;

[0009] Based on at least one target action information, generate the target approval result corresponding to the target credit approval task.

[0010] Secondly, embodiments of the present invention also provide a credit approval processing device, the device comprising:

[0011] The status information acquisition module is used to acquire at least one status information belonging to the state space of the decision model, based on the state space of the target credit approval task; the state space includes at least one of the following states: identity verification status, risk detection status, financial analysis status, compliance review status, and tool execution status;

[0012] The target action information output module is used to input at least one status information into the decision model to output at least one target action information belonging to the action space of the decision model; the action space includes at least one of the following actions: approval action, approval rejection action, and supplementary material action;

[0013] The target approval result generation module is used to generate the target approval result corresponding to the target credit approval task based on at least one target action information.

[0014] Thirdly, embodiments of the present invention also provide a credit approval processing device, which includes:

[0015] At least one processor; and

[0016] A memory that is communicatively connected to at least one processor; wherein,

[0017] The memory stores a computer program that can be executed by at least one processor, such that the at least one processor is able to perform the credit approval processing method of any embodiment of the present invention.

[0018] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the credit approval processing method of any embodiment of the present invention.

[0019] Fifthly, embodiments of the present invention also provide a computer program product, characterized in that the computer program product includes a computer program, which, when executed by a processor, implements the credit approval processing method of any embodiment of the present invention.

[0020] The technical solution of this invention, by constructing a decision model including a state space and an action space, and determining the target action information and target approval result based on state information, can realize state-based dynamic decision processing for target credit approval tasks, thereby improving the adaptability of the credit approval process to complex business scenarios; by using state information to drive action space decision-making, the dependence of traditional fixed-rule approval processes on predefined logical paths can be reduced, thereby improving the flexibility of approval processing; by using the decision model to automatically reason about actions and generate approval results for target approval tasks, the degree of automation and efficiency of approval processing can be improved to a certain extent; and the technical problem of poor accuracy of the generated approval results due to a large gap between the generated approval results and the actual approval results is solved, thus improving the accuracy of the generated credit approval results.

[0021] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart of a credit approval processing method provided in an embodiment of the present invention;

[0024] Figure 2 A flowchart illustrating a training method for a decision-making model of an intelligent agent, provided as an embodiment of the present invention;

[0025] Figure 3 This is a schematic diagram of the structure of a credit approval processing device provided in an embodiment of the present invention;

[0026] Figure 4 This is a schematic diagram of a credit approval processing device provided in an embodiment of the present invention. Detailed Implementation

[0027] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0028] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0029] In the technical solutions of the embodiments of the present invention, the acquisition, storage and application of at least one state information of the target credit approval task belonging to the state space all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0030] Figure 1 This is a flowchart illustrating a credit approval processing method provided in an embodiment of the present invention. This embodiment is applicable to situations where credit approval tasks are processed automatically to generate corresponding approval results. The method can be executed by a credit approval processing device, which can be implemented in hardware and / or software.

[0031] See Figure 1 The credit approval processing method shown includes:

[0032] S101. Based on the state space of the decision model, obtain at least one state information belonging to the state space of the target credit approval task; the state space includes at least one of the following states: identity verification state, risk detection state, financial analysis state, compliance review state, and tool execution state.

[0033] The decision model can refer to a semantic decision model based on a large model, deployed within an intelligent agent. This model dynamically infers and makes decisions regarding subsequent approval actions for the target credit approval task based on its current state information. Unlike traditional fixed-rule models, this model leverages the contextual understanding, semantic reasoning, and state association analysis capabilities of a large model to understand the business state under different approval scenarios and determine the target action information from the action space that matches the current state, thereby achieving dynamic control of the credit approval process.

[0034] For example, in a target credit approval task, the anti-fraud module detects that the applicant is currently using a newly encountered login device, and that the login address is significantly different from their place of residence. In this case, the decision model does not simply reject the application based on a fixed risk threshold. Instead, it uses a large model to analyze the semantics of the current state and reason within the current approval context. After reasoning, the decision model can determine that "login from a different location" and "login from a new device" are potentially suspicious behaviors, but do not yet meet the criteria for direct rejection. Therefore, it selects "supplementary identity verification action" from the action space as the target action information to further complete the subsequent approval process.

[0035] In this invention, the state space refers to a set of states that describe the different business states that a target credit approval task may be in throughout the entire approval process. Different states in the state space represent the business semantic information corresponding to different approval stages, risk scenarios, and business processing stages. The state space is a crucial foundation for large-scale models to perform approval semantic reasoning. Based on the relationships between different states, large-scale models can dynamically analyze the current business stage of the target credit approval task and the approval actions that may be executed subsequently. The state space can include multiple state nodes and multiple state transition relationships. Different state nodes correspond to different business processing stages, such as identity verification, risk detection, financial analysis, compliance review, and tool execution. State transition relationships describe the allowed flow paths between different states. In this invention, the state space is mainly used to achieve state-based management and state semantic modeling of the credit approval process. Complex approval processes can be broken down into multiple approval states with clear business semantics, facilitating the large-scale model to dynamically generate corresponding approval actions based on different states. Furthermore, the state space can provide state context information for the large-scale model, thereby improving its semantic understanding of complex approval scenarios.

[0036] The target credit approval task refers to a specific loan application process that requires approval by the intelligent agent. This task includes business data, approval data, risk data, and approval context information related to the loan application. The target credit approval task requires user permission to use. When a user submits a personal business loan application through an online lending platform, the system generates a corresponding target credit approval task. This task may include data such as ID photos, bank statements, tax returns, credit reports, and historical loan records. The large model can then perform approval processes such as identity verification, risk detection, financial analysis, and compliance review around this target credit approval task.

[0037] In this context, state information refers to data content characterizing the current business state of the target credit approval task. It primarily reflects the current approval stage, risk status, business characteristics, and contextual semantic information of the task. The formation principle of state information mainly involves the system extracting features from the target business data and determining the current state of the target credit approval task based on the correspondence between business state features and the state space. In this invention, state information is primarily used as the core input information for the large model to perform approval reasoning. Through state information, the large model can understand the current business scenario, risk status, and approval stage of the target credit approval task, thereby dynamically generating corresponding approval actions.

[0038] Among them, the identity verification status can refer to the business status corresponding to the applicant's identity authenticity verification stage of the target credit approval task. In this status, the intelligent agent can perform consistency verification on the applicant's identity information, facial information, device information and real name information to determine whether the applicant's identity is authentic and trustworthy.

[0039] Among them, the risk detection status can refer to the business status corresponding to the risk identification and risk analysis stage of the target credit approval task. In this status, the intelligent agent can analyze the applicant's behavioral data, device data, correlation data and historical business data to identify whether there is fraud risk, abnormal transaction risk or loan fraud risk.

[0040] Among them, the financial analysis status can refer to the business status corresponding to the financial capability analysis stage of the target credit approval task. In this status, the intelligent agent can analyze the applicant's income data, bank statement data, asset and liability data, and tax data to assess the applicant's repayment ability and financial stability.

[0041] The compliance review status refers to the business status corresponding to the business rule verification and regulatory rule review stage of the target credit approval task. In this status, the intelligent agent can review whether the current approval process complies with the preset regulatory requirements, risk control rules and business specifications.

[0042] The tool execution status can refer to the business status corresponding to the external tool call or auxiliary function execution stage of the target credit approval task. In this state, the intelligent agent can call credit inquiry tools, anti-fraud detection tools, or data analysis tools to obtain the corresponding business processing results.

[0043] S102. Input at least one state information into the decision model to output at least one target action information belonging to the action space of the decision model; the action space includes at least one of the following actions: approval action, approval rejection action, and supplementary material action.

[0044] In this invention, target action information refers to the information content corresponding to the target approval action dynamically determined by the decision model from the action space based on the current state information and the approval context. Target action information is primarily used to drive the execution of subsequent approval processes. The system can execute corresponding approval processing logic based on different target action information, such as executing a supplementary material process, executing a manual review process, or directly generating an approval result. Simultaneously, target action information can also drive the flow of approval status, thereby achieving dynamic control of the entire approval process. For example, when the large model analysis finds that the applicant's fraud score does not reach the automatic rejection threshold, but there are risk characteristics such as login from a different location, login from a new device, and abnormal fund flows, the large model can comprehensively reason based on the current state semantics and select "supplementary identity verification action" as the target action information from the action space, instead of directly outputting "approval rejection action."

[0045] The approval action can refer to the intelligent agent determining, based on the analysis results of the current approval status, that the target credit approval task meets the loan approval conditions, and then executing the approval process.

[0046] Among them, the approval rejection action can refer to the intelligent agent determining, based on the analysis results of the current approval status, that the target credit approval task does not meet the loan approval conditions, and executing the action of not approving the approval.

[0047] The action of supplementing materials can refer to the action by which the intelligent agent, based on the analysis results of the current approval status, determines that the current business data is insufficient to support the completion of the approval decision, and initiates a request to the applicant for supplementary identity information, financial information, or supporting materials.

[0048] S103. Based on at least one target action information, generate the target approval result corresponding to the target credit approval task.

[0049] The target approval result can refer to the final approval processing result generated based on the target action information. In this invention, the target approval result is mainly used to complete the final approval output of the target credit approval task and serves as the final processing result of the entire approval process.

[0050] As can be seen, in this embodiment, by constructing a decision model including a state space and an action space, and determining the target action information and target approval result based on state information, state-based dynamic decision processing of the target credit approval task can be achieved, thereby improving the adaptability of the credit approval process to complex business scenarios; by using state information to drive action space decision-making, the dependence of traditional fixed-rule approval processes on predefined logical paths can be reduced, thereby improving the flexibility of approval processing; by using the decision model to automatically reason about actions and generate approval results for the target approval task, the degree of automation and efficiency of approval processing can be improved to a certain extent; the technical problem of a large gap between the generated approval result and the actual approval result, resulting in poor accuracy of the generated approval result, is solved, thus the embodiments of the present invention can improve the accuracy of the generated credit approval result.

[0051] In an optional embodiment, Figure 2 A flowchart illustrating a training method for a decision-making model of an intelligent agent, provided in an embodiment of the present invention.

[0052] It should be noted that for parts not described in detail in the embodiments of the present invention, please refer to the descriptions in other embodiments.

[0053] See Figure 2 The training method for the decision-making model of the agent shown includes:

[0054] S201. Obtain training samples, which include state information.

[0055] The training samples refer to a set of historical approval data used to train the decision-making model. The training samples reflect the relationships between different approval states, scenarios, and actions. In this invention, the training samples are primarily used to train and optimize the decision-making model. Through the training samples, the large model can learn approval action strategies corresponding to different business states, thereby improving its ability to identify states, reason about actions, and generate approval results in subsequent approval processes. Furthermore, since the training samples also contain historical approval context, they can enhance the large model's semantic understanding of complex approval scenarios and gray-scale risk scenarios.

[0056] S202. Construct the state space and action space of the decision-making model.

[0057] S203. Input the state information from the training samples into the decision model to output the target action information determined from the action space.

[0058] S204. Evaluate the target action information according to the preset reward function, obtain the evaluation result, and feed the evaluation result back to train the decision model.

[0059] The reward function, in this invention, refers to an evaluation function that assesses the effectiveness of the decision model's output actions. It primarily generates a reward value based on the matching between the approval result corresponding to the target action information and the actual approval result, thereby guiding the large model to continuously optimize its approval strategy. In this invention, the reward function is mainly used to evaluate the effectiveness of the approval actions output by the decision model and serves as an important basis for updating model parameters. Through the reward function, the large model can continuously optimize its action decision-making strategy based on historical approval results, thereby improving approval accuracy, risk identification capabilities, and approval adaptability in complex business scenarios.

[0060] The evaluation result refers to the data generated by the reward function after evaluating the approval effect corresponding to the target action information. The principle behind the formation of the evaluation result is that the reward function, based on the matching relationship between the target approval result and the actual approval result, comprehensively analyzes the current action strategy to generate the corresponding evaluation result. In this invention, the evaluation result is mainly used as data for feedback training of the decision model. The system can dynamically update the strategy parameters in the large model based on the evaluation result, thereby improving the accuracy of action reasoning, risk identification ability, and approval process adaptability in subsequent approval processes.

[0061] As can be seen, in this embodiment, a large-scale decision model for credit approval scenarios is constructed based on state space and action space, and a reward function is used to train the approval actions. This enables the decision model to dynamically learn corresponding approval action strategies based on different approval states. This mainly solves the problem that the fixed rule approval method in the existing credit approval system is difficult to dynamically adjust the approval actions according to complex approval states, resulting in low approval flexibility in complex business scenarios. By training the decision model based on state space, action space, and reward feedback mechanism, the decision model can dynamically generate corresponding approval actions according to different approval states, thereby improving the approval processing adaptability and dynamic decision-making ability in complex credit approval scenarios.

[0062] In some embodiments, the target action information is evaluated according to a preset reward function to obtain an evaluation result, and the evaluation result is fed back to train the decision model, including:

[0063] Based on the target approval result corresponding to the target action information and the actual approval result corresponding to the target action information, determine the result matching information corresponding to the target action information;

[0064] The matching results are input into a preset reward function to determine the reward value corresponding to the target action information;

[0065] Based on the reward value, the policy parameters corresponding to the decision model are updated to complete the feedback training of the decision model.

[0066] The actual approval result refers to the final approval processing result of the target credit approval task during the actual business execution process. It reflects the actual approval conclusion of the target credit approval task in the actual business scenario. In this invention, the actual approval result is not directly generated by the current decision model, but rather is the actual result data corresponding to the target credit approval task after actual business processing, manual review, subsequent risk verification, or the actual loan fulfillment process. Since the actual approval result can truly reflect the execution effect of the approval action in actual business, it can serve as an important reference for the subsequent feedback training of the large model. In this invention, the actual approval result is mainly used as a true evaluation basis in the feedback training of the decision model. Through the actual approval result, the system can determine whether the target approval result generated by the current large model conforms to the real business scenario, thereby further analyzing whether the current action strategy is reasonable. At the same time, the actual approval result can also be used to assist the reward function in calculating the reward value, so as to achieve continuous optimization of the decision model strategy parameters.

[0067] Among them, result matching information refers to information that characterizes the degree of matching between the target approval result and the actual approval result. It mainly reflects the consistency between the current decision model's output action and the actual business result. The principle behind the formation of result matching information is that the system compares and analyzes the target approval result and the actual approval result, and generates corresponding matching results based on the correspondence between the two. Specifically, the system first obtains the target approval result corresponding to the target action information; then it obtains the actual approval result subsequently formed by the target credit approval task; and the system analyzes the degree of consistency between the two. For example, when both the target approval result and the actual approval result are "rejected by risky users," the degree of matching between the two can be considered high; while when the target approval result is "approved," but the actual approval result corresponds to "subsequent bad debt," it is determined that there is a significant deviation between the two. The system further generates corresponding result matching information based on the above matching degree and uses it for subsequent reward value calculation.

[0068] The reward value refers to the evaluation value calculated by the reward function based on the result matching information. In some embodiments, the reward value may include positive reward values, negative reward values, and neutral reward values. A positive reward value indicates that the current approval action is highly effective; a negative reward value indicates that the current approval action has a significant decision-making bias; and a neutral reward value indicates that the current approval action is of average effectiveness. In this invention, the reward value is mainly used as core evaluation data during the feedback training process of the decision model. The system can update the policy parameters in the large model based on the reward value, thereby strengthening effective approval actions and weakening erroneous approval actions. Simultaneously, the reward value can also help the system continuously optimize its action reasoning ability and risk identification ability in complex approval scenarios.

[0069] The strategy parameters refer to the model parameters used within the decision-making model to control the logic of generating approval actions. Strategy parameters can include state weight parameters, action weight parameters, risk bias parameters, action probability parameters, and contextual parameters. Specifically, state weight parameters represent the degree of influence of different state characteristics on approval actions; action probability parameters represent the execution probability of different actions in the current state; and contextual parameters represent the contextual relationships between different approval stages. In this invention, strategy parameters are primarily used to control the approval action generation logic of the large model. By dynamically adjusting the strategy parameters, the system can continuously optimize its action reasoning capabilities under different approval states, thereby improving approval accuracy, risk identification capabilities, and approval process adaptability in complex business scenarios.

[0070] As can be seen, in this embodiment, by determining the result matching information based on the target approval result and the actual approval result, generating the corresponding reward value using the reward function based on the result matching information, and then updating the strategy parameters of the decision model according to the reward value, the effect feedback and strategy adjustment of the approval action output by the decision model can be provided. This allows the decision model to gradually optimize the action generation logic under different approval states during continuous training, thereby improving the adaptability of approval actions and dynamic decision-making ability in complex credit approval scenarios.

[0071] In some embodiments, based on the state space of the decision model, obtaining at least one state information belonging to the state space of the target credit approval task includes:

[0072] Obtain the target business data corresponding to the target credit approval task;

[0073] Feature extraction is performed on the target business data to obtain at least one business status feature;

[0074] Based on the state space of the decision model and the correspondence between business state characteristics and the state space, determine at least one state information belonging to the state space for the target credit approval task.

[0075] The target business data refers to the original business data content corresponding to the target credit approval task. This data reflects the loan applicant's business situation, identity, financial situation, and risk situation in the current approval scenario. In some embodiments, the target business data may include identity information data, credit data, bank transaction data, financial statement data, behavior log data, device information data, and historical loan data. Identity information data may include ID card number, facial recognition result, and contact information; behavior log data may include login time, login location, and device switching records; device information data may include device number, IP address, and device risk label. In this invention, the target business data is primarily used as the basic input data in the entire intelligent approval process. The system can extract business status features based on the target business data and further determine the status information corresponding to the current approval task. Simultaneously, the target business data can provide a complete business context for the large model, thereby improving the large model's semantic understanding and dynamic reasoning capabilities for complex approval scenarios.

[0076] Among them, business status features refer to data features extracted from target business data that characterize the current business status of the target credit approval task. Business status features are used to reflect the risk situation, financial situation, identity situation, and business behavior situation corresponding to the current approval task.

[0077] As can be seen, in this embodiment, by acquiring the target business data corresponding to the target credit approval task and extracting features from the target business data to obtain the corresponding business state features, and then determining the state information corresponding to the target credit approval task based on the correspondence between the business state features and the state space, the decision model can identify the business state of the current credit approval task based on the target business data, thereby providing a state basis for subsequent approval action reasoning and improving the accuracy of state identification in complex credit approval scenarios.

[0078] In some embodiments, at least one state information is input into a decision model to output at least one target action information belonging to the action space of the decision model, including:

[0079] Input at least one state information into the state encoding layer corresponding to the decision model to obtain information features corresponding to at least one state information;

[0080] Based on information features and action space, action strategy reasoning is performed on information features to obtain at least one candidate action information;

[0081] Based on the action probability corresponding to at least one candidate action information, determine at least one target action information belonging to the action space of the decision model from at least one candidate action information.

[0082] The state encoding layer can refer to a data encoding structure within a decision-making model. It can include a feature extraction unit, a vector encoding unit, a semantic representation unit, and a context fusion unit. Specifically, the feature extraction unit extracts key state features from the state information; the vector encoding unit converts these features into vector representations; the semantic representation unit generates corresponding semantic features; and the context fusion unit fuses historical approval context information. In this invention, the state encoding layer primarily enables the semantic encoding and feature representation of state information. Through this layer, the system can convert complex approval states into information features suitable for large-scale model processing, thereby improving the accuracy and stability of subsequent action strategy reasoning.

[0083] Information features refer to the characteristic representation data generated after the state coding layer encodes state information. Information features can include risk semantic features, financial semantic features, behavioral semantic features, and approval context features. Specifically, risk semantic features characterize the potential risks in the current approval task; financial semantic features characterize the applicant's income, assets, and liabilities; and behavioral semantic features characterize user login behavior, device behavior, and operational behavior. In this invention, information features are primarily used as important input data for action strategy reasoning. Through information features, the large model can understand the business semantics and risk semantics corresponding to the current approval task, thereby improving the accuracy of approval action generation. Simultaneously, information features can also help the large model establish contextual relationships between different approval stages, thereby improving continuous reasoning capabilities in complex approval scenarios.

[0084] Action strategy reasoning refers to the reasoning process by which a decision model analyzes, evaluates, and selects different candidate approval actions based on current information features and the action space. It is primarily used to determine the appropriate approval action to execute under the current approval status. In this invention, action strategy reasoning is mainly used to dynamically generate suitable approval actions based on the current approval status. Through action strategy reasoning, the large model can output different approval strategies for different business scenarios, thereby improving the flexibility and risk identification capabilities of approvals in complex business scenarios. Simultaneously, action strategy reasoning can also reduce the misjudgment problems existing in traditional fixed-rule approval systems.

[0085] In this invention, candidate action information refers to the information content corresponding to multiple pending approval actions generated by the decision-making model during the action strategy reasoning process. Candidate action information is primarily used as the basic data for target action selection. The system can further determine the final target action information based on the action probability and risk associated with different candidate actions. Simultaneously, candidate action information also helps the system retain multiple potential approval schemes, thereby improving decision-making flexibility in complex approval scenarios.

[0086] In this invention, action probability refers to the probability value obtained by quantifying the likelihood of different candidate actions being suitable for the current approval status based on current information features. Action probability is primarily used to rank and filter multiple candidate actions. The system can determine the most suitable target action to execute in the current approval scenario based on the action probabilities corresponding to different candidate actions. Furthermore, action probability can improve the dynamism and flexibility of the approval action selection process, thereby increasing the decision-making accuracy in complex approval scenarios.

[0087] As can be seen, in this embodiment, by inputting the state information into the state encoding layer corresponding to the decision model to obtain the corresponding information features, and performing action strategy reasoning on the current approval state based on the information features and action space to generate corresponding candidate action information, and then determining the target action information according to the action probability corresponding to the candidate action information, the matching accuracy between the target action information and the current credit approval scenario can be improved, thereby improving the accuracy of the credit approval result.

[0088] In some embodiments, generating the target approval result corresponding to the target credit approval task based on at least one of the target action information includes:

[0089] Based on the target confidence of each target action information, the target action information corresponding to the highest target confidence is determined as candidate action information;

[0090] Based on the candidate action information, the target approval result corresponding to the target credit approval task is generated.

[0091] In some embodiments, after the agent performs action strategy reasoning based on state information, it can generate multiple target action information, where different target action information can correspond to different target confidence levels. For example, after analyzing the applicant's identity information, credit information, bank statement information, and behavior log information, the agent can generate multiple target action information corresponding to approval, rejection, and supplementary materials actions. The target confidence level for approval can be 0.21, for rejection 0.13, and for supplementary materials 0.66. Subsequently, the agent can compare the target confidence levels corresponding to each target action information and determine the target action information corresponding to the highest target confidence level as candidate action information. For example, since the target confidence level for supplementary materials is the highest, the agent can determine the target action information corresponding to supplementary materials as candidate action information and generate a corresponding target approval result based on this candidate action information, such as generating a target approval result of "Current business data is insufficient; please supplement income verification materials and bank statements for the past six months."

[0092] As can be seen, in this embodiment, by determining candidate action information from multiple target action information based on the target confidence level corresponding to each target action information, and generating corresponding target approval results based on the candidate action information, the matching degree between different approval actions and the current credit approval scenario can be quantitatively analyzed, thereby improving the accuracy of target approval result generation.

[0093] Figure 3 This invention provides a schematic diagram of a credit approval processing device. This invention is applicable to automating credit approval tasks to generate corresponding approval results. The device can execute credit approval processing methods and can be implemented in hardware and / or software.

[0094] See Figure 3 The credit approval processing device shown includes: a status information acquisition module 301, a target action information output module 302, and a target approval result generation module 303, wherein...

[0095] The status information acquisition module 301 is used to acquire at least one status information belonging to the state space of the decision model based on the state space of the target credit approval task; the state space includes at least one of the following states: identity verification status, risk detection status, financial analysis status, compliance review status, and tool execution status.

[0096] The target action information output module 302 is used to input at least one status information into the decision model to output at least one target action information belonging to the action space of the decision model; the action space includes at least one of the following actions: approval action, approval rejection action, and supplementary material action;

[0097] The target approval result generation module 303 is used to generate a target approval result corresponding to a target credit approval task based on at least one target action information. The technical solution of this embodiment of the invention, by constructing a decision model including a state space and an action space, and determining the target action information and target approval result based on state information, can achieve state-based dynamic decision processing for target credit approval tasks, thereby improving the adaptability of the credit approval process to complex business scenarios; by using state information to drive action space decision-making, the dependence of traditional fixed-rule approval processes on predefined logical paths can be reduced, thereby improving the flexibility of approval processing; by using the decision model to automatically reason about actions and generate approval results for target approval tasks, the degree of automation and efficiency of approval processing can be improved to a certain extent; and the technical problem of a large discrepancy between the generated approval result and the actual approval result, leading to poor accuracy of the generated approval result, is solved, thus this embodiment of the invention can improve the accuracy of the generated credit approval result.

[0098] In some embodiments, the decision model of the agent is obtained in the following way:

[0099] Obtain training samples, which include state information;

[0100] Construct the state space and action space of the decision-making model;

[0101] The state information from the training samples is input into the decision model to output the target action information determined from the action space;

[0102] The target action information is evaluated according to the preset reward function to obtain the evaluation result, and the evaluation result is used to train the decision model.

[0103] In some embodiments, the target action information is evaluated according to a preset reward function to obtain an evaluation result, and the evaluation result is fed back to train the decision model, including:

[0104] Based on the target approval result corresponding to the target action information and the actual approval result corresponding to the target action information, determine the result matching information corresponding to the target action information;

[0105] The matching results are input into a preset reward function to determine the reward value corresponding to the target action information;

[0106] Based on the reward value, the policy parameters corresponding to the decision model are updated to complete the feedback training of the decision model.

[0107] In some embodiments, in obtaining at least one state information belonging to the state space of the target credit approval task based on the state space of the decision model, the state information acquisition module 301 is specifically used for:

[0108] Obtain the target business data corresponding to the target credit approval task;

[0109] Feature extraction is performed on the target business data to obtain at least one business status feature;

[0110] Based on the state space of the decision model and the correspondence between business state characteristics and the state space, determine at least one state information belonging to the state space for the target credit approval task.

[0111] In some embodiments, in inputting at least one state information into a decision model to output at least one target action information belonging to the action space of the decision model, the target action information output module 302 is specifically used for:

[0112] Input at least one state information into the state encoding layer corresponding to the decision model to obtain information features corresponding to at least one state information;

[0113] Based on information features and action space, action strategy reasoning is performed on information features to obtain at least one candidate action information;

[0114] Based on the action probability corresponding to at least one candidate action information, determine at least one target action information belonging to the action space of the decision model from at least one candidate action information.

[0115] In some embodiments, the state space includes at least one of the following states: identity verification state, risk detection state, financial analysis state, compliance review state, and tool execution state; the action space includes at least one of the following actions: approval action, approval rejection action, and supplementary material action.

[0116] The credit approval processing device provided in the embodiments of the present invention can execute the credit approval processing method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the credit approval processing method.

[0117] Figure 4 This is a schematic diagram of a credit approval processing device provided in an embodiment of the present invention.

[0118] like Figure 4As shown, the credit approval processing device 400 includes at least one processor 401 and a memory, such as a read-only memory (ROM) 402 and a random access memory (RAM) 403, communicatively connected to the at least one processor 401. The memory stores computer programs executable by the at least one processor. The processor 401 can perform various appropriate actions and processes based on the computer program stored in the ROM 402 or loaded into the RAM 403 from storage unit 408. The RAM 403 can also store various programs and data required for the operation of the credit approval processing device 400. The processor 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0119] Multiple components in the credit approval processing device 400 are connected to the I / O interface 405, including: an input unit 406, such as a keyboard, mouse, etc.; an output unit 407, such as various types of displays, speakers, etc.; a storage unit 408, such as a disk, optical disk, etc.; and a communication unit 409, such as a network card, modem, wireless transceiver, etc. The communication unit 409 allows the credit approval processing device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0120] Processor 401 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 401 performs the various methods and processes described above, such as credit approval processing methods.

[0121] In some embodiments, the credit approval processing method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 408. In some embodiments, part or all of the computer program may be loaded and / or installed on the credit approval processing device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by processor 401, one or more steps of the credit approval processing method described above may be performed. Alternatively, in other embodiments, processor 401 may be configured to perform the credit approval processing method by any other suitable means (e.g., by means of firmware).

[0122] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0123] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0124] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0125] To provide user interaction, the systems and techniques described herein can be implemented on an operational detection device. This credit approval processing device includes: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the credit approval processing device. Other types of devices can also be used to provide user interaction; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including voice input, speech input, or tactile input).

[0126] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0127] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system. It addresses the shortcomings of traditional physical hosts and Virtual Private Server (VPS) services, such as high management difficulty and weak business scalability.

[0128] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0129] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A credit approval processing method characterized by comprising: The method applied to an intelligent agent comprises: According to the state space of the decision model, at least one state information of a target credit approval task is obtained, which belongs to the state space; the state space comprises at least one of the following states: identity verification state, risk detection state, financial analysis state, compliance audit state and tool execution state; At least one state information is input into the decision model to output at least one target action information belonging to the action space of the decision model; the action space comprises at least one of the following actions: approval pass action, approval reject action and supplementary material action; According to at least one target action information, a target approval result corresponding to the target credit approval task is generated.

2. The method of claim 1, wherein, The decision model of the intelligent agent is obtained by: Obtaining training samples, the training samples comprising state information; Building the state space and action space of the decision model; The state information in the training sample is input into the decision model to output target action information determined from the action space; According to a preset reward function, the target action information is evaluated to obtain an evaluation result, and the decision model is trained according to the evaluation result.

3. The method of claim 2, wherein, According to the target approval result corresponding to the target action information and the actual approval result corresponding to the target action information, the result matching information corresponding to the target action information is determined; The result matching information is input into a preset reward function to determine the reward value corresponding to the target action information; According to the reward value, the strategy parameters corresponding to the decision model are updated to complete the feedback training of the decision model. According to the state space of the decision model, at least one state information of a target credit approval task is obtained, which belongs to the state space; the state space comprises at least one of the following states: identity verification state, risk detection state, financial analysis state, compliance audit state and tool execution state; 4. The method of claim 1, wherein, Obtaining target business data corresponding to the target credit approval task; Feature extraction is performed on the target business data to obtain at least one business state feature; According to the state space of the decision model, and the corresponding relationship between the business state feature and the state space, at least one state information of the target credit approval task belonging to the state space is determined. The at least one state information is input into the state encoding layer corresponding to the decision model to obtain the information feature corresponding to the at least one state information; 5. The method of claim 1, wherein, According to the information feature and the action space, the information feature is inferred by action strategy to obtain at least one candidate action information; According to the action probability corresponding to at least one candidate action information, at least one target action information belonging to the action space of the decision model is determined from at least one candidate action information. According to at least one target action information, a target approval result corresponding to the target credit approval task is generated. ​ 6. The method of claim 1, wherein, ​ The target action information corresponding to the maximum target confidence is determined as candidate action information based on target confidences of the target action information; According to the candidate action information, a target approval result corresponding to the target credit approval task is generated.

7. A credit approval processing apparatus characterized by comprising: Comprise: The state information acquisition module is used for acquiring at least one state information of the target credit approval task belonging to the state space according to the state space of the decision model; the state space comprises at least one state of the following: identity verification state, risk detection state, financial analysis state, compliance audit state and tool execution state; The target action information output module is used for inputting at least one state information into the decision model to output at least one target action information belonging to the action space of the decision model; the action space comprises at least one action of the following: approval pass action, approval reject action and supplementary material action; The target approval result generation module is used for generating a target approval result corresponding to the target credit approval task according to at least one target action information.

8. A credit approval processing apparatus characterized by comprising: The credit approval processing device comprises: At least one processor; and The memory is in communication connection with the at least one processor; wherein The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the credit approval processing method in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for enabling the processor to execute the credit approval processing method in any one of claims 1-7 when executed.

10. A computer program product, characterised in that, The computer program product comprises a computer program, and the computer program is executed by the processor to implement the credit approval processing method according to any one of claims 1-6.

Images