Bank intelligent authorization method, system and device based on shared operation system and medium

Abstract

The invention discloses a bank intelligent authorization method, system and device based on a shared operation system and a medium, and the method comprises the steps: obtaining the business data of a to-be-handled business based on a business authorization request sent by a business counter, and carrying out the risk level evaluation and risk portrait evaluation of the to-be-handled business based on the business data, matching an authorization auditing strategy corresponding to the to-be-handled business according to the risk portrait and the business risk level, and performing authorization auditing on the to-be-handled business according to the authorization auditing strategy and the business data to obtain an authorization auditing result; performing quality inspection on the authorization auditing result based on the abnormal time sequence behavior analysis result; the authorization auditing result is corrected based on the quality inspection result, a target authorization result is obtained, and the target authorization result is fed back to the business counter, so that the efficiency and accuracy of business authorization are effectively improved, potential risks are accurately identified through risk level and risk portrait evaluation, quality inspection is performed through abnormal time sequence behavior analysis, and the risk of false authorization is reduced.

Description

Technical Field

[0001] The present invention relates to the field of computers, and in particular, to a bank intelligent authorization method, system, device, medium, and program product based on a shared job system. Background Art

[0002] With the development of science and technology, the authorization review of banking operations has gradually shifted from manual review to intelligent centralized authorization review. Intelligent centralized authorization review relies on artificial intelligence, OCR, face recognition, and big data technologies, and replaces traditional manual authorization through a remote centralized mode to achieve automation of business processes. Combining intelligent risk control models with multi-modal algorithms, the system can automatically compare transaction information, certificates, and biometric features, thereby realizing automated authorization of business. However, the current intelligent centralized authorization review has a low accuracy, and there is a problem of data islands. The risk analysis dimension is single, and it is impossible to accurately identify potential risks and anomalies in the review operations. Moreover, the current quality inspection of review results still relies on manual review, which is limited by the subjectivity of manual review, resulting in great limitations in authorization quality inspection. Summary of the Invention

[0003] The main objective of the present invention is to provide a bank intelligent authorization method, system, device, medium, and program product based on a shared job system, aiming to solve the technical problems of the existing technology that the current intelligent centralized authorization review has a low accuracy, a single risk analysis dimension, and is unable to accurately identify potential risks and anomalies in the review operations, and there are limitations in authorization quality inspection.

[0004] To achieve the above objective, the present invention provides a bank intelligent authorization method based on a shared job system. The method is applied to the shared job system and includes the following steps: In response to a business authorization request sent by a business counter, obtain business data of the business to be processed based on the business authorization request. The business data includes the business type; Based on the business type, evaluate the risk level of the business to be processed to determine the business risk level of the business to be processed; Input the branch data of the business branch to which the business counter belongs, the customer data of the customer handling the business to be processed, and the teller data of the teller handling the business to be processed into a risk portrait evaluation model to generate a risk portrait of the business to be processed. The risk portrait evaluation model is constructed based on multi-dimensional feature data of multiple external review nodes. The multi-dimensional feature data includes customer feature data, branch feature data, and teller feature data; Match the authorization review strategy corresponding to the business to be processed according to the risk profile and the business risk level. The authorization review strategy includes a high-risk authorization review strategy and a medium-low-risk authorization review strategy. The high-risk business processing authorization review strategy includes a target individual authorization review strategy and a target business authorization review strategy. The target individuals include agent customers and customers below a preset age threshold. The target businesses include fund flow business, loan business, and corporate business. The medium-low-risk authorization review strategy includes determining whether to directly authorize based on the daily authorization ratio of the business counter; Conduct an authorization review on the business to be processed according to the authorization review strategy and the business data, and obtain an authorization review result; Conduct an abnormal time-series behavior analysis based on the branch historical business data of the business branch and the customer historical business data of the customer handling the business, and perform quality inspection on the authorization review result based on the abnormal time-series behavior analysis result; Correct the authorization review result based on the quality inspection result to obtain a target authorization result, and feedback the target authorization result to the business counter.

[0005] Optionally, before inputting the branch data of the business branch to which the business counter belongs, the customer data of the customer handling the business to be processed, and the teller data of the teller handling the business to be processed into a risk profile evaluation model to generate a risk profile of the business to be processed, it further includes: Obtain business logs, where the business logs include branch business logs, customer business logs, and teller business logs; Generate an authorization behavior time series based on the business logs, and train a pre-constructed original portrait model according to the authorization behavior time series; Obtain initial model parameters according to the trained original portrait model; Perform hash encryption on the initial model parameters to generate encrypted model parameters, and send the encrypted model parameters to each external review node; In response to the external node model parameters sent by each external review node, aggregate the external node model parameters to obtain global model parameters. The external node model parameters are obtained by each external review node decrypting the encrypted model parameters and training a local portrait model based on the initial model parameters obtained after decryption; Train the trained original portrait model based on the global model parameters to obtain a risk profile evaluation model.

[0006] Optionally, the generating an authorization behavior time series based on the business logs and training a pre-constructed original portrait model according to the authorization behavior time series includes: Extract log semantic information and timestamp information from the business logs; Obtain historical event information based on the log semantic information, where the historical event information includes branch event information, teller event information, and customer event information; Conduct event behavior analysis based on the historical event information to obtain the event behavior characteristics of each historical event; Construct an event causal graph based on the event behavior characteristics; Generate an authorization behavior time series based on the timestamp information, and construct an event graph based on the authorization behavior time series; Establish a path index according to the causal path in the causal graph and the temporal path in the event graph; Extract event nodes from the event graph and extract causal pair nodes from the causal graph; Merge the event graph and the causal graph based on the path index, the event nodes, and the causal pair nodes to generate a target hypergraph, where the target hypergraph includes causal hyperedges and temporal hyperedges; Aggregate the node information of each hyperedge neighbor node in the target hypergraph to obtain a hypergraph embedding feature vector; Fuse the original features of each node in the target hypergraph with the hypergraph embedding feature vector to obtain a comprehensive feature vector; Generate a dataset based on the comprehensive feature vector, and train a pre-constructed original portrait model based on the dataset.

[0007] Optionally, the aggregating the external node model parameters to obtain global model parameters includes: Determine the institutional types of each external audit node, and obtain the historical audit behavior information of each external audit node; Conduct feature analysis on the historical audit behavior information to obtain multi-dimensional behavior feature information, where the multi-dimensional behavior feature information includes response delay characteristics, parameter noise characteristics, historical abnormal behavior characteristics, and network stability characteristics of each external audit node; Normalize the multi-dimensional behavior feature information, and calculate the direct trust degree of each external audit node based on the normalized multi-dimensional behavior feature information: Wherein, represents the direct trust degree, respectively represent the weight coefficients of each behavior feature, represents the normalized response delay feature, represents the normalized parameter noise feature, represents the normalized historical abnormal behavior feature, Indicates the network stability characteristics after normalization; Obtain the historical credibility of each external audit node and the evaluation information between each external audit node, and calculate the recommended trust degree of each external audit node based on the evaluation information: Among them, Indicates the external audit node Of the recommended trust degree, Indicates the external audit node interacting with Set of other external audit nodes, Indicates the external audit node Of the historical trust degree, Indicates the external audit node To the external audit node Of the evaluation information; Aggregate the direct trust degree and the recommended trust degree to obtain the comprehensive trust degree of each external audit node: Among them, Indicates the fusion weight coefficient, Indicates the comprehensive trust degree; Aggregate the external node model parameters based on the comprehensive trust degree and the institution type to obtain the global model parameters.

[0008] Optionally, the aggregating the external node model parameters based on the comprehensive trust degree and the institution type to obtain the global model parameters includes: Determine the historical audit behavior time interval of each external audit node based on the historical audit behavior information; Calculate the time-sensitive trust degree of each external audit node according to the behavior time interval and the comprehensive trust degree: Among them, Indicates the time-sensitive trust degree, Indicates the time decay factor, Indicates the weight of the current audit behavior, Indicates the historical audit trust degree, Indicates the decay rate, Indicates the behavior time interval; Classify the external audit nodes according to the institution type and the time-sensitive trust degree to obtain multiple local node sets, and the local node set includes one or more external audit nodes; Perform local aggregation on the external node model based on the comprehensive trust degrees of the nodes in each local node set to obtain the local model parameters of each local node set; Obtain the historical contribution degrees of the nodes in the local node set in the historical model parameters; Determine the set contribution degree of each local node set based on the historical contribution degrees, and determine the set trust degree of each local node set based on the comprehensive trust degrees; Aggregate the local model parameters based on the set contribution degree and the comprehensive trust degree to obtain the global model parameters: Wherein, represents the global model parameters, represents the local model parameters, represents the balance coefficient, and the balance coefficient is used to balance the comprehensive trust degree and the set contribution degree, represents the set contribution degree, represents the set trust degree.

[0009] Optionally, the authorizing and auditing the to-be-handled service according to the authorization auditing policy and the service data to obtain an authorization auditing result includes: If the authorization auditing policy is a target individual authorization auditing policy, obtain the individual type of the current customer of the to-be-handled service according to the service data; If the individual type of the current customer is an agent customer, obtain the original image of the power of attorney of the to-be-handled service; Perform gray processing on each pixel point in the original image to obtain a grayscale image of the power of attorney: Wherein, represents the pixel value of the grayscale image of the power of attorney at the coordinate , respectively represent the weight coefficients of the red, green, and blue channels, respectively represent the pixel values of the red, green, and blue channels at the coordinate in the original image; Obtain the grayscale value range of the grayscale image of the power of attorney, and traverse each pixel point of the grayscale image of the power of attorney; Perform grayscale distribution analysis on each pixel point in the grayscale image of the power of attorney according to the grayscale value range and the traversal result, and set an initial grayscale threshold based on the grayscale distribution analysis result; Perform maximum inter-class variance calculation on the grayscale image of the power of attorney based on the initial grayscale threshold and the grayscale distribution analysis result to determine the target grayscale threshold: Among them, represents the target grayscale threshold, represents the initial grayscale threshold, represents the initial grayscale threshold the proportion of left - hand side pixels, represents the initial grayscale threshold the proportion of right - hand side pixels, represents the average grayscale value of the left - hand side pixels, represents the average grayscale value of the right - hand side pixels; Segment the power of attorney grayscale image into a foreground layer and a background layer according to the target grayscale threshold; Divide the foreground layer into a handwritten font area and a printed font area; Obtain the handwritten structure element of the handwritten font area according to the structure information of the handwritten font area, and obtain the printed structure element of the printed font area according to the structure information of the printed font area; Perform morphological opening operation processing on the handwritten font area and the printed font area respectively based on the handwritten structure element and the printed structure element to obtain a target handwritten image and a target printed image, and the morphological opening operation processing includes erosion operation processing and dilation operation processing; Perform content detection on the target handwritten image and the target printed image to obtain the power of attorney content information of the business to be handled; Perform authorization review on the business to be handled based on the power of attorney content information to obtain an authorization review result.

[0010] Optionally, the dividing the foreground layer into a handwritten font area and a printed font area includes: Calculate the attribution cost of each pixel point in the foreground layer, and divide the foreground layer into multiple super - pixel blocks based on the attribution cost: Among them, represents the coordinate the attribution cost of the pixel at, represents the coordinate the gradient value of the pixel at, represents the average gradient value of the super - pixel cluster, represents the global gradient standard deviation of the foreground layer, represents the coordinate the spatial distance between the pixel at and the center of the super - pixel cluster, represents the size parameter of the super - pixel, respectively represent the balance weight coefficients of the gradient and the distance; Perform character target detection on each superpixel block, and perform border annotation on each superpixel block based on the character target detection results to generate a plurality of character borders; Extract features from each character border and the characters within the character border to obtain a border feature vector and a character structure feature vector for each character border, and determine the similarity of each character border based on the border feature vector and the character structure feature vector; Calculate the information entropy of each superpixel block based on the similarity; wherein, represents the information entropy of the superpixel block , represents the probability that a character border with a similarity lower than a preset similarity threshold appears in the superpixel block ; Classify the superpixel blocks into a handwritten font area and a printed font area based on the information entropy, where the information entropy of the handwritten font area is greater than a preset entropy threshold, and the information entropy of the printed font area is not greater than the preset entropy threshold.

[0011] In addition, to achieve the above object, the present invention also proposes a shared job system, and the shared job system includes: An authorization request response module, configured to respond to a business authorization request sent by a business counter, and obtain business data of a business to be processed based on the business authorization request, where the business data includes a business type; A risk level assessment module, configured to perform a risk level assessment on the business to be processed based on the business type to determine the business risk level of the business to be processed; A risk portrait generation module, configured to input the branch data of the business branch to which the business counter belongs, the customer data of the customer handling the business to be processed, and the teller data of the teller handling the business to be processed into a risk portrait assessment model to generate a risk portrait of the business to be processed, where the risk portrait assessment model is constructed based on multi-dimensional feature data of a plurality of external review nodes, and the multi-dimensional feature data includes customer feature data, branch feature data, and teller feature data; ​An audit policy matching module, configured to match an authorization audit policy corresponding to the business to be processed according to the risk profile and the business risk level. The authorization audit policy includes a high-risk authorization audit policy and a medium-low-risk authorization audit policy. The high-risk business processing authorization audit policy includes a target individual authorization audit policy and a target business authorization audit policy. The target individuals include agent customers and customers whose age is lower than a preset age threshold. The target businesses include fund flow business, loan business, and corporate business. The medium-low-risk authorization audit policy includes determining whether to directly authorize based on the daily authorization ratio of the business counter; An authorization audit module, configured to perform an authorization audit on the business to be processed according to the authorization audit policy and the business data, and obtain an authorization audit result; An authorization quality inspection module, configured to perform abnormal time-series behavior analysis based on the historical business data of the business outlet and the historical business data of the customer handling the business, and perform quality inspection on the authorization audit result based on the abnormal time-series behavior analysis result; An authorization feedback module, configured to correct the authorization audit result based on the quality inspection result, obtain a target authorization result, and feedback the target authorization result to the business counter.

[0012] In addition, to achieve the above object, the present application also provides a bank intelligent authorization device based on a shared job system. The device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor. The computer program is configured to implement the steps of the bank intelligent authorization method based on the shared job system as described above.

[0013] In addition, to achieve the above object, the present application also provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps of the bank intelligent authorization method based on the shared job system as described above are implemented.

[0014] In addition, to achieve the above object, the present application also provides a computer program product. The computer program product includes a computer program. When the computer program is executed by a processor, the steps of the bank intelligent authorization method based on the shared job system as described above are implemented.

[0015] The present invention responds to a business authorization request sent by a business counter, obtains business data of a business to be processed based on the business authorization request, the business data including a business type, evaluates a risk level of the business to be processed based on the business type, determines a business risk level of the business to be processed, inputs branch data of the business branch to which the business counter belongs, customer data of the customer handling the business to be processed, and teller data of the teller handling the business to be processed into a risk portrait evaluation model to generate a risk portrait of the business to be processed. The risk portrait evaluation model is constructed based on multi-dimensional feature data of multiple external audit nodes, and the multi-dimensional feature data includes customer feature data, branch feature data, and teller feature data. A corresponding authorization audit policy for the business to be processed is matched according to the risk portrait and the business risk level. The authorization audit policy includes a high-risk authorization audit policy and a medium-low risk authorization audit policy. The medium-low risk authorization audit policy includes determining whether to directly authorize based on the same-day authorization ratio of the business counter. The high-risk business handling authorization audit policy includes a target individual authorization audit policy and a target business authorization audit policy. The target individuals include agent customers and customers whose age is lower than a preset age threshold. The target businesses include fund flow businesses, loan businesses, and corporate businesses. An authorization audit is performed on the business to be processed according to the authorization audit policy and the business data to obtain an authorization audit result. An abnormal time-series behavior analysis is performed based on the branch historical business data of the business branch and the customer historical business data of the handling customer, and the authorization audit result is quality-checked based on the abnormal time-series behavior analysis result. The authorization audit result is corrected based on the quality-check result to obtain a target authorization result, and the target authorization result is fed back to the business counter; since the present invention effectively avoids the problem of data islands by performing risk portrait evaluation on a business from multiple dimensions respectively, overcomes the limitation problem existing in the risk assessment of single-dimensional data, accurately identifies potential risks and anomalies in the business through dynamic evaluation of the risk level and the risk portrait, accurately matches a corresponding authorization audit policy for the business to be processed, classifies the business audit into a high-risk authorization audit and a medium-low risk authorization audit, thereby strengthening the strictness of the audit for high-risk businesses, determining whether to directly authorize based on the same-day authorization ratio, thereby reducing the redundant audit links for medium-low risk businesses, thereby greatly improving the efficiency of the authorization audit, and accurately quality-checking the authorization result by analyzing historical business data for abnormal time-series behavior analysis, accurately identifying the deviation degree between the customer's current behavior and the historical behavior quality check, and effectively reducing the risk of misauthorization through the quality-check and correction mechanism. Description of the Drawings

[0016] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0017] Figure 1 It is a schematic structural diagram of a bank intelligent authorization device based on a shared job system in the hardware operating environment related to the embodiment solution of the present invention; Figure 2 It is a schematic flowchart of the first embodiment of the bank intelligent authorization method based on the shared job system of the present invention; Figure 3 It is a schematic flowchart of the second embodiment of the bank intelligent authorization method based on the shared job system of the present invention; Figure 4 It is a schematic flowchart of the third embodiment of the bank intelligent authorization method based on the shared job system of the present invention; Figure 5 It is a structural block diagram of the first embodiment of the shared job system of the present invention.

[0018] The realization, functional features and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. Detailed Embodiments

[0019] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0020] Refer to Figure 1 , Figure 1 It is a schematic structural diagram of a bank intelligent authorization device based on a shared job system in the hardware operating environment related to the embodiment solution of the present invention.

[0021] As Figure 1As shown in the figure, the bank intelligent authorization device based on the shared job system may include: a processor 1001, such as a Central Processing Unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard). Optionally, the user interface 1003 may further include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a Wireless-Fidelity (WI-FI) interface). The memory 1005 may be a high-speed Random Access Memory (RAM) or a stable Non-Volatile Memory (NVM), such as a disk memory. Optionally, the memory 1005 may also be a storage system independent of the aforementioned processor 1001.

[0022] Those skilled in the art can understand that Figure 1 the structure shown in the figure does not constitute a limitation on the bank intelligent authorization device based on the shared job system, and may include more or fewer components than shown in the figure, or combine some components, or have different component arrangements.

[0023] As Figure 1 shown, the memory 1005, as a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a bank intelligent authorization program based on the shared job system.

[0024] In Figure 1 the bank intelligent authorization device based on the shared job system shown in the figure, the network interface 1004 is mainly used for data communication with a network server; the user interface 1003 is mainly used for data interaction with users; the processor 1001 and the memory 1005 in the bank intelligent authorization device of the present invention based on the shared job system may be arranged in the bank intelligent authorization device based on the shared job system. The bank intelligent authorization device based on the shared job system calls the bank intelligent authorization program stored in the memory 1005 through the processor 1001 and executes the bank intelligent authorization method provided by the embodiments of the present invention.

[0025] The embodiments of the present invention provide a bank intelligent authorization method based on a shared job system. Referring to Figure 2 , Figure 2 which is a schematic flowchart of the first embodiment of the bank intelligent authorization method of the present invention based on the shared job system.

[0026] In this embodiment, the bank intelligent authorization method based on the shared job system includes the following steps: Step S10: In response to a business authorization request sent by a business counter, obtain business data of the business to be processed based on the business authorization request.

[0027] It should be understood that the execution subject of this embodiment can be a computing service device with data processing, network communication, and program running functions, such as a computer, a server, a mobile phone, etc., or a terminal electronic device capable of implementing the above functions. Hereinafter, taking the bank intelligent authorization device (hereinafter referred to as the authorization device) based on the shared job system as an example, this embodiment and the following embodiments will be described.

[0028] It should be noted that the business authorization request can be a business operation request initiated by a business counter (such as a bank manual counter or an intelligent counter) that requires centralized authorization review and verification and authorization execution, such as business operations such as account opening, transfer, loan approval, etc.

[0029] It should be noted that the business data can be a data set or information set related to the business to be processed of the business authorization request. The business data includes but is not limited to business type, customer type, user biometric information (such as face information or fingerprint information), OCR data, user identity information, transaction amount, historical behavior records, device information, geographical location, timestamp, etc.

[0030] In some embodiments, the business counter sends a business authorization request through a standardized interface (such as the HTTP protocol). After receiving the request through the API, the authorization device collects multi-source business data in real time (including structured data and unstructured data. The structured data includes user ID, transaction amount, account balance, historical transaction frequency, etc.; the unstructured data may include the scanned ID card uploaded by the user, the camera image of the transaction scenario, the voice call record, etc.).

[0031] In some embodiments, the authorization device can perform OCR recognition on the image data in the business data, extract text data, and use natural language processing (NLP) (such as the BERT model) on the text data to extract keywords (such as "urgent transfer", "large amount cash withdrawal", etc.).

[0032] Step S20: Based on the business type, evaluate the risk level of the business to be processed, and determine the business risk level of the business to be processed.

[0033] It should be noted that the business risk level can be the degree of business operation risk evaluated based on business data, and is usually divided into levels such as low risk, medium risk, and high risk. In some embodiments, the business risk level can be used to determine the strictness of the subsequent authorization process (such as manual review, real-time interception, etc.).

[0034] In specific implementation, the authorization device evaluates the risk level based on the business type. For example, business operations with risks such as transfers, corporate business, and withdrawals are evaluated as high-risk business risk levels; business operations with relatively low risks such as checking balances and consulting services are evaluated as low-risk or medium-risk business risk levels.

[0035] In some embodiments, the authorization device can evaluate the business risk level of the business to be processed from multiple dimensions based on business data. For example, based on the business type, large-amount transfers, large-amount withdrawals, loans, and corporate business are evaluated as high-risk levels.

[0036] In some embodiments, the authorization device can calculate the risk probability based on a pre-constructed logistic regression model and evaluate the risk level based on the risk probability. The risk probability calculation formula is as follows: Wherein, represents the risk probability of the business to be processed, represents the multi-dimensional risk feature vector, represents the intercept term, and the intercept term can be optimized through maximum likelihood estimation (MLE) or cross-entropy loss function, represents the combined weight of the multi-dimensional risk feature vector, represents the time factor, represents the business risk level.

[0037] Step S30: Input the branch data of the business network to which the business counter belongs, the customer data of the customer handling the business to be processed, and the teller data of the teller handling the business to be processed into the risk portrait evaluation model to generate the risk portrait of the business to be processed.

[0038] It should be noted that the risk portrait evaluation model is constructed based on multi-dimensional feature data of multiple external review nodes, and the multi-dimensional feature data includes customer feature data, branch feature data, and teller feature data.

[0039] It should be noted that the risk portrait can be the comprehensive evaluation result of the business risk of the business to be processed generated based on multi-dimensional data (customer data, branch data, teller data). For example, the risk portrait can include information such as risk labels (such as "fraudulent tendency", "abnormal operation"), risk probability, and risk source.

[0040] It should be noted that the branch data may be the characteristic information of the branch to which the business counter belongs, including the historical risk records of the branch, geographical location, business volume, equipment security level, etc. The above customer data may be the customer information for handling business, including identity information, transaction behavior data (such as historical transaction frequency, amount distribution), credit score, anti-fraud label, etc. The above teller data may be the relevant characteristic information of the teller handling the business, for example, it may include working years, historical operation compliance records, abnormal operation times, permission level, etc.

[0041] In some embodiments, the authorization device can preprocess the branch data, customer data, and teller data, then extract the characteristics of various types of data to obtain multi-dimensional characteristics, and then construct a knowledge graph to associate customer characteristics, branch characteristics, and teller characteristics. Use a graph neural network (GNN) to aggregate the multi-dimensional characteristics and generate a risk portrait of the business to be handled based on the aggregated characteristics.

[0042] Step S40: Match the authorization review policy corresponding to the business to be handled according to the risk portrait and the business risk level.

[0043] It should be noted that the authorization review policy includes a high-risk authorization review policy and a medium-low risk authorization review policy. The medium-low risk authorization review policy includes judging whether to directly authorize based on the daily authorization ratio of the business counter. The high-risk business handling authorization review policy includes a target individual authorization review policy and a target business authorization review policy. The target individuals include agent customers and customers whose age is lower than a preset age threshold. The target businesses include fund flow business, loan business, and corporate business.

[0044] In some embodiments, the authorization device can classify the business to be handled into high-risk business, medium-risk business, and low-risk business according to the risk portrait and the business risk level; classify the business to be handled into business handled by an agent, business handled by a minor, and business handled by an elderly customer according to the individual type of the current customer handling the business to be handled; classify the business to be handled into transfer business, loan business, corporate business, etc. according to the business type. According to the business classification results, match the corresponding authorization review policy for the business to be handled.

[0045] In some embodiments, the authorization review policy may include an agent business review policy, a high-risk business review policy, a low-age customer review policy, a loan review policy, a corporate business review policy, a transfer business review policy, an elderly customer review policy, etc.

[0046] It should be noted that for medium- and low-risk services, the authorized device can perform authorization processing based on the medium- and low-risk authorization review policy, which may include automatically matching basic service rules. If the verification is passed, authorization-free processing can be adopted. For authorization-free services, to prevent moral hazards, the risk score calculated according to the risk model is matched with the risk interval to obtain the corresponding system authorization ratio. The authorized device can dynamically select whether to automatically authorize the service to be processed according to the actual authorization ratio on the current day, thus reducing the redundant review process.

[0047] Step S50: Perform an authorization review on the service to be processed according to the authorization review policy and the service data to obtain an authorization review result.

[0048] In some embodiments, the authorization review policy may include a basic review policy and a dynamic review policy. The basic review policy can be basic review policies such as biometric recognition, OCR recognition, and NLP recognition. The above dynamic review policy can be a review policy that is adaptively adjusted according to service type, risk profile, risk level, and customer type. The authorized device can add the dynamic review policy on the basis of the basic review policy to generate an authorization review policy suitable for the current scenario. For example, if the service to be processed is a high-risk service and the customer type is an agent handling, the authorized device can add the review policy for the agent's power of attorney on the basis of the basic review policy.

[0049] In some embodiments, low-risk services can be directly authorized automatically; medium-risk services can be authorized and reviewed based on the basic review policy; high-risk services can be reviewed by adding the dynamic review policy on the basis of the basic review policy.

[0050] Step S60: Perform an analysis of abnormal time-series behavior based on the historical service data of the business outlet and the historical service data of the handling customer, and perform quality inspection on the authorization review result based on the analysis result of the abnormal time-series behavior.

[0051] It should be noted that the historical service data of the business outlet may include the historical service volume, transaction amount, risk events, and the corresponding event occurrence time of the business outlet. The above historical service data of the customer may include the customer's transaction records, risk label time series, operation device change records, etc.

[0052] It should be noted that the above analysis result of abnormal time-series behavior may include the analysis result of abnormal time-series behavior of the business outlet and the analysis result of abnormal time-series behavior of the handling customer. In this embodiment, it can be determined whether there are abnormal behaviors in the business outlet and / or the handling customer based on the analysis results of abnormal time-series behavior of the business outlet and the handling customer.

[0053] In some embodiments, the authorization device can obtain the business volume, transaction amount, risk event occurrence time, etc. of the network point in the past 30 days from the time series database; the authorization device can extract the transaction records, risk label time series, operation device change records, etc. of the customer in the past 180 days.

[0054] In some embodiments, the authorization device can align time series data with different granularities, fill in missing time series data, and then standardize behavior and event data (such as transaction amount, transaction times, business behavior, etc.). The Long Short Term Memory (LSTM) model is used to capture the behavioral dependencies of business network points and handled customers (such as periodic fluctuations in holiday transaction volumes, transition probabilities of transaction frequencies, etc.), and anomalies are detected through reconstruction errors: if the error exceeds a threshold (such as 0.3), it is marked as an anomaly.

[0055] Step S70: Correct the authorization review result based on the quality inspection result to obtain a target authorization result, and feedback the target authorization result to the business counter.

[0056] In some embodiments, the authorization device can correct through a decision tree by combining the original authorization review result and the quality inspection result to generate a final target authorization result: Target authorization result = DecisionTree(authorization review result, quality inspection result, confidence).

[0057] For example, if the original authorization review result is "passed" and the quality inspection result is "normal", the output target authorization result is "authorization passed"; If the original authorization review result is "passed", the quality inspection result is "abnormal customer behavior", and it is an anomaly with high confidence, the output target authorization result is "intercept authorization", and it can return to step S50 for review to form a quality inspection closed-loop review.

[0058] If there is a conflict between the quality inspection and the original result but the confidence levels are similar, artificial arbitration is triggered, and an artificial arbitration request is sent to the business counter or a network point or counter with higher authority.

[0059] In some embodiments, the authorization device obtains business logs, which include branch business logs, customer business logs, and teller business logs; generates an authorization behavior time series based on the business logs, and trains a pre-constructed original portrait model according to the authorization behavior time series; obtains initial model parameters according to the trained original portrait model; performs hash encryption on the initial model parameters to generate encrypted model parameters, and sends the encrypted model parameters to each external audit node; in response to the external node model parameters sent by each external audit node, aggregates the external node model parameters to obtain global model parameters, where the external node model parameters are obtained by each external audit node decrypting the encrypted model parameters and training a local portrait model based on the initial model parameters obtained after decryption; trains the trained original portrait model based on the global model parameters to obtain a risk portrait evaluation model.

[0060] In some embodiments, the authorization device extracts log semantic information and timestamp information from the business logs; obtains historical event information based on the log semantic information, where the historical event information includes branch event information, teller event information, and customer event information; performs event behavior analysis according to the historical event information to obtain event behavior characteristics of each historical event; constructs an event causal graph based on the event behavior characteristics; generates an authorization behavior time series according to the timestamp information, and constructs an event graph based on the authorization behavior time series; establishes a path index according to the causal path in the causal graph and the temporal path in the event graph; extracts event nodes from the event graph, and extracts causal pair nodes from the causal graph; merges the event graph and the causal graph based on the path index, the event nodes, and the causal pair nodes to generate a target hypergraph, where the target hypergraph includes causal hyperedges and temporal hyperedges; aggregates the node information of each hyperedge neighbor node in the target hypergraph to obtain a hypergraph embedding feature vector; fuses the original features of each node in the target hypergraph with the hypergraph embedding feature vector to obtain a comprehensive feature vector; generates a data set based on the comprehensive feature vector, and trains a pre-constructed original portrait model based on the data set.

[0061] In some embodiments, the authorization device determines the institutional types of each external audit node and obtains the historical audit behavior information of each external audit node; performs feature analysis on the historical audit behavior information to obtain multi-dimensional behavior feature information, where the multi-dimensional behavior feature information includes response delay characteristics, parameter noise characteristics, historical abnormal behavior characteristics, and network stability characteristics of each external audit node; normalizes the multi-dimensional behavior feature information, and calculates the direct trust degree of each external audit node based on the normalized multi-dimensional behavior feature information: Among them, represents the direct trust degree, , , , respectively represent the weight coefficients of each behavior feature, represents the response delay feature after normalization, represents the parameter noise feature after normalization, represents the historical abnormal behavior feature after normalization, represents the network stability feature after normalization; Obtain the historical credibility of each external audit node and the evaluation information between each external audit node, and calculate the recommended trust degree of each external audit node based on the evaluation information: Among them, represents the external audit node 's recommended trust degree, represents the set of other external audit nodes that have interacted with the external audit node , represents the historical trust degree of the external audit node , represents the evaluation information of the external audit node on the external audit node ; Aggregate the direct trust degree and the recommended trust degree to obtain the comprehensive trust degree of each external audit node: Among them, represents the fusion weight coefficient, represents the comprehensive trust degree; Aggregate the external node model parameters based on the comprehensive trust degree and the institution type to obtain the global model parameters.

[0062] In some embodiments, the authorized device determines the behavior time interval of the historical audits of each external audit node based on the historical audit behavior information; calculates the time-sensitive trust degree of each external audit node according to the behavior time interval and the comprehensive trust degree: Among them, represents the time-sensitive trust degree, represents the time decay factor, represents the weight of the current audit behavior, represents the historical audit trust degree, represents the decay rate, Indicates the behavior time interval; Classify the external audit nodes according to the institution type and the time-sensitive trust level to obtain multiple local node sets, where each local node set includes one or more external audit nodes; perform local aggregation on the external node model based on the comprehensive trust level of each node in each local node set to obtain the local model parameters of each local node set; obtain the historical contribution degrees of the nodes in the local node set in the historical model parameters; determine the set contribution degree of each local node set based on the historical contribution degree, and determine the set trust level of each local node set based on the comprehensive trust level; Aggregate the local model parameters based on the set contribution degree and the comprehensive trust level to obtain the global model parameters: Wherein, Indicates the global model parameter, Indicates the local model parameter, Indicates the balance coefficient, which is used to balance the comprehensive trust level and the set contribution degree, Indicates the set contribution degree, Indicates the set trust level.

[0063] In this embodiment, by performing risk portrait evaluation on the business from multiple dimensions respectively, the problem of data islands is effectively avoided, and the limitations existing in the risk assessment of single-dimensional data are overcome. Through dynamic evaluation of the risk level and risk portrait, potential risks and anomalies in the business are accurately identified, and the corresponding authorization audit strategy is accurately matched for the business to be processed. The business audit is classified into high-risk authorization audit and medium-low-risk authorization audit, thereby strengthening the audit strictness of high-risk businesses. Based on the authorization ratio on the current day, it is judged whether to directly authorize, thereby reducing the redundant audit links of medium-low-risk businesses, and thus greatly improving the efficiency of authorization audit. By analyzing historical business data for abnormal time-series behavior analysis, accurate quality inspection of authorization results is realized, and the deviation degree between the current behavior of the customer and the historical behavior quality inspection is accurately identified. Through the quality inspection and correction mechanism, the risk of misauthorization is effectively reduced.

[0064] Reference Figure 3 , Figure 3 is the flowchart of the second embodiment of the bank intelligent authorization method based on the shared job system of the present invention.

[0065] Based on the above first embodiment, in this embodiment, before the step S30, it further includes: Step S31: Obtain business logs.

[0066] It should be noted that the business logs include branch business logs, customer business logs, and teller business logs.

[0067] It should be noted that the above-mentioned branch business log can be a behavior log recording the operations of a business branch, which may include transaction records, device status, risk events, teller operation time, business type distribution, etc. For example, the total number of transactions on a certain day at a branch, the proportion of high-risk businesses, and the device failure alarm records.

[0068] The above-mentioned customer business log can be a behavior log recording the behavior of a customer when handling business, which may include transaction time, amount, device fingerprint for operation, IP address, number of input errors, business type preference, etc. For example, the transfer records of a certain customer within 30 days, the number of failed account logins, and the device models used.

[0069] The above-mentioned teller business log can be a behavior log recording the business operations of a teller, which may include the type of business processed, duration, permission usage records, number of abnormal operations, approval passing rate, etc. For example, a certain teller processed 50 transfers on a certain day, 3 of which were marked as "operation delay", and 2 triggered risk interception, etc.

[0070] In some embodiments, the authorization device can preprocess the collected original log data, such as performing data cleaning, feature extraction, abnormal data processing, etc. on the original log data.

[0071] Step S32: Generate an authorization behavior time series based on the business log, and train a pre-constructed original portrait model according to the authorization behavior time series.

[0072] It can be understood that in this embodiment, the time series features related to authorization behavior are extracted from the business log, such as extracting the proportion of business types, risk score fluctuations, authorization result distribution, etc. from the business log.

[0073] In some embodiments, the authorization device can use a sliding window to aggregate features in different dimensions to generate a multi-dimensional time series. Referring to the following formula, where represents the feature vector of the t-th time window, including business type, risk score, authorization result, etc.

[0074] Furthermore, in order to improve the timeliness and accuracy of risk assessment and provide effective time series support for model training, the above step S32 may include: Step S3201: Extract log semantic information and timestamp information from the business log.

[0075] It should be noted that the log semantic information can be structured semantic content extracted from business logs, such as "Customer A handled a transfer business at network point B through teller C" and "Device D malfunctioned at time T". The above timestamp information can be the timestamps corresponding to each log content or log behavior recorded in the business log.

[0076] Step S3202: Obtain historical event information based on the log semantic information.

[0077] It should be noted that the historical event information includes network point event information, teller event information, and customer event information. In this embodiment, historical events can be classified into three categories, including network point events, teller events, and customer events. Among them, network point events include abnormal network point transaction behaviors, historical authorization frequencies of network points, etc.; teller events include teller operation delays, teller privilege overstepping, teller authorization transaction frequencies, etc.; customer events include off-site login events, abnormal device login events, abnormal transaction events, etc.

[0078] Step S3203: Perform event behavior analysis based on the historical event information to obtain the event behavior characteristics of each historical event.

[0079] It should be noted that the event behavior characteristics include network point behavior characteristics, teller behavior characteristics, and customer behavior characteristics.

[0080] In some embodiments, the network point behavior characteristics may include equipment failure rates, the proportion of high-risk services, trading volume fluctuation coefficients, etc.; the teller behavior characteristics may include the standard deviation of operation duration, the number of privilege overstepping times, the audit passing rate, etc.; the customer behavior characteristics may include the number of IP changes, the number of device changes, the number of abnormal inputs, the number of authorization rejections, the number of authorization requests, etc.

[0081] Step S3204: Construct an event causal graph based on the event behavior characteristics.

[0082] In some embodiments, the authorization device can use Granger causality test or Bayesian network to analyze the causal relationship between events, construct causal edges between events. For example, equipment failure at a network point event causes customer transaction interruption; teller privilege overstepping causes customers to have risky transactions, etc. Construct a causal graph based on the causal edges and event nodes.

[0083] Step S3205: Generate an authorization behavior time series based on the timestamp information and construct an event graph based on the authorization behavior time series.

[0084] In some embodiments, the authorization device can construct each event node and time series edge based on the authorization behavior time series, and construct an event graph based on the event node and time series edge.

[0085] Step S3206: Establish a path index according to the causal paths in the causal graph and the temporal paths in the event graph.

[0086] It can be understood that in this embodiment, a path index is established by combining the causal paths in the causal graph (such as "equipment failure → transaction interruption") and the temporal paths in the event graph (such as "T1 → T2 → T3"), where the path index = {causal path, temporal path, association weight}.

[0087] Step S3207: Extract event nodes from the event graph and extract causal pair nodes from the causal graph; Step S3208: Based on the path index, the event nodes, and the causal pair nodes, merge the event graph and the causal graph to generate a target hypergraph.

[0088] In some embodiments, the target hypergraph includes causal hyperedges and temporal hyperedges; in this embodiment, the causal edges (hyperedges) of the causal graph and the temporal edges (hyperedges) of the temporal graph can be merged into a target hypergraph based on a hypergraph neural network: Among them, represents the target hypergraph, represents the combination of event nodes, represents the set of causal hyperedges, represents the set of temporal hyperedges.

[0089] Step S3209: Aggregate the node information of the neighbor nodes of each hyperedge in the target hypergraph to obtain a hypergraph embedding feature vector.

[0090] In some embodiments, the authorized device can aggregate the neighbor node information through a hypergraph convolutional network: Among them, represents the node v 's hypergraph embedding vector, represents the node's original feature matrix, and represents the node v 's hyperedge neighbor node's node information.

[0091] Step S3210: Perform feature fusion on the original features of each node in the target hypergraph and the hypergraph embedding feature vector to obtain a comprehensive feature vector; Step S3211: Generate a dataset based on the comprehensive feature vector and train a pre-constructed original portrait model based on the dataset.

[0092] In some embodiments, the authorized device may extract the comprehensive feature vectors of the nodes from the target hypergraph, generate a data set based on the comprehensive feature vectors, and train a pre-constructed original portrait model based on the data set.

[0093] Step S33: Obtain the initial model parameters according to the trained original portrait model.

[0094] It should be noted that the initial model parameters may include the weight parameters, biases, etc. of the trained original portrait model.

[0095] Step S34: Perform hash encryption on the initial model parameters to generate encrypted model parameters, and send the encrypted model parameters to each external audit node.

[0096] In some embodiments, the authorized device may use homomorphic encryption to encrypt the initial model parameters, and at the same time calculate the hash value to ensure integrity: Among them, represents the encrypted model parameters, represents the initial model parameters.

[0097] Step S35: In response to the external node model parameters sent by each external audit node, aggregate the external node model parameters to obtain global model parameters.

[0098] It should be noted that the external node model parameters are obtained by each external audit node decrypting the encrypted model parameters and training a local portrait model based on the initial model parameters obtained after decryption.

[0099] In some embodiments, the authorized device is the central node, and the external audit nodes are external nodes communicatively connected to the central node. The external audit nodes can be other outlets or third-party audit structures (such as regulatory agencies, etc.) other than the business handling outlets.

[0100] In some embodiments, the external audit node may use the private key to decrypt the received encrypted model parameters, obtain the initial model parameters, and train a local model based on the initial model parameters and local privacy data (such as customer transaction records, the regulatory data and evaluation data of the audit agency itself) to obtain the external node model parameters: Among them, is the external node model parameter obtained by the external audit node training the local model, is the local privacy parameter of the external audit node.

[0101] The external audit node extracts the external node model parameters in the locally trained model, adds noise to the external node model parameters, and sends the external node model parameters with added noise to the central node (i.e., the authorized device).

[0102] In some embodiments, the authorized device may use the federated averaging algorithm to aggregate the external node model parameters sent by each external audit node; or it may aggregate the external node model parameters of each external audit node through weighted averaging.

[0103] Furthermore, in order to effectively aggregate the model parameters of multiple external audit nodes, the above step S35 may include:[[]]END]] Step S3501: Determine the institutional types of each external audit node and obtain the historical audit behavior information of each external audit node; Step S3502: Perform feature analysis on the historical audit behavior information to obtain multi-dimensional behavior feature information; Step S3503: Normalize the multi-dimensional behavior feature information and calculate the direct trust degree of each external audit node based on the normalized multi-dimensional behavior feature information; Step S3504: Obtain the historical credibility of each external audit node and the evaluation information between each external audit node, and calculate the recommended trust degree of each external audit node based on the evaluation information; Step S3505: Aggregate the direct trust degree and the recommended trust degree to obtain the comprehensive trust degree of each external audit node; Step S3506: Aggregate the external node model parameters based on the comprehensive trust degree and the institutional type to obtain the global model parameters.

[0104] It should be noted that the institutional types may include financial institutions, regulatory agencies, third-party certification structures, etc. In some embodiments, the authorized device may determine the institutional type based on the institutional name, certification qualifications, and filing information of the external audit node.

[0105] It should be noted that the multi-dimensional behavior feature information includes the response delay feature, parameter noise feature, historical abnormal behavior feature, and network stability feature of each external audit node. The response delay feature may be the average response time, delay volatility, timeout rate, etc. of the node; the parameter noise feature may be the noise ratio, noise distribution, noise amplitude, invalid data ratio, etc. of the parameters provided by the node; the historical abnormal behavior feature may be the abnormal behavior ratio, response mode abnormal ratio, etc. of the node; the network stability feature may be the packet loss rate, bandwidth fluctuation, network protocol compatibility, etc. of the node.

[0106] It should be noted that the normalization of the multi-dimensional behavior feature information refers to the following formula: Among them, represents the direct trust degree, respectively represent the weight coefficients of each behavior feature, represents the response delay feature after normalization, represents the parameter noise feature after normalization, represents the historical abnormal behavior feature after normalization, represents the network stability feature after normalization.

[0107] It should be noted that the recommended trust degree of each external audit node is calculated with reference to the following formula: Among them, represents the external audit node 's recommended trust degree, represents the set of other external audit nodes that have interacted with the external audit node ; represents the historical trust degree of the external audit node ; represents the evaluation information of the external audit node on the external audit node .

[0108] It should be noted that the comprehensive trust degree of each external audit node is calculated with reference to the following formula: Among them, represents the fusion weight coefficient, represents the comprehensive trust degree.

[0109] It can be understood that in this embodiment, by analyzing the direct trust degree and recommended trust degree of each external audit node, parameter aggregation is performed from the trust degrees of two dimensions, so as to accurately configure the parameter weights of each node, and through the weight aggregation driven by the comprehensive trust degree, the efficiency and accuracy of global parameter aggregation are improved.

[0110] Furthermore, in order to improve the parameter quality, global parameter aggregation is performed from the time dimension, and the above step S3506 may include: Step S35061: Determine the behavior time interval of the historical audit of each external audit node based on the historical audit behavior information; Step S35062: Calculate the time-sensitive trust degree of each external audit node according to the behavior time interval and the comprehensive trust degree; Step S35063: Classify the external audit nodes according to the institution type and the time-sensitive trust degree to obtain multiple local node sets, where each local node set includes one or more external audit nodes; Step S35064: Based on the comprehensive trust degrees of the nodes in each local node set, perform local aggregation on the external node model to obtain the local model parameters of each local node set; Step S35065: Obtain the historical contribution degrees of the nodes in the local node set in the historical model parameters; Step S35066: Determine the set contribution degree of each local node set based on the historical contribution degree, and determine the set trust degree of each local node set based on the comprehensive trust degree; Step S35067: Aggregate the local model parameters based on the set contribution degree and the comprehensive trust degree to obtain the global model parameters.

[0111] It should be noted that the calculation formula for the time-sensitive trust degree is as follows: Among them, represents the time-sensitive trust degree, represents the time decay factor, represents the weight of the current audit behavior, represents the historical audit trust degree, represents the decay rate, represents the behavior time interval.

[0112] It should be noted that the aggregation of the local model parameters refers to the following formula: Among them, represents the global model parameters, represents the local model parameters, represents the balance coefficient, and the balance coefficient is used to balance the comprehensive trust degree and the set contribution degree, represents the set contribution degree, represents the set trust degree.

[0113] It can be understood that the authorized device can identify the activity and response speed of the nodes by analyzing the historical behavior time interval. For example, nodes that frequently participate in audits may be more reliable, while nodes that have not participated for a long time may be abnormal (such as hardware failures or malicious behaviors).

[0114] It should be understood that the authorized device can introduce a time decay factor by combining the time interval and the comprehensive trust level to conduct trust analysis from the time dimension. For example, a node that has not participated in the review for a long time may be marked as "inactive", while a node that frequently participates but has abnormal parameters may be identified as "malicious", thereby reducing its negative impact on the global model.

[0115] It can be understood that in this embodiment, the contribution degree analysis can be carried out from the level of the local set by aggregating the node parameters locally. Based on the sum of the historical contributions of the members, the overall value of the set is evaluated, and the reliability of the set is evaluated through the weighted average of the comprehensive trust levels of the members, thereby improving the efficiency and stability of the global aggregation. For example, a set with high contribution degree but low trust level may be marked as "to be monitored", while a set with high trust level and high contribution degree can be preferentially aggregated.

[0116] It should be understood that in this embodiment, the dynamic balance between trust and contribution is achieved through the time-sensitive trust level and the historical contribution degree, adapting to the real-time changes in the node behavior. Through multi-layer screening (individual trust level, set contribution degree), malicious nodes are effectively filtered, the attack risk is reduced, the local aggregation reduces the global communication overhead, and at the same time, inefficient nodes are avoided from participating through the contribution degree screening.

[0117] Step S36: Train the trained original portrait model based on the global model parameters to obtain a risk portrait evaluation model.

[0118] It can be understood that in this embodiment, the model parameters of the trained original portrait model are updated based on the global model parameters, a data set is generated according to the business log and the authorization behavior time series, and the updated original portrait model is trained based on the data set to obtain a risk portrait evaluation model.

[0119] In this embodiment, by collecting business logs, multi-dimensional time series analysis of the authorization behaviors of branches, customers, and tellers is carried out based on the business logs, so as to capture the dynamic characteristics of business risks. The original portrait model is trained according to the authorization behavior time, so as to provide time series-dependent characteristics for model training, significantly improving the timeliness and accuracy of risk prediction. Through federated learning, the global parameter aggregation is carried out by combining the external node model parameters obtained by multiple external review nodes based on local parameters. While ensuring data privacy, the problem of data islands is effectively avoided, and the accuracy of risk assessment is improved.

[0120] Reference Figure 4 , Figure 4 is a schematic flowchart of the third embodiment of the bank intelligent authorization method based on the shared job system of the present invention.

[0121] Based on the above embodiments, in this embodiment, the step S50 further includes: Step S501: If the authorization review policy is the target individual authorization review policy, obtain the individual type of the current customer of the business to be processed according to the business data.

[0122] It should be noted that the target individual authorization review policy can be applied to the authorization review scenario for a target individual's customer to handle a business. The individual types of the target individual can include agent customers (i.e., the customer entrusts an agent to handle the business), customers below a preset age threshold (i.e., minor customers), elderly customers (e.g., customers over 70 years old), etc.

[0123] Step S502: If the individual type of the current customer is an agent customer, obtain the original image of the power of attorney for the business to be processed.

[0124] It should be noted that if the individual type of the current customer is an agent customer, it is determined that the business to be processed is a business scenario where the customer entrusts an agent to handle the business (i.e., a non-self-handling business scenario). In this scenario, the principal needs to issue a power of attorney, and the entrusted person (i.e., the agent) assists in handling the business.

[0125] It should be understood that since there is no standard template for the power of attorney, it is difficult for machine recognition to accurately identify the content of the power of attorney, and the efficiency and accuracy of manual recognition are relatively low. Therefore, in this embodiment, the original image of the power of attorney is subjected to grayscale processing and calculation of the maximum between-class variance. Based on the target grayscale threshold, the power of attorney image is segmented into a foreground layer and a background layer, thereby effectively narrowing the recognition range. By dividing the foreground layer into a handwritten font area and a stamped font area and performing morphological opening operation processing on each, the noise in the image is effectively reduced, and the missing and broken characters in the font are complemented, improving the recognition accuracy of the handwritten font and the stamped font, and thus accurately identifying the content of the power of attorney, effectively improving the accuracy of authorization review in the business handling scenario by the agent.

[0126] Step S503: Perform grayscale processing on each pixel point in the original image to obtain a grayscale image of the power of attorney.

[0127] It should be noted that this embodiment can preprocess the original image to enhance the image quality and remove noise. In some embodiments, the preprocessing can include grayscale processing of the original image, direction correction processing (e.g., aligning and correcting the font direction of the handwritten font with that of the printed font), Gaussian filter smoothing processing, binarization processing, etc.

[0128] It can be understood that this embodiment can perform grayscale processing on the original image and calculate the grayscale value of each pixel point in the original image to achieve grayscale conversion. The grayscale value calculation refers to the following formula: Where, represents the pixel value of the power of attorney grayscale image at the coordinate , respectively represent the weight coefficients of the red, green, and blue channels, respectively represent the pixel values of the red, green, and blue channels at the coordinate in the original image.

[0129] Step S504: Obtain the grayscale value range of the power of attorney grayscale image and traverse each pixel point of the power of attorney grayscale image; Step S505: Perform grayscale distribution analysis on each pixel point in the power of attorney grayscale image according to the grayscale value range and traversal result, and set an initial grayscale threshold based on the grayscale distribution analysis result.

[0130] In some embodiments, the authorization device can determine the grayscale value range [0, L - 1] of the power of attorney grayscale image, initialize the grayscale value t to 0, start from t = 0, gradually increase t until t = L - 2, calculate the left pixel ratio, right pixel ratio, average grayscale value of the left pixels, average grayscale value of the right pixels, and between-class variance for each t, and find the maximum between-class variance among all possible t values.

[0131] Step S506: Perform maximum between-class variance calculation on the power of attorney grayscale image based on the initial grayscale threshold and the grayscale distribution analysis result to determine the target grayscale threshold.

[0132] It should be noted that the maximum between-class variance is calculated with reference to the following formula: where, represents the target grayscale threshold, represents the initial grayscale threshold, represents the initial grayscale threshold left pixel ratio, represents the initial grayscale threshold right pixel ratio, represents the average grayscale value of the left pixels, represents the average grayscale value of the right pixels.

[0133] Step S507: Segment the power of attorney grayscale image into a foreground layer and a background layer according to the target grayscale threshold.

[0134] It should be noted that in this embodiment, by determining the target grayscale threshold, the pixels in the power of attorney grayscale image are divided into two categories: pixels greater than the target grayscale threshold are marked as the foreground (usually white), and pixels less than the target grayscale threshold are marked as the background (usually black), thereby realizing the segmentation of the layers.

[0135] Step S508: Divide the foreground layer into a handwritten font area and a printed font area.

[0136] It should be noted that the handwritten font area can be the distribution area of the customer's handwritten font in the power of attorney, and the printed font area can be the distribution area of the font printed by the printer in the power of attorney. In an actual power of attorney, there will be handwritten Chinese characters of both the customer himself and the agent. Since there are significant differences in writing styles among different individuals and it is difficult to directly identify them, while printed fonts are usually standard fonts (such as regular script, Song typeface, etc.), this embodiment can improve the accuracy of content recognition in the power of attorney by partitioning and recognizing the handwritten and printed fonts in the power of attorney.

[0137] In some embodiments, the authorization device can extract texture features, edge features, and shape features from the foreground layer to obtain multi-dimensional feature information, classify the fonts based on the multi-dimensional feature information, and locate the handwritten font area and the printed font area.

[0138] For example, for each area in the image, the authorization device can calculate the mean and variance of its LBP histogram to obtain the texture features of the area, compare the texture features of different areas, and distinguish the handwritten font area and the printed font area. Generally speaking, the texture features of the handwritten font area are more complex and the variance of the LBP histogram is larger; while the texture features of the printed font area are relatively simple and the variance of the LBP histogram is smaller.

[0139] For example, for each area in the image, the authorization device can calculate the mean and variance of its edge intensity to obtain the edge features of the area, compare the edge features of different areas, and distinguish the handwritten font area and the printed font area. Generally speaking, the edge intensity of the handwritten font area is larger and the edge direction is more complex; while the edge intensity of the printed font area is smaller and the edge direction is relatively simple.

[0140] For example, for each area in the image, the authorization device can calculate the mean and variance of its area, perimeter, and aspect ratio to obtain the shape features of the area, compare the shape features of different areas, and distinguish the handwritten font area and the printed font area. Generally speaking, the shape features of the handwritten font area are more complex and the variances of the area, perimeter, and aspect ratio are larger; while the shape features of the printed font area are relatively simple and the variances of the area, perimeter, and aspect ratio are smaller.

[0141] Step S509: Obtain the handwritten structural elements of the handwritten font area according to the structural information of the handwritten font area, and obtain the printed structural elements of the printed font area according to the structural information of the printed font area.

[0142] It should be noted that the structuring element is a binary image, and the structuring element can be a simple shape, such as a square, a circle or a cross, which is used to define the neighborhood of the operation.

[0143] It should be explained that the above structural information can be the structural information of handwritten fonts and printed fonts, and can be the font structure feature information of handwritten fonts and printed fonts.

[0144] It can be understood that in this embodiment, an appropriate structuring element can be selected according to the characteristics of handwritten fonts and printed fonts. Generally, the strokes of printed fonts are thicker and more regular, while the strokes of handwritten fonts are thinner and more irregular. Therefore, a larger structuring element can be used to process the printed font area, and a smaller structuring element can be used to process the handwritten font area. For example, a 5x5 square structuring element can be selected to process the printed font area, and a 3x3 square structuring element can be selected to process the handwritten font area.

[0145] Step S510: Perform morphological opening operation processing on the handwritten font area and the printed font area respectively based on the handwritten structuring element and the printed structuring element to obtain a target handwritten image and a target printed image.

[0146] It should be noted that the morphological opening operation processing includes erosion operation processing and dilation operation processing. The erosion operation can be used to eliminate small objects in the image, separate the connections between objects, and smooth the boundaries of objects. In this embodiment, it can be used to eliminate invalid strokes, fonts, and noises in the power of attorney. The dilation operation can be used to fill small holes in the image, connect breakpoints between objects, and smooth the boundaries of objects. In this embodiment, it can be used to connect broken fonts and strokes and repair eroded fonts.

[0147] It can be understood that the erosion operation may include: traversing each pixel in the handwritten font area and the printed font area, aligning the center of the structuring element with the pixel, and checking all pixel values in the image area covered by the structuring element. If all pixel values covered by the structuring element are 1, then keep the pixel as 1; otherwise, set the pixel to 0.

[0148] It should be understood that the dilation operation may include: traversing each pixel in the handwritten font area and the printed font area, aligning the center of the structuring element with the pixel, and checking all pixel values in the image area covered by the structuring element. If any pixel value in the area covered by the structuring element is 1, then set the pixel to 1; otherwise, keep the pixel as 0.

[0149] In some embodiments, the authorization device may use a larger structure element to perform an opening operation on the printed font area to remove noise and connect broken strokes. For example, the erosion operation may use a 5x5 square structure element to perform an erosion operation on the printed font area to remove noise and small objects in the image; the dilation operation may use the same 5x5 square structure element to perform a dilation operation on the eroded image to connect broken strokes, making the printed font area more complete and clear.

[0150] In some embodiments, the authorized device may use a smaller structure element to perform an opening operation on the handwritten font area to remove noise and retain the details of the handwritten font. For example, the erosion operation may use a 3x3 square structure element to perform an erosion operation on the handwritten font area to remove noise and small objects in the image; the dilation operation may use the same 3x3 square structure element to perform a dilation operation on the eroded image to retain the details of the handwritten font, making the handwritten font area clearer.

[0151] Step S511: performing content detection on the target handwritten image and the target printed image to obtain content information of the power of attorney for the pending business.

[0152] In some embodiments, the authorization device may extract print features of the target printed image, perform content recognition on the target printed image based on the print features and a pre-built deep learning model (eg, a CRNN model), and obtain the print content text.

[0153] In some embodiments, the authorization device may extract handwriting features of the target handwriting image, perform content recognition on the target handwriting image based on the handwriting features and a pre-built attention mechanism model (eg, a Transformer model), and obtain handwriting content text.

[0154] Step S512: Perform authorization review on the pending business based on the content information of the power of attorney to obtain the authorization review result.

[0155] In some embodiments, the authorization device can extract the key information of the entrustment (such as the entrusted business, the entrusted amount, the entrustment date, the signature, etc.) based on the content information of the power of attorney, match and compare the key information of the entrustment with the key business information in the business data, and detect whether there is any key information that fails to match (such as the entrustment amount does not match the business amount, etc.).

[0156] In some embodiments, the authorizing device may locate the principal signature area and the agent (i.e., the entrusted party) signature area based on the target printed image, detect the handwriting density of the principal signature area and the agent signature area, and determine whether there is valid handwriting in both the principal signature area and the agent signature area based on the handwriting density. If there is valid handwriting, the historical signature of the principal and the historical signature of the agent are obtained, the features of the historical signature of the principal are compared with the signature in the principal signature area, and the features of the historical signature of the agent are compared with the signature in the agent signature area to detect whether there is a forgery risk in the signature in the target handwritten image.

[0157] In this embodiment, by performing grayscale processing and foreground segmentation on the power of attorney image, the handwritten font area and the printed font area in the power of attorney image are accurately classified. By performing morphological opening operations on the handwritten font area and the printed font area respectively, the noise in the image is effectively reduced, and the missing and broken characters in the font are complemented, improving the recognition accuracy of the handwritten font and the printed font, so as to accurately recognize the content in the power of attorney and effectively improve the accuracy of authorization review in the scenario of the agent handling business.

[0158] In addition, an embodiment of the present invention further provides a computer-readable storage medium, on which a bank intelligent authorization program based on a shared job system is stored. When the bank intelligent authorization program based on the shared job system is executed by a processor, the steps of the bank intelligent authorization method based on the shared job system as described above are implemented.

[0159] The computer-readable storage medium provided by this application can be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or components, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: electrical connections with 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 of the above. In this embodiment, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, device, or component. The program code contained on the computer-readable storage medium can be transmitted by any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0160] The above computer-readable storage medium can be included in a bank intelligent authorization device based on a shared job system; it can also exist independently and not be assembled into a bank intelligent authorization device based on a shared job system.

[0161] In addition, an embodiment of the present invention also proposes a computer program product, including a bank intelligent authorization program based on a shared job system. When the bank intelligent authorization program based on the shared job system is executed by a processor, it implements the steps of the bank intelligent authorization method based on the shared job system as described above.

[0162] The specific implementation manner of the computer program product of the present invention is basically the same as that of each embodiment of the above bank intelligent authorization method based on a shared job system, and will not be elaborated here.

[0163] Refer to Figure 5 , Figure 5 , which is the structural block diagram of the first embodiment of the shared job system of the present invention.

[0164] As Figure 5 shown, the shared job system proposed by the embodiment of the present invention includes: An authorization request response module 10, configured to respond to a service authorization request sent by a business counter, and obtain service data of a service to be processed based on the service authorization request, where the service data includes a service type; A risk level assessment module 20, which is used to assess the risk level of the business to be processed based on the business type and determine the business risk level of the business to be processed; A risk profile generation module 30, which is used to input the branch data of the business branch to which the business counter belongs, the customer data of the customer handling the business to be processed, and the teller data of the teller handling the business to be processed into a risk profile assessment model to generate a risk profile of the business to be processed. The risk profile assessment model is constructed based on multi-dimensional feature data of multiple external audit nodes, and the multi-dimensional feature data includes customer feature data, branch feature data, and teller feature data; An audit policy matching module 40, which is used to match the authorization audit policy corresponding to the business to be processed according to the risk profile and the business risk level. The authorization audit policy includes a high-risk authorization audit policy and a medium-low risk authorization audit policy. The high-risk business handling authorization audit policy includes a target individual authorization audit policy and a target business authorization audit policy. The target individuals include agent customers and customers whose age is lower than a preset age threshold. The target businesses include fund flow business, loan business, and corporate business. The medium-low risk authorization audit policy includes determining whether to directly authorize based on the daily authorization ratio of the business counter; An authorization audit module 50, which is used to conduct an authorization audit on the business to be processed according to the authorization audit policy and the business data to obtain an authorization audit result; An authorization quality inspection module 60, which is used to conduct an abnormal time series behavior analysis based on the historical business data of the business branch and the customer historical business data of the handling customer, and conduct quality inspection on the authorization audit result based on the abnormal time series behavior analysis result; An authorization feedback module 70, which is used to correct the authorization audit result based on the quality inspection result to obtain a target authorization result and feedback the target authorization result to the business counter.

[0165] In this embodiment, by conducting risk profile assessments on the business from multiple dimensions respectively, the problem of data islands is effectively avoided, and the limitations existing in the risk assessment of single-dimensional data are overcome. Through dynamic assessment of the risk level and risk profile, potential risks and anomalies in the business are accurately identified, and the corresponding authorization audit policy is accurately matched for the business to be processed. The business audit is classified into high-risk authorization audit and medium-low risk authorization audit, thereby strengthening the strictness of the audit for high-risk businesses. Whether to directly authorize is determined based on the daily authorization ratio, thereby reducing the redundant audit links for medium-low risk businesses, and thus greatly improving the efficiency of authorization audit. The accurate quality inspection of the authorization result is achieved by analyzing historical business data for abnormal time series behavior analysis, and the deviation degree between the customer's current behavior and the historical behavior quality inspection is accurately identified. The risk of misauthorization is effectively reduced through the quality inspection and correction mechanism.

[0166] The shared job system provided by this application adopts the bank intelligent authorization method based on the shared job system in the above embodiments, and can solve the technical problems of bank intelligent authorization. Compared with the prior art, the beneficial effects of the shared job system provided by this application are the same as those of the bank intelligent authorization method based on the shared job system provided by the above embodiments, and other technical features in the shared job system are the same as the features disclosed in the above embodiment method, which will not be elaborated herein.

[0167] It should be understood that the above is only an example for illustration and does not constitute any limitation to the technical solution of the present invention. In specific applications, those skilled in the art can set according to needs, and the present invention does not limit this.

[0168] It should be noted that the above-described work process is only illustrative and does not limit the protection scope of the present invention. In actual applications, those skilled in the art can select some or all of them according to actual needs to achieve the purpose of the solution of this embodiment, and there is no limitation here.

[0169] In addition, for the technical details not described in detail in this embodiment, reference can be made to the bank intelligent authorization method based on the shared job system provided in any embodiment of the present invention, which will not be elaborated herein.

[0170] It should be noted that in this article, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or system. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or system including that element.

[0171] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

[0172] Through the description of the above embodiments, those skilled in the art can clearly understand that the above method of the embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is a better implementation. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as a read-only memory / random access memory, magnetic disk, optical disk), and includes several instructions for causing a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0173] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. A bank intelligent authorization method based on a shared job system, applied to the shared job system, characterized in that The method includes: In response to a business authorization request sent by a business counter, obtaining business data of a business to be processed based on the business authorization request, where the business data includes a business type; Evaluating the risk level of the business to be processed based on the business type to determine the business risk level of the business to be processed; Inputting the branch data of the business branch to which the business counter belongs, the customer data of the customer handling the business to be processed, and the teller data of the teller handling the business to be processed into a risk portrait evaluation model to generate a risk portrait of the business to be processed, where the risk portrait evaluation model is constructed based on multi-dimensional feature data of multiple external audit nodes, and the multi-dimensional feature data includes customer feature data, branch feature data, and teller feature data; Matching an authorization audit policy corresponding to the business to be processed according to the risk portrait and the business risk level, where the authorization audit policy includes a high-risk authorization audit policy and a medium-low risk authorization audit policy, the high-risk business handling authorization audit policy includes a target individual authorization audit policy and a target business authorization audit policy, the target individuals include agent customers and customers whose age is lower than a preset age threshold, the target businesses include fund flow businesses, loan businesses, and corporate businesses, and the medium-low risk authorization audit policy includes determining whether to directly authorize based on the daily authorization ratio of the business counter; Conducting an authorization audit on the business to be processed according to the authorization audit policy and the business data to obtain an authorization audit result; Performing abnormal time series behavior analysis based on the branch historical business data of the business branch and the customer historical business data of the customer handling the business, and performing quality inspection on the authorization audit result based on the result of the abnormal time series behavior analysis; Correcting the authorization audit result based on the quality inspection result to obtain a target authorization result, and feeding back the target authorization result to the business counter.

2. The bank intelligent authorization method based on the shared job system according to claim 1, wherein Before inputting the branch data of the business branch to which the business counter belongs, the customer data of the customer handling the business to be processed, and the teller data of the teller handling the business to be processed into a risk portrait evaluation model to generate a risk portrait of the business to be processed, it further includes: Obtaining business logs, where the business logs include branch business logs, customer business logs, and teller business logs; Generating an authorization behavior time series based on the business logs, and training a pre-constructed original portrait model according to the authorization behavior time series; Obtaining initial model parameters according to the trained original portrait model; Performing hash encryption on the initial model parameters to generate encrypted model parameters, and sending the encrypted model parameters to each external audit node; In response to external node model parameters sent by each external audit node, aggregating the external node model parameters to obtain global model parameters, where the external node model parameters are obtained by each external audit node decrypting the encrypted model parameters and training a local portrait model based on the decrypted initial model parameters; Training the trained original portrait model based on the global model parameters to obtain a risk portrait evaluation model.

3. The bank intelligent authorization method based on the shared job system according to claim 2, characterized in that Generating an authorization behavior time series based on the business log and training a pre-constructed original portrait model according to the authorization behavior time series includes: Extracting log semantic information and timestamp information from the business log; Obtaining historical event information based on the log semantic information, where the historical event information includes branch event information, teller event information, and customer event information; Performing event behavior analysis according to the historical event information to obtain event behavior characteristics of each historical event; Constructing an event causal graph based on the event behavior characteristics; Generating an authorization behavior time series according to the timestamp information and constructing an event graph based on the authorization behavior time series; Establishing a path index according to the causal path in the causal graph and the temporal path in the event graph; Extracting event nodes from the event graph and extracting causal pair nodes from the causal graph; Merging the event graph and the causal graph based on the path index, the event nodes, and the causal pair nodes to generate a target hypergraph, where the target hypergraph includes causal hyperedges and temporal hyperedges; Aggregating the node information of each hyperedge neighbor node in the target hypergraph to obtain a hypergraph embedding feature vector; Performing feature fusion on the original features of each node in the target hypergraph and the hypergraph embedding feature vector to obtain a comprehensive feature vector; Generating a data set based on the comprehensive feature vector and training a pre-constructed original portrait model based on the data set.

4. The bank intelligent authorization method based on a shared job system according to claim 3, wherein Aggregating the external node model parameters to obtain global model parameters, including: Determining the institution type of each external audit node and obtaining the historical audit behavior information of each external audit node; Performing feature analysis on the historical audit behavior information to obtain multi-dimensional behavior feature information, where the multi-dimensional behavior feature information includes response delay characteristics, parameter noise characteristics, historical abnormal behavior characteristics, and network stability characteristics of each external audit node; Normalizing the multi-dimensional behavior feature information and calculating the direct trust degree of each external audit node based on the normalized multi-dimensional behavior feature information: Among them, represents the direct trust level, respectively represent the weight coefficients of each behavioral feature, represents the normalized response delay feature, represents the normalized parameter noise feature, represents the normalized historical abnormal behavior feature, represents the normalized network stability feature; Obtaining the historical credibility of each external audit node and the evaluation information between each external audit node, and calculating the recommended trust degree of each external audit node based on the evaluation information: Among them, represents the recommended trust level of the external audit node ; represents the set of other external audit nodes that have interacted with the external audit node ; represents the historical trust level of the external audit node ; represents the evaluation information of the external audit node on the external audit node ; Aggregating the direct trust degree and the recommended trust degree to obtain the comprehensive trust degree of each external audit node: Among them, represents the fusion weight coefficient, represents the comprehensive trust degree; Aggregating the external node model parameters based on the comprehensive trust degree and the institution type to obtain global model parameters.

5. The bank intelligent authorization method based on the shared job system according to claim 4, characterized in that Aggregating the external node model parameters based on the comprehensive trust degree and the institution type to obtain global model parameters, including: Determining the behavior time interval of each external audit node's historical audit based on the historical audit behavior information; Calculating the time-sensitive trust degree of each external audit node according to the behavior time interval and the comprehensive trust degree: Among them, represents the time-sensitive trust level, represents the time decay factor, represents the weight of the current audit behavior, represents the historical audit trust level, represents the decay rate, represents the behavior time interval; Classifying the external audit nodes according to the institution type and the time-sensitive trust degree to obtain a plurality of local node sets, where each local node set includes one or more external audit nodes; Locally aggregate the external node model based on the comprehensive trustworthiness of each node in each local node set to obtain the local model parameters of each local node set; Obtain the historical contribution degrees of each node in the local node set in the historical model parameters; Determine the set contribution degree of each local node set based on the historical contribution degree, and determine the set trustworthiness of each local node set based on the comprehensive trustworthiness; Aggregate the local model parameters based on the set contribution degree and the comprehensive trustworthiness to obtain the global model parameters: Among them, represents the global model parameters, represents the local model parameters, represents the balance coefficient, and the balance coefficient is used to balance the comprehensive trust degree and the set contribution degree, represents the set contribution degree, represents the set trust degree.

6. The bank intelligent authorization method based on the shared job system according to any one of claims 1 to 5, characterized in that, The authorization review of the business to be processed according to the authorization review policy and the business data to obtain an authorization review result includes: If the authorization review policy is the target individual authorization review policy, obtain the individual type of the current customer of the business to be processed according to the business data; If the individual type of the current customer is an agent customer, obtain the original image of the power of attorney for the business to be processed; Perform gray-scale processing on each pixel point in the original image to obtain a gray-scale power of attorney image: Among them, represents the pixel value of the power of attorney grayscale image at the coordinate , respectively represent the weight coefficients of the red, green, and blue channels, respectively represent the pixel values of the red, green, and blue channels at the coordinate in the original image; Obtain the gray-scale value range of the gray-scale power of attorney image, and traverse each pixel point of the gray-scale power of attorney image; Perform gray-scale distribution analysis on each pixel point in the gray-scale power of attorney image according to the gray-scale value range and the traversal result, and set an initial gray-scale threshold based on the gray-scale distribution analysis result; Perform maximum between-class variance calculation on the gray-scale power of attorney image based on the initial gray-scale threshold and the gray-scale distribution analysis result to determine the target gray-scale threshold: Among them, represents the target gray threshold, represents the initial gray threshold, represents the initial gray threshold proportion of left - hand pixels, represents the initial gray threshold proportion of right - hand pixels, represents the average gray value of left - hand pixels, represents the average gray value of right - hand pixels; Slice the gray-scale power of attorney image into a foreground layer and a background layer according to the target gray-scale threshold; Divide the foreground layer into a handwritten font area and a printed font area; Obtain the handwritten structure element of the handwritten font area according to the structure information of the handwritten font area, and obtain the printed structure element of the printed font area according to the structure information of the printed font area; Perform morphological opening operation processing on the handwritten font area and the printed font area respectively based on the handwritten structure element and the printed structure element to obtain a target handwritten image and a target printed image, and the morphological opening operation processing includes erosion operation processing and dilation operation processing; Perform content detection on the target handwritten image and the target printed image to obtain the power of attorney content information of the business to be processed; Perform authorization review on the business to be processed based on the power of attorney content information to obtain an authorization review result.

7. The bank intelligent authorization method based on a shared job system according to claim 6, wherein The dividing the foreground layer into a handwritten font area and a printed font area includes: Calculate the attribution cost of each pixel point in the foreground layer, and divide the foreground layer into multiple superpixel blocks based on the attribution cost: Among them, represents the attribution cost of the pixel at the coordinate . represents the gradient value of the pixel at the coordinate . represents the average gradient value of the superpixel cluster, represents the global gradient standard deviation of the foreground layer, represents the spatial distance between the pixel at the coordinate and the center of the superpixel cluster, represents the size parameter of the superpixel, respectively represent the balance weight coefficients of the gradient and the distance; Perform character target detection on each superpixel block, and perform border annotation on each superpixel block based on the character target detection result to generate multiple character borders; Extract features of each character border and the characters within the character border to obtain the border feature vector and the character structure feature vector of each character border, and determine the similarity of each character border based on the border feature vector and the character structure feature vector; Calculate the information entropy of each superpixel block based on the similarity; Among them, represents the information entropy of the superpixel block , represents the character border with a similarity lower than the preset similarity threshold appearing in the superpixel block with a probability; Classify the superpixel blocks into a handwritten font area and a printed font area based on the information entropy, where the information entropy of the handwritten font area is greater than a preset entropy threshold, and the information entropy of the printed font area is not greater than the preset entropy threshold.

8. A shared job system, characterized in that, The shared operation system includes: An authorization request response module, configured to respond to a business authorization request sent by a business counter, and obtain business data of a to-be-handled business based on the business authorization request, where the business data includes a business type; A risk level assessment module, configured to perform a risk level assessment on the to-be-handled business based on the business type, and determine the business risk level of the to-be-handled business; A risk portrait generation module, configured to input the branch data of the business branch to which the business counter belongs, the customer data of the customer handling the to-be-handled business, and the teller data of the teller handling the to-be-handled business into a risk portrait assessment model to generate a risk portrait of the to-be-handled business, where the risk portrait assessment model is constructed based on multi-dimensional feature data of multiple external audit nodes, and the multi-dimensional feature data includes customer feature data, branch feature data, and teller feature data; An audit policy matching module, configured to match an authorization audit policy corresponding to the to-be-handled business according to the risk portrait and the business risk level, where the authorization audit policy includes a high-risk authorization audit policy and a medium-low risk authorization audit policy, the high-risk business handling authorization audit policy includes a target individual authorization audit policy and a target business authorization audit policy, the target individuals include agent customers and customers whose age is lower than a preset age threshold, the target businesses include fund flow businesses, loan businesses, and corporate businesses, and the medium-low risk authorization audit policy includes determining whether to directly authorize based on the daily authorization ratio of the business counter; An authorization audit module, configured to perform an authorization audit on the to-be-handled business according to the authorization audit policy and the business data, and obtain an authorization audit result; An authorization quality inspection module, configured to perform abnormal time series behavior analysis based on the historical business data of the business branch and the historical business data of the customer handling the business, and perform quality inspection on the authorization audit result based on the abnormal time series behavior analysis result; An authorization feedback module, configured to correct the authorization audit result based on the quality inspection result to obtain a target authorization result, and feedback the target authorization result to the business counter.

9. A bank intelligent authorization device based on a shared job system, characterized in that The bank intelligent authorization device based on the shared operation system includes: a memory, a processor, and a bank intelligent authorization program based on the shared operation system stored on the memory and executable on the processor, where the bank intelligent authorization program based on the shared operation system is configured to implement the bank intelligent authorization method based on the shared operation system according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer-readable storage medium stores a bank intelligent authorization program based on the shared operation system, and when the bank intelligent authorization program based on the shared operation system is executed by a processor, it implements the bank intelligent authorization method based on the shared operation system according to any one of claims 1 to 7.

Images