Banking business digital management system based on deep learning algorithm

Abstract

The invention provides a bank business digital management system based on a deep learning algorithm, which relates to the field of digital management and specifically comprises a data acquisition and processing module, a data evaluation module, a deep learning model construction module, an intelligent decision module and an automatic error adjustment module. According to the system, basic information, transaction records and imaging receipts of customers are acquired in real time through the integrated data acquisition and processing module; a multi-layer neural network is adopted to carry out multi-modal fusion analysis on the customer information, a dynamic credit score is generated, and repayment behavior prediction is optimized through an LSTM network; constructing a credibility correction model based on the negative correlation between the identity verification duration and the accuracy; dynamically matching a differentiated loan limit rule according to the credit score; the system aims to solve the problems of customer layering deficiency, low credit evaluation efficiency, insufficient identity verification credibility and the like in the existing bank business management.

Description

Technical Field

[0001] The present invention relates to the field of digital management, and in particular to a digital banking management system based on a deep learning algorithm. Background Art

[0002] With the rapid development of information science and communication technology, in order to improve the service efficiency of banks and user experience, smart banking is becoming an increasingly popular trend in banking development. Smart banking is an advanced stage of traditional banking and online banking. It is a banking enterprise that uses intelligent means and new thinking patterns to examine its own needs, and uses innovative technologies to shape new services, new products, new operations and business models to achieve economies of scale, improve efficiency and reduce costs, and achieve the goals of effective customer management and efficient marketing performance.

[0003] Among them, the intelligence of the business system is the top priority of smart banking. The business system is the core part of the entire bank. In recent years, with the gradual development and maturity of mobile payment technology, people's payment habits have gradually changed, and people's dependence on cash has gradually decreased. Therefore, how to realize the digital management of business has become a technical problem that needs to be solved urgently. It is urgent to develop a digital management system for banking business that integrates multimodal deep learning algorithms, has dynamic decision-making capabilities and meets regulatory requirements to solve core pain points such as data silos, delayed decision-making, and uncontrollable risks. Summary of the Invention

[0004] To solve the above problems, the present invention provides a digital banking management system based on deep learning algorithms, which aims to solve the problems existing in existing banking business management, such as lack of customer stratification, low credit assessment efficiency and insufficient identity authentication credibility.

[0005] To achieve the above objectives, the present invention adopts a technical solution: a digital banking management system based on a deep learning algorithm, comprising: A data collection and processing module is used to collect multi-source banking business data and construct a local feature map based on feature vectors after pre-processing the multi-source banking business data; The data evaluation module is used to establish a dynamic credit assessment model based on the local feature map, calculate the customer's actual personal credit score, and then calculate the repayment time required for various bank credit lines based on the personal credit score; A deep learning model building module, used to build a deep learning model and map time demand into a prediction of the customer's future overdue risk; The intelligent decision-making module is used to calculate the comprehensive score of each credit line business handled by the customer based on the customer's actual personal ability credit score, and output the business with the highest score as the recommended business; The automated error adjustment module is used to compare the customer's actual personal ability credit score with the digital personal ability credit score predicted by the deep learning model, and to build a feedback difference mechanism to iteratively correct the deep learning model and time requirements.

[0006] Preferably, in the data collection and processing module, the multi-source banking business data includes basic customer information, transaction records, risk assessment indicators, loan purpose characteristics and imaged documents.

[0007] Preferably, the data evaluation module establishes a dynamic credit evaluation model based on the local feature map, and predicts a personal credit score by comprehensively considering multiple sources of customer information, wherein the multiple sources of customer information include basic information, transaction records, risk assessment indicators, and loan purposes. The specific calculation equation for the personal credit score is as follows:

[0008] Among them, Score i represents the personal credit score of customer i, k represents the personal deviation coefficient, D i represents the average monthly income of customer i, Davg represents the local per capita income, and H i represents the additional income of user i, represents the impact function of user i’s average monthly consumption on his personal credit score, where , is a constant used to adjust the impact of the user's average monthly consumption on the personal ability credit score; α represents the mutual influence coefficient between the average monthly income and the average monthly consumption; C i represents the average monthly consumption of user i.

[0009] More optimally, the personal credit score is converted into the time required to repay various bank credit lines, that is, the average monthly income and average monthly consumption combination required by user i to repay the required business. For this purpose, the following time demand prediction formula is proposed:

[0010] Among them, T i represents the time requirement of region i; c1 is a constant, representing the personal ability credit score Score i The influence coefficient on time demand; c2 is a constant, which represents the influence coefficient of average monthly income and average monthly consumption on time demand; represents the comprehensive impact function of average monthly income and average monthly consumption on time demand, where , which is the weighted combination of average monthly income and average monthly consumption; Represents the influence coefficient of user deposit on time demand.

[0011] Preferably, a deep learning model is established through the deep learning model construction module, time requirements are input into the deep learning model, and the customer repayment behavior time series data is analyzed through the LSTM network to generate a digital personal ability credit score.

[0012] Preferably, the intelligent decision-making module calculates the customer's actual personal ability credit score based on the data evaluation module, obtains the comprehensive score of the customer's various credit limit businesses through a weighted fusion algorithm, and pushes the business with the highest score as the optimal business to the customer for processing.

[0013] Preferably, the automated error adjustment module generates a feedback correction signal by establishing a feedback difference mechanism; uses the feedback difference mechanism as a correction factor for the personal ability credit score output by the dynamic credit assessment model, and adjusts the prediction result of the time demand in the data assessment module.

[0014] More preferably, when the feedback error is greater than a preset threshold, it indicates that the deviation between the predicted state and the actual state is large, and the setting parameters of the dynamic credit assessment model need to be adjusted to reduce the error and optimize the prediction effect; when the feedback error is less than the preset threshold, it indicates that the current prediction effect is close to the actual state, and the dynamic credit assessment model can maintain the existing parameters without further adjustment.

[0015] The beneficial effects of the present invention are: 1. Improved multimodal data fusion and precise decision-making capabilities: By integrating multiple data sources, including basic customer information, transaction records, and imaged receipts, and combining the Gaussian kernel function anomaly detection algorithm with U-Net privacy masking technology, this system achieves efficient data cleaning and secure processing, effectively improving the efficiency of data correlation analysis and reducing manual labeling costs. Furthermore, it dynamically generates credit scores through multimodal deep learning models (such as RNN and CNN), improving the accuracy of marketing strategy matching. 2. Dynamic credit assessment and risk prevention and control optimization: An LSTM network is used to analyze customer repayment behavior time series data, combined with a variance-attention mechanism to extract loan usage characteristics (such as production / infrastructure classification). This allows for the construction of a dynamic credit scoring model, effectively improving the accuracy of overdue risk predictions and reducing bad debt losses caused by credit assessment bias. This model also supports differentiated loan amount calculations, increasing the success rate of high-value customer marketing. 3. Adaptive security verification and enhanced customer experience: Based on the negative correlation between identity verification time and accuracy, a credibility correction model has been developed. When facial recognition time exceeds a threshold, it automatically switches to multi-factor authentication (such as voiceprint + dynamic password). This shortens the verification process, reduces the false positive rate, and reduces the proportion of customers abandoning business due to cumbersome verification. At the same time, blockchain technology ensures the traceability of operation logs, meeting financial regulatory compliance requirements. 4. Business process automation and operational efficiency improvement: The intelligent decision-making module pushes approval results to the target device in real time, triggering the generation of electronic contracts and automated cross-system processes (such as fund transfers and risk warnings). This significantly shortens the loan approval cycle, reduces the workload of manual review, and enables cross-departmental data collaboration through federated learning technology, improving business process response efficiency. 5. Personalized service and enhanced customer stickiness: Utilize deep learning algorithms to analyze historical customer behavior data (such as transaction preferences and channel usage habits), generate dynamic customer portraits, and match personalized product recommendations. This improves customer satisfaction, increases marketing response rates, and reduces customer churn rates. Furthermore, through A / B testing mechanisms, channel combination strategies are optimized to reduce marketing costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a block diagram of a digital banking management system based on a deep learning algorithm in the present invention. DETAILED DESCRIPTION

[0017] See also Figure 1 As shown, the present invention relates to a digital banking management system based on a deep learning algorithm, comprising: A data collection and processing module is used to collect multi-source banking business data and construct a local feature map based on feature vectors after pre-processing the multi-source banking business data; Through the bank system, we retrieve different multi-source information of customers, such as basic customer information, deposit size, transaction records, risk assessment indicators, loan purpose characteristics and imaged documents (contract scans), to form a local feature map F. i This data will provide the foundation for the subsequent development of dynamic credit assessment models and deep learning models. The innovation of the data acquisition and processing module lies in the collection and mapping of multi-source information from customers based on banks, providing accurate local data for subsequent adaptive control.

[0018] The basic information of customers can be read through the bank system or bank operation terminal, such as collecting the scanned copy of the customer's business license, tax data in the past three years, and voiceprint verification of the legal representative through the mobile terminal; each piece of information collected constitutes a feature vector F i ,in , that is, local feature map. The data includes: D i : average monthly income of customer i; H i : Additional income of user i; C i : Average monthly consumption of user i.

[0019] Multi-source bank data for each customer provides foundational information for subsequent time demand forecasting and error forecast adjustment. These feature maps can reflect the customer's true repayment ability and credit status, thus helping the system make personalized adjustments.

[0020] The data collection and processing module collects the user's bank multi-source data and combines it with the weighted model to transform the customer data features into a refined local feature map F i , providing accurate regionalized data for subsequent steps. This innovative design avoids the traditional "global adjustment" problem.

[0021] The data evaluation module is used to establish a dynamic credit assessment model based on the local feature map, calculate the customer's actual personal credit score, and then calculate the repayment time required for various bank credit lines based on the personal credit score; Using the local feature map F provided by the data acquisition and processing module i (i.e., the average monthly income of customer i, D i 、Extra income H i , average monthly consumption C i Based on the data, a dynamic credit assessment model is established. This model is used to predict customer i's personal credit score and time requirements (the time required to repay various bank lines, and the combination of average monthly income and average monthly consumption).

[0022] The specific steps are as follows: An improved credit assessment equation is proposed. By comprehensively considering these factors from multiple sources of bank data, the customer's actual personal credit score is calculated. The specific equation is as follows:

[0023] Among them, Score i represents the personal credit score of customer i, k represents the personal deviation coefficient, D i represents the average monthly income of customer i, D avg represents the local per capita income, H i represents the additional income of user i, represents the impact function of user i’s average monthly consumption on his personal credit score, where , is a constant used to adjust the impact of the user's average monthly consumption on the personal ability credit score; α represents the mutual influence coefficient between the average monthly income and the average monthly consumption; C i represents the average monthly consumption of user i.

[0024] Operation process: Calculate personal credit score i :First, based on the average monthly income D of customer i i and local per capita income D avg Calculate the income difference , the greater the income gap, the higher the personal ability credit score; Consider additional income H i :By considering the positive effect of extra income on personal ability credit score, it is directly multiplied by the extra income value. The greater the extra income, the higher the personal ability credit score, and the extra income value should be greater than or equal to 1. If there is no extra income, H i =1; The impact of average monthly consumption on personal credit score: through the consumption function , the higher the consumption, the smaller the evaporation rate; The interaction between average monthly income and average monthly consumption: Item, considering the accelerating effect of average monthly income on average monthly consumption, especially when the average monthly income is high, changes in consumption have a stronger driving effect on personal ability credit scores.

[0025] This comprehensive equation takes into account the interaction term between income and consumption, providing a more refined prediction of personal ability credit score.

[0026] Local feature map composed of multi-source information retrieved from the banking system The average monthly income, additional income and average monthly consumption of customers have been obtained, which provides a basis for time demand forecasting. i Converted into time requirement T i , that is, the average monthly income and average monthly consumption required by customer i to achieve the required repayment business. For this purpose, the following time demand prediction formula is proposed. The specific steps are as follows: The personal credit score is converted into the time required to repay various bank credit lines. For this purpose, the following energy demand prediction formula is proposed:

[0027] Among them, T i represents the time requirement of region i; c1 is a constant, representing the personal ability credit score Score iThe influence coefficient on time demand; c2 is a constant, which represents the influence coefficient of average monthly income and average monthly consumption on time demand; represents the comprehensive impact function of average monthly income and average monthly consumption on time demand, where , which is the weighted combination of average monthly income and average monthly consumption; Represents the influence coefficient of user deposit on time demand.

[0028] Credit score based on personal ability i The weighted relationship between income and consumption is used to calculate the time required for user i to repay the required service. i ; Weighted effect of additional income: , can dynamically adjust the customer's time demand T i , avoid excessive repayment time or early repayment; The combined impact of average monthly income and average monthly consumption: , combined with the interactive effect of average monthly income and average monthly consumption, to provide a more accurate forecast of the combination of average monthly income and average monthly consumption.

[0029] The data evaluation module predicts the personal credit score of customer i through the dynamic credit evaluation model. i , and further personal ability credit score Score i Converted into the time required to repay each bank's business line T i , providing a precise basis for subsequent business push notifications. Innovative designs introduced into the model, such as the nonlinear interaction term between average monthly income and average monthly consumption, and the dynamic weighting of additional income, make individual credit scores and time demand predictions more accurate, adapting in real time to changes in a customer's income process, thereby achieving true adaptive prediction.

[0030] A deep learning model building module, used to build a deep learning model and map time demand into a prediction of the customer's future overdue risk; The deep learning model construction module is to calculate the time required to repay each bank's business line in the data evaluation module. i The data is input into the LSTM network to analyze the customer's repayment behavior time series data, generate a digital personal ability credit score, and predict the overdue risk of various banking businesses.

[0031] Overdue risk prediction Input: The customer's repayment record for the past 24 months, the sequence of their operations, and the time required to repay each line of credit required by the bank. i ; Output: Probability of overdue payment in the next three months (KS value reaches 0.42 when threshold = 0.65); Decision-making application: Initiate collection strategies for high-risk customers in advance (such as shortening the repayment reminder cycle).

[0032] Differentiated quota management Dynamic adjustment rules: When the predicted overdue probability is less than 5%, the credit limit is increased by 20%. When 5%≤Probability<15%: Maintain the original amount; When the probability is ≥15%, manual review is triggered.

[0033] Anti-fraud scenarios: Real-time detection: When the number of logins per day increases suddenly (>10 times) and the repayment amount is abnormal (<50% of the repayment amount), the LSTM model is triggered to reassess the risk level; Example: A customer had a stable repayment history, but suddenly and repeatedly changed their contact information and attempted to settle their loan early. The model identified this as a risk of account theft (with 92% accuracy). The intelligent decision-making module is used to calculate the comprehensive score of each credit line business handled by the customer based on the customer's actual personal ability credit score, and output the business with the highest score as the recommended business; Define a regression model to calculate the comprehensive score H of each credit line business t The formula is as follows:

[0034] in, is the score of the previous banking business handled by the customer; C is the ratio of the distance between the additional income, which is used to consider the impact of additional income on consumption; It is the ratio of the customer's average monthly income to the local per capita income, used to quantify the impact of income changes on the customer's creditworthiness; is the error term, which represents external factors or personalized responses of users; These are the customers' comprehensive scores for each credit line business. t The weight coefficient of The model is further refined to account for individual user responses. For example, some users may be interested in higher-risk investments, which represent a greater source of additional income, while others may prefer lower-risk, more stable part-time jobs. In this case, the model introduces a "preference correction term" to dynamically adjust the risk factor: By introducing a preference correction term to dynamically adjust risky businesses, high-risk products and businesses will only be recommended to customers when the user's risk score exceeds a preset threshold, where:

[0035] in, represents the preference correction coefficient.

[0036] An automated error adjustment module compares a customer's actual personal credit score with the digital personal credit score predicted by the deep learning model, and establishes a feedback difference mechanism to iteratively correct the deep learning model and time requirements; A feedback correction signal is generated through the feedback difference mechanism; the feedback difference mechanism is used as a correction factor for the personal ability credit score output by the dynamic credit evaluation model to adjust the prediction result of the time demand in the data evaluation module.

[0037] By defining a feedback error function to evaluate the difference between the predicted state and the actual state, this function only models state prediction and does not perform execution control:

[0038] in, Represents the feedback error, which is used to evaluate the deviation between the current prediction result and the model expectation; H t Indicates the comprehensive score of customers for each credit line business predicted by the model; Indicates the risk value score of the customer predicted by the model; Represent the customer's actual comprehensive score and risk value score respectively; A weighting factor for the bias, used to control error sensitivity.

[0039] When the feedback error When the feedback error is greater than the preset threshold, it means that the deviation between the predicted state and the actual state is large, and the setting parameters of the dynamic credit evaluation model need to be adjusted to reduce the error and optimize the prediction effect; when the feedback error is greater than the preset threshold, it means that the deviation between the predicted state and the actual state is large, and the setting parameters of the dynamic credit evaluation model need to be adjusted to reduce the error and optimize the prediction effect. When it is less than the preset threshold, it means that the current control effect is close to the expectation and the system can maintain the existing output without further adjustment.

[0040] The following is the working principle of the system of the present invention: The data collection and processing module collects multi-source banking business data, including basic customer information, deposit size, transaction records, risk assessment indicators, loan usage characteristics, and imaged documents (contract scans), to form a local feature map F i ; The data evaluation module is based on the local feature map F i Establish a dynamic credit assessment model to calculate the customer's actual personal credit score i Then, the time required to repay each bank's credit line business is calculated based on the individual's credit score. i ; Then the deep learning model construction module builds the deep learning model and sets the time requirement T iThe data is input into the LSTM network to analyze the customer's repayment behavior time series data, generate a digital personal ability credit score, and predict the overdue risk of various banking services; finally, according to the intelligent decision-making module, it is used to calculate the comprehensive score of the customer's various credit limit businesses based on the customer's actual personal ability credit score, and output the business with the highest score as the recommended business; a "preference correction item" is introduced to dynamically adjust the risk coefficient: by introducing a preference correction coefficient to dynamically adjust risky businesses, high-risk product businesses will only be recommended to customers when the user's risk value score is greater than the preset threshold.

[0041] And through the automated error adjustment module, a feedback difference mechanism is established to compare the customer's actual personal ability credit score with the digital personal ability credit score predicted by the deep learning model, and a feedback difference mechanism is constructed to iteratively correct the deep learning model and time requirements; when the feedback error When the feedback error is greater than the preset threshold, it means that the deviation between the predicted state and the actual state is large, and the setting parameters of the dynamic credit evaluation model need to be adjusted to reduce the error and optimize the prediction effect; when the feedback error is greater than the preset threshold, it means that the deviation between the predicted state and the actual state is large, and the setting parameters of the dynamic credit evaluation model need to be adjusted to reduce the error and optimize the prediction effect. When it is less than the preset threshold, it means that the current control effect is close to the expectation and the system can maintain the existing output without further adjustment.

[0042] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of the systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, as well as combinations of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or actions, or can be implemented using a combination of dedicated hardware and computer instructions.

[0043] In addition, the functional modules in various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0044] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering technicians in this field should fall within the scope of protection determined by the claims of the present invention.

Claims

1. A digital banking management system based on deep learning algorithms, characterized by: include: A data collection and processing module is used to collect multi-source banking business data and construct a local feature map based on feature vectors after pre-processing the multi-source banking business data; The data evaluation module is used to establish a dynamic credit assessment model based on the local feature map, calculate the customer's actual personal credit score, and then calculate the repayment time required for various bank credit lines based on the personal credit score; A deep learning model building module, used to build a deep learning model and map time demand into a prediction of the customer's future overdue risk; The intelligent decision-making module is used to calculate the comprehensive score of each credit line business handled by the customer based on the customer's actual personal ability credit score, and output the business with the highest score as the recommended business; The automated error adjustment module is used to compare the customer's actual personal ability credit score with the digital personal ability credit score predicted by the deep learning model, and to build a feedback difference mechanism to iteratively correct the deep learning model and time requirements.

2. A banking business digital management system based on deep learning algorithm according to claim 1, characterized in that: In the data collection and processing module, the multi-source banking business data includes basic customer information, transaction records, risk assessment indicators, loan usage characteristics and imaged documents.

3. The banking business digital management system based on deep learning algorithm according to claim 1 is characterized in that: The data evaluation module establishes a dynamic credit evaluation model based on the local feature graph and predicts the personal credit score by comprehensively considering the customer's multi-source information, including basic information, transaction records, risk assessment indicators, and loan purposes. The specific calculation formula for the personal credit score is as follows: ， Among them, Score i represents the personal credit score of customer i, k represents the personal deviation coefficient, D i represents the average monthly income of customer i, D avg represents the local per capita income, H i represents the additional income of user i, represents the impact function of user i’s average monthly consumption on his personal credit score, where , is a constant used to adjust the impact of the user's average monthly consumption on the personal ability credit score; α represents the mutual influence coefficient between the average monthly income and the average monthly consumption; C i represents the average monthly consumption of user i.

4. A banking business digital management system based on deep learning algorithm according to claim 3, characterized in that: The personal credit score is converted into the time required to repay various bank credit lines, that is, the average monthly income and average monthly consumption combination required by user i to repay the required business. For this purpose, the following time demand prediction formula is proposed: ， Among them, T i represents the time requirement of region i; c1 is a constant, representing the personal ability credit score Score i The influence coefficient on time demand; c2 is a constant, which represents the influence coefficient of average monthly income and average monthly consumption on time demand; represents the comprehensive impact function of average monthly income and average monthly consumption on time demand, where , which is the weighted combination of average monthly income and average monthly consumption; Represents the influence coefficient of user deposit on time demand.

5. The banking business digital management system based on deep learning algorithm according to claim 1, characterized in that: A deep learning model is established through the deep learning model construction module, time requirements are input into the deep learning model, and the customer repayment behavior time series data is analyzed through the LSTM network to generate a digital personal ability credit score.

6. The banking business digital management system based on deep learning algorithm according to claim 1, characterized in that: The intelligent decision-making module calculates the customer's actual personal ability credit score based on the data evaluation module, calculates the comprehensive score of the customer's various credit limit businesses through a weighted fusion algorithm, and pushes the business with the highest score as the optimal business to the customer for processing.

7. The banking business digital management system based on deep learning algorithm according to claim 1 is characterized in that: The automated error adjustment module establishes a feedback difference mechanism to generate a feedback correction signal through the feedback difference mechanism; the feedback difference mechanism is used as a correction factor for the personal ability credit score output by the dynamic credit assessment model to adjust the prediction result of the time demand in the data assessment module.

8. The banking business digital management system based on deep learning algorithm according to claim 7 is characterized in that: When the feedback error is greater than the preset threshold, the setting parameters of the dynamic credit evaluation model are adjusted to reduce the error and optimize the prediction effect; when the feedback error is less than the preset threshold, there is no need to adjust the existing parameters of the dynamic credit evaluation model.

Images