Method, system and equipment for simulating customer approval result based on historical data learning

Abstract

The invention discloses a method, a system and equipment for simulating a customer approval result based on historical data learning. The method comprises the steps of user data collection and data preprocessing; constructing and training a model; model evaluation: performing prediction accuracy evaluation on the trained model by using an independent test data set, and if an evaluation result is not good, adjusting and optimizing model parameters until the evaluation result meets requirements; adjusting a marketing strategy according to a simulation result; applying the strategy to a marketing operation scene, and applying the adjusted marketing strategy to actual marketing operation; effect monitoring and iterative optimization are carried out, feedback and iterative optimization are carried out on the model, the marketing strategy or the whole process according to the monitoring result, and the prediction accuracy and the marketing effect are continuously improved. According to the method, intelligent simulation and prediction of a complex approval process are realized, the accuracy and reliability of an approval result are greatly improved, a more personalized marketing strategy and service scheme are formulated, and the marketing effect and customer satisfaction are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of data analysis, and relates to a method, system and device for simulating customer approval results based on historical data learning. Background Art

[0002] In the current financial, insurance, credit and other fields that require customer approval, the prediction of customer approval results is a key link in business operations. The traditional approval process mainly relies on manual experience judgment. Approval personnel make subjective judgments on customer approval results based on multi-dimensional information such as customer credit records, income status, historical borrowing behavior, etc., combined with their own professional knowledge and experience. However, this is not only time-consuming and laborious, but also easily affected by the personal experience and subjective judgment of the approvers, resulting in inconsistency and uncertainty in the approval results.

[0003] In order to overcome the limitations of manual approval, the industry has begun to try to use traditional statistical methods to predict customer approval results. Usually based on a large amount of historical data, mathematical models are built to mine the laws and patterns in the data to predict customer approval results. However, traditional statistical methods often have problems such as complex model construction, large amount of calculation, and limited prediction accuracy, which makes it difficult to meet the high requirements of modern business for approval efficiency and accuracy.

[0004] With the rapid development of big data and machine learning technologies, more and more fields are trying to use these advanced technologies to optimize business processes and improve decision-making efficiency. In the field of customer approval, machine learning algorithms are used to mine and analyze historical data to identify key factors that affect approval results and build prediction models based on them. However, there are still some shortcomings, such as limited learning ability of the model, insufficient processing ability for real-time data, and poor interpretability and ease of use of the model.

[0005] Therefore, a model is needed that can autonomously learn and simulate customer approval results in real time, make full use of online historical data, automatically mine the rules and patterns in the data through machine learning algorithms, accurately predict customer approval results, and effectively apply it to marketing operations and other scenarios. Summary of the invention

[0006] The purpose of the present invention is to solve the problem that the simulation model of customer approval results in the prior art has insufficient processing capability for real-time data and limited prediction accuracy, and to provide a method, system and device for simulating customer approval results based on historical data learning.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] The method for simulating customer approval results based on historical data learning includes the following steps:

[0009] User data collection, collecting multi-dimensional user data including basic customer information, historical transaction records, and credit status;

[0010] Data preprocessing: clean the collected user data to remove duplicate, erroneous or invalid data; transform the data and extract features with predictive value for approval results from the data;

[0011] Model construction and training: using preprocessed data and extracted features, build and train a model to predict customer approval results;

[0012] Model evaluation: Use an independent test data set to evaluate the prediction accuracy of the trained model. If the evaluation result meets the preset prediction accuracy standard, proceed to the next step. If the evaluation result is not good, adjust and optimize the model parameters, and use the optimized model to re-simulate the prediction until the evaluation result meets the requirements.

[0013] Adjust marketing strategies based on simulation results. According to the customer approval results predicted by the model, combined with business needs and marketing goals, formulate or adjust corresponding marketing strategies;

[0014] Apply strategies to marketing operation scenarios and apply adjusted marketing strategies to actual marketing operations to guide customer approval and marketing activities;

[0015] Effect monitoring and iterative optimization: monitor the actual effect after the strategy is executed; based on the monitoring results, provide feedback and iterative optimization of the model, marketing strategy or the entire process to continuously improve prediction accuracy and marketing effectiveness.

[0016] The model for predicting customer approval results specifically includes:

[0017] The data input layer is used to receive data on basic customer information, transaction behavior, and credit records, and to pre-process the data, where transaction behavior and credit record data are arranged and coded in time series;

[0018] Feature extraction layer, which uses convolutional neural networks to extract local features of input data, uses recurrent neural networks to process time series data to capture changing trends, and combines decision trees and random forest algorithms to extract key features and patterns from raw data;

[0019] The fusion layer combines the features from convolutional neural networks, recurrent neural networks, decision trees, and random forests, and uses the attention mechanism to assign weights to the features;

[0020] The prediction layer uses a multi-layer perceptron to process the fused features and outputs the probability distribution of the customer approval results. The model performance is optimized by adjusting the number of layers and neurons of the multi-layer perceptron.

[0021] The specific extraction steps of the feature extraction layer are:

[0022] Using convolutional neural networks, the input data is treated as a two-dimensional image when being visualized, and local features in the data are gradually extracted through convolutional layers and pooling layers;

[0023] Using the long short-term memory network or gated recurrent unit in the recurrent neural network, the customer credit record data is input in chronological order to capture the trend of credit record changes over time;

[0024] Combining decision tree and random forest algorithms, the input data is segmented and combined to extract key decision rules and feature combinations.

[0025] The fusion layer specifically includes:

[0026] Feature fusion unit, used to merge the features extracted from convolutional neural networks, recurrent neural networks, decision trees, and random forests by concatenation or weighted summation;

[0027] An attention mechanism processing unit, used to process the fused features and calculate the weight of each feature through an attention mechanism, wherein the attention mechanism adopts a query, key, and value-based attention mechanism framework, wherein the fused features are used as values ​​and interact with each feature through a learnable query vector, thereby calculating the weight of each feature;

[0028] The feature weighting and output unit performs weighted processing on the fused features according to the weights calculated by the attention mechanism processing unit, and uses the weighted features as the input of the subsequent prediction layer.

[0029] The prediction layer adopts a multi-layer perceptron architecture, specifically including:

[0030] An input unit, receiving the fused multi-dimensional customer feature data;

[0031] Multiple hidden units, each consisting of a preset number of fully connected neurons, are used to process input data and extract high-level features;

[0032] The output unit outputs the probability value of whether the customer approval result is approved or rejected through the Sigmoid activation function.

[0033] The training and prediction model for customer approval results specifically includes the following steps:

[0034] Define the loss function and use the cross entropy loss function to measure the difference between the model prediction results and the actual approval results. Specifically:

[0035]

[0036] Where N is the number of samples, y i The actual approval result, represented by 0 or 1. The probability of approval result predicted by the model;

[0037] Set the parameter tuning objective function, and use the weighted sum of the model's accuracy and recall on the validation set as the parameter tuning objective function, specifically:

[0038] F=ω×Precision+(1-ω)×Recall (2)

[0039] Where ω is the weight coefficient, Precision is the accuracy of the model on the validation set, and Recall is the recall rate of the model on the validation set;

[0040] Data partitioning: divide the preprocessed data set into training set, validation set and test set, which are used for model training, parameter adjustment and performance evaluation respectively;

[0041] Model initialization, randomly initialize the model parameters, including the weights and biases of convolutional neural networks, recurrent neural networks, decision trees, random forests, and multilayer perceptrons;

[0042] Iterative training: Use the training set to iteratively train the model. Each iteration updates the model parameters to gradually reduce the loss function. Use the small batch stochastic gradient descent algorithm for optimization. Divide the training set into multiple small batches of data, calculate the prediction results and loss function of each small batch of data, and derive the model parameters according to the loss function to obtain the gradient. Then use the gradient descent algorithm to update the model parameters.

[0043] Model evaluation: After each iteration, the model is evaluated using the validation set to calculate the accuracy, recall, and objective function value, and the model parameters are adjusted based on the evaluation results. The accuracy calculation formula is:

[0044]

[0045] Among them, TP represents true positive samples, that is, the number of samples that are actually positive and predicted as positive by the model; FP represents false positive samples, that is, the number of samples that are actually negative but predicted as positive by the model;

[0046] The calculation of recall is:

[0047]

[0048] Among them, FN represents false negative examples, that is, the number of samples that are actually positive but predicted as negative samples by the model;

[0049] Model selection: When the performance of the model on the validation set no longer improves, training is stopped and the model with the best performance on the validation set is selected as the final model.

[0050] Model testing: Use the test set to test the final model and evaluate the generalization ability and performance of the model, including accuracy, recall, F1 value indicators, as well as customer satisfaction and marketing effects in actual applications. The F1 value calculation formula is:

[0051]

[0052] The F1 value is the harmonic mean of precision and recall.

[0053] The iterative optimization of the model is specifically as follows:

[0054] Difference calculation, calculate the difference between the actual approval result and the model prediction result, and use the loss function to measure the difference. The loss function uses the mean square error:

[0055]

[0056] Where n is the number of samples;

[0057] Parameter adjustment, based on the result of the difference calculation, uses the stochastic gradient descent algorithm to adjust the parameters of the model to minimize the loss function. The update formula of the stochastic gradient descent algorithm is:

[0058]

[0059] Among them, θ is the model parameter and α is the learning rate;

[0060] Model update: Apply the adjusted parameters to the model, update the model structure and weights, and make the model more accurately predict customer approval results;

[0061] The real-time update mechanism monitors the changes in the approval logic in real time. When changes are detected, the data source is updated in time, and the updated data is used to perform incremental learning on the model. Based on the original model, the new data is trained. The update formula of the incremental learning is:

[0062]

[0063] Among them, θ0 is the original model parameter, D new is the new data, and L is the loss function.

[0064] The system for simulating customer approval results based on historical data learning includes the following modules:

[0065] User data collection module, used to collect multi-dimensional user data including basic customer information, historical transaction records, and credit status;

[0066] The data preprocessing module cleans the collected user data, removes duplicate, erroneous or invalid data, converts the data, and extracts features from the data that have predictive value for approval results;

[0067] Model building and training module, which uses preprocessed data and extracted features to build and train a model to predict customer approval results;

[0068] The model evaluation module uses an independent test data set to evaluate the prediction accuracy of the trained model; if the evaluation result meets the preset prediction accuracy standard, the next step is executed; if the evaluation result does not meet the standard, the model parameters are adjusted and optimized, and the optimized model is used for retraining and evaluation until the evaluation result meets the requirements;

[0069] The marketing strategy adjustment module formulates or adjusts corresponding marketing strategies based on the customer approval results predicted by the model, combined with business needs and marketing goals;

[0070] The marketing operation application module applies the adjusted marketing strategy to actual marketing operations to guide customer approval and marketing activities;

[0071] The effect monitoring and iterative optimization module monitors the actual effect after the implementation of the marketing strategy, and provides feedback and iterative optimization to the model, marketing strategy or the entire process based on the monitoring results, so as to continuously improve the prediction accuracy and marketing effect.

[0072] A device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method as described in any one of the preceding items when executing the computer program.

[0073] A computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of the method described in any one of the preceding items.

[0074] Compared with the prior art, the present invention has the following beneficial effects:

[0075] The present invention adopts a method of simulating customer approval results based on historical data learning, which can comprehensively and accurately capture and extract key features and patterns in the data, including but not limited to important dimensions such as basic customer information, transaction behavior, and credit records. Through advanced machine learning technology, this method realizes intelligent simulation and prediction of complex approval processes, greatly improving the accuracy and reliability of approval results.

[0076] In the process of autonomous simulation and optimization, it can continuously adjust and improve its own parameters and structure to adapt to changes in different data sets and task requirements. At the same time, using the feedback mechanism, it can make corrections and improvements based on the differences between the actual approval results and the predicted results, ensuring that the prediction model is always highly consistent with the actual approval logic.

[0077] In addition, the method of the present invention has strong real-time updating and adaptability, which can keep up with the changes in approval standards and business needs, and ensure the timeliness and accuracy of the prediction model. Based on the simulation results, more personalized marketing strategies and service plans can be formulated, and relevant products and preferential information can be accurately pushed, thereby significantly improving marketing effects and customer satisfaction.

[0078] This method also supports real-time monitoring and analysis of approval results, which can timely identify problems and risks in the operation process, and adjust the operation strategy and optimize the business process accordingly. This not only helps to improve work efficiency and reduce resource waste, but also provides strong data support and decision-making basis, helping users maintain their leading position and achieve sustainable development in the fierce market competition. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0080] Figure 1 This is a flow chart of the method for simulating customer approval results based on historical data learning of the present invention. DETAILED DESCRIPTION

[0081] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0082] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0083] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0084] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0085] See also Figure 1 , is a flow chart of a method for simulating customer approval results based on historical data learning in the present invention, comprising the following steps:

[0086] S1, user data collection, collects multi-dimensional user data including basic customer information, historical transaction records, and credit status.

[0087] S2, data preprocessing, cleans the collected user data to remove duplicate, erroneous or invalid data; transforms the data and extracts features from the data that have predictive value for the approval results.

[0088] S3, model construction and training, uses the preprocessed data and extracted features to build and train a model to predict customer approval results.

[0089] The model for predicting customer approval results specifically includes:

[0090] The data input layer is used to receive data on customer basic information, transaction behavior and credit records, and pre-process the data, where the transaction behavior and credit record data are arranged and encoded in time series.

[0091] Feature extraction layer, which uses convolutional neural networks to extract local features of input data, uses recurrent neural networks to process time series data to capture changing trends, and combines decision trees and random forest algorithms to extract key features and patterns from raw data;

[0092] The specific extraction steps of the feature extraction layer are:

[0093] Using convolutional neural networks, the input data is treated as a two-dimensional image when being visualized, and local features in the data are gradually extracted through convolutional layers and pooling layers;

[0094] Using the long short-term memory network or gated recurrent unit in the recurrent neural network, the customer credit record data is input in chronological order to capture the trend of credit record changes over time;

[0095] Combining decision tree and random forest algorithms, the input data is segmented and combined to extract key decision rules and feature combinations.

[0096] The fusion layer combines the features from convolutional neural networks, recurrent neural networks, decision trees, and random forests, and uses the attention mechanism to assign weights to the features.

[0097] Fusion layer, including:

[0098] Feature fusion unit, used to merge the features extracted from convolutional neural networks, recurrent neural networks, decision trees, and random forests by concatenation or weighted summation;

[0099] An attention mechanism processing unit, used to process the fused features and calculate the weight of each feature through an attention mechanism, wherein the attention mechanism adopts a query, key, and value-based attention mechanism framework, wherein the fused features are used as values ​​and interact with each feature through a learnable query vector, thereby calculating the weight of each feature;

[0100] The feature weighting and output unit performs weighted processing on the fused features according to the weights calculated by the attention mechanism processing unit, and uses the weighted features as the input of the subsequent prediction layer.

[0101] The prediction layer uses a multi-layer perceptron to process the fused features and outputs the probability distribution of the customer approval results. The model performance is optimized by adjusting the number of layers and neurons of the multi-layer perceptron.

[0102] The prediction layer adopts a multi-layer perceptron architecture, which includes:

[0103] An input unit, receiving the fused multi-dimensional customer feature data;

[0104] Multiple hidden units, each consisting of a preset number of fully connected neurons, are used to process input data and extract high-level features;

[0105] The output unit outputs the probability value of whether the customer approval result is approved or rejected through the Sigmoid activation function.

[0106] Training a model to predict customer approval results includes the following steps:

[0107] Define the loss function and use the cross entropy loss function to measure the difference between the model prediction results and the actual approval results. Specifically:

[0108]

[0109] Where N is the number of samples, y i The actual approval result, represented by 0 or 1. The probability of approval result predicted by the model;

[0110] Set the parameter tuning objective function, and use the weighted sum of the model's accuracy and recall on the validation set as the parameter tuning objective function, specifically:

[0111] F=ω×Precision+(1-ω)×Recall (2)

[0112] Where ω is the weight coefficient, Precision is the accuracy of the model on the validation set, and Recall is the recall rate of the model on the validation set;

[0113] Data partitioning: divide the preprocessed data set into training set, validation set and test set, which are used for model training, parameter adjustment and performance evaluation respectively;

[0114] Model initialization, randomly initialize the model parameters, including the weights and biases of convolutional neural networks, recurrent neural networks, decision trees, random forests, and multilayer perceptrons;

[0115] Iterative training: Use the training set to iteratively train the model. Each iteration updates the model parameters to gradually reduce the loss function. Use the small batch stochastic gradient descent algorithm for optimization. Divide the training set into multiple small batches of data, calculate the prediction results and loss function of each small batch of data, and derive the model parameters according to the loss function to obtain the gradient. Then use the gradient descent algorithm to update the model parameters.

[0116] Model evaluation: After each iteration, the model is evaluated using the validation set to calculate the accuracy, recall, and objective function value, and the model parameters are adjusted based on the evaluation results. The accuracy calculation formula is:

[0117]

[0118] Among them, TP represents true positive samples, that is, the number of samples that are actually positive and predicted as positive by the model; FP represents false positive samples, that is, the number of samples that are actually negative but predicted as positive by the model;

[0119] The calculation of recall is:

[0120]

[0121] Among them, FN represents false negative examples, that is, the number of samples that are actually positive but predicted as negative samples by the model;

[0122] Model selection: When the performance of the model on the validation set no longer improves, training is stopped and the model with the best performance on the validation set is selected as the final model.

[0123] Model testing: Use the test set to test the final model and evaluate the generalization ability and performance of the model, including accuracy, recall, F1 value indicators, as well as customer satisfaction and marketing effects in actual applications. The F1 value calculation formula is:

[0124]

[0125] The F1 value is the harmonic mean of precision and recall.

[0126] S4, model evaluation, uses an independent test data set to evaluate the prediction accuracy of the trained model; if the evaluation result meets the preset prediction accuracy standard, proceed to the next step; if the evaluation result is not good, adjust and optimize the model parameters, and use the optimized model to re-simulate the prediction until the evaluation result meets the requirements.

[0127] S5, adjust the marketing strategy according to the simulation results, formulate or adjust the corresponding marketing strategy based on the customer approval results predicted by the model, combined with business needs and marketing goals.

[0128] S6, apply strategies to marketing operation scenarios and apply the adjusted marketing strategies to actual marketing operations to guide customer approval and marketing activities.

[0129] S7, effect monitoring and iterative optimization, monitor the actual effect after the strategy is implemented; based on the monitoring results, provide feedback and iterative optimization of the model, marketing strategy or the entire process to continuously improve prediction accuracy and marketing effectiveness.

[0130] The iterative optimization of the model is specifically as follows:

[0131] Difference calculation, calculate the difference between the actual approval result and the model prediction result, and use the loss function to measure the difference. The loss function uses the mean square error:

[0132]

[0133] Where n is the number of samples;

[0134] Parameter adjustment, based on the result of the difference calculation, uses the stochastic gradient descent algorithm to adjust the parameters of the model to minimize the loss function. The update formula of the stochastic gradient descent algorithm is:

[0135]

[0136] Among them, θ is the model parameter and α is the learning rate;

[0137] Model update: Apply the adjusted parameters to the model, update the model structure and weights, and make the model more accurately predict customer approval results;

[0138] The real-time update mechanism monitors the changes in the approval logic in real time. When changes are detected, the data source is updated in time, and the updated data is used to perform incremental learning on the model. Based on the original model, the new data is trained. The update formula of the incremental learning is:

[0139]

[0140] Among them, θ0 is the original model parameter, D new is the new data, and L is the loss function.

[0141] The reason why the model structure in the present invention can effectively and accurately simulate the approval results is that it combines the advantages of multiple machine learning technologies. Deep learning technology can automatically mine complex features in the data, while decision trees and random forests provide good interpretability and robustness. Through integrated learning strategies, the generalization ability and prediction accuracy of the model are significantly enhanced. In particular, the feature extraction layer uses advanced deep learning technologies such as convolutional neural networks CNN and recurrent neural networks RNN to accurately capture local features and time series patterns in the data, which is crucial for analyzing the time dependencies and local patterns of customer transaction behaviors and credit records. In addition, the attention mechanism introduced by the fusion layer can be dynamically adjusted according to the importance of the features, focusing on key features, effectively weakening the impact of noise and irrelevant features, thereby further improving the accuracy of the prediction.

[0142] In application scenarios such as marketing operation promotion, the use of simulated data of customer approval prediction results provides strong support and basis for decision-making. By deeply analyzing the customer's basic information, transaction behavior, credit record and other data, it is possible to accurately grasp the customer's characteristics and needs, and then divide the customers into groups with similar characteristics and needs. On this basis, personalized product matching and push strategies are implemented: for customers with high approval rates, higher credit limits and more optimized promotion plans are provided; for customers with low approval rates, unnecessary promotion times are reduced to save costs, and the most appropriate products and preferential information are recommended to customers, thereby achieving precision marketing and efficient operations.

[0143] The present invention adopts a combination of computer automatic evaluation and personnel subjective evaluation to comprehensively monitor and evaluate the performance and effect of the model.

[0144] In terms of computer automatic evaluation, key indicators such as accuracy, recall, and F1 value were selected to measure the difference between the model prediction results and the actual results. To ensure the objectivity and accuracy of the evaluation results, the data set was scientifically divided into training set, validation set, and test set, and the test set was used to predict and evaluate the model. During the evaluation process, various performance indicators of the model were calculated and compared with the preset thresholds to determine whether the performance of the model met the standards. If the evaluation results did not meet expectations, they would be optimized by adjusting model parameters, increasing the amount of data, and other methods.

[0145] In terms of subjective evaluation by personnel, intuitive indicators such as customer satisfaction and marketing effectiveness were selected to reflect the effectiveness and value of the model in practical applications. Through questionnaire surveys, customer feedback and other channels, we widely collected customer evaluations on the satisfaction and marketing effectiveness of personalized push notifications. Subsequently, we conducted detailed statistics and analysis on the customer evaluation results to comprehensively evaluate the performance and effectiveness of the model. At the same time, we also established a feedback mechanism to timely adjust the strategy and content of personalized push notifications based on the results of subjective evaluation by personnel, so as to continuously improve customer satisfaction and marketing effectiveness.

[0146] An embodiment of the present invention is a system for simulating customer approval results based on historical data learning, which specifically includes the following modules:

[0147] The user data collection module is used to collect multi-dimensional user data including basic customer information, historical transaction records, and credit status.

[0148] The data preprocessing module cleans the collected user data, removes duplicate, erroneous or invalid data, converts the data, and extracts features from the data that have predictive value for approval results.

[0149] The model building and training module uses the preprocessed data and extracted features to build and train a model for predicting customer approval results.

[0150] The model evaluation module uses an independent test data set to evaluate the prediction accuracy of the trained model; if the evaluation result meets the preset prediction accuracy standard, the next step is executed; if the evaluation result does not meet the standard, the model parameters are adjusted and optimized, and the optimized model is used for retraining and evaluation until the evaluation result meets the requirements.

[0151] The marketing strategy adjustment module formulates or adjusts corresponding marketing strategies based on the customer approval results predicted by the model, combined with business needs and marketing goals.

[0152] The marketing operation application module applies the adjusted marketing strategies to actual marketing operations to guide customer approval and marketing activities.

[0153] The effect monitoring and iterative optimization module monitors the actual effect after the implementation of the marketing strategy, and provides feedback and iterative optimization to the model, marketing strategy or the entire process based on the monitoring results, so as to continuously improve the prediction accuracy and marketing effect.

[0154] An embodiment of the present invention provides a terminal device. The terminal device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned method embodiments are implemented. Alternatively, when the processor executes the computer program, the functions of the modules / units in the above-mentioned device embodiments are implemented.

[0155] The computer program may be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to accomplish the present invention.

[0156] The device / terminal equipment may be a computing device such as a desktop computer, a notebook, a palm computer, a cloud server, etc. The device / terminal equipment may include, but is not limited to, a processor and a memory.

[0157] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0158] The memory may be used to store the computer programs and / or modules, and the processor implements various functions of the apparatus / terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory.

[0159] If the module / unit integrated in the device / terminal equipment is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0160] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for simulating customer approval results based on historical data learning, characterized in that: The following steps are involved: User data collection, collecting multi-dimensional user data including basic customer information, historical transaction records, and credit status; Data preprocessing: cleaning the collected user data to remove duplicate, erroneous or invalid data; Transform the data and extract features from the data that have predictive value for approval results; Model construction and training: using preprocessed data and extracted features, build and train a model to predict customer approval results; Model evaluation: Use an independent test data set to evaluate the prediction accuracy of the trained model. If the evaluation result meets the preset prediction accuracy standard, proceed to the next step. If the evaluation result is not good, adjust and optimize the model parameters, and use the optimized model to re-simulate the prediction until the evaluation result meets the requirements. Adjust marketing strategies based on simulation results. According to the customer approval results predicted by the model, combined with business needs and marketing goals, formulate or adjust corresponding marketing strategies; Apply strategies to marketing operation scenarios and apply adjusted marketing strategies to actual marketing operations to guide customer approval and marketing activities; Effect monitoring and iterative optimization: monitor the actual effect after the strategy is executed; based on the monitoring results, provide feedback and iterative optimization of the model, marketing strategy or the entire process to continuously improve prediction accuracy and marketing effectiveness.

2. The method for simulating customer approval results based on historical data learning as claimed in claim 1, characterized in that: The model for predicting customer approval results specifically includes: The data input layer is used to receive data on basic customer information, transaction behavior, and credit records, and to pre-process the data, where transaction behavior and credit record data are arranged and coded in time series; Feature extraction layer, which uses convolutional neural networks to extract local features of input data, uses recurrent neural networks to process time series data to capture changing trends, and combines decision trees and random forest algorithms to extract key features and patterns from raw data; The fusion layer combines the features from convolutional neural networks, recurrent neural networks, decision trees, and random forests, and uses the attention mechanism to assign weights to the features; The prediction layer uses a multi-layer perceptron to process the fused features and outputs the probability distribution of the customer approval results. The model performance is optimized by adjusting the number of layers and neurons of the multi-layer perceptron.

3. The method for simulating customer approval results based on historical data learning as claimed in claim 2, characterized in that: The specific extraction steps of the feature extraction layer are: Using convolutional neural networks, the input data is treated as a two-dimensional image when being visualized, and local features in the data are gradually extracted through convolutional layers and pooling layers; Using the long short-term memory network or gated recurrent unit in the recurrent neural network, the customer credit record data is input in chronological order to capture the trend of credit record changes over time; Combining decision tree and random forest algorithms, the input data is segmented and combined to extract key decision rules and feature combinations.

4. The method for simulating customer approval results based on historical data learning as claimed in claim 2, characterized in that: The fusion layer specifically includes: Feature fusion unit, used to merge the features extracted from convolutional neural networks, recurrent neural networks, decision trees, and random forests by concatenation or weighted summation; An attention mechanism processing unit, used to process the fused features and calculate the weight of each feature through an attention mechanism, wherein the attention mechanism adopts a query, key, and value-based attention mechanism framework, wherein the fused features are used as values ​​and interact with each feature through a learnable query vector, thereby calculating the weight of each feature; The feature weighting and output unit performs weighted processing on the fused features according to the weights calculated by the attention mechanism processing unit, and uses the weighted features as the input of the subsequent prediction layer.

5. The method for simulating customer approval results based on historical data learning as claimed in claim 2, characterized in that: The prediction layer adopts a multi-layer perceptron architecture, specifically including: An input unit, receiving the fused multi-dimensional customer feature data; Multiple hidden units, each consisting of a preset number of fully connected neurons, are used to process input data and extract high-level features; The output unit outputs the probability value of whether the customer approval result is approved or rejected through the Sigmoid activation function.

6. The method for simulating customer approval results based on historical data learning as claimed in claim 1, characterized in that: The training and prediction model for customer approval results specifically includes the following steps: Define the loss function and use the cross entropy loss function to measure the difference between the model prediction results and the actual approval results. Specifically: Where N is the number of samples, y i The actual approval result, represented by 0 or 1. The probability of approval result predicted by the model; Set the parameter tuning objective function, and use the weighted sum of the model's accuracy and recall on the validation set as the parameter tuning objective function, specifically: F=ω×Precision+(1-ω)×Recall (2) Where ω is the weight coefficient, Precision is the accuracy of the model on the validation set, and Recall is the recall rate of the model on the validation set; Data partitioning: divide the preprocessed data set into training set, validation set and test set, which are used for model training, parameter adjustment and performance evaluation respectively; Model initialization, randomly initialize the model parameters, including the weights and biases of convolutional neural networks, recurrent neural networks, decision trees, random forests, and multilayer perceptrons; Iterative training: Use the training set to iteratively train the model. Each iteration updates the model parameters to gradually reduce the loss function. Use the small batch stochastic gradient descent algorithm for optimization. Divide the training set into multiple small batches of data, calculate the prediction results and loss function of each small batch of data, and derive the model parameters according to the loss function to obtain the gradient. Then use the gradient descent algorithm to update the model parameters. Model evaluation: After each iteration, the model is evaluated using the validation set to calculate the accuracy, recall, and objective function value, and the model parameters are adjusted based on the evaluation results. The accuracy calculation formula is: Among them, TP represents true positive samples, that is, the number of samples that are actually positive and predicted as positive by the model; FP represents false positive samples, that is, the number of samples that are actually negative but predicted as positive by the model; The calculation of recall is: Among them, FN represents false negative examples, that is, the number of samples that are actually positive but predicted as negative samples by the model; Model selection: When the performance of the model on the validation set no longer improves, training is stopped and the model with the best performance on the validation set is selected as the final model. Model testing: Use the test set to test the final model and evaluate the generalization ability and performance of the model, including accuracy, recall, F1 value indicators, as well as customer satisfaction and marketing effects in actual applications. The F1 value calculation formula is: The F1 value is the harmonic mean of precision and recall.

7. The method for simulating customer approval results based on historical data learning as claimed in claim 1, characterized in that: The iterative optimization of the model is specifically as follows: Difference calculation, calculate the difference between the actual approval result and the model prediction result, and use the loss function to measure the difference. The loss function uses the mean square error: Where n is the number of samples; Parameter adjustment, based on the result of the difference calculation, uses the stochastic gradient descent algorithm to adjust the parameters of the model to minimize the loss function. The update formula of the stochastic gradient descent algorithm is: Among them, θ is the model parameter and α is the learning rate; Model update: Apply the adjusted parameters to the model, update the model structure and weights, and make the model more accurately predict customer approval results; The real-time update mechanism monitors the changes in the approval logic in real time. When changes are detected, the data source is updated in time, and the updated data is used to perform incremental learning on the model. Based on the original model, the new data is trained. The update formula of the incremental learning is: Among them, θ0 is the original model parameter, D new is the new data, and L is the loss function.

8. A system for simulating customer approval results based on historical data learning, characterized in that: Includes the following modules: User data collection module, used to collect multi-dimensional user data including basic customer information, historical transaction records, and credit status; The data preprocessing module cleans the collected user data, removes duplicate, erroneous or invalid data, converts the data, and extracts features from the data that have predictive value for approval results; Model building and training module, which uses preprocessed data and extracted features to build and train a model to predict customer approval results; The model evaluation module uses an independent test data set to evaluate the prediction accuracy of the trained model; if the evaluation result meets the preset prediction accuracy standard, the next step is executed; if the evaluation result does not meet the standard, the model parameters are adjusted and optimized, and the optimized model is used for retraining and evaluation until the evaluation result meets the requirements; The marketing strategy adjustment module formulates or adjusts corresponding marketing strategies based on the customer approval results predicted by the model, combined with business needs and marketing goals; The marketing operation application module applies the adjusted marketing strategy to actual marketing operations to guide customer approval and marketing activities; The effect monitoring and iterative optimization module monitors the actual effect after the implementation of the marketing strategy, and provides feedback and iterative optimization to the model, marketing strategy or the entire process based on the monitoring results, so as to continuously improve the prediction accuracy and marketing effect.

9. A device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images