Policy execution method and business promotion system for potential service object

Abstract

The invention relates to the technical field of computers, in particular to a strategy execution method and a business promotion system for potential service objects, and aims to solve the problems of low accuracy and insufficient efficiency during processing of large-scale and high-dimensional data in the prior art. After a prediction request for a target object is received, inherent feature information and time sequence feature information of the target object are collected, the inherent feature information and the time sequence feature information are input into the two-channel deep learning model through the model layer, and the prediction probability and decision feature information that the target object becomes a potential service object are output. And if the prediction probability is greater than or equal to a preset threshold value, inputting the decision feature information into a strategy matching model, outputting a target service strategy, and calling an execution interface to execute the target service strategy so as to trigger a business promotion operation on the target object. According to the scheme, accurate identification of potential service objects and automatic execution of personalized strategies are realized, and decision accuracy and processing efficiency in a large-scale data scene are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a strategy execution method for potential service objects and a business promotion system. BACKGROUND

[0002] In current various business scenarios, there is a core demand: accurately identifying potential service objects with high value from a large number of candidate objects, and formulating personalized service strategies for each service object. The existing object screening and intelligent decision-making technology usually analyzes the attribute data of a single dimension of the candidate object. Specifically, the state information of the candidate object at a certain time is usually collected, and a shallow neural network or a traditional machine learning model is used to complete the classification identification of the object and the matching selection of the strategy.

[0003] However, the model used in the prior art has limitations in structure and capacity, resulting in low decision accuracy and low processing efficiency when dealing with high-dimensional and large-scale data. SUMMARY

[0004] The present application provides a strategy execution method for potential service objects and a business promotion system to solve the problem of low accuracy and insufficient efficiency in processing large-scale and high-dimensional data in the prior art.

[0005] In a first aspect, the present application provides a strategy execution method for potential service objects, applied to a business promotion system of an institution, the business promotion system comprising a service layer, a data layer and a model layer; comprising: Through the service layer, a prediction request for a target object is received; Through the data layer, in response to the prediction request, inherent characteristic information and time sequence characteristic information of the target object are collected, the inherent characteristic information representing the basic attributes of the target object, and the time sequence characteristic information including index data reflecting the behavior change of the target object in multiple consecutive time periods; Through the model layer, the inherent characteristic information and the time sequence characteristic information are input into a dual-channel deep learning model to generate a prediction probability that the target object becomes a potential service object and decision characteristic information of the target object, wherein the dual-channel deep learning model comprises a first channel and a second channel, the first channel being used for processing the inherent characteristic information, and the second channel being used for processing the time sequence characteristic information; if the prediction probability is greater than or equal to a preset threshold, the decision characteristic information is input into a strategy matching model, and a target service strategy of the target object is output; Through the service layer, a preconfigured execution interface is called to execute the target service strategy to trigger a business promotion operation on the target object.

[0006] The strategy execution method for potential service targets provided in this application is applied to an organization's business promotion system. The business promotion system includes a service layer, a data layer, and a model layer. First, the service layer receives prediction requests for target targets. Second, the data layer responds to the prediction requests, collecting the inherent and temporal feature information of the target targets. Then, the model layer inputs the inherent and temporal feature information into a dual-channel deep learning model to generate a predicted probability that the target target will become a potential service target and decision feature information for the target target. The dual-channel deep learning model includes a first channel and a second channel. The first channel processes inherent feature information, and the second channel processes temporal feature information. If the predicted probability is greater than or equal to a preset threshold, the decision feature information is input into a strategy matching model, outputting the target service strategy for the target target. Finally, the service layer calls a pre-configured execution interface to execute the target service strategy, triggering business promotion operations for the target target. This solution, by using the first channel of the dual-channel deep learning model to process inherent feature information and the second channel to process temporal feature information, achieves efficient parallel processing of large-scale, high-dimensional data, improving the system's processing capacity and computational efficiency. Furthermore, by employing a dual-channel deep learning model and a policy matching model, the system achieves accurate identification of potential service targets and automated execution of personalized policies, thereby improving decision-making accuracy and processing efficiency in large-scale data scenarios.

[0007] In one possible implementation, inputting the inherent feature information and the temporal feature information into a dual-channel deep learning model to generate a predicted probability that the target object becomes a potential service object includes: The inherent feature information is input into the first channel, and a nonlinear transformation is performed through the deep feedforward neural network in the first channel to obtain the first feature vector; The temporal feature information is input into the second channel and encoded by the temporal neural network in the second channel to obtain the second feature vector; The first feature vector and the second feature vector are input into the fusion layer of the dual-channel deep learning model to generate a first prediction probability, and the first prediction probability is used as the prediction probability that the target object becomes a potential service object.

[0008] In one possible implementation, after generating the first predicted probability, the method further includes: Based on the fused feature vector, at least one second prediction probability is output using at least one auxiliary prediction model. The first predicted probability and the at least one second predicted probability are fused together to generate a predicted probability that the target object becomes a potential service object.

[0009] In one possible implementation, the step of inputting the first feature vector and the second feature vector into the fusion layer of the dual-channel deep learning model to generate a first prediction probability includes: The correlation weight between the first feature vector and the second feature vector is calculated through the fusion layer of the dual-channel deep learning model. Based on the correlation weights, the first feature vector and the second feature vector are weighted and fused to obtain a fused feature vector. The fused feature vector is input into the output layer of the dual-channel deep learning model and subjected to fully connected linear computation and logistic function transformation to generate the first prediction probability.

[0010] In one possible implementation, inputting the decision feature information into the policy matching model and outputting the target service policy of the target object includes: Using the strategy matching model, the matching score between the decision feature information and multiple candidate service strategies is calculated respectively. The candidate service strategy with the highest matching score is selected as the target service strategy.

[0011] In one possible implementation, the business promotion system further includes a monitoring layer, and the method further includes: The monitoring layer collects and monitors the operating status of the business promotion system, model prediction results, and feedback information on the execution of the target service strategy, and generates a monitoring analysis report. Based on the monitoring and analysis report, the dual-channel deep learning model is updated.

[0012] In one possible implementation, the method further includes at least one of the following steps: The inherent feature information and the temporal feature information are encrypted through the data layer; Through the model layer, noise injection processing is performed on the training dataset during the training of the dual-channel deep learning model; Access control mechanisms are implemented through the service layer and the model layer to restrict access to feature information, model interfaces and prediction results. The feature information includes at least one of the following: inherent feature information, time-series feature information and decision feature information.

[0013] In one possible implementation, the step of inputting the inherent feature information into the first channel and performing a nonlinear transformation through a deep feedforward neural network in the first channel to obtain a first feature vector includes: The inherent feature information is input into the input layer of the deep feedforward neural network; The intrinsic feature information is transformed nonlinearly layer by layer through at least three sequentially connected hidden layers, wherein each hidden layer uses a linear rectified function as the activation function. The output information after the transformation of the last hidden layer is used as the first feature vector.

[0014] In one possible implementation, the step of inputting the temporal feature information into the second channel and encoding it through a temporal neural network in the second channel to obtain a second feature vector includes: The temporal feature information is input into the temporal neural network according to a preset sequence data format, so that the temporal neural network encodes the temporal relationship in the temporal feature information; The second feature vector is generated based on the encoded temporal feature information.

[0015] Secondly, this application provides a business promotion system, including: The service layer is used to receive prediction requests for the target object; The data layer is used to respond to the prediction request and collect the inherent feature information and temporal feature information of the target object. The inherent feature information represents the basic attributes of the target object, and the temporal feature information includes indicator data that reflects the changes in the behavior of the target object over multiple consecutive time periods. The model layer is used to input the inherent feature information and the temporal feature information into a dual-channel deep learning model to generate the predicted probability that the target object becomes a potential service object and the decision feature information of the target object. The dual-channel deep learning model includes a first channel and a second channel. The first channel is used to process the inherent feature information, and the second channel is used to process the temporal feature information. If the predicted probability is greater than or equal to a preset threshold, the decision feature information is input into a policy matching model to output the target service policy of the target object. The service layer is also used to call a pre-configured execution interface to execute the target service strategy, thereby triggering a business promotion operation for the target object.

[0016] Thirdly, this application provides an electronic device, comprising: Memory, used to store program instructions; A processor is configured to invoke program instructions stored in the memory and execute the steps of the method described in any one of the first aspects according to the obtained program instructions; A display used to respond to user actions.

[0017] Fourthly, this application provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the method described in any one of the first aspects.

[0018] Fifthly, this application provides a computer program product comprising: computer program code, which, when run on a computer, causes the computer to perform the method described in any one of the first aspects. Attached Figure Description

[0019] Figure 1 A flowchart illustrating a strategy execution method for a potential service object provided in an embodiment of this application; Figure 2 A schematic diagram illustrating the process of training a dual-channel deep learning model, provided in an embodiment of this application; Figure 3 This application provides a schematic diagram of the structure of a business promotion system. Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0021] The terms "first" and "second" in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the term "comprising" and any variations thereof are intended to cover non-exclusive protection. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices. The term "multiple" in this application can mean at least two, for example, two, three, or more, and the embodiments of this application do not impose limitations.

[0022] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of this application, including various details to aid understanding. These embodiments should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this application. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description. It should be noted that in the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solutions of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0023] The acquisition, transmission, storage, and use of data in this application all comply with the requirements of relevant national laws and regulations.

[0024] Before introducing the policy execution method for potential service objects provided in the embodiments of this application, the technical background of the embodiments of this application will be described in detail below for ease of understanding.

[0025] In the business development and customer operations of various organizations, there is a common core need: how to accurately and efficiently identify potential customers who are most likely to accept specific products or services from a massive number of target objects (such as customers and users), and to develop and implement personalized service strategies for them.

[0026] Traditional automated screening and decision-making schemes typically rely on analyzing static, inherent attribute data (such as basic information and financial data) of target objects at a single point in time. Specifically, these schemes collect such static data and use models such as logistic regression, support vector machines, or shallow neural networks for scoring and classification, and then match strategies based on pre-defined simple rules.

[0027] However, with the rapid expansion of data scale and the increase in business complexity, the models in traditional methods are limited by structure and capacity, making it difficult to efficiently process large-scale, high-dimensional, nonlinear data, resulting in low computational efficiency and low decision accuracy.

[0028] To address the aforementioned issues, embodiments of this application provide a strategy execution method and business promotion system for potential service targets. For ease of understanding, the strategy execution method, apparatus, and electronic device for resource transfer targeting potential service targets provided in this application embodiment will be described in detail below with reference to the accompanying drawings.

[0029] like Figure 1The diagram shown is a flowchart illustrating a strategy execution method for potential service targets provided in an embodiment of this application. This method is applied to an organization's business promotion system, which includes a service layer, a data layer, and a model layer. The specific method is as follows: S101: Receive prediction requests for the target object through the service layer; S102: Through the data layer, respond to the prediction request and collect the inherent characteristic information and temporal characteristic information of the target object. The inherent characteristic information represents the basic attributes of the target object, and the temporal characteristic information includes indicator data that reflects the changes in the behavior of the target object over multiple consecutive time periods. S103: Through the model layer, inherent feature information and temporal feature information are input into the dual-channel deep learning model to generate the predicted probability of the target object becoming a potential service object and the decision feature information of the target object. The dual-channel deep learning model includes a first channel and a second channel. The first channel is used to process inherent feature information, and the second channel is used to process temporal feature information. If the predicted probability is greater than or equal to a preset threshold, the decision feature information is input into the policy matching model to output the target service policy of the target object. S104: Through the service layer, the pre-configured execution interface is called to execute the target service strategy, thereby triggering business promotion operations on the target object.

[0030] The strategy execution method for potential service targets provided in this application is applied to an organization's business promotion system. The business promotion system includes a service layer, a data layer, and a model layer. First, the service layer receives prediction requests for target targets. Second, the data layer responds to the prediction requests, collecting the inherent and temporal feature information of the target targets. Then, the model layer inputs the inherent and temporal feature information into a dual-channel deep learning model to generate a predicted probability that the target target will become a potential service target and decision feature information for the target target. The dual-channel deep learning model includes a first channel and a second channel. The first channel processes inherent feature information, and the second channel processes temporal feature information. If the predicted probability is greater than or equal to a preset threshold, the decision feature information is input into a strategy matching model, outputting the target service strategy for the target target. Finally, the service layer calls a pre-configured execution interface to execute the target service strategy, triggering business promotion operations for the target target. This solution, by using the first channel of the dual-channel deep learning model to process inherent feature information and the second channel to process temporal feature information, achieves efficient parallel processing of large-scale, high-dimensional data, improving the system's processing capacity and computational efficiency. Furthermore, by employing a dual-channel deep learning model and a policy matching model, the system achieves accurate identification of potential service targets and automated execution of personalized policies, thereby improving decision-making accuracy and processing efficiency in large-scale data scenarios.

[0031] In this embodiment, the organization's business promotion system adopts a layered architecture design. This architecture decouples and organizes the core functions of the system, specifically dividing them into a service layer, a data layer, and a model layer. This not only clearly defines the technical responsibilities and functional boundaries of each layer, but also ensures the orderly flow and efficient collaboration of data flow, computing logic, and control instructions within the system, together forming a complete intelligent decision-making and execution system.

[0032] The service layer serves as the unified interface and process controller for interaction between the system and the external business environment. Its primary responsibility is to receive business requests (such as prediction requests for specific target objects) and coordinate and schedule lower-level resources. After making a decision, this layer is also responsible for invoking pre-configured execution interfaces to transform the generated target strategy into concrete and accessible business promotion operations, thereby forming a business closed loop.

[0033] The data layer is responsible for the collection, management, and provision of data related to the target object. After being scheduled by the response service layer, it collects multi-dimensional feature information from internal or external data sources and performs data preprocessing operations.

[0034] The model layer integrates the core algorithm model of this application. It receives feature information processed by the data layer and uses a dual-channel deep learning model for parallel analysis and deep feature extraction to generate predicted probabilities and decision features. Then, a policy matching model is used to achieve intelligent mapping from features to personalized policies. This layer encapsulates all machine learning and inference capabilities and is key to improving decision accuracy and adaptability.

[0035] The layered architecture design described above separates interface services, data management, and processing computation, which not only improves the independent maintainability and scalability of each part, but also ensures the stability and processing efficiency of the overall process, providing a technical foundation for dealing with large-scale, high-concurrency business promotion scenarios.

[0036] In one embodiment, such as Figure 2 The diagram shown is a flowchart illustrating a training process for a dual-channel deep learning model provided in this application. S201: Obtain sample data of multiple target objects within a historical time period; Specifically, each target object corresponds to a sample data. For each sample data, the data comes from the integration of multi-source heterogeneous data from the institution (e.g., a bank) and external sources, including extracting multi-dimensional data of the target object from the institution's internal databases (e.g., customer relationship management systems, core transaction systems, credit approval systems). Specifically, this may include inherent characteristic information representing the basic attributes of the target object, such as basic customer information (age, gender, occupation, etc.), static account attributes, and fixed asset status, as well as time-series characteristic information representing indicator data reflecting changes in the target object's behavior over multiple consecutive time periods, such as account transaction records, credit history, and product browsing and query logs within a continuous time window. Labeling information used to indicate whether the target object eventually becomes a "potential service recipient," such as whether it has purchased a product or signed a service contract.

[0037] S202: After preprocessing multiple sets of sample data, a training set, a validation set, and a test set are generated; The preprocessing operation involves cleaning the data based on inherent and temporal characteristics, including removing duplicate records, handling outliers, correcting obvious errors, and filling in missing values ​​using methods such as interpolation, mean imputation, or model-based prediction to ensure data quality and integrity.

[0038] Optionally, feature encoding and representation can be performed on inherent feature information and temporal feature information. For example, categorical variables (such as occupation and region) in the inherent feature information can be numerically represented using one-hot encoding or label encoding. For possible text data, it can be converted into TF-IDF vectors or Word2Vec embedding vector representations. For temporal feature information, sequence alignment and sliding window segmentation are required to form a regular temporal tensor.

[0039] Optionally, standardization and normalization operations can be performed on the inherent feature information and time-series feature information. Continuous inherent feature information (such as income and age) can be standardized or normalized to eliminate the influence of dimensions and accelerate model convergence.

[0040] Optionally, derived features can be constructed based on relevant knowledge of the business domain. For example, "monthly consumption volatility" and "credit limit utilization trend in the past six months" can be calculated based on transaction records; and "recent service attention index" can be constructed based on interaction logs. This provides the model with more discriminative information.

[0041] Optionally, for ultra-high dimensional feature information, dimensionality reduction techniques such as principal component analysis and linear discriminant analysis can be used, or feature selection can be performed through methods such as recursive feature elimination and LASSO regression to remove redundant or weakly correlated features and improve model efficiency and generalization ability.

[0042] S203: Construct the initial dual-channel deep learning model; The specific structure and working method are as follows: First Channel: This channel consists of multiple fully connected layers, forming a deep feedforward neural network. The hidden layers all use the Rectified Linear Unit (ReLU) activation function. The first channel is used to process intrinsic feature information, and its specific operation is as follows: Inherent feature information is input into the input layer of a deep feedforward neural network; The inherent feature information is transformed layer by layer nonlinearly by at least three sequentially connected hidden layers; The output information after the transformation of the last hidden layer is used as the first feature vector.

[0043] Second Channel: This channel is composed of a temporal neural network, which can be an architecture such as a Recurrent Neural Network (RNN), a Long Short-Term Memory (LSTM) network, or a Transformer Encoder. The second channel is used to process temporal feature information, and its specific working method is as follows: The temporal feature information is input into the temporal neural network according to a preset sequence data format (e.g., constructed as a matrix of time and feature dimensions) so that the temporal neural network can encode the temporal relationship in the temporal feature information; A second feature vector is generated based on the encoded temporal feature information.

[0044] Fusion layer: This is used to receive and fuse the first feature vector output from the first channel with the second feature vector output from the second channel. Fusion methods include, but are not limited to: weighted fusion based on attention mechanisms, feature vector concatenation followed by a fully connected layer for dimensionality reduction, or direct weighted summation. The goal is to generate a fused feature vector that comprehensively reflects both the inherent and temporal features of the target object.

[0045] Output layer: This layer receives the fused feature vector output from the fusion layer and generates the first prediction probability through the following operations: After linearly combining and transforming the fused feature vector through a fully connected layer, the result of the linear transformation is nonlinearly mapped by the sigmoid function, outputting a scalar value between 0 and 1. This scalar value serves as the first predicted probability that the target object becomes a potential service object.

[0046] S204: Train and optimize the initial dual-channel deep learning model; After the model architecture is built, the initial dual-channel deep learning model is iteratively trained and optimized using the training and validation sets. The specific process is as follows: For each target object, the inherent feature information in the sample data of the target object is input into the first channel of the model, and the first feature vector is output. The temporal feature information in the sample data of the target object is input into the second channel of the model, and the second feature vector is output. Based on the first feature vector and the second feature vector, the first predicted probability corresponding to the target object is output. Subsequently, the difference between the first predicted probability output by the model and the labeled information is calculated using the binary cross-entropy loss function to generate the loss value.

[0047] Then, using the backpropagation algorithm, based on the calculated loss value, an adaptive optimization algorithm (such as the Adam optimizer) is used to iteratively update the model parameters to minimize the loss value, thereby gradually bringing the model's predictive ability closer to the true distribution of the training data.

[0048] After each training cycle or after several training batches, the current state of the dual-channel deep learning model is evaluated using a validation set. Performance metrics of the dual-channel deep learning model on the validation set are calculated to monitor for overfitting. When the main performance metrics on the validation set no longer improve over several consecutive training cycles, the model is considered to have reached its current optimal generalization ability. Training is then terminated, and the model parameters are rolled back to the snapshot with the best performance on the validation set.

[0049] To further improve model performance, hyperparameters affecting model structure or training process can be systematically tuned. These include, but are not limited to, learning rate, batch size, number of hidden layer neurons, dropout rate, and regularization coefficient. Tuning methods can employ grid search, random search, or more advanced Bayesian optimization, with the goal of finding the optimal combination of hyperparameters on the validation set.

[0050] S205: Evaluate the trained dual-channel deep learning model using a test set.

[0051] After the model training and optimization are completed, the final dual-channel deep learning model is evaluated using a test set.

[0052] First, the feature information of all sample data in the test set is input into the trained dual-channel deep learning model to obtain the model's prediction probability for each sample. Then, based on the predicted probabilities and the labeled information, the model evaluation metrics are calculated. Typically, model evaluation metrics include, but are not limited to, precision, recall, F1 score, and accuracy.

[0053] When the evaluation metrics of the dual-channel deep learning model on the test set reach or exceed the preset business requirement threshold, the training of the model is confirmed to be complete.

[0054] S206: Encapsulate the trained dual-channel deep learning model into a Representational State Transfer Application Programming Interface (RESTful API) and integrate it into the model layer of the business promotion system to support real-time prediction and batch processing.

[0055] In the above embodiments, after the training and evaluation of the dual-channel deep learning model are completed, the trained dual-channel deep learning model is integrated into the model layer of the business promotion system and can provide services through a standardized interface.

[0056] In one embodiment, after taking the first predicted probability output by the trained dual-channel deep learning model as the predicted probability that the target object becomes a potential service object, if the predicted probability is greater than or equal to a preset threshold, the decision feature information is input into the policy matching model to output the target service policy of the target object.

[0057] Specifically, the matching score between decision feature information and multiple candidate service strategies is calculated using a strategy matching model. The candidate service strategy with the highest matching score is selected as the target service strategy.

[0058] It should be noted that this application does not specify a particular algorithm for calculating the matching score. It can be implemented using cosine similarity, neural network-based matching networks, attention mechanism scoring, or any other applicable similarity or relevance calculation algorithm.

[0059] The following section will elaborate on a specific implementation method for using a trained dual-channel deep learning model to execute the entire process from customer identification to strategy triggering, taking into account the specific business scenario of bank credit marketing.

[0060] This embodiment is applied to the marketing scenario of personal credit loans (such as consumer loans and business loans) by commercial banks. The goal is to accurately predict the probability of each customer applying for and successfully signing a designated credit loan product within a certain period (such as 30 days) by analyzing massive amounts of customer data, and to automatically match the best marketing strategy.

[0061] First, after the service layer of the bank's business promotion system receives a prediction request for the target audience, the data layer responds to the prediction request and collects the target audience's inherent characteristic information and time-series characteristic information. For example, inherent characteristic information may include customer profile (age, occupation, education), asset status (average annual income, total financial assets, real estate information), credit foundation (central bank credit score, internal bank rating, social security and housing provident fund payment status), and product holdings (existing deposits, wealth management products, credit cards, etc.). Time-series characteristic information may include fund flow sequence (monthly salary payments, total inflows and outflows, and balance fluctuations over the past 12 months), transaction behavior sequence (average weekly spending amount over the past 90 days, online payment frequency, and number of transfers), credit behavior sequence (historical loan application records, repayment records, and recent credit inquiry times), and interaction behavior sequence (recent frequency and duration of browsing loan product pages on the mobile banking APP, as well as records of inquiries to customer service regarding loan issues). Then, the collected intrinsic feature information and temporal feature information are preprocessed. The preprocessed intrinsic feature information is input into the first channel of the trained dual-channel deep learning model, and the preprocessed temporal feature information is input into the second channel of the trained dual-channel deep learning model. The dual-channel deep learning model outputs the probability value of the target customer successfully signing the target credit loan product in the next 30 days and the decision feature information of the target customer.

[0062] Secondly, the predicted probability is compared with a preset marketing threshold (e.g., 0.72, which can be dynamically adjusted according to marketing resources). If it is lower than the preset marketing threshold, the process is terminated and the target customer is not included in this marketing list. If it is greater than or equal to the preset marketing threshold, the target customer is marked as a "high-potential credit customer" and enters the strategy matching stage.

[0063] Input the decision feature information into the strategy matching model, calculate the matching degree between the decision feature information and each strategy, and output the target marketing strategy, where each strategy can be one of the following: Strategy A: Push exclusive loan quota coupons and interest rate discounts to mobile phones; Strategy B: The account manager will contact the customer by phone to recommend loan products; Strategy C: Conduct SMS recommendations Finally, the service layer of the business promotion system calls the execution interface of the corresponding channel to automatically reach customers for marketing activities.

[0064] Through the above steps, this embodiment realizes an automated closed loop from data-driven prediction to intelligent strategy execution, effectively improving the accuracy and efficiency of bank credit loan marketing.

[0065] In another embodiment, after generating the first prediction probability through the trained dual-channel deep learning model, at least one second prediction probability is output based on the fused feature vector through at least one auxiliary prediction model. The auxiliary prediction model can be a gradient boosting decision tree (GBDT), a random forest, or a support vector machine (SVM), etc.

[0066] Then, the first predicted probability and at least one second predicted probability are fused to generate the predicted probability that the target object becomes a potential service object.

[0067] It should be noted that the specific method of fusion processing can be selected and designed according to actual needs. For example, ensemble learning strategies such as weighted averaging, voting, or stacking can be used. This application does not impose specific limitations on this.

[0068] This embodiment combines a dual-channel deep learning model with at least one auxiliary model for collaborative prediction, and fuses the first prediction probability and at least one second prediction probability to generate the prediction probability that the target object becomes a potential service object, thereby enhancing the accuracy of generating the prediction probability that the target object becomes a potential service object.

[0069] In one embodiment, the business promotion system further includes a monitoring layer; Through the monitoring layer, the operational status of the business promotion system, model prediction results, and feedback information on the execution of target service strategies are collected and monitored, and monitoring analysis reports are generated. The operational status of the system includes, but is not limited to, the response time and success rate of service layer interfaces, the execution status and time consumption of data layer feature extraction tasks, and the load and performance indicators of model layer inference services, as well as other infrastructure-level operational health status. The model prediction results include, but are not limited to, the prediction probability output by the dual-channel deep learning model, the matched target service strategy, and the target object identifier. The feedback information on the execution of target service strategies includes the business results after execution.

[0070] Then, based on the monitoring and analysis report, the dual-channel deep learning model is updated.

[0071] The above embodiments ensure that the business promotion system can not only operate automatically, but also learn and iterate based on real business results, thereby maintaining the accuracy of predictions and the effectiveness of strategies in the long term, demonstrating the system's high level of intelligence and adaptability.

[0072] In one embodiment, to protect the privacy and data security of the target audience, the security of the business promotion system is ensured through the following methods: In the data layer, to ensure the security of the inherent and temporal characteristics of the target object, the inherent and temporal characteristics are encrypted. At the model layer, during the training of the dual-channel deep learning model, noise injection is performed on the training dataset. For example, during the gradient descent iteration, noise conforming to a specific distribution (such as Laplace distribution or Gaussian distribution) is added to the calculated model gradient before parameter updates are performed. This ensures that the model as a whole learns the general distribution law of the data, while strictly limiting the influence of any single training sample on the final model from the algorithm level. This reduces the risk that the model can infer the training data of a specific individual through the prediction results, thus providing privacy protection for the model prediction service.

[0073] Access control mechanisms are implemented through the service layer and model layer to restrict the access to feature information, model interfaces and prediction results. The feature information includes at least one of the following: inherent feature information, time series feature information and decision feature information.

[0074] Based on the same inventive concept, this application also provides a business promotion system. The implementation principle of this business promotion system is similar to that of the strategy execution method for potential service objects. The specific implementation of this business promotion system can be found in the aforementioned embodiment of the strategy execution method for potential service objects, and the repeated parts will not be described again.

[0075] like Figure 3 The diagram shown is a structural schematic of a business promotion system provided in an embodiment of this application, comprising: Service layer 301 is used to receive prediction requests for the target object; Data layer 302 is used to respond to the prediction request and collect the inherent feature information and temporal feature information of the target object. The inherent feature information represents the basic attributes of the target object, and the temporal feature information includes indicator data reflecting the behavioral changes of the target object in multiple consecutive time periods. Model layer 303 is used to input the inherent feature information and the temporal feature information into a dual-channel deep learning model to generate the predicted probability that the target object becomes a potential service object and the decision feature information of the target object. The dual-channel deep learning model includes a first channel and a second channel. The first channel is used to process the inherent feature information, and the second channel is used to process the temporal feature information. If the predicted probability is greater than or equal to a preset threshold, the decision feature information is input into a policy matching model to output the target service policy of the target object. Service layer 301 is also used to call a pre-configured execution interface to execute the target service strategy, so as to trigger a business promotion operation on the target object.

[0076] In one embodiment, model layer 303 is specifically used for: The inherent feature information is input into the first channel, and a nonlinear transformation is performed through the deep feedforward neural network in the first channel to obtain the first feature vector; The temporal feature information is input into the second channel and encoded by the temporal neural network in the second channel to obtain the second feature vector; The first feature vector and the second feature vector are input into the fusion layer of the dual-channel deep learning model to generate a first prediction probability, and the first prediction probability is used as the prediction probability that the target object becomes a potential service object.

[0077] In one embodiment, after generating the first predicted probability, model layer 303 is further configured to: Based on the fused feature vector, at least one second prediction probability is output using at least one auxiliary prediction model. The first predicted probability and the at least one second predicted probability are fused together to generate a predicted probability that the target object becomes a potential service object.

[0078] In one embodiment, model layer 303 is specifically used for: The correlation weight between the first feature vector and the second feature vector is calculated through the fusion layer of the dual-channel deep learning model. Based on the correlation weights, the first feature vector and the second feature vector are weighted and fused to obtain a fused feature vector. The fused feature vector is input into the output layer of the dual-channel deep learning model and subjected to fully connected linear computation and logistic function transformation to generate the first prediction probability.

[0079] In one embodiment, model layer 303 is specifically used for: Using the strategy matching model, the matching score between the decision feature information and multiple candidate service strategies is calculated respectively. The candidate service strategy with the highest matching score is selected as the target service strategy.

[0080] In one embodiment, the business promotion system further includes a monitoring layer 304, which is specifically used for: The monitoring layer collects and monitors the operating status of the business promotion system, model prediction results, and feedback information on the execution of the target service strategy, and generates a monitoring analysis report. Based on the monitoring and analysis report, the dual-channel deep learning model is updated.

[0081] In one embodiment, the data layer 302 is further configured to: The inherent feature information and the temporal feature information are encrypted. Model layer 303 is also used for: During the training of the dual-channel deep learning model, noise injection is performed on the training dataset; Service layer 301 and model layer 303 are also used for: An access control mechanism is implemented to restrict access to feature information, model interfaces, and prediction results, wherein the feature information includes at least one of the following: the inherent feature information, the time-series feature information, and the decision feature information.

[0082] In one embodiment, model layer 303 is specifically used for: The inherent feature information is input into the input layer of the deep feedforward neural network; The intrinsic feature information is transformed nonlinearly layer by layer through at least three sequentially connected hidden layers, wherein each hidden layer uses a linear rectified function as the activation function. The output information after the transformation of the last hidden layer is used as the first feature vector.

[0083] In one embodiment, model layer 303 is specifically used for: The temporal feature information is input into the temporal neural network according to a preset sequence data format, so that the temporal neural network encodes the temporal relationship in the temporal feature information; The second feature vector is generated based on the encoded temporal feature information.

[0084] like Figure 4 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this application, comprising: Memory 401 is used to store program instructions; Processor 402 is configured to call program instructions stored in the memory and execute the following steps according to the obtained program instructions: The service layer receives prediction requests for the target object. Through the data layer, in response to the prediction request, the inherent feature information and temporal feature information of the target object are collected. The inherent feature information characterizes the basic attributes of the target object, and the temporal feature information includes indicator data reflecting the behavioral changes of the target object over multiple consecutive time periods. Through the model layer, the inherent feature information and the temporal feature information are input into a dual-channel deep learning model to generate the predicted probability that the target object becomes a potential service object and the decision feature information of the target object. The dual-channel deep learning model includes a first channel and a second channel. The first channel is used to process the inherent feature information, and the second channel is used to process the temporal feature information. If the predicted probability is greater than or equal to a preset threshold, the decision feature information is input into a policy matching model to output the target service policy of the target object. The service layer invokes a pre-configured execution interface to execute the target service strategy, thereby triggering a business promotion operation for the target object.

[0085] Based on the same inventive concept, this application also provides a computer-readable storage medium. The implementation principle of the computer-readable storage medium is similar to that of the policy execution method for potential service objects. For the specific implementation of the computer-readable storage medium, please refer to the aforementioned policy execution method embodiment for potential service objects. Repeated details will not be repeated.

[0086] Based on the same inventive concept, this application also provides a computer program product. The implementation principle of the computer program product is similar to that of the strategy execution method for potential service objects. The specific implementation of the computer program product can be found in the aforementioned embodiment of the strategy execution method for potential service objects, and the repeated parts will not be described again.

[0087] The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a computer, cause the computer to perform the policy execution method for a potential service object as described in any of the preceding claims.

[0088] The computer program product includes: computer program code, which, when run on a computer, causes the computer to execute the policy execution method for potential service objects as described above.

[0089] This application provides a strategy execution method and business promotion system for potential service objects. The business promotion system includes a service layer, a data layer, and a model layer. First, the service layer receives prediction requests for target objects. Second, the data layer responds to the prediction requests, collecting the inherent feature information and temporal feature information of the target objects. Then, the model layer inputs the inherent feature information and temporal feature information into a dual-channel deep learning model to generate the predicted probability of the target object becoming a potential service object and the decision feature information of the target object. The dual-channel deep learning model includes a first channel and a second channel. The first channel processes the inherent feature information, and the second channel processes the temporal feature information. If the predicted probability is greater than or equal to a preset threshold, the decision feature information is input into a strategy matching model, outputting the target service strategy for the target object. Finally, the service layer calls a pre-configured execution interface to execute the target service strategy, triggering the business promotion operation for the target object. The above solution, by using the first channel of the dual-channel deep learning model to process inherent feature information and the second channel to process temporal feature information, achieves efficient parallel processing of large-scale, high-dimensional data, improving the system's processing capacity and computational efficiency. Furthermore, by employing a dual-channel deep learning model and a policy matching model, the system achieves accurate identification of potential service targets and automated execution of personalized policies, thereby improving decision-making accuracy and processing efficiency in large-scale data scenarios.

[0090] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0091] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0092] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0093] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0094] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A strategy execution method for potential service objects, characterized in that, A business promotion system applied to an organization, the business promotion system comprising a service layer, a data layer, and a model layer; including: The service layer receives prediction requests for the target object. Through the data layer, in response to the prediction request, the inherent feature information and temporal feature information of the target object are collected. The inherent feature information characterizes the basic attributes of the target object, and the temporal feature information includes indicator data reflecting the behavioral changes of the target object over multiple consecutive time periods. Through the model layer, the inherent feature information and the temporal feature information are input into a dual-channel deep learning model to generate the predicted probability that the target object becomes a potential service object and the decision feature information of the target object. The dual-channel deep learning model includes a first channel and a second channel. The first channel is used to process the inherent feature information, and the second channel is used to process the temporal feature information. If the predicted probability is greater than or equal to a preset threshold, the decision feature information is input into a policy matching model to output the target service policy of the target object. The service layer invokes a pre-configured execution interface to execute the target service strategy, thereby triggering a business promotion operation for the target object.

2. The method according to claim 1, characterized in that, The step of inputting the inherent feature information and the temporal feature information into a dual-channel deep learning model to generate a predicted probability that the target object will become a potential service object includes: The inherent feature information is input into the first channel, and a nonlinear transformation is performed through the deep feedforward neural network in the first channel to obtain the first feature vector; The temporal feature information is input into the second channel and encoded by the temporal neural network in the second channel to obtain the second feature vector; The first feature vector and the second feature vector are input into the fusion layer of the dual-channel deep learning model to generate a first prediction probability, and the first prediction probability is used as the prediction probability that the target object becomes a potential service object.

3. The method according to claim 2, characterized in that, After generating the first predicted probability, the method further includes: Based on the fused feature vector, at least one second prediction probability is output using at least one auxiliary prediction model. The first predicted probability and the at least one second predicted probability are fused together to generate a predicted probability that the target object becomes a potential service object.

4. The method according to claim 2, characterized in that, The step of inputting the first feature vector and the second feature vector into the fusion layer of the dual-channel deep learning model to generate the first prediction probability includes: The correlation weight between the first feature vector and the second feature vector is calculated through the fusion layer of the dual-channel deep learning model. Based on the correlation weights, the first feature vector and the second feature vector are weighted and fused to obtain a fused feature vector. The fused feature vector is input into the output layer of the dual-channel deep learning model and subjected to fully connected linear computation and logistic function transformation to generate the first prediction probability.

5. The method according to any one of claims 1 to 4, characterized in that, The step of inputting the decision feature information into the strategy matching model and outputting the target service strategy of the target object includes: Using the strategy matching model, the matching score between the decision feature information and multiple candidate service strategies is calculated respectively. The candidate service strategy with the highest matching score is selected as the target service strategy.

6. The method according to any one of claims 1 to 4, characterized in that, The business promotion system also includes a monitoring layer, and the method further includes: The monitoring layer collects and monitors the operating status of the business promotion system, model prediction results, and feedback information on the execution of the target service strategy, and generates a monitoring analysis report. Based on the monitoring and analysis report, the dual-channel deep learning model is updated.

7. The method according to any one of claims 1 to 4, characterized in that, The method further includes at least one of the following steps: The inherent feature information and the temporal feature information are encrypted through the data layer; Through the model layer, noise injection processing is performed on the training dataset during the training of the dual-channel deep learning model; Access control mechanisms are implemented through the service layer and the model layer to restrict access to feature information, model interfaces and prediction results. The feature information includes at least one of the following: inherent feature information, time-series feature information and decision feature information.

8. The method according to claim 2, characterized in that, The step of inputting the inherent feature information into the first channel and performing a nonlinear transformation through a deep feedforward neural network in the first channel to obtain a first feature vector includes: The inherent feature information is input into the input layer of the deep feedforward neural network; The intrinsic feature information is transformed nonlinearly layer by layer through at least three sequentially connected hidden layers, wherein each hidden layer uses a linear rectified function as the activation function. The output information after the transformation of the last hidden layer is used as the first feature vector.

9. The method according to claim 2, characterized in that, The step of inputting the temporal feature information into the second channel and encoding it through the temporal neural network in the second channel to obtain the second feature vector includes: The temporal feature information is input into the temporal neural network according to a preset sequence data format, so that the temporal neural network encodes the temporal relationship in the temporal feature information; The second feature vector is generated based on the encoded temporal feature information.

10. A business promotion system, characterized in that, include: The service layer is used to receive prediction requests for the target object; The data layer is used to respond to the prediction request and collect the inherent feature information and temporal feature information of the target object. The inherent feature information represents the basic attributes of the target object, and the temporal feature information includes indicator data that reflects the changes in the behavior of the target object over multiple consecutive time periods. The model layer is used to input the inherent feature information and the temporal feature information into a dual-channel deep learning model to generate the predicted probability that the target object becomes a potential service object and the decision feature information of the target object. The dual-channel deep learning model includes a first channel and a second channel. The first channel is used to process the inherent feature information, and the second channel is used to process the temporal feature information. If the predicted probability is greater than or equal to a preset threshold, the decision feature information is input into a policy matching model to output the target service policy of the target object. The service layer is also used to call a pre-configured execution interface to execute the target service strategy, thereby triggering a business promotion operation for the target object.

11. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor is configured to invoke program instructions stored in the memory and execute the steps of the method according to any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a computer, cause the computer to perform the method as described in any one of claims 1-9.

13. A computer program product, characterized in that, The computer program product includes: computer program code, which, when run on a computer, causes the computer to perform the method described in any one of claims 1-9.

Images