A data processing method and system for intelligent marketing products

Abstract

The application relates to a data processing method for intelligent marketing products. First, the financial behavior data of target users is obtained, and a machine learning algorithm is used for clustering. A preset threshold is used to select a target cluster. A matching degree model is defined to evaluate the comprehensive matching degree between users and different loan products, and the preference value of the target cluster is predicted. Then, promotion information is customized according to the sales target, and a marketing influence factor is calculated. According to the matching degree and the marketing influence factor, the recommendation score of each user for each loan product is calculated, the comprehensive score is calculated according to the recommendation score, the loan products are prioritized, and the corresponding marketing strategy is selected. Finally, KPIs related to the marketing strategy are collected, the effect is evaluated, and the strategy parameters are adjusted. Through accurate user data analysis and personalized marketing strategy, the application improves the recommendation accuracy and marketing effect of the loan product, and enhances the user satisfaction and loyalty.

Description

Technical Field

[0001] This invention relates to the field of intelligent financial risk control technology, and more specifically, to a data processing method and system for intelligent marketing products. Background Technology

[0002] In the design of consumer finance products, institutions often adopt differentiated pricing strategies for different customer groups to balance profit requirements and user acceptance (indicators such as utilization rate and loan scale). The most common method is to set corresponding interest rates and credit limits based on the estimated default rate of different customer groups, taking into account costs and profit requirements. In actual business, credit limits, default rates, interest rates, and user acceptance (hereinafter referred to as conversion rates) are interdependent. Higher interest rates are associated with higher user risk, and user conversion rates decrease accordingly, and vice versa. Higher credit limits lead to higher conversion rates, but default rates may also increase. Currently, most financial institutions only assess the user's risk level to decide whether to approve or reject the loan, rather than combining credit limits and interest rates. This results in excessively low conversion rates, wasted customer acquisition costs, failure to maximize the value of low-risk users, and a lack of organic integration of credit limits, default rates, interest rates, and conversion rates for medium- and high-risk users, failing to calculate the cost of assets and the profitability after loan disbursement. Therefore, we hope to have a risk pricing and segmented marketing strategy that can further segment the customer base, match different product pricing based on the risk of different user categories, and then maximize user conversion rates according to marketing strategies, find the balance point of revenue, and improve business profitability. Summary of the Invention

[0003] This invention provides a data processing method for intelligent marketing products, which can improve customers' financial management capabilities and acceptance rates, reduce risk coefficients, and conversely, reduce risk coefficients to improve customers' financial management capabilities and acceptance rates, thereby finding a balance point of returns and improving business profitability.

[0004] This invention provides a data processing method for intelligent marketing products, comprising the following steps:

[0005] Step S01: Obtain the financial behavior data of the target user, which includes credit card usage records, consumption records, historical loan records and historical repayment records. Convert the raw data into vector features through feature extraction and form a set. Use a clustering algorithm to divide the vector feature set into multiple clusters. Label the user with the corresponding cluster label according to the clustering results.

[0006] Step S02: Calculate the value score of each cluster using a preset value assessment model, identify candidate clusters with value scores greater than a preset threshold as target clusters, and remove non-target clusters from the dataset;

[0007] Step S03: Evaluate the overall matching degree between users in each target cluster and multiple loan products through a preset matching degree model, and predict the preference value of users in each target cluster for different loan products based on the matching degree model. The matching degree model is established based on the historical loan behavior of users in each target cluster.

[0008] Step S04: Customize promotional information based on the sales target of the loan product within a preset period. The promotional information includes discount information and time-limited information. Predict the promotional effect value of the loan product within the preset period and calculate the promotional impact factor based on the promotional effect value.

[0009] Step S05: Combine matching degree and marketing influence factor to calculate the recommendation score of users in each target cluster for each loan product. Calculate the comprehensive score of each loan product based on the recommendation score. Based on the comprehensive score, classify the loan products into different priority categories and select the corresponding marketing strategy according to the priority category.

[0010] Step S06: Collect KPIs related to the marketing strategies within the preset period to evaluate the marketing effectiveness of each marketing strategy, and adjust the parameters of the marketing strategies for the next preset period based on the marketing effectiveness.

[0011] Preferably, in step S01, a clustering algorithm is used to divide the vector feature set into multiple clusters, and the user is labeled with the corresponding cluster label based on the clustering results, including:

[0012] Obtain the user's raw dataset Through feature extraction function Original data Convert to vector features ,in Indicates the first The data for each user, specifically represented by the feature vector, is as follows: ,in It is the number of features;

[0013] Use clustering algorithms to group feature vector sets Divided into K clusters;

[0014] Clustering results are represented as functions ,function Data for each user Mapping to cluster labels , represented as: ;

[0015] The K-means clustering algorithm is applied to minimize the variance within each cluster, as follows: ,in It is the first One cluster, It is a cluster center;

[0016] Each user is assigned a corresponding cluster label based on their cluster. This assignment is done using the CreditLevel function, as follows:

[0017] .

[0018] Preferably, in step S02, the value assessment model is established based on the user's credit rating, historical response rate, and purchasing power, and is expressed as follows:

[0019]

[0020] in, , , These are the weighting coefficients. , and This represents the average of the credit rating, response rate, and purchasing power of all users within a cluster.

[0021] The value score of each cluster is calculated using a pre-defined value assessment model. Candidate clusters with value scores greater than a pre-defined threshold are identified as target clusters. Non-target clusters in the dataset are then removed. This is represented as follows:

[0022]

[0023] in, This represents the new user dataset after removing users who were not in the target cluster.

[0024] This represents the set of all users in the non-target cluster.

[0025] Preferably, in step S03, a matching degree model is defined to evaluate the matching degree between the user and multiple loan products, including:

[0026] definition A collection of loan products. Each loan product It has multiple attributes, including interest rate. Loan amount and repayment period ;

[0027] By using the comprehensive matching degree function Evaluation of the first The target clustering and the first The matching degree between individual loan products, specifically expressed by the matching degree function, is as follows:

[0028]

[0029] in, To calculate the matching degree function for interest rate, loan amount, and repayment period, It is the weight used to calculate the interest rate, loan amount, and repayment period;

[0030] Interest rate adaptation Represented as: ;

[0031] Loan amount matching Represented as: ;

[0032] Repayment period matching Represented as: ;

[0033] The preference values ​​for different loan products among users in each target cluster are predicted based on the matching degree model, and are expressed as follows:

[0034]

[0035] in, and It's about adjusting parameters. Indicates based on the first Historical lending behavior of each target cluster and the first Compatibility between loan products.

[0036] Preferably, in step S04, promotional information is customized based on sales targets and market demand, wherein the promotional information is expressed as follows:

[0037]

[0038] in, The corresponding descriptive text is generated based on the user's clustering tags. For products In time Discount rate, For products Promotion period, To customize promotional information;

[0039] Get the promotional performance value within the previous preset period. The predicted promotional effect value within the preset period is expressed as:

[0040]

[0041] in, For prediction functions;

[0042] Calculate the predicted promotional influencing factors within the current preset period. , represented as:

[0043]

[0044] in, , and It refers to adjusting parameters.

[0045] Preferably, combining matching degree and marketing influence factor, a recommendation score is calculated for each target cluster of users for each loan product, and each cluster... For each product The recommended score is expressed as , represented as:

[0046]

[0047] A comprehensive score for each loan product is calculated based on the recommendation score. , represented as:

[0048]

[0049] Based on a comprehensive score, loan products are categorized into different types, including "high priority," "medium priority," and "low priority," with high priority being... Medium priority Low priority is , and A preset scoring threshold is used to distinguish products with different priorities;

[0050] At the same time, select each target cluster The top n products with the highest recommendation scores are then pushed to the device in a personalized way.

[0051] Preferably, KPIs related to the marketing strategy are collected, and these KPIs are used to evaluate the effectiveness of each marketing strategy. The parameters of the marketing strategy are adjusted based on changes in the KPIs, including:

[0052] Collect marketing-related KPIs, including click-through rate, conversion rate, and user engagement. Use a statistical model to correlate marketing strategies with changes in these KPIs, and represent the results as follows:

[0053]

[0054] in, This indicates the change in KPIs due to marketing strategies. This is the point in time before the marketing strategy is implemented;

[0055] Define a policy adjustment function to adjust the policy based on the base The result updates the parameters of the marketing strategy, expressed as: ;

[0056] in, In time For products Marketing strategy parameter vector, It's the learning rate. It is a loss function Regarding strategy parameters The gradient of the loss function measures the difference between the change in KPI and the policy objective.

[0057] Preferably, the discount rate in the initial promotional information is set based on the product's sales speed and target sales speed in the previous preset period, expressed as:

[0058]

[0059] in, and These are the minimum and maximum discount rates, respectively. Based on the current sales speed of the product, The target sales speed for the product;

[0060] The promotion period in the initial promotional information is set based on the product's seasonal demand and expected market response, as follows: ;

[0061] in, and These are the minimum and maximum values ​​of the promotion period, respectively. Based on the current sales speed of the product, The target sales speed for the product;

[0062] Discount rates and promotional periods are adjusted regularly through a feedback mechanism.

[0063] This invention also provides a data processing system for intelligent marketing products, based on the aforementioned data processing method for intelligent marketing products, characterized by comprising the following modules:

[0064] Clustering module: Acquires financial behavior data of target users, including credit card usage records, consumption records, historical loan records and historical repayment records. Converts the raw data into vector features and forms a set through feature extraction function. Uses clustering algorithm to divide the vector feature set into multiple clusters. Labels users with corresponding clusters based on the clustering results.

[0065] Screening module: Calculates the value score of each cluster using a preset value assessment model, identifies candidate clusters with value scores greater than a preset threshold as target clusters, and removes non-target clusters from the dataset;

[0066] Matching degree calculation module: It evaluates the comprehensive matching degree between users in each target cluster and multiple loan products through a preset matching degree model, and predicts the preference value of users in each target cluster for different loan products based on the matching degree model. The matching degree model is established based on the historical loan behavior of users in each target cluster.

[0067] Promotion module: Customizes promotional information based on the sales target of the loan product within a preset period. The promotional information includes discount information and time-limited information. Predicts the promotional effect value of the loan product within the preset period and calculates the promotional impact factor based on the promotional effect value.

[0068] Marketing Strategy Module: Combining matching degree and marketing influence factor, calculate the recommendation score of users in each target cluster for each loan product, calculate the comprehensive score of each loan product based on the recommendation score, divide the loan products into different priority categories based on the comprehensive score, and select the corresponding marketing strategy according to the priority category;

[0069] Feedback Update Module: Collects KPIs related to marketing strategies within a preset period, uses a data processing platform to evaluate the effectiveness of each marketing strategy, and adjusts the parameters of the marketing strategies for the next preset period based on changes in KPIs.

[0070] The beneficial effects of this invention are as follows: By using clustering algorithms to analyze users' financial behavior data, the method can accurately group users, ensuring that marketing campaigns target user groups with similar financial behaviors and needs. Precise target segmentation helps marketing campaigns become more effective because promotional information and product recommendations are tailored to users' actual needs and historical behavior. By evaluating the value score of each cluster and eliminating those with low value, marketing resources can be more effectively allocated to target clusters most likely to generate high returns. This increases the ROI of marketing campaigns and reduces unnecessary spending on low-return groups. The matching model predicts users' preferences for different loan products based on their historical loan behavior. This means users are more likely to receive product information they are actually interested in, thereby improving user satisfaction and potentially increasing user loyalty to the brand. By monitoring sales targets and promotional effectiveness, and then calculating promotional impact factors, marketing teams can continuously adjust and optimize their strategies. This dynamic adjustment can respond to market changes, ensuring that marketing campaigns remain highly efficient and effective. Collecting KPIs related to marketing strategies and adjusting marketing strategies based on these indicators ensures that the decision-making process is entirely based on actual data and empirical results. This data-driven approach reduces the space for bias and guesswork, making marketing decisions more scientific and precise. Because adjustments to marketing strategies and promotional activities are based on real-time or periodic KPI feedback, organizations can respond quickly to market changes, rapidly capitalize on new market opportunities, or adjust strategies to meet challenges. Therefore, it can significantly enhance the market competitiveness of financial institutions, achieve higher economic benefits, improve customer experience, and further drive business growth and market expansion. Attached Figure Description

[0071] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0072] Figure 1 This is a schematic diagram of a data processing method for an intelligent marketing product according to an embodiment of the present invention. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, a clear and complete description will be provided below in conjunction with the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0074] like Figure 1 As shown, the present invention provides a data processing method for intelligent marketing products, including the following steps:

[0075] Step S01: Obtain the financial behavior data of the target user, which includes credit card usage records, consumption records, historical loan records and historical repayment records. Convert the raw data into vector features through feature extraction and form a set. Use a clustering algorithm to divide the vector feature set into multiple clusters. Label the user with the corresponding cluster label according to the clustering results.

[0076] Step S02: Calculate the value score of each cluster using a preset value assessment model, identify candidate clusters with value scores greater than a preset threshold as target clusters, and remove non-target clusters from the dataset;

[0077] Step S03: Evaluate the overall matching degree between users in each target cluster and multiple loan products through a preset matching degree model, and predict the preference value of users in each target cluster for different loan products based on the matching degree model. The matching degree model is established based on the historical loan behavior of users in each target cluster.

[0078] Step S04: Customize promotional information based on the sales target of the loan product within a preset period. The promotional information includes discount information and time-limited information. Predict the promotional effect value of the loan product within the preset period and calculate the promotional impact factor based on the promotional effect value.

[0079] Step S05: Combine matching degree and marketing influence factor to calculate the recommendation score of users in each target cluster for each loan product. Calculate the comprehensive score of each loan product based on the recommendation score. Based on the comprehensive score, classify the loan products into different priority categories and select the corresponding marketing strategy according to the priority category.

[0080] Step S06: Collect KPIs related to the marketing strategies within the preset period to evaluate the marketing effectiveness of each marketing strategy, and adjust the parameters of the marketing strategies for the next preset period based on the marketing effectiveness.

[0081] This invention analyzes users' financial behavior data using clustering algorithms. The method accurately groups users, ensuring marketing campaigns target user groups with similar financial behaviors and needs. Precise target segmentation makes marketing campaigns more effective because promotional information and product recommendations are tailored to users' actual needs and historical behavior. By evaluating the value score of each cluster and eliminating low-value clusters, marketing resources can be more effectively allocated to target clusters most likely to generate high returns. This increases the ROI of marketing campaigns and reduces unnecessary spending on low-return groups. The matching model predicts users' preferences for different loan products based on their historical loan behavior. This means users are more likely to receive product information they are actually interested in, thereby increasing user satisfaction and potentially increasing brand loyalty. By monitoring sales targets and promotional effectiveness, and then calculating promotional impact factors, marketing teams can continuously adjust and optimize their strategies. This dynamic adjustment responds to market changes, ensuring marketing campaigns remain highly efficient and effective. Collecting KPIs related to marketing strategies and adjusting strategies based on these indicators ensures the decision-making process is entirely based on actual data and empirical results. This data-driven approach reduces the space for bias and guesswork, making marketing decisions more scientific and precise. Because adjustments to marketing strategies and promotional activities are based on real-time or periodic KPI feedback, organizations can respond quickly to market changes, rapidly capitalize on new market opportunities, or adjust strategies to address challenges. This significantly enhances the market competitiveness of financial institutions, achieves higher economic efficiency, improves customer experience, and further drives business growth and market expansion.

[0082] The user's financial behavior data is obtained from user data submitted during loan applications or other public channels, and is stored using a data storage device. The user data is encrypted using homomorphic encryption and dynamic segmentation technology. Each user's data includes identity data, credit history, and recent (e.g., within one year) financial records. A decentralized identity verification framework is constructed using blockchain technology to store user data. An identity verification smart contract is designed to verify user identity. The user's credit score is dynamically updated based on transaction history and external data. The data storage device specifically includes: creating a decentralized identity database using blockchain; encrypting and storing user data on the blockchain; and transmitting encryption keys to authorized verification parties through a key exchange protocol to ensure the secure storage and management of the keys.

[0083] This invention employs a data processing platform to process and analyze user data to support various decision-making processes. The platform's main functions and features include:

[0084] The data processing platform uses machine learning algorithms to cluster and analyze collected data, identifying key characteristics such as user credit ratings. It aims to discover patterns and correlations within the data, helping to label users with cluster tags. The platform supports defining and executing matching models to assess the fit between users and loan products, predict user product preferences, and calculate the potential impact of promotional activities. With real-time data processing capabilities, the platform can instantly receive and analyze user credit ratings and financial needs information to quickly respond to market changes and user demands. The platform uses a decision support system to calculate a recommendation score for each user for each loan product and categorizes loan products into high, medium, and low priorities based on these scores. It supports the development and optimization of personalized marketing strategies. The platform also integrates performance monitoring tools to collect and analyze key performance indicators (KPIs) related to marketing strategies and automatically adjust marketing strategy parameters based on KPI changes to ensure optimal marketing campaign results.

[0085] Data processing platforms can include Apache Spark, Apache Hadoop, Google BigQuery, and Amazon AWS. The platform should be able to efficiently process large amounts of data and support real-time data analysis. It should also be able to easily scale as the amount of data increases. Furthermore, the platform needs to support multiple data analysis tools and languages, and have built-in or compatibility with advanced machine learning and data mining tools.

[0086] Data storage devices can be selected from Amazon S3, Google Cloud Storage, Microsoft Azure Blob Storage, and MongoDB. When using data storage devices, structured, semi-structured, and unstructured data are supported to ensure high availability and disaster recovery capabilities.

[0087] In this embodiment, data processing platforms such as Apache Spark or Google BigQuery are combined with data storage devices such as Amazon S3 or Google Cloud Storage. This combination provides powerful data processing capabilities, flexible storage options, and high scalability, while ensuring high system performance and security.

[0088] In step S01 of this embodiment, a clustering algorithm is used to divide the vector feature set into multiple clusters, and the user is labeled with the corresponding cluster label according to the clustering results, including:

[0089] Obtain the user's raw dataset Through feature extraction function Original data Convert to vector features ,in Indicates the first The data for each user, specifically represented by the feature vector, is as follows: ,in It is the number of features;

[0090] Use clustering algorithms to group feature vector sets Divided into K clusters;

[0091] Clustering results are represented as functions ,function Data for each user Mapping to cluster labels , represented as: ;

[0092] The K-means clustering algorithm is applied to minimize the variance within each cluster, as follows: ,in It is the first One cluster, It is a cluster center;

[0093] Each user is assigned a corresponding cluster label based on their cluster. This assignment is done using the CreditLevel function, as follows:

[0094] .

[0095] The extraction function Principal Component Analysis (PCA) involves: standardizing (normalizing) each feature in the original dataset to ensure that the mean of each feature is 0 and the standard deviation is 1. Based on the standardized data, a covariance matrix is ​​calculated between the features. The covariance matrix reveals the linear relationship between different variables. Eigenvalue decomposition is performed on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors. The eigenvectors represent the orientation of the principal components in the data, while the eigenvalues ​​represent the importance of each principal component (the amount of variance explained). Multiple principal components are selected based on the eigenvalues, and the original data is projected using the selected principal components to obtain dimensionality-reduced data. The purpose of the CreditLevel function is to convert user cluster labels obtained from clustering algorithms into specific credit ratings. This process needs to be both intuitive and reliable for easy interpretation and validation. The following detailed instructions and steps demonstrate how to design, implement, and validate this key function: The CreditLevel function takes a cluster label as input and outputs a corresponding credit rating. These credit ratings can be categorical labels such as "high credit," "medium credit," and "low credit," or more specific credit scores.

[0096] Each cluster is assigned a credit rating based on its characteristics—such as the average delinquency rate of users in the cluster, credit card usage rate, historical loan performance, etc. For example, if users in cluster 1 have a low delinquency rate and a healthy credit history, then CreditLevel(1) is defined as "high credit". If users in cluster 2 have a high delinquency rate, then CreditLevel(2) is defined as "low credit".

[0097] Validation and Adjustment: The accuracy of the CreditLevel function is validated using historical datasets containing users' actual credit performance. The model-assigned credit ratings are compared with users' subsequent credit performance to assess the accuracy and reliability of the credit rating predictions. The mapping logic of the CreditLevel function is adjusted based on the validation results. For example, if users in a certain cluster exhibit better credit behavior than expected, the credit rating for that cluster may need to be increased. Machine learning models can be introduced to automatically adjust the logic of credit rating determination, ensuring that credit ratings dynamically reflect the latest credit risk trends.

[0098] Regularly review and update the CreditLevel function to ensure it aligns with the latest credit risk models and market conditions. Monitor any feedback or complaints during the credit rating process to improve the function's transparency and fairness. Through these steps, the CreditLevel function not only provides accurate credit ratings for loan recommendation systems, but its process is also clearly interpretable and easily validated and adjusted using historical data.

[0099] Publicly disclose the criteria used to determine the credit rating for each cluster, including the main variables and statistical data considered. Use visualization tools to show the key characteristics of each cluster and how they map to a specific credit rating. Provide an interface or report detailing why a user was assigned to a specific credit rating and the key factors that may influence their credit score.

[0100] Therefore, precise cluster analysis allows for a better understanding of users' behavioral patterns and financial situations, leading to more accurate credit ratings. Based on user cluster tags and credit ratings, recommendation systems can provide more personalized and relevant loan products. Understanding the preferences and creditworthiness of users in different clusters helps marketing teams target marketing campaigns more effectively, reducing marketing costs and increasing conversion rates. When users receive loan product recommendations that better suit their financial situation and needs, their satisfaction is likely to increase, leading to greater loyalty to the service provider. More accurate credit ratings enable lending institutions to better manage risk and develop suitable loan terms, thereby reducing default rates. In conclusion, through precise credit rating via cluster analysis, loan product recommendation systems not only improve operational efficiency but also enhance overall business performance and customer satisfaction.

[0101] In step S02 of this embodiment, the value assessment model is established based on the user's credit rating, historical response rate, and purchasing power, and is expressed as follows:

[0102]

[0103] in, , , These are the weighting coefficients. , and This represents the average of the credit rating, response rate, and purchasing power of all users within a cluster.

[0104] The value score of each cluster is calculated using a pre-defined value assessment model. Candidate clusters with value scores greater than a pre-defined threshold are identified as target clusters. Non-target clusters in the dataset are then removed. This is represented as follows:

[0105]

[0106] in, This represents the new user dataset after removing users who were not in the target cluster.

[0107] This represents the set of all users in the non-target cluster.

[0108] The value assessment model is calculated by weighting and summing the average credit rating, response rate, and purchasing power for each cluster. Credit rating represents the reliability of users within a cluster, response rate reflects their participation in past marketing campaigns, and purchasing power estimates their spending capacity. These three weights correspond to the business importance of credit rating, response rate, and purchasing power, respectively. These weights are derived from historical data analysis.

[0109] A value score is calculated for each cluster, and a threshold is set for comparison of the value scores of each cluster. All clusters with value scores higher than the preset threshold are identified as target clusters and included in the new user dataset. Clusters with value scores lower than the preset threshold are removed from the dataset. Considering the dynamic changes in the market and user behavior, it is necessary to update the cluster information and recalculate the value scores regularly. This can be done quarterly, semi-annually, or annually, depending on business needs and the rate of data change.

[0110] By identifying and focusing on high-value user clusters, marketing resources are allocated more effectively, thereby reducing costs and improving the overall ROI of marketing campaigns. Concentrating marketing efforts on clusters with high response rates can improve the overall response rate and engagement of marketing campaigns. Understanding and meeting the specific needs of high-value clusters can enhance the satisfaction and loyalty of these users. Using quantitative methods to assess the value of user groups makes decision-making more scientific and objective, reducing the bias of subjective judgment. Overall, step S02 not only improves the accuracy of marketing strategies but also optimizes resource allocation through scientific data processing methods, further promoting the achievement of the company's business objectives.

[0111] In step S03 of this embodiment, a matching degree model is defined to evaluate the matching degree between a user and multiple loan products, including:

[0112] definition A collection of loan products. Each loan product It has multiple attributes, including interest rate. Loan amount and repayment period ;

[0113] By using the comprehensive matching degree function Evaluation of the first The target clustering and the first The matching degree between individual loan products, specifically expressed by the matching degree function, is as follows:

[0114]

[0115] in, To calculate the matching degree function for interest rate, loan amount, and repayment period, It is the weight used to calculate the interest rate, loan amount, and repayment period;

[0116] Machine learning models, such as random forests or gradient boosting machines, can be used to automatically adjust the weights in the matching function of each attribute. To better accommodate the preferences of different users; represented as:

[0117]

[0118] Here, MLModel represents the machine learning model, and Feedback is the user's feedback on the recommended product.

[0119] Interest rate adaptation Represented as:

[0120] ;

[0121] in, It is a loan product. interest rates; It is based on the user Clustering tags The resulting expected interest rate. This means that the system has an expected interest rate based on the user's credit rating; This is a positive adjustment coefficient used to control the sensitivity of the matching function. The magnitude of this coefficient determines the degree to which the interest rate deviation affects the fitted value. Function The core is the exponential function, which ensures The value is always between 0 and 1. Interest rate The closer to the user's expected interest rate The closer the value is to 1, the higher the fit; conversely, the lower the value is, the lower the fit.

[0122] Loan amount matching Represented as: ;

[0123] in, It is a product Loan amount; It is clustering Ideal loan amount; and These are the maximum and minimum loan amounts in the system, used to standardize differences in loan amounts.

[0124] The ratio used to measure the difference between the actual loan amount and the ideal loan amount is between 0 and 1, reflecting the degree of matching of the loan amount. The smaller the difference, the higher the fit.

[0125] Repayment period matching Represented as: ;

[0126] in, It is a product The repayment period; This is user x's ideal repayment period. and These are the maximum and minimum repayment periods in the system, used for standardization. By standardizing the differences in repayment terms, the degree of matching between repayment terms is assessed. The closer the repayment term is to the user's ideal term, the higher the fit.

[0127] The preference values ​​for different loan products among users in each target cluster are predicted based on the matching degree model, and are expressed as follows:

[0128]

[0129] in, and It's about adjusting parameters. Indicates based on the first Historical lending behavior of each target cluster and the first Compatibility between loan products.

[0130] Preference values ​​can be used to analyze user purchasing behavior and market trends, helping businesses understand which product features are most popular or which loan conditions need adjustment. By analyzing the preference values ​​of different user groups, businesses can identify changes in market demand and adjust product strategies or develop new loan products to meet market needs. Preference values ​​can also serve as an indicator for monitoring system performance, helping businesses evaluate the effectiveness of the recommendation system and make necessary adjustments. Regularly checking the correlation between preference values ​​and actual user choices and satisfaction assesses the accuracy and effectiveness of the recommendation algorithm. Based on this data, algorithm parameters can be optimized or user interface design improved. Preference values ​​can also help assess potential credit risk, ensuring that loan product recommendations comply with regulatory requirements. This ensures that high-risk products are not over-recommended to users with lower credit ratings, thereby managing credit risk and complying with relevant financial and credit regulations.

[0131] In summary, by taking a more refined approach to users' credit history and historical behavior, this invention's matching model can more accurately reflect users' actual needs and preferences, thereby improving the relevance of recommendations and user satisfaction. User experience will be enhanced by receiving more personalized recommendations and promotional information. This not only increases user trust in the recommendation system but may also increase user interaction with the platform. Precise matching and personalized marketing strategies will increase user acceptance of recommended loan products, thus improving conversion rates. Simultaneously, by providing products that meet user expectations, user retention rates can be improved. Using machine learning techniques to dynamically adjust weights and strategies makes the system more flexible, adapting to changes in market and user behavior, thereby maintaining the system's competitiveness and efficiency.

[0132] In step S04 of this embodiment, promotional information is customized based on sales targets and market demand, wherein the promotional information is represented as follows:

[0133]

[0134] in, The corresponding descriptive text is generated based on the user's clustering tags. For products In time Discount rate, For products Promotion period, To customize promotional information;

[0135] in, The corresponding introductory text is generated based on the user's credit rating and financial behavior. For products In time Discount rate, For products Promotion period, To customize promotional information;

[0136] To generate personalized intros based on users' credit ratings and financial behavior, we can employ a systematic approach that combines specific financial data with behavioral patterns to develop more targeted and appealing promotional messages. The detailed steps are as follows:

[0137] First, collect detailed financial data about the user, including:

[0138] Credit rating: Obtained from cluster tags, reflecting the user's credit risk.

[0139] Consumption patterns: Analyze users' consumption records, including purchase type, frequency, and amount.

[0140] Loan history: This includes the type of loan, amount, repayment period, and repayment records.

[0141] Credit card usage frequency: Data on credit card activities, such as utilization rate, commonly used credit card types, and repayment behavior.

[0142] Based on the above data, users are divided into different categories, each representing a specific credit and consumption behavior pattern. These include:

[0143] High-credit consumers: Excellent credit rating, low credit card usage, and timely repayment.

[0144] Stability-seeking consumers: medium credit rating, stable consumption patterns, and a preference for long-term loans.

[0145] Risk-taking consumers: lower credit rating, high credit card usage, and frequent spending.

[0146] For each user group, corresponding introductory texts are developed based on the AI ​​model to ensure that the information aligns with the user's financial situation and preferences.

[0147] High-credit consumers: "As our valued customer, you are eligible for our exclusive offers. Excellent credit earns you lower interest rates, enhancing your financial management."

[0148] For consumers seeking stability: "Thank you for your continued trust. We offer specially tailored loan solutions for those who prefer prudent financial management. Safe and reliable, helping you move forward steadily."

[0149] For adventurous consumers: "Looking for flexible loan options? Our new products might be just what you're looking for, adapting to your ever-changing financial needs and helping you seize every opportunity."

[0150] The discount rate in the initial promotional information is set based on the product's sales speed and target sales speed in the previous preset period, expressed as:

[0151]

[0152] in, and These are the minimum and maximum discount rates, respectively. Based on the current sales speed of the product, The target sales speed for the product;

[0153] The promotion period in the initial promotional information is set based on the product's seasonal demand and expected market response, as follows: ;

[0154] in, and These are the minimum and maximum values ​​of the promotion period, respectively. Based on the current sales speed of the product, The target sales speed for the product;

[0155] Discount rates and promotional periods are adjusted regularly through a feedback mechanism.

[0156] Get the promotional performance value within the previous preset period. The predicted promotional effect value within the preset period is expressed as:

[0157]

[0158] in, For prediction functions;

[0159] Calculate the predicted promotional influencing factors within the current preset period. , represented as:

[0160]

[0161] in, , and These are adjustment parameters, which respectively adjust the influence weights of the logarithm of the predicted promotional effect value, the discount rate, and the limited-time information;

[0162] It is a logarithmic transformation of the predicted promotional effect value to smooth the effect and avoid the influence of extreme values;

[0163] It is the discount rate of the product in the current period;

[0164] It indicates the urgency of the promotion, that is, the reciprocal of the promotion period. The shorter the period, the higher the urgency.

[0165] To predict the effectiveness of a promotion, the first step is to obtain the promotion effectiveness values ​​from the previous preset period and even earlier periods based on historical data, such as... Wait, use prediction function The prediction function can be a time series analysis method (such as ARIMA, exponential smoothing), regression analysis, or a more complex machine learning model (such as random forest, neural network). The prediction function uses historical promotional performance values ​​to predict the performance value for the current period. Data preprocessing is performed according to the model's needs, such as data cleaning, handling missing values, and handling outliers.

[0166] In summary, targeted descriptions and promotional information can significantly increase user attention and engagement with marketing campaigns. Personalized promotional messages are more likely to resonate with target users, thereby improving response rates and user satisfaction. By setting precise discount rates and promotional periods, businesses can more effectively manage inventory and cash flow, optimizing the allocation of marketing resources. This not only helps control marketing costs but also maximizes sales revenue through precisely targeted promotional activities. Predicting the effectiveness of promotional activities using historical data makes marketing decisions more data-driven and results-oriented. Guiding resource allocation through predicted promotional influencing factors allows marketing resources to be focused on activities most likely to generate the greatest return. Accurate forecasting and effective resource allocation will improve the overall effectiveness of promotional activities, increase sales, and improve ROI.

[0167] In step S05 of this embodiment, the matching degree and marketing influence factor are combined to calculate the recommendation score of users in each target cluster for each loan product. For each product The recommended score is expressed as , represented as:

[0168]

[0169] and These are model parameters, representing the weights of different factors in the recommendation score, and can be optimized using training data.

[0170] For machine learning models, effective feature engineering is needed to extract the most informative features for the recommendation system from historical data.

[0171] Features include, but are not limited to:

[0172] User behavior characteristics include purchase history, click history, and browsing time.

[0173] Product characteristics: such as product category, price, and past sales performance.

[0174] Time factors: such as seasonality, special holiday effects, etc.

[0175] Promotional characteristics: types, frequency, and discount depth of historical promotional activities.

[0176] In this embodiment, factorization machines (FM) are applied to optimize the recommendation score model, which can be expressed as:

[0177]

[0178] in, It is a global bias term; It is the number of features; It is the first The weights of each feature; It is the input vector (features), including user and product features; Representation of features and characteristics The dot product in the latent space is used to capture the interaction between features; and They are the first and The latent vector of each feature.

[0179] To evaluate the generalization ability and accuracy of the factorization machine, cross-validation was used. In cross-validation, the dataset was divided into... There are three subsets of equal size. Each subset is used as the test set in turn, and the rest... A subset is used as the training set. This process is repeated. Each time, the model's performance metrics are calculated, and the average results are expressed as follows:

[0180]

[0181] The performance metrics include accuracy, root mean square error, precision, and recall. The model parameters include the dimension of the latent vector and the regularization term. These parameters are optimized using hyperparameter tuning techniques such as grid search to find the best model settings.

[0182] A comprehensive score for each loan product is calculated based on the recommendation score. , represented as:

[0183]

[0184] Based on a comprehensive score, loan products are categorized into different types, including "high priority," "medium priority," and "low priority," with high priority being... Medium priority Low priority is , and A preset scoring threshold is used to distinguish products with different priorities;

[0185] High-priority loan products have the highest overall scores and are expected to generate the greatest marketing return. Allocate the highest percentage of your advertising budget to multi-channel advertising such as television, internet, social media, and out-of-home advertising. Organize large-scale promotional campaigns such as special discounts, limited-time offers, gifts, or loyalty rewards. Utilize data-driven personalized marketing tools, such as customized emails, SMS campaigns, and app push notifications, based on known customer preferences. Provide priority customer service, including dedicated line support and a dedicated account manager.

[0186] The mid-priority loan product has a moderate overall score, possessing some market appeal but insufficient to compete with high-priority products. Allocate a moderate advertising budget, primarily focusing on cost-effective channels such as online advertising and email marketing. Conduct moderate-scale promotional activities, such as buy-one-get-one-free or small cash rebates. Strengthen social media engagement to increase brand online visibility and engagement. Increase research efforts in the target market to better adjust product and marketing strategies.

[0187] Low-priority loan products have lower overall credit scores and limited market appeal. Minimize investment by advertising only through very low-cost channels, such as organic social media posts. Limit promotional activities to essential market maintenance activities, such as maintaining existing customer relationships. Regularly evaluate the market performance of these products and consider whether product features need to be adjusted or phased out. Consider reallocating resources to higher-potential high-priority or mid-priority products.

[0188] This strategy ensures that marketing resources are used most effectively, while simultaneously improving the market performance and profitability of each type of loan product. This tiered resource allocation method also helps the company flexibly adjust its market strategies and respond quickly to market changes.

[0189] At the same time, select each target cluster The top n products with the highest recommendation scores are then pushed to the device in a personalized way.

[0190] In this embodiment, KPIs related to marketing strategies are collected, and the effectiveness of each marketing strategy is evaluated using these KPIs. The parameters of the marketing strategies are adjusted based on changes in the KPIs, including:

[0191] Collect marketing-related KPIs, including metrics such as click-through rate, conversion rate, and user engagement. Use statistical models to correlate marketing strategies with changes in these KPIs, and represent the results as follows:

[0192]

[0193] in, This indicates the change in KPIs due to marketing strategies. This is the point in time before the marketing strategy is implemented;

[0194] Define a policy adjustment function to adjust the policy based on the base The result updates the parameters of the marketing strategy, expressed as: ;

[0195] in, In time For products Marketing strategy parameter vector, It's the learning rate. It is a loss function Regarding strategy parameters The gradient of the loss function measures the difference between the change in KPI and the policy objective.

[0196] The loss function can be designed as a function of the difference between the KPI and the target, expressed as:

[0197]

[0198] in, It is the target KPI value. This is the current KPI value.

[0199] Based on the above parameter update rules, the advertising budget and promotion parameters will be adjusted as follows:

[0200] If conversion rates increase significantly, increase the advertising budget, and vice versa, as shown below:

[0201]

[0202] in, It is an adjustment factor that determines how sensitive the budget is to changes in conversion rate.

[0203] Adjusting the discount rate to respond to changes in sales volume is expressed as:

[0204]

[0205]

[0206] Similarly, the promotion duration can be adjusted to respond to increases or decreases in sales, as shown below:

[0207]

[0208]

[0209] in, and These are adjustment coefficients that control the response of discount rates and promotion duration to sales growth and inventory levels, respectively.

[0210] The parameters of promotional strategies can also be dynamically adjusted using reinforcement learning models. In the construction of reinforcement learning models:

[0211] Definition of environment: The environment is the behavior of the market and consumers, which is reflected in changes in KPIs (such as click-through rate, conversion rate, sales, etc.).

[0212] Definition of an agent: A system used to execute marketing activities and select parameters for promotional strategies (such as discount rates, promotion periods, etc.).

[0213] Definition of State: State refers to the environmental information that an agent considers when making decisions, such as current market conditions, consumer behavior, and competitor activities. This information can be represented by vectors. It means that among them Indicates the current time.

[0214] Definition of Action: An action is a decision that an agent can perform, such as setting a specific discount rate. or promotion period Behavior can be represented as ,in Indicates the moment of decision-making.

[0215] Reward function: Reward function Measure and take action The immediate rewards gained are such as increased sales or improved conversion rates. Rewards are calculated based on improvements in KPIs.

[0216] Policy: It is a mapping from the current state to the action taken. The goal of reinforcement learning is to learn the optimal policy to maximize the cumulative reward in the future.

[0217] In this embodiment, a Markov Decision Process (MDP) is used for modeling, with the goal of finding a policy that maximizes the expected cumulative discounted reward starting from any initial state, expressed as:

[0218]

[0219] in, It is a discount factor used to balance the importance of immediate rewards and future rewards.

[0220] Q-learning is used as the policy optimization algorithm to directly learn the value function of state-action pairs. , represented as:

[0221]

[0222] Update status information using real-time data. And reassess and adjust strategies based on the latest environmental changes. Continuously monitor strategy performance and adjust behavior based on feedback. and learning parameters.

[0223] By monitoring specific KPIs, such as click-through rate (CTR), conversion rate (CR), and user engagement, the actual effectiveness of each marketing strategy can be accurately evaluated. This real-time feedback allows marketing teams to quickly identify which strategies are effective and which need improvement, thereby increasing overall marketing accuracy and effectiveness.

[0224] By KPI change Marketing strategy parameters Linking resources allows for smarter adjustments to marketing efforts. If a product's conversion rate increases, the advertising budget for that product may increase, and vice versa. This dynamic adjustment helps ensure that resources are always allocated to areas most likely to generate the highest returns.

[0225] Defining strategy adjustment functions makes marketing decisions more data-driven. This approach not only reduces the influence of bias and subjective judgment but also makes strategy adjustments more scientific and systematic. Marketing strategy parameters include, but are not limited to, budget allocation, target audience, content strategy, and channel strategy.

[0226] Therefore, this invention improves a company's responsiveness to market changes through real-time data analysis and immediate strategy adjustments. It helps to significantly increase the return on each marketing investment by continuously optimizing marketing strategies and adjusting investment directions. In a highly competitive market environment, it enables rapid adaptation to changes in consumer behavior and preferences.

[0227] This invention also provides a data processing system for intelligent marketing products, based on the aforementioned data processing method for intelligent marketing products, comprising the following modules:

[0228] Clustering module: Acquires financial behavior data of target users, including credit card usage records, consumption records, historical loan records and historical repayment records. Converts the raw data into vector features and forms a set through feature extraction function. Uses clustering algorithm to divide the vector feature set into multiple clusters. Labels users with corresponding clusters based on the clustering results.

[0229] Screening module: Calculates the value score of each cluster using a preset value assessment model, identifies candidate clusters with value scores greater than a preset threshold as target clusters, and removes non-target clusters from the dataset;

[0230] Matching degree calculation module: It evaluates the comprehensive matching degree between users in each target cluster and multiple loan products through a preset matching degree model, and predicts the preference value of users in each target cluster for different loan products based on the matching degree model. The matching degree model is established based on the historical loan behavior of users in each target cluster.

[0231] Promotion module: Customizes promotional information based on the sales target of the loan product within a preset period. The promotional information includes discount information and time-limited information. Predicts the promotional effect value of the loan product within the preset period and calculates the promotional impact factor based on the promotional effect value.

[0232] Marketing Strategy Module: Combining matching degree and marketing influence factor, calculate the recommendation score of users in each target cluster for each loan product, calculate the comprehensive score of each loan product based on the recommendation score, divide the loan products into different priority categories based on the comprehensive score, and select the corresponding marketing strategy according to the priority category;

[0233] Feedback Update Module: Collects KPIs related to marketing strategies within a preset period, uses a data processing platform to evaluate the effectiveness of each marketing strategy, and adjusts the parameters of the marketing strategies for the next preset period based on changes in KPIs.

[0234] This invention dynamically adjusts recommendation and marketing strategies by monitoring changes in market conditions and user behavior in real time. This flexibility enables businesses to quickly adapt to market changes and maintain competitiveness. The system operates based on data-driven decision support, from cluster analysis of user data to KPI evaluation of marketing strategies. This evidence-based approach reduces the influence of chance and subjective judgment, improving the quality and efficiency of business decisions. Through accurate credit rating and financial needs analysis, the loan product recommendation system can effectively reduce non-performing loans and credit losses. Correct credit assessment helps businesses avoid lending to customers with poor credit, thereby reducing risk. In summary, this invention not only improves the efficiency and accuracy of loan product recommendation systems but also helps businesses optimize resource allocation, improve user satisfaction and loyalty, and ultimately achieve business growth and profit maximization while maintaining market competitiveness.

[0235] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A data processing method for an intelligent marketing product, characterized in that, Includes the following steps: Step S01: Obtain the financial behavior data of the target user, which includes credit card usage records, consumption records, historical loan records and historical repayment records. Convert the raw data into vector features through feature extraction and form a set. Use a clustering algorithm to divide the vector feature set into multiple clusters. Label the user with the corresponding cluster label according to the clustering results. Step S02: Calculate the value score of each cluster using a preset value assessment model, identify candidate clusters with value scores greater than a preset threshold as target clusters, and remove non-target clusters from the dataset; Step S03: Evaluate the overall matching degree between users in each target cluster and multiple loan products through a preset matching degree model, and predict the preference value of users in each target cluster for different loan products based on the matching degree model. The matching degree model is established based on the historical loan behavior of users in each target cluster. Step S04: Customize promotional information based on the sales target of the loan product within a preset period. The promotional information includes discount information and time-limited information. Predict the promotional effect value of the loan product within the preset period and calculate the promotional impact factor based on the promotional effect value. Step S05: Combine matching degree and marketing influence factor to calculate the recommendation score of users in each target cluster for each loan product. Calculate the comprehensive score of each loan product based on the recommendation score. Based on the comprehensive score, classify the loan products into different priority categories and select the corresponding marketing strategy according to the priority category. Step S06: Collect KPIs related to the marketing strategies within the preset period to evaluate the marketing effectiveness of each marketing strategy, and adjust the parameters of the marketing strategies for the next preset period based on the marketing effectiveness values; In step S04, promotional information is customized based on sales targets and market demand, wherein the promotional information is represented as follows: ； in, The corresponding descriptive text is generated based on the user's clustering tags. For products In time Discount rate, For products Promotion period, To customize promotional information; Get the promotional performance value within the previous preset period. The predicted promotional effect value within the preset period is expressed as: ,in, For the prediction function, It represents the promotional effectiveness value from an earlier period; Calculate the predicted promotional influencing factors within the current preset period. , represented as: ； in, , and It refers to adjusting parameters.

2. The data processing method for the intelligent marketing product according to claim 1, characterized in that, In step S01, a clustering algorithm is used to divide the vector feature set into multiple clusters, and the user is labeled with the corresponding cluster label based on the clustering results, including: Obtain the user's raw dataset Through feature extraction function Original data Convert to vector features ,in Indicates the first The data for each user, specifically represented by the feature vector, is as follows: ,in It is the number of features; Use clustering algorithms to group feature vector sets Divided into K clusters; Clustering results are represented as functions ,function Data for each user Mapping to cluster labels , represented as: ; The K-means clustering algorithm is applied to minimize the variance within each cluster, as follows: ,in It is the first One cluster, It is a cluster center; Each user is assigned a corresponding cluster label based on their cluster. This assignment is done using the CreditLevel function, as follows: 。 3. The data processing method for the intelligent marketing product according to claim 2, characterized in that, In step S02, the value assessment model is established based on the user's credit rating, historical response rate, and purchasing power, and is expressed as follows: ； in, , , These are the weighting coefficients. , and This represents the average of the credit rating, response rate, and purchasing power of all users within a cluster. The value score of each cluster is calculated using a pre-defined value assessment model. Candidate clusters with value scores greater than a pre-defined threshold are identified as target clusters. Non-target clusters in the dataset are then removed. This is represented as follows: ； in, This represents the new user dataset after removing users who were not in the target cluster. This represents the set of all users in the non-target cluster.

4. The data processing method for the intelligent marketing product according to claim 3, characterized in that, In step S03, a matching degree model is defined to evaluate the matching degree between a user and multiple loan products, including: definition A collection of loan products. Each loan product It has multiple attributes, including interest rate. Loan amount and repayment period ; By using the comprehensive matching degree function Evaluation of the first The target clustering and the first The matching degree between individual loan products, specifically expressed by the matching degree function, is as follows: ； in, To calculate the matching degree function for interest rate, loan amount, and repayment period, It is the weight used to calculate the interest rate, loan amount, and repayment period; Interest rate adaptation Represented as: ; It is a positive adjustment coefficient used to control the sensitivity of the matching function; Loan amount matching Represented as: ; in, It is clustering Ideal loan amount; and These are the maximum and minimum loan amounts in the system, respectively. Repayment period matching Represented as: ; in, and These are the maximum and minimum repayment periods in the system; The preference values ​​for different loan products among users in each target cluster are predicted based on the matching degree model, and are expressed as follows: .

5. The data processing method for the intelligent marketing product according to claim 1, characterized in that, Collect KPIs related to marketing strategies, use KPIs to evaluate the effectiveness of each marketing strategy, and adjust the parameters of the marketing strategies based on changes in KPIs, including: Collect marketing-related KPIs, including click-through rate, conversion rate, and user engagement. Use a statistical model to correlate marketing strategies with changes in these KPIs, and represent the results as follows: ； in, This indicates the change in KPIs due to marketing strategies. This is the point in time before the marketing strategy is implemented; Define a policy adjustment function to adjust the policy based on the base The result updates the parameters of the marketing strategy, expressed as: ; in, In time For products Marketing strategy parameter vector, It's the learning rate. It is a loss function Regarding strategy parameters The gradient of the loss function measures the difference between the change in KPI and the policy objective.

6. The data processing method for the intelligent marketing product according to claim 1, characterized in that, The discount rate in the initial promotional information is set based on the product's sales speed and target sales speed in the previous preset period, expressed as: ； in, and These are the minimum and maximum discount rates, respectively. Based on the current sales speed of the product, The target sales speed for the product; The promotion period in the initial promotional information is set based on the product's seasonal demand and expected market response, as follows: ; in, and These are the minimum and maximum values ​​of the promotion period, respectively. Based on the current sales speed of the product, The target sales speed for the product; discount rates and promotional periods are adjusted regularly through a feedback mechanism.

7. A data processing system for an intelligent marketing product, based on the data processing method for the intelligent marketing product according to any one of claims 1-6, characterized in that, Includes the following modules: Clustering module: Acquires financial behavior data of target users, including credit card usage records, consumption records, historical loan records and historical repayment records. Converts the raw data into vector features and forms a set through feature extraction function. Uses clustering algorithm to divide the vector feature set into multiple clusters. Labels users with corresponding clusters based on the clustering results. Screening module: Calculates the value score of each cluster using a preset value assessment model, identifies candidate clusters with value scores greater than a preset threshold as target clusters, and removes non-target clusters from the dataset; Matching degree calculation module: It evaluates the comprehensive matching degree between users in each target cluster and multiple loan products through a preset matching degree model, and predicts the preference value of users in each target cluster for different loan products based on the matching degree model. The matching degree model is established based on the historical loan behavior of users in each target cluster. Promotion module: Customizes promotional information based on the sales target of the loan product within a preset period. The promotional information includes discount information and time-limited information. Predicts the promotional effect value of the loan product within the preset period and calculates the promotional impact factor based on the promotional effect value. Marketing Strategy Module: Combining matching degree and marketing influence factor, calculate the recommendation score of users in each target cluster for each loan product, calculate the comprehensive score of each loan product based on the recommendation score, divide the loan products into different priority categories based on the comprehensive score, and select the corresponding marketing strategy according to the priority category; Feedback Update Module: Collects KPIs related to marketing strategies within a preset period, uses a data processing platform to evaluate the effectiveness of each marketing strategy, and adjusts the parameters of the marketing strategies for the next preset period based on changes in KPIs.

Images