Data processing method and system for intelligent marketing products

Abstract

The invention relates to a data processing method for intelligent marketing products, and the method comprises the steps: obtaining the financial behavior data of a target user, carrying out the clustering through a machine learning algorithm, and selecting a target cluster through a preset threshold value; and defining a matching degree model to evaluate the comprehensive matching degree between the user and different loan products, and predicting the preference value of the target cluster. Then, sales promotion information is customized according to the sales target, and marketing influence factors of the sales promotion information are calculated; and calculating a recommendation score of each user for each loan product according to the matching degree and the marketing influence factor, calculating a comprehensive score according to the recommendation scores, dividing the loan products into priorities, and selecting a corresponding marketing strategy. And finally, collecting KPIs (Key Performance Indicators) related to marketing strategies, evaluating effects and adjusting strategy parameters. According to the invention, through accurate user data analysis and a personalized marketing strategy, the recommendation accuracy and marketing effect of loan products are improved, and the satisfaction and loyalty of users are enhanced.

Description

Technical Field

[0001] The present 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 Art

[0002] In the process of designing consumer finance products, institutions often adopt differentiated pricing strategies for different customer groups in order to balance profit requirements and user acceptance (indicators such as expenditure rate and loan size). The most common method is to set corresponding interest rates and credit limits based on the estimated default rates, comprehensive costs and profit requirements of different customer groups. In actual business, the credit limit, default rate, interest rate and user acceptance (hereinafter referred to as conversion rate) are mutually influential. The higher the interest rate, the higher the user risk, and the user conversion rate will also decrease, and vice versa. The higher the credit limit, the higher the conversion rate, and the default rate may also increase. At present, most financial institutions only evaluate the high or low risk of users to decide whether to approve or reject users. Instead of combining credit limit and interest rate, the conversion rate will be too low, and too much customer acquisition cost will be wasted. Low-risk users cannot maximize their value, and medium- and high-risk users do not organically combine credit limit, default, interest rate and conversion rate to calculate the cost of assets and the profit of users after lending. Therefore, we hope to have a set of risk pricing and classification marketing strategies that can further segment the customer base, match different product pricing based on the risks of different categories of users, and then maximize user conversion rates based on marketing strategies, find the profit balance point, and improve business profitability. Summary of the invention

[0003] The present invention provides a data processing method for intelligent marketing products, which can improve customers' financial management ability and acceptance rate and reduce risk factors. At the same time, by reducing risk factors, customers' financial management ability and acceptance rate can be improved in reverse, thereby finding a profit balance point and improving business profitability.

[0004] The present invention provides a data processing method for an intelligent marketing product, comprising the following steps:

[0005] Step S01: obtaining the target user's financial behavior data, which includes credit card usage records, consumption records, historical loan records, and historical repayment records; converting the original data into vector features through feature extraction and forming a set; using a clustering algorithm to divide the vector feature set into multiple clusters; and labeling the user with a corresponding cluster label according to the clustering result;

[0006] Step S02: Calculate the value score of each cluster through a preset value evaluation model, determine the candidate cluster with a value score greater than a preset threshold as the target cluster, and remove non-target clusters in the data set;

[0007] Step S03: Evaluate the comprehensive matching degree between the users of each target cluster and multiple loan products through a preset matching degree model, and predict the preference value of the users of each target cluster for different loan products according to the matching degree model, wherein the matching degree model is established according to the historical loan behavior of the users of each target cluster;

[0008] Step S04: Customizing promotion information according to the sales target within a preset period of the loan product, the promotion information including discount information and time-limited information, predicting the promotion effect value of the loan product within the preset period, and calculating the promotion impact factor according to the promotion effect value;

[0009] Step S05: Calculate the recommendation score of each target cluster user for each loan product based on the matching degree and the marketing influence factor, 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;

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

[0011] Preferably, in step S01, a clustering algorithm is used to divide the vector feature set into multiple clusters, and corresponding cluster labels are labeled for users according to the clustering results, including:

[0012] Get the user's original data set X = {x1, x2, x3, ..., x n}, the original data x is extracted by the feature extraction function f i Convert to vector feature f(x i ), where x i represents the data of the i-th user, and the feature vector is specifically expressed as f(x i )=[f1(x i ),f2(x i ),…,f m (x i )] T , where m is the number of features;

[0013] Use clustering algorithm to cluster the feature vector set {f(x1),f(x2),…,f(x N )} is divided into K clusters;

[0014] The clustering result is expressed as a function φ, which transforms each user data x i Mapped to cluster label φ(x i ), expressed as:φ(x i )=k,where k∈{1,2,…,K};

[0015] The clustering algorithm K-means is applied to minimize the variance within each cluster, expressed as: Where S K is the kth cluster, μ k is the cluster center;

[0016] Each user is assigned a corresponding cluster label according to the cluster to which he belongs, and each cluster label is assigned through the CreditLevel function, which is expressed as:

[0017] r i (t) = CreditLevel(φ(x i ))=CreditLevel(φ(x i )).

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

[0019] V(C k )=α1·CreditLevel(C k )+α2·ResponseRate(C k )+α3·PurchasingPower(C k )

[0020] Among them, α1, α2, and α3 are weight coefficients, respectively, and CreditLevel(C k )、ResponseRate(C k ) and PurchasingPower(C k ) represents the average of credit rating, response rate and purchasing power of all users in the cluster;

[0021] The value score of each cluster is calculated through the preset value evaluation model, and the candidate clusters with value scores greater than the preset threshold are determined as target clusters, and non-target clusters in the data set are removed, which is expressed as:

[0022]

[0023] Among them, X new represents the new user data set after removing non-target clustering users,

[0024] Represents the set of users in all non-target clusters.

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

[0026] Define P as a set of loan products. Each loan product p in the loan product set P j There are multiple properties, including the interest rate r j , loan amount j , and repayment period t j ;

[0027] Through the comprehensive matching function m(C k ,p j ) evaluates the matching degree between the kth target cluster and the jth loan product. The matching degree function is specifically expressed as:

[0028] m(C k ,p j )=w1f1(C k ,p j )+w2f2(C k ,p j )+w3f3(C k ,p j )

[0029] Among them, f1, f2, f3 are matching functions for calculating interest rate, loan amount and repayment period, and w1, w2, w3 are weights for calculating interest rate, loan amount and repayment period;

[0030] The interest rate adaptation f1(C k ,p j ) is expressed as: f1(C k ,p j )=exp(-α(r j -r expected (φ(C k ))));

[0031] Loan amount adaptation f2(C k ,p j ) is expressed as:

[0032] Repayment term adaptation f3(C k ,p j ) is expressed as:

[0033] According to the matching model, the preference value of users in each target cluster for different loan products is predicted, which is expressed as:

[0034] P(C k ,p j )=β·m(C k ,p j )+γ·H(C k ,p j )

[0035] Among them, β and γ are adjustment parameters, H(x i ,p j ) represents the compatibility between the historical loan behavior based on the i-th target cluster and the j-th loan product.

[0036] Preferably, in step S04, promotion information is customized according to sales targets and market demand, wherein the promotion information is represented as:

[0037] m i (p j ,t)=Intro(u i )+"save"+d(p j ,t)×100+"Next loan! The discount ends at"+τ(p j ,t)+"day

[0038] Among them, Intro(u i ) is the corresponding introduction word generated based on the user's clustering label, d(p j ,t) is the discount rate of product j at time t, τ(p j ,t) is the promotion period of product j, m i (p j ,t) to customize promotional information;

[0039] Get the promotion effect value E(p j ,t-1), predict the promotion effect value within this preset period, expressed as:

[0040]

[0041] Where, f is the prediction function;

[0042] Calculate the predicted promotion impact factor ψ(p j ,t), expressed as:

[0043]

[0044] Among them, β, γ and δ are adjustment parameters.

[0045] Preferably, the matching degree and marketing influence factor are combined to calculate the recommendation score of each target cluster user for each loan product. The recommendation score of each cluster k for each product j is expressed as R(u i ,p j ,t), expressed as:

[0046] R(u i ,p j ,t)=ω1·m(C k (t),p j)×ω2·ψ(C k ,p j ,t)

[0047] Calculate the comprehensive score S(p) of each loan product based on the recommendation score j ,t), expressed as:

[0048]

[0049] Based on the comprehensive score, loan products are divided into different categories, including "high priority", "medium priority" and "low priority", with high priority being S(p j ,t)≥Threshold high , medium priority is Threshold low ≤S(p j ,t) <Threshold high , low priority is S(p j ,t) <Threshold low , Threshold high and Threshold low It is a preset score threshold used to distinguish products of different priorities;

[0050] At the same time, the n products with the highest recommendation scores corresponding to each target cluster k are selected for personalized push.

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

[0052] Collect marketing-related KPIs, including click-through rate, conversion rate, and user engagement. Use statistical models to correlate marketing strategies with KPI changes, expressed as:

[0053] ΔKPI(p j ,t)=KPI(p j ,t)-KPI(p j ,t0)

[0054] Among them, ΔKPI(p j ,t) represents the KPI change caused by the marketing strategy, and t0 is the time point before the implementation of the marketing strategy;

[0055] Define a policy adjustment function to adjust the policy according to the base ΔKPI(p j ,t) results to update the parameters of the marketing strategy, expressed as:

[0056] Among them, θ(p j,t+1) is the product p at time t j The marketing strategy parameter vector, η is the learning rate, is the loss function With respect to the gradient of the policy parameters θ, the loss function measures the difference between the KPI change and the policy target.

[0057] Preferably, the setting of the discount rate in the initial promotion information is based on the sales speed of the product in the previous preset period and the target sales speed, expressed as:

[0058]

[0059] Among them, d min and d max are the minimum and maximum discount rates, v(p j ,t) is the current sales speed of the product, v * (p j ) is the target sales velocity of the product;

[0060] The promotion period in the initial promotion information is set based on the seasonal demand of the product and the expected market response, expressed as:

[0061] Among them, τ min and τ max are the minimum and maximum values ​​of the promotion period, S(p j ,t) is the current sales speed of the product, T(p j ,t) is the target sales speed of the product;

[0062] Discount rates and promotion durations are regularly adjusted through a feedback mechanism.

[0063] The present invention also provides a data processing system for intelligent marketing products, based on the data processing method for intelligent marketing products described above, characterized in that it includes the following modules:

[0064] Clustering module: obtains the target user's financial behavior data, including credit card usage records, consumption records, historical loan records, and historical repayment records. The original data is converted into vector features through feature extraction functions to form a set. The vector feature set is divided into multiple clusters using a clustering algorithm. The user is labeled with a corresponding cluster label based on the clustering results.

[0065] Screening module: Calculate the value score of each cluster through the preset value evaluation model, determine the candidate clusters with value scores greater than the preset threshold as target clusters, and remove non-target clusters in the data set;

[0066] Matching degree calculation module: evaluates the comprehensive matching degree between the users of each target cluster and multiple loan products through a preset matching degree model, and predicts the preference value of the users of each target cluster for different loan products according to the matching degree model, wherein the matching degree model is established according to the historical loan behavior of the users of each target cluster;

[0067] Promotion module: customizes promotion information according to the sales target within the preset period of the loan product, including discount information and time-limited information, predicts the promotion effect value of the loan product within the preset period, and calculates the promotion impact factor according to the promotion effect value;

[0068] Marketing strategy module: Combine the matching degree and marketing influence factor to calculate the recommendation score of each target cluster user 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 the marketing strategies within the preset period, uses the data processing platform to evaluate the effectiveness of each marketing strategy, and adjusts the parameters of the marketing strategies within the next preset period based on the changes in KPIs.

[0070] The beneficial effects of the present invention are as follows: the present invention uses a clustering algorithm to analyze the financial behavior data of users, and the method can accurately group users to ensure that marketing activities are targeted at user groups with similar financial behaviors and needs. Accurate target division helps marketing activities to be more effective, because promotion information and product recommendations are customized based on the actual needs and historical behaviors of users. By evaluating the value score of each cluster and eliminating those with low value, marketing resources can be more effectively allocated to those target clusters that are most likely to generate high returns. The ROI of marketing activities is increased and unnecessary expenses on low-return groups are reduced. The matching model predicts users' preferences for different loan products based on their historical loan behaviors. This means that users are more likely to receive product information that they are actually interested in, thereby improving user satisfaction and possibly increasing user loyalty to the brand. By monitoring sales targets and promotion effects and then calculating promotion impact factors, the marketing team can continuously adjust and optimize its strategy. This dynamic adjustment can respond to market changes and ensure that marketing activities always remain efficient and effective. The KPIs related to the marketing strategy are collected and the marketing strategy is adjusted according to these indicators, so that the decision-making process is completely based on actual data and empirical results. This data-based approach reduces the room for bias and guesswork, making marketing decisions more scientific and precise. Since adjustments to marketing strategies and promotions are based on real-time or periodic KPI feedback, organizations can respond quickly to market changes, quickly take advantage of new market opportunities or adjust strategies to face challenges. This can significantly enhance the market competitiveness of financial institutions, achieve higher economic benefits, and improve customer experience, further driving business growth and market expansion. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in 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 ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work:

[0072] Figure 1 It is a flowchart of a data processing method for an intelligent marketing product according to an embodiment of the present invention. DETAILED DESCRIPTION

[0073] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the following will be described clearly and completely in combination with the technical solution in the embodiments of the present invention. Obviously, the described embodiments are partial embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to 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, comprising the following steps:

[0075] Step S01: obtaining the target user's financial behavior data, which includes credit card usage records, consumption records, historical loan records, and historical repayment records; converting the original data into vector features through feature extraction and forming a set; using a clustering algorithm to divide the vector feature set into multiple clusters; and labeling the user with a corresponding cluster label according to the clustering result;

[0076] Step S02: Calculate the value score of each cluster through a preset value evaluation model, determine the candidate cluster with a value score greater than a preset threshold as the target cluster, and remove non-target clusters in the data set;

[0077] Step S03: Evaluate the comprehensive matching degree between the users of each target cluster and multiple loan products through a preset matching degree model, and predict the preference value of the users of each target cluster for different loan products according to the matching degree model, wherein the matching degree model is established according to the historical loan behavior of the users of each target cluster;

[0078] Step S04: Customizing promotion information according to the sales target within a preset period of the loan product, the promotion information including discount information and time-limited information, predicting the promotion effect value of the loan product within the preset period, and calculating the promotion impact factor according to the promotion effect value;

[0079] Step S05: Calculate the recommendation score of each target cluster user for each loan product based on the matching degree and the marketing influence factor, 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;

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

[0081] The present invention uses a clustering algorithm to analyze the financial behavior data of users. The method can accurately group users and ensure that marketing activities are targeted at user groups with similar financial behaviors and needs. Accurate target division helps marketing activities to be more effective because promotion information and product recommendations are customized based on the actual needs and historical behaviors of users. By evaluating the value score of each cluster and eliminating those with low value, marketing resources can be more effectively allocated to those target clusters that are most likely to generate high returns. The ROI of marketing activities is increased and unnecessary expenses on low-return groups are reduced. The matching model predicts users' preferences for different loan products based on their historical loan behaviors. This means that users are more likely to receive product information that they are actually interested in, thereby improving user satisfaction and possibly increasing user loyalty to the brand. By monitoring sales targets and promotion effects and then calculating promotion impact factors, the marketing team can continuously adjust and optimize its strategy. This dynamic adjustment can respond to market changes and ensure that marketing activities always remain efficient and effective. Collecting KPIs related to marketing strategies and adjusting marketing strategies based on these indicators makes the decision-making process completely based on actual data and empirical results. This data-based approach reduces the space for bias and speculation and makes marketing decisions more scientific and accurate. Since the adjustment of marketing strategies and promotional activities is based on real-time or periodic KPI feedback, organizations can quickly respond to market changes, quickly take advantage of new market opportunities or adjust strategies to face challenges. Therefore, it can significantly enhance the market competitiveness of financial institutions, achieve higher economic benefits, improve customer experience, and further promote business growth and market expansion.

[0082] The financial behavior data of the user is obtained based on the user data submitted by the user when applying for a loan or other public channels, and a data storage device is used for data storage, and the user data is encrypted through homomorphic encryption dynamic segmentation technology; wherein each user's data includes identity data, credit history and financial records in recent time (such as within one year); a decentralized identity authentication framework is constructed using blockchain technology to store user data, and then an identity authentication smart contract is designed to verify the user's identity, and the user's credit score is dynamically updated based on the user's transaction history and external data. The data storage device specifically includes: using blockchain to create a decentralized identity database, user data is encrypted and stored on the blockchain, and the encryption key is passed to the authorized verification party through the key exchange protocol to ensure the secure storage and management of the key;

[0083] The present invention uses a data processing platform to process and analyze user data to support various decision-making steps. The main functions and features of the platform include:

[0084] The data processing platform uses machine learning algorithms to perform cluster analysis on the collected data to identify important features such as the user's credit rating. It aims to discover patterns and associations from the data and help annotate the user's clustering labels. The data processing platform supports the definition and execution of matching models to evaluate the matching between users and loan products, predict users' product preferences, and calculate the potential impact of promotional activities. The data processing platform has real-time data processing capabilities and can instantly receive and analyze users' credit ratings and financial needs information to quickly respond to market changes and user needs. The data processing platform uses a decision support system to calculate each user's recommendation score for each loan product, and classifies loan products into high, medium, and low priority based on these scores. Supports the formulation and optimization of personalized marketing strategies. The data processing platform also integrates performance monitoring tools to collect and analyze key performance indicators (KPIs) related to marketing strategies, and automatically adjusts the parameters of marketing strategies according to changes in KPIs to ensure the optimal effect of marketing activities.

[0085] Data processing platforms can be Apache Spark, Apache Hadoop, Google BigQuery, AmazonAWS, etc. The data processing platform should be able to efficiently process large amounts of data and support real-time data analysis. As the amount of data increases, the platform should be able to scale easily. The platform needs to support a variety of data analysis tools and languages, and have built-in or compatible 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. The data storage device supports structured, semi-structured, and unstructured data to ensure high data availability and disaster recovery capabilities.

[0087] In this embodiment, a data processing platform such as Apache Spark or Google BigQuery is used in combination with a data storage device such as Amazon S3 or Google Cloud Storage. This combination can provide powerful data processing capabilities, flexible storage options, and high scalability while ensuring high performance and security of the system.

[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 a corresponding cluster label according to the clustering result, including:

[0089] Get the user's original data set X = {x1, x2, x3, ..., x n}, the original data x is extracted by the feature extraction function f iConvert to vector feature f(x i ), where x i represents the data of the i-th user, and the feature vector is specifically expressed as f(x i )=[f1(x i ),f2(x i ),…,f m (x i )] T , where m is the number of features;

[0090] Use clustering algorithm to cluster the feature vector set {f(x1),f(x2),…,f(x N )} is divided into K clusters;

[0091] The clustering result is expressed as a function φ, which transforms each user data x i Mapped to cluster label φ(x i ), expressed as:φ(x i )=k,where k∈{1,2,…,K};

[0092] The clustering algorithm K-means is applied to minimize the variance within each cluster, expressed as: Where S K is the kth cluster, μ k is the cluster center;

[0093] Each user is assigned a corresponding cluster label according to the cluster to which he belongs, and each cluster label is assigned through the CreditLevel function, which is expressed as:

[0094] r i (t) = CreditLevel(φ(x i ))=CreditLevel(φ(x i )).

[0095] The extraction function f is a principal component analysis (PCA), which includes: standardizing (normalizing) each feature in the original data set to ensure that the mean of each feature is 0 and the standard deviation is 1. Based on the standardized data, the covariance matrix between the features is calculated. The covariance matrix reveals the linear relationship between different variables. The covariance matrix is ​​eigendecomposed to obtain eigenvalues ​​and corresponding eigenvectors. The eigenvector represents the direction of the principal component in the data, while the eigenvalue represents the importance of each principal component (the amount of variance explained). Multiple principal components are selected according to the eigenvalues, and the original data is projected using the selected principal components to obtain the reduced-dimensional data. The purpose of the CreditLevel function is to convert the user cluster labels obtained from the clustering algorithm into specific credit ratings. This process needs to be both intuitive and reliable to facilitate interpretation and verification. The following are detailed instructions and steps to show how to design, implement and verify this key function: The CreditLevel function accepts a cluster label as input and outputs a corresponding credit rating. These credit ratings can be classification labels such as "high credit", "medium credit", "low credit", or more specific credit scores.

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

[0097] Verify and adjust: Use historical data sets that contain users’ actual credit performance to verify the accuracy of the CreditLevel function. Compare the credit rating assigned by the model with the user’s subsequent credit performance to evaluate the accuracy and reliability of the credit rating prediction. Adjust the mapping logic of the CreditLevel function based on the verification results. For example, if users in a cluster show better credit behavior than expected, the credit rating of the cluster may need to be increased. Machine learning models can be introduced to automatically adjust the logic of credit rating determination to ensure that credit ratings dynamically reflect the latest credit risk trends.

[0098] Review and update the CreditLevel function regularly to ensure it is consistent with the latest credit risk models and market conditions. Monitor any feedback or complaints about the credit rating process to improve the transparency and fairness of the function. Through these steps, the CreditLevel function can not only provide accurate credit ratings for the loan recommendation system, but also its process is clear and explainable, and easy to verify and adjust through historical data.

[0099] Disclose the basis for determining the credit rating of each cluster, including the main variables and statistics considered. Use visualization tools to show the main characteristics of each cluster and how they map to specific credit ratings. Provide an interface or report that details why users are assigned to specific credit ratings and the key factors that may affect their credit ratings.

[0100] Therefore, through accurate cluster analysis, users' behavior patterns and financial status can be better understood, thereby more accurately assessing their credit rating. Based on users' cluster labels and credit ratings, the recommendation system can provide more personalized and relevant loan products. Understanding the user preferences and credit status of different clusters can help marketing teams target marketing activities more effectively, thereby reducing marketing costs and increasing conversion rates. When users receive loan product recommendations that are more suitable for their financial status and needs, their satisfaction may increase, thereby increasing their loyalty to the service provider. With more accurate credit ratings, lenders are able to better manage risks and formulate appropriate loan terms, thereby reducing default rates. In short, with accurate credit ratings through cluster analysis, the loan product recommendation system can not only improve operational efficiency, but also improve 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:

[0102] V(C k )=α1·CreditLevel(C k )+α2·ResponseRate(C k )+α3·PurchasingPower(C k )

[0103] Among them, α1, α2, and α3 are weight coefficients, respectively, and CreditLevel(C k )、ResponseRate(C k ) and PurchasingPower(C k ) represents the average of credit rating, response rate and purchasing power of all users in the cluster;

[0104] The value score of each cluster is calculated through the preset value evaluation model, and the candidate clusters with value scores greater than the preset threshold are determined as target clusters, and non-target clusters in the data set are removed, which is expressed as:

[0105]

[0106] Among them, X new represents the new user data set after removing non-target clustering users,

[0107] Represents the set of users in all non-target clusters.

[0108] The value assessment model is calculated by weighting and summing the average credit rating, response rate, and purchasing power of each cluster. The credit rating represents the credit reliability of the cluster users, the response rate reflects the degree of participation of the cluster users in previous marketing activities, and the purchasing power estimates the consumption capacity of the cluster users. The three weight coefficients correspond to the business importance of the credit rating, response rate, and purchasing power. The three weights are derived based on historical data analysis.

[0109] Calculate the value score for each cluster and set a threshold to compare 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 data set. Those clusters with value scores lower than the preset threshold are removed from the data set. Considering the dynamic changes in the market and user behavior, it is necessary to regularly update cluster information and recalculate value scores. This can be done quarterly, half-yearly or annually, depending on business needs and the speed of data changes.

[0110] By identifying and focusing on high-value user clusters, marketing resources are allocated more effectively, thereby reducing costs and improving the overall return on marketing activities. Focusing marketing efforts on clusters with high response rates can improve the overall response rate and participation of marketing activities. By understanding and meeting the specific needs of high-value clusters, the satisfaction and loyalty of these users can be improved. Using quantitative methods to evaluate the value of user groups makes decisions 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 realization of the company's business goals.

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

[0112] Define P as a set of loan products. Each loan product p in the loan product set P j There are multiple properties, including the interest rate r j , loan amount j , and repayment period t j ;

[0113] Through the comprehensive matching function m(C k ,p j ) evaluates the matching degree between the kth target cluster and the jth loan product. The matching degree function is specifically expressed as:

[0114] m(C k ,pj )=w1f1(C k ,p j )+w2f2(C k ,p j )+w3f3(C k ,p j )

[0115] Among them, f1, f2, f3 are matching functions for calculating interest rate, loan amount and repayment period, and w1, w2, w3 are weights for calculating 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 w1, w2, and w3 in the attribute matching function to better suit the preferences of different users; expressed as:

[0117] LearnedWeights(x i )=MLModel(x i ,Feedback)

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

[0119] Interest rate adaptation f1(C k ,p j ) is expressed as:

[0120] f1(C k ,p j )=exp(-α(r j -r expected (φ(C k ))));

[0121] Among them, r j Is a loan product j The interest rate; r expected (φ(x i )) is based on user x i The cluster label φ(x i ) is the expected interest rate. This means that according to the user's credit rating, the system has an expected interest rate; α is a positive adjustment coefficient used to control the sensitivity of the matching function. The size of this coefficient determines the degree of influence of the interest rate deviation on the adaptation value. Function f1(x i ,p j ) is the core of the exponential function, which ensures that the value of f1 is always between 0 and 1. The interest rate r j The closer it is to the user's expected interest rate r expected The closer the value is to 1, the higher the degree of adaptation; otherwise, the lower the degree of adaptation.

[0122] Loan amount adaptation f2(C k ,p j ) is expressed as:

[0123] Among them, l j Is the product j The loan amount; ideal is cluster C k Ideal loan amount; max and l min They are the maximum and minimum loan amounts in the system, respectively, and are used to normalize loan amount differences.

[0124] f2(x i ,p j ) is used to reflect the matching degree of the loan amount by normalizing the difference between the actual loan amount and the ideal loan amount to a ratio between 0 and 1. The smaller the difference, the higher the degree of fit.

[0125] Repayment term adaptation f3(C j ,p j ) is expressed as:

[0126] Among them, t j Is the product j The repayment period; ideal (x i ) is the ideal repayment period for user x. max and t min is the maximum and minimum repayment period in the system, used for normalization. i ,p j ) Evaluate the matching degree of repayment period by standardizing the difference of repayment period. The closer the period is to the user's ideal period, the higher the matching degree.

[0127] According to the matching model, the preference value of users in each target cluster for different loan products is predicted, which is expressed as:

[0128] P(C j ,p j )=β·m(C j ,p j )+γ·H(C k ,p j )

[0129] Among them, β and γ are adjustment parameters, H(x i ,p j ) represents the compatibility between the historical loan behavior based on the i-th target cluster and the j-th loan product.

[0130] Preference values ​​can be used to analyze user purchasing behavior and market trends, helping companies understand which product features are most popular or which loan conditions need to be adjusted. By analyzing the preference values ​​of different user groups, companies can discover changes in market demand and adjust product strategies or develop new loan products to meet market demand. Preference values ​​can also be used as an indicator to monitor system performance, helping companies evaluate the effectiveness of recommendation systems and make necessary adjustments. Regularly check the correlation between preference values ​​and actual user choices and satisfaction to evaluate the accuracy and effectiveness of recommendation algorithms. Based on this data, optimize algorithm parameters or improve user interface design. Preference values ​​can also help assess potential credit risks and ensure that loan product recommendations meet regulatory requirements. Ensure that high-risk products are not over-recommended to users with lower credit ratings, thereby managing credit risks and complying with relevant financial and credit regulations.

[0131] In summary, the present invention can more accurately reflect the actual needs and preferences of users by taking into account the credit status and historical behavior of users more carefully, so that the matching model can improve the relevance of recommendations and user satisfaction. The user experience will be enhanced by receiving more personalized recommendations and promotional information. This will not only increase users' trust in the recommendation system, but may also increase users' interaction with the platform. Accurate matching and personalized marketing strategies will increase users' acceptance of recommended loan products, thereby improving conversion rates. At the same time, by providing products that meet user expectations, user retention rates can be improved. Using machine learning technology to dynamically adjust weights and strategies can make the system more flexible and adapt to changes in the market and user behavior, thereby maintaining the competitiveness and efficiency of the system.

[0132] In step S04 of this embodiment, promotion information is customized according to sales targets and market demands, wherein the promotion information is represented as:

[0133] m i (p j ,t)=Intro(u i )+"save"+d(p j ,t)×100+"Next loan! The discount ends at"+τ(p j ,t)+"day

[0134] Among them, Intro(u i ) is the corresponding introduction word generated based on the user's clustering label, d(p j ,t) is the discount rate of product j at time t, τ(p j ,t) is the promotion period of product j, m i (p j ,t) to customize promotional information;

[0135] Among them, Intro(u i) is the corresponding introduction word generated based on the user's credit rating and financial behavior, d(p j ,t) is the discount rate of product j at time t, τ(p j ,t) is the promotion period of product j, m i (p j ,t) to customize promotional information;

[0136] In order to generate personalized intros based on the user's credit rating and financial behavior, we can use a systematic approach that combines the user's specific financial data with behavioral patterns to develop more targeted and attractive promotional information. Here are the detailed steps:

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

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

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

[0140] Loan history: including loan type, amount, repayment period and repayment record.

[0141] Credit card usage frequency: Data on credit card activity, such as utilization rates, types of cards frequently used, and repayment behavior.

[0142] Based on the above data, users are divided into different categories, each of which represents a specific credit and consumption behavior pattern. Including:

[0143] High credit consumers: excellent credit rating, low credit card utilization, and timely repayment.

[0144] Stable consumers: medium credit rating, stable consumption pattern, and preference for long-term loans.

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

[0146] For each type of user, we formulate corresponding introduction words based on the AI ​​big model to ensure that the information is consistent with the user's financial status and preferences:

[0147] High credit consumers: "As our high-quality customer, you are eligible for our exclusive offers. With outstanding credit, you can enjoy low interest rates and add points to your financial management."

[0148] Consumers who seek stability: "Thank you for your continued trust. We provide you with specially customized loan solutions for your prudent financial management. It is safe and reliable, helping you to move forward steadily."

[0149] Adventurous consumers: "Looking for flexible loan options? Our new products may be just what you need, adapting to your changing financial needs and helping you seize every opportunity."

[0150] The setting of the discount rate in the initial promotion information is based on the sales speed of the product in the previous preset period and the target sales speed, expressed as:

[0151]

[0152] Among them, d min and d max are the minimum and maximum discount rates, v(p j ,t) is the current sales speed of the product, v*(p j ) is the target sales velocity of the product;

[0153] The promotion period in the initial promotion information is set based on the seasonal demand of the product and the expected market response, expressed as:

[0154] Among them, τ min and τ max are the minimum and maximum values ​​of the promotion period, S(p j ,t) is the current sales speed of the product, T(p j ,t) is the target sales speed of the product;

[0155] Discount rates and promotion durations are regularly adjusted through a feedback mechanism.

[0156] Get the promotion effect value E(p j ,t-1), predict the promotion effect value within this preset period, expressed as:

[0157]

[0158] Where, f is the prediction function;

[0159] Calculate the predicted promotion impact factor ψ(p j ,t), expressed as:

[0160]

[0161] Among them, β, γ and δ are adjustment parameters, which adjust the logarithm of the predicted promotion effect value, the discount rate and the influence weight of the limited time information respectively;

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

[0163] d(p j,t) is the discount rate of the product in the current period;

[0164] Indicates the urgency of the promotion, that is, the countdown of the promotion period. The shorter the period, the higher the urgency.

[0165] To predict the promotion effect value, we first obtain the promotion effect value of the previous preset cycle and earlier cycles from historical data, such as (E(p j ,t-1),E(p j , t-2), etc., using the prediction function f, which 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 predicts the effect value in the current cycle based on the historical promotion effect value. Perform data preprocessing according to the needs of the model, such as data cleaning, missing value processing, outlier processing, etc.

[0166] In summary, targeted introductions and discount information can significantly increase user attention and participation in marketing activities. Personalized promotional information is more likely to resonate with target users, thereby increasing response rates and user satisfaction. By setting precise discount rates and promotion deadlines, companies can more effectively manage inventory and cash flow and optimize the allocation of marketing resources. This not only helps control marketing costs, but also maximizes sales revenue through targeted promotions. By using historical data to predict the effectiveness of promotions, marketing decisions can be made more data-driven and result-oriented. By guiding the allocation of resources through predicted promotional influencing factors, marketing resources can be focused on activities that are most likely to generate the greatest returns. Accurate forecasts and effective resource allocation will improve the overall effectiveness of promotions, increase sales, and improve ROI.

[0167] In step S05 of this embodiment, the recommendation score of each target cluster user for each loan product is calculated by combining the matching degree and the marketing influence factor. The recommendation score of each cluster k for each product j is expressed as R(u i ,p j ,t), expressed as:

[0168] R(u i ,p j ,t)=ω1·m(C j (t),p j )×ω2·ψ(C k ,p j ,t)

[0169] ω1 and ω2 are model parameters, which represent the weights of different factors in the recommendation score and can be optimized through training data.

[0170] For machine learning models, effective feature engineering is required 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: including purchase history, click history, browsing time, etc.

[0173] Product features: such as product category, price, past sales performance, etc.

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

[0175] Promotion characteristics: type, frequency, discount depth, etc. of historical promotions.

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

[0177]

[0178] Where w0 is the global bias term; N is the number of features; w n is the weight of the nth feature; x n is the input vector (feature), including the features of users and products; <v n ,v m > represents the dot product of feature n and feature m in the latent space, which is used to capture the interaction between features; v n and v m are the latent vectors of the nth and mth features respectively.

[0179] In order to evaluate the generalization ability and accuracy of the factorization machine, a cross-validation method is used. In cross-validation, the dataset is divided into K subsets of equal size. Each subset is used as a test set in turn, and the remaining K-1 subsets are used as training sets. This process is repeated K times, and the performance indicators of the model are calculated each time, and then the results are averaged, expressed as:

[0180]

[0181] Among them, the performance indicators include accuracy, root mean square error, precision and recall rate, and the model parameters include the dimension of the latent vector and the regularization term, which are optimized through hyperparameter tuning techniques such as grid search to find the best model settings.

[0182] Calculate the comprehensive score S(p) of each loan product based on the recommendation score j ,t), expressed as:

[0183]

[0184] Based on the comprehensive score, loan products are divided into different categories, including "high priority", "medium priority" and "low priority", with high priority being S(p j ,t)≥Threshold high , medium priority is Threshold low ≤S(p j ,t) <Threshold high , low priority is S(p j ,t) <Threshold low , Threshold high and Threshold low It is a preset score threshold used to distinguish products of different priorities;

[0185] High-priority loan products have the highest overall scores and are expected to bring the greatest marketing returns. Allocate the highest proportion of advertising budget and use multi-channel advertising such as TV, Internet, social media and outdoor advertising. Organize large-scale promotions such as special discounts, limited-time offers, giveaways or loyalty rewards. Use data-driven personalized marketing tools such as customized emails, SMS promotions and APP push based on known customer preferences. Provide priority customer service, including dedicated support lines and dedicated account managers.

[0186] The comprehensive score of the mid-priority loan product is moderate, which has certain market appeal but not enough to compete with the high-priority products. Allocate a moderate advertising budget, focusing mainly on cost-effective channels such as online advertising and email marketing. Carry out moderate-scale promotional activities, such as gift-with-purchase or small cash rebates. Strengthen social media interaction to increase the brand's online visibility and engagement. Increase research on the target market to better adjust product and marketing strategies.

[0187] Low-priority loan products have low overall scores and limited market appeal. Minimize investment and advertise only in very low-cost channels, such as organic social media posts. Limit promotional activities to necessary 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 high-priority or medium-priority products with greater potential.

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

[0189] At the same time, the n products with the highest recommendation scores corresponding to each target cluster k are selected for personalized push.

[0190] In this embodiment, KPIs related to marketing strategies are collected, the KPIs are used to evaluate the effect of each marketing strategy, and the parameters of the marketing strategy are adjusted based on the changes in the KPIs, including:

[0191] Collect marketing-related KPIs, such as click-through rate, conversion rate, and user engagement. Use statistical models to correlate marketing strategies with KPI changes, expressed as:

[0192] ΔKPI(p j ,t)=KPI(p j ,t)-KPI(p j ,t0)

[0193] Among them, ΔKPI(p j ,t) represents the KPI change caused by the marketing strategy, and t0 is the time point before the implementation of the marketing strategy;

[0194] Define a policy adjustment function to adjust the policy according to the base ΔKPI(p j ,t) results to update the parameters of the marketing strategy, expressed as:

[0195] Among them, θ(p j ,t+1) is the product p at time t j The marketing strategy parameter vector, η is the learning rate, is the loss function With respect to the gradient of the policy parameters θ, the loss function measures the difference between the KPI change and the policy target.

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

[0197] L(ΔKPI(p j ,t),θ(p j ,t))=(TargetKPI(p j ,t)-Current KPI(p j ,t)) 2

[0198] Among them, TargetKPI (p j ,t) is the target KPI value, Current KPI (p j ,t) is the current KPI value.

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

[0200] If the conversion rate increases significantly, then increase the advertising budget, and vice versa, expressed as:

[0201] Budget(p j ,t+1)=Budget(p j ,t)+α·(CR(p j ,t)-CR(p j ,t-1))

[0202] Among them, α is the adjustment coefficient, which determines the sensitivity of the budget to changes in conversion rate.

[0203] Adjust the discount rate in response to increases or decreases in sales, expressed as:

[0204] Discount Rate(p j ,t+1)=Discount Rate(p j ,t)+β·(Sales Increase(p j ,t)-Target Sales(p j ,t))

[0205] Similarly, the promotion duration can be adjusted to respond to increases or decreases in sales, expressed as:

[0206] Promotion Duration(p j ,t+1)=Promotion Duration(p j ,t)+γ·(ExcessInventory(p j ,t)-Safety Stock Level(p j ,t))

[0207] Among them, β and γ are adjustment coefficients that control the responses of discount rate and promotion duration to sales growth and inventory level, respectively.

[0208] The parameters of the promotion strategy can also be dynamically adjusted using a reinforcement learning model. In the construction of the reinforcement learning model:

[0209] Definition of environment: The environment is the behavior of the market and consumers, which is reflected in the changes in KPIs (such as click-through rate, conversion rate, sales, etc.).

[0210] Definition of Agent: A system used to execute marketing activities and select the parameters of promotion strategies (such as discount rate, promotion time, etc.).

[0211] Definition of State: The state is the environmental information that the agent considers when making decisions, such as current market conditions, consumer behavior, competitor activities, etc. This information can be represented by the vector s t Indicates, where t represents the current time.

[0212] Action definition: An action is a decision that an agent can make, such as setting a specific discount rate d or promotion period τ. An action can be represented as a t =(d t ,τ t ), where t represents the decision time.

[0213] Reward function: Reward function R(s t ,a t )Measure and take action t The rewards are calculated based on the improvement of KPIs.

[0214] Policy: Policy π(s t |a t ) is a mapping from the current state to taking action. The goal of reinforcement learning is to learn the optimal strategy to maximize the future cumulative reward.

[0215] In this embodiment, a Markov decision process (MDP) is used for modeling, and its goal is to find a strategy that maximizes the expected cumulative discounted reward starting from any initial state, expressed as:

[0216]

[0217] Among them, γ is a discount factor used to balance the importance of immediate rewards and future rewards.

[0218] Q learning is used as the policy optimization algorithm to directly learn the value function Q(s,a) of the state-action pair, expressed as:

[0219]

[0220] Update status information using real-time data t , and re-evaluate and adjust strategies based on the latest environmental changes. Continuously monitor the performance of strategies and adjust behaviors based on feedback t and learning parameters.

[0221] 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 enables the marketing team to quickly identify which strategies are effective and which need improvement, thereby increasing overall marketing accuracy and effectiveness.

[0222] By calculating the KPI change ΔKPI(p j ,t) and marketing strategy parameter θ(p j,t) association, you can adjust marketing resources more intelligently. If the conversion rate of a product increases, the advertising budget for that product may be increased, and vice versa. This dynamic adjustment helps ensure that resources are always invested in the areas most likely to generate the highest returns.

[0223] Define a strategy adjustment function to make marketing decisions more data-driven. This approach not only reduces the impact 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.

[0224] Therefore, the present invention improves the speed of enterprises' response to market changes through real-time data analysis and immediate strategy adjustment. It helps to significantly improve the return on each marketing investment by continuously optimizing marketing strategies and adjusting investment directions. In a highly competitive market environment, it can quickly adapt to changes in consumer behavior and preferences.

[0225] The present invention also provides a data processing system for intelligent marketing products, based on the data processing method for the aforementioned intelligent marketing products, comprising the following modules:

[0226] Clustering module: obtains the target user's financial behavior data, including credit card usage records, consumption records, historical loan records, and historical repayment records. The original data is converted into vector features through feature extraction functions to form a set. The vector feature set is divided into multiple clusters using a clustering algorithm. The user is labeled with a corresponding cluster label based on the clustering results.

[0227] Screening module: Calculate the value score of each cluster through the preset value evaluation model, determine the candidate clusters with value scores greater than the preset threshold as target clusters, and remove non-target clusters in the data set;

[0228] Matching degree calculation module: evaluates the comprehensive matching degree between the users of each target cluster and multiple loan products through a preset matching degree model, and predicts the preference value of the users of each target cluster for different loan products according to the matching degree model, wherein the matching degree model is established according to the historical loan behavior of the users of each target cluster;

[0229] Promotion module: customizes promotion information according to the sales target within the preset period of the loan product, including discount information and time-limited information, predicts the promotion effect value of the loan product within the preset period, and calculates the promotion impact factor according to the promotion effect value;

[0230] Marketing strategy module: Combine the matching degree and marketing influence factor to calculate the recommendation score of each target cluster user 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;

[0231] Feedback update module: collects KPIs related to the marketing strategies within the preset period, uses the data processing platform to evaluate the effectiveness of each marketing strategy, and adjusts the parameters of the marketing strategies within the next preset period based on the changes in KPIs.

[0232] The present invention dynamically adjusts recommendation and marketing strategies by monitoring changes in market conditions and user behavior in real time. This flexibility enables enterprises to quickly adapt to market changes and remain competitive. The operation of the system is based on data-driven decision support, from cluster analysis of user data to KPI evaluation of marketing strategies. This evidence-based approach can reduce the impact of contingency and subjective judgment and improve 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 companies avoid issuing loans to customers with poor credit, thereby reducing risks. In summary, the present invention not only improves the efficiency and accuracy of the loan product recommendation system, but also helps companies optimize resource allocation, improve user satisfaction and loyalty while maintaining market competitiveness, and ultimately achieve business growth and profit maximization.

[0233] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A data processing method for an intelligent marketing product, characterized in that: The following steps are involved: Step S01: obtaining the target user's financial behavior data, which includes credit card usage records, consumption records, historical loan records, and historical repayment records; converting the original data into vector features through feature extraction and forming a set; using a clustering algorithm to divide the vector feature set into multiple clusters; and labeling the user with a corresponding cluster label according to the clustering result; Step S02: Calculate the value score of each cluster through a preset value evaluation model, determine the candidate cluster with a value score greater than a preset threshold as the target cluster, and remove non-target clusters in the data set; Step S03: Evaluate the comprehensive matching degree between the users of each target cluster and multiple loan products through a preset matching degree model, and predict the preference value of the users of each target cluster for different loan products according to the matching degree model, wherein the matching degree model is established according to the historical loan behavior of the users of each target cluster; Step S04: Customizing promotion information according to the sales target within a preset period of the loan product, the promotion information including discount information and time-limited information, predicting the promotion effect value of the loan product within the preset period, and calculating the promotion impact factor according to the promotion effect value; Step S05: Calculate the recommendation score of each target cluster user for each loan product based on the matching degree and the marketing influence factor, 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; Step S06: Collect KPIs related to the marketing strategies within a preset period to evaluate the marketing effectiveness value of each marketing strategy, and adjust the parameters of the marketing strategies within the next preset period based on the marketing effectiveness value.

2. The data processing method of the smart 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 corresponding cluster labels are marked for the user according to the clustering results, including: Get the user's original data set X = {x1, x2, x3, ..., x n }, the original data x is extracted by the feature extraction function f i Convert to vector feature f(x i ), where x i represents the data of the i-th user, and the feature vector is specifically expressed as f(x i )=[f1(x i ), f2(x i ), ..., f m (x i )] T , where m is the number of features; Use the clustering algorithm to cluster the feature vector set {f(x1), f(x2), ..., f(x N )) is divided into K clusters; The clustering result is expressed as a function φ, which transforms each user data x i Mapped to cluster label φ(x i ), expressed as: φ(x i )=k,wherek∈{1, 2,...,K}; The clustering algorithm K-means is applied to minimize the variance within each cluster, expressed as: Where S K is the kth cluster, μk is the cluster center; Each user is assigned a corresponding cluster label according to the cluster to which he belongs, and each cluster label is assigned through the CreditLevel function, which is expressed as: r i (t)=CreditLevel(φ(x i ))=CreditLevel(φ(x i ))。 3. The data processing method of the smart marketing product according to claim 2, characterized in that: In step S02, a value assessment model is established based on the user's credit rating, historical response rate and purchasing power, and is expressed as: V(C k )=α1·CreditLevel(C k )+α2·ResponseRate(C k )+α3·PurchasingPower(C k ) Among them, α1, α2, and α3 are weight coefficients, respectively, and CreditLevel(C k )、ResponseRate(C k ) and PurchasingPower(C k ) represents the average of the credit rating, response rate, and purchasing power of all users in the cluster: The value score of each cluster is calculated through the preset value evaluation model, and the candidate clusters with value scores greater than the preset threshold are determined as target clusters, and non-target clusters in the data set are removed, which is expressed as: Among them, X new represents the new user data set after removing non-target clustering users, Represents the set of users in all non-target clusters.

4. The data processing method of the smart marketing product according to claim 3, characterized in that: In step S03, a matching model is defined to evaluate the matching between the user and multiple loan products, including: Define P as a set of loan products. Each loan product p in the loan product set P j There are multiple properties, including the interest rate r j , loan amount j , and repayment period t j ; Through the comprehensive matching function m(C k , p j ) evaluates the matching degree between the kth target cluster and the jth loan product. The matching degree function is specifically expressed as: m(C k ,p j )=w1f1(C k ,p j )+w2f2(C k ,p j )+w3f3(C k ,p j ) Among them, f1, f2, f3 are matching functions for calculating interest rate, loan amount and repayment period, and w1, w2, W3 are weights for calculating interest rate, loan amount and repayment period; The interest rate adaptation f1(C k , p j ) is expressed as: f1(C k , p j )=exp(-α(r j -r exDected (φ(C k )))); Loan amount adaptation f2(C k , p j ) is expressed as: Repayment term adaptation f3(C k , p j ) is expressed as: According to the matching model, the preference value of users in each target cluster for different loan products is predicted, expressed as: P(C k , p j )=β·m(C k , p j )+γ·H(C k , p j ); Among them, β and γ are adjustment parameters, H(x i , p j ) represents the compatibility between the historical loan behavior based on the i-th target cluster and the j-th loan product.

5. The data processing method of the smart marketing product according to claim 4, characterized in that: In step S04, promotion information is customized according to sales targets and market demands, where the promotion information is represented as: m i (p j , t) = Intro(u i )+"saving"+d(p j ,t)×100+″next loan! The discount ends at″+τ(p j , t)+″day Among them, Intro(u i ) is the corresponding introduction word generated based on the user's clustering label, d(p j , t) is the discount rate of product j at time t, τ(p j , t) is the promotion period of product j, m i (p j , t) to customize promotional information; Get the promotion effect value E(p j , t-1), predict the promotion effect value within this preset period, expressed as: Where, f is the prediction function; Calculate the predicted promotion impact factor ψ(p j , t), expressed as: Among them, β, γ and δ are adjustment parameters.

6. The data processing method of the smart marketing product according to any one of claims 1 to 5, characterized in that: Combined with the matching degree and marketing influence factor, the recommendation score of each target cluster user for each loan product is calculated. The recommendation score of each cluster k for each product j is expressed as R(u i , p j , t), expressed as: R(u i ,p j ,t)=ω1·m(C k (t),p j )×ω2·ψ(C k ,p j ,t) Calculate the comprehensive score S(p) of each loan product based on the recommendation score j , t), expressed as: Based on the comprehensive score, loan products are divided into different categories, including "high priority", "medium priority" and "low priority", with high priority being S(p j , t)≥Threshold high , medium priority is Threshold low ≤S(p j , t) <Threshold high , low priority is S(p j , t) <Threshold low , Threshold high and Threshold low It is a preset score threshold used to distinguish products of different priorities and select marketing strategies corresponding to the products of different priorities for marketing; At the same time, the n products with the highest recommendation scores corresponding to each target cluster k are selected for personalized push.

7. The data processing method of the smart marketing product according to claim 6, characterized in that: Collect KPIs related to marketing strategies, use KPIs to evaluate the effectiveness of each marketing strategy, and adjust marketing strategy parameters based on changes in KPIs, including: Collect marketing-related KPIs, including click-through rate, conversion rate, and user engagement. Use statistical models to correlate marketing strategies with KPI changes, expressed as: ΔKPI(p j ,t)=KPI(p j ,t)-KPI(p j ,t0) Among them, ΔKPI(p j , t) represents the KPI change caused by the marketing strategy, and t0 is the time point before the implementation of the marketing strategy; Define a policy adjustment function to adjust the policy according to the base ΔKPI(p j , t) results update the parameters of the marketing strategy, expressed as: Among them, θ(p j , t+1) is the product p at time t j The marketing strategy parameter vector, η is the learning rate, It is the gradient of the loss function L with respect to the policy parameter θ. The loss function measures the difference between the change in KPl and the policy target.

8. The data processing method of the smart marketing product according to claim 5, characterized in that: The setting of the discount rate in the initial promotion information is based on the sales speed of the product in the previous preset period and the target sales speed, expressed as: Among them, d min and d max are the minimum and maximum discount rates, v(p j , t) is the current sales speed of the product, v * (p j ) is the target sales velocity of the product; The promotion period in the initial promotion information is set based on the seasonal demand of the product and the expected market response, expressed as: Among them, τ min and τ max are the minimum and maximum values ​​of the promotion period, S(p j , t) is the current sales speed of the product, T(p j , t) is the target sales speed of the product; the discount rate and promotion period are adjusted regularly through the feedback mechanism.

9. A data processing system for smart marketing products, based on the data processing method for smart marketing products according to any one of claims 1 to 8, characterized in that: Includes the following modules: Clustering module: obtains the target user's financial behavior data, including credit card usage records, consumption records, historical loan records, and historical repayment records. The original data is converted into vector features through feature extraction functions to form a set. The vector feature set is divided into multiple clusters using a clustering algorithm. The user is labeled with a corresponding cluster label based on the clustering results. Screening module: Calculate the value score of each cluster through the preset value evaluation model, determine the candidate clusters with value scores greater than the preset threshold as target clusters, and remove non-target clusters in the data set; Matching degree calculation module: evaluates the comprehensive matching degree between the users of each target cluster and multiple loan products through a preset matching degree model, and predicts the preference value of the users of each target cluster for different loan products according to the matching degree model, wherein the matching degree model is established according to the historical loan behavior of the users of each target cluster; Promotion module: customizes promotion information according to the sales target within the preset period of the loan product, including discount information and time-limited information, predicts the promotion effect value of the loan product within the preset period, and calculates the promotion impact factor according to the promotion effect value; Marketing strategy module: Combine the matching degree and marketing influence factor to calculate the recommendation score of each target cluster user 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 the marketing strategies within the preset period, uses the data processing platform to evaluate the effectiveness of each marketing strategy, and adjusts the parameters of the marketing strategies within the next preset period based on the changes in KPIs.

Images