A banking business marketing data processing method and system based on multiple channels

Abstract

The present invention relates to the technical field of banking business marketing data processing, specifically a banking business marketing data processing method and system based on multiple channels, which integrates multiple channel data, obtains and verifies the source of banking business marketing data in real time, ensures the accuracy and reliability of marketing strategies, and then uses machine learning algorithms to generate customer portraits, analyzes their behavior patterns and potential needs, generates personalized marketing strategies, and combines natural language processing technology to ensure the real-time and pertinence of marketing content, pushes according to customer preferences, and optimizes the push timing. After pushing, the system monitors feedback behavior in real time, evaluates the effect of the strategy, optimizes marketing activities using A / B testing, forms closed-loop feedback, continuously improves strategies, and ensures the efficiency and accuracy of marketing plans. The present invention realizes automated strategy generation and optimization through AI technology, greatly improves the marketing efficiency and customer experience of banking business, and is suitable for multi-channel data processing scenarios.

Description

Technical Field

[0001] The present invention relates to the technical field of banking business marketing data processing, and in particular to a banking business marketing data processing method and system based on multiple channels. Background Art

[0002] With the development of financial technology, the digital and intelligent transformation of banking business has become an industry trend. Traditional banking marketing methods rely on a single channel and extensive customer management, which is difficult to meet the increasingly diverse needs of modern customers. Customer behavior data comes from a wide range of sources, including offline outlets, mobile apps, ATMs, social media and other channels. How to effectively integrate and utilize these scattered customer data to improve the accuracy and personalization of marketing has become an important challenge for banking business;

[0003] The Chinese invention application with publication number CN114399376A discloses a bank marketing data processing method, which strictly distinguishes data into static data from within the bank and dynamic data from marketing channels, generates label information of the sedimentation platform based on the two types of data, and performs main bank docking output based on the label information, thereby automatically distinguishing different data and outputting them in a targeted manner, thereby improving the intelligence of bank data and enhancing the data docking and output capabilities for bank marketing scenarios;

[0004] However, in actual production life, banking business marketing still has problems such as scattered customer data that is difficult to manage in a unified manner, customer needs that are difficult to accurately identify, marketing strategies that are difficult to accurately match, traditional marketing methods that are slow to respond and have poor results, and the inability to effectively track and optimize marketing activities, which in turn affects the intelligent processing of banking business marketing data. Summary of the invention

[0005] The purpose of the present invention is to address the problems existing in the background technology and to propose a multi-channel banking business marketing data processing method and system.

[0006] The technical solution of the present invention is a multi-channel banking business marketing data processing method, which includes the following specific implementation steps:

[0007] S1. The business marketing data collection module obtains customer-related data and various market analysis data from multiple channels in real time, aggregates them to obtain business marketing data, and generates exclusive label information for the business marketing data, and then transmits {exclusive label, data type identifier, marketing data} to the business marketing data processing module;

[0008] S2. The business marketing data processing module checks the source of the received marketing data, distinguishes the customer-related data and the market analysis data in the marketing data according to the data type identifier, and then transmits the customer-related data to the customer portrait generation module, and transmits the market analysis data to the marketing strategy recommendation module;

[0009] S3, the customer portrait generation module uses customer identity matching data, uses the K-means algorithm to classify customers, analyzes customer behavior through the SARIMA model, predicts future trends, uses the random forest model to predict customer potential needs, generates accurate customer portraits, and transmits the customer portraits to the marketing strategy recommendation module;

[0010] S4, the marketing strategy recommendation module uses the pre-trained sentiment analysis model based on customer classification and demand analysis, combined with the BERT pre-trained model and CNN network, outputs the sentiment analysis results, formulates targeted marketing strategies based on them, and then transmits the marketing strategies to the multi-channel push module;

[0011] S5. The multi-channel push module pushes the formulated personalized marketing strategies to customers through multiple channels, and sends feedback information collection instructions to the effect monitoring and feedback module;

[0012] S6. After the marketing campaign is launched, the effect monitoring and feedback module monitors the customer's response behavior in real time, automatically collects and analyzes customer feedback data, measures the effectiveness of the marketing strategy, and compares the effects of different marketing strategies or push methods through A / B testing, and then transmits the feedback data to the marketing strategy recommendation module;

[0013] S7. The marketing strategy recommendation module continuously adjusts and optimizes marketing strategies through continuous analysis of customer feedback.

[0014] Preferably, the specific implementation steps of the customer portrait generation module using the K-means algorithm to classify customers are as follows:

[0015] S21. Randomly select K data points as the initial cluster centers. Each cluster center represents the centroid of a cluster, and its vector is expressed as: C k ={c k1 , c k2 , ..., c kf , ..., c kn};

[0016] Among them, c kf represents the initial value of the kth cluster center on the fth feature, and n is the total number of data;

[0017] S22. For any customer CID i Data point X i , calculate its center C to each clusterk The Euclidean distance of each data point X is selected as the cluster with the smallest distance. i Assigned to the cluster center C with the smallest distance k , update the set of data points within the cluster;

[0018] The calculation formula of Euclidean distance is:

[0019] ;

[0020] In the formula, x if Indicates customer CID i At data point X i The value of the fth feature in ;

[0021] S23. Calculate the new cluster center of each cluster, that is, the mean of all data points in the cluster on each feature. The formula for updating the new cluster center is as follows:

[0022] ;

[0023] In the formula, S k is the set of all data points currently belonging to the kth cluster, |S k | represents the number of data points belonging to the kth cluster, is the value of the updated cluster center on the fth feature;

[0024] S24, repeat S22 and S23 until the position of the cluster center no longer changes significantly, that is, the moving distance of the cluster center is less than a preset threshold value| - |<ε, or the maximum number of iterations is reached;

[0025] in, is the value of the cluster center selected in the previous iteration on the fth feature, and ε is the preset minimum value, i.e., the threshold;

[0026] S25. After clustering is completed, the customers are divided into three clusters.

[0027] Preferably, the customers are divided into three clusters: a VIP customer cluster, a potential churn customer cluster and a common customer cluster.

[0028] Preferably, the training process of the SARIMA model is as follows:

[0029] S41. From any client CID i Data point X i Extracting time series data of customer behavior, including but not limited to the number of transactions per month, the monthly transaction amount, and the monthly account balance changes;

[0030] S42. Preprocess the raw data, handle missing data, and standardize the data to ensure that it is suitable for time series analysis;

[0031] S43. Use the ADF test to detect the stationarity of the time series. If the time series is not stable, differential processing is required. The differential formula is as follows:

[0032] ;

[0033] In the formula, represents the difference between the value at the current time t and the value at the previous time t-1, Y(t) represents the customer behavior data at the current time t, and Y(t-1) represents the customer behavior data at the previous time t-1;

[0034] S44. Train the SARIMA model. The model learns seasonal patterns and trends in time series based on the input customer behavior data. The model expression is:

[0035] SARIMA(p,d,q)(p',d',q',s);

[0036] Where p is the autoregressive order, d is the non-seasonal difference order, q is the moving average order, p' is the seasonal autoregressive order, d' is the seasonal difference order, q' is the seasonal moving average order, and s is the length of the season;

[0037] S45. Deploy the trained SARIMA model to the customer portrait generation module.

[0038] Preferably, the method for the business marketing data collection module to generate exclusive tag information for business marketing data is as follows:

[0039] S51, randomly select temporary coding element Ecode i ∈[1,λ-1], calculate the temporary coding information Icode i =Ecode i ×P;

[0040] Where P is defined in the finite field F p An elliptic curve E(F p ) on the λ-order base point;

[0041] S52, calculate the first-level label information I lable1i =x(Icode i ) (mod λ);

[0042] Among them, x(Icode i ) indicates Icode i In the elliptic curve E(F p ) on the x-coordinate;

[0043] S53, select random numbers α, β∈[1,λ-1], and calculate the secondary label information:

[0044] I lable2i =α×I lable1i +(w+I lable1i )×Ctg (mod λ);

[0045] Among them, α and β satisfy Ecode i =α×I lable1i +β×mdata i ,w is H(mdata i ), H is the hash function, Ctg is the label generation code, Ctg∈[1,λ-1], mdata i Any marketing data in binary string form;

[0046] S54. Generate any marketing data mdata i Exclusive label Lable i ={β,I lable1i , I lable2i}.

[0047] Preferably, the method for the business marketing data processing module to check the source of the received marketing data is as follows:

[0048] S61, test β, I lable1i ,I lable2i ∈[1,λ-1] is true, if not, then directly determine the marketing data mdata i The source is illegal, so the marketing data mdata is discarded i Otherwise, the following step S62 is executed;

[0049] S62, calculate H(mdata i )’s Hamming weight w’;

[0050] S63, calculate the following intermediate parameters:

[0051] Ip 1i =(I lable2i +β×mdata i ) (mod λ);

[0052] Ip 2i =(w'+I lable1i ) (mod λ);

[0053] Ip 3i =Ip 1i ×P-Ip 2i ×Ctp;

[0054] S64. Calculate verification information Ic i =x(Ip 3i ) (mod λ);

[0055] Wherein, Ctp is the label parsing code, Ctp=Ctg×P;

[0056] S65, if Ic i =I lable1i , then the marketing data mdata i Collected by the business marketing data collection module.

[0057] Preferably, the data type identifier is: the data type identifier of the customer-related data is defined as 0x00, and the data type identifier of the market analysis data is defined as 0x01.

[0058] The technical solution of the present invention is a multi-channel banking business marketing data processing system, which is used to execute the multi-channel banking business marketing data processing method proposed above, comprising:

[0059] Business marketing data collection module, automatically collects customer behavior and transaction data from multiple channels;

[0060] The business marketing data processing module cleans, removes duplications, and formats the collected data to ensure data accuracy and consistency;

[0061] The customer portrait generation module collects customer behavior and transaction data and uses AI and machine learning algorithms to mine customers' potential needs and behavioral preferences and build customer portraits.

[0062] The marketing strategy recommendation module uses natural language processing technology to analyze market dynamics and changes in customer demand in real time, generate personalized marketing strategies, and automatically recommend the best marketing time, channels, and content based on different customer groups and market environments;

[0063] The effect monitoring and feedback module monitors the effect of each marketing activity in real time and evaluates customer feedback and conversion rate through data analysis;

[0064] The multi-channel push module selects appropriate channels for precise marketing content push based on customer portraits, and automatically delivers marketing content through multiple channels.

[0065] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0066] The present invention proposes a multi-channel banking business marketing data processing method and system. The system integrates multi-channel data, obtains banking business marketing data in real time, and verifies the source of banking business marketing data to ensure that the marketing data is collected by the system, thereby ensuring the accuracy and reliability of subsequent marketing strategy formulation. Subsequently, a machine learning algorithm is used to generate customer portraits and analyze their behavior patterns and potential needs. Based on the customer portraits, the system generates personalized marketing strategies with the help of AI, and combines natural language processing technology to ensure the real-time and pertinence of marketing content. The push link is accurately pushed through the channels preferred by customers, and the push timing is optimized through algorithms. After the push, the system monitors the customer's feedback behavior in real time, evaluates the effectiveness of the strategy, and optimizes marketing activities through A / B testing. The system forms a feedback closed loop, continuously improves strategies, and ensures the effectiveness and accuracy of marketing plans. The present invention uses AI technology to realize automatic generation and real-time optimization of marketing strategies, which can significantly improve the marketing efficiency and customer experience of banking business, and is suitable for complex multi-channel data processing scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is an architecture diagram of a multi-channel banking business marketing data processing system proposed by the present invention;

[0068] Figure 2 This is a network structure diagram of the sentiment analysis model proposed in the present invention. DETAILED DESCRIPTION

[0069] Embodiment 1, as Figure 1 As shown, the present invention proposes a multi-channel banking business marketing data processing system, including a business marketing data collection module, a business marketing data processing module, a customer portrait generation module, a marketing strategy recommendation module, an effect monitoring and feedback module and a multi-channel push module;

[0070] The business marketing data collection module automatically collects customer behavior and transaction data from multiple channels (including but not limited to ATM, mobile banking, official website, third-party payment and customer service);

[0071] The business marketing data processing module cleans, removes duplications, and formats the collected data to ensure data accuracy and consistency;

[0072] The customer portrait generation module collects customer behavior and transaction data and uses AI and machine learning algorithms to mine customers' potential needs and behavioral preferences and build customer portraits.

[0073] The marketing strategy recommendation module uses natural language processing technology to analyze market dynamics and changes in customer demand in real time, generate personalized marketing strategies, and automatically recommend the best marketing time, channels, and content based on different customer groups and market environments;

[0074] The effect monitoring and feedback module monitors the effect of each marketing campaign in real time and evaluates customer feedback and conversion rate through data analysis;

[0075] The multi-channel push module selects appropriate channels (including but not limited to SMS, email, bank APP push and social media advertising) to push precise marketing content based on customer portraits, thereby realizing automated delivery of marketing content through multiple channels.

[0076] Embodiment 2, a multi-channel banking marketing data processing method proposed by the present invention is applicable to a multi-channel banking marketing data processing system proposed in Embodiment 1, and its specific implementation steps are as follows:

[0077] S1. The business marketing data collection module obtains data from multiple business channels of the bank (including but not limited to offline bank branches, mobile banking apps, ATMs, customer service hotlines, social media, third-party payment platforms, e-commerce platform cooperation data and SMS services) in real time, and achieves seamless connection of data from various channels through API interfaces to obtain customer-related data. On the other hand, it uses crawlers to regularly crawl competitor product information, market reports and related discussions on social media to obtain market analysis data. The business marketing data collection module aggregates customer-related data and various market analysis data, distinguishes data types, obtains business marketing data, and generates exclusive label information for the business marketing data, and then transmits {exclusive label, data type identifier, marketing data} to the business marketing data processing module;

[0078] The data type identifier of customer-related data is defined as 0x00, and the data type identifier of market analysis data is defined as 0x01;

[0079] The above business marketing data collection module is any marketing data mdata i (mdata i Generate a unique label Lable in binary string form i The method is as follows:

[0080] (1) Randomly select a temporary encoding element Ecode i ∈[1,λ-1], calculate the temporary coding information Icode i =Ecode i ×P;

[0081] Where P is defined in the finite field F p An elliptic curve E(F p ) on the λ-order base point;

[0082] (2) Calculate the first-level label information I lable1i=x(Icode i ) (mod λ);

[0083] Among them, x(Icode i ) indicates Icode i In the elliptic curve E(F p ) on the x-coordinate;

[0084] (3) Select random numbers α, β∈[1,λ-1] and calculate the secondary label information:

[0085] I lable2i =α×I lable1i +(w+I lable1i )×Ctg (mod λ);

[0086] Among them, α and β satisfy Ecode i =α×I lable1i +β×mdata i ,w is H(mdata i ), H is the hash function, Ctg is the label generation code, Ctg∈[1,λ-1], and the label resolution code Ctp=Ctg×P is defined accordingly;

[0087] (4) Generate a unique label i ={β,I lable1i , I lable2i};

[0088] It should be noted that Hamming Weight, also known as binary bit weight, refers to the number of "1"s in the binary representation of a number;

[0089] It should be noted that the collected customer-related data includes but is not limited to the customer's transaction records, account information, login behavior, consultation records, customer service call transcripts, social media interactions and SMS feedback;

[0090] S2. The business marketing data processing module first checks the source of the received marketing data to ensure that the marketing data is collected by the business marketing data collection module in the system, thereby ensuring the accuracy and reliability of the subsequent marketing strategy formulation. It then deduplicates the collected customer-related data and market analysis data to eliminate repeated behavior records of customers in different channels to ensure that each customer's behavior is unique. It then automatically detects missing values ​​in the marketing data, uses statistical methods or machine learning models to fill in missing data, and unifies the data formats from different sources, such as unifying currency symbols, date formats, customer identity identifiers, etc. It distinguishes customer-related data and market analysis data in the marketing data based on data type identifiers, then transmits customer-related data to the customer portrait generation module, and transmits market analysis data to the marketing strategy recommendation module.

[0091] The method for the business marketing data processing module to check the source of the received marketing data is as follows:

[0092] (1) Test β and I lable1i ,I lable2i ∈[1,λ-1] is true, if not, then directly determine the marketing data mdata i The source is illegal, so the marketing data mdata is discarded i Otherwise, execute the following step (2);

[0093] (2) Calculate H(mdata i )’s Hamming weight w’;

[0094] (3) Calculate the following intermediate parameters:

[0095] Ip 1i =(I lable2i +β×mdata i ) (mod λ);

[0096] Ip 2i =(w'+I lable1i ) (mod λ);

[0097] Ip 3i =Ip 1i ×P-Ip 2i ×Ctp;

[0098] (4) Calculate the verification information Ic i =x(Ip 3i ) (mod λ);

[0099] (5) If Ic i =I lable1i , then the marketing data mdata i The source is legal;

[0100] S3, the customer portrait generation module uses customer identity matching data, uses the K-means algorithm to classify customers, and stratifies them based on transaction frequency, account balance and other characteristics. It analyzes customer behavior through time series analysis, predicts future trends, and uses the random forest model to predict customer potential needs. It generates accurate customer portraits and transmits the customer portraits to the marketing strategy recommendation module. The specific implementation process is as follows:

[0101] S3-1. Use customer identity to match and integrate to ensure that the data of the same customer in different channels can be accurately aggregated:

[0102] ;

[0103] Where CID i For any customer's identity, Data i CID of customers collected from multiple channels i Data, D j is the data set of the jth channel, and m is the total number of channels;

[0104] S3-2, based on any customer CID i Data i , construct a feature vector, including but not limited to transaction frequency, account balance, financial product usage, loan record, age and credit score, and obtain the customer's CID i Data point X i ={x i1 , x i2 , ..., x if , ..., x in};

[0105] Among them, x if Indicates customer CID i At data point X i The value of the fth feature in ;

[0106] S3-3. In order to avoid the result deviation caused by different dimensions when clustering different features, the feature values ​​need to be standardized:

[0107] ;

[0108] In the formula, x if Indicates any customer CID i The original value of the f-th feature, μ f represents the mean of the fth feature, δ f represents the standard deviation of the f-th feature;

[0109] S3-4. Use K-means algorithm to classify customers, as follows:

[0110] (1) Randomly select K data points as the initial cluster centers. Each cluster center represents the centroid of a cluster, and its vector is expressed as: C k ={c k1 , c k2 , ..., c kf , ..., c kn};

[0111] Among them, c kf represents the initial value of the k-th cluster center on the f-th feature;

[0112] It should be noted that the K value in the present invention is 3;

[0113] (2) For each data point X i , calculate its center C to each cluster k The Euclidean distance of each data point X is selected as the cluster with the smallest distance. i Assigned to the cluster center C with the smallest distance k , update the set of data points within the cluster;

[0114] The calculation formula of Euclidean distance is:

[0115] ;

[0116] (3) Calculate the new cluster center of each cluster, that is, the mean of all data points in the cluster on each feature. The formula for updating the new cluster center is as follows:

[0117] ;

[0118] In the formula, S k is the set of all data points currently belonging to the kth cluster, |S k | represents the number of data points belonging to the kth cluster, is the value of the updated cluster center on the fth feature;

[0119] (4) Repeat (2) and (3) until the position of the cluster center no longer changes significantly, that is, the moving distance of the cluster center is less than the preset threshold | - |<ε, or the maximum number of iterations is reached;

[0120] in, is the value of the cluster center selected in the previous iteration on the fth feature, and ε is the preset minimum value, i.e., the threshold;

[0121] (5) After clustering, the customers are divided into three clusters, and the characteristics of each cluster are analyzed based on the center value of each cluster:

[0122] VIP customer cluster: This cluster of customers has high transaction frequency, large account balances, active use of financial products, and high credit scores, representing the bank's high-value customers;

[0123] Potential churn customer cluster: This cluster of customers has a decrease in transaction frequency, a decrease in account balance, a long time since the last transaction, and a decrease in the frequency of using financial products, indicating that these customers are at risk of churn;

[0124] Ordinary customer cluster: The transaction frequency, account balance, and use of financial products of customers in this cluster are at an intermediate level, and they are regular bank customers;

[0125] S3-5. Analyze the customer's historical transactions, account changes, product usage and other behaviors, extract the customer's behavioral characteristics, and use the SARIMA (Seasonal Autoregressive Integrated Moving Average) model to predict future customer behavior;

[0126] It should be noted that the training process of the SARIMA model is as follows:

[0127] (1) From any customer CID i Data point X i Extracting time series data of customer behavior, including but not limited to the number of transactions per month, the monthly transaction amount, and the monthly account balance changes;

[0128] (2) Preprocess the raw data, handle missing data, and standardize the data to ensure that it is suitable for time series analysis;

[0129] (3) Use the ADF (Augmented Dickey-Fuller) test to detect the stationarity of the time series. If the time series is detected to be non-stationary, differential processing is required. The differential formula is as follows:

[0130] ;

[0131] In the formula, represents the differential value at time t, that is, the difference between the value at the current time t and the value at the previous time t-1. Y(t) represents the actual value of the time series at time t, that is, the customer behavior data at the current time t. Y(t-1) represents the actual value of the time series at time t-1, that is, the customer behavior data at the previous time t-1.

[0132] (4) Train the SARIMA model. During the training process, the model will learn seasonal patterns and trends in the time series based on the input customer behavior data. The model expression is:

[0133] SARIMA(p,d,q)(p',d',q',s);

[0134] Wherein, p is the autoregressive order, d is the non-seasonal difference order, q is the moving average order, p' is the seasonal autoregressive order, d' is the seasonal difference order, q' is the seasonal moving average order, and s is the season length (the present invention uses monthly data, and its season length is 12);

[0135] (5) Deploy the trained SARIMA model to the customer profile generation module;

[0136] S3-6. Use random forest (RF) for integrated prediction, combine feature importance analysis and feedback mechanism to understand customer demand characteristics. The specific implementation process is as follows:

[0137] (1) Obtain any customer CID i Data point X i ={x i1 , x i2 , ..., x if , ..., x in};

[0138] (2) Assign a prediction label to each customer based on business needs. This label includes, but is not limited to, whether the customer will purchase a certain financial product, participate in a certain banking service, or respond to a specific marketing campaign;

[0139] (3) Random forest predicts customer demand by generating multiple decision trees. Each tree makes a prediction based on the characteristics of the data sample, specifically through a "question-answer" decision path.

[0140] For example, “Does the customer account balance exceed a certain threshold?”, “Is the number of transactions above a certain level?”;

[0141] (4) The decision tree selects the optimal feature at each split and evaluates the contribution of each feature to the classification effect by calculating the "information gain". The information gain formula is as follows:

[0142] ;

[0143] In the formula, T is the sample set of the current node, A is the feature to be split, Tv is the sample subset corresponding to the vth value of feature A, and Entropy is the entropy function used to measure the uncertainty of the data;

[0144] (5) When the split cannot continue, each leaf node of the decision tree will give a prediction result;

[0145] For example, “customers will buy a product” or “customers will not buy it”;

[0146] (6) Each decision tree generates a prediction result, and the random forest outputs the final prediction result by integrating the results of multiple decision trees:

[0147] ;

[0148] Where M is the total number of decision trees, T m (CID i ) is the mth tree for customer CID i Demand forecasting, Predict results for the final ensemble;

[0149] S4. The marketing strategy recommendation module generates personalized marketing strategies through detailed customer classification and demand analysis to ensure that each customer can obtain products and services suitable for them, and transmits the marketing strategies to the multi-channel push module. The implementation process of the marketing strategy recommendation module is as follows:

[0150] S4-1, such as Figure 2 As shown, based on market analysis data, a sentiment analysis model is built to analyze social media content, identify customer emotions and concerns, and analyze customer attitudes toward different products;

[0151] It should be noted that the sentiment analysis model used in the present invention combines the BERT (Bidirectional Encoder Representations from Transformers) pre-training model and CNN (Convolutional Neural Networks), and its analysis process is as follows:

[0152] (1) Social media text content Text = {text 1 ,text 2 ,...,text i ,...,text n} Input to BERT pre-trained model;

[0153] Among them, text i is the word embedding vector of the i-th word;

[0154] (2) The BERT pre-trained model encodes the social media text content Text through the self-attention mechanism and obtains the hidden state matrix H = {h 1 ,h 2, ..., h i , ..., h n};

[0155] (3) The pre-training layer adjusts the BERT model output and introduces the weight matrix W to obtain the hidden state matrix H hidden =ReLU(W·H);

[0156] (4) Convolutional neural network CNN for hidden state matrix H hidden Perform convolution operation to obtain the convolution feature matrix C=ReLU(W CNN *H hidden );

[0157] Among them, W CNN is the convolution kernel of the convolutional neural network;

[0158] (5) The output layer receives the output C of the convolutional neural network and passes through the fully connected layer W FC Mapped to the sentiment category space, the sentiment analysis result Y is finally obtained:

[0159] Y=softmax(W FC C);

[0160] S4-2, combine the analysis results, identify emerging trends, and develop targeted marketing strategies:

[0161] (1) VIP customer strategy: Based on the results of sentiment analysis, we obtain products or services that customers are interested in, and design exclusive preferential services for high-value customers, including but not limited to customized financial plans, exclusive account managers, and VIP event invitations. We also recommend related or complementary products based on the customer's behavior prediction results and demand prediction results.

[0162] (2) Potential churn customer strategy: proactively contact potential churn customers to understand their needs and dissatisfactions, and provide personalized solutions, including but not limited to improving service quality and providing additional customer support. Based on the results of sentiment analysis, obtain products or services that customers are interested in, provide limited-time discounts, fee reductions, reward points, etc. to potential churn customers to encourage them to continue using banking services, and recommend relevant information based on the customer behavior prediction results and demand prediction results;

[0163] (3) General customer strategy: Based on sentiment analysis results, obtain products or services that customers are interested in, and recommend basic products suitable for them based on their historical transaction data. Based on the customer's behavior prediction results and demand prediction results, send financial knowledge articles or videos to help customers better understand banking products and increase their willingness to use them;

[0164] S5. The multi-channel push module pushes the formulated personalized marketing strategies to customers through multiple channels, and sends feedback information collection instructions to the effect monitoring and feedback module;

[0165] (1) Choose the appropriate push method based on the customer’s channel usage preferences;

[0166] For example, some customers prefer to receive notifications via SMS or email, while others prefer to use banking apps or social media. Push channels include SMS, email, notifications within the banking app, social media messages, etc.

[0167] (2) Select the appropriate push time based on the customer's behavior habits (including but not limited to the time when they log into the APP on a daily basis and the peak trading period);

[0168] For example, if a customer usually checks his phone at noon on weekdays, the system will push marketing content during this period to increase its open rate;

[0169] (3) Set trigger conditions for different customer groups and select effective push methods based on customer behavior preferences;

[0170] For example, VIP customers can be automatically triggered to recommend high-end products when they reach a certain balance, and potential churn customers can be triggered to contact care when they have no transactions for a certain number of days. VIP customers can be pushed through phone calls or customized emails, potential churn customers can be stimulated through SMS and APP notifications, and ordinary customers can be promoted through emails and social media ads;

[0171] S6. After the marketing campaign is launched, the effect monitoring and feedback module monitors the customer's response behavior in real time, automatically collects and analyzes customer feedback data, measures the effectiveness of the marketing strategy, compares the effects of different marketing strategies or push methods through A / B testing, and transmits the feedback data to the marketing strategy recommendation module;

[0172] For example, the marketing strategy recommendation module formulates two different promotional activities, and the effect monitoring and feedback module analyzes customer responses to different activities, determines the most effective strategy, and outputs feedback results;

[0173] Among them, feedback data includes but is not limited to the opening rate, click-through rate and conversion rate of pushed content (i.e. whether to purchase products or apply for services);

[0174] S7, Marketing strategy recommendation module continuously adjusts and optimizes marketing strategies through continuous analysis of customer feedback;

[0175] For example: If the conversion rate of a certain type of marketing content is low, the marketing strategy recommendation module automatically generates a new strategy or changes the push method.

[0176] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A multi-channel banking marketing data processing method, characterized in that: The specific implementation steps include the following: S1. The business marketing data collection module obtains customer-related data and various market analysis data from multiple channels in real time, aggregates them to obtain business marketing data, and generates exclusive label information for the business marketing data, and then transmits {exclusive label information, data type identifier, marketing data} to the business marketing data processing module; S2. The business marketing data processing module checks the source of the received marketing data, distinguishes the customer-related data and the market analysis data in the marketing data according to the data type identifier, and then transmits the customer-related data to the customer portrait generation module, and transmits the market analysis data to the marketing strategy recommendation module; S3, the customer portrait generation module uses customer identity matching data, uses the K-means algorithm to classify customers, analyzes customer behavior through the SARIMA model, predicts future trends, uses the random forest model to predict customer potential needs, generates accurate customer portraits, and transmits the customer portraits to the marketing strategy recommendation module; S4, the marketing strategy recommendation module uses the pre-trained sentiment analysis model based on customer classification and demand analysis, combined with the BERT pre-trained model and CNN network, outputs the sentiment analysis results, formulates targeted marketing strategies based on them, and then transmits the marketing strategies to the multi-channel push module; S5. The multi-channel push module pushes the formulated personalized marketing strategies to customers through multiple channels, and sends feedback information collection instructions to the effect monitoring and feedback module; S6. After the marketing campaign is launched, the effect monitoring and feedback module monitors the customer's response behavior in real time, automatically collects and analyzes customer feedback data, measures the effectiveness of the marketing strategy, and compares the effects of different marketing strategies or push methods through A / B testing, and then transmits the feedback data to the marketing strategy recommendation module; S7. The marketing strategy recommendation module continuously adjusts and optimizes marketing strategies through continuous analysis of customer feedback.

2. A multi-channel banking marketing data processing method according to claim 1, characterized in that: The specific implementation steps for classifying customers using the K-means algorithm are as follows: S21. The customer portrait generation module randomly selects K data points as the initial cluster centers. Each cluster center represents the centroid of a cluster, and its vector is expressed as: C k ={c k1 , c k2 , ..., c kf , ..., c kn }; Among them, c kf represents the initial value of the kth cluster center on the fth feature, and n is the total number of data; S22. For any customer CID i Data point X i , calculate its center C to each cluster k The Euclidean distance of each data point X is selected as the cluster with the smallest distance. i Assigned to the cluster center C with the smallest distance k , update the set of data points within the cluster; The calculation formula of Euclidean distance is: ； In the formula, x if Indicates customer CID i At data point X i The value of the fth feature in ; S23. Calculate the new cluster center of each cluster, that is, the mean of all data points in the cluster on each feature. The formula for updating the new cluster center is as follows: ； In the formula, S k is the set of all data points currently belonging to the kth cluster, |S k | represents the number of data points belonging to the kth cluster, is the value of the updated cluster center on the fth feature; S24, repeat S22 and S23 until the position of the cluster center no longer changes significantly, that is, the moving distance of the cluster center is less than a preset threshold value| - |<ε, or the maximum number of iterations is reached; in, is the value of the cluster center selected in the previous iteration on the fth feature, and ε is the preset minimum value, i.e., the threshold; S25. After clustering is completed, the customers are divided into three clusters.

3. A multi-channel banking marketing data processing method according to claim 2, characterized in that: The specific classification of customers into three clusters is: VIP customer cluster, potential churn customer cluster and ordinary customer cluster.

4. A multi-channel banking marketing data processing method according to claim 1, characterized in that: The training process of the SARIMA model is as follows: S41. From any client CID i Data point X i Extracting time series data of customer behavior, including but not limited to the number of transactions per month, the monthly transaction amount, and the monthly account balance changes; S42. Preprocess the raw data, handle missing data, and standardize the data to ensure that it is suitable for time series analysis; S43. Use the ADF test to detect the stationarity of the time series. If the time series is not stable, differential processing is required. The differential formula is as follows: ； In the formula, represents the difference between the value at the current time t and the value at the previous time t-1, Y(t) represents the customer behavior data at the current time t, and Y(t-1) represents the customer behavior data at the previous time t-1; S44. Train the SARIMA model. The model learns the seasonal patterns and trends in the time series based on the input customer behavior data. The model expression is: SARIMA(p,d,q)(p',d',q',s); Where p is the autoregressive order, d is the non-seasonal difference order, q is the moving average order, p' is the seasonal autoregressive order, d' is the seasonal difference order, q' is the seasonal moving average order, and s is the length of the season; S45. Deploy the trained SARIMA model to the customer portrait generation module.

5. A multi-channel banking marketing data processing method according to claim 1, characterized in that: The method for generating exclusive label information is as follows: S51, the business marketing data collection module randomly selects the temporary coding element Ecode i ∈[1,λ-1], calculate the temporary coding information Icode i =Ecode i ×P; Where P is defined in the finite field F p An elliptic curve E(F p ) on the λ-order basis point; S52, calculate the first-level label information I lable1i =x(Icode i ) (mod λ); Among them, x(Icode i ) indicates Icode i In the elliptic curve E(F p ) on the x-coordinate; S53, select random numbers α, β∈[1,λ-1], and calculate the secondary label information: I lable2i =α×I lable1i +(w+I lable1i )×Ctg (mod λ); Among them, α and β satisfy Ecode i =α×I lable1i +β×mdata i ,w is H(mdata i ), H is the hash function, Ctg is the label generation code, Ctg∈[1,λ-1], mdata i Any marketing data in binary string form; S54. Generate any marketing data mdata i Exclusive label Lable i ={β,I lable1i , I lable2i }.

6. A multi-channel banking marketing data processing method according to any one of claims 1 or 5, characterized in that: The method by which the business marketing data processing module verifies the source of the received marketing data is as follows: S61, test β, I lable1i ,I lable2i ∈[1,λ-1] is true, if not, then directly determine the marketing data mdata i The source is illegal, so the marketing data mdata is discarded i Otherwise, the following step S62 is executed; S62, calculate H(mdata i )’s Hamming weight w’; S63, calculate the following intermediate parameters: IP 1i =(I lable2i +β×mdata i ) (versus λ); IP 2i =(w'+I lable1i ) (mod λ); Ip 3i =Ip 1i ×P-Ip 2i ×Ctp; S64. Calculate verification information Ic i =x(Ip 3i ) (mod λ); Wherein, Ctp is the label parsing code, Ctp=Ctg×P; S65, if Ic i =I lable1i , then the marketing data mdata i Collected by the business marketing data collection module.

7. A multi-channel banking marketing data processing method according to claim 1, characterized in that: The data type identifier is: the data type identifier of customer-related data is defined as 0x00, and the data type identifier of market analysis data is defined as 0x01.

8. A multi-channel banking marketing data processing system, which is used to execute a multi-channel banking marketing data processing method according to any one of claims 1 to 7, characterized in that: include: Business marketing data collection module, automatically collects customer behavior and transaction data from multiple channels; The business marketing data processing module cleans, removes duplications, and formats the collected data to ensure data accuracy and consistency; The customer portrait generation module collects customer behavior and transaction data and uses AI and machine learning algorithms to mine customers' potential needs and behavioral preferences and build customer portraits. The marketing strategy recommendation module uses natural language processing technology to analyze market dynamics and changes in customer demand in real time, generate personalized marketing strategies, and automatically recommend the best marketing time, channels, and content based on different customer groups and market environments; The effect monitoring and feedback module monitors the effect of each marketing activity in real time and evaluates customer feedback and conversion rate through data analysis; The multi-channel push module selects appropriate channels for precise marketing content push based on customer portraits, and automatically delivers marketing content through multiple channels.

Images