Loan demand prediction and product recommendation method, apparatus and device, and computer program product

Abstract

The invention discloses a loan demand prediction and product recommendation method, device and equipment, and a computer program product, and the method comprises the steps: obtaining multi-source feature data of a customer, including first multi-source feature data and second multi-source feature data; according to the first multi-source feature data, utilizing a preset loan demand prediction model to predict the loan demand of the customer, and obtaining loan demand prediction results of the customer in different periods; and according to the second multi-source feature data and the loan demand prediction result, performing product recommendation by using a preset product recommendation model to obtain a product recommendation result of the customer. According to the method, loan demand mining of the customer in different periods is considered, and the loan demand condition of the customer can be known more comprehensively and deeply; the deep learning model is adopted to carry out loan demand mining and product recommendation in sequence, hidden association which is difficult to cover by manual rules is avoided, the overall efficiency of financial services is effectively improved, meanwhile, the demands of customers for different financial products are considered, and the customers are served comprehensively.

Description

Technical Field

[0001] The present application relates to the field of financial technology, and in particular to a loan demand forecasting and product recommendation method, apparatus and equipment, and a computer program product. Background Art

[0002] As a crucial component of the financial services sector, inclusive finance is committed to providing broad, affordable, and high-quality financial services to all segments of society. It plays a key role in promoting balanced economic development and enhancing overall social well-being. With the continuous development of the financial market and the increasing diversification of customer needs, the scale of inclusive finance continues to expand, encompassing individual business owners, small and micro business owners, and a broad range of low- and middle-income individuals. These customers, at different stages of their development, have diverse loan needs, ranging from short-term working capital to long-term financing for business expansion, and from traditional production and operation loans to emerging consumer upgrade loans. Accurately identifying and meeting these needs is crucial for financial institutions to enhance the quality of inclusive financial services and strengthen their market competitiveness.

[0003] In previous industry operations, financial institutions primarily relied on traditional methods for inclusive finance customer marketing and product recommendations. Specifically, they often selected clients based on sales staff's experience or subjective judgment. When selecting target customers, they typically applied fixed criteria based on the limited information available at the time. For example, they identified potential customer groups based on simple indicators such as their industry, tax bracket, and account balance. For these selected target customers, the recommended inclusive finance products were also relatively simple and rigid. For example, if a customer had recently incurred a large expense, they would recommend a short-term working capital loan; if the customer's industry was "retail or wholesale," they would simply recommend supply chain financing products. This recommendation approach lacked in-depth analysis of individual customer needs and their dynamic evolution, making it difficult to meet their diverse financial needs.

[0004] However, the above method has at least the following technical problems:

[0005] (1) Incomplete demand exploration

[0006] Existing technologies primarily focus on identifying the immediate loan needs of inclusive finance clients, significantly neglecting their potential long-term loan needs. In practice, meeting loan requirements of a certain amount often requires a lengthy process involving multiple steps and complex approval procedures. Failure to proactively identify clients and promptly introduce appropriate financial products to them can lead to late discovery of their needs, missing the optimal opportunity to serve them. Furthermore, current product recommendations for inclusive finance clients by industry peers overlook their need for value-added financial products. Value-added financial products, such as investment and wealth management planning and risk management consulting, are crucial for preserving and increasing the value of clients' assets and effectively managing risk. Existing technologies fail to fully serve clients, limiting the depth and breadth of inclusive financial services.

[0007] (2) Data sparsity leads to prediction difficulties

[0008] Before applying for a loan, banks retain relatively little information about potential inclusive finance clients. This limited information fails to fully and accurately reflect their true financial status, creditworthiness, and potential loan needs. This creates considerable difficulty in predicting loan demand. Due to a lack of sufficient data support, existing technologies struggle to build effective forecasting models, resulting in low accuracy in predicting customer loan demand and a failure to provide a reliable basis for financial institutions' marketing decisions.

[0009] (3) Recommendation efficiency is low and relies on manual labor

[0010] Existing technology primarily relies on business experience to recommend products to inclusive finance clients. This leads to a strong correlation between the success rate of recommendations and the professional expertise of marketers. Differences in the skills and experience of different marketers can lead to widely varying recommendations for the same client. From a bank's overall perspective on inclusive finance, this reliance on manual experience is inefficient. On the one hand, it requires marketers to expend considerable time and effort analyzing and assessing customer information, increasing operational costs. On the other hand, due to human interference, the accuracy and consistency of recommendation results are difficult to guarantee, making them incapable of meeting the demands of large-scale inclusive finance development.

[0011] To sum up, existing technologies have many shortcomings in mining, predicting and matching loan demands of inclusive finance customers, and cannot meet the current financial market's requirements for efficient, accurate and comprehensive inclusive financial services. Summary of the Invention

[0012] The embodiments of the present application provide a loan demand forecasting and product recommendation method, apparatus and device, and computer program product to improve the accuracy of financial product recommendations and customer satisfaction.

[0013] The embodiments of this application adopt the following technical solutions:

[0014] In a first aspect, an embodiment of the present application provides a method for predicting loan demand and recommending products, the method comprising:

[0015] Acquire multi-source feature data of a customer, wherein the multi-source feature data includes first multi-source feature data and second multi-source feature data;

[0016] Predicting the customer's loan demand using a preset loan demand prediction model based on the first multi-source feature data to obtain a loan demand prediction result for the customer, the loan demand prediction result including loan demand prediction results for the customer at different time periods;

[0017] Based on the second multi-source feature data and the loan demand prediction result of the customer, a product recommendation is performed using a preset product recommendation model to obtain a product recommendation result for the customer.

[0018] Optionally, the preset loan demand prediction model is a network structure of a twin-tower multi-objective model, and the twin-tower multi-objective model is used to predict the customer's loan demand and loan amount in different periods. The preset product recommendation model is a network structure of a twin-tower single-objective model, and the twin-tower single-objective model is used to recommend financial products corresponding to the customer.

[0019] Optionally, the preset loan demand prediction model is trained in the following manner:

[0020] Constructing training sample data of a loan demand prediction model and corresponding true value data of loan demand, wherein the training sample data of the loan demand prediction model includes first multi-source feature data;

[0021] Inputting the training sample data of the loan demand prediction model into the loan demand prediction model to obtain a loan demand prediction result, wherein the loan demand prediction result includes loan demand prediction results of customers in different periods;

[0022] The prediction loss of the loan demand prediction model is calculated according to the loan demand prediction result and the corresponding true value data of the loan demand, and the parameters of the loan demand prediction model are iteratively updated using the prediction loss of the loan demand prediction model to obtain a trained loan demand prediction model.

[0023] Optionally, the dual-tower multi-objective model includes a static tower network and a time-series tower network, the first multi-source feature data includes static feature data and time-series feature data, and inputting the training sample data of the loan demand prediction model into the loan demand prediction model to obtain the loan demand prediction result includes:

[0024] Inputting the static characteristic data into the static tower network to obtain a loan demand prediction result of the static tower network;

[0025] Inputting the time series characteristic data into the time series tower network to obtain the loan demand prediction result of the time series tower network;

[0026] The loan demand prediction result of the static tower network and the loan demand prediction result of the sequential tower network are combined as the loan demand prediction result of the loan demand prediction model.

[0027] Optionally, the preset product recommendation model is trained in the following manner:

[0028] Constructing training sample data for a product recommendation model and corresponding product recommendation true value data, wherein the training sample data for the product recommendation model includes the second multi-source feature data and the loan demand feature data;

[0029] Inputting the training sample data of the product recommendation model into the product recommendation model to obtain a product recommendation result;

[0030] The recommendation loss of the product recommendation model is calculated according to the product recommendation results and the corresponding product recommendation true value data, and the recommendation loss of the product recommendation model is used to iteratively update the parameters of the product recommendation model to obtain a trained product recommendation model.

[0031] Optionally, the dual-tower single-objective model includes a user tower network and a product tower network, the second multi-source feature data includes customer key feature data and product feature data, and inputting the training sample data of the product recommendation model into the product recommendation model to obtain a product recommendation result includes:

[0032] Inputting the customer key feature data and the loan demand feature data into the user tower network to obtain feature processing results of the user tower network;

[0033] Inputting the product characteristic data into the product tower network to obtain characteristic processing results of the product tower network;

[0034] The product recommendation result is determined according to the feature processing result of the user tower network and the feature processing result of the product tower network.

[0035] Optionally, determining the product recommendation result according to the feature processing result of the user tower network and the feature processing result of the product tower network includes:

[0036] performing a correlation calculation on the feature processing result of the user tower network and the feature processing result of the product tower network to obtain a correlation calculation result;

[0037] The product recommendation result is determined according to the correlation calculation result.

[0038] In a second aspect, an embodiment of the present application further provides a loan demand prediction and product recommendation device, the loan demand prediction and product recommendation device comprising:

[0039] an acquiring unit, configured to acquire multi-source feature data of a customer, wherein the multi-source feature data includes first multi-source feature data and second multi-source feature data;

[0040] a loan demand prediction unit, configured to predict the customer's loan demand using a preset loan demand prediction model based on the first multi-source feature data, and obtain a loan demand prediction result for the customer, wherein the loan demand prediction result includes loan demand prediction results for the customer at different time periods;

[0041] A product recommendation unit is configured to perform product recommendations based on the second multi-source feature data and the loan demand prediction result of the customer using a preset product recommendation model to obtain a product recommendation result for the customer.

[0042] In a third aspect, an embodiment of the present application further provides a device, including:

[0043] A processor; and a memory arranged to store computer-executable instructions, which, when executed, cause the processor to perform any of the aforementioned loan demand prediction and product recommendation methods.

[0044] In a fourth aspect, an embodiment of the present application further provides a computer program product, comprising a computer program / instruction, which, when executed by a processor, implements any of the aforementioned loan demand prediction and product recommendation methods.

[0045] At least one of the above-mentioned technical solutions employed in the embodiments of the present application can achieve the following beneficial effects: The loan demand prediction and product recommendation method of the embodiments of the present application first obtains multi-source feature data of a customer, the multi-source feature data including first multi-source feature data and second multi-source feature data; then, based on the first multi-source feature data, uses a preset loan demand prediction model to predict the customer's loan demand, obtaining a loan demand prediction result for the customer, which includes loan demand prediction results for the customer at different time periods; finally, based on the second multi-source feature data and the customer's loan demand prediction result, uses a preset product recommendation model to make product recommendations, obtaining a product recommendation result for the customer. The loan demand prediction and product recommendation method of the present application takes into account the loan demand of inclusive finance customers at different time periods, breaking through the limitation of relying solely on business experience to discover customers' long-term loan needs, and can provide a more comprehensive and in-depth understanding of customers' loan needs. A deep learning model is used to sequentially perform loan demand mining and product recommendation. The deep learning model can automatically capture the complex nonlinear relationships between customer behavior, financial data, industry characteristics, etc., avoiding implicit correlations that are difficult to cover with manual rules, effectively improving the overall efficiency of financial services, while also taking into account customers' needs for different financial products and achieving comprehensive customer service. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0047] Figure 1 A flowchart of a loan demand prediction and product recommendation method in an embodiment of the present application;

[0048] Figure 2 A schematic diagram of a training framework for a loan demand prediction model according to an embodiment of the present application;

[0049] Figure 3 Schematic diagram of a training framework for a product recommendation model in an embodiment of the present application;

[0050] Figure 4 This is a structural diagram of a loan demand prediction and product recommendation device in an embodiment of the present application;

[0051] Figure 5 This is a structural diagram of a device in an embodiment of the present application. DETAILED DESCRIPTION

[0052] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0053] The following describes in detail the technical solutions provided by various embodiments of the present application in conjunction with the accompanying drawings.

[0054] This application embodiment provides a loan demand prediction and product recommendation method, such as Figure 1 As shown, a flow chart of a method for predicting loan demand and recommending products in an embodiment of the present application is provided. The method for predicting loan demand and recommending products includes at least the following steps S110 to S130:

[0055] Step S110 : acquiring multi-source feature data of a customer, where the multi-source feature data includes first multi-source feature data and second multi-source feature data.

[0056] Collecting relevant customer data from multiple channels constitutes multi-source feature data. Multi-source feature data is divided into first-source feature data and second-source feature data. First-source feature data primarily focuses on predicting customer loan demand, while second-source feature data primarily focuses on product recommendations. Data sources include not only customer transaction flows, payroll services, corporate online banking, and international settlements, but can also be integrated with the credit information on small and micro-enterprises' funds flows in the inclusive finance sector publicly released by the People's Bank of China. This allows for more comprehensive data, providing rich foundational data support for subsequent steps.

[0057] Step S120: predicting the customer's loan demand using a preset loan demand prediction model based on the first multi-source feature data to obtain a loan demand prediction result for the customer, wherein the loan demand prediction result includes loan demand prediction results for the customer at different periods.

[0058] Using a preset loan demand prediction model, the first multi-source feature data obtained in step S110 is input into the model. The model performs calculations and analyses based on this data to predict the customer's loan demand at different time periods, ultimately outputting a loan demand prediction result for the customer. This result covers the customer's loan demand prediction for different time periods, such as short-term (1 month), medium-term (2-3 months), and long-term (1 year).

[0059] Step S130 , performing product recommendations using a preset product recommendation model based on the second multi-source feature data and the loan demand prediction result of the customer, to obtain a product recommendation result for the customer.

[0060] The second multi-source feature data acquired in step S110 and the customer loan demand prediction results obtained in step S120 are jointly input into the preset product recommendation model. The product recommendation model comprehensively considers this information and, through its internal algorithms and logic, analyzes the customer's needs and potential financial products of interest to generate a product recommendation strategy for the customer. Based on the product recommendation strategy generated by the two-stage model, a product marketing list is generated and published to the maintainer's workstation.

[0061] The loan demand forecasting and product recommendation methods of the present application embodiment take into account the exploration of loan demand at different stages of inclusive finance customers, breaking through the limitation of relying solely on business experience to discover customers' long-term loan needs, and achieving a more comprehensive and in-depth understanding of customers' loan needs. A deep learning model is used to sequentially explore loan demand and recommend products. The deep learning model can automatically capture the complex nonlinear relationships between customer behavior, financial data, industry characteristics, etc., avoiding implicit associations that are difficult to cover with manual rules, effectively improving the overall efficiency of financial services, while also taking into account customers' needs for different financial products, and achieving comprehensive customer service.

[0062] In some embodiments of the present application, obtaining multi-source feature data of a customer includes:

[0063] (1) Multi-source data integration

[0064] 1) In-bank data integration for corporate clients

[0065] Collect basic information retained by corporate clients within the bank, including business registration information such as company name, registered address, and registered capital, as well as industry classification information, to clarify the industry sector to which the company belongs. Integrate corporate clients' settlement transaction flow data, recording in detail the amount, time, counterparty, and other information for each transaction to reflect the company's capital flow. Incorporate corporate clients' payroll data to understand the frequency and amount of salary payments to employees, which indirectly reflects the company's operating conditions and payment capabilities. Collect tax data, including corporate tax return records and tax amounts, to reflect the company's tax compliance and operating efficiency. Integrate social security payment records to clarify the number of employees and the base number for social security contributions, reflecting the company's level of employee protection and its fulfillment of social responsibility. Incorporate wealth product holdings data to record the various wealth products held by the company, such as wealth management products and funds, to understand the company's asset allocation.

[0066] 2) Integration of highly relevant personal customer data

[0067] Collect basic information on individual customers that is highly relevant to the company, such as the name, age, and contact information of the company's head, to build a basic profile of the individual customer. Integrate individual customer transaction data, record every transaction within the bank, and analyze their spending habits and cash flow. Incorporate data on individual customer wealth product holdings to understand the types and amounts of wealth products held by individual customers and reflect their asset status. Collect data related to individual customer loans, including loan application records, loan amounts, and repayment status, to assess individual customers' credit status and repayment ability.

[0068] 3) Product data integration

[0069] Collect basic information on historically traded products, including product types such as loan products and wealth management products, product terms such as short-term, medium-term, and long-term, product interest rates, and other relevant terms and conditions of the products, to provide data support for subsequent analysis of the relationship between products and customer needs.

[0070] 4) Integration of key market index data

[0071] Collect key market index data, such as the "Inclusive Finance - Small and Micro Loan Index," which reflects the overall performance and development trends of the small and micro loan market within the inclusive finance sector. Incorporate "Loan Market Reference Rate" data to understand the market benchmark for loan interest rates, providing a reference for analyzing client loan needs and costs.

[0072] 5) People's Bank of China credit information integration

[0073] Integrating the credit information on cash flows of small and micro enterprises in the inclusive finance sector publicly released by the People's Bank of China, this information reflects the credit status of small and micro enterprises in terms of cash flow, including the stability of cash flow, credit records, etc., providing an important basis for assessing the credit risk of corporate customers.

[0074] (2) Feature Engineering

[0075] 1) Data cleaning

[0076] Check the integrated multi-source data to identify and address missing values. For cases with a small number of missing values, use the mean, median, or mode to fill them in. For cases with a large number of missing values, consider deleting the data record or using more complex interpolation methods to fill them in.

[0077] Detect and process outliers by setting reasonable thresholds or using statistical methods to identify outliers. Outliers can be corrected or deleted according to actual conditions to ensure the accuracy and reliability of the data.

[0078] Unify data formats and standardize data formats from different sources, such as date format, currency format, etc., so that the data has a consistent representation, which facilitates subsequent analysis and processing.

[0079] 2) Data preprocessing

[0080] Transform the original data, such as standardizing or normalizing the numerical data, scaling the data to a specific range, such as [0, 1] or [-1, 1], to improve the data distribution and avoid affecting the model performance due to different data dimensions.

[0081] Encode categorical data and convert text-type categorical variables into numerical variables, such as using one-hot encoding or label encoding, so that the model can process categorical data.

[0082] 3) Feature extraction

[0083] New indicators are derived based on business implications and data time windows. For example, based on a company's settlement transaction flow data, indicators such as the cash inflow and outflow ratio and capital turnover rate over a specific period of time are calculated to reflect the company's capital operation efficiency. Individual customer transaction data and loan-related data are combined to calculate indicators such as the individual customer's consumer credit ratio and repayment delinquency rate to assess their credit risk and spending power. Product data and key market index data are used to construct correlation indicators between products and market indices to analyze product performance and changes in customer demand under different market environments.

[0084] Through the above multi-source data fusion and feature engineering processing, we can finally obtain the customer's multi-source feature data, providing high-quality data support for subsequent loan demand prediction and product recommendations.

[0085] In some embodiments of the present application, the preset loan demand prediction model is a network structure of a dual-tower multi-objective model, and the dual-tower multi-objective model is used to predict the customer's loan demand and loan amount in different periods. The preset product recommendation model is a network structure of a dual-tower single-objective model, and the dual-tower single-objective model is used to recommend financial products corresponding to the customer.

[0086] The loan demand prediction model uses a dual-tower structure, with each tower processing different input features. One tower can focus on processing static features related to the customer's current status, while the other tower focuses on processing the customer's temporal feature data. This design can separately mine information from different types of features and reduce interference between features.

[0087] The loan demand forecasting model sets multiple prediction objectives, namely, predicting customer loan demand and loan amounts over different time periods (short, medium, and long term). During model training, optimization algorithms simultaneously adjust model parameters to achieve good prediction results across multiple objectives. For example, a multi-task learning loss function is used to comprehensively consider the errors in loan demand forecasts and loan amount forecasts over different time periods to train and optimize the model.

[0088] The product recommendation model also uses a dual-tower architecture. One tower focuses on processing customer characteristics, such as basic information, transaction history, wealth status, and loan demand predictions; the other tower focuses on processing financial product characteristics, such as product type, maturity, interest rate, and return characteristics. This dual-tower architecture extracts feature representations for both customers and products.

[0089] The goal of a product recommendation model is to recommend the most suitable financial product to a customer, thus falling into the category of single-objective recommendation. During model training, model parameters are optimized to accurately predict a customer's preference for or likelihood of purchasing different financial products.

[0090] The dual-tower, multi-objective model can simultaneously predict both customer loan demand and loan amounts over different time periods, overcoming the limitations of traditional single-objective models that can only predict a single indicator. This model provides financial institutions with more comprehensive and in-depth information on customer loan demand. By integrating multi-source feature data and designing a dual-tower architecture, the model can fully explore the complex relationships between customer characteristics and market environment characteristics, improving the accuracy of loan demand forecasts.

[0091] The Twin Tower Single-Objective Model combines loan demand forecasts with customer and product characteristics to make product recommendations, providing personalized financial product recommendations. The model considers a variety of customer characteristics and needs, as well as the characteristics of financial products, making recommendations more aligned with customers' actual circumstances and preferences, thereby increasing customer acceptance and satisfaction with the recommended products.

[0092] In some embodiments of the present application, the preset loan demand prediction model is trained in the following manner: constructing training sample data of the loan demand prediction model and corresponding loan demand true value data, the training sample data of the loan demand prediction model including first multi-source feature data; inputting the training sample data of the loan demand prediction model into the loan demand prediction model to obtain a loan demand prediction result, the loan demand prediction result including the loan demand prediction results of the customer at different periods; calculating the prediction loss of the loan demand prediction model based on the loan demand prediction result and the corresponding loan demand true value data, and using the prediction loss of the loan demand prediction model to iteratively update the parameters of the loan demand prediction model to obtain a trained loan demand prediction model.

[0093] When training the loan demand prediction model, you can first collect multi-source feature data of customers according to the above embodiment as the basis for constructing training samples. These data cover the basic information of corporate customers retained in the bank (such as industrial and commercial registration, industry classification), settlement transaction flow, salary data, tax data, social security payment records, wealth product holdings, etc.; basic information of personal customers with strong correlation with the enterprise, transaction data, wealth product holdings, loan-related data, etc.; product data, including basic information of historically traded products; key market indexes, including "Inclusive Finance-Small and Micro Loan Index", "Loan Market Reference Rate", etc.; credit information on the capital flow of small and micro enterprises in the inclusive finance line publicly released by the People's Bank of China, etc.

[0094] True loan demand data can be obtained through a variety of reliable channels. For existing loan customers, the actual loan amount and term of their applications and approvals can be used as a reference for the true value of short-term, medium-term, or long-term loan demand. For example, if a customer recently applied for and received a short-term working capital loan, the loan amount and term can serve as the true value of their short-term loan demand.

[0095] For clients who haven't yet applied for a loan but have potential needs, a comprehensive assessment can be conducted based on the experience of industry experts, market research data, and the client's historical operating data and financial status. For example, based on the average growth rate of the client's industry, the client's revenue growth trends in recent years, and their balance sheet, we can estimate the loan amount and term they are likely to need at different times in the future, using this as the true value of their loan demand.

[0096] The training sample data of the constructed loan demand prediction model, namely the first multi-source feature data, is input into the preset loan demand prediction model. The model adopts a network structure of a double-tower multi-objective model, which can handle complex feature relationships and predict multiple targets.

[0097] After receiving input data, the model processes different types of data using its internal dual-tower architecture. One tower focuses on analyzing the customer's current status, such as basic information and transaction behavior; the other tower processes transaction-related temporal features, such as historical transaction flow data. After calculation and analysis, the model outputs forecasts of the customer's loan demand over different time periods (short, medium, and long term), including the predicted loan demand (yes / no) and the loan amount.

[0098] The model's output of loan demand predictions is compared with the corresponding true loan demand data. An appropriate loss function is used to calculate the difference between the two, known as the prediction loss. Common loss functions include the mean squared error (MSE), which measures prediction accuracy by averaging the squares of the differences between the predicted and true values.

[0099] Using the calculated prediction loss, the backpropagation algorithm and optimization algorithms (such as stochastic gradient descent (SGD) and the Adam optimization algorithm) are used to iteratively update the parameters of the loan demand prediction model. The backpropagation algorithm gradually adjusts the parameter values ​​from the output layer to the input layer based on the gradient of the loss function with respect to the model parameters, thereby gradually reducing the loss function value. The optimization algorithm controls the step size and direction of the parameter update to improve the model's convergence speed and stability. Through multiple iterative updates, the model's performance is continuously optimized until the model's prediction loss reaches a preset threshold or meets other stopping conditions, resulting in a trained loan demand prediction model.

[0100] By constructing comprehensive training sample data that incorporates information about customers, the market environment, and products, the model learns richer feature relationships. Using real loan demand data as ground truth for training, combined with an appropriate loss function and optimization algorithm to continuously adjust model parameters, this effectively reduces prediction error, improves the accuracy of customer loan demand forecasts over time, and provides a reliable basis for financial institutions to formulate reasonable credit strategies. The dual-tower, multi-objective model structure can process different types of feature data separately, reducing interference between features and improving the model's adaptability to complex data.

[0101] Accurate loan demand forecasts can help financial institutions gain a deeper understanding of each customer's specific needs and provide them with personalized credit products and services. For example, based on short-term, medium-term, and long-term loan demand forecasts, financial institutions can tailor loan plans with varying terms and amounts to meet the funding needs of customers at different stages of their development, thereby improving customer satisfaction and loyalty.

[0102] In some embodiments of the present application, the dual-tower multi-objective model includes a static tower network and a time-series tower network, the first multi-source feature data includes static feature data and time-series feature data, and the inputting the training sample data of the loan demand prediction model into the loan demand prediction model to obtain the loan demand prediction result includes: inputting the static feature data into the static tower network to obtain the loan demand prediction result of the static tower network; inputting the time-series feature data into the time-series tower network to obtain the loan demand prediction result of the time-series tower network; and merging the loan demand prediction result of the static tower network and the loan demand prediction result of the time-series tower network as the loan demand prediction result of the loan demand prediction model.

[0103] like Figure 2As shown, a schematic diagram of the training framework of a loan demand prediction model in an embodiment of the present application is provided. The first multi-source feature data in the embodiment of the present application is divided into static feature data (X1) and time series feature data (X2). The static feature data includes all feature variables in the current state of the enterprise, such as the basic information of the enterprise, the transaction flow in the current state, the wealth product holdings, etc.; the time series feature data is the historical transaction data of the enterprise, such as the daily transaction flow variables in the past year, which reflects the changes in the capital flow of the enterprise over time.

[0104] A static tower network model based on a deep neural network (DNN) was constructed. This model was used to process static feature data (X1). For the six predictor variables of the sampled corporate customers (Y1: short-term loan availability, Y2: short-term loan amount, Y3: medium-term loan availability, Y4: medium-term loan amount, Y5: long-term loan availability, and Y6: long-term loan amount), a deep learning model was established to predict Y1 to Y6 based on X1.

[0105] The static feature data (X1) is input into the Static Tower Network. The model uses nonlinear transformations of multiple layers of neurons to explore the complex relationships between static features and outputs the Static Tower Network's loan demand prediction results for the six predictor variables (Y1-Y6).

[0106] A time series tower network model based on the long short-term memory (LSTM) network was constructed. The LSTM model effectively processes time series data and captures long-term dependencies within the data. This model uses the six predictor variables (Y1-Y6) of the enterprise customers in the sample to establish a time series model where X2 predicts Y1 to Y6.

[0107] The time series feature data (X2) is input into the Time Series Tower Network. The LSTM model uses its unique gating mechanism to learn and analyze the time series data and outputs the Time Series Tower Network's loan demand prediction results for the six predictor variables (Y1-Y6).

[0108] The loan demand forecast results from the static tower network and the time-series tower network are combined. Each sample receives prediction results for six predictor variables from both the static and time-series towers, resulting in a total of 12 prediction results for each sample. These 12 combined prediction results are output in parallel as the final loan demand forecast result of the loan demand forecast model for subsequent analysis and decision-making.

[0109] The static tower network, using a DNN model, can deeply explore the complex nonlinear relationships between static enterprise features, such as the correlation between basic information and financial status. The time-series tower network, using an LSTM model, can effectively capture the long-term dependencies within dynamic time-series features, such as how daily transaction flows change over time. The combined dual-tower architecture comprehensively captures various data features, improving the accuracy of forecasts of customer loan demand.

[0110] This model can simultaneously predict whether a company will seek loans, as well as the corresponding loan amounts, in the short, medium, and long term. This multi-objective forecasting approach can provide financial institutions with more comprehensive and detailed information on customer loan needs, meeting their needs in various business scenarios, such as developing long-term credit plans and short-term capital allocation.

[0111] In some embodiments of the present application, the preset product recommendation model is trained in the following manner: constructing training sample data of the product recommendation model and corresponding product recommendation true value data, the training sample data of the product recommendation model including second multi-source feature data and loan demand feature data; inputting the training sample data of the product recommendation model into the product recommendation model to obtain a product recommendation result; calculating the recommendation loss of the product recommendation model based on the product recommendation result and the corresponding product recommendation true value data, and using the recommendation loss of the product recommendation model to iteratively update the parameters of the product recommendation model to obtain a trained product recommendation model.

[0112] When training a product recommendation model, the second multi-source feature data of customers can also be obtained according to the aforementioned embodiment. This second multi-source feature data can be further extracted through key features based on the multi-source feature data integrated in the aforementioned embodiment. That is, compared to the first multi-source feature data in the aforementioned embodiment, the second multi-source feature data primarily focuses on the key feature data of corporate customers, thereby improving the training efficiency of the product recommendation model while ensuring product recommendation accuracy.

[0113] Using the previously trained loan demand prediction model, we predict customer data and obtain the customer's loan demand prediction results in different periods (short-term, medium-term, and long-term), including information such as whether to apply for a loan and the loan amount, which is used as loan demand feature data.

[0114] The true value data for product recommendations can be determined by the financial products actually purchased or selected by customers. For example, for customers who have already purchased financial products, the products they actually selected can be used as the true value. For new customers, market research, customer profiles, and the preferences of similar customer groups can be combined to estimate the products they may be interested in and use as the true value data for product recommendations.

[0115] The training sample data for the constructed product recommendation model—the second multi-source feature data and loan demand feature data—is input into the pre-set product recommendation model. This model utilizes a dual-tower, single-objective network structure, capable of processing complex feature relationships and making product recommendations. After receiving the input data, the model's internal dual-tower structure processes different types of data separately. One tower focuses on analyzing key customer characteristics, such as basic information, transaction history, wealth status, and loan demand characteristics; the other tower processes financial product characteristics, such as product type, term, interest rate, and yield characteristics. After calculation and analysis, the model outputs the most suitable financial product for the customer, which is the product recommendation result.

[0116] Compare the product recommendations output by the model with the corresponding ground-truth product recommendations and use an appropriate loss function to calculate the difference between the two, known as the recommendation loss. For example, you can use the cross-entropy loss function, which is commonly used in classification problems and measures the difference between the model's predicted probability distribution and the true distribution.

[0117] Using the calculated recommendation loss, the backpropagation algorithm and optimization algorithm are used to iteratively update the parameters of the product recommendation model. Based on the gradient of the loss function with respect to the model parameters, the backpropagation algorithm gradually adjusts the parameter values ​​from the output layer to the input layer, gradually reducing the loss function value. Through multiple iterative updates, the model's performance is continuously optimized until the model's recommendation loss reaches a preset threshold or meets other stopping conditions, resulting in a trained product recommendation model.

[0118] By integrating secondary multi-source feature data with loan demand feature data, the model can fully understand all aspects of a customer's situation and needs. Training with real product recommendation data enables the model to learn the mapping relationship between customer characteristics and products, thereby providing customers with accurate financial product recommendations and improving customer acceptance and satisfaction with recommended products.

[0119] Taking into account the characteristics of a customer's loan needs, the model can recommend matching financial products based on the customer's loan needs at different times. This personalized recommendation method can better meet the customer's actual needs, enhance the personalized service level of financial institutions, and strengthen the stickiness between customers and financial institutions.

[0120] Trained product recommendation models can directly provide recommendations to financial institutions' inclusive finance managers, reducing the time and effort required to manually screen and recommend products. Based on the model's recommendations, inclusive finance managers can quickly provide customers with suitable products, improving transaction efficiency and ultimately enhancing the overall efficiency of the bank's inclusive finance business.

[0121] In some embodiments of the present application, the dual-tower single-objective model includes a user tower network and a product tower network, the second multi-source feature data includes customer key feature data and product feature data, and the inputting the training sample data of the product recommendation model into the product recommendation model to obtain the product recommendation result includes: inputting the customer key feature data and the loan demand feature data into the user tower network to obtain the feature processing result of the user tower network; inputting the product feature data into the product tower network to obtain the feature processing result of the product tower network; and determining the product recommendation result based on the feature processing result of the user tower network and the feature processing result of the product tower network.

[0122] like Figure 3 The figure shows a training framework diagram of a product recommendation model in an embodiment of the present application. The dual-tower single-target model in the embodiment of the present application includes a user tower network and a product tower network. Both the user tower and the product tower adopt a neural network model with a DNN as the main body, and each inserts a SENet (Squeeze-and-Excitation Network) module in the hidden layer to weaken data noise, and strengthens important features through dynamic weight adjustment.

[0123] Second, multi-source feature data is divided into customer key feature data and product feature data. Customer key feature data covers key features such as basic customer information, transaction history, and wealth status. Product feature data includes basic information such as product type, term, interest rate, and yield characteristics. Product types here include not only inclusive loan products but also wealth-related products such as wealth management funds and trusts. Simultaneously, customer loan demand feature data is derived using a trained loan demand prediction model.

[0124] The user tower network inputs key customer feature data and loan demand data. This data contains multiple key aspects of the customer's information and their loan demand over time, comprehensively reflecting their financial needs and characteristics. The user tower network uses multi-layered neural network nonlinear transformations to deeply mine and analyze the input key customer feature data and loan demand data, extracting a deep representation of the customer's characteristics and outputting the user tower network's feature processing results. This result is an abstract representation of the customer's comprehensive characteristics, better representing the customer's financial needs and preferences.

[0125] Product feature data, which describes the various attributes and characteristics of a financial product, is input into the Product Tower Network. The Product Tower Network processes this input, extracting a deep representation of the product's characteristics through computation and transformation within neurons. The output is the Product Tower Network's feature processing results, which reflect the product's core characteristics and potential appeal.

[0126] Finally, the relationship between the feature processing results of the user tower network and the feature processing results of the product tower network is further analyzed to determine the final product recommendation results.

[0127] The User Tower Network comprehensively considers key customer characteristics and loan demand characteristics, enabling a comprehensive and in-depth understanding of their financial needs and preferences. The Product Tower Network focuses on extracting product features and accurately grasping their attributes and advantages. By fully considering individual customer differences and dynamic loan needs, recommendations are made based on a combination of their diverse characteristics and loan demand profiles, significantly enhancing the level of personalization and increasing customer acceptance and satisfaction with recommended products.

[0128] In some embodiments of the present application, determining the product recommendation result based on the feature processing results of the user tower network and the feature processing results of the product tower network includes: performing correlation calculation on the feature processing results of the user tower network and the feature processing results of the product tower network to obtain a correlation calculation result; and determining the product recommendation result based on the correlation calculation result.

[0129] Continue to refer Figure 3 When determining the final product recommendation results, a correlation calculation can be performed between the feature processing results output by the user tower network and the feature processing results output by the product tower network. By calculating the similarity or matching degree between the two, the degree of association between the customer and each financial product can be determined. For example, methods such as cosine similarity can be used to measure the similarity between user feature vectors and product feature vectors. Based on the calculated correlation results, the financial product with the highest correlation is selected as the product recommendation result, providing customers with product recommendations that best meet their needs and characteristics.

[0130] This efficient and accurate correlation calculation method can quickly process large amounts of customer and product feature data to quickly determine recommendation results. Compared to traditional manual recommendation methods, this method significantly improves recommendation efficiency while ensuring quality and reducing errors and subjectivity caused by human factors.

[0131] In some embodiments of the present application, indicators such as Precision@N (precision of the top N recommended products), Recall@N (recall rate of the top N recommended products), MAP (average precision mean), and MRR (mean reciprocal ranking) can be used to evaluate the model results, so as to continuously adjust the model variables and parameters until each indicator achieves the expected effect.

[0132] In summary, the loan demand forecasting and product recommendation method of this application has achieved at least the following technical effects:

[0133] First of all, this application takes into account the exploration of inclusive finance customers' short-term, medium-term and long-term loan needs, solving the problem that it is difficult to discover their possible long-term loan needs based on business experience. In the product recommendation process, it also takes into account the inclusive finance customers' needs for value-added financial products, thus providing comprehensive services to customers.

[0134] Secondly, this application not only makes full use of the transaction flow, payroll payment, corporate online banking, international settlement and other data information left by inclusive finance customers in the bank, but can also be integrated with the credit information of small and micro enterprises in the inclusive finance line publicly released by the People's Bank of China, so the scope of data information is wider.

[0135] Finally, this application uses a deep learning model to identify loan needs and recommend products for inclusive finance clients. This model automatically captures the complex, nonlinear relationships between customer behavior, financial data, and industry characteristics, avoiding implicit correlations that are difficult to capture with manual rules. The resulting recommendation list is directly accessible to branch inclusive finance managers, effectively improving the overall efficiency of the bank's inclusive finance business.

[0136] The embodiment of the present application also provides a loan demand prediction and product recommendation device 400, such as Figure 4 , a schematic diagram of the structure of a loan demand prediction and product recommendation device in an embodiment of the present application is provided. The loan demand prediction and product recommendation device 400 includes: an acquisition unit 410, a loan demand prediction unit 420, and a product recommendation unit 430, wherein:

[0137] An acquiring unit 410 is configured to acquire multi-source feature data of a client, wherein the multi-source feature data includes first multi-source feature data and second multi-source feature data;

[0138] a loan demand prediction unit 420 configured to predict the customer's loan demand using a preset loan demand prediction model based on the first multi-source feature data, and obtain a loan demand prediction result for the customer, wherein the loan demand prediction result includes loan demand prediction results for the customer at different time periods;

[0139] The product recommendation unit 430 is configured to perform product recommendation based on the second multi-source feature data and the loan demand prediction result of the customer using a preset product recommendation model to obtain a product recommendation result for the customer.

[0140] In some embodiments of the present application, the preset loan demand prediction model is a network structure of a dual-tower multi-objective model, and the dual-tower multi-objective model is used to predict the customer's loan demand and loan amount in different periods. The preset product recommendation model is a network structure of a dual-tower single-objective model, and the dual-tower single-objective model is used to recommend financial products corresponding to the customer.

[0141] In some embodiments of the present application, the preset loan demand prediction model is trained in the following manner: constructing training sample data of the loan demand prediction model and corresponding loan demand true value data, the training sample data of the loan demand prediction model including first multi-source feature data; inputting the training sample data of the loan demand prediction model into the loan demand prediction model to obtain a loan demand prediction result, the loan demand prediction result including the loan demand prediction results of the customer at different periods; calculating the prediction loss of the loan demand prediction model based on the loan demand prediction result and the corresponding loan demand true value data, and using the prediction loss of the loan demand prediction model to iteratively update the parameters of the loan demand prediction model to obtain a trained loan demand prediction model.

[0142] In some embodiments of the present application, the dual-tower multi-objective model includes a static tower network and a time-series tower network, the first multi-source feature data includes static feature data and time-series feature data, and the inputting the training sample data of the loan demand prediction model into the loan demand prediction model to obtain the loan demand prediction result includes: inputting the static feature data into the static tower network to obtain the loan demand prediction result of the static tower network; inputting the time-series feature data into the time-series tower network to obtain the loan demand prediction result of the time-series tower network; and merging the loan demand prediction result of the static tower network and the loan demand prediction result of the time-series tower network as the loan demand prediction result of the loan demand prediction model.

[0143] In some embodiments of the present application, the preset product recommendation model is trained in the following manner: constructing training sample data of the product recommendation model and corresponding product recommendation true value data, the training sample data of the product recommendation model including second multi-source feature data and loan demand feature data; inputting the training sample data of the product recommendation model into the product recommendation model to obtain a product recommendation result; calculating the recommendation loss of the product recommendation model based on the product recommendation result and the corresponding product recommendation true value data, and using the recommendation loss of the product recommendation model to iteratively update the parameters of the product recommendation model to obtain a trained product recommendation model.

[0144] In some embodiments of the present application, the dual-tower single-objective model includes a user tower network and a product tower network, the second multi-source feature data includes customer key feature data and product feature data, and the inputting the training sample data of the product recommendation model into the product recommendation model to obtain the product recommendation result includes: inputting the customer key feature data and the loan demand feature data into the user tower network to obtain the feature processing result of the user tower network; inputting the product feature data into the product tower network to obtain the feature processing result of the product tower network; and determining the product recommendation result based on the feature processing result of the user tower network and the feature processing result of the product tower network.

[0145] In some embodiments of the present application, determining the product recommendation result based on the feature processing results of the user tower network and the feature processing results of the product tower network includes: performing correlation calculation on the feature processing results of the user tower network and the feature processing results of the product tower network to obtain a correlation calculation result; and determining the product recommendation result based on the correlation calculation result.

[0146] It can be understood that the above-mentioned loan demand prediction and product recommendation device can implement the various steps of the loan demand prediction and product recommendation method provided in the aforementioned embodiment. The relevant explanations about the loan demand prediction and product recommendation method are applicable to the loan demand prediction and product recommendation device and will not be repeated here.

[0147] Figure 5 This is a schematic diagram of the structure of a device in the embodiment of the present application. Figure 5 As shown, the device includes one or more processors (or processing units), may further include one or more memories coupled to the processors, and may further include a communication module coupled to the processors.

[0148] The communication module can be used to communicate with other devices or apparatuses, such as sending or receiving data and / or signals. The communication module can include at least one communication module for communication. The communication module can include any interface necessary for communicating with other devices. Exemplarily, the communication module can be a transceiver, circuit, bus, module, or other type of communication module.

[0149] The processor may include, but is not limited to, at least one of the following: a general-purpose computer, a special-purpose computer, a microcontroller, a digital signal processor (DSP), or one or more of a controller-based multi-core controller architecture. A device may have multiple processors, such as application-specific integrated circuit chips, which are time-slave to a clock synchronized with a main processor.

[0150] The memory may include one or more non-volatile memories and one or more volatile memories. Examples of non-volatile memories include, but are not limited to, at least one of the following: read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, hard disk, compact disc (CD), digital video disc (DVD), or other magnetic storage and / or optical storage. Examples of volatile memories include, but are not limited to, at least one of the following: random access memory (RAM), or other volatile memories that do not persist during a power outage.

[0151] A computer program includes computer-executable instructions that are executed by an associated processor. The program may be stored in ROM. The processor may perform any suitable actions and processes by loading the program into RAM.

[0152] The possible implementation of the present application can be realized by means of a program, so that the communication device can perform any process discussed in the above embodiments. The possible implementation of the present application can also be realized by hardware or by a combination of software and hardware.

[0153] In some embodiments, the program may be tangibly contained in a computer-readable storage medium that may be included in the device (such as in a memory) or other storage device accessible by the device. The program may be loaded from the computer-readable storage medium into RAM for execution. The computer-readable storage medium may include any type of tangible non-volatile memory, such as ROM, EPROM, flash memory, hard disk, CD, DVD, etc.

[0154] The present application also provides a computer-readable storage medium having computer instructions or program codes stored thereon, which, when executed by a processor, causes the processor to perform the methods and functions described in any of the above embodiments. A computer-readable medium may be any tangible medium containing or storing a program for or related to an instruction execution system, apparatus, or device. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. Computer-readable media may include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any suitable combination thereof. The computer-readable storage medium may be any available medium that a computer can access, or a data storage device such as a server or data center that includes one or more available media integrated therein. More detailed examples of computer-readable storage media include electrical connections with one or more wires, magnetic media (e.g., magnetic disks, floppy disks, hard disks, tapes, magnetic storage devices), optical media (e.g., optical storage devices, DVDs), semiconductor media (e.g., solid-state drives), random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), or any suitable combination thereof.

[0155] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The embodiments of the present application also provide at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes one or more computer-executable instructions, such as instructions included in a program module, which are executed in a device on a real or virtual processor of the target to perform the processes, methods and functions involved in any of the above embodiments. When the computer program instructions are loaded and executed on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) method.

[0156] The present application also provides a computer program product, including a computer program or instructions, which, when run on a computer, causes the computer to perform the processes, methods, and functions in the above-described embodiments. Typically, a program module includes routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of the program modules can be combined or divided between program modules as needed. The machine executable instructions for the program modules can be executed in local or distributed devices. In distributed devices, the program modules can be located in local and remote storage media.

[0157] In general, various embodiments of the present application can be implemented in hardware or dedicated circuits, software, logic, or any combination thereof. Some aspects can be implemented in hardware, while other aspects can be implemented in firmware or software, which can be executed by a controller, microprocessor, or other computing device. Although various aspects of the embodiments of the present disclosure are shown and described as block diagrams, flow charts, or using some other graphical representation, it should be understood that the blocks, devices, systems, techniques, or methods described herein can be implemented as, by way of non-limiting example, hardware, software, firmware, dedicated circuits or logic, general-purpose hardware or a controller or other computing device, or some combination thereof.

[0158] It should be noted that although the embodiments of the present application are described above in conjunction with the accompanying drawings, the above embodiments are not independent of each other, and they can also be combined to obtain other embodiments. The methods, situations, categories, and divisions of the embodiments in the embodiments of the present application are only for the convenience of description and should not constitute special limitations. The features of the various methods, categories, situations, and embodiments can be combined with each other when they are logical. The various embodiments of the present application can be combined arbitrarily to achieve different technical effects. The embodiments of the present application no longer list various combinations.

[0159] In addition, although the operations of the method of the present disclosure are described in a particular order in the accompanying drawings, this does not require or imply that these operations must be performed in this particular order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flowchart can change the order of execution. Additionally or alternatively, certain steps can be omitted, multiple steps can be combined into one step, and / or one step can be decomposed into multiple steps. It should also be noted that the features and functions of two or more devices according to the present disclosure can be embodied in one device. Conversely, the features and functions of a device described above can be further divided into being embodied by multiple devices.

[0160] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0161] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A loan demand prediction and product recommendation method, characterized in that: The loan demand prediction and product recommendation method includes: Acquire multi-source feature data of a customer, wherein the multi-source feature data includes first multi-source feature data and second multi-source feature data; Predicting the customer's loan demand using a preset loan demand prediction model based on the first multi-source feature data to obtain a loan demand prediction result for the customer, the loan demand prediction result including loan demand prediction results for the customer at different time periods; Based on the second multi-source feature data and the loan demand prediction result of the customer, a product recommendation is performed using a preset product recommendation model to obtain a product recommendation result for the customer.

2. The loan demand forecasting and product recommendation method according to claim 1, characterized in that: The preset loan demand prediction model is a network structure of a dual-tower multi-objective model, and the dual-tower multi-objective model is used to predict customers' loan demand and loan amount in different periods. The preset product recommendation model is a network structure of a dual-tower single-objective model, and the dual-tower single-objective model is used to recommend financial products corresponding to customers.

3. The loan demand forecasting and product recommendation method according to claim 2, characterized in that: The preset loan demand prediction model is trained in the following way: Constructing training sample data of a loan demand prediction model and corresponding true value data of loan demand, wherein the training sample data of the loan demand prediction model includes first multi-source feature data; Inputting the training sample data of the loan demand prediction model into the loan demand prediction model to obtain a loan demand prediction result, wherein the loan demand prediction result includes loan demand prediction results of customers in different periods; The prediction loss of the loan demand prediction model is calculated according to the loan demand prediction result and the corresponding true value data of the loan demand, and the parameters of the loan demand prediction model are iteratively updated using the prediction loss of the loan demand prediction model to obtain a trained loan demand prediction model.

4. The loan demand forecasting and product recommendation method according to claim 3, characterized in that: The dual-tower multi-objective model includes a static tower network and a time-series tower network, the first multi-source feature data includes static feature data and time-series feature data, and the inputting of the training sample data of the loan demand prediction model into the loan demand prediction model to obtain a loan demand prediction result includes: Inputting the static characteristic data into the static tower network to obtain a loan demand prediction result of the static tower network; Inputting the time series characteristic data into the time series tower network to obtain the loan demand prediction result of the time series tower network; The loan demand prediction result of the static tower network and the loan demand prediction result of the sequential tower network are combined as the loan demand prediction result of the loan demand prediction model.

5. The loan demand forecasting and product recommendation method according to claim 2, characterized in that: The preset product recommendation model is trained in the following way: Constructing training sample data for a product recommendation model and corresponding product recommendation true value data, wherein the training sample data for the product recommendation model includes the second multi-source feature data and the loan demand feature data; Inputting the training sample data of the product recommendation model into the product recommendation model to obtain a product recommendation result; The recommendation loss of the product recommendation model is calculated according to the product recommendation results and the corresponding product recommendation true value data, and the recommendation loss of the product recommendation model is used to iteratively update the parameters of the product recommendation model to obtain a trained product recommendation model.

6. The loan demand forecasting and product recommendation method according to claim 5, characterized in that: The dual-tower single-objective model includes a user tower network and a product tower network, the second multi-source feature data includes customer key feature data and product feature data, and the inputting of the training sample data of the product recommendation model into the product recommendation model to obtain a product recommendation result includes: Inputting the customer key feature data and the loan demand feature data into the user tower network to obtain feature processing results of the user tower network; Inputting the product characteristic data into the product tower network to obtain characteristic processing results of the product tower network; The product recommendation result is determined according to the feature processing result of the user tower network and the feature processing result of the product tower network.

7. The loan demand forecasting and product recommendation method according to claim 6, characterized in that: Determining the product recommendation result according to the feature processing result of the user tower network and the feature processing result of the product tower network includes: performing a correlation calculation on the feature processing result of the user tower network and the feature processing result of the product tower network to obtain a correlation calculation result; The product recommendation result is determined according to the correlation calculation result.

8. A loan demand prediction and product recommendation device, characterized in that: The loan demand prediction and product recommendation device includes: an acquiring unit, configured to acquire multi-source feature data of a customer, wherein the multi-source feature data includes first multi-source feature data and second multi-source feature data; a loan demand prediction unit, configured to predict the customer's loan demand using a preset loan demand prediction model based on the first multi-source feature data, and obtain a loan demand prediction result for the customer, wherein the loan demand prediction result includes loan demand prediction results for the customer at different time periods; A product recommendation unit is configured to perform product recommendations based on the second multi-source feature data and the loan demand prediction result of the customer using a preset product recommendation model to obtain a product recommendation result for the customer.

9. A device comprising: processor; and a memory arranged to store computer-executable instructions, which, when executed, cause the processor to execute the loan demand prediction and product recommendation method according to any one of claims 1 to 7.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the loan demand prediction and product recommendation method according to any one of claims 1 to 7 is implemented.

Images