Product recommendation method, device, equipment, medium and program product

By integrating dynamic and static features of customers and products and using a matching probability model for automated recommendations, the problem of incomplete customer profiles and human judgment bias in product recommendations for financial institutions has been solved, achieving efficient and accurate product recommendations.

CN122288829APending Publication Date: 2026-06-26INDUSTRIAL AND COMMERCIAL BANK OF CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610409567.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-03-31
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Financial institutions rely on customer information recorded at branches for product recommendations, resulting in incomplete customer profiles and difficulty in ensuring the accuracy of recommendation results. Furthermore, human judgment is subject to subjective bias, making it difficult to accurately match customer needs.

Method used

By integrating customers' historical data and attribute information on application software, customer features are extracted from both dynamic and static dimensions. Combined with product attributes and click purchase records, a pre-trained matching probability determination model is used to quantify the matching probability between customers and products, thereby achieving automated recommendations.

Benefits of technology

This improved the accuracy and efficiency of recommendation results, reduced subjective bias in human judgment, and achieved high-quality matching with customer needs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122288829A_ABST
    Figure CN122288829A_ABST
Patent Text Reader

Abstract

This application provides a product recommendation method, apparatus, device, medium, and program product, relating to the field of artificial intelligence. The method includes: determining the dynamic characteristics of a customer based on historical data of the customer's use of application software; determining the static characteristics of a customer based on the customer's attribute information; determining the static characteristics of a product based on the product's attribute information; determining the dynamic characteristics of a product based on the product's click and purchase records; pairing the customer and product based on their dynamic and static characteristics to obtain customer-product pairs; inputting the dynamic and static characteristics of the customer-product pairs into a pre-trained matching probability determination model to obtain the matching probability of the customer-product pairs; and recommending products to the customer based on the matching probability of each customer-product pair. This application accurately characterizes the features of customers and products through dynamic and static dual-dimensional information. Based on this, the matching results of customers and products are input into the model, resulting in more accurate and reliable matching probabilities.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and more particularly to a product recommendation method, apparatus, device, medium, and program product. Background Technology

[0002] With the diversification of financial products and the personalization of customer needs, banks and other financial institutions offer targeted product recommendation services, which involve recommending fund investment products that match the specific needs of each customer. This service achieves a win-win situation for both financial institutions and customers: for financial institutions, it enables efficient marketing; for customers, it largely satisfies their needs.

[0003] Currently, financial institutions' precise product recommendation services largely rely on customer information recorded at branches, with products matched to customers based on manual judgment. However, this approach has the following problems: First, the customer information recorded at branches is relatively incomplete, resulting in incomplete customer profiles and directly affecting the accuracy of matching recommended products with customer needs; second, the recommendation process depends on human experience, such as business personnel selecting target customers for products based on subjective judgment, making it difficult to guarantee the accuracy of the recommendation results. Summary of the Invention

[0004] This application provides a product recommendation method, apparatus, device, medium, and program product. By obtaining the static and dynamic characteristics of users and products, it comprehensively improves the information profile of customers and products from both static and dynamic dimensions, providing an effective data foundation for subsequent product recommendations. At the same time, it utilizes the powerful learning and reasoning capabilities of the model to obtain the matching probability between customers and products, overcoming the limitations of manual judgment and making the recommendation results more objective and reliable.

[0005] Firstly, this application provides a product recommendation method, including:

[0006] Based on customers' historical data of using application software, determine customers' dynamic characteristics; based on customers' attribute information, determine customers' static characteristics.

[0007] Based on the product's attribute information, determine the product's static characteristics; based on the product's click and purchase records, determine the product's dynamic characteristics.

[0008] Based on the dynamic and static characteristics of customers and products, customers and products are matched to obtain customer-product pairs; dynamic and static characteristics include dynamic features and static features.

[0009] For each customer product pair, the dynamic and static features of the customer product pair are input into a pre-trained matching probability determination model to obtain the matching probability of the customer product pair; the dynamic and static features of the customer product pair include the dynamic and static features of the customer and the dynamic and static features of the product.

[0010] Based on the matching probability of each customer's product pair, products are recommended to the customer.

[0011] Secondly, this application provides a product recommendation device, comprising:

[0012] The customer characteristic determination module is used to determine the dynamic characteristics of customers based on their historical data of using application software, and to determine the static characteristics of customers based on their attribute information.

[0013] The product feature determination module is used to determine the static features of a product based on its attribute information, and to determine the dynamic features of a product based on its click and purchase records.

[0014] The customer-product pairing module is used to pair customers and products based on the dynamic and static characteristics of the customers and the products to obtain customer-product pairs; dynamic and static characteristics include dynamic features and static features;

[0015] The matching probability determination module is used to input the dynamic and static features of each customer product pair into a pre-trained matching probability determination model to obtain the matching probability of the customer product pair. The dynamic and static features of the customer product pair include the dynamic and static features of the customer and the dynamic and static features of the product.

[0016] The recommendation module is used to recommend products to customers based on the matching probability of each customer's product pair.

[0017] Thirdly, this application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0018] The memory stores the instructions that the computer executes;

[0019] The processor executes computer-executable instructions stored in memory to implement the method provided in the first aspect above.

[0020] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method provided in the first aspect above.

[0021] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the method provided in the first aspect above.

[0022] The product recommendation method, apparatus, equipment, medium, and program products provided in this application integrate historical data of customer application software usage and customer attribute information. They extract customer features from both dynamic and static dimensions, resulting in more effective and accurate customer characterization. Simultaneously, they integrate product attribute information and click and purchase records, extracting product features from both dynamic and static dimensions. These dynamic and static features further enhance product characterization, providing a high-quality and more complete data foundation for subsequent product recommendations. Then, customers and products are paired to obtain customer-product pairs. The dynamic and static features of these pairs are input into a model. Leveraging the model's powerful probabilistic modeling and confidence estimation capabilities, the match between customers and products in a customer-product pair is quantified as a matching probability. This matching probability effectively characterizes whether a product matches a customer's needs, transforming subjective experience into quantifiable predictive logic. Finally, based on the matching probability, matching products are recommended to customers. This automated process replaces manual judgment, increasing efficiency. Furthermore, this method significantly improves the matching degree between recommendation results and customer needs, reduces subjective bias from manual judgment, and makes recommendation results more accurate. Attached Figure Description

[0023] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0024] Figure 1 A flowchart illustrating a product recommendation method provided in an embodiment of this application;

[0025] Figure 2 A schematic diagram of the hardware architecture for implementing the product recommendation method provided in an embodiment of this application;

[0026] Figure 3 A flowchart illustrating another product recommendation method provided in this application embodiment;

[0027] Figure 4 A schematic diagram illustrating the temporal relationship between training samples and their labels provided in the embodiments of this application;

[0028] Figure 5 This application provides a schematic diagram of the architecture of a product recommendation system.

[0029] Figure 6 This is a schematic diagram of the internal flow of the model unit provided in the embodiments of this application;

[0030] Figure 7 This is a schematic diagram of the internal flow of the recommendation result display unit provided in the embodiments of this application;

[0031] Figure 8This is a schematic diagram of the structure of a product recommendation device provided in an embodiment of this application;

[0032] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0033] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0034] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0035] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with the relevant laws, regulations, and standards of the relevant countries and regions, have taken necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation access points for users to choose to authorize or refuse.

[0036] Furthermore, the technical solution involved in this application, which involves big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.) and the use of artificial intelligence technology for automated decision-making, and makes decisions that have a significant impact on personal rights based on the results of automated decision-making, provides users with corresponding operation entry points for users to choose to agree to or reject the results of automated decision-making; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0037] It should be noted that the product recommendation methods, apparatus, equipment, media and program products provided in this application can be used in the field of artificial intelligence, or in any field other than artificial intelligence. The application fields of the product recommendation methods, apparatus, equipment, media and program products in this application are not limited.

[0038] This application applies to scenarios where banks and other financial institutions perform product maintenance or customer maintenance. In the product maintenance scenario, banks and other financial institutions often need to promote their products to customers, i.e., find customers for their products. In the customer maintenance scenario, banks and other financial institutions often need to promote products to their customers, i.e., recommend products that meet their needs.

[0039] Currently, banks and other financial institutions maintain their products and customers by acquiring limited information about customers and products, manually assessing customer needs, matching those needs with the customers, and then recommending products to customers based on the matching results. For example, each bank branch maintains its limited customer information and relies on the subjective judgment of its staff to recommend products.

[0040] However, the above methods have significant limitations. On the one hand, the business personnel's understanding of customer needs is limited to the limited information available at that particular branch, resulting in an incomplete understanding of customer needs and a discrepancy between the recommended results and actual customer requirements. On the other hand, subjective judgment by business personnel can also lead to discrepancies between the recommended results and actual customer needs. Furthermore, subjective judgment often focuses on highly active customers, easily overlooking inactive customers—those who have not made transactions for a long time but still have transaction needs. In addition, manual judgment has a long turnaround time, potentially missing the optimal time for maintenance and repair.

[0041] The product recommendation method provided in this application aims to solve the aforementioned technical problems. Specifically, it first integrates static customer attribute information and dynamic application software usage data to generate dynamic and static customer features; and secondly, it integrates static product attribute information and dynamic click and purchase records to generate dynamic and static product features. This dual-dimensional approach refines the characteristics of both customers and products, making them more comprehensive and thus improving the accuracy of subsequent recommendations. Next, it constructs customer-product pairs, inputting the dynamic and static features of these pairs into the model. Leveraging the model's powerful capabilities, it quantifies the matching probability between the corresponding customer and product. Finally, based on the matching probability, it recommends products that match the customer's needs, achieving automated product recommendation with higher efficiency. Furthermore, the matching probability determined by the model's objective logic is more objective and effective, further enhancing the effectiveness of the recommendation results.

[0042] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0043] Figure 1 This is a flowchart illustrating a product recommendation method provided in an embodiment of this application. Figure 1 As shown, the method provided in this embodiment includes the following steps:

[0044] Step S101: Based on the customer's historical data on application software usage, determine the customer's dynamic characteristics; based on the customer's attribute information, determine the customer's static characteristics.

[0045] The product recommendation method provided in this application embodiment can be executed by an electronic device with corresponding processing capabilities, such as a processing server in a banking or other financial institution system.

[0046] The customers refer to all customers maintained by the bank or other financial institutions. This approach avoids ignoring dormant customers, uncovers potential needs, and reduces the risk of customer churn.

[0047] In some embodiments, banks or other financial institutions often include a head office and multiple branch offices, each maintaining its own dedicated customer base. In this embodiment, customers include all customers maintained by the head office and branch offices.

[0048] The application software is software launched by the bank or other financial institution for conducting financial transactions. Customers conduct transactions through this application software.

[0049] Customer usage history data refers to the records of a customer's past actions while using the application. For example, customer A clicked on and browsed 5 products within 3 days. Customer usage history data is unstructured data; unstructured data is data without fixed rules or formats.

[0050] Historical data can be records of customer actions when using the application software within a preset time period. For example, historical data can be records of actions taken within a period of one or two months prior to the current moment.

[0051] Customer attribute information is structured data recorded within the bank or other financial institution and remains unchanged over a relatively long period. Structured data is data stored according to fixed rules and formats. Attribute information includes, but is not limited to, basic identity attributes, financial credit attributes, financial behavior attributes, and risk preference attributes.

[0052] In this step, historical data on customer usage of the application software is read from the application software's operation logs. Customer attribute information is also retrieved from the bank's or other financial institution's internal database. Dynamic customer characteristics are extracted from the historical application software usage data, and static customer characteristics are extracted from the customer attribute information.

[0053] In some embodiments, historical data on customer usage of application software and customer attribute information can be stored in a data lake. The data lake uniformly maintains customer information.

[0054] In this embodiment, historical data on customer usage of the application software can be read from the application software's operation logs in a first cycle and stored in the data lake. Customer attribute information can be read in a second cycle and stored in the data lake. The first and second cycles are configurable parameters.

[0055] In one example, since the attribute information remains unchanged over a long period, the second period can be longer than the first period, for example, the second period is 3 months and the first period is 1 day.

[0056] In other embodiments, two interfaces may be provided. One interface connects the application software's server and the data lake, and is used to update the historical data of customer usage of the application software in the application software's operation logs to the data lake. The other interface connects the bank's or other financial institution's internal database (including the database of the head office and the proprietary databases of each branch) and the data lake, and is used to update the customer attribute information in the internal database to the data lake.

[0057] In some embodiments, historical data on customer usage of application software and customer attribute information can be obtained from the data lake at a preset period, such as 1 day, and subsequent steps can be performed.

[0058] In some embodiments, historical data and attribute information can be preprocessed, such as data cleaning and data normalization, before being stored in the data lake, to ensure the quality of the information when used later.

[0059] Since the customer's attribute information is structured data, we can extract the items corresponding to the customer's static features from the customer's attribute information, and then concatenate the extracted items to obtain the customer's static features.

[0060] A customer's static characteristics include basic characteristics (such as gender and age), assets, asset distribution (such as funds, insurance, wealth management products, and precious metals), and acceptable risk level. Information corresponding to each item in the static characteristics is extracted from the customer's attribute information and then concatenated to obtain the customer's static characteristics.

[0061] Since the historical data of customer usage of the application software is unstructured data, data related to the customer's dynamic characteristics can be retrieved from the historical data. This data can then be processed to obtain the various components of the dynamic characteristics. These components are then concatenated to obtain the customer's dynamic characteristics.

[0062] Customer dynamic characteristics include transaction frequency, browsing frequency, held products, and browsed products within the first time period. The first time period includes at least one time segment.

[0063] For example, a customer's dynamic characteristics include transaction frequency, browsing frequency, held products, and browsed products in the 3 days, 7 days, and 1 month prior to the current moment.

[0064] Based on this, historical transaction data and browsing data of customers are retrieved from historical data. Based on the historical transaction data, the transaction frequency and held products in the first time period are determined; based on the historical browsing data, the browsing frequency and browsed products in the first time period are determined. These data are then combined to obtain the customer's dynamic characteristics.

[0065] Step S102: Determine the static characteristics of the product based on its attribute information; determine the dynamic characteristics of the product based on its click and purchase records.

[0066] The product is the full range of products owned by banks and other financial institutions.

[0067] Product attribute information refers to structured data about a product that is recorded within the bank or other financial institution and remains unchanged over a relatively long period. This includes product name, product type, risk attributes, transaction rule attributes, and size attributes.

[0068] Product click and purchase records are the historical data on product clicks and purchases within the application software.

[0069] In this step, click and purchase records for the product are read from the application software's historical operation logs. Product attribute information is also retrieved from the bank's or other financial institution's internal database. Static features of the product are extracted from the attribute information, while dynamic features are extracted from the click and purchase records.

[0070] The static characteristics of a product include product type, risk level, and scale. By extracting the corresponding information from each item in the product's attribute information and concatenating this information, the product's static characteristics are obtained.

[0071] The product's dynamic characteristics include the average price, click frequency, purchase frequency, and performance (such as 5-day annualized rate of return and Sharpe ratio) within the first time period. The first time period includes at least one time segment. Based on the product's purchase records, the average price, purchase frequency, and performance within the first time period are determined. Based on the product's click records, the click frequency within the first time period is determined. These factors are then combined to obtain the product's dynamic characteristics.

[0072] In some embodiments, the read product attribute information, click records, and purchase records can be stored in a data lake. The data lake uniformly maintains product information.

[0073] In this embodiment, click and purchase records for the product can be read from the application software's historical operation logs in a third cycle and stored in the data lake. Product attribute information can be read from the bank's or other financial institution's internal database in a fourth cycle and stored in the data lake. The third and fourth cycles are configurable parameters.

[0074] In one example, since the attribute information remains unchanged over a long period, the fourth period can be longer than the third period, for example, the fourth period is 3 months and the third period is 1 day.

[0075] In other embodiments, two interfaces may be provided. One interface connects the application software's server and the data lake, and is used to update product click records and purchase records from the application software's operation logs to the data lake. The other interface connects the bank's or other financial institution's internal database and the data lake, and is used to update product attribute information from the internal database to the data lake.

[0076] In some embodiments, before storing click records, purchase records, and attribute information into the data lake, the click records, purchase records, and attribute information can be preprocessed, such as through data cleaning and data normalization, to ensure the quality of the information when used subsequently.

[0077] In some embodiments, product attribute information, product click records, and purchase records can be retrieved from the data lake at a preset period, such as 1 day, and subsequent steps can be performed.

[0078] Step S103: Based on the dynamic and static characteristics of the customer and the product, the customer and product are matched to obtain customer-product pairs.

[0079] Dynamic and static features include dynamic features and static features.

[0080] Specifically, the dynamic and static characteristics of a customer are combined to form the customer's dynamic and static characteristics; similarly, the dynamic and static characteristics of a product are combined to form the product's dynamic and static characteristics.

[0081] In this step, customers and products whose dynamic and static features satisfy a pairing relationship are paired to obtain customer-product pairs. The dynamic and static features of the customers in the customer-product pairs are then concatenated with the dynamic and static features of the products in the customer-product pairs to obtain the dynamic and static features of the customer-product pairs.

[0082] A pairing relationship is used to indicate the likelihood that a customer will purchase the product.

[0083] For example, if a product is not among the products held by a customer in the dynamic characteristics, then it is determined that the dynamic and static characteristics of the customer and the product satisfy a pairing relationship, and the customer and the product are paired.

[0084] Step S104: For each customer product pair, input the dynamic and static features of the customer product pair into the pre-trained matching probability determination model to obtain the matching probability of the customer product pair.

[0085] The dynamic and static characteristics of a customer's product pair include the dynamic and static characteristics of the customer and the dynamic and static characteristics of the product.

[0086] The probabilistic matching determination model is a type of machine learning model. By learning from training data, probabilistic modeling, and mathematical optimization, the probabilistic matching determination model can map input data to a probability distribution.

[0087] Specifically, dynamic and static data of customer samples and product samples are acquired, and positive and negative customer-product pair samples are generated. The dynamic and static data of these samples are used as training samples to train the matching probability determination model. The trained matching probability determination model can then map the input dynamic and static data of customer-product pairs to matching probabilities.

[0088] Specifically, the dynamic and static features of the customer-product pair are input into a pre-trained matching probability determination model. The matching probability determination model obtains the matching probability of the customer-product pair based on the mapping relationship between the dynamic and static features of the customer and the dynamic and static features of the product in the customer-product pair.

[0089] For example, the matching probability determination module can be a deep learning model. The dynamic and static features of the customer product pair are input into the pre-trained matching probability determination module. The matching probability determination model maps the dynamic and static features of the customer product pair to a high-dimensional feature space. Through contrastive learning or cross-entropy encoding, the mapping relationship between the dynamic and static features of the customer and the dynamic and static features of the product in the customer product pair is quantified to obtain the matching probability of the customer product pair.

[0090] For example, the matching probability determination model can be a probabilistic model, such as a Bayesian model. The dynamic and static features of the customer product pair are input into the pre-trained matching probability determination model, which calculates the matching probability based on the pre-learned prior distribution using the Bayesian formula.

[0091] Optionally, a matching probability determination model can be constructed by integrating multiple learning frameworks. Matching probabilities are predicted based on multiple machine learning models (such as random forests, gradient boosting trees, and neural networks). The final matching probability of the customer-product pair is obtained based on the outputs of these multiple models.

[0092] In one example, the matching probabilities output by multiple models can be weighted and fused to obtain the final matching probability of the customer product pair.

[0093] For example, a random forest is used to predict the likelihood of purchase, a neural network is used to predict how well a product matches the customer's preferences, and a dynamic weighting strategy (such as adjusting model weights based on historical prediction errors) is used to generate the final customer-product pair matching probability.

[0094] Ensemble learning algorithms (such as Stacking and Bagging) and dynamic weight allocation mechanisms are employed to enhance prediction robustness through model diversity. Specific implementations include: dividing the training dataset, training multiple models separately, and designing weight calculation rules to dynamically adjust the weights of each model based on their accuracy on the validation set.

[0095] The multi-model ensemble learning framework addresses the overfitting or underfitting issues of single models in complex scenarios. For example, when a customer group contains a mix of high-risk and low-risk preferences, a single model may struggle to capture the characteristics of different groups. Multi-model ensemble, however, can balance prediction bias through weighted fusion, improving the breadth and adaptability of recommendation results. Furthermore, the dynamic weight allocation mechanism ensures high predictive stability across various market environments.

[0096] Step S105: Recommend products to customers based on the matching probability of each customer's product pair.

[0097] In this step, customer-product pairs with a matching probability greater than or equal to a preset threshold are identified, and products from that customer-product pair are recommended to customers within those pairs. The preset threshold is a pre-set configurable parameter.

[0098] In this step, the matching probability of each customer's product pairs can also be sorted. Based on the sorting results, a preset number of customer product pairs are determined, and products from those customer product pairs are recommended to the customer.

[0099] In this step, the matching probability of customer product pairs for each product can also be sorted. Based on the sorting results, a preset number of customer product pairs are determined, and the product is recommended to each customer in each customer product pair.

[0100] In this embodiment, the method of recommending products to customers can be to send the recommended product information to the relevant sales personnel for each customer, so that the relevant sales personnel can recommend the product to that customer.

[0101] Another way to recommend products to customers is to broadcast the product information in the corresponding customer's application software.

[0102] Another way to recommend products to customers is to announce each product in the corresponding customer's application software, or to send information about a portion of customers to relevant sales personnel so that they can recommend products to customers offline.

[0103] By combining online and offline recommendations, the cost and benefits of recommendations are balanced, making this product recommendation method more effective.

[0104] In some embodiments, products can be recommended to customers through a multi-channel collaborative recommendation mechanism.

[0105] Specifically, based on customers' dynamic characteristics, such as their transaction frequency and browsing frequency over the past 30 days, the recommendation method (such as SMS, app announcements, or offline recommendations) is dynamically selected. For example, for highly active customers (those with high transaction or browsing frequency), products are recommended to them through app announcements and offline recommendations; for inactive customers, products are recommended to them through app announcements.

[0106] By employing a multi-channel collaborative recommendation mechanism, personalized recommendations are made to each customer, resolving the issue of varying customer acceptance rates caused by relying on a single recommendation channel and improving the reach of recommended content. Furthermore, dynamically adjusting recommendation methods can further enhance customer satisfaction and purchase conversion rates.

[0107] Figure 2 This is a schematic diagram of a hardware architecture for implementing a product recommendation method, provided as an embodiment of this application. Figure 2 As shown, the product recommendation method can be executed on the backend server at a set period, such as one day. The products recommended to customers are stored in the database. Specifically, the database stores the recommended products to customers in the form of tables. The database grants permissions to the frontend server, which directly reads the information from the tables, processes it, and displays it on the frontend interface.

[0108] The product recommendation method provided in this application integrates historical data of customer application usage and customer attribute information, extracting customer features from both dynamic and static dimensions. The resulting dynamic and static features more effectively and accurately represent customers. Simultaneously, it integrates product attribute information and click and purchase records, extracting product features from both dynamic and static dimensions. The resulting dynamic and static features more effectively and accurately represent products, providing a high-quality and more complete data foundation for subsequent product recommendations. Then, customers and products are paired to obtain customer-product pairs. The dynamic and static features of these customer-product pairs are input into the model. Leveraging the model's powerful probabilistic modeling and confidence estimation capabilities, the match between customers and products in a customer-product pair is quantified as a matching probability. This matching probability effectively represents whether a product matches a customer's needs, transforming subjective experience into quantitative predictive logic. Finally, based on the matching probability, matching products are recommended to customers, automating the process to replace manual judgment, thus increasing efficiency. Furthermore, this method significantly improves the matching degree between recommendation results and customer needs, reduces subjective bias from manual judgment, and makes recommendation results more accurate.

[0109] Figure 3 This is a flowchart illustrating another product recommendation method provided in an embodiment of this application. The method provided in this embodiment is... Figure 1 Based on the illustrated embodiment, a more detailed explanation of steps S103 and S105 is provided. For example... Figure 3 As shown, the method provided in this embodiment includes:

[0110] Step S301: Based on the customer's historical data on application software usage, determine the customer's dynamic characteristics; based on the customer's attribute information, determine the customer's static characteristics.

[0111] Step S302: Determine the static characteristics of the product based on its attribute information; determine the dynamic characteristics of the product based on its click and purchase records.

[0112] Step S303: Based on the dynamic and static characteristics of the customer and the dynamic and static characteristics of the product, determine the customers and products that have a pairing relationship.

[0113] In actual referral marketing, several static rules exist. These static rules explicitly limit the range of products recommended to specific customers, or define the customer group that can be matched with specific products. Customers and products that meet these static rules are considered to have a pairing relationship.

[0114] Specifically, customers and products whose dynamic and static characteristics satisfy the static rules are identified as having a pairing relationship.

[0115] Static rules are pre-set rules based on dynamic and static characteristics, including but not limited to product performance exceeding the initial performance threshold and customer assets exceeding the asset threshold.

[0116] This static rule avoids generating customer-product pairs between products with low returns and customers with no investment needs, reducing the number of customer-product pairs, thereby reducing the amount of data to be processed and improving the efficiency of subsequent processing.

[0117] For example, assume there are customer 1 and customer 2, customer 1 has assets of 0 and customer 2 has assets of 10,000; and there are products 1 to 3, product 1 has a Sharpe ratio of -1, product 2 has a Sharpe ratio of 1, and product 3 has a Sharpe ratio of 2. Static rules include a product performance (Shape ratio) greater than 0 and a customer's assets greater than 1,000. Therefore, customer 2 and product 2, and customer 2 and product 3 are identified as having a paired customer and product relationship, respectively.

[0118] Optionally, the static characteristics of the customer include the acceptable risk level; the static characteristics of the product include the risk level; based on the dynamic and static characteristics of the customer and the dynamic and static characteristics of the product, the customer and product with a matching relationship are determined, including: if the risk level of a product is less than or equal to the customer's acceptable risk level, then the customer and product are determined to be a customer and product with a matching relationship.

[0119] The customer's acceptable risk level is the maximum risk threshold they are willing to bear, set based on their own risk tolerance and investment goals. This acceptable risk level is entered by the customer in the application software.

[0120] The risk level of a product is a classification and labeling of its risk level. Product risk levels include low risk, low-to-medium risk, medium risk, medium-to-high risk, and high risk.

[0121] In this embodiment, if a product's risk level is less than or equal to the customer's acceptable risk level, it indicates that the product matches the customer's risk tolerance, and thus it is determined that the customer and the product have a pairing relationship.

[0122] By matching risk levels with acceptable risk levels, it is determined whether a customer and product are matched, thus avoiding the need to calculate the matching probability between customers with low risk tolerance and products with high risk levels, reducing computational load and hardware resource consumption.

[0123] Optionally, the static characteristics of a customer may also include the average holding period; the dynamic and static characteristics of a product may also include the investment period; determining the matching relationship between customers and products based on the dynamic and static characteristics of customers and products may further include: if the average holding period of a customer is greater than or equal to the investment period of a product, then the customer and product are determined to be a matching customer and product.

[0124] The average holding period is the average investment period of all products held by this customer. The investment period of a product is the time period from purchasing the product to planning to redeem or withdraw funds.

[0125] In this embodiment, if a customer's average holding period is greater than or equal to the product's investment period, it indicates that the customer's willingness to hold the product for a certain period matches the product's capital requirements, and thus it is determined that the customer and the product have a pairing relationship.

[0126] In this embodiment, if a customer's average holding period is greater than or equal to the product's investment period, or if the product's risk level is less than or equal to the customer's acceptable risk level, then it is determined that the customer and the product have a pairing relationship.

[0127] Optionally, based on the dynamic and static characteristics of customers and products, determine the customers and products with matching relationships, including:

[0128] If a customer's average holding period is greater than or equal to the product's investment period, and the product's risk level is less than or equal to the customer's acceptable risk level, then it is determined that the customer and the product have a matching relationship.

[0129] By matching the average holding period of customers with the investment period of products, it can determine whether a pairing relationship exists. This directly excludes products that exceed the customer's holding period intention, eliminating the need to calculate the matching probability of customers and products in this case, thus reducing the amount of computation and the hardware resources required.

[0130] Optionally, the dynamic characteristics of a customer include the products they hold and the products they click; the dynamic characteristics of a product include priority; based on the dynamic and static characteristics of the customer and the products, the system determines customers and products with matching relationships, including: for each customer, generating a first set of products other than the products they hold based on the products they hold; determining a second set of products based on the dynamic and static characteristics of the products in the first set and the dynamic and static characteristics of the customer; and determining a preset number of customers and products with matching relationships based on the priority of the products in the second set and whether the products are the customer's clicked products.

[0131] The products a customer holds are those they possess at the moment of the event. The products a customer clicks on are those they clicked on or viewed at the moment of the event.

[0132] The product priority is determined by the ranking rules of the bank or other financial institutions, which dictate the order of recommendations.

[0133] In this embodiment, for each customer, a first product set is generated based on the products the customer already owns, excluding those products. This first product set contains products the customer has not purchased. Then, a second product set is determined based on the dynamic and static characteristics of the products in the first product set and the dynamic and static characteristics of the customer. Finally, the products in the second product set are sorted according to their priority. Based on the sorting result and whether the products in the second product set are clicked products by the customer, a preset number of customer-product pairs with matching relationships are determined.

[0134] For example, based on the dynamic and static characteristics of the products in the first product set and the dynamic and static characteristics of the customer, products with a risk level less than or equal to the customer's acceptable risk level and an investment period less than the customer's average holding period can be identified as the second product set.

[0135] For example, based on whether the products in the ranking results and the second product set are the clicked products of the customer, the clicked products can be preferentially identified as customers and products with a pairing relationship. Then, the products at the top of the ranking results can be identified as customers and products with a pairing relationship.

[0136] By using a primary product set, the system avoids recommending products already held by customers, thus preventing ineffective recommendations and making them more effective. Furthermore, by prioritizing and identifying clicked products, it identifies products with existing customer pairings, prioritizing high-intent products for promotion. This ensures that the identified paired customers and products align with the bank's or other financial institution's objectives and customer needs, resulting in a higher degree of matching.

[0137] Step S304: Match customers with existing pairing relationships with products to obtain customer-product pairs.

[0138] Specifically, establish pairing relationships between existing customers and products, obtain customer-product pairs, and store these pairs in a mapping table.

[0139] After obtaining the customer product pair, the process also includes: concatenating the dynamic and static features of the customer and the dynamic and static features of the product in the customer product pair to obtain the dynamic and static features of the customer product pair.

[0140] Step S305: For each customer product pair, input the dynamic and static features of the customer product pair into the pre-trained matching probability determination model to obtain the matching probability of the customer product pair.

[0141] The dynamic and static characteristics of a customer's product pair include the dynamic and static characteristics of the customer and the dynamic and static characteristics of the product.

[0142] Step S306: For each customer, recommend matching products to customers whose matching probability is greater than or equal to the first threshold.

[0143] A customer-product pair consists of products that are paired with the customer. In other words, a customer and its paired product constitute a customer-product pair. For example, if customer 1 and product 1 constitute a customer-product pair, then product 1 is the paired product of customer 1.

[0144] The first threshold is a pre-set configurable parameter, such as 0.7, 0.8, etc.

[0145] In this step, at the customer level, for each customer's product pair, customer product pairs with a matching probability greater than or equal to a first threshold are identified, and the matched products in the customer product pair are recommended to that customer.

[0146] In this embodiment, only customers and products with existing pairing relationships are matched, achieving precise matching. This eliminates the need to establish a large number of customer-product pairs, reducing trial-and-error costs and improving the efficiency of subsequent processing. By using customers as the granularity and determining products to recommend to customers based on threshold judgment, personalized recommendations are achieved for each individual. At the same time, the threshold judgment method is easy to implement, simplifying the process of determining recommended products.

[0147] In one possible implementation, step S105, recommending products to customers based on the matching probability of each customer product pair, includes: for each product, identifying customers paired with products whose matching probability is greater than or equal to a second threshold as a high-probability recommended customer group; determining potential demand recommended customer groups based on the dynamic and static characteristics of customers paired with products whose matching probability is greater than or equal to a third threshold and less than the second threshold; and recommending products to customers based on the customer group to which the customer belongs.

[0148] In a customer-product pair, the customer is the paired customer of the product. In other words, a product and its paired customer constitute a customer-product pair. For example, if customer 1 and product 1 constitute a customer-product pair, then customer 1 is the paired customer of product 1.

[0149] In this embodiment, the corresponding customer is determined for each product at the product level.

[0150] Specifically, for each product's customer-product pair, customer-product pairs with a matching probability greater than or equal to the second threshold are identified, and the paired customers in these pairs are designated as high-probability recommended customer groups. Customer-product pairs with a matching probability greater than or equal to the third threshold and less than the second threshold are identified, and based on the dynamic and static characteristics of the paired customers in these pairs, potential demand recommended customer groups are determined.

[0151] For example, it can be determined whether a customer group represents a potential customer group based on whether the clicked products of the matched customers in the customer product pair with a matching probability greater than or equal to a third threshold and less than a second threshold contain that product.

[0152] Based on the customer's customer group, products are recommended to the customer. Specifically, if the customer belongs to the high-probability recommendation customer group, the product is recommended to the customer through the first recommendation channel; if the customer belongs to the potential demand recommendation customer group, the product is recommended to the customer through the second recommendation channel.

[0153] For example, the first recommendation channel could be broadcasting the product in the corresponding customer's application software, or recommending the product to the customer through relevant personnel. The second recommendation channel could be broadcasting the product in the corresponding customer's application software.

[0154] This method enables product recommendations to customers with low activity levels but potential needs, further addressing the issue of overlooking inactive customers and improving the comprehensiveness of the method's coverage.

[0155] In one possible implementation, in order to enable the matching probability determination model to work accurately and reliably, this application provides a training method for the matching probability determination model, so that the model can output accurate matching probabilities.

[0156] Specifically, multiple training samples and their labels are obtained; each training sample includes the dynamic and static features of the corresponding customer product pair, and the labels of the training samples are determined based on the customer's interaction with the recommended products after recommending the products to the customers in the corresponding customer product pair; the matching probability determination model is trained using the training samples and their labels.

[0157] In this embodiment, based on historical data of customer samples using the application software, the dynamic characteristics of the customer samples at the observation time point are determined; based on the attribute information of the customer samples, the static characteristics of the customer samples at the observation time point are determined; based on the attribute information of the product samples, the static characteristics of the product samples at the observation time point are determined; based on the click and purchase records of the product samples, the dynamic characteristics of the product samples at the observation time point are determined; based on the dynamic and static characteristics of the customer samples and the product samples, the customer samples and product samples are paired to obtain the customer-product pairs included in each training sample. The dynamic and static characteristics of the customer samples and the product samples are concatenated to obtain the dynamic and static characteristics of the customer-product pairs included in each training sample.

[0158] The labels of the training samples are determined based on the customer's interaction with the recommended products after recommending products to customers in the corresponding customer product pair within a preset time period (e.g., 10 days) following the observation time point.

[0159] For example, if a customer's interaction with a recommended product indicates that they purchase, favorite, or browse the product, the label for that training sample is 1. If a customer's interaction with a recommended product indicates that they do not purchase or browse the product, the label for that training sample is 0.

[0160] In some embodiments, if a customer's interaction with a recommended product instructs them to purchase or favorite the product, or instructs them to browse the product for a duration greater than or equal to a preset browsing time threshold, then the label of the training sample is determined to be 1.

[0161] Figure 4 This is a schematic diagram illustrating the temporal relationship between training samples and their labels provided in embodiments of this application. For example... Figure 4 As shown, the dynamic and static features of the customer product pairs included in the training samples are before the observation time point, while the labels of the training samples are after the observation time point.

[0162] Based on the labels, positive and negative samples are identified. The matching probability is then used to train the model.

[0163] After the matching probability determination model has been trained, its accuracy can be determined through inter-period validation and practical validation.

[0164] Specifically, cross-period validation involves acquiring validation samples at validation observation time points that differ from the observation time points of the training samples (e.g., after the observation time points of the training samples). The validation samples include the dynamic and static features of customer product pairs at the validation observation time points, as well as the labels of the validation samples determined based on the customer's interactive operations on the recommended products after recommending products to customers in the corresponding customer product pairs.

[0165] The validation samples are input into the matching probability determination model to obtain the output results. Based on the output results and labels, the precision and recall rates are calculated. The parameters of the matching probability determination model are then adjusted based on the precision and recall rates.

[0166] Practice has shown that by putting the matching probability determination model into use, and then updating the matching probability model based on the actual purchase situation of customers after a period of time, the model can be optimized.

[0167] By training the matching probability determination model, the model can output matching probabilities more accurately, thereby improving the accuracy of recommendations.

[0168] Figure 5 This is a schematic diagram of the architecture of a product recommendation system provided in an embodiment of this application. Figure 5 As shown, the product recommendation system includes a data acquisition unit, a feature processing unit, a model unit, a model result verification unit, a data analysis unit, and a recommendation result display unit.

[0169] The data acquisition unit is used to obtain historical data on customer usage of application software, customer attribute information, product attribute information, and product click and purchase records from the data lake.

[0170] The feature processing unit is used to determine the dynamic characteristics of customers based on their historical data of using application software; to determine the static characteristics of customers based on their attribute information; to determine the static characteristics of products based on their attribute information; and to determine the dynamic characteristics of products based on their click and purchase records.

[0171] The model unit is used to pair customers and products based on the dynamic and static features of customers and products to obtain customer-product pairs; dynamic and static features include dynamic features and static features; for each customer-product pair, the dynamic and static features of the customer-product pair are input into a pre-trained matching probability determination model to obtain the matching probability of the customer-product pair; the dynamic and static features of the customer-product pair include the dynamic and static features of the customer and the dynamic and static features of the product.

[0172] The recommendation results display unit is used to recommend products to customers based on the matching probability of each customer's product pair.

[0173] The model results validation unit is used to determine the accuracy of the model's output by verifying the matching probability. For example, the accuracy of the model's output can be determined through intertemporal validation and practical validation.

[0174] The data analysis unit is used to adjust the matching probability determination model after the model result verification unit determines that the matching probability determination model's output is inaccurate.

[0175] Specifically, the data analysis unit instructs the data acquisition unit to acquire multiple training samples and label them to obtain labels. The matching probability determination model is then retrained using the training samples and their labels.

[0176] Optionally, the model unit is also used to train the matching probability determination model.

[0177] Figure 6 This is a schematic diagram of the internal flow of the model unit provided in an embodiment of this application. For example... Figure 5 As shown, the model unit is used to train and use the matching probability to determine the model. Specifically, in the training phase of the matching probability determination model, training samples are obtained from the input data and labeled. That is, the labels of the training samples are determined based on the customer's interaction with the recommended products after recommending products to customers in the corresponding customer-product pair. Then, the matching probability model is trained and generated using the training samples and labels. After training is completed, the matching probability determination model is generated.

[0178] In the phase of using the matching probability determination model, the dynamic and static features of the customer product pair are obtained from the input data. These dynamic and static features are then input into the trained matching probability determination model to obtain the matching probability of the customer product pair.

[0179] Optional, Figure 7 This is a schematic diagram of the internal flow of the recommendation result display unit provided in an embodiment of this application. Figure 7 As shown, the recommendation results display unit includes offline recommendations and application software broadcast recommendations. Offline recommendations involve sending recommended product information to relevant sales personnel so that they can recommend the product to the customer.

[0180] Figure 8 This is a schematic diagram of a product recommendation device provided in an embodiment of this application. Figure 8 As shown, the product recommendation device provided in this embodiment includes a customer feature determination module 801, a product feature determination module 802, a customer-product pair determination module 803, a matching probability determination module 804, and a recommendation module 805.

[0181] The customer feature determination module 801 is used to determine the dynamic features of customers based on their historical data of using application software; and to determine the static features of customers based on their attribute information. The product feature determination module 802 is used to determine the static features of products based on their attribute information; and to determine the dynamic features of products based on their click and purchase records. The customer-product pair determination module 803 is used to pair customers and products based on their dynamic and static features and the dynamic and static features of products to obtain customer-product pairs; dynamic and static features include both dynamic and static features. The matching probability determination module 804 is used to input the dynamic and static features of each customer-product pair into a pre-trained matching probability determination model to obtain the matching probability of the customer-product pair; the dynamic and static features of the customer-product pair include both the dynamic and static features of the customer and the dynamic and static features of the product. The recommendation module 805 is used to recommend products to customers based on the matching probability of each customer-product pair.

[0182] Optionally, the customer product determines module 803, specifically for:

[0183] Based on the dynamic and static characteristics of customers and products, identify customers and products with matching relationships; pair customers and products with matching relationships to obtain customer-product pairs.

[0184] Optionally, the customer's static characteristics include an acceptable risk level; the product's static characteristics include a risk level; the customer's product is used to determine module 803, specifically for:

[0185] If a product's risk level is less than or equal to the customer's acceptable risk level, then the customer and product are identified as having a pairing relationship; pairing customers and products with a pairing relationship results in a customer-product pair.

[0186] Optionally, the static characteristics of the customer also include the average holding period; the dynamic and static characteristics of the product also include the investment period; the customer product determination module 803 is specifically used for:

[0187] If a customer's average holding period is greater than or equal to the product's investment period, then the customer and product are identified as having a pairing relationship; pairing customers and products with a pairing relationship results in a customer-product pair.

[0188] Optionally, the customer's dynamic characteristics include held products and clicked products; the product's dynamic characteristics include priority; the customer product determination module 803 is specifically used for:

[0189] For each customer, a first product set is generated based on the products held by the customer, excluding the held products; a second product set is determined based on the dynamic and static characteristics of the products in the first product set and the dynamic and static characteristics of the customer; a preset number of customers and products with matching relationships are determined based on the priority of the products in the second product set and whether the product is the customer's clicked product; customers and products with matching relationships are paired to obtain customer-product pairs.

[0190] Optional, recommended module 805, specifically used for:

[0191] For each customer, recommend matched products to customers whose matching probability is greater than or equal to the first threshold; the products in the customer's product pair are the matched products for that customer.

[0192] Optional, recommended module 805, specifically used for:

[0193] For each product, customers matched with products whose matching probability is greater than or equal to the second threshold are identified as high-probability recommended customer groups; customers in a customer's product pair are the matched customers of that product; based on the dynamic and static characteristics of matched customers of products whose matching probability is greater than or equal to the third threshold and less than the second threshold, potential demand recommended customer groups are identified; and products are recommended to customers based on the customer groups they belong to.

[0194] Optionally, the product recommendation device also includes a model training module for:

[0195] Obtain multiple training samples and their labels; each training sample includes the dynamic and static features of the corresponding customer product pair, and the labels of the training samples are determined based on the customer's interaction with the recommended products after recommending them to the customer in the corresponding customer product pair; use the training samples and their labels to train the matching probability determination model.

[0196] The product recommendation device provided in this application embodiment can be used to execute the technical solution of the product recommendation method provided in any of the above embodiments of this application. Its implementation principle and technical effect are similar, and will not be described again here.

[0197] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 9 As shown, the electronic device of this embodiment may include: at least one processor 901; and a memory 902 communicatively connected to at least one processor; wherein the memory 902 stores instructions that can be executed by at least one processor 901, and the instructions are executed by at least one processor 901 to cause the electronic device to perform the method as described in any of the above embodiments.

[0198] Optionally, the memory 902 can be either standalone or integrated with the processor 901. When the memory 902 is configured independently, the device also includes a bus for connecting the memory 902 and the processor 901.

[0199] The implementation principle and technical effects of the electronic device provided in this embodiment can be found in the foregoing embodiments, and will not be repeated here.

[0200] This application also provides a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are executed by a processor, the methods provided in any of the foregoing embodiments can be implemented.

[0201] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the method provided in any of the foregoing embodiments.

[0202] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0203] It should be further noted that although the steps in the flowchart are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowchart may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0204] It should be understood that the above-described device embodiments are merely illustrative, and the device of this application can also be implemented in other ways. For example, the division of units / modules in the above embodiments is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units, modules, or components may be combined, or integrated into another system, or some features may be ignored or not executed.

[0205] Furthermore, unless otherwise specified, the functional units / modules in the various embodiments of this application can be integrated into one unit / module, or each unit / module can exist physically separately, or two or more units / modules can be integrated together. The integrated units / modules described above can be implemented in hardware or as software program modules.

[0206] When integrated units / modules are implemented in hardware, the hardware can be digital circuits, analog circuits, etc. The physical implementation of the hardware structure includes, but is not limited to, transistors, memristors, etc. Unless otherwise specified, the processor can be any suitable hardware processor, such as a CPU, GPU, FPGA, DSP, and ASIC, etc. Unless otherwise specified, the storage unit can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc.

[0207] If the integrated unit / module is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0208] In the above embodiments, the descriptions of each embodiment have their own emphasis. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0209] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0210] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A product recommendation method, characterized in that, include: Based on the customer's historical data on application software usage, determine the customer's dynamic characteristics; Based on the customer's attribute information, the customer's static characteristics are determined; Based on the product's attribute information, determine the product's static characteristics; Based on the product's click and purchase records, the dynamic characteristics of the product are determined; Based on the dynamic and static characteristics of the customer and the product, the customer and the product are matched to obtain customer-product pairs; the dynamic and static characteristics include dynamic characteristics and static characteristics. For each customer product pair, the dynamic and static features of the customer product pair are input into a pre-trained matching probability determination model to obtain the matching probability of the customer product pair; the dynamic and static features of the customer product pair include the dynamic and static features of the customer and the dynamic and static features of the product. Based on the matching probability of each customer product pair, products are recommended to the customer.

2. The method according to claim 1, characterized in that, The process of pairing the customer and the product based on the dynamic and static characteristics of the customer and the product to obtain customer-product pairs includes: Based on the dynamic and static characteristics of the customers and the dynamic and static characteristics of the products, customers and products with a pairing relationship are determined; The existing customers are paired with products to obtain customer-product pairs.

3. The method according to claim 2, characterized in that, The customer's static characteristics include an acceptable risk level; The static characteristics of the product include a risk level; the process of determining a pairing relationship between a customer and a product based on the dynamic and static characteristics of the customer and the product includes: If a product's risk level is less than or equal to the customer's acceptable risk level, then the customer and the product are identified as a paired customer and product.

4. The method according to claim 3, characterized in that, The static characteristics of the customer also include the average holding period; the dynamic and static characteristics of the product also include the investment period; the step of determining the matching relationship between customers and products based on the dynamic and static characteristics of the customer and the dynamic and static characteristics of the product further includes: If a customer's average holding period is greater than or equal to the product's investment period, then the customer and the product are identified as having a paired relationship.

5. The method according to claim 2, characterized in that, The dynamic characteristics of the customer include the products they hold and the products they click on; the dynamic characteristics of the product include priority; based on the dynamic and static characteristics of the customer and the dynamic and static characteristics of the product, determining the customers and products with a matching relationship includes: For each of the aforementioned customers, a first set of products other than the products held by the customer is generated based on the products held by the customer. The second product set is determined based on the dynamic and static characteristics of the products in the first product set and the dynamic and static characteristics of the customer; Based on the priority of the products in the second product set and whether the product is the customer's clicked product, a preset number of customers and products with matching relationships are determined.

6. The method according to claim 1, characterized in that, The step of recommending products to customers based on the matching probability of each customer product pair includes: For each customer, a matching product with a matching probability greater than or equal to a first threshold is recommended to that customer; the product in the customer product pair is the matching product of that customer.

7. The method according to claim 1, characterized in that, The step of recommending products to customers based on the matching probability of each customer product pair includes: For each product, customers whose matching probability is greater than or equal to the second threshold are identified as high-probability recommended customer groups; customers in the customer-product pair are the matched customers of the products in that pair. Based on the dynamic and static characteristics of the matched customers of the product whose matching probability is greater than or equal to the third threshold and less than the second threshold, a potential demand recommended customer group is determined. Recommend products to the customer based on the customer group they belong to.

8. The method according to any one of claims 1-7, characterized in that, The method further includes: Multiple training samples and their labels are obtained; each training sample includes the dynamic and static features of the corresponding customer product pair, and the label of the training sample is determined based on the customer's interaction with the recommended product after recommending the product to the customer in the corresponding customer product pair. The matching probability determination model is trained using the training samples and their labels.

9. A product recommendation device, characterized in that, include: The customer characteristic determination module is used to determine the dynamic characteristics of the customer based on the customer's historical data on the application software. Based on the customer's attribute information, the customer's static characteristics are determined; The product feature determination module is used to determine the static features of the product based on the product's attribute information; Based on the product's click and purchase records, the dynamic characteristics of the product are determined; The customer-product pairing module is used to pair the customer and the product based on the dynamic and static characteristics of the customer and the dynamic and static characteristics of the product to obtain customer-product pairs; the dynamic and static characteristics include dynamic characteristics and static characteristics; The matching probability determination module is used to input the dynamic and static features of the customer product pair into a pre-trained matching probability determination model for each customer product pair to obtain the matching probability of the customer product pair; the dynamic and static features of the customer product pair include the dynamic and static features of the customer and the dynamic and static features of the product. The recommendation module is used to recommend products to customers based on the matching probability of each customer product pair.

10. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 8.

12. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method of any one of claims 1 to 8.