Financial product recommendation method, device, equipment, storage medium and program product

Abstract

The application provides a financial product recommendation method and device, equipment, a storage medium and a program product, which can be applied to the fields of artificial intelligence technology and big data technology. The method comprises the following steps: obtaining a plurality of multidimensional data of a target customer; inputting the a plurality of multidimensional data of the target customer into a first financial product recommendation model generated by pre-training; outputting n first recommended financial products; and sending the n first recommended financial products to the target customer respectively. n first feedback data of the target customer after receiving the n first recommended financial products is obtained. Based on the n first feedback data, the first financial product recommendation model is fine-tuned to generate a target financial product recommendation model. The a plurality of multidimensional data of the target customer are inputted into the target financial product recommendation model, and a target recommended financial product is outputted.

Description

Technical Field

[0001] This disclosure relates to the field of big data technology, specifically to the field of artificial intelligence technology, and in particular to a method, apparatus, device, storage medium, and program product for recommending financial products. Background Technology

[0002] With the development of computer technology, enterprises, including banks, are increasingly demanding the ability to dynamically classify customer profiles and accurately recommend suitable financial products to different customer categories. Traditional customer profile classification often uses static models, which require first fixing customer categories based on historical data and then matching preset financial products to these fixed categories.

[0003] However, this method cannot respond to changes in customer behavior in real time. It relies solely on historical data and is difficult to capture potential customer needs. This results in poor matching between the recommended financial products and the customer's real-time needs, leading to technical problems such as low accuracy, poor reliability, low suitability, poor customer conversion rate, and poor user satisfaction in product recommendations. Summary of the Invention

[0004] In view of the above problems, this application provides a method, apparatus, device, medium and program product for recommending financial products.

[0005] According to a first aspect of this application, a method for recommending financial products is provided. The method includes: acquiring *a* multidimensional data points of a target customer; inputting the *a* multidimensional data points of the target customer into a pre-trained first financial product recommendation model to output *n* first recommended financial products; and sending the *n* first recommended financial products to the target customer, respectively, where *a* is an integer and *a* is greater than 0, and *n* is an integer and *n* is greater than 0; acquiring *n* first feedback data points from the target customer after receiving the *n* first recommended financial products; fine-tuning the first financial product recommendation model based on the *n* first feedback data points to generate a target financial product recommendation model; and inputting the *a* multidimensional data points of the target customer into the target financial product recommendation model to output the target recommended financial products.

[0006] According to an embodiment of this application, inputting a multidimensional data points of the target customer into a pre-trained first financial product recommendation model and outputting n first recommended financial products includes: extracting features from the a multidimensional data points of the target customer to generate c feature vectors of the target customer, where c is an integer, c is greater than 0 and c is less than a; determining the change range of the c feature vectors of the target customer; and if the change range of at least one of the c feature vectors exceeds a first preset threshold, then inputting the c feature vectors of the target customer into the pre-trained first financial product recommendation model and outputting n first recommended financial products.

[0007] According to an embodiment of this application, feature extraction is performed on a multidimensional data points of the target customer to generate c feature vectors of the target customer, including: data cleaning of the a multidimensional data points of the target customer to generate b cleaned multidimensional data points, where b is an integer, b is greater than c and b is less than a; feature encoding is performed on the b cleaned multidimensional data points to generate b first vectors of the target customer; and feature filtering is performed on the b first vectors of the target customer to generate c feature vectors of the target customer.

[0008] According to an embodiment of this application, feature filtering is performed on b first vectors of the target customer to generate c feature vectors of the target customer, including: obtaining historical purchase data of the target customer, calculating the correlation values ​​between the b first vectors and the historical purchase data of the target customer respectively, generating b correlation values; and obtaining c first vectors corresponding to the correlation values ​​c among the b correlation values ​​that are greater than a first preset threshold, as c feature vectors of the target customer.

[0009] According to an embodiment of this application, inputting c feature vectors of the target customer into a pre-trained first financial product recommendation model and outputting n first recommended financial products includes: labeling the c feature vectors of the target customer with attention weights to generate c feature vectors of the target customer labeled with attention weights; and inputting the c feature vectors of the target customer labeled with attention weights into the pre-trained first financial product recommendation model and outputting n first recommended financial products.

[0010] According to an embodiment of this application, training and generating the first financial product recommendation model includes: acquiring the historical behavior data of the target customer, the historical behavior data of the target customer including: i historical multidimensional data of the target customer, j historical recommended financial products of the target customer, and j historical feedback data of the target customer, wherein i is an integer and i is greater than 0, and j is an integer and j is greater than 0; and using the i historical multidimensional data of the target customer as the state input of the model, using the j historical recommended financial products of the target customer as the action output of the model, and using the j historical feedback data of the target customer as the reward of the action output of the model, training the initial financial product recommendation model to generate the first financial product recommendation model.

[0011] According to an embodiment of this application, based on the n first feedback data, the first financial product recommendation model is fine-tuned to generate a target financial product recommendation model, including: performing clustering calculations on the n first feedback data using a clustering algorithm to generate the category features of the target customer; optimizing the state parameters of the first financial product recommendation model based on the category features of the target customer to generate a second financial product recommendation model; and optimizing the action value function of the second financial product recommendation model using a temporal difference learning algorithm to generate the target financial product recommendation model.

[0012] According to a second aspect of this application, a financial product recommendation device is provided, comprising: a first generation module, configured to acquire a multidimensional data of a target customer, input the a multidimensional data of the target customer into a pre-trained first financial product recommendation model, output n first recommended financial products, and send the n first recommended financial products to the target customer respectively, wherein a is an integer and a is greater than 0, and n is an integer and n is greater than 0; a first acquisition module, configured to acquire n first feedback data after the target customer receives the n first recommended financial products; a second generation module, configured to fine-tune the first financial product recommendation model based on the n first feedback data to generate a target financial product recommendation model; and a third generation module, configured to input the a multidimensional data of the target customer into the target financial product recommendation model and output the target recommended financial products.

[0013] According to a third aspect of this application, an electronic device is provided, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0014] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0015] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0016] According to embodiments of this application, by first recommending a batch of products to customers using an initial model, collecting customer feedback data on these recommendations, then fine-tuning the model using this feedback data to generate an optimized model, and finally using this optimized model to provide final product recommendations to customers, this technical approach can effectively improve the accuracy and reliability of the financial product recommendations, significantly reduce the push of invalid financial products, save computer and network resources, and improve customer conversion rates and customer experience. It solves the technical problems of traditional static model-based financial product recommendation methods, which cannot respond to changes in customer behavior in real time, rely solely on historical data association, and struggle to capture potential customer needs, resulting in poor matching between recommended financial products and real-time customer needs, leading to low accuracy, poor reliability, low adaptability, low customer conversion rates, and poor user satisfaction. Attached Figure Description

[0017] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0018] Figure 1 The illustrations depict application scenarios of financial product recommendation methods, apparatuses, devices, media, and program products according to embodiments of this application.

[0019] Figure 2 A flowchart illustrating a financial product recommendation method according to an embodiment of this application is shown schematically.

[0020] Figure 3 This illustration schematically shows a flowchart of generating n first recommended financial products in a financial product recommendation method according to an embodiment of this application;

[0021] Figure 4 This illustration schematically shows a flowchart of generating c feature vectors of a target customer in a financial product recommendation method according to an embodiment of this application;

[0022] Figure 5 This illustration schematically shows a flowchart of the feature vector labeling process in a financial product recommendation method according to an embodiment of this application;

[0023] Figure 6 This illustration schematically shows a flowchart of the process for generating a target financial product recommendation model in a financial product recommendation method according to an embodiment of this application;

[0024] Figure 7 This illustration schematically shows a diagram of one embodiment of the financial product recommendation method according to embodiments of this application;

[0025] Figure 8 This schematic diagram illustrates a structural block diagram of a financial product recommendation device according to an embodiment of this application;

[0026] Figure 9 A block diagram schematically illustrates an electronic device suitable for implementing a financial product recommendation method according to an embodiment of this application. Detailed Implementation

[0027] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0028] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0029] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0030] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0031] The accompanying drawings show some block diagrams and / or flowcharts. It should be understood that some blocks or combinations thereof in the block diagrams and / or flowcharts can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable control device, so that when executed by the processor, these instructions can create means for implementing the functions / operations described in these block diagrams and / or flowcharts.

[0032] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved 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 related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0033] In scenarios where personal information is used for automated decision-making, the methods, devices, and systems provided in this application all provide users with corresponding operation entry points for users to choose to agree to or reject the automated decision results; if the user chooses to reject, the process enters the expert decision-making process.

[0034] First, the technical terms used in this article are explained as follows:

[0035] One-hot encoding is a method for converting categorical data into a binary vector representation. It is used to represent a categorical feature with N possible values ​​using a vector of length N. The current value in vector N is set to 1 at its corresponding position, and all other positions are set to 0.

[0036] Min-Max normalization is a data standardization method that linearly transforms the original data to a specified range.

[0037] DQN (Deep Q-Network) is a reinforcement learning algorithm that combines deep neural networks with Q-learning to solve decision-making problems in high-dimensional state spaces.

[0038] APP (Application) refers to a computer program designed to complete a specific task, generally referring to third-party software on mobile devices such as smartphones and tablets.

[0039] This application provides a financial product recommendation method, which includes: acquiring *a* multidimensional data points of a target customer; inputting the *a* multidimensional data points of the target customer into a pre-trained first financial product recommendation model to output *n* first recommended financial products; and sending the *n* first recommended financial products to the target customer, where *a* is an integer and *a* is greater than 0, and *n* is an integer and *n* is greater than 0. The method also includes acquiring *n* first feedback data points from the target customer after receiving the *n* first recommended financial products; fine-tuning the first financial product recommendation model based on the *n* first feedback data points to generate a target financial product recommendation model; and finally inputting the *a* multidimensional data points of the target customer into the target financial product recommendation model to output the target recommended financial products.

[0040] According to the embodiments of this application, by first recommending a batch of products to customers using an initial model, collecting customer feedback data on these recommendations, then using this feedback data to fine-tune the model to generate an optimized model, and finally using this optimized model to provide final product recommendations to customers, the technical means can effectively improve the accuracy and reliability of the output target financial product recommendations, significantly reduce the push of invalid financial products, save computer and network resources, and improve customer conversion rates and customer experience. This solves the technical problems of traditional static model-based financial product recommendation methods, which cannot respond to changes in customer behavior in real time, rely solely on historical data association, and struggle to capture potential customer needs, resulting in poor matching between financial product recommendations and real-time customer needs, leading to low product recommendation accuracy, poor reliability, low adaptability, poor customer conversion rates, and poor user satisfaction.

[0041] Figure 1 An exemplary distributed system architecture 100 that can be applied to the financial product recommendation method according to embodiments of this application is illustrated. It should be noted that... Figure 1 The examples shown are merely examples of the architecture that can be applied to the methods according to the embodiments of this application, in order to help those skilled in the art understand the technical content of this application, but do not imply any limitation on the application scenarios of the embodiments of this application.

[0042] Figure 1 The illustration depicts an application scenario of the financial product recommendation method according to an embodiment of this application. For example... Figure 1As shown, application scenario 100 according to an embodiment of this application may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables. For example, a user can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send information, etc.

[0043] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be electronic devices such as smartphones, wearable devices, personal computers, intelligent voice interaction devices, smart home appliances, intelligent vehicles, in-vehicle terminals, aircraft, unmanned vending terminals, and extended reality devices. Extended reality devices can include virtual reality devices, augmented reality devices, and mixed reality devices. A client application for the target application can be installed and run on the terminal devices. This target application can include, but is not limited to, financial transaction applications, payment applications, shopping applications, web browser applications, search applications, instant messaging tools, email clients, and social media platform software (these are just examples). Furthermore, this application embodiment does not limit the form of the target application, and it can include, but is not limited to, applications, mini-programs, etc., installed on the terminal devices, and can also be in the form of web pages.

[0044] Server 105 can be a server providing various services, such as a backend management server supporting websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process received user requests and other data, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services such as cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks, and basic cloud computing services such as big data. The server can be the backend server of the aforementioned target application, used to provide backend services to the clients of the target application.

[0045] It should be noted that the financial product recommendation method provided in this application embodiment can generally be executed by server 105 and / or terminal devices 101-103. Accordingly, the financial product recommendation device provided in this application embodiment can generally be set in server 105 and / or terminal devices 101-103.

[0046] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0047] The following will be based on Figure 1 The described scene, through Figures 2-7 The financial product recommendation method of the disclosed embodiments is described in detail. It should be noted that the above application scenarios are shown only to facilitate understanding of the spirit and principles of this application, and the implementation methods of this application are not limited in any way. On the contrary, the implementation methods of this application can be applied to any applicable scenario.

[0048] Figure 2 A flowchart illustrating a financial product recommendation method according to an embodiment of this application is shown.

[0049] like Figure 2 As shown, the method 200 includes steps S210 to S240.

[0050] Step S210: Obtain a multidimensional data of the target customer, input the a multidimensional data of the target customer into the pre-trained first financial product recommendation model, output n first recommended financial products, and send the n first recommended financial products to the target customer respectively, where a is an integer and a is greater than 0, and n is an integer and n is greater than 0.

[0051] According to the embodiments of this application, obtaining *a* multidimensional data points of a target customer can include: real-time collection of *a* multidimensional data points of the customer through bank apps, counter systems, and customer service records. The *a* multidimensional data points of the customer can include: basic customer attributes, such as age, occupation, and region; customer asset characteristics, such as deposit balance and wealth management product holdings; transaction behavior data, such as average monthly transfer frequency and consumption frequency; behavioral preferences, such as historical product click records and consultation types; and incremental behavioral data, such as new transaction records and product operation logs. These multidimensional data points can be processed through feature normalization into vector data that the model can recognize for model input.

[0052] Figure 3 The flowchart illustrating the generation of n first recommended financial products in the financial product recommendation method according to an embodiment of this application is shown.

[0053] like Figure 3As shown, the method 300 includes steps S310 to S330.

[0054] Step S310: Extract features from the a multidimensional data of the target customer to generate c feature vectors of the target customer, where c is an integer, c is greater than 0 and c is less than a.

[0055] Figure 4 The flowchart illustrating the generation of c feature vectors of a target customer in a financial product recommendation method according to an embodiment of this application is shown.

[0056] like Figure 4 As shown, the method 400 includes steps S410 to S430.

[0057] Step S410: Perform data cleaning on the a multidimensional data of the target customer to generate b multidimensional data after data cleaning, where b is an integer, b is greater than c and b is less than a.

[0058] For example, data cleaning can be performed on a multidimensional dataset of a target customer by handling missing values ​​and outliers. In handling missing values, missing basic attributes can be coded using the "unknown" label, and missing asset characteristics can be filled in using the median of the customer's customer group. In handling outliers, outliers in transaction habits can be filtered out, or outliers can be replaced to complete the data cleaning process.

[0059] Step S420: Perform feature encoding on the b multidimensional data after data cleaning to generate b first vectors for the target customer.

[0060] For example, we can first classify the b multidimensional data after data cleaning into categorical feature data or continuous feature data. Categorical feature data generally includes occupational data or geographical data, etc. Continuous feature data generally includes age data or bank balance data, etc. We use One-Hot encoding for categorical feature data and Min-Max normalization for continuous feature data, mapping the data to the [0, 1] interval to generate the first vector of the target customer.

[0061] Step S430: Perform feature filtering on the b first vectors of the target customer to generate c feature vectors of the target customer.

[0062] According to the embodiments of this application, historical purchase data of the target customer can be obtained, and the correlation values ​​between the b first vectors and the historical purchase data of the target customer can be calculated to generate b correlation values. The c first vectors corresponding to the correlation values ​​c that are greater than a second preset threshold among the b correlation values ​​are obtained and used as the c feature vectors of the target customer.

[0063] For example, the target customer's historical purchase data can be historical customer purchase behavior data. We can calculate the information entropy of b first vectors and the target customer's historical purchase data, generating b information entropies as correlation values. A preset information entropy threshold is set, and c first vectors corresponding to information entropies greater than the threshold are obtained from the b information entropies. These are used as c feature vectors for the target customer, eliminating redundant features and completing data feature dimensionality reduction.

[0064] Alternatively, the cosine similarity between each of the b first vectors and the vectors generated from the historical purchase data of the target customer can be calculated to generate b cosine similarity values ​​as association values. A preset cosine similarity threshold is then set, and the c first vectors corresponding to the cosine similarities greater than the threshold are obtained from the b cosine similarity values ​​and used as the c feature vectors of the target customer.

[0065] By using the historical purchase data of target customers as a benchmark to filter the features of b first vectors and generate c feature vectors of target customers, the technology can effectively achieve personalized feature filtering based on individual behavioral preferences, effectively eliminate feature dimensions that are weakly or unrelated to historical purchase behavior, reduce the computational complexity of downstream models, enhance the interpretability of downstream model output, and optimize the utilization of computer resources.

[0066] Data cleaning effectively improves data quality and eliminates noise interference; data feature encoding enables unified representation of heterogeneous, multi-dimensional data; and feature filtering effectively reduces feature dimensionality and optimizes model input. This three-tiered, progressive data processing workflow—data cleaning, data encoding, and feature filtering—achieves the technical effects of purifying data input, unifying data feature representation, and reducing model complexity. It effectively solves the technical problems of inconsistent raw data, difficulty in processing multiple heterogeneous data sets uniformly, and excessively high feature dimensionality leading to decreased model performance.

[0067] Return to reference Figure 3 In step S320, the variation range of the c feature vectors of the target customer is determined.

[0068] For example, the magnitude of change of c feature vectors of a target customer can be calculated, which may specifically include: calculating the magnitude of fluctuation of the asset balance feature vector or the magnitude of change of the transaction frequency feature vector.

[0069] Step S330: If the change magnitude of at least one of the c feature vectors exceeds the first preset threshold, then the c feature vectors of the target customer are input into the pre-trained first financial product recommendation model, and n first recommended financial products are output.

[0070] For example, when the asset balance fluctuates by more than a preset threshold of 20% or the frequency of transaction habits changes by more than a preset threshold of 50%, a potential category recalculation is triggered to update the customer's category. Then, the c feature vectors of the target customer are input into the first financial product recommendation model, outputting n first recommended financial products. Updating the customer's category can include changing "savings-oriented customer" to "cash flow-oriented customer." Furthermore, before inputting the feature vectors into the first financial product recommendation model, the feature vectors can be labeled to add an attention layer, facilitating the assignment of dynamic weights to different dimensions of features, highlighting recent behavioral characteristics and high-value asset characteristics, and improving the model's ability to capture key information. Recent behavioral characteristics can include product clicks in the past 7 days, and high-value asset characteristics can include large-amount wealth management holdings data.

[0071] Figure 5 The flowchart illustrating the labeling of feature vectors in a financial product recommendation method according to an embodiment of this application is shown.

[0072] like Figure 5 As shown, the method 500 includes steps S510 to S520.

[0073] Step S510: Mark the c feature vectors of the target customer with attention weights to generate c feature vectors of the target customer marked with attention weights.

[0074] Step S520: Input the c feature vectors of the target customer labeled with attention weights into the pre-trained first financial product recommendation model, and output n first recommended financial products.

[0075] By using the attention weight labeling mechanism, the model can be guided to prioritize key features with higher weights during the inference process. This strengthens features that are strongly correlated with the customer's true preferences, while weakening the interference of redundant or noisy features. This effectively enhances the model's ability to focus on key features and improves the accuracy and stability of the recommendation output.

[0076] By extracting and reducing features, core feature vectors can be identified, optimizing model input data, reducing model inference computation, improving model response speed, and effectively enhancing the stability and accuracy of model output. Simultaneously, by employing change-aware threshold processing, invalid model calculations and pushes can be effectively avoided during periods of stable client states, significantly reducing the computational and network transmission resource consumption of the computing system.

[0077] According to an embodiment of this application, training and generating the first financial product recommendation model includes: acquiring the historical behavior data of the target customer, the historical behavior data of the target customer including: i historical multidimensional data of the target customer, j historical recommended financial products of the target customer, and j historical feedback data of the target customer, wherein i is an integer and i is greater than 0, and j is an integer and j is greater than 0; and using the i historical multidimensional data of the target customer as the state input of the model, using the j historical recommended financial products of the target customer as the action output of the model, and using the j historical feedback data of the target customer as the reward of the action output of the model, training the initial financial product recommendation model to generate the first financial product recommendation model.

[0078] By using i historical multidimensional data points of the target customer as the state, j historical financial product recommendations of the target customer as the action, and j historical feedback data of the target customer as the reward, a closed loop of state, action, and reward model training is constructed. This enables the initial financial product recommendation model to be trained based on historical data of real interactions, significantly improving the utilization rate and training credibility of training samples, and effectively enhancing the reliability and accuracy of the generated first financial product recommendation model.

[0079] Return to reference Figure 2 In step S220, n first feedback data are obtained after the target customer receives the n first recommended financial products.

[0080] Step S230: Based on the n first feedback data, fine-tune the first financial product recommendation model to generate the target financial product recommendation model.

[0081] Figure 6 The flowchart illustrating the generation of a target financial product recommendation model in a financial product recommendation method according to an embodiment of this application is shown in the illustration.

[0082] like Figure 6 As shown, the method 600 includes steps S610 to S630.

[0083] Step S610: Cluster the n first feedback data using a clustering algorithm to generate the category features of the target customer.

[0084] For example, if n is 3, based on the customer's feedback on the first 3 recommended actions ( , , The system uses clustering algorithms such as K-means to generate temporary "potential categories" such as "sensitive to equity funds but not buying them for the time being" and "those who prefer stable financial management" as category features of target customers, and incorporates these category features as auxiliary features into the subsequent environmental state.

[0085] Step S620: Optimize the state parameters of the first financial product recommendation model based on the category characteristics of the target customer to generate a second financial product recommendation model.

[0086] For example, by incorporating the category characteristics of target customers as auxiliary features into the subsequent environmental state, the state parameters of the first financial product recommendation model can be optimized to generate the second financial product recommendation model.

[0087] Step S630: Optimize the action value function of the second financial product recommendation model using a temporal difference learning algorithm to generate the target financial product recommendation model.

[0088] For example, the action value function can be a Q-value function, which can be updated using a temporal difference learning algorithm with the goal of "maximizing cumulative reward". The formula is as follows:

[0089]

[0090] in:

[0091] : Represents the state Take action below Current Q value estimate;

[0092] α: Represents the learning rate, which can be set to 0.1 to control the step size of each update;

[0093] : Represents the state Take action The instant reward obtained afterward;

[0094] γ: Represents the discount factor, which can be set to 0.9 to weigh the importance of future rewards;

[0095] In subsequent states The largest Q value among all possible actions represents the estimate of the optimal reward in the future.

[0096] The formula as a whole expresses the following: The current Q-value estimate is corrected using "immediate reward + discounted future optimal reward," enabling the second financial product recommendation model to gradually learn a better strategy. This progressively optimizes the "state-action" mapping strategy, making the recommended actions more aligned with the customer's potential category needs, thereby generating the target financial product recommendation model.

[0097] By using clustering feedback to generate category features and then optimizing model state parameters and model action value functions based on these features, the model's state representation and recommendation action evaluation can be made more accurate. This achieves a gradual improvement in model performance and enhances its dynamic adaptability to customer preferences, thereby increasing the reliability and stability of the generated target financial product recommendation model.

[0098] Return to reference Figure 2 In step S240, the a-dimensional data of the target customer are input into the target financial product recommendation model, and the target recommended financial product is output.

[0099] Figure 7 The illustration shows a schematic diagram of one embodiment of the financial product recommendation method according to embodiments of this application.

[0100] like Figure 7 As shown, a memory module and a calling mechanism can be added to the financial product recommendation model to generate a reinforcement learning agent for financial product recommendation. The reinforcement learning agent can adopt a DQN architecture, which includes: an input layer that receives the environment state vector, a hidden layer with three fully connected layers and activation functions, an attention layer with a multi-head attention mechanism, and an output layer that outputs the Q-values ​​of each product recommendation action.

[0101] The reinforcement learning agent receives customer feature data from the customer feature input module as its environmental state. This customer feature data can include basic attributes, asset characteristics, transaction habits, and behavioral preferences. The agent processes this data and outputs product recommendation actions. These actions are defined by a set of financial products that the bank can promote. For example, product recommendation actions could include "recommend a stable wealth management product," "recommend personal consumer loans," "recommend large-denomination certificates of deposit," or "not recommend any products for now." Each action corresponds to a unique operation identifier.

[0102] Furthermore, the reinforcement learning agent updates its Q-score to optimize the financial product recommendation model by acquiring feedback and reward data. Feedback and reward data typically employ a combination mechanism of "immediate rewards + delayed rewards + long-term value rewards." Immediate rewards are the customer's immediate feedback on the recommended product. For example, clicking a product link earns 1 point, purchasing the product earns 5 points, and rejecting the recommendation earns -0.5 points. Delayed rewards are the customer's behavioral feedback over the next 30 days. For example, holding the product for 30 days after purchase earns 3 points, early redemption deducts 2 points, and proactively inquiring about the product earns 2 points. Long-term value rewards are the customer's annual increase in comprehensive financial assets. For example, 1 point is added for every 10,000 yuan increase, up to a maximum of 10 points.

[0103] In addition, a closed loop of real-time data feedback and model updates can be established through real-time data collection, dynamic category adjustment, and regular model iteration to ensure dynamic adaptation between customer categories and recommendation strategies.

[0104] Real-time data collection can include: collecting customer behavior data such as new transaction records and product operation logs in real time through bank apps, counter systems, and customer service records, and ensuring that customer feature vectors are updated every 24 hours.

[0105] Dynamic category adjustment can include: when the change in a customer's feature vector exceeds a preset threshold, the model automatically triggers a recalculation of the latent category and updates the customer's category.

[0106] Regular model iterations may include: fine-tuning the network parameters of the reinforcement learning model weekly based on the "state-action-reward" interaction data of all customers, updating the action space such as adding popular financial products and removing slow-selling products, and retraining the attention layer weights to ensure that the model strategy iterates in sync with business needs.

[0107] According to embodiments of this application, this method, by capturing changes in customer behavior in real time, shortens the response time for customer category adjustments from the traditional "monthly updates" or even "yearly updates" to "daily updates," effectively improving category matching accuracy. Through reinforcement learning, it autonomously learns the optimal mapping strategy, increasing customer click-through rates for product recommendations and translating this into higher actual purchase rates, effectively reducing the interference of "ineffective recommendations" on the customer experience. A delayed feedback reward mechanism encourages the model to focus on long-term customer behavior, improving 30-day retention rates and repeat purchase rates after product purchases, significantly enhancing customer stickiness and business sustainability.

[0108] Figure 8 A schematic block diagram of a financial product recommendation device according to an embodiment of this application is shown.

[0109] like Figure 8 As shown, the device 800 includes: a first generation module 810, a first acquisition module 820, a second generation module 830, and a third generation module 840.

[0110] The first generation module 810 is used to acquire *a* multidimensional data points of the target customer, input the *a* multidimensional data points of the target customer into a pre-trained first financial product recommendation model, output *n* first recommended financial products, and send the *n* first recommended financial products to the target customer respectively, where *a* is an integer and *a* is greater than 0, and *n* is an integer and *n* is greater than 0. In one embodiment, the first generation module 810 can be used to execute step S210 described above.

[0111] The first generation module 810 includes: a fourth generation module, a first determination module, and a fifth generation module.

[0112] The fourth generation module is used to extract features from the *a* multidimensional data of the target customer and generate *c* feature vectors of the target customer, where *c* is an integer, *c* is greater than 0 and *c* is less than *a*. In one embodiment, the fourth generation module can be used to perform step S310 described above.

[0113] The fourth generation module includes: the sixth generation module, the seventh generation module, and the eighth generation module.

[0114] The sixth generation module is used to clean the a multidimensional data points of the target customer and generate b cleaned multidimensional data points, where b is an integer, b is greater than c and b is less than a. In one embodiment, the sixth generation module can be used to execute step S410 described above, which will not be repeated here.

[0115] The seventh generation module is used to perform feature encoding on the b multidimensional data after data cleaning to generate b first vectors for the target customer. In one embodiment, the seventh generation module can be used to execute step S420 described above, which will not be repeated here.

[0116] The eighth generation module is used to perform feature filtering on the b first vectors of the target customer to generate c feature vectors of the target customer. In one embodiment, the eighth generation module can be used to perform step S430 described above.

[0117] The eighth generation module includes: the ninth generation module and the second acquisition module.

[0118] The ninth generation module is used to obtain the historical purchase data of the target customer, calculate the correlation values ​​between the b first vectors and the historical purchase data of the target customer, and generate b correlation values.

[0119] The second acquisition module is used to acquire c first vectors corresponding to c association values ​​that are greater than a first preset threshold among the b association values, and use them as c feature vectors of the target customer.

[0120] The first determining module is used to determine the change magnitude of the c feature vectors of the target customer. In one embodiment, the first determining module can be used to execute step S320 described above, which will not be repeated here.

[0121] The fifth generation module is configured to, if the change magnitude of at least one of the c feature vectors exceeds a first preset threshold, input the c feature vectors of the target customer into a pre-trained first financial product recommendation model and output n first recommended financial products. In one embodiment, the fifth generation module can be used to execute step S330 described above.

[0122] The fifth generation module includes the tenth and eleventh generation modules.

[0123] The tenth generation module is used to label the c feature vectors of the target customer with attention weights, and generate c feature vectors of the target customer labeled with attention weights. In one embodiment, the tenth generation module can be used to execute step S510 described above, which will not be repeated here.

[0124] The eleventh generation module is used to input c feature vectors of the target customer labeled with attention weights into the pre-trained first financial product recommendation model, and output n first recommended financial products. In one embodiment, the eleventh generation module can be used to execute step S520 described above, which will not be repeated here.

[0125] The first generation module also includes a third acquisition module and a twelfth generation module.

[0126] The third acquisition module is used to acquire the historical behavior data of the target customer. The historical behavior data of the target customer includes: i historical multidimensional data of the target customer, j historical recommended financial products of the target customer, and j historical feedback data of the target customer, wherein i is an integer and i is greater than 0, and j is an integer and j is greater than 0.

[0127] The twelfth generation module is used to train the initial financial product recommendation model by using i historical multidimensional data of the target customer as the state input of the model, j historical recommended financial products of the target customer as the action output of the model, and j historical feedback data of the target customer as the reward of the model action output, to generate the first financial product recommendation model.

[0128] The first acquisition module 820 is used to acquire n first feedback data points after the target customer receives the n first recommended financial products. In one embodiment, the first acquisition module 820 can be used to execute step S220 described above, which will not be repeated here.

[0129] The second generation module 830 is used to fine-tune the first financial product recommendation model based on the n first feedback data to generate a target financial product recommendation model. In one embodiment, the second generation module 830 can be used to execute step S230 described above.

[0130] The second generation module 830 includes: the thirteenth generation module, the fourteenth generation module, and the fifteenth generation module.

[0131] The thirteenth generation module is used to perform clustering calculations on the n first feedback data using a clustering algorithm to generate the category features of the target customer. In one embodiment, the thirteenth generation module can be used to execute step S610 described above, which will not be repeated here.

[0132] The fourteenth generation module is used to optimize the state parameters of the first financial product recommendation model based on the category characteristics of the target customer, and generate a second financial product recommendation model. In one embodiment, the fourteenth generation module can be used to execute step S620 described above, which will not be repeated here.

[0133] The fifteenth generation module is used to optimize the action value function of the second financial product recommendation model using a temporal difference learning algorithm to generate the target financial product recommendation model. In one embodiment, the fifteenth generation module can be used to execute step S630 described above, which will not be repeated here.

[0134] The third generation module 840 is used to input the a-dimensional data of the target customer into the target financial product recommendation model and output the target recommended financial product. In one embodiment, the third generation module 840 can be used to execute step S240 described above, which will not be repeated here.

[0135] According to embodiments of this application, any plurality of modules among the first generation module 810, the first acquisition module 820, the second generation module 830, and the third generation module 840 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in one module. According to embodiments of this application, at least one of the first generation module 810, the first acquisition module 820, the second generation module 830, and the third generation module 840 can be at least partially implemented as hardware circuitry, such as field-programmable gate arrays, programmable logic arrays, systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits, or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in any one of software, hardware, and firmware methods, or in a suitable combination of any of these methods. Alternatively, at least one of the first generation module 810, the first acquisition module 820, the second generation module 830, and the third generation module 840 may be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0136] Figure 9 A block diagram schematically illustrates an electronic device suitable for implementing a financial product recommendation method according to an embodiment of this application.

[0137] like Figure 9 As shown, an electronic device 900 according to an embodiment of this application includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory 902 or a program loaded from a storage portion 908 into a random access memory 903. The processor 901 may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a dedicated microprocessor. The processor 901 may also include onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for executing different steps of the method flow according to an embodiment of this application.

[0138] Random access memory 903 stores various programs and data required for the operation of electronic device 900. Processor 901, read-only memory 902, and random access memory 903 are interconnected via bus 904. Processor 901 executes various steps of the method flow according to embodiments of this application by executing programs stored in read-only memory 902 and / or random access memory 903. It should be noted that the programs may also be stored in one or more memories other than read-only memory 902 and random access memory 903. Processor 901 may also execute various steps of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0139] According to embodiments of this application, the electronic device 900 may further include an input / output interface 905, which is also connected to a bus 904. The electronic device 900 may also include one or more of the following components connected to the input / output interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube, liquid crystal display, etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card, such as a local area network card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 910 as needed so that computer programs read from it can be installed into the storage section 908 as needed.

[0140] Embodiments of this application also provide a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0141] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof. In embodiments of this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include the read-only memory 902 described above, and / or random access memory 903, and / or one or more memories other than read-only memory 902 and random access memory 903.

[0142] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the methods provided in the embodiments of this application.

[0143] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 909, and / or installed from a removable medium 911. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0144] In embodiments of this application, the computer program can be downloaded and installed from a network via communication section 909, and / or installed from removable medium 911. When the computer program is executed by processor 901, it performs the functions defined in the system of embodiments of this application. According to embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0145] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0146] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0147] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A method for recommending financial products, characterized in that, The method includes: Acquire a multidimensional data points of the target customer, input the a multidimensional data points of the target customer into a pre-trained first financial product recommendation model, output n first recommended financial products, and send the n first recommended financial products to the target customer respectively, where a is an integer and a is greater than 0, and n is an integer and n is greater than 0; Obtain n first feedback data points from the target customer after receiving the n first recommended financial products; Based on the n first feedback data, the first financial product recommendation model is fine-tuned to generate a target financial product recommendation model; and Input the target customer's a multidimensional data points into the target financial product recommendation model, and output the target recommended financial product.

2. The method according to claim 1, characterized in that, Inputting a multidimensional data points of the target customer into a pre-trained first financial product recommendation model, the model outputs n first recommended financial products, including: Feature extraction is performed on a multidimensional data points of the target customer to generate c feature vectors of the target customer, where c is an integer, c is greater than 0 and c is less than a; Determine the magnitude of change of the c feature vectors of the target customer; and If the change in at least one of the c feature vectors exceeds a first preset threshold, then the c feature vectors of the target customer are input into a pre-trained first financial product recommendation model, which outputs n first recommended financial products.

3. The method according to claim 2, characterized in that, Feature extraction is performed on a multidimensional data points of the target customer to generate c feature vectors for the target customer, including: Data cleaning is performed on a multidimensional data points of the target customer to generate b multidimensional data points after data cleaning, where b is an integer, b is greater than c and b is less than a; Feature encoding is performed on the b multidimensional data after data cleaning to generate b first vectors for the target customer; and Feature filtering is performed on b first vectors of the target customer to generate c feature vectors of the target customer.

4. The method according to claim 3, characterized in that, Feature filtering is performed on b first vectors of the target customer to generate c feature vectors of the target customer, including: Obtain the historical purchase data of the target customer, calculate the correlation values ​​between the b first vectors and the historical purchase data of the target customer, and generate b correlation values; and Obtain c first vectors corresponding to c association values ​​that are greater than a first preset threshold from among the b association values, and use them as c feature vectors of the target customer.

5. The method according to claim 2, characterized in that, The c feature vectors of the target customer are input into a pre-trained first financial product recommendation model, which outputs n first recommended financial products, including: Attention weights are assigned to the c feature vectors of the target customer to generate c feature vectors of the target customer labeled with attention weights; and Input c feature vectors of the target customer labeled with attention weights into the pre-trained first financial product recommendation model, and output n first recommended financial products.

6. The method according to claim 1, characterized in that, The training process generates the first financial product recommendation model, including: Obtain the historical behavior data of the target customer, which includes: i historical multidimensional data points of the target customer, j historical recommended financial products of the target customer, and j historical feedback data of the target customer, wherein i is an integer and i is greater than 0, and j is an integer and j is greater than 0; and Using i historical multidimensional data points of the target customer as the state input of the model, j historical recommended financial products of the target customer as the action output of the model, and j historical feedback data of the target customer as the reward of the action output of the model, the initial financial product recommendation model is trained to generate the first financial product recommendation model.

7. The method according to any one of claims 1 to 6, characterized in that, Based on the n first feedback data, the first financial product recommendation model is fine-tuned to generate a target financial product recommendation model, including: Clustering algorithms are used to perform clustering calculations on the n first feedback data to generate the category characteristics of the target customer; Based on the category characteristics of the target customers, the state parameters of the first financial product recommendation model are optimized to generate a second financial product recommendation model; and The action value function of the second financial product recommendation model is optimized by using a temporal difference learning algorithm to generate the target financial product recommendation model.

8. A financial product recommendation device, characterized in that, The device includes: The first generation module is used to acquire a multidimensional data of the target customer, input the a multidimensional data of the target customer into the pre-trained first financial product recommendation model, output n first recommended financial products, and send the n first recommended financial products to the target customer respectively, where a is an integer and a is greater than 0, and n is an integer and n is greater than 0. The first acquisition module is used to acquire n first feedback data after the target customer receives the n first recommended financial products; The second generation module is used to fine-tune the first financial product recommendation model based on the n first feedback data points, and generate a target financial product recommendation model; and The third generation module is used to input the target customer's a multidimensional data points into the target financial product recommendation model and output the target recommended financial product.

9. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images