Financial product determination method and device, program product and electronic equipment

Abstract

The invention discloses a financial product determination method and device, a program product and electronic equipment, and relates to the field of artificial intelligence, and the determination method comprises the steps: obtaining customer data of a target customer under the condition of obtaining authorization of the target customer, obtaining financial product data, carrying out the embedding processing of the customer data and the financial product data, and obtaining a financial product; the method comprises the following steps: obtaining a customer feature embedded vector and a product feature embedded vector, splicing the customer feature embedded vector and the product feature embedded vector to obtain a combined feature, carrying out noise reduction processing on the combined feature to obtain a noise reduction feature, and processing the combined feature to obtain a preset feature having the same dimension as the noise reduction feature; and processing the noise reduction features and preset features by adopting a preset scoring model to obtain a score of each financial product, and determining the financial product indicated by the maximum score as a target financial product. According to the method and the device, the technical problem of relatively low accuracy of determining the financial product in related technologies is solved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence, and more specifically, to a method, apparatus, program, and electronic device for determining financial products. Background Technology

[0002] With the rapid development of fintech, the number of financial products is increasing. Commonly used technologies for identifying financial products include collaborative filtering, content filtering, and machine learning. However, these methods often overlook the presence of noise in the existing data (e.g., asset class in customer data only represents the customer's assets at the current financial institution, not their total assets). Introducing this noise into the model training process significantly reduces the model's predictive accuracy, thus affecting the effectiveness of identifying financial products.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This invention provides a method, apparatus, program, and electronic device for determining financial products, in order to at least solve the technical problem of low accuracy in determining financial products in related technologies.

[0005] According to one aspect of the present invention, a method for determining a financial product is provided, comprising: acquiring customer data of the target customer and acquiring financial product data, provided that authorization is obtained from the target customer; embedding the customer data and financial product data to obtain a customer feature embedding vector and a product feature embedding vector, and concatenating the customer feature embedding vector and the product feature embedding vector to obtain a combined feature; performing noise reduction processing on the combined feature to obtain a noise-reduced feature, and processing the combined feature to obtain a preset feature with the same dimension as the noise-reduced feature; processing the noise-reduced feature and the preset feature using a preset scoring model to obtain a score for each financial product, and determining the financial product with the highest score as the target financial product, wherein the target financial product is used to be displayed on the target customer's terminal client.

[0006] Furthermore, before acquiring customer data from the target customers, the process includes: constructing an initial scoring model, wherein the model structure of the initial scoring model includes at least: an embedding layer, a noise reduction feature extraction module, a fully connected network, and a score prediction module; collecting historical customer data and financial product data to obtain training data; labeling the training data to obtain labeling information, wherein the labeling information includes at least: the actual score of each financial product; training the initial scoring model using the training data and labeling information based on a preset loss function until the loss value is less than a preset loss threshold to obtain the target model parameters, wherein the loss value is obtained by using the preset loss function based on the actual score and the predicted score output by the initial scoring model; the target model parameters include at least: the first weight matrix and the first bias of the embedding layer, the second weight matrix and the second bias of each layer of the network in the noise reduction feature extraction module, the third weight matrix and the third bias of each layer of the network in the fully connected network, and the fourth weight matrix and the fourth bias of each layer of the network in the score prediction module; and adjusting the initial scoring model based on the target model parameters to obtain the preset scoring model.

[0007] Furthermore, the customer data includes multiple customer sub-data, and the financial product data includes multiple product sub-data. The steps of embedding the customer data and financial product data to obtain customer feature embedding vectors and product feature embedding vectors include: based on a first weight matrix and a first bias, performing feature mapping on each customer sub-data using an embedding layer to obtain multiple first embedding vectors; based on the first weight matrix and the first bias, performing feature mapping on each product sub-data using an embedding layer to obtain multiple second embedding vectors; concatenating all the first embedding vectors to obtain the customer feature embedding vector; and concatenating all the second embedding vectors to obtain the product feature embedding vector.

[0008] Furthermore, the preset scoring model also includes a feature fusion structure, which concatenates the customer feature embedding vector and the product feature embedding vector to obtain combined features. This includes: using the flattening layer in the feature fusion structure to flatten the customer feature embedding vector and the product feature embedding vector to obtain a first flattened vector and a second flattened vector; and using the combining layer in the feature fusion structure to combine the first flattened vector and the second flattened vector to obtain combined features.

[0009] Furthermore, the denoising feature extraction module includes an encoder, a decoder, and a residual structure. The step of denoising the combined features to obtain denoised features includes: processing the combined features using an activation function based on the second weight matrix and second bias corresponding to the first layer of the encoder to obtain initial features; processing the initial features using an activation function based on the second weight matrix and second bias corresponding to the next layer of the encoder until the last layer of the encoder outputs features, and representing these features as the first feature; processing the first feature using the decoder to obtain the second feature; and processing the second feature using the residual structure to obtain the denoised features.

[0010] Furthermore, the steps of processing the denoised features and preset features using a preset scoring model to obtain the score for each financial product include: fusing the denoised features and preset features using a score prediction module to obtain fused features; and processing the fused features based on the fourth weight matrix and fourth bias corresponding to the score prediction module to obtain the score for each financial product.

[0011] Furthermore, the step of fusing the denoised features and preset features using the score prediction module to obtain the fused features includes: flattening the denoised features to obtain a third flattened vector, and flattening the preset features to obtain a fourth flattened vector; and combining the third flattened vector and the fourth flattened vector to obtain the fused features.

[0012] According to another aspect of the present invention, a financial product determination apparatus is also provided, comprising: an acquisition unit, configured to acquire customer data of a target customer and financial product data upon obtaining authorization from a target customer; a first processing unit, configured to perform embedding processing on the customer data and financial product data to obtain a customer feature embedding vector and a product feature embedding vector, and to concatenate the customer feature embedding vector and the product feature embedding vector to obtain a combined feature; a second processing unit, configured to perform noise reduction processing on the combined feature to obtain a noise-reduced feature, and to process the combined feature to obtain a preset feature with the same dimension as the noise-reduced feature; and a third processing unit, configured to process the noise-reduced feature and the preset feature using a preset scoring model to obtain a score for each financial product, and to determine the financial product with the highest score as the target financial product, wherein the target financial product is used to be displayed on the target customer's terminal client.

[0013] Furthermore, the determining device also includes: a first construction module, used to construct an initial scoring model before acquiring customer data of the target customer, wherein the model structure of the initial scoring model includes at least: an embedding layer, a noise reduction feature extraction module, a fully connected network, and a score prediction module; a first acquisition module, used to acquire historical customer data and financial product data to obtain training data; a first annotation module, used to annotate the training data to obtain annotation information, wherein the annotation information includes at least: the actual score of each financial product; a first training module, used to train the initial scoring model based on a preset loss function, using the training data and annotation information, until the loss value is less than a preset loss threshold, to obtain target model parameters, wherein the loss value is obtained by using the preset loss function based on the actual score and the predicted score output by the initial scoring model; the target model parameters include at least: a first weight matrix and a first bias of the embedding layer, a second weight matrix and a second bias of each layer of the network in the noise reduction feature extraction module, a third weight matrix and a third bias of each layer of the network in the fully connected network, and a fourth weight matrix and a fourth bias of each layer of the network in the score prediction module; and a first adjustment module, used to adjust the initial scoring model based on the target model parameters to obtain a preset scoring model.

[0014] Furthermore, the customer data includes multiple customer sub-data, and the financial product data includes multiple product sub-data. The first processing unit includes: a first mapping module, used to perform feature mapping on each customer sub-data using an embedding layer based on a first weight matrix and a first bias, to obtain multiple first embedding vectors; a second mapping module, used to perform feature mapping on each product sub-data using an embedding layer based on the first weight matrix and the first bias, to obtain multiple second embedding vectors; a first concatenation module, used to concatenate all the first embedding vectors to obtain a customer feature embedding vector; and a second concatenation module, used to concatenate all the second embedding vectors to obtain a product feature embedding vector.

[0015] Furthermore, the preset scoring model also includes a feature fusion structure, and the first processing unit further includes: a first flattening module, used to flatten the customer feature embedding vector and the product feature embedding vector using the flattening layer in the feature fusion structure to obtain a first flattened vector and a second flattened vector; and a first combining module, used to combine the first flattened vector and the second flattened vector using the combining layer in the feature fusion structure to obtain a combined feature.

[0016] Furthermore, the noise reduction feature extraction module includes an encoder, a decoder, and a residual structure. The second processing unit includes: a first processing module, used to process the combined features using an activation function based on the second weight matrix and second bias corresponding to the first layer of the encoder to obtain initial features; a second processing module, used to process the initial features using an activation function based on the second weight matrix and second bias corresponding to the next layer of the encoder, until the last layer of the encoder outputs features, and represents these features as the first feature; a third processing module, used to process the first feature using a decoder to obtain the second feature; and a fourth processing module, used to process the second feature using a residual structure to obtain the noise reduction features.

[0017] Furthermore, the third processing unit includes: a first fusion module, used to fuse the denoised features and preset features using the score prediction module to obtain fused features; and a fifth processing module, used to process the fused features based on the fourth weight matrix and fourth bias corresponding to the score prediction module to obtain the score of each financial product.

[0018] Furthermore, the first fusion module includes: a first flattening submodule, used to flatten the noise reduction features to obtain a third flattening vector, and to flatten the preset features to obtain a fourth flattening vector; and a first combining submodule, used to combine the third flattening vector and the fourth flattening vector to obtain fused features.

[0019] According to another aspect of the present invention, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the method for determining any of the above-mentioned financial products.

[0020] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method for determining any of the above-described financial products.

[0021] In this invention, with the authorization of the target customer, customer data and financial product data of the target customer are obtained. The customer data and financial product data are embedded to obtain customer feature embedding vectors and product feature embedding vectors. The customer feature embedding vectors and product feature embedding vectors are concatenated to obtain combined features. The combined features are denoised to obtain denoised features. The combined features are further processed to obtain preset features with the same dimension as the denoised features. A preset scoring model is used to process the denoised features and preset features to obtain a score for each financial product. The financial product with the highest score is identified as the target financial product, thereby solving the technical problem of low accuracy in identifying financial products in related technologies.

[0022] In this invention, while ensuring compliance with user privacy protection regulations, customer data (such as asset level) and financial product data (such as product type and risk level) of target customers are obtained. These data are then processed using an embedding layer to transform them into customer feature embedding vectors and product feature embedding vectors. A noise reduction feature extraction module then performs noise reduction processing on the concatenated combined features to obtain cleaner noise-reduced features. Simultaneously, the combined features are further processed to obtain preset features of the same dimension as the noise-reduced features. Finally, a scoring model integrates the noise-reduced features and the preset features to accurately predict and select the financial product with the highest score. This financial product is then presented to the target customer as the target financial product, achieving the goal of improving the accuracy and personalization of financial product identification, thereby enhancing customer satisfaction. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0024] Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing a method for determining financial products is shown.

[0025] Figure 2 This is a flowchart of a method for determining financial products according to Embodiment 1 of the present invention;

[0026] Figure 3 This is a schematic diagram of the scoring model structure based on the autoencoder for noise reduction feature extraction according to Embodiment 1 of the present invention;

[0027] Figure 4 This is a schematic diagram of an optional financial product determination device according to an embodiment of the present invention;

[0028] Figure 5This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0031] It should be noted that all relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, and displayed data) collected and involved in this invention are 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 this data comply with the relevant laws, regulations, and standards of the relevant regions, necessary confidentiality measures have been taken, and it does not violate public order and good morals. Corresponding operation entry points are provided for users to choose to authorize or refuse. For example, this system has an interface with relevant users or organizations. Before obtaining relevant information, a request to obtain the information needs to be sent to the aforementioned user or organization through the interface. The relevant information is obtained only after receiving consent from the aforementioned user or organization. If the user chooses to refuse, the process proceeds to an expert decision-making process.

[0032] This invention proposes a method for determining financial products based on a noise-reducing feature extraction autoencoder. This method processes potentially noisy data in customer and financial product information, significantly improving the predictive accuracy of financial product identification, thereby providing customers with more precise and personalized financial products.

[0033] In this invention, a noise reduction feature extraction module extracts noise reduction features from the original features, effectively reducing the impact of noise on the model and improving prediction accuracy. Furthermore, considering that the noise reduction feature extraction module is implemented through a complex neural network, the original features and the noise reduction features can be concatenated and fused before being fed into the prediction network for prediction, mitigating the potential impact of overfitting on the prediction results.

[0034] The present invention will now be described in detail with reference to various embodiments.

[0035] Example 1

[0036] According to an embodiment of this application, an embodiment of a method for determining financial products is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0037] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a method for determining financial products is shown. Figure 1 As shown, computer terminal 10 (or mobile device) may include one or more ( Figure 1 The processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions may also be included. In addition, it may include: a display, a keyboard, a cursor control device, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera, wherein the network interface can be connected to wired and / or wireless networks. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0038] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0039] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the financial product determination method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned financial product determination method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0040] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0041] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0042] Under the aforementioned operating environment, this application provides the following: Figure 2 The method for determining the financial products shown. Figure 2 This is a flowchart of a method for determining financial products according to Embodiment 1 of the present invention, as shown below. Figure 2 As shown, the method includes the following steps:

[0043] Step S201: With the authorization of the target customer, obtain the target customer's customer data and financial product data.

[0044] In this embodiment of the invention, after obtaining explicit authorization from the target customer regarding the use of their personal data, customer information, such as customer ID, geographical location, asset level, and risk preference, is collected. This information constitutes customer data. Simultaneously, detailed information on all financial products offered by financial institutions is collected, including product ID, product category, risk level, minimum investment amount, and historical annual interest rate. This information constitutes financial product data.

[0045] Step S202: Embedding processing is performed on customer data and financial product data to obtain customer feature embedding vectors and product feature embedding vectors. The customer feature embedding vectors and product feature embedding vectors are then concatenated to obtain combined features.

[0046] In this embodiment of the invention, an embedding layer is used to convert discrete features in customer data and financial product data into low-dimensional, dense, continuous vector representations to better capture the potential relationships between features. For example, for the discrete feature of customer ID, the embedding layer maps the ID to a numerical vector through a fully connected layer. This vector represents the complex information behind the customer ID. The same processing is applied to features such as product ID. Then, all the embedded customer feature vectors are concatenated to form a customer feature embedding vector that integrates customer attributes, and all the embedded product feature vectors are concatenated to form a product feature embedding vector that integrates product attributes. Finally, the customer feature embedding vector and the product feature embedding vector are concatenated to obtain the combined feature.

[0047] Step S203: Perform noise reduction processing on the combined features to obtain noise-reduced features, and process the combined features to obtain preset features with the same dimension as the noise-reduced features.

[0048] In this embodiment of the invention, an autoencoder can be used as the noise reduction feature extraction module. This module consists of an encoder, a decoder, and multiple residual blocks. These residual blocks can use fully connected networks instead of CNNs (Convolutional Neural Networks) because the data type is mainly numerical rather than image-based. The neural network structure of the encoder and decoder helps extract and reconstruct features, removing unnecessary noise to obtain noise-reduced features. Furthermore, through another different processing procedure, the combined features are processed to generate a set of preset features with the same dimensionality as the noise-reduced features.

[0049] Step S204: The noise reduction features and preset features are processed using a preset scoring model to obtain a score for each financial product, and the financial product with the highest score is identified as the target financial product. The target financial product is used to be displayed on the target customer's terminal client.

[0050] In this embodiment of the invention, a pre-trained scoring model is used to process denoising features and preset features. Through multi-layer processing (such as activation functions and fully connected layers), a predicted score for each financial product is finally output. Then, based on these scores, the financial product with the highest score is automatically identified and determined as the target financial product, so that the target financial product can be displayed on the target customer's terminal client (such as the application interface of a mobile phone).

[0051] In summary, while ensuring compliance with user privacy protection regulations, this approach acquires customer data (such as asset level) and financial product data (such as product type and risk level) from target customers. This data is then processed using an embedding layer to transform it into customer feature embedding vectors and product feature embedding vectors. A noise reduction feature extraction module then denoises the concatenated features to obtain cleaner, denoised features. Further processing of the combined features yields preset features of the same dimension as the denoised features. Finally, a scoring model integrates the denoised features and the preset features to accurately predict and select the highest-scoring financial product. This product is then presented to target customers, improving the accuracy and personalization of financial product selection and ultimately enhancing customer satisfaction.

[0052] To improve the prediction accuracy of the scoring model, in the method for determining financial products provided in Embodiment 1 of this application, before obtaining the customer data of the target customer, an initial scoring model is constructed. The model structure of the initial scoring model includes at least: an embedding layer, a noise reduction feature extraction module, a fully connected network, and a score prediction module. Historical customer data and financial product data are collected to obtain training data. The training data is labeled to obtain labeling information, which includes at least: the actual score of each financial product. Based on a preset loss function, the initial scoring model is trained using the training data and labeling information until the loss value is less than a preset loss threshold to obtain the target model parameters. The loss value is obtained by using the preset loss function based on the actual score and the predicted score output by the initial scoring model. The target model parameters include at least: the first weight matrix and the first bias of the embedding layer, the second weight matrix and the second bias of each layer of the noise reduction feature extraction module, the third weight matrix and the third bias of each layer of the fully connected network, and the fourth weight matrix and the fourth bias of each layer of the score prediction module. Based on the target model parameters, the initial scoring model is adjusted to obtain the preset scoring model.

[0053] In this embodiment of the invention, an initial scoring model can be constructed first. This model includes an embedding layer, a denoising feature extraction module, a fully connected network, and a score prediction module. The embedding layer maps discrete features in customer data and financial product data into dense continuous vector representations to improve the model's data processing capabilities and prediction performance. The denoising feature extraction module uses an autoencoder principle to denoise the embedded combined features, removing noise from the data and improving the purity and effectiveness of the features. The fully connected network, as an intermediate processing layer, fuses the denoised features with other pre-processed features, ensuring that the model learns both the denoised, clean features and information from the original features. The score prediction module, located at the end of the model, is the output part and is used to predict the customer satisfaction score for each financial product based on the input feature set.

[0054] In this embodiment of the invention, historical customer behavior records, basic customer information, and historical data on various financial products offered by financial institutions can be extracted from a database. This data includes not only successful transaction cases but also information on unselected products, enabling the model to comprehensively learn the correlation between customer preferences and product attributes. By preprocessing the data, all non-numerical features are converted into numerical values, ensuring data consistency and integrity, thereby forming a training dataset that can be used for model training.

[0055] Next, the training data is labeled to obtain annotation information. This annotation information provides guidance for model training. The annotation information includes: the true satisfaction rating for each financial product, which can be obtained directly from historical data or indirectly estimated through methods such as surveys. The true rating is the target of model training; the model will attempt to minimize the difference between its predicted rating and these true ratings.

[0056] Next, the initial scoring model can be trained using training data and labeled information based on a preset loss function. During model training, the preset loss function quantifies the difference between the model's predicted output and the true label. For example, Mean Absolute Error (MAE) can be used as the model's loss function, and the accuracy of score prediction can be measured using MAE and Root Mean Square Error (RMSE). The formulas for MAE and RMSE are shown below:

[0057] ;

[0058] ;

[0059] in, To score the actual products, The product score predicted by the model, Indicates a sample, This represents the total number of samples in the dataset that need to be predicted.

[0060] In this embodiment of the invention, model training continuously adjusts model parameters through a backpropagation algorithm, including but not limited to: the first weight matrix and first bias of the embedding layer (corresponding to the embedding of customer and product features), the second weight matrix and second bias of each layer in the denoising feature extraction module (responsible for feature denoising), the third weight matrix and third bias of each layer in the fully connected network (used for feature extraction), and the fourth weight matrix and fourth bias of each layer in the score prediction module (determining the final prediction output). The entire training process continues until the loss value (i.e., prediction error) converges to below a preset loss threshold, at which point the set of model parameters becomes the target model parameters.

[0061] Once training is complete and the target model parameters are obtained, these parameters can be used to replace the corresponding parts of the initial scoring model, thus obtaining the preset scoring model. The preset scoring model has been sufficiently trained and can effectively extract denoised, clean signals from customer and product features for accurate satisfaction prediction. This model design based on a denoised feature extraction autoencoder helps maintain prediction accuracy even in noisy datasets, thereby providing customers with more precise and personalized financial product selection.

[0062] In this embodiment, the negative impact of data noise on the accuracy of financial product determination is effectively resolved, greatly improving the accuracy of financial product determination and customer experience.

[0063] Optionally, the customer data includes multiple customer sub-data, and the financial product data includes multiple product sub-data. To improve the accuracy of determining the customer feature embedding vector and the product feature embedding vector, in the method for determining financial products provided in Embodiment 1 of this application, based on the first weight matrix and the first bias, an embedding layer is used to perform feature mapping on each customer sub-data to obtain multiple first embedding vectors; based on the first weight matrix and the first bias, an embedding layer is used to perform feature mapping on each product sub-data to obtain multiple second embedding vectors; all first embedding vectors are concatenated to obtain the customer feature embedding vector; and all second embedding vectors are concatenated to obtain the product feature embedding vector.

[0064] In this embodiment of the invention, to enhance the model's ability to mine and learn information from data, customer features and product features can be respectively input into the embedding layer for embedding operations. Here, the embedding layer is an unbiased fully connected layer used to map discrete features into low-dimensional dense continuous vector representations.

[0065] Specifically, for each customer sub-data (i.e., each customer feature, such as the customer ID feature) included in the customer data, an embedding layer can be used to perform feature mapping on each customer sub-data to obtain multiple first embedding vectors. Taking the customer ID feature as an example, the formula is as follows:

[0066] ;

[0067] in, The embedding vector (i.e., the first embedding vector) represents the customer ID feature. This represents the corresponding original input vector (i.e., customer ID feature). This represents the activation function in the embedding layer. The embedding weight matrix (i.e., the first weight matrix) used in the representation process. This is the bias term (i.e., the first bias).

[0068] For each product sub-data (i.e. each product feature, such as product number, product type, risk level, etc.) included in the financial product data, multiple second embedding vectors can be obtained after processing by the embedding layer.

[0069] Then, by concatenating the embedding vectors of each customer feature and the embedding vectors of each product feature using the concatenation operation, the customer feature embedding vector is obtained. The product feature embedding vector P is then concatenated with all the first embedding vectors (i.e., horizontally connected) to form a complete customer feature embedding vector S. This vector contains the embedded representations of all customer attributes, providing a unified customer view for subsequent noise reduction and score prediction. Finally, all the second embedding vectors are concatenated to obtain the product feature embedding vector P. This product feature embedding vector condenses the key attributes of all financial products, laying the foundation for the model to learn product characteristics.

[0070] In this embodiment, by using feature mapping and concatenation in the embedding layer, the original customer and product data are transformed into numerical embedding vectors, improving the model's data processing capabilities and efficiency. The customer feature embedding vectors and product feature embedding vectors not only retain the rich information of the original data but also enhance the model's performance in subsequent noise reduction and prediction stages through dimensionality reduction and numericalization. This effectively solves the problems of data noise and feature sparsity, ensuring that the model can make accurate and personalized predictions in complex financial environments.

[0071] Optionally, the preset scoring model also includes a feature fusion structure. In order to improve the accuracy of determining the combined features, in the method for determining financial products provided in Embodiment 1 of this application, a flattening layer in the feature fusion structure is used to flatten the customer feature embedding vector and the product feature embedding vector to obtain a first flattened vector and a second flattened vector; a combining layer in the feature fusion structure is used to combine the first flattened vector and the second flattened vector to obtain the combined features.

[0072] In this embodiment of the invention, a flattening layer in the feature fusion structure of a preset scoring model can be used to convert the multidimensional embedding vectors (customer feature embedding vectors and product feature embedding vectors) into one-dimensional vectors to ensure smooth processing by subsequent neural network layers. Specifically, the flattening layer is applied to both the customer feature embedding vector S and the product feature embedding vector P. The flattening operation does not change the information in the vectors, only adjusts their format. Then, a combining layer in the feature fusion structure is used to combine the first flattened vector and the second flattened vector to obtain the combined feature B.

[0073] The formula for concatenating the customer feature embedding vector S and the product feature embedding vector P to obtain the combined feature B is as follows:

[0074] ;

[0075] in, and These represent the combination operation and the flattening operation, respectively. This indicates a dimension of 1.

[0076] In this embodiment, by employing a flattening layer and a combining layer in the feature fusion structure, effective integration of customer and product features is achieved, providing a more general and unified feature input for subsequent deep learning of the model. This not only simplifies the data processing flow but also enhances the model's ability to capture complex relationships, contributing to improved accuracy and personalization in financial product identification.

[0077] Optionally, the noise reduction feature extraction module includes an encoder, a decoder, and a residual structure. To improve the accuracy of determining the noise reduction features, in the financial product determination method provided in Embodiment 1 of this application, based on the second weight matrix and second bias corresponding to the first layer network in the encoder, an activation function is used to process the combined features to obtain initial features; based on the second weight matrix and second bias corresponding to the next layer network in the encoder, an activation function is used to process the initial features until the last layer network in the encoder outputs features, and these features are represented as first features; the decoder is used to process the first features to obtain second features; and the residual structure is used to process the second features to obtain noise reduction features.

[0078] In this embodiment of the invention, the combined feature B is fed into the denoising feature extraction module to extract the denoising features. Here, the denoising feature extraction module consists of an encoder, a decoder, and a prediction module (residual structure) composed of multiple residual blocks. The prediction result obtained by the prediction module does not represent the final prediction result of the model, but is only used to obtain denoising features that can more accurately predict the product score through backpropagation. Both the encoder and decoder are composed of multilayer perceptrons. The implementation of the encoder is shown below:

[0079] ;

[0080] ;

[0081] ;

[0082] in, Represents the activation function, in practice, It can be set to the ReLU (Rectified Linear Unit) function to mitigate the effects of problems such as vanishing gradients. Indicates the first The weight matrix (i.e., the second weight matrix) of the layer network. Indicates the first The bias value of the layer network (i.e., the second bias). Indicates the first The output of the layer network, This represents the total number of layers in the network. This represents the characteristics of the encoder output. Since the decoder structure is completely symmetrical to the encoder, it will not be elaborated upon further here.

[0083] In the prediction module, the residual blocks are composed of CNNs. However, since the data used in the determination of financial products are all simple numbers, rather than images or spatial data, a fully connected network can be used to construct the residual blocks and connect multiple residual blocks to complete the prediction of the product score.

[0084] Specifically, the combined features can be first input into the first layer of the encoder. This layer processes the combined features using an activation function based on its second weight matrix and second bias to obtain initial features. Then, these initial features are fed as input into the next layer of the encoder, which also has its corresponding second weight matrix and second bias. This process is repeated until every layer of the encoder has been traversed. The output of each layer is used as the input to the next layer, until the final layer of the encoder outputs its features. The output of the final layer of the encoder, i.e., the features processed by multiple layers of networks, is represented as the first feature.

[0085] The decoder is a neural network symmetrical to the encoder structure, taking the first feature as input. Similar to the encoder, each layer in the decoder is computed using a specific second weight matrix and a second bias, decoding the first feature layer by layer until it is restored to a shape close to the original combined features. The final output of this process is called the second feature.

[0086] In the residual structure, the encoder's input (i.e., combined features) and the decoder's output (i.e., second features) are compared. By learning the difference between the input and output, denoising features can be obtained. This operation is usually performed in the residual block, which consists of two fully connected layers. The input is the decoder's output, and the output is the processed result of the difference between the input and the decoder's output, i.e., the denoising feature.

[0087] In this embodiment, the noise reduction feature extraction method based on encoder, decoder, and residual structure can effectively remove noisy data from customer and product features, improving the prediction accuracy of the model. The encoder converts the combined features into a first feature, and the decoder then decodes it into a second feature that is close to the original state. The residual structure learns and extracts the difference between the two, resulting in a cleaner noise reduction feature that is less affected by noise.

[0088] To improve the accuracy of determining the score of each financial product, in the method for determining financial products provided in Embodiment 1 of this application, a score prediction module is used to fuse the noise reduction features and preset features to obtain fused features; based on the fourth weight matrix and fourth bias corresponding to the score prediction module, the fused features are processed to obtain the score of each financial product.

[0089] In this embodiment of the invention, considering that the denoising feature extraction module is composed of a complex neural network and may suffer from overfitting, the features before denoising and the extracted denoised features can be concatenated and fed into the subsequent network for final prediction. Then, the prediction module will adjust the learning weight parameters of the two features in the concatenated features to balance the relationship between denoising and overfitting, thereby obtaining a model with higher accuracy and stronger generalization ability.

[0090] Specifically, the denoised feature B' is fused with the preset feature E' to obtain the fused feature. Here, the preset feature refers to those features that were not denoised during model training and were obtained directly through other methods (such as fully connected network processing).

[0091] Then, based on the fourth weight matrix and fourth bias corresponding to the score prediction module, the fused features are processed to obtain the score for each financial product. Here, the score prediction module typically consists of one or more fully connected layers, each with its corresponding fourth weight matrix and fourth bias. The fused features are taken as input, multiplied by the fourth weight matrix, and the fourth bias is added. Then, they are processed by an activation function (usually a linear activation function to output continuous score values) to obtain the score prediction result for each financial product.

[0092] In this embodiment, by fusing denoised features with preset features, and then processing based on the fourth weight matrix and fourth bias corresponding to the score prediction module, a more accurate financial product score prediction can be obtained. Thus, not only does the denoised feature extraction autoencoder remove noise from the original data, maintaining the model's sensitivity to effective information, but the fusion of preset features also introduces additional external information, enhancing the model's decision-making ability. Ultimately, the score prediction results for each product will be closer to the actual satisfaction level, helping financial institutions more accurately identify financial products that match customer preferences, thereby improving customer satisfaction and the success rate of product selection.

[0093] To improve the accuracy of determining the fusion features, in the method for determining financial products provided in Embodiment 1 of this application, the noise reduction features are flattened to obtain a third flattened vector, and the preset features are flattened to obtain a fourth flattened vector; the third flattened vector and the fourth flattened vector are combined to obtain the fusion features.

[0094] In this embodiment of the invention, a flattening operation is performed on the denoising feature B', transforming it from a multi-dimensional structure into a one-dimensional vector, i.e., the third flattened vector. The flattening operation ensures that the feature can be correctly processed by the fully connected layer, preparing it for subsequent score prediction. Similarly, a flattening operation is performed on the preset feature E', resulting in a fourth flattened vector. The preset feature may include additional information that influences the scoring of financial products, such as market trends and historical performance, and also needs to be converted into a one-dimensional form for fusion.

[0095] Then, a combination operation (such as a concatenation operation) is used to fuse the third and fourth flattened vectors, forming a fused feature X that includes denoising features and preset feature information. The concatenation process is shown below:

[0096] ;

[0097] in, Representative combined operation, This represents the flattening operation.

[0098] In this embodiment, denoising features and preset features are effectively combined, providing a richer and cleaner input for the score prediction module. By flattening the denoising features and combining them with preset features, the model can simultaneously consider denoising information of customer and product characteristics, as well as external market and economic conditions that may affect product attractiveness, in score prediction. This improves the accuracy and reliability of score prediction.

[0099] Figure 3 This is a schematic diagram of the scoring model structure based on a noise reduction feature extraction autoencoder according to Embodiment 1 of the present invention, as shown below. Figure 3 As shown, the model includes: an input / output layer, a data embedding layer, a denoising feature extraction and feature fusion structure, and a score prediction structure. The denoising feature extraction and feature fusion structure comprises a denoising feature extraction module and a fully connected network. The denoising feature extraction module consists of an encoder, a decoder, and a prediction module composed of multiple residual blocks. The specific processing steps of the scoring model are as follows:

[0100] (1) Data input. The input data is divided into two dimensions: customer data and product data, and customer features and product features are constructed respectively. Among them, customer features include customer ID (number), asset level, risk preference and other information; product features include product ID, product type, risk level, minimum purchase amount, historical annual interest rate and other information.

[0101] (2) Data Embedding. To enhance the model's ability to mine and learn information from data, customer features and product features are respectively input into the embedding layer for embedding operations. For example, embedding the customer ID yields the embedded features. Embedding operations are performed on asset grades to obtain embedded features. Embedding operation is performed on risk preference to obtain embedded feature P; embedding operation is performed on product ID to obtain embedded feature P. Embedding operation is performed on product type to obtain embedded feature T; embedding operation is performed on risk level to obtain embedded feature. Embedding operation is performed on the minimum purchase amount to obtain the embedding feature M; embedding operation is performed on the historical annual interest rate to obtain the embedding feature. Then, the embedding vectors of each customer feature and the embedding vectors of each product feature are concatenated using a combination operation to obtain the customer feature embedding vector. And the product feature embedding vector P.

[0102] (3) The customer feature embedding vector and the product feature embedding vector are concatenated to obtain the combined feature.

[0103] (4) The combined features are fed into the noise reduction feature extraction module to extract the noise reduction features. Here, the score obtained by the prediction module does not represent the final prediction result of the model, but is only used to obtain noise reduction features that can more accurately predict the product score through backpropagation.

[0104] (5) The combined features are fed into a fully connected network to obtain the features extracted by the noise reduction feature extraction module. Features with the same dimensions .

[0105] (6) Using fusion operation C, the noise reduction features are processed. and characteristics The fusion process is performed to obtain the fusion characteristics.

[0106] (7) The fusion features are processed using a score prediction structure to obtain the final score.

[0107] In this embodiment, the impact of noisy data in customer and product characteristics on the accuracy of model product score prediction is effectively mitigated when determining financial products. This allows for the provision of more accurate and personalized financial products to customers, thereby improving customer satisfaction.

[0108] The method for determining financial products provided in this application, while ensuring compliance with user privacy protection regulations, acquires customer data (such as asset level) and financial product data (such as product type and risk level) of the target customer. This data is then processed using an embedding layer to transform it into customer feature embedding vectors and product feature embedding vectors. A noise reduction feature extraction module then performs noise reduction processing on the concatenated combined features to obtain cleaner noise-reduced features. Simultaneously, the combined features are further processed to obtain preset features of the same dimension as the noise-reduced features. Finally, a scoring model integrates the noise-reduced features and the preset features to accurately predict and select the financial product with the highest score. This financial product is then presented to the target customer as the target financial product, achieving the goal of improving the accuracy and personalization of financial product determination, thereby enhancing customer satisfaction.

[0109] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0110] Example 2

[0111] This application also provides a device for determining financial products. It should be noted that this device can be used to execute the method for determining financial products provided in this application. The following describes the device for determining financial products provided in this application.

[0112] According to an embodiment of this application, an apparatus for implementing the above-described method for determining financial products is also provided. Figure 4 This is a schematic diagram of an optional financial product determination device according to an embodiment of the present invention, such as... Figure 4 As shown, the determining device may include: an acquisition unit 40, a first processing unit 41, a second processing unit 42, and a third processing unit 43.

[0113] The acquisition unit 40 is used to acquire customer data of the target customer and financial product data when the target customer authorizes the acquisition.

[0114] The first processing unit 41 is used to perform embedding processing on customer data and financial product data to obtain customer feature embedding vectors and product feature embedding vectors, and to concatenate the customer feature embedding vectors and product feature embedding vectors to obtain combined features.

[0115] The second processing unit 42 is used to perform noise reduction processing on the combined features to obtain noise-reduced features, and to process the combined features to obtain preset features with the same dimension as the noise-reduced features.

[0116] The third processing unit 43 is used to process the noise reduction features and preset features using a preset scoring model to obtain a score for each financial product, and to determine the financial product with the highest score as the target financial product, wherein the target financial product is used to be displayed on the target customer's terminal client.

[0117] The financial product determination device provided in this application, while ensuring compliance with user privacy protection regulations, acquires customer data (such as asset level) and financial product data (such as product type and risk level) of the target customer. It then processes this data using an embedding layer, converting it into customer feature embedding vectors and product feature embedding vectors. Next, a noise reduction feature extraction module performs noise reduction processing on the concatenated combined features to obtain cleaner noise-reduced features. Simultaneously, the combined features are further processed to obtain preset features of the same dimension as the noise-reduced features. Finally, a scoring model integrates the noise-reduced features and the preset features to accurately predict and select the financial product with the highest score. This financial product is then presented to the target customer as the target financial product, achieving the goal of improving the accuracy and personalization of financial product determination, thereby enhancing customer satisfaction.

[0118] Optionally, the determining device further includes: a first construction module, used to construct an initial scoring model before acquiring customer data of the target customer, wherein the model structure of the initial scoring model includes at least: an embedding layer, a noise reduction feature extraction module, a fully connected network, and a score prediction module; a first acquisition module, used to acquire historical customer data and financial product data to obtain training data; a first annotation module, used to annotate the training data to obtain annotation information, wherein the annotation information includes at least: the actual score of each financial product; a first training module, used to train the initial scoring model based on a preset loss function, using the training data and annotation information, until the loss value is less than a preset loss threshold, to obtain target model parameters, wherein the loss value is obtained by using the preset loss function based on the actual score and the predicted score output by the initial scoring model; the target model parameters include at least: a first weight matrix and a first bias of the embedding layer, a second weight matrix and a second bias of each layer of the network in the noise reduction feature extraction module, a third weight matrix and a third bias of each layer of the network in the fully connected network, and a fourth weight matrix and a fourth bias of each layer of the network in the score prediction module; and a first adjustment module, used to adjust the initial scoring model based on the target model parameters to obtain a preset scoring model.

[0119] Optionally, the customer data includes multiple customer sub-data, and the financial product data includes multiple product sub-data. The first processing unit includes: a first mapping module, used to perform feature mapping on each customer sub-data using an embedding layer based on a first weight matrix and a first bias, to obtain multiple first embedding vectors; a second mapping module, used to perform feature mapping on each product sub-data using an embedding layer based on the first weight matrix and the first bias, to obtain multiple second embedding vectors; a first concatenation module, used to concatenate all the first embedding vectors to obtain a customer feature embedding vector; and a second concatenation module, used to concatenate all the second embedding vectors to obtain a product feature embedding vector.

[0120] Optionally, the preset scoring model also includes a feature fusion structure, and the first processing unit further includes: a first flattening module, used to flatten the customer feature embedding vector and the product feature embedding vector using the flattening layer in the feature fusion structure to obtain a first flattened vector and a second flattened vector; and a first combining module, used to combine the first flattened vector and the second flattened vector using the combining layer in the feature fusion structure to obtain a combined feature.

[0121] Optionally, the noise reduction feature extraction module includes an encoder, a decoder, and a residual structure. The second processing unit includes: a first processing module, used to process the combined features using an activation function based on the second weight matrix and the second bias corresponding to the first layer of the encoder to obtain initial features; a second processing module, used to process the initial features using an activation function based on the second weight matrix and the second bias corresponding to the next layer of the encoder, until the last layer of the encoder outputs features, and represents these features as the first feature; a third processing module, used to process the first feature using a decoder to obtain the second feature; and a fourth processing module, used to process the second feature using a residual structure to obtain the noise reduction features.

[0122] Optionally, the third processing unit includes: a first fusion module, used to fuse the noise reduction features and preset features using the score prediction module to obtain fused features; and a fifth processing module, used to process the fused features based on the fourth weight matrix and fourth bias corresponding to the score prediction module to obtain the score of each financial product.

[0123] Optionally, the first fusion module includes: a first flattening submodule, used to flatten the noise reduction features to obtain a third flattening vector, and to flatten the preset features to obtain a fourth flattening vector; and a first combining submodule, used to combine the third flattening vector and the fourth flattening vector to obtain fused features.

[0124] The aforementioned determining device may further include a processor and a memory. The aforementioned acquisition unit 40, first processing unit 41, second processing unit 42, third processing unit 43, etc., are all stored in the memory as program units, and the processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0125] The aforementioned processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured. By adjusting kernel parameters, a preset scoring model is used to process the noise reduction features and preset features to obtain a score for each financial product. The financial product with the highest score is then identified as the target financial product.

[0126] The aforementioned memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0127] It should be noted that the acquisition unit 40, the first processing unit 41, the second processing unit 42, and the third processing unit 43 mentioned above correspond to steps S201 to S204 in Embodiment 1. The instances and application scenarios implemented by the above units and the corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above units can be hardware or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above units can also be part of a device and run in the computer terminal 10 provided in Embodiment 1.

[0128] Example 3

[0129] Embodiments of this application may provide an electronic device. Figure 5 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Figure 5 As shown, the electronic device may include: one or more ( Figure 5 (Only one is shown) processor 502, memory 504, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0130] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the financial product determination method and apparatus in this application embodiment. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the aforementioned financial product determination method. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0131] The processor can access information and applications stored in the memory via a transmission device to perform the following steps: With the authorization of the target customer, acquire the target customer's customer data and financial product data; embed the customer data and financial product data to obtain customer feature embedding vectors and product feature embedding vectors, and concatenate the customer feature embedding vectors and product feature embedding vectors to obtain combined features; perform noise reduction processing on the combined features to obtain noise-reduced features, and process the combined features to obtain preset features with the same dimension as the noise-reduced features; process the noise-reduced features and preset features using a preset scoring model to obtain a score for each financial product, and determine the financial product with the highest score as the target financial product, wherein the target financial product is used for display on the target customer's terminal client.

[0132] The processor can access information and applications stored in memory via a transmission device to execute the following steps: Constructing an initial scoring model, wherein the initial scoring model structure includes at least: an embedding layer, a noise reduction feature extraction module, a fully connected network, and a score prediction module; collecting historical customer data and financial product data to obtain training data; labeling the training data to obtain labeling information, wherein the labeling information includes at least: the actual score of each financial product; training the initial scoring model using the training data and labeling information based on a preset loss function until the loss value is less than a preset loss threshold, obtaining the target model parameters, wherein the loss value is obtained by using the preset loss function based on the actual score and the predicted score output by the initial scoring model; the target model parameters include at least: the first weight matrix and first bias of the embedding layer, the second weight matrix and second bias of each layer in the noise reduction feature extraction module, the third weight matrix and third bias of each layer in the fully connected network, and the fourth weight matrix and fourth bias of each layer in the score prediction module; adjusting the initial scoring model based on the target model parameters to obtain the preset scoring model.

[0133] The processor can invoke information and application programs stored in the memory through the transmission device to perform the following steps: based on the first weight matrix and the first bias, perform feature mapping on each customer sub-data using an embedding layer to obtain multiple first embedding vectors; based on the first weight matrix and the first bias, perform feature mapping on each product sub-data using an embedding layer to obtain multiple second embedding vectors; concatenate all the first embedding vectors to obtain a customer feature embedding vector; concatenate all the second embedding vectors to obtain a product feature embedding vector.

[0134] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: using the flattening layer in the feature fusion structure to flatten the customer feature embedding vector and the product feature embedding vector to obtain a first flattened vector and a second flattened vector; using the combining layer in the feature fusion structure to combine the first flattened vector and the second flattened vector to obtain a combined feature.

[0135] The processor can access the information and application program stored in the memory via the transmission device to perform the following steps: based on the second weight matrix and second bias corresponding to the first layer network in the encoder, the combined features are processed using an activation function to obtain initial features; based on the second weight matrix and second bias corresponding to the next layer network in the encoder, the initial features are processed using an activation function until the last layer network in the encoder outputs features, and these features are represented as the first feature; the first feature is processed using a decoder to obtain the second feature; and the second feature is processed using a residual structure to obtain the denoised feature.

[0136] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: use the score prediction module to fuse the noise reduction features and the preset features to obtain the fused features; based on the fourth weight matrix and the fourth bias corresponding to the score prediction module, process the fused features to obtain the score of each financial product.

[0137] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: flatten the noise reduction features to obtain a third flattened vector, and flatten the preset features to obtain a fourth flattened vector; combine the third flattened vector and the fourth flattened vector to obtain a fused feature.

[0138] This application provides a scheme for identifying financial products. By acquiring target customer data (such as asset level) and financial product data (such as product type and risk level) while ensuring compliance with user privacy protection regulations, and processing this data using an embedding layer to convert it into customer feature embedding vectors and product feature embedding vectors, a denoising feature extraction module is used to denoise the concatenated combined features, obtaining cleaner denoised features. Simultaneously, the combined features are further processed to obtain preset features of the same dimension as the denoised features. Then, a scoring model is used to comprehensively analyze the denoised features and preset features to accurately predict and select the financial product with the highest score. This financial product is then presented to the target customer as the target financial product, achieving the goal of improving the accuracy and personalization of financial product identification, thereby enhancing customer satisfaction.

[0139] Those skilled in the art will understand that Figure 5 The structure shown is for illustrative purposes only. Electronic devices can also be terminal devices such as smartphones, tablets, PDAs, and mobile internet devices (MIDs). Figure 5 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 5 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 5 The different configurations shown.

[0140] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0141] Example 4

[0142] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the method for determining financial products provided in Embodiment 1.

[0143] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0144] This application also provides a computer program product that, when executed on a data processing device, is suitable for performing the steps of a method for determining a financial product.

[0145] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0146] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0147] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0148] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0149] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0150] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. 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 storage medium 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 described in the various embodiments of this application. The aforementioned storage medium 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.

[0151] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for determining a financial product, characterized in that, include: With the authorization of the target customer, the customer data of the target customer and financial product data are obtained. The customer data and the financial product data are embedded to obtain customer feature embedding vectors and product feature embedding vectors, and the customer feature embedding vectors and the product feature embedding vectors are concatenated to obtain combined features; The combined features are subjected to noise reduction processing to obtain noise-reduced features, and the combined features are further processed to obtain preset features with the same dimension as the noise-reduced features; The noise reduction features and the preset features are processed using a preset scoring model to obtain a score for each financial product, and the financial product with the highest score is identified as the target financial product, wherein the target financial product is used to be displayed on the target customer's terminal client.

2. The determination method according to claim 1, characterized in that, Before obtaining the target customer's customer data, the process also includes: Construct an initial scoring model, wherein the model structure of the initial scoring model includes at least: an embedding layer, a noise reduction feature extraction module, a fully connected network, and a score prediction module; Historical customer data and the financial product data are collected to obtain training data; The training data is labeled to obtain labeling information, wherein the labeling information includes at least: the actual score of each financial product; Based on a preset loss function, the initial scoring model is trained using the training data and the labeled information until the loss value is less than a preset loss threshold, thereby obtaining the target model parameters. The loss value is obtained by using the preset loss function based on the actual score and the predicted score output by the initial scoring model. The target model parameters include at least: a first weight matrix and a first bias of the embedding layer; a second weight matrix and a second bias of each layer in the denoising feature extraction module; a third weight matrix and a third bias of each layer in the fully connected network; and a fourth weight matrix and a fourth bias of each layer in the score prediction module. Based on the target model parameters, the initial scoring model is adjusted to obtain the preset scoring model.

3. The determination method according to claim 2, characterized in that, The customer data includes multiple customer sub-data, and the financial product data includes multiple product sub-data. The step of embedding the customer data and the financial product data to obtain customer feature embedding vectors and product feature embedding vectors includes: Based on the first weight matrix and the first bias, the embedding layer is used to perform feature mapping on each type of customer sub-data to obtain multiple first embedding vectors; Based on the first weight matrix and the first bias, the embedding layer is used to perform feature mapping on each product sub-data to obtain multiple second embedding vectors; Concatenate all the first embedding vectors to obtain the customer feature embedding vector; All the second embedding vectors are concatenated to obtain the product feature embedding vector.

4. The determination method according to claim 1, characterized in that, The preset scoring model also includes a feature fusion structure, which involves concatenating the customer feature embedding vector and the product feature embedding vector to obtain combined features, including: The flattening layer in the feature fusion structure is used to flatten the customer feature embedding vector and the product feature embedding vector to obtain a first flattened vector and a second flattened vector. The first flattened vector and the second flattened vector are combined using the combination layer in the feature fusion structure to obtain the combined feature.

5. The determination method according to claim 2, characterized in that, The noise reduction feature extraction module includes an encoder, a decoder, and a residual structure. The step of performing noise reduction processing on the combined features to obtain the noise-reduced features includes: Based on the second weight matrix and the second bias corresponding to the first layer network in the encoder, the combined features are processed by an activation function to obtain initial features; Based on the second weight matrix and the second bias corresponding to the next layer of the encoder, the initial feature is processed by the activation function until the last layer of the encoder outputs a feature, and this feature is represented as the first feature. The first feature is processed using the decoder to obtain the second feature; The second feature is processed using the residual structure to obtain the noise reduction feature.

6. The determination method according to claim 2, characterized in that, The steps of processing the denoised features and the preset features using a pre-defined scoring model to obtain a score for each financial product include: The score prediction module is used to fuse the noise reduction features and the preset features to obtain fused features; Based on the fourth weight matrix and the fourth bias corresponding to the score prediction module, the fusion features are processed to obtain the score of each financial product.

7. The determination method according to claim 6, characterized in that, The step of fusing the denoising features and the preset features using the score prediction module to obtain fused features includes: The noise reduction feature is flattened to obtain a third flattened vector, and the preset feature is flattened to obtain a fourth flattened vector. The third and fourth flattening vectors are combined to obtain the fused feature.

8. A device for determining a financial product, characterized in that, include: The acquisition unit is used to acquire customer data of the target customer and financial product data, provided that the target customer has authorized the acquisition. The first processing unit is used to perform embedding processing on the customer data and the financial product data to obtain customer feature embedding vectors and product feature embedding vectors, and to concatenate the customer feature embedding vectors and the product feature embedding vectors to obtain combined features. The second processing unit is used to perform noise reduction processing on the combined features to obtain noise-reduced features, and to process the combined features to obtain preset features with the same dimension as the noise-reduced features. The third processing unit is used to process the noise reduction features and the preset features using a preset scoring model to obtain a score for each financial product, and to determine the financial product with the highest score as the target financial product, wherein the target financial product is used to be displayed on the target customer's terminal client.

9. A computer program product, characterized in that, The method includes a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for determining a financial product as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, It includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method for determining a financial product as described in any one of claims 1 to 7.

Images