Financial resource configuration method and device and electronic equipment

Abstract

The invention provides a financial resource configuration method and device and electronic equipment. Comprising the steps that customer portrait data is determined, and the customer portrait data is a data set representing different features of customers; the customer portrait data is analyzed through a target recognition model to obtain a prediction result, the prediction result represents whether the customer is a potential customer, and the target recognition model is obtained through machine learning training according to multiple groups of data; each group of data in the multiple groups of data comprises historical portrait data and a label of a result indicating whether a customer corresponding to the historical portrait data is a potential customer; and under the condition that the prediction result is greater than a preset threshold value, determining that the customer is a potential customer, and configuring corresponding financial resources for the potential customer, the financial resources at least comprising financial products. Through the method and the device, the problems that whether the customer is a potential customer or not cannot be accurately identified in the prior art, financial resources cannot be accurately configured for the customer, and customer satisfaction is poor are solved.

Description

Technical Field

[0001] This application relates to the technical field of financial data processing. Specifically, it relates to a method, apparatus, computer-readable storage medium, and electronic device for allocating financial resources. Background Art

[0002] The bill business occupies an important position in financial institutions. It is not only an important way for financial institutions to serve the real economy and optimize resource allocation, but also an effective means to enhance their own profitability and risk management capabilities. However, to achieve the booming development of the bill business, it is inevitable to rely on a large user group and excellent service quality. Therefore, how to analyze and mine customer-related data to find potential customers with business value has become a crucial link in the high-quality development of the bill business.

[0003] In the prior art, technologies such as LightGBM (Light Gradient Boosting Machine, based on the gradient boosting decision tree algorithm, abbreviated as LGBM) or SVM (Support Vector Machine, abbreviated as SVM) are used. In practical applications, they often perform better on medium and small-scale data. Moreover, as the data scale further increases, not only does the required computational volume increase sharply, but the model performance actually decreases. In addition, in the feature engineering step of machine learning technology, more reliance is placed on manually selecting and extracting features from the original data and assigning weights before training the machine learning model. This is not easy in the context of the complex business data of financial institutions. Finally, due to the problem of positive and negative example samples with high similarity in the data of financial institutions, the models in these solutions often cannot correctly classify this data, thus unable to accurately mine potential customers. Summary of the Invention

[0004] The main purpose of this application is to provide a method, apparatus, computer-readable storage medium, and electronic device for allocating financial resources, so as to at least solve the problem in the prior art that it is impossible to accurately identify whether a customer is a potential customer, and thus impossible to accurately allocate financial resources for the customer, resulting in poor customer satisfaction.

[0005] To achieve the above object, according to one aspect of the present application, a method for allocating financial resources is provided, including: determining customer portrait data, where the customer portrait data is a data set representing different characteristics of customers; analyzing the customer portrait data through a target recognition model to obtain a prediction result, where the prediction result indicates whether the customer is a potential customer, and the target recognition model is obtained through machine learning training based on multiple sets of data, and each set of data in the multiple sets of data includes: historical portrait data and a label of the result indicating whether the customer corresponding to the historical portrait data is a potential customer; in the case where the prediction result is greater than a preset threshold, determining that the customer is the potential customer and allocating corresponding financial resources to the potential customer, where the financial resources at least include financial products.

[0006] Optionally, before analyzing the customer portrait data through the target recognition model, the method further includes: obtaining a feature extraction model, training the feature extraction model through the historical portrait data and the label of the result indicating whether the customer corresponding to the historical portrait data is a potential customer until the loss function value of the feature extraction model is minimized, where the feature extraction model at least includes an input layer, an encoding layer, a decoding layer, and an output layer; keeping the structural parameters of the input layer, the encoding layer, and the decoding layer unchanged, replacing the output layer with the structural layer of a multi-layer perceptron network model, and training until the loss function value of the multi-layer perceptron network model is minimized to obtain the target recognition model.

[0007] Optionally, training the feature extraction model through the historical portrait data and the label of the result indicating whether the customer corresponding to the historical portrait data is a potential customer until the loss function value of the feature extraction model is minimized includes: calculating the loss function value of the feature extraction model using a negative example sample loss function until the loss function value of the feature extraction model is minimized, where the negative example sample loss function is used to increase the difference between positive example samples and negative example samples, the positive example samples represent samples with a prediction result indicating a potential customer, and the negative example samples represent samples that are not potential customers.

[0008] Optionally, replacing the output layer with the structural layer of a multi-layer perceptron network model and training until the loss function value of the multi-layer perceptron network model is minimized includes: replacing the output layer with the structural layer of a multi-layer perceptron network model having two fully connected layers; continuing to train the multi-layer perceptron network model through the historical portrait data and the label of the result indicating whether the customer corresponding to the historical portrait data is a potential customer until the cross-entropy loss function value of the multi-layer perceptron network model is minimized.

[0009] Optionally, after determining the customer portrait data, the method further includes: performing noise removal on the customer portrait data by using a wavelet transform method to obtain the denoised customer portrait data.

[0010] Optionally, determining the customer portrait data includes: obtaining customer basic information, and generating an identity feature of the customer according to the customer basic information, where the customer basic information at least includes a name and an age; obtaining customer financial information, and generating a financial feature of the customer according to the customer financial information, where the customer financial information at least includes income and liabilities; obtaining customer transaction bill information, and generating a transaction feature of the customer according to the customer transaction bill information, where the customer transaction bill information at least includes a transaction time and a transaction type.

[0011] Optionally, before analyzing the customer portrait data by using a target recognition model, the method further includes: generating a data sequence corresponding to the customer portrait data, and inputting the data sequence into the target recognition model.

[0012] According to another aspect of the present application, there is provided an apparatus for allocating financial resources, including: a first determination unit, configured to determine customer portrait data, where the customer portrait data is a data set representing different features of a customer; an analysis unit, configured to analyze the customer portrait data by using a target recognition model to obtain a prediction result, where the prediction result indicates whether the customer is a potential customer, and the target recognition model is obtained by machine learning training based on multiple sets of data, and each set of data in the multiple sets of data includes: historical portrait data and a label of a result indicating whether the customer corresponding to the historical portrait data is a potential customer; a second determination unit, configured to determine that the customer is the potential customer and allocate corresponding financial resources to the potential customer when the prediction result is greater than a preset threshold, where the financial resources at least include financial products.

[0013] According to still another aspect of the present application, there is provided a computer-readable storage medium, where the computer-readable storage medium includes a stored program, and when the program runs, it controls a device where the computer-readable storage medium is located to execute any one of the methods for allocating financial resources.

[0014] According to yet another aspect of the present application, there is provided an electronic device, including: one or more processors, a memory, and one or more programs, where the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include those for executing any one of the methods for allocating financial resources.

[0015] Applying the technical solution of the present application to determine customer portrait data, analyzing the customer portrait data through a target recognition model to obtain a prediction result. When the prediction result is greater than a preset threshold, it is determined that the customer is a potential customer, and corresponding financial resources are configured for the potential customer. The target recognition model is trained based on historical portrait data and labels of whether the customers corresponding to the historical portrait data are potential customers. The target recognition model determines whether a customer is a potential customer by identifying features, so as to accurately configure corresponding financial resources for the customer and improve customer satisfaction. Compared with the prior art where it is impossible to accurately identify whether a customer is a potential customer, and thus unable to accurately configure financial resources for the customer, resulting in poor customer satisfaction, the present application predicts whether the customer corresponding to the customer portrait data is a potential customer through the trained target recognition model. Therefore, it can solve the problems in the prior art of being unable to accurately identify whether a customer is a potential customer, and thus unable to accurately configure financial resources for the customer, and poor customer satisfaction, achieving the effects of improving prediction accuracy and enhancing customer satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings forming a part of this application are used to provide a further understanding of the application. The illustrative embodiments and descriptions thereof of the application are used to explain the application and do not constitute an improper limitation to the application. In the drawings:

[0017] Figure 1 A hardware structure block diagram of a mobile terminal for implementing a method of allocating financial resources provided by an embodiment of the present application is shown;

[0018] Figure 2 A flowchart of a method for allocating financial resources provided by an embodiment of the present application is shown;

[0019] Figure 3 A flowchart of a specific method for allocating financial resources provided by an embodiment of the present application is shown;

[0020] Figure 4 A structure block diagram of a device for allocating financial resources provided by an embodiment of the present application is shown.

[0021] Among them, the above-mentioned accompanying drawings include the following reference numerals:

[0022] 102, processor; 104, memory; 106, transmission device; 108, input / output device. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0024] To enable those skilled in the art to better understand the solution of this application, the following will clearly and completely describe the technical solutions in the embodiments of this application with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this application.

[0025] It should be noted that the terms "first", "second", etc. in the specification, claims and the above-mentioned drawings of this application are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances for the embodiments of this application described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those clearly listed steps or units, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0026] For the convenience of description, some nouns or terms related to the embodiments of this application are described below:

[0027] Loss function: It is used to quantify the difference between the predicted value of the model and the true value. The loss function is a function that maps the prediction result of the model and the true label to a real number loss value (usually non-negative), and this loss value reflects the performance of the model on a single data point. During the training process, the goal of the model is to minimize the average loss of all data points, thereby improving the generalization ability of the model.

[0028] Hard negative example: Also known as a difficult sample, it usually refers to a negative sample that is extremely similar to the positive sample and is difficult to be learned by the model in supervised learning.

[0029] Gradient vanishing: It refers to the phenomenon that during the training of a deep learning model, when performing backpropagation based on the loss calculated by the loss function according to the output features of the model, when the gradient becomes very small and approaches zero, the weight update in the network becomes very slow or even almost stops updating.

[0030] Gradient explosion: It refers to the phenomenon that during the training of a deep learning model, when performing backpropagation based on the loss calculated by the loss function according to the output features of the model, when the gradient becomes very large and approaches infinity, the weight update in the network becomes very large and uncontrollable, resulting in the model being unable to learn effectively.

[0031] It should be noted that there is an error in your original text. The description of "gradient explosion" in item is the same as that of "gradient vanishing" in item . The correct description of "gradient explosion" should be: It refers to the phenomenon that during the training of a deep learning model, when performing backpropagation based on the loss calculated by the loss function according to the output features of the model, when the gradient becomes very large and approaches infinity, the weight update in the network becomes very large and uncontrollable, resulting in the model being unable to learn effectively. I have translated it correctly according to the corrected content.As introduced in the background art, in the prior art, it is impossible to accurately identify whether a customer is a potential customer, and thus it is impossible to accurately allocate financial resources for the customer, resulting in poor customer satisfaction. To solve the problem of poor customer satisfaction, embodiments of the present application provide a method, an apparatus, a computer-readable storage medium, and an electronic device for allocating financial resources.

[0032] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention.

[0033] The method embodiments provided in the embodiments of the present application can be executed on a mobile terminal, a computer terminal, or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 is a hardware structure block diagram of a mobile terminal of a method for allocating financial resources according to an embodiment of the present invention. As Figure 1 shown, the mobile terminal may include one or more ( Figure 1 only one is shown in Figure 1 a processor 102 (the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. Among them, the above mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those of ordinary skill in the art can understand that Figure 1 the structure shown is only schematic and does not limit the structure of the above mobile terminal. For example, the mobile terminal may further include more or fewer components than Figure 1 shown, or have a different configuration from

[0034] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the financial resource allocation method in the embodiments of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, the above-mentioned method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely disposed relative to the processor 102, and these remote memories can be connected to the mobile terminal through a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the above-mentioned network may include a wireless network provided by a communication provider of the mobile terminal. In one instance, the transmission device 106 includes a network adapter (Network Interface Controller, abbreviated as NIC), which can be connected to other network devices through a base station and thus can communicate with the Internet. In one instance, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0035] In this embodiment, a financial resource allocation method running on a mobile terminal, a computer terminal, or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0036] Figure 2 It is a flowchart of the financial resource allocation method according to the embodiments of the present application. As Figure 2 shown, the method includes the following steps:

[0037] Step S201, determine customer portrait data, where the customer portrait data is a data set characterizing different features of the customer;

[0038] Specifically, the customer portrait data is a collection of multi-dimensional data such as the customer's personal information, transaction records, credit status, and financial indicators. In the scenario of bill business, the customer portrait data may include the enterprise's financial reports, credit scores, details of historical bill transactions (such as transaction amounts, frequencies, time points), and other behavior indicators related to bills. The customer can be a specific individual, or an enterprise or company, etc.

[0039] Step S202, analyze the customer portrait data through the target recognition model to obtain a prediction result, where the prediction result indicates whether the customer is a potential customer, the target recognition model is obtained through machine learning training based on multiple sets of data, and each set of data in the multiple sets of data includes: historical portrait data and a label of the result indicating whether the customer corresponding to the historical portrait data is a potential customer;

[0040] Specifically, the target recognition model is a model obtained through machine learning training, aiming to identify and predict which customers have potential bill business needs, that is, whether they are potential customers. During the training phase, the model will learn a large amount of labeled customer profile data to understand which features are associated with potential customers. The importance of identifying potential customers in the bill business is mainly reflected in the following aspects: 1. It is conducive to expanding the market share: By identifying and connecting with the potential customers that have not been fully developed in the market, financial institutions can broaden the coverage of the bill business and increase the market share. 2. It is conducive to optimizing resource allocation: Identifying potential customers helps to accurately match customer needs, rationally allocate financial resources, and provide customized bill financing solutions for enterprises with different credit ratings and scales, thus improving the efficiency of fund use. 3. It is conducive to risk control in advance: Conducting in-depth analysis and evaluation before introducing new customers helps financial institutions to identify and control credit risks in advance, ensure the quality of the new customer group, and reduce the non-performing asset ratio. 4. It is conducive to enhancing profitability: By identifying more high-quality potential customers, financial institutions can expand the volume of bill discounting, re-discounting and other businesses, increase the sources of commission fees and interest income, and achieve business growth and profit improvement. 5. It is conducive to innovating customer service: Deeply understanding the characteristics of potential customers is conducive to financial institutions to continuously innovate product and service models, improve customer satisfaction and loyalty, and build long-term and stable customer relationships. The above-mentioned target recognition model can specifically be a model obtained by combining the training of the Transformer model and the MLP model. Transformer is a model with an Encoder-Decoder structure. Its sophisticated self-attention parallel structure has stronger long-term dependence modeling ability, is more suitable for processing long-sequence related tasks, and can solve the problem that the class RNN (Recurrent Neural Network) model cannot completely eliminate gradient vanishing and gradient explosion when facing long sequences, and performs excellently in sequence data processing problems. Specifically, the combined training of the Transformer model and the MLP model will be described in detail below. During the machine learning training process, the model needs to learn and understand the historical customer profile data. These data not only contain the feature information of the customers, but each data sample also has a corresponding label indicating whether the customer was or is currently a potential customer in the bill business. The historical profile data is the basis for model training, helping the model learn the patterns that can distinguish potential customers from non-potential customers.

[0041] Step S203, when the prediction result is greater than the preset threshold, determine that the customer is the potential customer, and configure corresponding financial resources for the potential customer, where the financial resources at least include financial products.

[0042] Specifically, after analyzing the customer profile data, the target recognition model will output a predicted probability or score. The magnitude of the predicted result reflects the model's judgment of the customer's potential needs. Generally, the closer the value is to 1 (or a set high threshold), the higher the likelihood that the customer is a potential customer. Between the output of the model and its actual application, a decision boundary is required, that is, a preset threshold. This threshold is a boundary set by financial institutions when applying the model's prediction results to distinguish whether a customer is classified as a potential customer. The setting of the threshold is usually based on factors such as business requirements, risk preferences, and the performance metrics of the model (such as precision, recall), etc. When the predicted result is greater than the preset threshold, the model determines that the customer has a high likelihood of business needs, and thus defines this customer as a potential customer. Potential customers are a key concept in the customer relationship management of financial institutions, representing the group of people who have the potential to become bank customers and use specific financial products. After determining that a customer is a potential customer, the financial institution will allocate appropriate financial resources to the potential customer according to the customer's specific needs and profile data. Here, the financial resources mainly refer to financial products, including but not limited to bill discounting, bill acceptance, bill pledge loans, commercial credit loans, etc. Allocating financial resources aims to provide customers with financial solutions that match their needs, while optimizing the resource allocation of financial institutions and achieving risk control and profit goals.

[0043] Through this embodiment, customer profile data is determined, and the target recognition model analyzes the customer profile data to obtain a predicted result. In the case where the predicted result is greater than the preset threshold, the customer is determined to be a potential customer, and corresponding financial resources are allocated to the potential customer. The target recognition model is trained based on historical profile data and the labels of whether the customers corresponding to the historical profile data are potential customers. The target recognition model determines whether a customer is a potential customer through feature recognition, so as to accurately allocate corresponding financial resources to the customer and improve customer satisfaction. Compared with the prior art where it is impossible to accurately identify whether a customer is a potential customer, and thus unable to accurately allocate financial resources to the customer, resulting in poor customer satisfaction, in this application, the target recognition model obtained through training predicts whether the customer corresponding to the customer profile data is a potential customer. Therefore, it can solve the problem in the prior art of being unable to accurately identify whether a customer is a potential customer, and thus unable to accurately allocate financial resources to the customer, resulting in poor customer satisfaction, and achieve the effects of improving prediction accuracy and enhancing customer satisfaction.

[0044] In the specific implementation process, the above method further includes step S204: Before analyzing the customer portrait data through the target recognition model, obtain a feature extraction model, and train the feature extraction model with the historical portrait data and the label of whether the customer corresponding to the historical portrait data is a potential customer until the loss function value of the feature extraction model is minimized, where the feature extraction model at least includes an input layer, an encoding layer, a decoding layer, and an output layer; step S205: Keep the structural parameters of the input layer, the encoding layer, and the decoding layer unchanged, replace the output layer with the structural layer of a multi-layer perceptron model, and train until the loss function value of the multi-layer perceptron model is minimized to obtain the target recognition model. This method completes model training in two steps. After the feature extraction model is trained, freeze the weight parameters of the input module, encoding module, and decoding module except the output module, and replace the original output module with an MLP (Multi-Layer Perceptron) structure, and train with the loss function until the loss is minimized, which can ensure the generalization ability of the model on different customer portrait data, help capture complex patterns in the data, and enable the model to make more accurate predictions when facing new data.

[0045] Specifically, first obtain an initial feature extraction model, which at least includes an input layer, an encoding layer, a decoding layer, and an output layer. The input layer receives the historical portrait data of customers, including but not limited to financial data, transaction records, credit scores, etc. The encoding layer and the decoding layer are responsible for performing deep feature extraction and representation learning on the input data, which is usually based on complex neural network structures, such as the Transformer model, to capture the long-term dependencies and context information in the data. Then, use the historical portrait dataset with labels to train the feature extraction model, with the goal of minimizing the loss function value, which reflects the gap between the model's prediction results and the true labels. The training process may include adjusting the learning rate, batch size, regularization parameters, etc., to optimize the model performance and ensure that the model can learn the features that distinguish potential customers from non-potential customers from the data. Keeping the structural parameters unchanged and replacing the output layer: After the loss function value of the feature extraction model reaches the minimum, that is, the model has learned an effective feature representation, at this time, keep the structural parameters of the input layer, encoding layer, and decoding layer unchanged, which ensures that the model's feature learning ability is retained. Then, replace the original output layer with the structural layer of a multi-layer perceptron network model. The MLP model consists of multiple fully connected layers and can perform non-linear transformation and combination on the extracted features to generate the final prediction result, such as whether a customer is a potential customer. Final training of the target recognition model: After replacing the output layer, continue to train the model using the new MLP structural layer, with the goal of minimizing the loss function value of the multi-layer perceptron network model. This process may require readjusting the parameters of the MLP layer, including the number of neurons, activation function, loss function type, etc., to ensure that the model can accurately predict whether a customer has potential needs based on the representation generated by the feature extraction layer. After training is completed, the resulting target recognition model will be able to efficiently process new customer portrait data and output the probability estimate of whether a customer is a potential customer, providing data support for the customer relationship management and market expansion strategies of financial institutions.

[0046] The input layer of the model mainly includes the source sequence data embedding layer and its position encoder, and the data embedding layer and its position encoder. Among them, the data embedding layer is used to transform the digital representation in the sequence data into a vector representation to capture the dependencies between data in the high-dimensional vector space. The encoding layer is used to extract features from the input sequence data. Its structure is composed of N encoder layers stacked together. Each encoder layer consists of two sub-layer connection structures. The first sub-layer connection structure includes a multi-head self-attention sub-layer, a normalization layer, and a residual connection. The second sub-layer connection structure includes a feed-forward fully-connected sub-layer, a normalization layer, and a residual connection. The decoding layer is used to represent the features of the next possible 'value' based on the output result of the encoder and the result of the previous prediction. Its structure is similar to that of the encoding module and is composed of N decoder layers stacked together. Each decoder layer consists of three sub-layer connection structures. The first sub-layer connection structure includes a multi-head self-attention sub-layer, a normalization layer, and a residual connection. The second sub-layer connection structure includes a multi-head attention sub-layer, a normalization layer, and a residual connection. The third sub-layer connection structure includes a feed-forward fully-connected sub-layer, a normalization layer, and a residual connection. The output layer is a linear layer, which is responsible for linearly transforming the output of the decoding module to obtain an output of a specified dimension, and plays a role in converting the output of the decoding module into a prediction result. Contrastive learning makes two similar things continue to be similar in the representation space. By continuously optimizing the contrastive loss function, a certain similarity measure between similar samples in the representation space is increased, and the similarity measure between dissimilar samples in the representation space is decreased.

[0047] In some alternative embodiments, step S204 can be implemented through the following steps. Step S2041: Calculate the loss function value of the feature extraction model using the negative example sample loss function until the loss function value of the feature extraction model is minimized. Among them, the negative example sample loss function is used to increase the difference between positive example samples and negative example samples. The positive example samples represent the samples whose prediction results are characterized as potential customers, and the negative example samples represent the samples that are not potential customers. By introducing and optimizing the negative example sample loss function, this method improves the ability of the model to extract features of positive and negative samples, reduces the similarity of the feature projections of positive and negative samples, and amplifies the difference between positive and negative samples to accurately extract sample features.

[0048] In the specific implementation process, when dealing with classification tasks such as bill potential customer mining, a major challenge faced by the model is to distinguish between positive and negative samples, that is, to correctly identify which customers are potential business opportunities and which are not. Traditional loss functions may perform poorly when faced with high similarity between positive and negative sample features, resulting in blurred classification boundaries and reduced classification accuracy. In order to meet this challenge, a negative sample loss function is specially introduced. The purpose of this loss function is to increase the distance between positive and negative samples in the feature space, thereby improving the classification ability of the model. Specifically, the negative sample loss function will focus on those non-potential customer samples that are misclassified as potential customers, that is, hard negative samples, when calculating. By designing a mechanism, the representation of positive samples in the feature space is as far away from the representation of negative samples as possible, which helps the model learn clearer and more discriminative feature representations, so that it can more accurately distinguish between potential customers and non-potential customers when predicting. The negative sample loss function is shown in the following formula:

[0049]

[0050] Wherein, τ is the temperature coefficient, which is used to adjust the calculated similarity; |P(i)| is the size of the set of positive samples of data sample i; z i is the output feature of the i-th sample output by the transformer model; γ is the adjustment coefficient of the regular term, represents the supervised loss function, z a represents the output feature of sample a, z p Represents the output feature of sample p. As can be seen from the formula, when there are multiple positive samples, the loss function maximizes the loss between i and all positive samples p∈P i The average similarity between the two samples brings the feature center distance of this sample closer to that of all positive samples. bound is a regular term responsible for penalizing hard negative samples (or negative samples), further reducing the average similarity between the two, and solving the hard negative problem that is difficult to optimize in contrastive training. The regular term expression is Where L is the label set in the dataset, z i represents a positive sample, z l Represents negative samples.

[0051] In some alternative embodiments, step S204 further includes: step S2042: replacing the output layer with a structural layer of a multi-layer perceptron network model having two fully-connected layers; step S2043: continuing to train the multi-layer perceptron network model with the historical portrait data and the label of whether the customer corresponding to the historical portrait data is a potential customer until the cross-entropy loss function value of the multi-layer perceptron network model is minimized. This method enhances the non-linear expression ability of the model by introducing an MLP (Multi-Layer Perceptron), enabling the model to learn more complex decision boundaries, thereby improving the prediction accuracy of potential customers.

[0052] Specifically, the feature extraction model has learned during the training process how to convert the input customer portrait data into a high-dimensional feature representation, but the original output layer may simply make classification decisions based on these features. To improve the prediction ability of the model, the original output layer is replaced with an MLP here. An MLP is a feed-forward neural network that contains at least two fully-connected layers. It can perform non-linear transformations on the high-dimensional feature representations generated by the feature extraction model, thereby better fitting complex data distributions and improving the classification performance of the model. Continue training until the minimum cross-entropy loss: After replacing the output layer, the model needs to be further trained to optimize the new structural layer. The historical portrait data and the corresponding label information (i.e., the classification result of whether the customer is a potential customer) used during the training process are used to adjust the weight parameters of the MLP by minimizing the cross-entropy loss function. The cross-entropy loss function is a commonly used metric for measuring the difference between the predicted probability distribution of the model and the true label distribution, and it is particularly effective for classification tasks. By comparing the predicted results of the model with the actual labels and adjusting the weights to minimize the cross-entropy loss, the model can more accurately predict potential customers on the training set, thereby improving the overall classification accuracy and the generalization ability of the model. The expression of the cross-entropy loss function is as follows:

[0053]

[0054] where y i represents the classification of sample i, with the positive class being 1 and the negative class being 0; p i represents the probability that sample i is predicted as the positive class, and N represents the number of samples.

[0055] In some other alternative embodiments, after determining the customer portrait data, the method further includes step S205: using the wavelet transform method to remove noise from the customer portrait data to obtain the denoised customer portrait data. This method removes noise through the wavelet transform method, which can not only improve the data quality, optimize the feature learning and model training processes, but also significantly enhance the accuracy of financial institutions in potential customer mining.

[0056] In the specific implementation process, the wavelet transform denoising technology has good locality characteristics in both time and frequency spaces. Therefore, it can remove noise effectively while retaining the low-frequency and stable characteristics of the original signal. The present invention uses wavelet transform to remove noise from sequence data, significantly improving the effectiveness of data denoising and providing a data basis for more accurate analysis and prediction of time series data. The input sequence is denoised using the wavelet function shown in the following formula. The formula of the wavelet function is as follows:

[0057]

[0058] where x(t) represents the data sequence corresponding to the original customer portrait data, and ψ t is a continuous mother wavelet, first scaled to a and then translated to b. In the DWT domain, the scale factor and translation factor use discrete values. As the time series granularity increases, the scale factor increases exponentially, i.e., a = 1, 2, 4..., and the translation factor increases by integers, i.e., b = 1, 2, 3....

[0059] In some other alternative embodiments, step S201 can be implemented through the following steps: step S2011: obtain the customer's basic information, and generate the identity characteristics of the customer according to the customer's basic information, where the customer's basic information includes at least name and age; step S2012: obtain the customer's financial information, and generate the financial characteristics of the customer according to the customer's financial information, where the customer's financial information includes at least income and liabilities; step S2013: obtain the customer's transaction bill information, and generate the transaction characteristics of the customer according to the customer's transaction bill information, where the customer's transaction bill information includes at least transaction time and transaction type.

[0060] Specifically, the customer can be an enterprise or an individual. The customer information mainly includes customer basic information, customer financial information, and customer transaction bill information. In addition, it can also include credit risk information, executive and shareholder information, and statistical characteristics of customer bill transactions. Among them, when the customer is an individual, the customer basic information includes at least name and age; when the customer is an enterprise, the customer basic information can specifically include: enterprise name, address information, enterprise nature, registered capital, total assets, operating income, holding type, enterprise scale, affiliated institution, enterprise legal person, credit granting situation, etc. In addition to income and liabilities, the customer financial information can also include: whether it is a listed company, credit rating, default probability, total liabilities, pre-tax profit, main business income, accounts receivable, total current assets, and total current liabilities, etc. In addition to transaction time and transaction type, the customer transaction bill information can also include: counterparty, face value, bill acceptance bank, etc. In addition to the above customer information, it can also include: customer credit risk information, which can specifically include: risk warning code, gray list hit flag, etc. Customer executive and shareholder information can specifically include: executive or shareholder name, executive category, actual controller flag, name of shareholder-related enterprise, type of shareholder-related enterprise, shareholder shareholding ratio, and association type, etc. Statistical characteristics of customer bill transactions can specifically include: average monthly bill issuing times of the customer, average monthly bill issuing amount of the customer, average monthly bill issuing time point of the customer, average monthly bill receiving times of the customer, average monthly bill receiving amount of the customer, average monthly bill receiving time point of the customer, average monthly bill pledging times of the customer, average monthly bill pledging amount of the customer, average monthly bill pledging time point of the customer, customer-signed discount type, cumulative discount amount in the current year, cumulative weighted discount rate in the current year, proportion of online banking bill receiving discounted by our bank, customer endorsement times statistics, customer stratification label, customer activity label, customer potential label, etc.

[0061] In some other alternative embodiments, before analyzing the customer portrait data through the target recognition model, the method further includes step S206: generating a data sequence corresponding to the customer portrait data and inputting the data sequence into the target recognition model. The method of generating a data sequence corresponding to the customer portrait data and inputting it into the target recognition model can make full use of the model's processing ability for sequence data, improve the depth of customer feature learning, and thus achieve higher accuracy and efficiency in potential customer identification.

[0062] Specifically, first, it is necessary to integrate the customer base information, financial information, transaction bill information, etc. obtained from different sources to ensure that all relevant information is taken into account. This may involve preprocessing steps such as data cleaning, format unification, and missing value filling to ensure data quality and consistency. Next, according to business requirements and model characteristics, the integrated customer information is constructed into a data sequence. During the construction process, it is important to determine the length, frequency, and element order of the sequence. For example, the historical transaction bill information of customers can be arranged in chronological order to form a time series, which helps the model capture the evolution trend and periodic characteristics of customer transaction behaviors. When constructing the sequence, features that are helpful for potential customer identification need to be selected and encoded. For categorical variables, one-hot encoding or other encoding methods may be used to convert them into numerical representations; for numerical variables, normalization or standardization may be required to ensure data comparability and model training effects. If the customer profile data contains a large amount of information, it may be necessary to divide it into multiple sequences, each sequence focusing on different customer attributes or transaction types, which helps the model learn customer characteristics from different perspectives and improve the accuracy and comprehensiveness of prediction.

[0063] In order to enable those skilled in the art to more clearly understand the technical solution of the present application, the implementation process of the financial resource allocation method of the present application will be described in detail below in conjunction with specific embodiments.

[0064] This embodiment relates to a specific financial resource allocation method, as Figure 3 shown, including the following steps:

[0065] Step S1: Data sampling. The collected customer profile data mainly includes the customer's basic information, financial information, credit risk information, executive and shareholder information, customer bill transaction history information, customer bill transaction statistical characteristics, etc.;

[0066] Step S2: Data preprocessing. Wavelet transform is used to remove noise from the sequence data corresponding to the customer profile data;

[0067] Step S3: Feature extraction model training. The Transformer model is used as the feature extraction model, including an input layer, an encoding layer, a decoding layer, and an output layer. Since in the scenario of potential customer identification, there is a phenomenon that some negative example data features are similar to positive example data features, a negative example sample loss function is adopted to increase the penalty for such negative example samples, thereby increasing the feature learning ability of the feature extraction model;

[0068] Step S4: Predictive model training. After the loss of the feature extraction model reaches the minimum, freeze the parameters (including weight parameters, etc.) of the input layer, encoding layer, and decoding layer except for the output layer in the model (i.e., the above structural parameters will no longer be updated). Replace the original output layer with an MLP structure with two fully connected layers, and train it with the cross-entropy loss function until the loss reaches the minimum to obtain the final target recognition model;

[0069] Step S5: Data prediction. Input the customer portrait data into the target recognition model. After the model conducts inference analysis and prediction, obtain the prediction result, and compare the prediction result with the size of the set preset threshold to determine whether the customer is a potential customer.

[0070] The embodiment of the present application also provides a financial resource allocation device. It should be noted that the financial resource allocation device in the embodiment of the present application can be used to execute the financial resource allocation method provided in the embodiment of the present application. This device is used to implement the above embodiments and preferred implementation manners, and those that have been described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0071] The following introduces the financial resource allocation device provided in the embodiment of the present application.

[0072] Figure 4 is a schematic diagram of the financial resource allocation device according to the embodiment of the present application. As Figure 4 shown, the device includes:

[0073] The first determination unit 10 is used to determine customer portrait data, where the customer portrait data is a data set representing different characteristics of a customer;

[0074] Specifically, the customer portrait data is a collection of multi-dimensional data such as a customer's personal information, transaction records, credit status, and financial indicators. In the bill business scenario, the customer portrait data may include an enterprise's financial report, credit score, details of historical bill transactions (such as transaction amount, frequency, time point), and other behavior indicators related to bills. The customer can be a specific individual, or an enterprise or company, etc.

[0075] The analysis unit 20 is used to analyze the customer portrait data through the target recognition model to obtain a prediction result, where the prediction result indicates whether the customer is a potential customer. The target recognition model is obtained through machine learning training based on multiple sets of data, and each set of data in the multiple sets of data includes: historical portrait data and a label of the result indicating whether the customer corresponding to the historical portrait data is a potential customer;

[0076] Specifically, the target recognition model is a model obtained through machine learning training, aiming to identify and predict which customers have potential bill business needs, that is, whether they are potential customers. During the training phase, the model will learn a large amount of labeled customer profile data to understand which features are associated with potential customers. The importance of identifying potential customers in the bill business is mainly reflected in the following aspects: 1. It is conducive to expanding the market share: By identifying and connecting with the potential customers that have not been fully developed in the market, financial institutions can broaden the coverage of the bill business and increase the market share. 2. It is conducive to optimizing resource allocation: Identifying potential customers helps to accurately match customer needs, rationally allocate financial resources, and provide customized bill financing solutions for enterprises with different credit ratings and scales, thereby improving the efficiency of fund use. 3. It is conducive to pre-positioning risk control: Conducting in-depth analysis and evaluation before introducing new customers helps financial institutions to identify and control credit risks in advance, ensure the quality of the new customer group, and reduce the non-performing asset ratio. 4. It is conducive to enhancing profitability: By identifying more high-quality potential customers, financial institutions can expand the volume of bill discounting, re-discounting and other businesses, increase the sources of commission fees and interest income, and achieve business growth and profit improvement. 5. It is conducive to innovating customer services: Deeply understanding the characteristics of potential customers is conducive to financial institutions continuously innovating product and service models, improving customer satisfaction and loyalty, and building long-term and stable customer relationships. The above-mentioned target recognition model can specifically be a model obtained by combining the Transformer model and the MLP model for training. The Transformer is a model with an Encoder-Decoder structure. Its sophisticated self-attention parallel structure has stronger long-term dependence modeling capabilities, is more suitable for processing long-sequence related tasks, and can solve the problem that the class RNN (Recurrent Neural Network, abbreviated as RNN) model cannot completely eliminate gradient vanishing and gradient explosion when facing long sequences, and performs excellently in sequence data processing problems. Specifically, the combined training of the Transformer model and the MLP model will be described in detail below. During the machine learning training process, the model needs to learn and understand the historical customer profile data, which not only contains the feature information of the customers, but also each data sample has a corresponding label indicating whether the customer was or is currently a potential customer of the bill business. The historical profile data is the basis for model training, helping the model learn the patterns that can distinguish potential customers from non-potential customers.

[0077] The second determination unit 30 is configured to determine that the customer is the potential customer and configure corresponding financial resources for the potential customer when the prediction result is greater than a preset threshold, where the financial resources at least include financial products.

[0078] Specifically, after analyzing the customer portrait data, the target recognition model will output a prediction probability or score. The numerical value of the prediction result reflects the model's judgment of the customer's potential needs. Generally, the closer the value is to 1 (or a set high threshold), the higher the likelihood that the customer is a potential customer. Between the output of the model and its actual application, a decision boundary is needed, that is, a preset threshold. This threshold is a boundary set by financial institutions when applying the model's prediction results to distinguish whether a customer is classified as a potential customer. The setting of the threshold is usually based on comprehensive consideration of factors such as business requirements, risk preferences, and the performance indicators of the model (such as precision and recall). When the prediction result is greater than the preset threshold, the model determines that the customer has a high likelihood of business needs, and thus defines the customer as a potential customer. Potential customers are a key concept in the customer relationship management of financial institutions, representing the group of people who have the potential to become bank customers and use specific financial products. After determining that a customer is a potential customer, the financial institution will allocate appropriate financial resources to the potential customer according to the customer's specific needs and portrait data. Here, the financial resources mainly refer to financial products, including but not limited to bill discounting, bill acceptance, bill pledge loans, commercial credit loans, etc. Allocating financial resources aims to provide customers with financial solutions that match their needs, while optimizing the resource allocation of financial institutions to achieve risk control and profit goals.

[0079] Through this embodiment, customer portrait data is determined, the customer portrait data is analyzed by the target recognition model to obtain a prediction result. When the prediction result is greater than the preset threshold, the customer is determined to be a potential customer, and corresponding financial resources are allocated to the potential customer. The target recognition model is trained based on historical portrait data and the labels of whether the customers corresponding to the historical portrait data are potential customers. The target recognition model determines whether a customer is a potential customer through feature recognition, so as to accurately allocate corresponding financial resources to the customer and improve customer satisfaction. Compared with the prior art where it is impossible to accurately identify whether a customer is a potential customer, and thus impossible to accurately allocate financial resources to the customer, resulting in poor customer satisfaction, this application predicts whether the customer corresponding to the customer portrait data is a potential customer through the trained target recognition model. Therefore, it can solve the problem in the prior art of being unable to accurately identify whether a customer is a potential customer, and thus unable to accurately allocate financial resources to the customer, with poor customer satisfaction, and achieve the effects of improving prediction accuracy and enhancing customer satisfaction.

[0080] In the specific implementation process, the above device further includes a first training unit and a second training unit. The first training unit is used to obtain a feature extraction model before analyzing the customer portrait data through the target recognition model, and train the feature extraction model with the historical portrait data and the label of whether the customer corresponding to the historical portrait data is a potential customer until the loss function value of the feature extraction model is minimized. Wherein, the feature extraction model at least includes an input layer, an encoding layer, a decoding layer and an output layer; the second training unit is used to keep the structural parameters of the input layer, the encoding layer and the decoding layer unchanged, replace the output layer with the structural layer of a multi-layer perceptron model, and train until the loss function value of the multi-layer perceptron model is minimized to obtain the target recognition model. The device completes model training in two steps. After the feature extraction model is trained, the weight parameters of the input module, the encoding module and the decoding module except the output module are frozen, and the original output module is replaced with an MLP (Multi-Layer Perceptron) structure, and trained with the loss function until the loss is the lowest, which can ensure the generalization ability of the model on different customer portrait data, help capture complex patterns in the data, and enable the model to make more accurate predictions when facing new data.

[0081] Specifically, first, an initial feature extraction model is obtained, which at least includes an input layer, an encoding layer, a decoding layer, and an output layer. The input layer receives the historical portrait data of customers, including but not limited to financial data, transaction records, credit scores, etc. The encoding layer and the decoding layer are responsible for performing deep feature extraction and representation learning on the input data, which is usually based on complex neural network structures, such as the Transformer model, to capture long-term dependencies and context information in the data. Then, the feature extraction model is trained using a historical portrait dataset with labels, and the goal is to minimize the value of the loss function, which reflects the gap between the model's prediction results and the true labels. The training process may include adjusting the learning rate, batch size, regularization parameters, etc., to optimize the model performance and ensure that the model can learn the features that distinguish potential customers from non-potential customers from the data. Keeping the structural parameters unchanged and replacing the output layer: After the value of the loss function of the feature extraction model reaches the minimum, that is, the model has learned an effective feature representation, at this time, the structural parameters of the input layer, encoding layer, and decoding layer are kept unchanged, which ensures that the model's feature learning ability is retained. Then, the original output layer is replaced with the structural layer of a multi-layer perceptron network model. The MLP model consists of multiple fully connected layers and can perform non-linear transformation and combination on the extracted features to generate the final prediction result, such as whether a customer is a potential customer. Final training of the target recognition model: After replacing the output layer, the model is continued to be trained using the new MLP structural layer, and the goal is to minimize the value of the loss function of the multi-layer perceptron network model. This process may require readjusting the parameters of the MLP layer, including the number of neurons, activation function, loss function type, etc., to ensure that the model can accurately predict whether a customer has potential needs based on the representation generated by the feature extraction layer. After training is completed, the obtained target recognition model will be able to efficiently process new customer portrait data and output the probability estimate of whether a customer is a potential customer, providing data support for the customer relationship management and market expansion strategies of financial institutions.

[0082] The input layer of the model mainly includes the source sequence data embedding layer and its position encoder, and the data embedding layer and its position encoder. Among them, the data embedding layer is used to transform the digital representation in the sequence data into a vector representation to capture the dependencies between data in the high-dimensional vector space. The encoding layer is used to extract features from the input sequence data. Its structure is composed of N encoder layers stacked together. Each encoder layer consists of two sub-layer connection structures. The first sub-layer connection structure includes a multi-head self-attention sub-layer, a normalization layer, and a residual connection. The second sub-layer connection structure includes a feed-forward fully connected sub-layer, a normalization layer, and a residual connection. The decoding layer is used to represent the features of the next possible 'value' based on the output result of the encoder and the result of the previous prediction. Its structure is similar to that of the encoding module and is composed of N decoder layers stacked together. Each decoder layer consists of three sub-layer connection structures. The first sub-layer connection structure includes a multi-head self-attention sub-layer, a normalization layer, and a residual connection. The second sub-layer connection structure includes a multi-head attention sub-layer, a normalization layer, and a residual connection. The third sub-layer connection structure includes a feed-forward fully connected sub-layer, a normalization layer, and a residual connection. The output layer is a linear layer, which is responsible for performing a linear transformation on the output of the decoding module to obtain an output of a specified dimension, and plays a role in converting the output of the decoding module into a prediction result. Contrastive learning enables two similar things to continue to be similar in the representation space. By continuously optimizing the contrastive loss function, a certain similarity measure between similar samples in the representation space is increased, and the similarity measure between dissimilar samples in the representation space is decreased.

[0083] In some alternative embodiments, the first training unit includes a calculation module for calculating the loss function value of the feature extraction model using a negative example sample loss function until the loss function value of the feature extraction model is minimized. Among them, the negative example sample loss function is used to increase the difference between positive example samples and negative example samples. The positive example samples represent samples whose prediction results are characterized as potential customers, and the negative example samples represent samples that are not potential customers. By introducing and optimizing the negative example sample loss function, the device improves the ability of the model to extract features of positive and negative samples, reduces the similarity of the feature projections of positive and negative samples, and amplifies the difference between positive and negative samples to accurately extract sample features.

[0084] In the specific implementation process, when dealing with classification tasks such as bill potential customer mining, a major challenge faced by the model is to distinguish between positive and negative samples, that is, to correctly identify which customers are potential business opportunities and which are not. Traditional loss functions may perform poorly when faced with high similarity between positive and negative sample features, resulting in blurred classification boundaries and reduced classification accuracy. In order to meet this challenge, a negative sample loss function is specially introduced. The purpose of this loss function is to increase the distance between positive and negative samples in the feature space, thereby improving the classification ability of the model. Specifically, the negative sample loss function will focus on those non-potential customer samples that are misclassified as potential customers, that is, hard negative samples, when calculating. By designing a mechanism, the representation of positive samples in the feature space is as far away from the representation of negative samples as possible, which helps the model learn clearer and more discriminative feature representations, so that it can more accurately distinguish between potential customers and non-potential customers when predicting. The negative sample loss function is shown in the following formula:

[0085]

[0086] Wherein, τ is the temperature coefficient, which is used to adjust the calculated similarity; |P(i)| is the size of the set of positive samples of data sample i; z i is the output feature of the i-th sample output by the transformer model; γ is the adjustment coefficient of the regular term, represents the supervised loss function, z a represents the output feature of sample a, z p Represents the output feature of sample p. As can be seen from the formula, when there are multiple positive samples, the loss function maximizes the loss between i and all positive samples p∈P i The average similarity between the two samples brings the feature center distance of this sample closer to that of all positive samples. bound is a regular term responsible for penalizing hard negative samples (or negative samples), further reducing the average similarity between the two, and solving the hard negative problem that is difficult to optimize in contrastive training. The regular term expression is Where L is the label set in the dataset, z i represents a positive sample, z l Represents negative samples.

[0087] In some alternative embodiments, the second training unit includes a replacement module and a training module. The replacement module is configured to replace the output layer with a structural layer of a multi-layer perceptron network model having two fully-connected layers. The training module is configured to continue training the multi-layer perceptron network model with the historical portrait data and the label of whether the customer corresponding to the historical portrait data is a potential customer until the cross-entropy loss function value of the multi-layer perceptron network model is minimized. By introducing an MLP (Multi-Layer Perceptron), this device enhances the non-linear expression ability of the model, enabling the model to learn more complex decision boundaries, thereby improving the prediction accuracy of potential customers.

[0088] Specifically, the feature extraction model has learned during training how to convert the input customer portrait data into a high-dimensional feature representation, but the original output layer may simply make classification decisions based on these features. To improve the prediction ability of the model, the original output layer is replaced with an MLP here. An MLP is a feedforward neural network that includes at least two fully-connected layers. It can perform non-linear transformations on the high-dimensional feature representations generated by the feature extraction model, thereby better fitting complex data distributions and improving the classification performance of the model. Continue training to the minimum cross-entropy loss: After replacing the output layer, the model needs to be further trained to optimize the new structural layer. The historical portrait data and the corresponding label information (i.e., the classification result of whether the customer is a potential customer) used during training are used to adjust the weight parameters of the MLP by minimizing the cross-entropy loss function. The cross-entropy loss function is a commonly used metric for measuring the difference between the predicted probability distribution of the model and the true label distribution, and it is particularly effective for classification tasks. By comparing the predicted results of the model with the actual labels and adjusting the weights to minimize the cross-entropy loss, the model can more accurately predict potential customers on the training set, thereby improving the overall classification accuracy and the generalization ability of the model. The expression of the cross-entropy loss function is as follows:

[0089]

[0090] where y i represents the classification of sample i, with the positive class being 1 and the negative class being 0; p i represents the probability that sample i is predicted as the positive class, and N represents the number of samples.

[0091] In some other alternative embodiments, after determining the customer portrait data, the device further includes a noise removal unit configured to remove noise from the customer portrait data using a wavelet transform device to obtain the denoised customer portrait data. By removing noise with the wavelet transform device, this device can not only improve the data quality, optimize the feature learning and model training processes, but also significantly enhance the accuracy of financial institutions in potential customer mining.

[0092] In the specific implementation process, the wavelet transform denoising technology has good locality characteristics in both time and frequency spaces. Therefore, it can retain the low-frequency and stable characteristics of the original signal while effectively removing noise. The present invention uses wavelet transform to remove noise from sequence data, significantly improving the effectiveness of data denoising and providing a data basis for more accurate analysis and prediction of time series data. The input sequence is denoised using the wavelet function shown in the following formula. The formula of the wavelet function is as follows:

[0093]

[0094] Among them, x(t) represents the data sequence corresponding to the original customer portrait data, and ψ t is a continuous mother wavelet, first scaled to a and then translated to b. In the DWT domain, the scale factor and translation factor use discrete values. As the time series granularity increases, the scale factor increases exponentially, i.e., a = 1, 2, 4..., and the translation factor increases by integers, i.e., b = 1, 2, 3....

[0095] In some other alternative embodiments, the first determination unit includes a first generation module, a second generation module, and a third generation module. The first generation module is configured to obtain customer basic information and generate the identity characteristics of the customer according to the customer basic information, where the customer basic information includes at least name and age; the second generation module is configured to obtain customer financial information and generate the financial characteristics of the customer according to the customer financial information, where the customer financial information includes at least income and liabilities; the third generation module is configured to obtain customer transaction bill information and generate the transaction characteristics of the customer according to the customer transaction bill information, where the customer transaction bill information includes at least transaction time and transaction type.

[0096] Specifically, the customer can be an enterprise or an individual. The customer information mainly includes customer basic information, customer financial information, and customer transaction bill information. In addition, it can also include credit risk information, executive and shareholder information, and statistical characteristics of customer bill transactions. Among them, when the customer is an individual, the customer basic information includes at least name and age; when the customer is an enterprise, the customer basic information can specifically include: enterprise name, address information, enterprise nature, registered capital, total assets, operating income, holding type, enterprise scale, affiliated institution, enterprise legal person, credit granting situation, etc. In addition to income and liabilities, the customer financial information can also include: whether it is a listed company, credit rating, default probability, total liabilities, pre-tax profit, main business income, accounts receivable, total current assets and total current liabilities, etc. In addition to transaction time and transaction type, the customer transaction bill information can also include: counterparty, face value, bill acceptance bank, etc. In addition to the above customer information, it can also include: customer credit risk information, which can specifically include: risk warning code, gray list hit flag, etc. Customer executive and shareholder information can specifically include: executive or shareholder name, executive category, actual controller flag, name of shareholder-related enterprise, type of shareholder-related enterprise, shareholder shareholding ratio and association type, etc. Statistical characteristics of customer bill transactions can specifically include: average monthly bill issuance times of the customer, average monthly bill issuance amount of the customer, average monthly bill issuance time point of the customer, average monthly bill receipt times of the customer, average monthly bill receipt amount of the customer, average monthly bill receipt time point of the customer, average monthly bill pledge times of the customer, average monthly bill pledge amount of the customer, average monthly bill pledge time point of the customer, customer-signed discount type, cumulative discount amount in the current year, cumulative weighted discount rate in the current year, proportion of online banking bill receipt discounted by our bank, customer endorsement times statistics, customer stratification label, customer activity label, customer potential label, etc.

[0097] In some other alternative embodiments, before analyzing the customer portrait data through the target recognition model, the device further includes an input unit for generating a data sequence corresponding to the customer portrait data and inputting the data sequence into the target recognition model. The device generates a data sequence corresponding to the customer portrait data and inputs it into the target recognition model, which can make full use of the model's processing ability for sequence data, enhance the depth of customer feature learning, and thus achieve higher accuracy and efficiency in potential customer identification.

[0098] Specifically, first, it is necessary to integrate customer base information, financial information, transaction bill information, etc. obtained from different sources to ensure that all relevant information is taken into account. This may involve preprocessing steps such as data cleaning, format unification, and filling missing values to ensure data quality and consistency. Next, according to business requirements and model characteristics, the integrated customer information is constructed into a data sequence. During the construction process, it is important to determine the length, frequency, and element order of the sequence. For example, the historical transaction bill information of customers can be arranged in chronological order to form a time series, which helps the model capture the evolution trend and periodic characteristics of customer transaction behaviors. When constructing the sequence, features that are helpful for identifying potential customers need to be selected and encoded. For categorical variables, one-hot encoding or other encoding methods may be used to convert them into numerical representations; for numerical variables, normalization or standardization processing may be required to ensure data comparability and the training effect of the model. If the customer portrait data contains a large amount of information, it may be necessary to divide it into multiple sequences, each sequence focusing on different customer attributes or transaction types, which helps the model learn customer characteristics from different perspectives and improve the accuracy and comprehensiveness of prediction.

[0099] The financial resource allocation device includes a processor and a memory. The above-mentioned first determination unit, analysis unit, second determination unit, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to implement corresponding functions. The above-mentioned modules are all located in the same processor; or, the above-mentioned each module is located in different processors in any combination form.

[0100] The processor contains a kernel, and the kernel retrieves the corresponding program unit from the memory. One or more kernels can be set, and by adjusting the kernel parameters, it can accurately identify whether a customer is a potential customer.

[0101] The memory may include non-permanent memory in a computer-readable medium, forms such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM), and the memory includes at least one storage chip.

[0102] An embodiment of the present invention provides a computer-readable storage medium, and the computer-readable storage medium includes a stored program. Wherein, when the program runs, it controls the device where the computer-readable storage medium is located to execute the financial resource allocation method.

[0103] Specifically, the financial resource allocation method includes:

[0104] Step S201, determine customer portrait data, where the customer portrait data is a data set representing different characteristics of customers;

[0105] Step S202, analyze the customer portrait data through a target recognition model to obtain a prediction result, where the prediction result indicates whether the customer is a potential customer. The target recognition model is obtained through machine learning training based on multiple sets of data, and each set of data in the multiple sets of data includes: historical portrait data and a label of the result indicating whether the customer corresponding to the historical portrait data is a potential customer;

[0106] Step S203, when the prediction result is greater than a preset threshold, determine that the customer is the potential customer and configure corresponding financial resources for the potential customer, where the financial resources at least include financial products.

[0107] An embodiment of the present invention provides an electronic device, including a processor, a memory, and a program stored on the memory and executable on the processor. When the processor executes the program, at least the following steps are implemented:

[0108] Step S201, determine customer portrait data, where the customer portrait data is a data set representing different characteristics of customers;

[0109] Step S202, analyze the customer portrait data through a target recognition model to obtain a prediction result, where the prediction result indicates whether the customer is a potential customer. The target recognition model is obtained through machine learning training based on multiple sets of data, and each set of data in the multiple sets of data includes: historical portrait data and a label of the result indicating whether the customer corresponding to the historical portrait data is a potential customer;

[0110] Step S203, when the prediction result is greater than a preset threshold, determine that the customer is the potential customer and configure corresponding financial resources for the potential customer, where the financial resources at least include financial products.

[0111] The device in this article can be a server, a PC, a PAD, a mobile phone, etc.

[0112] The present application also provides a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the methods in the various embodiments of the present application are implemented:

[0113] Step S201, determine customer portrait data, where the customer portrait data is a data set representing different characteristics of customers;

[0114] Step S202: Analyze the customer portrait data through the target recognition model to obtain a prediction result, where the prediction result indicates whether the customer is a potential customer. The target recognition model is obtained through machine learning training based on multiple sets of data, and each set of data in the multiple sets of data includes: historical portrait data and a label of the result indicating whether the customer corresponding to the historical portrait data is a potential customer.

[0115] Step S203: When the prediction result is greater than a preset threshold, determine that the customer is the potential customer and configure corresponding financial resources for the potential customer, where the financial resources at least include financial products.

[0116] Obviously, those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computing device. They can be concentrated on a single computing device or distributed on a network composed of multiple computing devices. They can be implemented by program codes executable by the computing device. Thus, they can be stored in a storage device and executed by the computing device. And in some cases, the steps shown or described herein can be executed in a different order, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module for implementation. In this way, the present invention is not limited to any specific combination of hardware and software.

[0117] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0118] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one or more flows or multiple flows and / or blocks Figure 1 one or more blocks or multiple blocks.

[0119] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means that implement the function specified in one or more of the blocks and / or processes. Figure 1 one or more of the processes and / or blocks Figure 1 specified in the one or more of the blocks and / or processes.

[0120] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable apparatus provide steps for implementing the function specified in one or more of the processes and / or blocks. Figure 1 one or more of the processes and / or blocks Figure 1 specified in the one or more of the blocks and / or processes.

[0121] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0122] The memory may include non-permanent memory in the computer-readable medium, random access memory (RAM) and / or non-volatile memory such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of a computer-readable medium.

[0123] Computer-readable media includes both permanent and non-permanent, removable and non-removable media implemented by any method or technology for storage of information such as computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0124] It should also be noted that the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, commodity or device comprising a series of elements not only includes those elements but also other elements not expressly listed, or elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, commodity or device comprising the element.

[0125] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0126] 1) In the financial resource allocation method of the present application, customer portrait data is determined, analyzed by a target recognition model to obtain a prediction result. When the prediction result is greater than a preset threshold, the customer is determined as a potential customer, and corresponding financial resources are allocated to the potential customer. The target recognition model is trained based on historical portrait data and labels of whether the customers corresponding to the historical portrait data are potential customers. The target recognition model determines whether a customer is a potential customer through feature recognition to accurately allocate corresponding financial resources to the customer and improve customer satisfaction. Compared with the prior art where it is impossible to accurately identify whether a customer is a potential customer, and thus impossible to accurately allocate financial resources to the customer, resulting in poor customer satisfaction, the present application predicts whether the customer corresponding to the customer portrait data is a potential customer through the trained target recognition model. Therefore, it can solve the problem in the prior art of being unable to accurately identify whether a customer is a potential customer, and thus unable to accurately allocate financial resources to the customer, resulting in poor customer satisfaction, achieving the effects of improving prediction accuracy and enhancing customer satisfaction.

[0127] 2) In the financial resource allocation device of the present application, customer portrait data is determined, analyzed by a target recognition model to obtain a prediction result. When the prediction result is greater than a preset threshold, the customer is determined as a potential customer, and corresponding financial resources are allocated to the potential customer. The target recognition model is trained based on historical portrait data and labels of whether the customers corresponding to the historical portrait data are potential customers. The target recognition model determines whether a customer is a potential customer through feature recognition to accurately allocate corresponding financial resources to the customer and improve customer satisfaction. Compared with the prior art where it is impossible to accurately identify whether a customer is a potential customer, and thus impossible to accurately allocate financial resources to the customer, resulting in poor customer satisfaction, the present application predicts whether the customer corresponding to the customer portrait data is a potential customer through the trained target recognition model. Therefore, it can solve the problem in the prior art of being unable to accurately identify whether a customer is a potential customer, and thus unable to accurately allocate financial resources to the customer, resulting in poor customer satisfaction, achieving the effects of improving prediction accuracy and enhancing customer satisfaction.

[0128] The above are only the preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A method for allocating financial resources, characterized in that: include: Determine customer profile data, wherein the customer profile data is a data set representing different characteristics of the customer; The customer portrait data is analyzed by a target recognition model to obtain a prediction result, wherein the prediction result indicates whether the customer is a potential customer, and the target recognition model is obtained by machine learning training based on multiple sets of data, and each set of data in the multiple sets of data includes: historical portrait data and a label indicating whether the customer corresponding to the historical portrait data is a potential customer; When the prediction result is greater than a preset threshold, the customer is determined to be the potential customer, and corresponding financial resources are configured for the potential customer, wherein the financial resources at least include financial products.

2. The method for allocating financial resources according to claim 1, characterized in that: Before analyzing the customer portrait data by the target recognition model, the method further includes: Obtain a feature extraction model, and train the feature extraction model using the historical profile data and a label indicating whether the customer corresponding to the historical profile data is a potential customer, until the loss function value of the feature extraction model is minimized, wherein the feature extraction model includes at least an input layer, an encoding layer, a decoding layer, and an output layer; The structural parameters of the input layer, the structural parameters of the encoding layer and the structural parameters of the decoding layer remain unchanged, the output layer is replaced by the structural layer of the multi-layer perception network model, and the multi-layer perception network model is trained until the loss function value is minimized to obtain the target recognition model.

3. The method for allocating financial resources according to claim 2, characterized in that: The feature extraction model is trained by using the historical profile data and the label of the result of whether the customer corresponding to the historical profile data is a potential customer until the loss function value of the feature extraction model is minimized, including: A negative sample loss function is used to calculate the loss function value of the feature extraction model until the loss function value of the feature extraction model is minimized, wherein the negative sample loss function is used to increase the difference between positive samples and negative samples, the positive samples represent samples whose prediction results are characterized as potential customers, and the negative samples represent samples that are not potential customers.

4. The method for allocating financial resources according to claim 2, characterized in that: The output layer is replaced by a structural layer of a multi-layer perception network model, and training is performed until the loss function value of the multi-layer perception network model is minimized, including: The output layer is replaced by a structure layer of a multi-layer perception network model having two fully connected layers; Continue to train the multi-layer perception network model using the historical portrait data and labels of the results of whether the customer corresponding to the historical portrait data is a potential customer, until the cross entropy loss function value of the multi-layer perception network model is minimized.

5. The method for allocating financial resources according to claim 1, characterized in that: After determining the customer portrait data, the method further includes: The wavelet transform method is used to remove noise from the customer portrait data to obtain the denoised customer portrait data.

6. The method for allocating financial resources according to claim 1, characterized in that: Determine customer profile data, including: Acquire basic customer information, and generate identity features of the customer based on the basic customer information, wherein the basic customer information includes at least name and age; Acquire financial information of a customer, and generate financial characteristics of the customer based on the financial information of the customer, wherein the financial information of the customer includes at least income and liabilities; Acquire customer transaction ticket information, and generate the customer's transaction features based on the customer transaction ticket information, wherein the customer transaction ticket information at least includes transaction time and transaction type.

7. The method for allocating financial resources according to claim 1, characterized in that: Before analyzing the customer portrait data by the target recognition model, the method further includes: Generate a data sequence corresponding to the customer portrait data, and input the data sequence into the target recognition model.

8. A financial resource allocation device, characterized in that: include: A first determining unit, configured to determine customer portrait data, wherein the customer portrait data is a data set representing different characteristics of a customer; An analysis unit is used to analyze the customer portrait data through a target recognition model to obtain a prediction result, wherein the prediction result indicates whether the customer is a potential customer, and the target recognition model is obtained through machine learning training based on multiple sets of data, and each set of data in the multiple sets of data includes: historical portrait data and a label indicating whether the customer corresponding to the historical portrait data is a potential customer; The second determination unit is used to determine that the customer is the potential customer when the prediction result is greater than a preset threshold, and configure corresponding financial resources for the potential customer, wherein the financial resources at least include financial products.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method for configuring financial resources according to any one of claims 1 to 7.

10. An electronic device, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include a method for configuring financial resources according to any one of claims 1 to 7.

Images