Risk level prediction model training method and financial product recommendation method

Abstract

The invention discloses a risk level prediction model training method and a financial product recommendation method. The method comprises the following steps: acquiring financial product data, and converting the financial product data into a product representation vector; acquiring purchase behavior data of a customer for financial products and basic purchase preferences of the customer, and converting the basic purchase preferences into customer basic representation vectors; determining a long-term preference vector and a short-term preference vector of the customer based on the customer basic representation vector, the product representation vector and the purchase behavior data, and determining a purchase preference vector of the customer for the financial product based on the long-term preference vector and the short-term preference vector; constructing a training data set based on the product representation vector and the purchase preference vector of the customer for the financial product; and training the graph neural network model based on the training data set to obtain a risk level prediction model. According to the method, the accuracy of risk level prediction of the financial products can be improved by capturing long-term and short-term purchase preferences of customers, so that the accuracy of financial product recommendation is improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a training method for a risk level prediction model and a method for recommending financial products. Background Art

[0002] As major banks develop their businesses, they are launching more and more financial products. In order to select appropriate recommendations for customers among these financial products, this patent invents a recommendation method based on multi-feature fusion and customers' long-term and short-term preferences to predict customers' preferences for various financial products.

[0003] Currently, financial product recommendation methods rely on neural networks, such as graph neural network models, to predict the scores of financial products. However, these methods fail to fully utilize the relevant information between financial products and customers, and fail to consider that customers' long-term and short-term preferences change over time, resulting in insufficiently accurate recommendations. Summary of the Invention

[0004] The present invention provides a training method for a risk level prediction model and a financial product recommendation method to solve the problem that customers' long-term and short-term preferences change over time.

[0005] According to one aspect of the present invention, a method for training a risk level prediction model is provided, comprising:

[0006] Acquiring financial product data and converting the financial product data into a product representation vector; wherein the product representation vector embeds multiple features of the financial product;

[0007] Obtaining customer purchasing behavior data for financial products and the customer's basic purchasing preferences, and converting the basic purchasing preferences into a customer base representation vector; the basic purchasing preferences are obtained by desensitizing the customer's responses to the most recent financial questionnaire;

[0008] Determining a customer's long-term preference vector and a short-term preference vector based on the customer base representation vector, the product representation vector, and the purchase behavior data, and determining the customer's purchase preference vector for financial products based on the long-term preference vector and the short-term preference vector;

[0009] constructing a training dataset based on the product representation vector and the customer's purchase preference vector for financial products;

[0010] The graph neural network model is trained based on the training data set to obtain a risk level prediction model.

[0011] According to another aspect of the present invention, a method for recommending financial products is provided, characterized by comprising:

[0012] Acquiring the customer's purchase behavior data and financial product data of at least one financial product to be recommended, and converting the purchase behavior data and the financial product data into a product representation vector and a customer's purchase preference vector for the financial product;

[0013] Inputting the product representation vector and the purchase preference vector into a risk level prediction model for prediction to obtain a risk level prediction result; wherein the risk level prediction model is trained using the risk level prediction model training method described in any embodiment of the present invention;

[0014] A target financial product is determined based on the risk level prediction result.

[0015] According to another aspect of the present invention, a training device for a risk level prediction model is provided, characterized in that it includes:

[0016] A product representation vector determination module, configured to obtain financial product data and convert the financial product data into a product representation vector; the product representation vector embeds multiple features of the financial product;

[0017] A customer base representation vector determination module is configured to obtain customer purchasing behavior data for financial products and the customer's basic purchasing preferences, and convert the basic purchasing preferences into a customer base representation vector; the basic purchasing preferences are obtained by desensitizing the customer's responses to the most recent financial questionnaire;

[0018] a purchase preference vector determination module, configured to determine a customer's long-term preference vector and short-term preference vector based on the customer base representation vector, the product representation vector, and the purchase behavior data, and determine the customer's purchase preference vector for financial products based on the long-term preference vector and the short-term preference vector;

[0019] A training data set construction module, configured to construct a training data set based on the product representation vector and the customer's purchase preference vector for financial products;

[0020] The risk level prediction model training module is used to train the graph neural network model based on the training data set to obtain a risk level prediction model.

[0021] According to another aspect of the present invention, a financial product recommendation device is provided, characterized by comprising:

[0022] a data acquisition module, configured to acquire the customer's purchase behavior data and financial product data of at least one financial product to be recommended, and convert the purchase behavior data and financial product data into a product representation vector and a customer's purchase preference vector for the financial product;

[0023] a risk level prediction result prediction module, configured to input the product representation vector and the purchase preference vector into a risk level prediction model for prediction, thereby obtaining a risk level prediction result; wherein the risk level prediction model is trained using the risk level prediction model training method described in an embodiment of the present invention;

[0024] The target financial product determination module is used to determine the target financial product based on the risk level prediction result.

[0025] According to another aspect of the present invention, an electronic device is provided, comprising:

[0026] at least one processor; and

[0027] a memory communicatively connected to the at least one processor; wherein,

[0028] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the risk level prediction model training method described in any embodiment of the present invention, and / or the financial product recommendation method described in any embodiment of the present invention.

[0029] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the risk level prediction model training method described in any embodiment of the present invention, and / or the financial product recommendation method described in any embodiment of the present invention when executed.

[0030] The technical solution of the embodiment of the present invention obtains financial product data and converts the financial product data into a product representation vector; obtains customer purchase behavior data for financial products and the customer's basic purchase preferences and converts the basic purchase preferences into a customer base representation vector; determines the customer's long-term preference vector and short-term preference vector based on the customer base representation vector, the product representation vector, and the purchase behavior data, and determines the customer's purchase preference vector for financial products based on the long-term preference vector and the short-term preference vector; constructs a training data set based on the product representation vector and the customer's purchase preference vector for financial products; and trains a graph neural network model based on the training data set to obtain a risk level prediction model. By capturing the customer's long-term and short-term purchase preferences, the present invention obtains the customer's purchase preference vector for financial products, can capture changes in the customer's purchase preferences over time, and can improve the accuracy of the prediction of the risk level of financial products, thereby improving the accuracy of financial product recommendations.

[0031] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 This is a flow chart of a method for training a risk level prediction model provided in Example 1 of the present invention;

[0034] Figure 2 This is a schematic diagram of converting financial product data into a product representation vector provided by the first embodiment of the present invention;

[0035] Figure 3 Schematic diagram of a method for expressing customer preferences based on long-term and short-term preferences provided in the first embodiment of the present invention;

[0036] Figure 4 This is a flowchart of a method for recommending financial products provided in Example 2 of the present invention;

[0037] Figure 5 This is a schematic diagram of the structure of a training device for a risk level prediction model provided by the third embodiment of the present invention;

[0038] Figure 6 This is a schematic diagram of the structure of a training device for a risk level prediction model provided by a fourth embodiment of the present invention;

[0039] Figure 7 This is a structural diagram of an electronic device provided in Example 5 of the present invention. DETAILED DESCRIPTION

[0040] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0041] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0042] Example 1

[0043] Figure 1 This is a flow chart of a method for training a risk level prediction model provided by the first embodiment of the present invention. This embodiment is applicable to training a risk level prediction model for a financial product, predicting the risk level of the financial product based on the risk level prediction model, and thus recommending the financial product based on the risk level of the financial product. This method can be executed by a training device for a risk level prediction model. The training device for the risk level prediction model can be implemented in the form of hardware and / or software. The training device for the risk level prediction model can be configured in electronic devices such as computers and servers. Figure 1 As shown, the method includes:

[0044] S110: Acquire financial product data, and convert the financial product data into a product representation vector; the product representation vector embeds multiple features of the financial product.

[0045] Financial product data refers to data related to financial products provided by financial institutions to their clients. Specifically, financial product data includes one or more of the following: a description of the financial product, its term, expected rate of return, and risk level. The description of the financial product is a text introducing the product. The term of the product and the risk level are discrete features, and the risk level is divided into five levels: stable, steady, balanced, aggressive, and radical. Expected rate of return is a continuous feature. In an embodiment of the present invention, financial product data is obtained, multiple features in the financial product data are integrated, and the financial product data is converted into a product representation vector.

[0046] Based on the above embodiment, optionally, the financial product data is converted into a product representation vector, including: performing word segmentation on the description text to obtain a word sequence, and converting the word sequence into a word vector; converting the word vector into a word context representation vector that captures context features through a convolutional network; assigning different attention weights to different words through attention weighting based on a word-level attention network to obtain a text representation vector; and / or, embedding the product term into a product term vector through an embedding layer; and / or, converting the expected rate of return into discrete features, and embedding the discrete features into an expected rate of return vector through an embedding layer; and / or, embedding the risk level into a risk level vector through an embedding layer; and determining the product representation vector based on one or more of the text vector, product term vector, expected rate of return vector, and risk level vector.

[0047] Figure 2 This is a schematic diagram of converting financial product data into a product representation vector provided by the first embodiment of the present invention. In this embodiment of the present invention, Figure 2 As shown, for the description text of financial products, the description text of financial products is segmented, and the word sequence can be recorded as {w1, w2, ..., w n}. Further, through word embedding, the word sequence is converted into a low-dimensional word vector. Specifically, during the conversion process, by querying the word embedding table The words in the word sequence are converted into word vectors, recorded as Where n is the number of words in the sequence, V is the number of words in the embedding table, and D is the embedding dimension. Further, through the convolutional network, the word vector is converted into a word context representation vector that captures the context features, denoted as Specifically, the local context features of the word are captured through convolution operation, and the convolution operation is set as the inner product between the convolution kernel and the word vector. i w i The context representation vector of is calculated as follows:

[0048] c i =ReLU(M c e i：i+l-1 +b c );

[0049] Among them, ReLU represents the activation function, e i：i+l-1 Represents the concatenation of l word vectors, Represent the convolution kernel and bias term respectively, which are trainable parameters in the convolutional network model (such as the CNN model), N f represents the number of convolution kernels, and l represents the convolution kernel window size.

[0050] Furthermore, based on the word-level attention network, different words are given different degrees of importance through attention weighting and a unified text representation vector is formed. Specifically, a fixed query vector is randomly initialized, and a fully connected network with a single hidden layer is used to calculate each context representation vector c i The attention weight relative to the query vector. i The attention weight is α i , the calculation formula is as follows:

[0051]

[0052] in, is a trainable parameter of the model, and q represents the query vector of the word-level attention network, and Q is the dimension of the query vector of the word-level attention network.

[0053] Based on the weighted sum of the word context representation vector and the word attention weight, the text representation vector is obtained. The calculation formula is as follows:

[0054]

[0055] Among them, r T Represents a text representation vector.

[0056] In the embodiment of the present invention, for discrete features such as product term and risk level, the product term is embedded into a product term vector through the embedding layer, which can be expressed as E P ; The risk level is embedded into a risk level vector through the embedding layer, which can be expressed as E R For continuous features such as expected rate of return, they are converted into discrete features according to the range of expected rate of return, and then embedded into vector E through the embedding layer. A .in,

[0057] Furthermore, a product feature-level attention network is used to assign different attention weights to different features of financial products through attention weighting to obtain product representation vectors. Specifically, a fixed query vector is randomly initialized, and a fully connected network with a single hidden layer is used to calculate each feature representation vector ei∈{r T , E P , E R , E A} is the attention weight relative to the query vector. i The attention weight is α′ i , the attention weight calculation process is as follows:

[0058]

[0059] in, is a trainable parameter of the model, and q′ represents the query vector of the product feature-level attention network, and Q′ is the dimension of the query vector of the product feature-level attention network.

[0060] The product representation vector is obtained based on the weighted sum of the feature context representation vector and the attention weight. The calculation formula is as follows:

[0061]

[0062] Among them, E i The product representation vector representing financial product i.

[0063] S120. Obtain the customer's purchasing behavior data for financial products and the customer's basic purchasing preferences, and convert the basic purchasing preferences into a customer base representation vector; the basic purchasing preferences are obtained by desensitizing the customer's responses to the most recent financial management questionnaire.

[0064] Among them, basic purchasing preferences refer to the basic and relatively stable preference characteristics shown by customers when purchasing financial products. In an embodiment of the present invention, the customer's responses in the most recent financial questionnaire are desensitized, and the customer's preference characteristics for financial products are extracted as basic purchasing preferences, and the basic purchasing preferences are converted into customer basic representation vectors. The customer basic representation vector is a vector representation of the basic purchasing preferences. Specifically, for discrete features in the basic purchasing preferences, embedding layer is used for embedding. For continuous features in the basic purchasing preferences, the continuous features are divided into different intervals and then embedded using embedding layer. Furthermore, a customer feature-level attention network is used to assign different weights to different features through attention weighting to form a customer basic representation vector.

[0065] S130. Determine a long-term preference vector and a short-term preference vector of the customer based on the customer base representation vector, the product representation vector, and the purchase behavior data, and determine a purchase preference vector of the customer for financial products based on the long-term preference vector and the short-term preference vector.

[0066] Among them, the purchase behavior data is time series data, specifically the click data on preferred financial products at different times, and the dynamic preferences of customers that change over time are determined based on the purchase behavior data of customers at different times. The long-term preference vector is used to capture the customer's relatively stable financial product purchase preference within a preset time range. The long-term preference vector reflects the customer's long-term financial management habits and investment strategy, which is of great significance for understanding the customer's overall investment style and risk tolerance. The short-term preference vector focuses on capturing the customer's financial product purchase preference at the current moment. The short-term preference vector can quickly capture the customer's latest investment intentions and demand changes, which is crucial for timely adjusting the recommendation strategy and providing targeted financial product recommendations. In an embodiment of the present invention, the customer's long-term preference vector and short-term preference vector are determined based on the customer base representation vector, the product representation vector and the purchase behavior data, and the customer's purchase preference vector for financial products is determined based on the long-term preference vector and the short-term preference vector. Specifically, the long-term and short-term customer purchase preference representations are spliced ​​to obtain the purchase preference vector, which can be expressed as E u =[r S :r L ],r S Represents short-term customer purchasing preferences L Purchase preference representation for long-term customers.

[0067] Based on the above embodiment, optionally, the determining of the customer's short-term preference vector based on the customer base representation vector, the product representation vector and the purchase behavior data includes: determining the customer preference vector representation based on the base representation vector and the product representation vector; modeling the customer's sequential preference from the purchase behavior data based on a gated recurrent unit network and the customer preference vector representation to obtain the customer's short-term preference vector.

[0068] In the embodiment of the present invention, the customer preference vector representation is obtained by concatenating the basic representation vector and the product representation vector, and the customer preference vector representation is used to model the customer's sequential preference from the purchase behavior data using the gated recurrent unit network and the customer preference vector representation to capture the temporal dependency and obtain the customer's short-term preference vector. Specifically, Figure 3 Schematic diagram of the method for expressing customer preferences based on long-term and short-term preferences provided by the first embodiment of the present invention. Figure 3 As shown in Figure 1, in order to capture the user's dynamic preferences that change over time, the present invention uses a gated recurrent unit network to model the user's sequential preferences from historical purchase behavior data. The gated recurrent unit network is good at processing data with time series characteristics. It combines the current moment's input with the previous moment's hidden state output, and after a specific calculation, obtains the current moment's hidden state output. This calculation process will be repeated, and by resetting the gate x t and update gate z tTo control the amount and direction of information flow. The specific calculation formula involved at each moment is as follows:

[0069] x t =σ(M1[h t-1 +E t ]);

[0070] z t =σ(M2[h t-1 +E t ]);

[0071]

[0072] Among them, E t It is the vector representation of customer u’s financial product purchase preference at time t and is the input of the gated recurrent unit network at time t. t-1 represents the hidden state output of the gated recurrent unit network at time t-1, that is, the customer's financial product purchase preference at the previous moment, ⊙ represents element-by-element multiplication, σ represents the sigmoid nonlinear activation function, is the candidate state of the gated recurrent unit network at time t, h t is the hidden state output of the gated recurrent unit network at time t, M1, M2 and M3 represent the trainable parameters of the gated recurrent unit network. m represents the customer's preference for purchasing financial products at the current moment, so h m As the customer's short-term preference vector, denoted as Here, H represents the latent vector dimension.

[0073] Based on the above embodiment, optionally, determining the customer's long-term preference vector based on the customer base representation vector, the product representation vector and the purchase behavior data includes: introducing a personalized attention network to assign different weight coefficients to each purchase behavior; and determining the long-term preference vector based on the short-term preference vector and the weight coefficient of the purchase behavior corresponding to the short-term preference vector.

[0074] Short-term purchasing preferences focus on the customer's current preferences and fail to fully consider the customer's long-term preferences. Figure 3 As shown, the present invention is in {h1, h2, ..., h m}, an attention network is introduced to assign different importance weights to each purchase behavior. When obtaining the product representation vector and customer base representation, the query vector is randomly initialized and trained together with other network parameters. Here, the query vector of the attention network is set to the customer's purchase preference representation at the last moment h m Therefore, the purchase preference vector at each other moment will be the same as h mPerform a similarity evaluation and normalize it through sigmoid to obtain the corresponding weight coefficient. The weight coefficient calculation formula is as follows:

[0075]

[0076] Among them, V j and v j and Both are projection matrices and are trainable parameters of the model.

[0077] Based on the weighted sum of the purchase preference and weight coefficient at each moment, the customer's long-term preference vector is obtained. The calculation formula of the long-term preference vector is as follows:

[0078]

[0079] Among them, r L represents the customer's long-term preference vector.

[0080] In some embodiments, optionally, the method further includes: performing prediction based on the product representation vector and the customer's purchase preference vector for the financial product to obtain a click-through rate score of the customer for the financial product.

[0081] In this embodiment of the present invention, the inner product of the product representation vector and the customer's purchase preference vector for the financial product is performed to obtain the customer's click-through rate score for the financial product. For example, the calculation formula for the click-through rate score is as follows:

[0082]

[0083] S140: Construct a training dataset based on the product representation vector and the customer's purchase preference vector for financial products;

[0084] In this embodiment of the present invention, training data is constructed based on the product representation vector and the customer's financial product purchase preference vector. The constructed training data is divided into a training set, a validation set, and a test set according to a preset ratio. The prediction ratio may be training set: validation set: test set = 7:2:1.

[0085] S150. Train the graph neural network model based on the training data set to obtain a risk level prediction model.

[0086] In this embodiment of the present invention, a risk level prediction model for financial products purchased by customers is trained using a training set, and then validated using a validation set after each round of training. Training is terminated when the model's accuracy on the validation set fails to improve further or decreases, resulting in a final risk level prediction model.

[0087] In some embodiments, because the ratio of positive to negative samples is often highly unbalanced, a negative sampling strategy is used for model training. For each financial product purchased by a customer (i.e., a positive sample), R financial products not purchased by the customer are randomly sampled (i.e., negative samples). The click-through rate prediction problem is then reformulated as an R+1 classification task.

[0088] remember and The click rate scores of the positive sample and R negative samples are respectively, and the cross entropy loss function is used to optimize the classification problem. First, these click probabilities are softmax normalized to calculate the posterior click probability of the positive sample. The present invention performs softmax normalization on these click probabilities to calculate the posterior click probability of the positive sample. The calculation formula of the posterior click probability is as follows:

[0089]

[0090] Among them, p i represents the posterior click probability of the i-th positive sample, represents the click-through rate score of the i-th positive sample, represents the click-through rate score of the jth negative sample among R negative samples.

[0091] Furthermore, the negative log-likelihood of all positive samples is used as the final loss function, and the formula is as follows:

[0092]

[0093] Among them, Pos represents the set of all positive samples.

[0094] The technical solution of this embodiment obtains financial product data and converts the financial product data into a product representation vector; obtains customer purchase behavior data for financial products and the customer's basic purchase preferences and converts the basic purchase preferences into a customer base representation vector; determines the customer's long-term preference vector and short-term preference vector based on the customer base representation vector, the product representation vector, and the purchase behavior data, and determines the customer's purchase preference vector for financial products based on the long-term preference vector and the short-term preference vector; constructs a training data set based on the product representation vector and the customer's purchase preference vector for financial products; and trains a graph neural network model based on the training data set to obtain a risk level prediction model. By capturing the customer's long-term and short-term purchase preferences, the present invention obtains the customer's purchase preference vector for financial products, can capture changes in the customer's purchase preferences over time, and can improve the accuracy of the prediction of the risk level of financial products, thereby improving the accuracy of financial product recommendations.

[0095] Example 2

[0096] Figure 4This is a flow chart of a method for recommending financial products provided by the second embodiment of the present invention. Figure 4 As shown, the method includes:

[0097] S210 obtains the customer's purchase behavior data and financial product data of at least one financial product to be recommended, and converts the purchase behavior data and financial product data into a product representation vector and a customer's purchase preference vector for the financial product.

[0098] S220. Input the product representation vector and the purchase preference vector into a risk level prediction model for prediction to obtain a risk level prediction result; wherein, the risk level prediction model is trained using the risk level prediction model training method described in any embodiment of the present invention.

[0099] S230: Determine a target financial product based on the risk level prediction result.

[0100] The financial products to be recommended refer to financial products to be recommended to customers, such as wealth management products. In an embodiment of the present invention, the customer's purchasing behavior data and financial product data of at least one financial product to be recommended are obtained, and the purchasing behavior data and financial product data are converted into a product representation vector and a customer's purchase preference vector for the financial product. The product representation vector and the purchase preference vector are input into a risk level prediction model for prediction, thereby obtaining a risk level prediction result; and a target financial product is determined based on the risk level prediction result. The risk level prediction result refers to the predicted risk level of the financial product to be recommended. The risk level prediction result can be used to screen the target financial product from the financial products to be recommended, i.e., the financial product recommended to the customer.

[0101] It should be noted that the conversion process between the product representation vector and the customer's financial product purchase preference vector is the same as that in the first embodiment and will not be described in detail here.

[0102] Example 3

[0103] Figure 5 This is a schematic diagram of the structure of a training device for a risk level prediction model provided by the third embodiment of the present invention. Figure 5 As shown, the device includes:

[0104] A product representation vector determination module 310 is configured to obtain financial product data and convert the financial product data into a product representation vector; the product representation vector embeds multiple features of the financial product;

[0105] The customer base representation vector determination module 320 is configured to obtain customer purchase behavior data for financial products and the customer's basic purchasing preferences, and convert the basic purchasing preferences into a customer base representation vector; the basic purchasing preferences are obtained by desensitizing the customer's responses to the most recent financial questionnaire;

[0106] a purchase preference vector determination module 330 configured to determine a customer's long-term preference vector and short-term preference vector based on the customer base representation vector, the product representation vector, and the purchase behavior data, and determine the customer's purchase preference vector for financial products based on the long-term preference vector and the short-term preference vector;

[0107] A training data set construction module 340 is configured to construct a training data set based on the product representation vector and the customer's purchase preference vector for financial products;

[0108] The risk level prediction model training module 350 is used to train the graph neural network model based on the training data set to obtain a risk level prediction model.

[0109] The technical solution of this embodiment obtains financial product data and converts the financial product data into a product representation vector; obtains customer purchase behavior data for financial products and the customer's basic purchase preferences and converts the basic purchase preferences into a customer base representation vector; determines the customer's long-term preference vector and short-term preference vector based on the customer base representation vector, the product representation vector, and the purchase behavior data, and determines the customer's purchase preference vector for financial products based on the long-term preference vector and the short-term preference vector; constructs a training data set based on the product representation vector and the customer's purchase preference vector for financial products; and trains a graph neural network model based on the training data set to obtain a risk level prediction model. By capturing the customer's long-term and short-term purchase preferences, the present invention obtains the customer's purchase preference vector for financial products, can capture changes in the customer's purchase preferences over time, and can improve the accuracy of the prediction of the risk level of financial products, thereby improving the accuracy of financial product recommendations.

[0110] Based on the above embodiment, optionally, the financial product data includes one or more of a description text of the financial product, a product term, an expected rate of return, and a risk level;

[0111] The product representation vector determination module 310 is used to:

[0112] The description text is segmented to obtain a word sequence, and the word sequence is converted into a word vector; the word vector is converted into a word context representation vector that captures context features through a convolutional network; based on a word-level attention network, different attention weights are assigned to different words through attention weighting to obtain a text representation vector;

[0113] and / or, embedding the product term into a product term vector through an embedding layer;

[0114] and / or, converting the expected rate of return into discrete features, and embedding the discrete features into an expected rate of return vector through an embedding layer;

[0115] and / or, embedding the risk level into a risk level vector via an embedding layer;

[0116] A product representation vector is determined based on one or more of the text vector, product term vector, expected return vector, and risk level vector.

[0117] Based on the above embodiment, optionally, the purchase preference vector determination module 330 includes a short-term preference vector determination unit configured to:

[0118] A customer preference vector representation is determined based on the basic representation vector and the product representation vector; and a customer's sequential preference is modeled from the purchase behavior data based on a gated recurrent unit network and the customer preference vector representation to obtain a customer's short-term preference vector.

[0119] Based on the above embodiment, optionally, the purchase preference vector determination module 330 includes a long-term preference vector determination unit configured to:

[0120] Introducing a personalized attention network to assign different weight coefficients to each purchase behavior;

[0121] A long-term purchase preference vector is determined based on the short-term preference vector and a weight coefficient of the purchase behavior corresponding to the short-term preference vector.

[0122] Based on the above embodiment, optionally, the device further includes a click-through rate score determination module, which is used to make a prediction based on the product representation vector and the customer's purchase preference vector for the financial product to obtain the customer's click-through rate score for the financial product.

[0123] The training device for the risk level prediction model provided in the embodiment of the present invention can execute the training method for the risk level prediction model provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0124] Example 4

[0125] Figure 6 This is a schematic diagram of the structure of a training device for a risk level prediction model provided by the fourth embodiment of the present invention. Figure 6 As shown, the device includes:

[0126] A data acquisition module 410 is configured to acquire a customer's purchase behavior data and financial product data of at least one financial product to be recommended, and convert the purchase behavior data and financial product data into a product representation vector and a customer's purchase preference vector for the financial product;

[0127] a risk level prediction result prediction module 420, configured to input the product representation vector and the purchase preference vector into a risk level prediction model for prediction, thereby obtaining a risk level prediction result; wherein the risk level prediction model is trained based on the risk level prediction model training method described in claims 1-5;

[0128] The target financial product determination module 430 is configured to determine a target financial product based on the risk level prediction result.

[0129] The financial product recommendation device provided in the embodiment of the present invention can execute the financial product recommendation method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0130] Example 5

[0131] Figure 7 1 is a structural diagram of an electronic device provided in Example 5 of the present invention. The electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0132] like Figure 7 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11. The memory stores a computer program that can be executed by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. Various programs and data required for the operation of the electronic device 10 can also be stored in the RAM 13. The processor 11, ROM 12, and RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0133] Multiple components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0134] Processor 11 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors that run machine learning model algorithms, digital signal processors (DSPs), and any other suitable processors, controllers, microcontrollers, etc. Processor 11 executes the various methods and processes described above, such as the risk level prediction model training method and / or the financial product recommendation method.

[0135] In some embodiments, the training method of the risk level prediction model and / or the method of recommending financial products may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the training method of the risk level prediction model and / or the method of recommending financial products described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute the training method of the risk level prediction model and / or the method of recommending financial products in any other appropriate manner (for example, by means of firmware).

[0136] Various embodiments of the systems and techniques described above can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0137] Computer programs for implementing the risk level prediction model training method and / or financial product recommendation method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer programs can be executed entirely on the machine, partially on the machine, as a standalone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0138] Example 6

[0139] Embodiment 6 of the present invention further provides a computer-readable storage medium storing computer instructions, the computer instructions being used to cause a processor to execute a method for training a risk level prediction model, the method comprising:

[0140] Acquiring financial product data and converting the financial product data into a product representation vector; wherein the product representation vector embeds multiple features of the financial product;

[0141] Obtaining customer purchasing behavior data for financial products and the customer's basic purchasing preferences, and converting the basic purchasing preferences into a customer base representation vector; the basic purchasing preferences are obtained by desensitizing the customer's responses to the most recent financial questionnaire;

[0142] Determining a customer's long-term preference vector and a short-term preference vector based on the customer base representation vector, the product representation vector, and the purchase behavior data, and determining the customer's purchase preference vector for financial products based on the long-term preference vector and the short-term preference vector;

[0143] constructing a training dataset based on the product representation vector and the customer's purchase preference vector for financial products;

[0144] The graph neural network model is trained based on the training data set to obtain a risk level prediction model.

[0145] and / or, performing a method for recommending a financial product, the method comprising:

[0146] Acquiring the customer's purchase behavior data and financial product data of at least one financial product to be recommended, and converting the purchase behavior data and the financial product data into a product representation vector and a customer's purchase preference vector for the financial product;

[0147] Inputting the product representation vector and the purchase preference vector into a risk level prediction model for prediction to obtain a risk level prediction result; wherein the risk level prediction model is trained based on the risk level prediction model training method described in any embodiment of the present invention;

[0148] A target financial product is determined based on the risk level prediction result.

[0149] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0150] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0151] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0152] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0153] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0154] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A training method for a risk level prediction model, characterized in that: include: Acquiring financial product data and converting the financial product data into a product representation vector; wherein the product representation vector embeds multiple features of the financial product; Obtaining customer purchasing behavior data for financial products and the customer's basic purchasing preferences, and converting the basic purchasing preferences into a customer base representation vector; the basic purchasing preferences are obtained by desensitizing the customer's responses to the most recent financial questionnaire; Determining a customer's long-term preference vector and a short-term preference vector based on the customer base representation vector, the product representation vector, and the purchase behavior data, and determining the customer's purchase preference vector for financial products based on the long-term preference vector and the short-term preference vector; constructing a training dataset based on the product representation vector and the customer's purchase preference vector for financial products; The graph neural network model is trained based on the training data set to obtain a risk level prediction model.

2. The method according to claim 1, characterized in that The financial product data includes one or more of the following: description text of the financial product, product term, expected rate of return and risk level; The converting the financial product data into a product representation vector includes: The description text is segmented to obtain a word sequence, and the word sequence is converted into a word vector; the word vector is converted into a word context representation vector that captures context features through a convolutional network; based on a word-level attention network, different attention weights are assigned to different words through attention weighting to obtain a text representation vector; and / or, embedding the product term into a product term vector through an embedding layer; and / or, converting the expected rate of return into discrete features, and embedding the discrete features into an expected rate of return vector through an embedding layer; and / or, embedding the risk level into a risk level vector via an embedding layer; A product representation vector is determined based on one or more of the text vector, product term vector, expected return vector, and risk level vector.

3. The method according to claim 1, characterized in that The determining of the customer's short-term preference vector based on the customer base representation vector, the product representation vector, and the purchase behavior data includes: A customer preference vector representation is determined based on the basic representation vector and the product representation vector; and a customer's sequential preference is modeled from the purchase behavior data based on a gated recurrent unit network and the customer preference vector representation to obtain a customer's short-term preference vector.

4. The method according to claim 3, characterized in that The determining of the customer's long-term preference vector based on the customer base representation vector, the product representation vector, and the purchase behavior data includes: Introducing a personalized attention network to assign different weight coefficients to each purchase behavior; A long-term preference vector is determined based on the short-term preference vector and a weight coefficient of the purchase behavior corresponding to the short-term preference vector.

5. The method according to claim 1, wherein The method further comprises: A prediction is made based on the product representation vector and the customer's purchase preference vector for the financial product to obtain a click-through rate score of the customer for the financial product.

6. A method for recommending financial products, characterized in that: include: Acquiring the customer's purchase behavior data and financial product data of at least one financial product to be recommended, and converting the purchase behavior data and the financial product data into a product representation vector and a customer's purchase preference vector for the financial product; Inputting the product representation vector and the purchase preference vector into a risk level prediction model for prediction to obtain a risk level prediction result; wherein the risk level prediction model is trained based on the risk level prediction model training method described in claims 1-5; A target financial product is determined based on the risk level prediction result.

7. A training device for a risk level prediction model, characterized in that: include: A product representation vector determination module, configured to obtain financial product data and convert the financial product data into a product representation vector; the product representation vector embeds multiple features of the financial product; A customer base representation vector determination module is configured to obtain customer purchasing behavior data for financial products and the customer's basic purchasing preferences, and convert the basic purchasing preferences into a customer base representation vector; the basic purchasing preferences are obtained by desensitizing the customer's responses to the most recent financial questionnaire; a purchase preference vector determination module, configured to determine a customer's long-term preference vector and short-term preference vector based on the customer base representation vector, the product representation vector, and the purchase behavior data, and determine the customer's purchase preference vector for financial products based on the long-term preference vector and the short-term preference vector; A training data set construction module, configured to construct a training data set based on the product representation vector and the customer's purchase preference vector for financial products; The risk level prediction model training module is used to train the graph neural network model based on the training data set to obtain a risk level prediction model.

8. A financial product recommendation device, characterized in that: include: a data acquisition module, configured to acquire the customer's purchase behavior data and financial product data of at least one financial product to be recommended, and convert the purchase behavior data and financial product data into a product representation vector and a customer's purchase preference vector for the financial product; a risk level prediction result prediction module, configured to input the product representation vector and the purchase preference vector into a risk level prediction model for prediction, thereby obtaining a risk level prediction result; wherein the risk level prediction model is trained based on the risk level prediction model training method according to claims 1-5; The target financial product determination module is used to determine the target financial product based on the risk level prediction result.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the risk level prediction model training method described in any one of claims 1-5, and / or the financial product recommendation method described in claim 6.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, which are used to enable a processor to implement the risk level prediction model training method described in any one of claims 1 to 5 and / or the financial product recommendation method described in claim 6 when executed.

Images