Green financial data processing method and system based on deep learning model

Abstract

The invention discloses a green financial data processing method and system based on a deep learning model. The method comprises the steps of collecting green customer purchase intention text data in green financial data, obtaining structured subject term data, converting the structured subject term data into subject term distribution represented by vectors, and obtaining a subject term sample set; constructing an attention mechanism and a deep neural network to form a deep learning model, and training and optimizing the deep learning model based on the subject term sample set; and based on the trained and optimized deep learning model, using green financial data to predict the purchase intention of the green customer. According to the method, the attention mechanism and the deep neural network are used for green financial data processing, the purchase intention of the green customer is predicted, the prediction accuracy and efficiency are improved, a new thought is provided for prediction of the purchase intention of the green customer, the relationship between the green customer and the financial product can be effectively captured, work development is facilitated, and the economic benefit is improved. And the working efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of using deep learning models to process financial data for predicting the purchase intention of green customer financial management, and particularly to a method and system for processing green financial data based on a deep learning model. Background Art

[0002] Green finance refers to economic activities that support environmental improvement, response to climate change, and efficient utilization of resources, that is, financial services provided to green finance customers in fields such as environmental protection, energy conservation, clean energy, green transportation, and green buildings, including multiple aspects such as financial management and financing. Currently, green finance is continuously optimized, improving green transition finance standards, and continuously promoting the development of various green transition financial product tools to support the construction of a beautiful China and green and low-carbon transformation.

[0003] In recent years, green finance customers hope to obtain information on financial management products and manage funds in a stable manner. When financial institutions seek to expand market share and improve service quality, they need to accurately process green customer green financial data, grasp the financial management needs and purchase intentions of green customers, so as to provide personalized products and services. However, the traditional processing of green financial data often focuses on the static statistics of customer purchase intentions and cannot effectively process green financial data rich in dynamic information, nor can it accurately capture the potential needs of green customers.

[0004] Moreover, the traditional processing of green financial data mainly analyzes based on the data structure and data characteristics of green financial data, and often ignores the weight of various intentions of green customers and the association between various intentions and financial management products within green financial data. In recent years, with the emergence of the attention mechanism and deep neural networks, it provides a new idea for green financial data processing. Based on the attention mechanism and deep neural networks, the relationship between green customers and financial management products in green financial data can be effectively captured.

[0005] Due to the complexity of green customer structured subject term data, traditional machine learning methods have some limitations in processing this type of data. Therefore, a method for processing green financial data based on a deep learning model is needed to further improve the processing of structured green financial data through the attention mechanism and deep neural networks, and improve the accuracy and efficiency of predicting the purchase intention of green customers for financial management. It can avoid a large amount of manual intervention and adjustment, be applicable to complex and large amounts of green financial data, and has stronger scalability.

[0006] The invention application with the application number 202211313530.2 discloses a financial planning method, system and system, belonging to the technical field of data processing. The method includes: determining target customers from customer behavior data based on a financial intention prediction model; pushing financial service recommendation information to the target customers based on a service push model; determining the risk level corresponding to the target customer label portrait based on a risk assessment model; outputting the asset allocation result corresponding to the financial service configuration information based on a preset deep investment model; classifying information data based on an information classification and rating model and outputting a service rating result. Through the proposed comprehensive financial planning method for the entire life cycle of customer finance, the present invention realizes the innovation of full-process automation and intelligence in financial planning. However, the scheme also has the following problems: 1. The customer intention prediction does not effectively combine the attention mechanism and the deep neural network; 2. The ability to comprehensively process financial data is poor. Summary of the Invention

[0007] The purpose of the present invention is to provide a method and system for processing green financial data based on a deep learning model, which uses the text data in green financial data to predict the purchase intention of green customers for financial management, and improves the accuracy and efficiency of predicting the purchase intention of green customers.

[0008] The purpose of the present invention can be achieved through the following technical solutions: A method and system for processing green financial data based on a deep learning model.

[0009] First aspect: A method for processing green financial data based on a deep learning model, including:

[0010] S1. Collect the text data of the purchase intention of green customers in green financial data, preprocess the text data to obtain structured subject word data, convert the structured subject word data into a subject word distribution represented by vectors, and obtain a subject word sample set;

[0011] S2. Construct a deep learning model composed of an attention mechanism and a deep neural network to perform weighted processing on the subject words and capture the complex relationships between the subject words;

[0012] S3. Train and optimize the deep learning model based on the subject word sample set;

[0013] S4. Based on the trained and optimized deep learning model, use green financial data to predict the purchase intention of green customers.

[0014] Further, the preprocessing of the text data in S1 includes:

[0015] Classify the text data, complete some missing key data to obtain complete text data; use Chinese word segmentation technology to segment the complete text data into a sequence of subject word vectors to obtain structured subject word data.

[0016] Further, the step of obtaining the set of subject term samples in S1 includes:

[0017] S11. Divide the structured subject term data into k categories of samples, and calculate the variance of the k-th category of samples The formula is:

[0018]

[0019] where the number of samples in each category is N 1 , N 2 , …, N k , corresponding to the k-th category (1 ≤ k ≤ M), D k = {(x 1 , y 1 ), (x 2 , y 2 ),..., (x m , y m )} is the data set of the k-th category of samples, and each sample is an n-dimensional vector; is the sample after change, then u is a unit vector, u T u = 1, and the mean vector of the samples after change is where:

[0020]

[0021] S12. Calculate the sum of the sample variances of each category. The formula is:

[0022]

[0023] where,

[0024] S13. Calculate the central distance between different categories of samples i and j. The formula is:

[0025]

[0026] S14. Calculate the sum of the distances between all categories of samples. The formula is:

[0027]

[0028] where,

[0029] S15. Under the known conditions, maximize u T S b u, and minimize u T Sw u, the calculation matrix The largest d eigenvalues and the eigenvectors corresponding to the d eigenvalues (w 1 , w 2 ,..., w d ), obtain the projection matrix W = (w 1 , w 2 ,…, w d ).

[0030] S16. For each sample feature x k in the sample set D i , transform it into a new sample z i = W T x i , and obtain the topic word sample set D k ' = {(z 1 , y 1 ), (z 2 , y 2 ),…, (z m , y m )}.

[0031] Furthermore, the construction process of the attention mechanism is as follows:

[0032] S21. Obtain the multi-head self-attention result, and the formula is:

[0033]

[0034] where is the initial feature vector, and the matrix is the learnable parameter of the i-th head in the Transfomer;

[0035] S22. Concatenate the multi-head self-attention results to obtain the self-attention first-order feature, and the formula is:

[0036]

[0037] where is the trainable parameter;

[0038] S23. Aggregate the self-attention first-order feature result using the Attentional Aggregation Layer, and the formula is:

[0039]

[0040] where:

[0041]

[0042] S24. Calculate the inner product of the result \(u\) of the Attentional Aggregation Layer 1 and the result \(X\) of the Transformer 1 and splice the calculation results of each feature to form a second-order interaction result. The formula is:

[0043]

[0044] S25. Reduce the dimension of the second-order interaction result through a fully connected layer, add it to the first-order interaction result, and obtain the calculation result of the attention mechanism module.

[0045] Further, the steps of constructing the deep neural network are as follows:

[0046] S26. Divide the input layer, hidden layer, and output layer of the deep neural network, where the hidden layer has three layers and is fully connected between layers;

[0047] S27. The activation function of the deep neural network is \(\sigma(z)\), and the linear relationship coefficients \(w\) and bias \(b\) satisfy the linear relationship formula:

[0048]

[0049] S28. The output of the \(j\)-th neuron in the \(l\)-th layer of the deep neural network The formula is:

[0050]

[0051] where \(m\) is the number of neurons.

[0052] Further, in S3, the steps of training and optimizing the deep learning model based on the topic word sample set are as follows:

[0053] S31. Divide the labeled topic word sample set into a training set, a validation set, and a test set;

[0054] S32. Use the training set to calculate the loss through the cross-entropy loss function (CrossEntropyLoss). The formula is:

[0055]

[0056] S33. Set the number of iterations for network training, use the gradient descent method to perform gradient backpropagation and update all trainable parameters in the model according to the calculation result of the loss function, obtain the trained deep learning model, and perform validation and testing.

[0057] Second aspect: A green finance data processing system based on a deep learning model, including:

[0058] Data preprocessing module: used to convert structured subject term data into vector representations for processing based on attention mechanisms and deep neural networks;

[0059] Attention mechanism module: used to automatically learn the weights of each subject term, making the model more flexible and effective;

[0060] Deep neural network module: used to capture the complex relationships between subject terms and accurately predict the purchase intentions of green customers;

[0061] Training module: used to train and optimize the deep learning model using the labeled subject term dataset.

[0062] Prediction module: used to predict the unlabeled data of green customers' purchase intentions.

[0063] Third aspect: An electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps of the method provided in the first aspect.

[0064] Fourth aspect: A non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the method provided in the first aspect.

[0065] Advantages of the present invention:

[0066] 1. The present invention uses attention mechanisms and deep neural networks to process structured text data of green finance, accurately predicts the financial management intentions of green customers, improves the accuracy and efficiency of prediction, provides a new idea for predicting the purchase intentions of green customers using green finance data analysis, can effectively capture the relationship between green customers and financial products, is conducive to the work development, and improves work efficiency.

[0067] 2. The present invention classifies the text data in green finance data, fills in some missing key data to obtain complete text data, uses Chinese word segmentation technology to segment the complete text data into text segments, converts them into a sequence of subject term vectors, and obtains structured subject term data, which is conducive to introducing attention mechanisms, conducive to deep learning model analysis, and conducive to accurately obtaining the judgment of green customers' purchase intentions in green finance data.

[0068] 3. Based on the LDA topic model, the present invention introduces an attention mechanism to weight each subject term to reflect the influence degree of the subject term on the purchase intentions of green customers; the attention mechanism can automatically learn the weights of each subject term, making important subject terms receive more attention, can effectively capture the relationship between green customers and financial products, is conducive to the work development, and improves work efficiency.

[0069] 4. The present invention constructs a deep neural network to learn the feature representation and relationship extraction of the purchase intention of green customers in green financial data. By learning the features of structured data and predicting the purchase intention of green customers, a large amount of manual intervention and adjustment are avoided, the error rate of the model is reduced, the model is applicable to various types of text data at the same time, has strong scalability, is conducive to the development of work, and improves work efficiency. Description of the Drawings

[0070] Figure 1 It is a schematic flow chart of a method and system for processing green financial data based on a deep learning model according to the present invention;

[0071] Figure 2 It is a schematic principle diagram of a method and system for processing green financial data based on a deep learning model according to the present invention;

[0072] Figure 3 It is a schematic structural diagram of an electronic device according to the present invention. Detailed Embodiments

[0073] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation to the present invention.

[0074] As Figure 1 shown, the present invention discloses a method for processing green financial data based on a deep learning model, including the steps of:

[0075] S1. Collect the text data of the purchase intention of green customers in green financial data, preprocess the text data to obtain structured subject word data, convert the structured subject word data into a topic word distribution represented by a vector, and obtain a topic word sample set.

[0076] As Figure 2 shown, collect and sort out a large amount of text data of the purchase intention of green customers in green financial data, including information such as product characteristics description, risk tolerance, purchase intention, etc., and obtain structured subject word data after Chinese word segmentation processing. The Chinese word segmentation processing includes text segmentation, stop word removal, etc., represents each text data as structured subject word data of a topic word vector sequence, and calculates the perplexity of each subject word in the structured subject word data.

[0077] Then use the LDA topic model to calculate the topic word distribution, and the process includes:

[0078] Assume that there are k categories of topic word samples, and the number of samples in each category is N1 , N 2 , …, N k 。 Corresponding to the first class Corresponding to the second class Corresponding to the k-th class, where each sample is an n-dimensional vector.

[0079] Let be the sample after change, then Here, let u be a unit vector, i.e., u T u = 1.

[0080] Suppose the dataset of the k-th class samples is D k , and the mean vector of the samples after change is: Then the variance of the k-th class samples is where:

[0081]

[0082] The variance of the k-th class samples:

[0083]

[0084] The sum of the sample variances of each class:

[0085]

[0086] where,

[0087] The central distance between different classes i, j:

[0088]

[0089] The sum of the distances between all classes is:

[0090]

[0091] where, Under the known conditions, maximize u T S b u, and minimize u T S w u.

[0092] Calculate the largest d eigenvalues and the corresponding d eigenvectors (w ) of the matrix 1 , w 2 ,..., w d ), and obtain the projection matrix W = (w 1 , w 2 ,..., wd )。For each sample feature x in the sample set i , it is transformed into a new sample z i = W T x i , and the topic word sample set D' = {(z 1 , y 1 ), (z 2 , y 2 ), …, (z m , y m )} is obtained.

[0093] S2. Construct a deep learning model composed of an attention mechanism and a deep neural network to perform weighted processing on topic words and capture the complex relationships between topic words.

[0094] Construct an attention mechanism model. For the input initial feature vector , the multi-head self-attention results of the feature encoding of each feature domain are obtained through Transfomer respectively:

[0095]

[0096] Matrix is the learnable parameter of the i-th head in Transfomer;

[0097] Concatenate the multi-head self-attention results to obtain the self-attention first-order features:

[0098]

[0099] where is a trainable parameter, where d = d k ;

[0100] Aggregate the self-attention results obtained from the previous layer using AttentionalAggregationLayer to obtain:

[0101]

[0102] Calculate the inner product of the results of AttentionalAggregationLayer and the results of Transformer and concatenate the calculation results of each feature to form the second-order interaction results:

[0103]

[0104] Finally, reduce the dimension of the second-order interaction results through a fully connected layer and add them to the first-order interaction results to obtain the calculation results of the attention mechanism module.

[0105] Build a deep neural network, which is divided into three layers: an input layer, a hidden layer, and an output layer, with three hidden layers in the middle.

[0106] The layers are fully connected, and any neuron in the i-th layer is connected to any neuron in the i+1-th layer. The number of network layers and parameters are relatively complex, and the linear relationship coefficients w and biases b satisfy a linear relationship plus an activation function σ(z).

[0107] For the output of the second layer we have:

[0108]

[0109] Generalize the above process. Assume that the l-th layer has m neurons in total. Then, for the output of the j-th neuron in the l-th layer we have:

[0110]

[0111] Correspondingly is the x of the input layer k ; the σ activation function is the relu function, and its expression is: f(x) = max(0, x).

[0112] The input layer has no w parameter, and the output layer has no bias parameter b. w is the linear relationship coefficient, and the linear coefficient from the 4th neuron in the second layer to the 2nd neuron in the third layer is defined as The superscript 3 represents the layer where the linear coefficient w is located, and the subscripts correspond to the index 2 of the output third layer and the index 4 of the input second layer. This is mainly to facilitate the matrix representation operation of the model. For linear operations, no transpose is required, that is, directly wx + b.

[0113] For matrix operations, assume that the l-1-th layer has m neurons in total, and the l-th layer has n neurons in total. Then, the linear coefficient w of the l-th layer forms an n×m matrix W l . The bias b of the l-th layer forms an n×1 vector b l , the output a of the l-1-th layer forms an m×1 vector a l-1 , the pre-activation linear output z of the l-th layer forms an n×1 vector z l , the output a of the l-th layer forms an n×1 vector a l . The output of the l-th layer is:

[0114] a l = σ(z l ) = σ(W l a l-1 + b l )

[0115] Starting from the input layer, calculate layer by layer backward until the output layer is reached to obtain the result of the output layer.

[0116] S3. Train and optimize the deep learning model based on the set of topic word samples.

[0117] The attention mechanism and the deep neural network together constitute the deep learning model. As Figure 2 shown, the deep learning model includes a multi-layer perceptron with N fully connected layers to learn the hierarchical features of the input. Dropout and Batch Normalization are used in the deep learning model to avoid overfitting and improve the multi-layer perception effect of the attention mechanism.

[0118] Use the Sigmoid function Combine the results of the attention mechanism and the deep neural network model to achieve classification, so as to obtain the probability that a green customer becomes a green customer's purchase intention, and then calculate the loss through the cross-entropy loss function (CrossEntropyLoss). Specifically:

[0119]

[0120] Finally, set the number of iterations for network training, and use the gradient descent method to perform gradient backpropagation and update all trainable parameters in the model according to the calculation result of the loss function. If the set number of iterations is reached, the training ends.

[0121] This model training uses the supervised learning algorithm to perform model training on the feature data extracted by the attention mechanism and the deep neural network model constructed in the previous step on the training set, and perform the prediction of the intention to purchase financial products on the test set. Specifically, use the labeled green customer intention data for supervised learning, and update the parameters of the attention mechanism and the deep neural network model through the backpropagation algorithm to minimize the loss function. The loss function can be the cross-entropy loss function, the mean squared error loss function or other types of loss functions. The present invention uses the cross-entropy loss function.

[0122] S4. Based on the trained and optimized deep learning model, use green finance data to predict the purchase intention of green customers.

[0123] Input the label data of the green customer to be predicted into the above green customer prediction method model, and calculate the possibility that it becomes a green customer to purchase financial products.

[0124] Use the trained feature embedding layer to perform topic word modeling on its label data, and input the topic word modeling data features into the trained attention mechanism and deep neural network model to perform a binary classification task. Finally, at the output layer, process the output feature representation of the fully connected layer, and judge whether the green customers have the intention to purchase financial products according to the classification results.

[0125] Use the trained attention mechanism and deep neural network model to predict the new green customer purchase intention data. Specifically, input the new green customer purchase intention data into the input layer of the attention mechanism and deep neural network model, extract important feature representations through the operations of the attention mechanism and deep neural network model, and finally obtain the prediction results at the output layer.

[0126] The present invention also discloses a green customer green financial data processing system based on the above processing method. The processing system includes:

[0127] Data preprocessing module: used to convert structured topic word data into vector representations for the convenience of processing based on the attention mechanism and deep neural network;

[0128] Attention mechanism module: used to automatically learn the weights of each topic word to make the model more flexible and effective;

[0129] Deep neural network module: used to capture the complex relationships between topic words and accurately predict the purchase intention of green customers;

[0130] Training module: used to train and optimize the deep learning model using the labeled topic word data set.

[0131] Prediction module: used to predict the unlabeled green customer purchase intention data.

[0132] The processing system of the present invention uses the data preprocessing module to collect and organize a large amount of green customer purchase intention text data, obtain the structured text data after Chinese word segmentation processing, where each text data represents a word vector; perform topic word modeling on the preprocessed text data by calculating the topic word perplexity and using the topic model, convert the text data into a topic word distribution with a fixed number, and convert the structured text data into vector representations for the convenience of processing by the attention mechanism and deep neural network.

[0133] Use the attention mechanism module to weight each topic word to reflect the influence degree of the topic word on the green customer purchase intention. The attention mechanism can automatically learn the weights of each topic word, so that important topic words receive more attention.

[0134] Utilize a deep neural network module to learn the feature representation and relationship extraction of green customer purchase intention, including multiple hidden layers and an output layer, for learning the features of structured data and predicting the green customer purchase intention.

[0135] Utilize a training module to train and optimize the attention mechanism and deep neural network model using the labeled green customer purchase intention data.

[0136] Utilize the trained prediction module to predict the green customer purchase intention for new data based on the attention mechanism and deep learning model using green financial data. Input the new green customer purchase intention text data into the input layer of the attention mechanism and deep learning model, extract important feature representations through the operations of the attention mechanism and deep learning model, and finally obtain the prediction result at the output layer.

[0137] The present invention also provides an electronic device. Figure 3 It is a schematic structural diagram of the electronic device provided by the embodiment of the present invention, as Figure 3 shown. The electronic device may include: a processor, a communications interface, a memory, and a communication bus. Among them, the processor, the communications interface, and the memory complete mutual communication through the communication bus. The processor can call the logical instructions in the memory, for example, to execute the following methods:

[0138] S1. Collect the green customer purchase intention text data in the green financial data, preprocess the text data to obtain structured subject term data, convert the structured subject term data into a vector representation of the subject term distribution, and obtain a subject term sample set;

[0139] S2. Construct a deep learning model composed of an attention mechanism and a deep neural network to perform weighted processing on the subject terms and capture the complex relationships between the subject terms;

[0140] S3. Train and optimize the deep learning model based on the subject term sample set;

[0141] S4. Based on the trained and optimized deep learning model, use the green financial data to predict the green customer purchase intention.

[0142] In addition, when the logical instructions in the above-mentioned memory can be implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0143] An embodiment of the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the methods provided in the above-mentioned embodiments, for example, including:

[0144] S1. Collect the text data of the green customer purchase intention in the green finance data, preprocess the text data to obtain the structured subject word data, convert the structured subject word data into a vector representation of the subject word distribution, and obtain the subject word sample set;

[0145] S2. Construct a deep learning model composed of an attention mechanism and a deep neural network, perform weighted processing on the subject words, and capture the complex relationships between the subject words;

[0146] S3. Train and optimize the deep learning model based on the subject word sample set;

[0147] S4. Based on the trained and optimized deep learning model, use the green finance data to predict the green customer purchase intention.

[0148] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0149] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A green finance data processing method and system based on a deep learning model, characterized in that: Includes steps: S1. Collect green customer purchase intention text data from green finance data, pre-process the text data, obtain structured subject word data, convert the structured subject word data into subject word distribution represented by vectors, and obtain a subject word sample set; S2. Construct a deep learning model composed of attention mechanism and deep neural network to perform weighted processing on topic words and capture the complex relationship between topic words; S3, training and optimizing the deep learning model based on the keyword sample set; S4. Based on the trained and optimized deep learning model, green financial data is used to predict green customer purchase intention.

2. According to claim 1, a method and system for processing green financial data based on a deep learning model is characterized in that: The preprocessing of text data in S1 includes: Classify the text data, complete some missing key data, and obtain complete text data; use Chinese word segmentation technology to segment the complete text data, convert it into a subject word vector sequence, and obtain structured subject word data.

3. According to claim 1, a method and system for processing green financial data based on a deep learning model is characterized in that: The step of obtaining a subject word sample set in S1 includes: S11. Divide the structured keyword data into k-class samples and calculate the variance of the k-th class sample The formula is: Among them, the number of samples in each category is N1, N2, ..., N k , Corresponding to the kth class (1≤k≤M), D k ={(x1,y1),(x2,y2),...,(x m ,y m )} is the data set of the k-th sample, each sample All are n-dimensional vectors; for After the change, the sample u is a unit vector, u T u=1, the sample mean vector after the change is in: S12. Calculate the sum of sample variances of each category. The formula is: in, S13. Calculate the center distance between samples i and j of different categories. The formula is: S14. Calculate the sum of the distances between all category samples. The formula is: in, S15. Maximize u under known conditions T S b u, minimize u T S w u, calculation matrix The largest d eigenvalues ​​and the eigenvectors corresponding to the d eigenvalues ​​(w1, w2, ..., w d ), and obtain the projection matrix W = (w1, w2, ..., w d ); S16. For sample set D k Each sample feature x in i , converted into a new sample z i =W T x i , get the subject word sample set D k '={(z1,y1),(z2,y2),...,(z m ,y m )}.

4. A green finance data processing method and system based on a deep learning model according to claim 1, characterized in that: The attention mechanism construction process is as follows: S21. Get the multi-head self-attention results. The formula is: <h2 style=";text-align:left;direction:ltr">Q=W<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> (Q) <h2 style=";text-align:left;direction:ltr"> X0,K=W<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> (K) <h2 style=";text-align:left;direction:ltr"> X0,V=W<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> (V) <h2 style=";text-align:left;direction:ltr"> X0 in, is the initial eigenvector, the matrix is the learnable parameter of the i-th head in Transformer; S22. Concatenate the results of multiple self-attention heads to obtain the first-order features of self-attention. The formula is: in is a trainable parameter; S23. Results of self-attention first-order features X1∈R d×m Use AttentionalAggregation Layer for aggregation, the formula is: in: S24. Calculate the inner product of the Attentional Aggregation Layer result u1 and the Transformer result X1, and concatenate the calculation results of each feature to form a second-order interaction result. The formula is: S25. Reduce the dimension of the second-order interaction result through the fully connected layer and add it to the first-order interaction result to obtain the calculation result of the attention mechanism module.

5. A green finance data processing method and system based on a deep learning model according to claim 1, characterized in that: The steps of constructing a deep neural network are: S26. Divide the deep neural network into input layer, hidden layer and output layer, where the hidden layer has three layers and all layers are fully connected; S27, the activation function of the deep neural network is σ(z), and the linear relationship coefficient w and the bias b satisfy the linear relationship formula: S28, the output of the jth neuron in the lth layer of the deep neural network The formula is: Among them, m is the number of neurons.

6. A green finance data processing method and system based on a deep learning model according to claim 1, characterized in that: In S3, the deep learning model is trained and optimized based on the subject word sample set, and the steps are as follows: S31, dividing the labeled subject word sample set into a training set, a validation set and a test set; S32. Using the training set, the loss is calculated by the cross entropy loss function (CrossEntropy Loss), the formula is: S33. Set the number of iterations of network training, use the gradient descent method to perform gradient backpropagation and update of all trainable parameters in the model according to the calculation results of the loss function, obtain the trained deep learning model, and perform verification and testing.

7. A green customer green finance data processing system based on the method according to any one of claims 1 to 6, characterized in that: The processing system comprises: Data preprocessing module: used to convert structured keyword data into vector representation for easy processing based on attention mechanism and deep neural network; Attention mechanism module: used to automatically learn the weight of each topic word, making the model more flexible and effective; Deep neural network module: used to capture the complex relationship between keywords and accurately predict the purchase intention of green customers; Training module: used to train and optimize deep learning models using labeled keyword datasets; Prediction module: used to predict unlabeled green customer purchase intention data.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of a processing method according to any one of claims 1 to 6 are implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a processing method according to any one of claims 1 to 6 are implemented.

Images