LSTM and multi-graph convolutional neural network combined stock selection method and system

Abstract

The invention discloses a stock selection method and system in combination with an LSTM and a multi-graph convolutional neural network, and the method comprises the following steps: obtaining original stock feature data, carrying out the classification of the original stock feature data according to stock codes, and constructing a data set through the preprocessed stock feature data; a stock selection model comprising an LSTM model, a graph convolutional network, an Attention mechanism and a full connection layer is trained, stock feature data of stocks to be detected are input into the trained stock selection model to predict future return rates, sorting is performed according to the future return rates, and the first K stocks to be detected with the highest return rates are selected for investment. According to the method, the stock relationship influence is dynamically captured, historical feature embedding and stock relationship embedding are combined, the model prediction accuracy can be effectively improved, investment selection is provided for investors, and therefore the higher investment return rate is obtained.

Description

Technical Field

[0001] The present invention belongs to the field of data processing, and in particular relates to a stock selection method and system combining LSTM and multi-graph convolutional neural network. Background Art

[0002] As the level of economic development improves and the financial market expands, stocks have become one of the most attractive investment options due to their potential high returns. Whether stock investors can make a profit essentially depends on whether the selected stock prices will rise in the future. Therefore, it is crucial to be able to effectively predict the future prices and trends of stocks.

[0003] Quantitative investment makes this process more scientific through modeling. Later, with the advent of the information age and the development of artificial intelligence technology, more algorithms were used for stock prediction. By capturing various stock characteristics, such as historical price data, trading volume, etc., and the relationship between returns, effective stock yield predictions can be achieved, and stocks can be selected based on the prediction results to ultimately obtain excess investment returns.

[0004] There are still some problems with the deep learning models currently used for stock prediction. First, when making stock predictions, only the relationship between stock prices and their historical data series is considered. However, in the real market, stocks do not exist independently, and there are a lot of connections between them. For example, the stock prices of two companies in a competitive relationship show a state of increase and decrease, while the stock prices of companies with supply relationships are likely to show the same trend. Secondly, the methods based on stock relationship graphs are usually based on static relationship graphs. The stock relationship graphs are constructed based on prior knowledge and cannot accurately reflect the changes in stock relationships. Moreover, the generated relationship graphs only start from a single perspective (such as industry relationships, Wiki relationships, investment relationships), and cannot fully include all the relationships between stocks.

[0005] LSTM is a variant of RNN. It was originally proposed to solve the problem of gradient vanishing or gradient exploding in the back-propagation process of RNN. It can capture and utilize long-term time series dependencies well. It can automatically filter information through input gate, output gate and forget gate, ignore invalid parts and retain important information. Stock prices are regarded as a typical time series. The LSTM model can capture the long-term dependencies of the sequence well, and reduce the complexity of the data while retaining the trend characteristics of the original data. Therefore, it is widely used in the financial field.

[0006] The efficient market hypothesis proves that stock prices can fully reflect market information. Based on this view, by calculating the relationship between historical stock price sequences, the relationship between stocks can be dynamically reflected. Attention can adaptively focus on the importance of input information, and can more effectively integrate the importance of different stocks calculated from different perspectives to obtain a more complete stock relationship graph. The vertices on the graph represent stock nodes, and the edge values ​​represent the strength of the correlation between each pair of stocks. GCN is a deep learning method widely used on graphs. It aggregates adjacent nodes in the graph, captures the relationship between nodes, and generates new feature representations for each node, thereby realizing the cross-information influence between nodes and obtaining the results after stock interaction.

[0007] In the process of stock prediction, stock prices are not only related to their own historical series, but also affected by other related stocks. Therefore, with the help of LSTM's ability to process long-term dependencies in time series and multi-graph convolutional neural networks' ability to capture the relationship between adjacent nodes, we can obtain market information more comprehensively, thereby achieving accurate and efficient stock data analysis and prediction, and obtaining a new quantitative stock selection method. Summary of the invention

[0008] The purpose of the present invention is to solve the problems existing in the prior art and provide a stock selection method and system combining LSTM and multi-graph convolutional neural network. The method of the present invention can effectively improve the prediction accuracy and increase the investment return rate.

[0009] In order to achieve the above-mentioned invention object, the present invention specifically adopts the following technical solutions:

[0010] In a first aspect, the present invention provides a stock selection method combining LSTM with a multi-graph convolutional neural network, comprising the following steps:

[0011] S1: Obtain original stock feature data and classify the original stock feature data according to stock codes;

[0012] S2: performing data preprocessing on the classified stock feature data to obtain preprocessed stock feature data, and constructing a data set from the preprocessed stock feature data;

[0013] S3: training a stock selection model on the data set, wherein the stock selection model includes an LSTM model, a graph convolutional network, an Attention mechanism, and a fully connected layer;

[0014] S4: Obtain the stock feature data of the stocks to be tested that have undergone data preprocessing, input the stock feature data of the stocks to be tested into the trained stock selection model, use the stock selection model to predict the future rate of return of each stock to be tested, sort the stock feature data of the stocks to be tested according to the size of the future rate of return, and select the top K stocks to be tested with the highest rate of return for investment.

[0015] Based on the above scheme, each step can be implemented in the following preferred specific manner.

[0016] As a preferred embodiment of the first aspect, in step S2, the specific process of data preprocessing is: after centering and normalizing the classified stock feature data in turn, the stock price moving average is calculated using the centralized and normalized stock feature data.

[0017] As a preferred embodiment of the first aspect, in step S3, the specific processing flow in the stock selection model is as follows:

[0018] S31: Input the preprocessed stock feature data into the LSTM model to extract the stock trend features of each stock;

[0019] S32: Obtain the historical closing price sequence of the stock from the original stock feature data, calculate the Pearson correlation coefficient between each pair of stocks based on the historical closing price sequence of the stock, and obtain the Pearson relationship matrix, calculate the Spearman rank correlation coefficient between each pair of stocks based on the historical closing price sequence of the stock, and obtain the Spearman rank relationship matrix, and obtain the similarity between the historical closing price sequences of the stock through the FastDTW algorithm to obtain the similarity relationship matrix;

[0020] S33: Integrate the Pearson relationship matrix, the Spearman rank relationship matrix, and the similarity relationship matrix into an attention relationship matrix through the Attention mechanism, where each element in the attention relationship matrix represents the strength of the relationship between two stocks of the corresponding index;

[0021] S34: taking each element in the attention relationship matrix as an edge value of the stock relationship graph, and taking the stock trend characteristics of each stock as a stock node in the stock relationship graph, so as to construct the stock relationship graph;

[0022] S35: Input the stock relationship graph into the graph convolutional network, and obtain the relationship embedding of each stock after performing two layers of graph convolution process;

[0023] S36: The stock trend characteristics of each stock and the relationship between each stock are embedded and spliced ​​together, and the spliced ​​results are processed through a fully connected layer to obtain the prediction results of the yield of each stock.

[0024] As a preferred embodiment of the first aspect, in step S3, when training the stock selection model, the loss function used is specifically:

[0025]

[0026]

[0027]

[0028] Where: L1 represents the prediction loss, L2 represents the relative ranking loss between each pair of stocks; represents the predicted results of all stock returns at time t+1; y t+1 represents the true value of all stock returns at time t+1; α represents the weight hyperparameter; They represent the predicted results of the returns of the i-th stock and the j-th stock at time t+1 respectively; They represent the true values ​​of the returns of the i-th stock and the j-th stock at time t+1 respectively.

[0029] In a second aspect, the present invention provides a stock selection system combining LSTM and multi-graph convolutional neural network, comprising:

[0030] A data acquisition module is used to acquire original stock feature data and classify the original stock feature data according to stock codes;

[0031] A preprocessing module is used to perform data preprocessing on the classified stock feature data to obtain the preprocessed stock feature data, and to construct a data set from the preprocessed stock feature data;

[0032] A model training module, used for training a stock selection model on the data set, wherein the stock selection model includes an LSTM model, a graph convolutional network, an Attention mechanism, and a fully connected layer;

[0033] The result acquisition module is used to obtain the stock feature data of the stocks to be tested after data preprocessing, input the stock feature data of the stocks to be tested into the trained stock selection model, and use the stock selection model to predict the future rate of return of each stock to be tested, sort the stock feature data of the stocks to be tested according to the size of the future rate of return, and select the top K stocks to be tested with the highest rate of return for investment.

[0034] In a third aspect, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, can implement a stock selection method combining LSTM and a multi-graph convolutional neural network as described in any one of the schemes of the first aspect above.

[0035] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, a stock selection method combining LSTM and a multi-graph convolutional neural network as described in any one of the schemes of the first aspect is implemented.

[0036] In a fifth aspect, the present invention provides a computer electronic device, comprising a memory and a processor;

[0037] The memory is used to store computer programs;

[0038] The processor is used to implement a stock selection method combining LSTM and multi-graph convolutional neural network as described in any one of the solutions of the first aspect above when executing the computer program.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention provides a new method for quantitative stock selection. Compared with traditional statistical regression models and machine learning methods, the use of LSTM models and multi-graph convolutional neural networks can simultaneously capture the event dependency of stock price data and the relationship between different stocks, thereby improving the accuracy of stock yield prediction results. In addition, adding the relative ranking of stocks to the loss function can more accurately select the top-ranked stocks, enabling investors to obtain higher returns on investment. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic diagram of the process of the present invention;

[0042] Figure 2 It is a system block diagram of the present invention. DETAILED DESCRIPTION

[0043] In order to make the above-mentioned purpose, features and advantages of the present invention more obvious and easy to understand, the specific implementation mode of the present invention is described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. The technical features in each embodiment of the present invention can be combined accordingly without conflicting with each other.

[0044] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for the purpose of distinguishing descriptions, and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features.

[0045] like Figure 1 As shown, in a preferred implementation of the present invention, the above-mentioned stock selection method combining LSTM and multi-graph convolutional neural network includes the following steps S1 to S4. The specific implementation process is described in detail below.

[0046] S1: Obtain original stock feature data, and classify the original stock feature data according to stock codes.

[0047] S2: performing data preprocessing on the classified stock feature data to obtain preprocessed stock feature data, and constructing a data set from the preprocessed stock feature data.

[0048] It should be noted that in step S2, the specific process of data preprocessing is: after the classified stock feature data is centralized and normalized in turn, the stock price moving average is calculated using the centralized and normalized stock feature data.

[0049] In this embodiment, the data preprocessing of the classified stock feature data specifically includes three processes: centering, normalizing, and feature calculation. Specifically, each classified stock feature data is subtracted from its minimum value in the entire data set to center it; each classified stock feature data is divided by the difference between its maximum and minimum values ​​so that all data are mapped to the same dimension, that is, normalized to the (0,1) interval. The stock price moving average is calculated using the centralized and normalized stock feature data.

[0050] S3: Training a stock selection model on the data set, wherein the stock selection model includes an LSTM model, a graph convolutional network, an Attention mechanism, and a fully connected layer.

[0051] It should be noted that in step S3, the LSTM model of the stock selection model first extracts stock trend features, then the Attention mechanism integrates the stock relationship graph, and further the Graph Convolutional Network (GCN) performs graph convolution operations to realize dynamic stock relationship feature extraction, and finally the fully connected layer outputs the prediction results.

[0052] In this embodiment, the specific processing flow in the stock selection model is as follows:

[0053] S31: Input the preprocessed stock feature data into the LSTM model to extract the stock trend features of each stock; wherein the result of the i-th stock at time t is The result of all s stocks at time t is n is the number of extracted stock trend features, s is the number of stocks, They represent the 1st, 2nd, …, nth stock trend characteristics of the i-th stock at time t, They represent the results of the 1st, 2nd,…,sth stocks at time t respectively.

[0054] S32: Obtain the historical closing price sequence of the stock from the original stock feature data, calculate the Pearson correlation coefficient between each pair of stocks based on the historical closing price sequence of the stock, and obtain the Pearson relationship matrix Based on the historical closing price series of stocks, the Spearman rank correlation coefficient between each pair of stocks is calculated to obtain the Spearman rank relationship matrix The similarity between the historical closing price sequences of stocks is obtained through the FastDTW algorithm, and the similarity relationship matrix is ​​obtained. in, They represent the Pearson correlation coefficients between the ith stock and the 1st, 2nd, …, sth stocks, respectively. Respectively represent the Spearman rank correlation coefficient between the ith stock and the 1st, 2nd, …, sth stocks, Respectively represent the similarity between the i-th stock and the 1st, 2nd,…,s-th stocks.

[0055] S33: The Pearson relationship matrix, the Spearman rank relationship matrix, and the similarity relationship matrix are integrated into an attention relationship matrix A through the Attention mechanism; wherein the element A in the i-th row and j-th column of the attention relationship matrix ij Represents the strength of the relationship between the i-th stock and the j-th stock.

[0056] S34: Each element in the attention relationship matrix is ​​used as an edge value of the stock relationship graph, and the stock trend characteristics of each stock are used as stock nodes in the stock relationship graph to construct the stock relationship graph.

[0057] S35: Input the stock relationship graph into the graph convolutional network, and obtain the relationship embedding of each stock after two layers of graph convolution process; among which, the relationship embedding of the i-th stock is The relationship embedding of all s stocks is Represent the relation embeddings of the 1st, 2nd,…,sth stocks respectively.

[0058] S36: The stock trend characteristics of each stock and the relationship between each stock are embedded and spliced ​​together, and the spliced ​​results are processed through a fully connected layer to obtain the prediction results of the yield of each stock.

[0059] In this embodiment, the fully connected layer maps the concatenated result into a vector of size s, which is used to represent the predicted result of the yield of each stock.

[0060] It should be noted that in step S3, when training the stock selection model, the loss function used is Specifically:

[0061]

[0062]

[0063]

[0064] Where: L1 represents the prediction loss, L2 represents the relative ranking loss between each pair of stocks; represents the predicted results of all stock returns at time t+1; y t+1 represents the true value of all stock returns at time t+1; α represents the weight hyperparameter; They represent the predicted results of the returns of the i-th stock and the j-th stock at time t+1 respectively; They represent the true values ​​of the returns of the i-th stock and the j-th stock at time t+1 respectively.

[0065] It should also be noted that in step S3, the Adam optimizer is used to train the stock selection model.

[0066] S4: Obtain the stock feature data of the stocks to be tested that have undergone data preprocessing, input the stock feature data of the stocks to be tested into the trained stock selection model, use the stock selection model to predict the future rate of return of each stock to be tested, sort the stock feature data of the stocks to be tested according to the size of the future rate of return, and select the top K stocks to be tested with the highest rate of return for investment.

[0067] The present invention will now use a specific example to demonstrate the application effect of the stock selection method combining LSTM and multi-graph convolutional neural network described in S1 to S4 in the above embodiments on a specific data set, so as to facilitate understanding of the essence of the present invention.

[0068] Example

[0069] The specific implementation process of the stock selection method combining LSTM and multi-graph convolutional neural network adopted in this embodiment is as described above and will not be repeated here.

[0070] The experimental data used in this embodiment are stocks from NASDAQ and NYSE markets from January 2, 2013 to December 8, 2017, with the number of stocks being 1026 and 1737 respectively. Closing price, 5-days, 10-days, 20-days and 30-days average moving price are used as stock features. The entire data set is divided into three parts for training and validation of the model, the training set (2013-2015), validation set (2016), and test set (2017), which include 756, 252 and 237 trading days respectively. The specific details are shown in Table 1.

[0071] Table 1. Specific division details of the dataset

[0072]

[0073] The performance results of the stock selection model of the present invention on the test set are shown in Table 2. MSE is the prediction result error, which is 3.77e-4 in the NASDAQ market and 2.25e-4 in the NYSE market. MRR is a ranking index, reflecting the ranking accuracy of the prediction results, which is 4.99e-2 in the NASDAQ market and 4.74e-2 in the NYSE market. IRR is the cumulative return on investment, which is 1.98 in the NASDAQ market and 2.85 in the NYSE market.

[0074] Table 2. Performance of the stock selection model of the present invention on the test set

[0075] MSE MRR IRR NASDAQ 3.77e-4±1.39e-6 4.99e-2±2.67e-3 1.98±0.27 NYSE 2.25e-4±1.15e-6 4.74e-2±3.56e-3 2.85±0.19

[0076] It should also be noted that the stock selection method combining LSTM and multi-graph convolutional neural network in the above embodiment can essentially be executed by a computer program or module. Therefore, similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides a stock selection system combining LSTM and multi-graph convolutional neural network corresponding to the stock selection method combining LSTM and multi-graph convolutional neural network provided in the above embodiment, such as Figure 2 As shown, it includes:

[0077] A data acquisition module is used to acquire original stock feature data and classify the original stock feature data according to stock codes;

[0078] A preprocessing module is used to perform data preprocessing on the classified stock feature data to obtain the preprocessed stock feature data, and to construct a data set from the preprocessed stock feature data;

[0079] A model training module, used for training a stock selection model on the data set, wherein the stock selection model includes an LSTM model, a graph convolutional network, an Attention mechanism, and a fully connected layer;

[0080] The result acquisition module is used to obtain the stock feature data of the stocks to be tested after data preprocessing, input the stock feature data of the stocks to be tested into the trained stock selection model, and use the stock selection model to predict the future rate of return of each stock to be tested, sort the stock feature data of the stocks to be tested according to the size of the future rate of return, and select the top K stocks to be tested with the highest rate of return for investment.

[0081] It can be understood that the stock selection method combining LSTM and multi-graph convolutional neural network described in S1 to S4 above can be substantially implemented by a computer program. Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer program product corresponding to the stock selection method combining LSTM and multi-graph convolutional neural network provided in the above embodiment, which includes a computer program / instruction. When the computer program / instruction is executed by a processor, it can implement the stock selection method combining LSTM and multi-graph convolutional neural network as described in the above embodiment.

[0082] Similarly, based on the same inventive concept, another preferred embodiment of the present invention further provides a computer electronic device corresponding to the stock selection method combining LSTM and multi-graph convolutional neural network provided in the above embodiment, which includes a memory and a processor;

[0083] The memory is used to store computer programs;

[0084] The processor is used to implement a stock selection method combining LSTM and multi-graph convolutional neural network in the above embodiment when executing the computer program.

[0085] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0086] Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer-readable storage medium corresponding to the stock selection method combining LSTM and multi-graph convolutional neural network provided in the above embodiment, and a computer program is stored on the storage medium. When the computer program is executed by the processor, it can implement the stock selection method combining LSTM and multi-graph convolutional neural network in the above embodiment.

[0087] It is understandable that the above storage medium may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage. The storage medium may also be a U disk, a mobile hard disk, a magnetic disk or an optical disk, etc., which can store program codes.

[0088] It is understandable that the above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0089] It should also be noted that those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the various embodiments provided in this application, the division of steps or modules in the system and method is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules or steps can be combined or integrated together, and a module or step can also be split.

[0090] The above-described embodiment is only a preferred solution of the present invention, but it is not intended to limit the present invention. A person skilled in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A stock selection method combining LSTM and multi-graph convolutional neural network, characterized in that: The following steps are involved: S1: Obtain original stock feature data and classify the original stock feature data according to stock codes; S2: performing data preprocessing on the classified stock feature data to obtain preprocessed stock feature data, and constructing a data set from the preprocessed stock feature data; S3: training a stock selection model on the data set, wherein the stock selection model includes an LSTM model, a graph convolutional network, an Attention mechanism, and a fully connected layer; S4: Obtain the stock feature data of the stocks to be tested that have undergone data preprocessing, input the stock feature data of the stocks to be tested into the trained stock selection model, use the stock selection model to predict the future rate of return of each stock to be tested, sort the stock feature data of the stocks to be tested according to the size of the future rate of return, and select the top K stocks to be tested with the highest rate of return for investment.

2. A stock selection method combining LSTM and multi-graph convolutional neural network as claimed in claim 1, characterized in that: In step S2, the specific process of data preprocessing is: after the classified stock feature data is centralized and normalized in turn, the stock price moving average is calculated using the centralized and normalized stock feature data.

3. A stock selection method combining LSTM and multi-graph convolutional neural network as claimed in claim 1, characterized in that: In step S3, the specific processing flow in the stock selection model is as follows: S31: Input the preprocessed stock feature data into the LSTM model to extract the stock trend features of each stock; S32: Obtain the historical closing price sequence of the stock from the original stock feature data, calculate the Pearson correlation coefficient between each pair of stocks based on the historical closing price sequence of the stock, and obtain the Pearson relationship matrix, calculate the Spearman rank correlation coefficient between each pair of stocks based on the historical closing price sequence of the stock, and obtain the Spearman rank relationship matrix, and obtain the similarity between the historical closing price sequences of the stock through the FastDTW algorithm to obtain the similarity relationship matrix; S33: Integrate the Pearson relationship matrix, the Spearman rank relationship matrix, and the similarity relationship matrix into an attention relationship matrix through the Attention mechanism, where each element in the attention relationship matrix represents the strength of the relationship between two stocks of the corresponding index; S34: taking each element in the attention relationship matrix as an edge value of the stock relationship graph, and taking the stock trend characteristics of each stock as a stock node in the stock relationship graph, so as to construct the stock relationship graph; S35: Input the stock relationship graph into the graph convolutional network, and obtain the relationship embedding of each stock after performing two layers of graph convolution process; S36: The stock trend characteristics of each stock and the relationship between each stock are embedded and spliced ​​together, and the spliced ​​results are processed through a fully connected layer to obtain the prediction results of the yield of each stock.

4. A stock selection method combining LSTM and multi-graph convolutional neural network as claimed in claim 1, characterized in that: In step S3, when training the stock selection model, the loss function used is specifically: Where: L1 represents the prediction loss, L2 represents the relative ranking loss between each pair of stocks; represents the predicted results of all stock returns at time t+1; y t+1 represents the true value of all stock returns at time t+1; α represents the weight hyperparameter; They represent the predicted results of the returns of the i-th stock and the j-th stock at time t+1 respectively; They represent the true values ​​of the returns of the i-th stock and the j-th stock at time t+1 respectively.

5. A stock selection system combining LSTM and multi-graph convolutional neural network, characterized in that: include: A data acquisition module is used to acquire original stock feature data and classify the original stock feature data according to stock codes; A preprocessing module is used to perform data preprocessing on the classified stock feature data to obtain the preprocessed stock feature data, and to construct a data set from the preprocessed stock feature data; A model training module, used for training a stock selection model on the data set, wherein the stock selection model includes an LSTM model, a graph convolutional network, an Attention mechanism, and a fully connected layer; The result acquisition module is used to obtain the stock feature data of the stocks to be tested after data preprocessing, input the stock feature data of the stocks to be tested into the trained stock selection model, and use the stock selection model to predict the future rate of return of each stock to be tested, sort the stock feature data of the stocks to be tested according to the size of the future rate of return, and select the top K stocks to be tested with the highest rate of return for investment.

6. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, it can implement a stock selection method combining LSTM and multi-graph convolutional neural network as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, a stock selection method combining LSTM and multi-graph convolutional neural network as described in any one of claims 1 to 4 is implemented.

8. A computer electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is used to implement a stock selection method combining LSTM and multi-graph convolutional neural network as described in any one of claims 1 to 4 when executing the computer program.

Images