Information query method, device, electronic device and storage medium

Abstract

The present invention discloses an information query method, device, electronic device and storage medium, which relate to the field of artificial intelligence technology. The method includes: obtaining historical per-thousand-share income data of a target financial product; decomposing and processing the historical per-thousand-share income data to obtain a first data component and a second data component; using a first type of income prediction model, based on the first data component, determining a first prediction result of the per-thousand-share income of the target financial product in a future time period; using a second type of income prediction model, based on the second data component, determining a second prediction result of the per-thousand-share income of the target financial product in a future time period; fusing the first prediction result and the second prediction result to obtain a final per-thousand-share income prediction result, and feeding it back to the user. By adopting the present invention, the accuracy of the per-thousand-share income prediction can be improved, and the predicted per-thousand-share income can be fed back to the user as a query result, so that the user can query the per-thousand-share income of the financial product in the future time period.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an information query method, device, electronic device and storage medium. Background Art

[0002] With the development of financial technology, cash management products have become popular with clients due to their low risk, high liquidity, and stable returns. Cash management products do not have expected returns or investment periods, and issuers do not promise fixed returns. Returns are expressed as a seven-day annualized return or a return per 10,000 shares. The return per 10,000 shares determines the risk and return profile of the product and is a key factor in investor selection.

[0003] Currently, for most wealth management products, customers can only view historical returns per 10,000 shares, but cannot query forecasts for future returns. Therefore, how to enable customers to query forecasts for future returns for wealth management products has become a pressing technical challenge. Summary of the Invention

[0004] The present invention provides an information query method, device, electronic device, storage medium and computer program product.

[0005] According to one aspect of the present invention, there is provided an information query method, comprising:

[0006] In response to a user's query request for the return per ten thousand shares of a target financial product in a future time period, obtaining historical return per ten thousand shares data of the target financial product;

[0007] Decomposing the historical per-thousand-share return data to obtain a first data component and a second data component;

[0008] Determine, by using the pre-trained first-type yield prediction model and the first data component, a first prediction result of the yield per ten thousand shares of the target financial product in a future time period;

[0009] Determine, by using the pre-trained second-type income prediction model and the second data component, a second prediction result of the income per ten thousand shares of the target financial product in the future time period;

[0010] The first prediction result and the second prediction result are fused to obtain the final per-thousand-share revenue prediction result, and the final per-thousand-share revenue prediction result is fed back to the user.

[0011] Optionally, the historical per-10,000-share return data is decomposed to obtain a first data component and a second data component, including:

[0012] Perform wavelet transform on the historical per-thousand-share earnings data to obtain high-frequency and low-frequency components;

[0013] The low-frequency component is used as the first data component, and the high-frequency component is used as the second data component.

[0014] Optionally, the first type of revenue prediction model is a first gated recurrent neural network model pre-trained using 10,000 revenue sample data;

[0015] Using the pre-trained first-category income prediction model, a first prediction result of the income per ten thousand shares of the target wealth management product in the future time period is determined based on the first data component, including:

[0016] Through the first gated recurrent neural network model, the first prediction result of the per-thousand-share yield of the target financial product in the future time period is determined according to the low-frequency components in the historical per-thousand-share yield data.

[0017] Optionally, the second type of revenue prediction model includes a second gated recurrent neural network model and a gradient boosting decision tree model pre-trained using 10,000 revenue sample data;

[0018] The second type of pre-trained income prediction model is used to determine a second prediction result of the income per ten thousand shares of the target financial product in the future time period based on the second data component, including:

[0019] Using the second gated recurrent neural network model, a third prediction result of the return per 10,000 shares of the target financial product in the future time period is determined based on the high-frequency components in the historical return per 10,000 shares data;

[0020] Using the gradient boosting decision tree model, the fourth prediction result of the target financial product's return per 10,000 shares in the future time period is determined based on the high-frequency components in the historical return per 10,000 shares data;

[0021] The third prediction result and the fourth prediction result are fused by means of covariance cross fusion to obtain the second prediction result of the return per ten thousand shares of the target financial product in the future time period.

[0022] Optionally, the third prediction result and the fourth prediction result are fused by covariance cross fusion to obtain a second prediction result of the target financial product's return per ten thousand shares in the future time period, including

[0023] Obtaining a first estimated variance and a first fusion weight of a prediction error of a second gated recurrent neural network model determined during the model training process;

[0024] Obtaining a second estimated variance and a second fusion weight of a prediction error of a gradient boosting decision tree model determined during model training;

[0025] A second prediction result of the return per ten thousand shares of the target financial product in a future time period is determined based on the first estimated variance, the first fusion weight, the second estimated variance, the second fusion weight, the third prediction result and the fourth prediction result.

[0026] Optionally, before decomposing the historical per-10,000-share earnings data, the method further includes:

[0027] The historical per-thousand-share earnings data were cleaned, and the missing values ​​identified during the cleaning process were supplemented using the mean completion method.

[0028] Optionally, the first prediction result and the second prediction result are fused to obtain a final per-thousand-share revenue prediction result, and the final per-thousand-share revenue prediction result is fed back to the user, including:

[0029] The first prediction result and the second prediction result are added and merged at the corresponding time points to obtain the final per-thousand-share revenue prediction result;

[0030] The final per-thousand-share profit forecast results are visualized and displayed to users.

[0031] According to another aspect of the present invention, there is provided an information query device, comprising:

[0032] A data acquisition module is used to respond to a user's query request for the return per ten thousand shares of a target financial product in a future time period and obtain historical return per ten thousand shares data of the target financial product;

[0033] A decomposition processing module, configured to decompose the historical per-thousand-share return data to obtain a first data component and a second data component;

[0034] A first prediction module is configured to determine a first prediction result of the return per ten thousand shares of the target financial product in a future time period based on the first data component using a pre-trained first type of return prediction model;

[0035] A second prediction module is configured to determine a second prediction result of the return per ten thousand shares of the target financial product in a future time period based on the second data component using a pre-trained second type of return prediction model;

[0036] The fusion feedback module is used to fuse the first prediction result and the second prediction result to obtain the final per-thousand-share revenue prediction result, and feed back the final per-thousand-share revenue prediction result to the user.

[0037] Optionally, the decomposition processing module is specifically used to:

[0038] Perform wavelet transform on the historical per-thousand-share earnings data to obtain high-frequency and low-frequency components;

[0039] The low-frequency component is used as the first data component, and the high-frequency component is used as the second data component.

[0040] Optionally, the first type of revenue prediction model is a first gated recurrent neural network model pre-trained using 10,000 revenue sample data;

[0041] The first prediction module is specifically used for:

[0042] Through the first gated recurrent neural network model, the first prediction result of the per-thousand-share yield of the target financial product in the future time period is determined according to the low-frequency components in the historical per-thousand-share yield data.

[0043] Optionally, the second type of revenue prediction model includes a second gated recurrent neural network model and a gradient boosting decision tree model pre-trained using 10,000 revenue sample data;

[0044] The second prediction module includes:

[0045] The first prediction unit is configured to determine a third prediction result of the return per 10,000 shares of the target wealth management product in a future time period based on the high-frequency components in the historical return per 10,000 shares data using the second gated recurrent neural network model;

[0046] The second prediction unit is configured to determine a fourth prediction result of the return per 10,000 shares of the target wealth management product in a future time period based on the high-frequency components in the historical return per 10,000 shares data by using a gradient boosting decision tree model;

[0047] The fusion unit is used to fuse the third prediction result and the fourth prediction result by covariance cross fusion to obtain a second prediction result of the target financial product's return per ten thousand shares in the future time period.

[0048] Optionally, the fusion unit is specifically used to:

[0049] Obtaining a first estimated variance and a first fusion weight of a prediction error of a second gated recurrent neural network model determined during the model training process;

[0050] Obtaining a second estimated variance and a second fusion weight of a prediction error of a gradient boosting decision tree model determined during model training;

[0051] A second prediction result of the return per ten thousand shares of the target financial product in a future time period is determined based on the first estimated variance, the first fusion weight, the second estimated variance, the second fusion weight, the third prediction result and the fourth prediction result.

[0052] Optionally, the device further includes:

[0053] The cleaning processing module is used to clean the historical per-thousand-share return data before decomposing the data, and to supplement the missing values ​​determined during the cleaning process by the mean completion method.

[0054] Optionally, the fusion feedback module is specifically used to:

[0055] The first prediction result and the second prediction result are added and merged at the corresponding time points to obtain the final per-thousand-share revenue prediction result;

[0056] The final per-thousand-share profit forecast results are visualized and displayed to users.

[0057] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0058] at least one processor; and

[0059] a memory communicatively connected to at least one processor; wherein,

[0060] The memory stores a computer program that can be executed by at least one processor. The computer program is executed by the at least one processor so that the at least one processor can execute the information query method according to the embodiment of the present invention.

[0061] According to another aspect of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the information query method according to an embodiment of the present invention when executed.

[0062] According to another aspect of the present invention, a computer program product is provided, comprising a computer program, which implements the steps of the above method when executed by a processor.

[0063] The technical solution of the embodiment of the present invention utilizes multiple deep learning-based profit prediction models to predict different components of the historical per-thousand-share profit data of the target wealth management product, and integrates multiple prediction results to obtain the final per-thousand-share profit prediction result, thereby improving the accuracy of predicting the per-thousand-share profit in future time periods. The predicted per-thousand-share profit is fed back to the user as a query result, enabling the user to query the per-thousand-share profit of the wealth management product in the future time period.

[0064] 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

[0065] 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.

[0066] Figure 1 This is a flow chart of an information query method provided by an embodiment of the present invention;

[0067] Figure 2 is a flow chart of another information query method provided by an embodiment of the present invention;

[0068] Figure 3a is a flow chart of another information query method provided by an embodiment of the present invention;

[0069] Figure 3b This is a logic block diagram of an information query method provided by an embodiment of the present invention;

[0070] Figure 4 This is a schematic structural diagram of an information query device provided by an embodiment of the present invention;

[0071] Figure 5 It is a structural diagram of an electronic device for implementing the information query method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0072] 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.

[0073] Example 1

[0074] Figure 1 A flowchart of an information query method is provided for an embodiment of the present invention. This embodiment is applicable to scenarios where the per-thousand-share yield of financial products is predicted. The method can be executed by an information query device, which can be implemented in the form of hardware and / or software, and can be configured in an electronic device.

[0075] like Figure 1 As shown, the information query method includes:

[0076] S101. In response to a user's query request for the return per ten thousand shares of a target financial product in a future time period, obtain historical return per ten thousand shares data of the target financial product.

[0077] In an embodiment of the present invention, the target financial product may be a cash management product. Cash management products are defined as products with low risk, high liquidity, and stable returns. The future time period refers to a number of days starting from tomorrow, based on the current time, such as the next three days, the next seven days, etc. The specific number of days can be determined by the user based on their query needs. To facilitate user queries, a financial product yield per 10,000 shares query interface can be displayed to the user. This interface generates a corresponding query request based on the product name / product code of the target financial product and the specific future time period entered by the user in the text input box of the query interface. Alternatively, a corresponding query request can be generated based on the product name / product code of the target financial product and the specific future time period entered by the user via voice. In response to the user's query request for the yield per 10,000 shares of the target financial product in the future time period, historical yield per 10,000 shares data for the target financial product can be obtained from a pre-specified financial data platform. For example, the historical yield per 10,000 shares data can include the yield per 10,000 shares data for the target financial product over a past period of time, such as the past week or month. It will be understood that historical yield per 10,000 shares data is essentially time series data.

[0078] S102: Decompose the historical return data per ten thousand shares to obtain a first data component and a second data component.

[0079] In an embodiment of the present invention, the historical per-thousand-share revenue data obtained through step S101 is data with a relatively high overall dimension and complexity, and contains information of various frequencies and properties. If a revenue prediction model is used to directly predict the historical per-thousand-share revenue data, there will be a problem of low prediction accuracy. Based on this, the present invention proposes to first decompose the historical per-thousand-share revenue data to obtain a first data component and a second data component. Exemplarily, any of the following methods can be used to decompose the historical per-thousand-share revenue data: a wavelet transform method, a principal component analysis method, and a singular value decomposition method. It can be understood that by decomposing the historical per-thousand-share revenue data, the complexity of the first data component and the second data component obtained by decomposition is reduced, and when predictions are subsequently made based on the low-complexity data components, the accuracy of the prediction results can be guaranteed.

[0080] S103. Determine a first prediction result of the return per ten thousand shares of the target financial product in a future time period according to the first data component through the pre-trained first type of return prediction model.

[0081] In an embodiment of the present invention, the first type of income prediction model is a pre-trained deep learning model that can predict the income per ten thousand shares based on the first data component. During the prediction, the first data component can be input into the first type of income prediction model, and the first prediction result of the income per ten thousand shares of the target financial product in the future time period can be determined based on the model output; wherein the first prediction result is also time series data. It should be noted that the number of first type of income prediction models can be one or more. In the case where there are multiple first type of income prediction models, the output results of each model can be fused to obtain the final first prediction result.

[0082] S104. Determine a second prediction result of the return per ten thousand shares of the target financial product in a future time period according to the second data component through the pre-trained second type of return prediction model.

[0083] In an embodiment of the present invention, the second type of income prediction model is a pre-trained deep learning model that can predict the income per ten thousand shares based on the second data component. During the prediction, the second data component can be input into the second type of income prediction model, and the second prediction result of the income per ten thousand shares of the target financial product in the future time period can be determined based on the model output; wherein the second prediction result is also time series data. It should be noted that the number of the second type of income prediction models can be one or more. In the case where there are multiple second type of income prediction models, the output results of each model can be fused to obtain the final second prediction result.

[0084] S105: Fusing the first prediction result and the second prediction result to obtain a final per-thousand-share revenue prediction result, and feeding back the final per-thousand-share revenue prediction result to the user.

[0085] In an embodiment of the present invention, through steps S103 and S104, different data components of the historical per-thousand-share return data are used to perform predictions respectively, obtaining a first prediction result and a second prediction result which are essentially time series data. Subsequently, it is only necessary to perform addition and fusion of the first prediction result and the second prediction result obtained in the two steps at the corresponding time points to obtain the final per-thousand-share return prediction result, and the final per-thousand-share return prediction result is fed back to the user as a query result, thereby completing a user's query on the prediction result of the per-thousand-share return of the financial product in the future time period. It should be noted that the final per-thousand-share return prediction result is also a time series data. This time series data can be directly fed back to the user as a query result, or this time series data can be visualized (for example, a line graph is drawn based on the time series data), and the visualization result is fed back to the user as a query result.

[0086] The present invention utilizes multiple deep learning-based return prediction models to predict different components of the historical per-thousand-share return data of the target financial product, and fuses multiple prediction results to obtain the final per-thousand-share return prediction result, thereby improving the accuracy of predicting the per-thousand-share return in future time periods. The predicted per-thousand-share return is fed back to the user as a query result, enabling the user to query the per-thousand-share return of the financial product in the future time period.

[0087] Example 2

[0088] Figure 2 The present invention provides a flowchart of an information query method. Figure 2 , the method comprises the following steps:

[0089] S201. In response to a user's query request for the return per ten thousand shares of a target financial product in a future time period, obtain historical return per ten thousand shares data of the target financial product.

[0090] S202. Perform wavelet transform on the historical per-thousand-share return data to obtain high-frequency components and low-frequency components; use the low-frequency components as the first data component and the high-frequency components as the second data component.

[0091] In an embodiment of the present invention, in order to reduce the complexity of the historical per-thousand-share return data, a wavelet transform method is used to decompose the historical per-thousand-share return data to obtain high-frequency components and low-frequency components. The low-frequency component represents the long-term trend in the historical per-thousand-share return data and reflects the overall direction of the return of the financial product over a long period of time. The high-frequency component contains short-term fluctuation information in the historical per-thousand-share return data. The low-frequency component is used as the first data component, and the high-frequency component is used as the second data component.

[0092] In an embodiment of the present invention, to predict the per-thousand-share return of a target financial product over a future time period, two types of return prediction models are pre-selected and trained. The first type of return prediction model is a first gated recurrent neural network model pre-trained using sample data on per-thousand-share returns. The second type of return prediction model consists of two models: a second gated recurrent neural network model pre-trained using sample data on per-thousand-share returns and a gradient boosted decision tree model. To train the first gated recurrent neural network model, a gated recurrent neural network is first constructed, and the number of input nodes, hidden nodes, and output nodes is determined, for example, 256 hidden nodes and 1 output node. The Relu function is used as the activation function in the hidden layer, and the model training cycle is set to 1000 times. In this embodiment of the present invention, the structure of the second gated recurrent neural network model is the same as that of the first gated recurrent neural network model. The gradient boosted decision tree model is composed of a gradient boosted decision tree network, and the model is also trained 1000 times. On this basis, the first gated recurrent neural network model, the second gated recurrent neural network model, and the gradient boosted decision tree model can be trained simultaneously. The training process includes: first, obtaining the historical per-thousand-share return data of the financial product, and performing wavelet transform on the historical per-thousand-share return data to obtain high-frequency components and low-frequency components; wherein, the high-frequency components and low-frequency components are essentially time series data; for the low-frequency components, a plurality of per-thousand-share return sample data are constructed through the window method, for example, the low-frequency components are the following time series data: [(t1, x1), (t2, x2) ... (t n , x n )]; where t n represents time, x nRepresents the low-frequency component of ten thousand shares of revenue; the low-frequency component of the 1st to 5th elements in the time series data is taken as a ten thousand share of revenue sample, and the low-frequency component of the 6th to 8th elements is taken as the real result corresponding to the ten thousand share of revenue sample; the low-frequency component of the 2nd to 6th elements in the time series data is taken as a ten thousand share of revenue sample, and the low-frequency component of the 7th to 9th elements is taken as the real result corresponding to the ten thousand share of revenue sample; and so on, a plurality of ten thousand share of revenue training samples based on low-frequency components can be obtained; similarly, according to the above method, a plurality of ten thousand share of revenue samples based on high-frequency components can be constructed. On this basis, the ten thousand share of revenue training samples based on low-frequency components are input into the first gated recurrent neural network model, and the loss between the output of the first gated recurrent neural network model and the real result corresponding to the ten thousand share of revenue training samples is calculated; the parameters of the first gated recurrent neural network model are adjusted inversely according to the loss; when the number of training times reaches the preset number, the training is stopped to obtain the first type of revenue prediction model of the present invention. Similarly, for the second gated recurrent neural network model and the gradient boosting decision tree model, ten thousand profit training samples based on high-frequency components are input into the second gated recurrent neural network model and the gradient boosting decision tree model respectively, and the first loss between the output of the second gated recurrent neural network model and the true results corresponding to the ten thousand profit training samples is calculated, and the second loss between the output of the gradient boosting decision tree model and the true results corresponding to the ten thousand profit training samples is calculated; the parameters of the second gated recurrent neural network model are reversely adjusted according to the first loss, and the parameters of the gradient boosting decision tree model are reversely adjusted according to the second loss; when the number of training times reaches the preset number, the training is stopped to obtain two second-category profit prediction models.

[0093] On the basis of the above, a validation set consisting of 10,000 training samples of returns based on high-frequency components is constructed. The 10,000 training samples of returns in the validation set are sequentially input into the second gated recurrent neural network model and the gradient boosting decision tree model, and the estimated variance of the prediction errors of the two models is calculated. Specifically, it can be calculated according to the following formula: Among them, t is the time step, i is the model number, and n is the number of samples in the validation set. is the actual result corresponding to 10,000 income training samples, The model outputs the prediction results based on 10,000 income samples; p i is the estimated variance of the model with serial number i. For example, the serial number of the second gated recurrent neural network model is 1, and the serial number of the gradient boosting decision tree model is 2, then p1 is the calculated first estimated variance, and p2 is the calculated second estimated variance. Further, based on the estimated variance of the prediction error of each model, the covariance of all models is calculated, which can be specifically performed according to the following formula: Among them, p -1 is the covariance, m is the number of models, w irepresents the fusion weight of the i-th model. Since minimizing the covariance means that the estimated value and the stabilization effect are balanced, the covariance calculation is converted into an optimization problem. Sequential least squares optimization is used to obtain the optimal first fusion weight for the second gated recurrent neural network model and the optimal second fusion weight for the gradient boosted decision tree model. Subsequently, the prediction results of the two models can be cross-fused based on the first estimated variance and first fusion weight of the prediction error of the second gated recurrent neural network model and the second estimated variance and second fusion weight of the prediction error of the gradient boosted decision tree model.

[0094] On this basis, a first prediction result of the target financial product's return per 10,000 shares in the future time period is determined based on the first data component using the pre-trained first-type return prediction model, which can be performed according to step S203. A second prediction result of the target financial product's return per 10,000 shares in the future time period is determined based on the second data component using the pre-trained second-type return prediction model, which can be performed according to steps S204-S206.

[0095] S203. Determine a first prediction result of the per-thousand-share return of the target financial product in a future time period based on the low-frequency components in the historical per-thousand-share return data using a first gated recurrent neural network model.

[0096] In an embodiment of the present invention, the low-frequency component in the historical per-thousand-share return data can be input into the first gated recurrent neural network model, and the first prediction result of the per-thousand-share return of the target financial product in the future time period can be determined based on the output of the first gated recurrent neural network model.

[0097] In this embodiment of the present invention, since the second type of revenue prediction model includes a second gated recurrent neural network model and a gradient boosted decision tree model pre-trained using 10,000 revenue training sample data, it is necessary to use these two models to perform predictions separately and then fuse the prediction results to obtain the second prediction result. See steps S204-S206 for details.

[0098] S204. Determine a third prediction result of the return per ten thousand shares of the target financial product in the future time period based on the high-frequency components in the historical return per ten thousand shares data through the second gated recurrent neural network model.

[0099] In an embodiment of the present invention, the high-frequency components in the historical per-thousand-share yield data can be input into the second gated recurrent neural network model, and based on the output of the second gated recurrent neural network model, a third prediction result of the per-thousand-share yield of the target financial product in the future time period can be determined.

[0100] S205. Determine a fourth prediction result of the return per ten thousand shares of the target financial product in the future time period based on the high-frequency components in the historical return per ten thousand shares data by using a gradient boosting decision tree model.

[0101] In an embodiment of the present invention, the high-frequency components in the historical per-thousand-share yield data can be input into a gradient boosting decision tree model, and based on the output of the gradient boosting decision tree model, the fourth prediction result of the per-thousand-share yield of the target financial product in the future time period can be determined.

[0102] S206. The third prediction result and the fourth prediction result are integrated by means of covariance cross-fusion to obtain a second prediction result of the return per ten thousand shares of the target financial product in the future time period.

[0103] In an optional implementation scheme, first, the first estimated variance and first fusion weight of the prediction error of the second gated recurrent neural network model determined during the model training process are obtained; the second estimated variance and second fusion weight of the prediction error of the gradient boosting decision tree model determined during the model training process are obtained; wherein, the process of determining the first estimated variance, the first fusion weight, the second estimated variance, and the second fusion weight can be referred to the above description. Then, based on the first estimated variance, the first fusion weight, the second estimated variance, the second fusion weight, the third prediction result, and the fourth prediction result, a second prediction result of the return per 10,000 shares of the target financial product in the future time period is determined. Exemplarily, the inverse of the first estimated variance, the product of the first fusion weight, and the third prediction result are first calculated, and then the inverse of the second estimated variance, the product of the second fusion weight, and the fourth prediction result are calculated; finally, the two product results are added together to obtain the second prediction result of the return per 10,000 shares of the target financial product in the future time period.

[0104] S207: Fusing the first prediction result and the second prediction result to obtain a final per-thousand-share revenue prediction result, and feeding back the final per-thousand-share revenue prediction result to the user.

[0105] Optionally, the first prediction result and the second prediction result are added and fused at corresponding time points to obtain a final per-thousand-share revenue prediction result, and the final per-thousand-share revenue prediction result is fed back to the user as a query result.

[0106] In an embodiment of the present invention, multiple deep learning models are used to predict different components of the historical per-thousand-share earnings data, and then the prediction results are integrated to obtain the final prediction result, which can ensure the accuracy of the prediction result.

[0107] Example 3

[0108] Figure 3a The present invention provides a flowchart of an information query method. Figure 3a, the method comprises the following steps:

[0109] S301. In response to a user's query request for the return per ten thousand shares of a target financial product in a future time period, obtain historical return per ten thousand shares data of the target financial product.

[0110] S302. Clean the historical per-thousand-share earnings data and supplement the missing values ​​identified during the cleaning process using the mean completion method.

[0111] In the embodiments of the present invention, the historical per-thousand-share revenue data is time series data. Cleaning the historical per-thousand-share revenue data primarily includes: identifying outliers in the historical per-thousand-share revenue data and deleting or replacing the identified outliers; or identifying missing values ​​in the historical per-thousand-share revenue data and supplementing the missing values, for example, by using a mean-complementation method to supplement the missing values ​​identified during the cleaning process. It is understood that by cleaning the acquired historical per-thousand-share revenue data, the integrity and accuracy of the data can be ensured.

[0112] S303: Decompose the historical return data per ten thousand shares to obtain a first data component and a second data component.

[0113] S304: Determine a first prediction result of the return per ten thousand shares of the target financial product in a future time period according to the first data component through the pre-trained first type of return prediction model.

[0114] S305. Determine a second prediction result of the return per ten thousand shares of the target financial product in a future time period according to the second data component through the pre-trained second type of return prediction model.

[0115] In the embodiment of the present invention, the specific implementation process of steps S303-S305 can be found in the description of the above embodiment and will not be repeated here.

[0116] S306: Add and fuse the first prediction result and the second prediction result at corresponding time points to obtain the final per-thousand-share revenue prediction result.

[0117] Since the first and second prediction results are essentially time series data, the final per-thousand-share revenue forecast result can be obtained by adding and fusing the two prediction results at corresponding time points. It is understandable that the final per-thousand-share revenue forecast result is also time series data.

[0118] S307: Visualize the final per-thousand-share revenue forecast result and present the visualization result to the user.

[0119] In an embodiment of the present invention, the final per-thousand-share return forecast result is visualized, that is, the per-thousand-share return forecast result, which is numerical time series data, is converted into a chart (such as a line chart or a bar chart). Exemplarily, a visualization conversion instruction is constructed, and the visualization conversion instruction and the final per-thousand-share return forecast result are input into a large language model. Based on the output of the large language model, a visualization result (including a per-thousand-share return forecast value chart) is obtained; wherein, the visualization conversion instruction can be exemplarily as follows: "This data is the per-thousand-share return forecast data for the next X days for the XX financial product. Your task is to convert the per-thousand-share return forecast data into a visual chart."

[0120] When presenting the visualization results to users, the following methods can be used: The charts included in the visualization results can be embedded into a designated webpage, which can then automatically redirect the user to the webpage to present the forecast results per 10,000 shares in a graphical format; or a link to the webpage can be sent to the user, allowing the user to click the link to view the forecast results per 10,000 shares in a graphical format. Alternatively, the charts included in the visualization results can be inserted into a relevant report document, accompanied by textual descriptions explaining the content displayed by the visualization charts and the key information of the forecast results, so that the document can be presented to the user in the form of a formal report.

[0121] In an embodiment of the present invention, by visualizing and displaying the final per-thousand-share revenue forecast results, users can intuitively understand the possible future changing trends of per-thousand-share revenue compared to directly viewing the per-thousand-share revenue forecast results in the form of a time series.

[0122] Further, in order to explain the process of the present invention in detail, see Figure 3b , which shows the logical block diagram of the information query method. The specific prediction process includes: obtaining the historical per-thousand-share return data of the target financial product; cleaning the historical per-thousand-share return data and supplementing missing values ​​during the cleaning process; performing wavelet transform processing on the cleaned historical per-thousand-share return data to obtain high-frequency components and low-frequency components; inputting the low-frequency components into the first gated recurrent neural network model for prediction, and simultaneously inputting the high-frequency components into the second gated recurrent neural network model and the gradient boosting decision tree model for prediction; performing covariance cross-fusion on the prediction results output by the second gated recurrent neural network model and the gradient boosting decision tree model; adding and fusing the covariance cross-fusion result with the prediction result output by the first gated recurrent neural network model at the corresponding time to obtain the final per-thousand-share return prediction result, and presenting it to the user in a visual manner, thereby completing the query of the per-thousand-share return of the financial product in the future time period.

[0123] Example 4

[0124] Figure 4This is a schematic diagram of the structure of an information query device provided by an embodiment of the present invention. This embodiment is applicable to the scenario of predicting the return per ten thousand of financial products. The device can execute any information query method of the present invention. Figure 4 As shown, the information query device includes:

[0125] The data acquisition module 401 is configured to respond to a user's query request for the return per ten thousand shares of a target financial product in a future time period and acquire historical return data per ten thousand shares of the target financial product;

[0126] Decomposition processing module 402, for decomposing the historical per-thousand-share return data to obtain a first data component and a second data component;

[0127] The first prediction module 403 is configured to determine a first prediction result of the return per 10,000 shares of the target financial product in a future time period based on the first data component using a pre-trained first type of return prediction model;

[0128] The second prediction module 404 is configured to determine a second prediction result of the return per 10,000 shares of the target financial product in a future time period based on the second data component using the pre-trained second type of return prediction model;

[0129] The fusion feedback module 405 is used to fuse the first prediction result and the second prediction result to obtain the final per-thousand-share revenue prediction result, and feed back the final per-thousand-share revenue prediction result to the user.

[0130] In some embodiments, the decomposition processing module 402 is specifically configured to:

[0131] Perform wavelet transform on the historical per-thousand-share earnings data to obtain high-frequency and low-frequency components;

[0132] The low-frequency component is used as the first data component, and the high-frequency component is used as the second data component.

[0133] In some embodiments, the first type of revenue prediction model is a first gated recurrent neural network model pre-trained using 10,000 revenue sample data;

[0134] The first prediction module 403 is specifically used for:

[0135] Through the first gated recurrent neural network model, the first prediction result of the per-thousand-share yield of the target financial product in the future time period is determined according to the low-frequency components in the historical per-thousand-share yield data.

[0136] In some embodiments, the second type of revenue prediction model includes a second gated recurrent neural network model and a gradient boosting decision tree model pre-trained using 10,000 revenue sample data;

[0137] The second prediction module 404 includes:

[0138] The first prediction unit is configured to determine a third prediction result of the return per 10,000 shares of the target wealth management product in a future time period based on the high-frequency components in the historical return per 10,000 shares data using the second gated recurrent neural network model;

[0139] The second prediction unit is configured to determine a fourth prediction result of the return per 10,000 shares of the target wealth management product in a future time period based on the high-frequency components in the historical return per 10,000 shares data by using a gradient boosting decision tree model;

[0140] The fusion unit is used to fuse the third prediction result and the fourth prediction result by covariance cross fusion to obtain a second prediction result of the target financial product's return per ten thousand shares in the future time period.

[0141] In some embodiments, the fusion unit is specifically configured to:

[0142] Obtaining a first estimated variance and a first fusion weight of a prediction error of a second gated recurrent neural network model determined during the model training process;

[0143] Obtaining a second estimated variance and a second fusion weight of a prediction error of a gradient boosting decision tree model determined during model training;

[0144] A second prediction result of the return per ten thousand shares of the target financial product in a future time period is determined based on the first estimated variance, the first fusion weight, the second estimated variance, the second fusion weight, the third prediction result and the fourth prediction result.

[0145] In some embodiments, the apparatus further comprises:

[0146] The cleaning processing module is used to clean the historical per-thousand-share return data before decomposing the data, and to supplement the missing values ​​determined during the cleaning process by the mean completion method.

[0147] In some embodiments, the fusion feedback module 405 is specifically configured to:

[0148] The first prediction result and the second prediction result are added and merged at the corresponding time points to obtain the final per-thousand-share revenue prediction result;

[0149] The final per-thousand-share profit forecast results are visualized and displayed to users.

[0150] The information query device provided by the embodiment of the present invention can execute the information query method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0151] According to an embodiment of the present invention, the present invention further provides an electronic device, a readable storage medium and a computer program product.

[0152] Example 5

[0153] Figure 5 The components, their connections and relationships, and their functions shown herein are merely examples and are not intended to limit implementation of the invention described and / or claimed herein.

[0154] like Figure 5 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.

[0155] Multiple components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16; 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.

[0156] The processor 11 may be any general-purpose and / or specialized processing component with processing and computing capabilities. Examples of the 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 for running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as executing the information query method.

[0157] In some embodiments, the information query method can be implemented as a computer program that 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 can 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 information query method described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to perform the information query method in any other appropriate manner (for example, by means of firmware).

[0158] 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), system-on-chip systems (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.

[0159] Computer programs for implementing the methods 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 information query device, so that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0160] 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.

[0161] 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).

[0162] 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.

[0163] 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.

[0164] 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.

[0165] 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. An information query method, characterized in that: include: In response to a user's query request for the return per ten thousand shares of a target financial product in a future time period, obtaining historical return per ten thousand shares data of the target financial product; Performing wavelet transform on the historical per-thousand-share earnings data to obtain high-frequency components and low-frequency components; Determining a first prediction result of the ten thousand share return of the target financial product in a future time period based on a low-frequency component in the historical ten thousand share return data by using a first gated recurrent neural network model pre-trained using ten thousand share return sample data; Determining a third prediction result of the ten thousand share return of the target wealth management product in a future time period based on high-frequency components in the historical ten thousand share return data by using a second gated recurrent neural network model pre-trained using the ten thousand share return sample data; Determining a fourth prediction result of the return per 10,000 shares of the target wealth management product in a future time period based on high-frequency components in the historical return per 10,000 shares data by using a gradient boosting decision tree model pre-trained using the return per 10,000 shares sample data; The third prediction result and the fourth prediction result are integrated by means of covariance cross fusion to obtain a second prediction result of the return per ten thousand of the target financial product in the future time period; The first prediction result and the second prediction result are fused to obtain a final per-thousand-share revenue prediction result, and the final per-thousand-share revenue prediction result is fed back to the user.

2. The method according to claim 1, characterized in that The third prediction result and the fourth prediction result are fused by covariance cross fusion to obtain a second prediction result of the target financial product's return on ten thousand shares in the future time period, including Obtaining a first estimated variance and a first fusion weight of a prediction error of the second gated recurrent neural network model determined during the model training process; Obtaining a second estimated variance and a second fusion weight of the prediction error of the gradient boosting decision tree model determined during the model training process; A second prediction result of the return per ten thousand shares of the target financial product in a future time period is determined based on the first estimated variance, the first fusion weight, the second estimated variance, the second fusion weight, the third prediction result and the fourth prediction result.

3. The method according to claim 1, characterized in that Before performing wavelet transform processing on the historical 10,000-share earnings data, the method further includes: The historical per-thousand-share earnings data is cleaned, and the missing values ​​identified during the cleaning process are supplemented using the mean completion method.

4. The method according to claim 1, wherein The fusing the first prediction result and the second prediction result to obtain a final per-thousand-share revenue prediction result, and feeding back the final per-thousand-share revenue prediction result to the user, includes: Adding and fusing the first prediction result and the second prediction result at corresponding time points to obtain a final per-thousand-share revenue prediction result; The final per-thousand-share revenue forecast result is visualized and the visualization result is presented to the user.

5. An information query device, characterized in that: include: A data acquisition module is configured to respond to a user's query request for the return per ten thousand shares of a target financial product in a future time period and acquire historical return per ten thousand shares data of the target financial product; A decomposition processing module, configured to perform wavelet transform processing on the historical 10,000-share return data to obtain high-frequency components and low-frequency components; A first prediction module is configured to determine a first prediction result of the per-thousand-share return of a target wealth management product in a future time period based on a low-frequency component in the historical per-thousand-share return data by using a first gated recurrent neural network model pre-trained using per-thousand-share return sample data; The second prediction module is used to determine a third prediction result of the return per ten thousand shares of the target wealth management product in a future time period based on the high-frequency components in the historical return per ten thousand shares data by using a second gated recurrent neural network model pre-trained using the return per ten thousand shares sample data; determine a fourth prediction result of the return per ten thousand shares of the target wealth management product in a future time period based on the high-frequency components in the historical return per ten thousand shares data by using a gradient boosting decision tree model pre-trained using the return per ten thousand shares sample data; and fuse the third prediction result and the fourth prediction result by covariance cross fusion to obtain a second prediction result of the return per ten thousand shares of the target wealth management product in the future time period; A fusion feedback module is used to fuse the first prediction result and the second prediction result to obtain a final per-thousand-share revenue prediction result, and feed back the final per-thousand-share revenue prediction result to the user.

6. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor. The computer program is executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method according to any one of claims 1 to 4 when executed.

8. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 4.

Images