Recommendation Method and Device for Financial Products, Storage Medium, Electronic Device

Abstract

The present invention discloses a method and device for recommending financial products, a storage medium, and an electronic device, which relate to the field of artificial intelligence. The method includes: obtaining feature information associated with specified fund features of a target object within a historical time period, where the feature information at least includes: a set of dynamic feature values, a set of trend feature values, a set of quantitative feature values, and a set of static feature values; inputting the set of dynamic feature values, the set of trend feature values, the set of quantitative feature values, and the set of static feature values into a feature integration model corresponding to the specified fund features to obtain an integrated feature value output by the feature integration model; and inputting the integrated feature values output by all feature integration models into a product recommendation model to obtain recommendation information output by the product recommendation model. The present invention solves the technical problem in the related art that the loss rate of data information used in the recommendation strategy for financial products is high, resulting in inaccurate recommendation directions.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular, to a method and device for recommending financial products, a storage medium, and an electronic device. Background Art

[0002] Due to the fact that marketing entities in the market economy are greatly affected by special environments (such as: epidemics, earthquakes, etc.) in a short period of time, however, when the influence of special environmental factors weakens, the customer acquisition models of marketing entities in the market economy will also change significantly accordingly. At the same time, the capital trends, transaction models, etc. of marketing entities can all undergo stage-like changes, but this change has brought new impacts and challenges to the traditional marketing prediction models.

[0003] Currently, in terms of customer acquisition, financial institutions are mainly divided into offline direct sales and online recommendations for recommending products to the main users. How to identify and activate potential customers has high practical significance for the customer acquisition level of financial institutions, while reducing labor costs and improving the operating efficiency of financial institutions.

[0004] In the related art, at the level of financial institution marketing models, generally models such as clustering analysis, scoring card models, and logistic regression are used, and the model construction is mainly based on data such as customer historical transaction information and transaction flow information. Among them, when screening marketing objects in the clustering analysis model, affected by special environmental changes, it can directly interfere with the screening of the target customer group. For example, in the special background of the epidemic environment, some basic transaction flows may be restricted due to policy reasons, resulting in possible different morphological changes in the clustering part, directly interfering with the screening of the target customer group. And for models such as logistic regression and scoring cards, generally, the target customer group is screened based on static features, but this method cannot capture the potential transaction logic of customers, and there are disadvantages of data information usage loss, resulting in inaccurate financial products recommended to customers.

[0005] In view of the above problems, no effective solution has been proposed yet. Summary of the Invention

[0006] Embodiments of the present invention provide a method and device for recommending financial products, a storage medium, and an electronic device, so as to at least solve the technical problem in the related art that the recommendation strategy for financial products has a high data information usage loss rate, resulting in inaccurate recommendation directions.

[0007] According to one aspect of an embodiment of the present invention, there is provided a method for recommending financial products, including: obtaining feature information related to a specified fund feature of a target object within a historical time period, where the feature information at least includes: a set of dynamic feature values, a set of trend feature values, a set of quantization feature values, and a set of static feature values, the set of trend feature values and the set of quantization feature values are calculated based on the set of dynamic feature values, the set of dynamic feature values includes: a plurality of asset volatility parameters of financial products for transactions, and the set of static feature values includes: a plurality of asset transaction parameters of the financial products for transactions within the historical time period; inputting the set of dynamic feature values, the set of trend feature values, the set of quantization feature values, and the set of static feature values into a feature integration model corresponding to the specified fund feature to obtain an integrated feature value output by the feature integration model, where each specified fund feature corresponds to the feature integration model; inputting the integrated feature values output by all the feature integration models into a product recommendation model to obtain recommendation information output by the product recommendation model, where the recommendation information at least includes: an object to be recommended and a financial product to be recommended.

[0008] Optionally, the step of obtaining feature information related to a specified fund feature of a target object within a historical time period includes: slicing the historical time period to obtain a plurality of historical time domain intervals; obtaining transaction information of the target object for generating fund transactions within each historical time domain interval, and calculating average asset data corresponding to each historical time domain interval; determining reference asset data corresponding to each historical time domain interval based on the transaction information and a preselected anchor time; and calculating the feature information of the specified fund feature by using a preset calculation strategy for the average asset data within each historical time domain interval and the corresponding reference asset data.

[0009] Optionally, the historical time domain intervals at least include: a plurality of first time domain intervals and a plurality of second time domain intervals, where the interval duration of the first time domain intervals is less than the interval duration of the second time domain intervals.

[0010] Optionally, the step of calculating average asset data corresponding to each historical time domain interval includes: calculating first average asset data of the target object within each first time domain interval based on the transaction information of the target object for generating fund transactions within the first time domain interval; and calculating second average asset data of the target object within each second time domain interval based on the transaction information of the target object for generating fund transactions within the second time domain interval.

[0011] Optionally, the step of calculating the average asset data and the corresponding benchmark asset data in each of the historical time domain intervals by adopting a pre-designed calculation strategy to determine the characteristic information of the specified fund characteristic includes: the average asset data in each of the multiple historical time domain intervals forms a dynamic eigenvalue set of the specified fund characteristic; any one of the average asset data in the dynamic eigenvalue set is selected as the target average asset data and compared with the corresponding benchmark asset data; when the target average asset data is greater than the corresponding benchmark asset data, determining that the trend eigenvalue of the target average asset data is the first trend value; when the target average asset data is less than or equal to the corresponding benchmark asset data, determining that the trend eigenvalue of the target average asset data is the second trend value; taking the difference between the target average asset data and the benchmark asset data to determine the quantization eigenvalue of the target average asset data; determining the trend eigenvalue and the quantization eigenvalue of each of the average asset data in the dynamic eigenvalue set of the specified fund characteristic, and determining the trend eigenvalue set of the specified fund characteristic based on the trend eigenvalue of each of the average asset data; determining the quantization eigenvalue set of the specified fund characteristic based on the quantization eigenvalue of each of the average asset data.

[0012] Optionally, for any one of the feature integration models corresponding to the integration eigenvalue, the training steps adopted include: obtaining the historical transaction data and historical transaction eigenvalues of multiple trading objects; based on the historical transaction data, determining the historical dynamic eigenvalue set, historical trend eigenvalue set, historical quantization eigenvalue set and historical static eigenvalue set of the specified fund characteristic associated with each trading object; using the historical dynamic eigenvalue set, historical trend eigenvalue set, historical quantization eigenvalue set and historical static eigenvalue set of each trading object as the input parameters of the initial feature integration model, and using the historical transaction eigenvalue as the output parameter of the initial feature integration model to train the initial feature integration model to obtain the feature integration model.

[0013] Optionally, for the product recommendation model, the training steps adopted include: obtaining the historical integration feature sets of multiple trading objects and the corresponding historical recommendation parameters, where at least the historical recommendation parameters include: the marketing probability of the recommended financial product; using each of the historical integration feature sets as the model input and the historical recommendation parameters as the model output to train the pre-constructed initial recommendation model to obtain the product recommendation model.

[0014] According to another aspect of an embodiment of the present invention, there is provided a recommendation device for financial products, including: a first acquisition unit, configured to acquire feature information associated with specified fund features of a target object within a historical time period, where the feature information at least includes: a dynamic feature value set, a trend feature value set, a quantization feature value set, and a static feature value set, the trend feature value set and the quantization feature value set are calculated based on the dynamic feature value set, the dynamic feature value set includes: multiple asset fluctuation parameters of financial products for transactions, and the static feature value set includes: multiple asset transaction parameters of the financial products for transactions within the historical time period; a first processing unit, configured to input the dynamic feature value set, the trend feature value set, the quantization feature value set, and the static feature value set into a feature integration model corresponding to the specified fund feature to obtain an integrated feature value output by the feature integration model, where each specified fund feature corresponds to the feature integration model; a second processing unit, configured to input the integrated feature values output by all the feature integration models into a product recommendation model to obtain recommendation information output by the product recommendation model, where the recommendation information at least includes: an object to be recommended and a financial product to be recommended.

[0015] Optionally, the first acquisition unit includes: a slicing subunit, configured to slice the historical time period to obtain multiple historical time domain intervals; a first acquisition subunit, configured to acquire transaction information of fund transactions generated by the target object within each historical time domain interval and calculate average asset data corresponding to each historical time domain interval; a first determination subunit, configured to determine benchmark asset data corresponding to each historical time domain interval based on the transaction information and a preselected anchor time; a second determination subunit, configured to calculate the average asset data within each historical time domain interval and the corresponding benchmark asset data by using a pre-designed calculation strategy to determine the feature information of the specified fund feature.

[0016] Optionally, the historical time domain intervals at least include: multiple first time domain intervals and multiple second time domain intervals, where the interval duration of the first time domain interval is less than the interval duration of the second time domain interval.

[0017] Optionally, the first acquisition subunit includes: a first calculation module, configured to calculate first average asset data of the target object within each first time domain interval based on the transaction information of fund transactions generated by the target object within the first time domain interval; a second calculation module, configured to calculate second average asset data of the target object within each second time domain interval based on the transaction information of fund transactions generated by the target object within the second time domain interval.

[0018] Optionally, the second determination subunit includes: a third processing module, configured to form a set of dynamic eigenvalue of the specified fund feature by using the average asset data within each of the plurality of historical time domain intervals; a comparison module, configured to select any one of the average asset data in the set of dynamic eigenvalue as a target average asset data, and compare it with the corresponding benchmark asset data; a third determination module, configured to determine that a trend value of the target average asset data is a first trend value when the target average asset data is greater than the corresponding benchmark asset data; a fourth determination module, configured to determine that a trend value of the target average asset data is a second trend value when the target average asset data is less than or equal to the corresponding benchmark asset data; a fifth determination module, configured to determine a quantization value of the target average asset data by using a difference between the target average asset data and the benchmark asset data; a sixth determination module, configured to determine a trend value and a quantization value of each of the average asset data in the set of dynamic eigenvalue of the specified fund feature; a seventh determination module, configured to determine a set of trend feature values of the specified fund feature based on the trend value of each of the average asset data; an eighth determination module, configured to determine a set of quantization feature values of the specified fund feature based on the quantization value of each of the average asset data.

[0019] Optionally, the financial product recommendation device further includes: for any one of the feature integration models corresponding to the integration eigenvalue, a training method includes: a second obtaining unit, configured to obtain historical transaction data and historical transaction eigenvalue of a plurality of transaction objects; a first determination unit, configured to determine, based on the historical transaction data, a set of historical dynamic eigenvalue, a set of historical trend eigenvalue, a set of historical quantization eigenvalue, and a set of historical static eigenvalue of each of the transaction objects associated with the specified fund feature; a second determination unit, configured to use the set of historical dynamic eigenvalue, the set of historical trend eigenvalue, the set of historical quantization eigenvalue, and the set of historical static eigenvalue of each of the transaction objects as input parameters of an initial feature integration model, and use the historical transaction eigenvalue as output parameters of the initial feature integration model, and train the initial feature integration model to obtain the feature integration model.

[0020] Optionally, a training step for the product recommendation model includes: a third obtaining unit, configured to obtain a set of historical integration features of a plurality of transaction objects and corresponding historical recommendation parameters, where the historical recommendation parameters at least include: a marketing probability of the recommended financial product; a third processing unit, configured to use each of the set of historical integration features as a model input, and use the historical recommendation parameters as model outputs, and train a pre-constructed initial recommendation model to obtain the product recommendation model.

[0021] In the present invention, first, feature information associated with a specified fund feature of a target object within a historical time period is obtained. The feature information at least includes: a set of dynamic feature values, a set of trend feature values, a set of quantitative feature values, and a set of static feature values. Then, the set of dynamic feature values, the set of trend feature values, the set of quantitative feature values, and the set of static feature values are input into a feature integration model corresponding to the specified fund feature to obtain an integrated feature value output by the feature integration model. Each specified fund feature corresponds to a feature integration model respectively. Finally, the integrated feature values output by all the feature integration models are input into a product recommendation model to obtain recommendation information output by the product recommendation model. The recommendation information at least includes: an object to be recommended and a financial product to be recommended. In the present invention, the integrated feature value can be jointly constructed by the feature information of different specified fund features extracted (at least including: the set of trend feature values and the set of quantitative feature values of each part of the feature information, combined with the original feature information (the set of dynamic feature values and the set of static feature values)), and then the integrated feature value of the specified fund feature is input into the product recommendation model to obtain accurate financial product recommendation information, thereby solving the technical problem in the related art that the loss rate of data information used in the recommendation strategy for financial products is high, resulting in an inaccurate recommendation direction. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings described herein are used to provide a further understanding of the present invention, form a part of this application, and the illustrative embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0023] Figure 1 is a flowchart of an optional method for recommending financial products according to an embodiment of the present invention;

[0024] Figure 2 is a flowchart of constructing an integrated feature and a marketing model according to an embodiment of the present invention;

[0025] Figure 3 is a schematic diagram of the specific structure of a hybrid marketing model according to an embodiment of the present invention;

[0026] Figure 4 is a flowchart of an optional method for analyzing a quantitative trend signal according to an embodiment of the present invention;

[0027] Figure 5 is a schematic diagram of an optional device for recommending financial products according to an embodiment of the present invention;

[0028] Figure 6 is a hardware structure block diagram of an electronic device (or mobile device) for a method of recommending financial products according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] To enable those skilled in the art to better understand the solution 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 accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0031] For the convenience of description, some nouns or terms related to the present application are explained below:

[0032] XGBoost: An open-source software library that executes an optimized distributed gradient boosting machine learning algorithm under the gradient boosting framework. It is an extensible distributed gradient boosting decision tree (GBDT) machine learning library that provides parallel tree boosting capabilities and is an advanced machine learning library for solving regression, classification, and ranking problems.

[0033] It should be noted that the financial product recommendation method and its device in the present disclosure can be used in the field of artificial intelligence when selecting the sales recommendation direction of financial products, and can also be used in any field other than the field of artificial intelligence when selecting the sales recommendation direction of financial products. The application field of the financial product recommendation method and its device in the present disclosure is not limited.

[0034] It should be noted that the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in the present disclosure are all information and data authorized by the user or fully authorized by all parties. For example, an interface is set between the present system and relevant users or institutions. Before obtaining relevant information, a request for acquisition needs to be sent to the aforementioned users or institutions through the interface, and after receiving the consent information feedback from the aforementioned users or institutions, the relevant information is obtained.

[0035] The present invention can be applied to the marketing of various software products, control systems, financial products, and client terminals (including but not limited to: mobile client terminals, PCs, etc.) of various financial institutions. Taking financial products as an example for illustration, through the analysis of the associated fund data of financial products, the recommendation of the business content of financial institutions can be realized (including but not limited to: business functions such as transfer, wealth management, funds, payment, account checking, advertising, etc.).

[0036] It should be noted that based on the transaction data of financial products, the present invention extracts the data dynamic trend change signals, completes the quantitative analysis of the potential information data of features, improves the quality of data usage, optimizes the feature quality to the greatest extent through the way of integrated learning, compresses the multi-dimensional space information for feature integration, completes the feature integration, reduces the model volume, and improves the model performance.

[0037] The present invention can improve the model prediction accuracy, fully excavate the information of static feature data, effectively improve the quality of using effective feature data, and fully compress the multi-data information to improve the quality of data usage.

[0038] The following describes the present invention in conjunction with various embodiments. Embodiment

[0039] According to an embodiment of the present invention, a method embodiment of an optional recommendation method for financial products is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0040] Figure 1 An optional recommendation method for financial products according to an embodiment of the present invention is as follows Figure 1 shown, and the method includes the following steps

[0041] Step S101, obtaining the feature information of the target object associated with the specified fund feature in the historical time period, where the feature information at least includes: a set of dynamic feature values, a set of trend feature values, a set of quantization feature values, and a set of static feature values. The set of trend feature values and the set of quantization feature values are calculated based on the set of dynamic feature values. The set of dynamic feature values includes: multiple asset fluctuation parameters of the financial products for transactions. The set of static feature values includes: multiple asset flow parameters of the financial products for transactions in the historical time period;

[0042] Step S102, inputting the set of dynamic feature values, the set of trend feature values, the set of quantization feature values, and the set of static feature values into the feature integration model corresponding to the specified fund feature, and obtaining the integrated feature value output by the feature integration model, where each specified fund feature corresponds to a feature integration model;

[0043] Step S103: Input the integrated feature values output by all feature integration models into the product recommendation model to obtain the recommendation information output by the product recommendation model, where the recommendation information at least includes: the object to be recommended and the financial product to be recommended.

[0044] Through the above steps, first obtain the feature information related to the specified fund feature of the target object in the historical time period, where the feature information at least includes: the dynamic feature value set, the trend feature value set, the quantitative feature value set, and the static feature value set. Then input the dynamic feature value set, the trend feature value set, the quantitative feature value set, and the static feature value set into the feature integration model corresponding to the specified fund feature to obtain the integrated feature values output by the feature integration model. Finally, input the integrated feature values output by all feature integration models into the product recommendation model to obtain the recommendation information output by the product recommendation model, where the recommendation information at least includes: the object to be recommended and the financial product to be recommended. In this embodiment, the integrated feature values can be jointly constructed by extracting the feature information of different specified fund features (at least including: the trend feature value set and the quantitative feature value set of each part of the feature information), combined with the original feature information (corresponding to the dynamic feature value set and the static feature value set), and then input the integrated feature values of the specified fund feature into the product recommendation model to obtain accurate financial product recommendation information, thereby solving the technical problem in the related art that the loss rate of data information used in the recommendation strategy for financial products is high, resulting in inaccurate recommendation directions.

[0045] The embodiments of the present invention will be further described below in combination with the above implementation steps.

[0046] The construction methods of the feature integration model and the product recommendation model used in the embodiments of the present invention will be described below.

[0047] It should be noted that the feature integration model in the embodiments of the present invention can be used to integrate the fund features related to financial products and output integrated feature values; the product recommendation model can be used to input the integrated feature values of multiple fund features and output the recommendation information of financial products, such as the object to be recommended and the financial product to be recommended. The model selection of the feature integration model and the product recommendation model is not limited here, and the XGBoost model can be selected.

[0048] As an alternative implementation for constructing the feature integration model in this embodiment, for any feature integration model corresponding to the integrated feature value, the training steps adopted include: obtaining the historical transaction data and historical transaction feature values of multiple trading objects; based on the historical transaction data, determining the historical dynamic feature value set, historical trend feature value set, historical quantitative feature value set, and historical static feature value set of each trading object associated with the specified fund feature; using the historical dynamic feature value set, historical trend feature value set, historical quantitative feature value set, and historical static feature value set of each trading object as the input parameters of the initial feature integration model, and using the historical transaction feature value as the output parameter of the initial feature integration model to train the initial feature integration model to obtain the feature integration model.

[0049] The above trading objects may refer to trading objects that conduct transactions on financial products. For example, financial institution customers, the public, enterprises, other financial institutions, etc.

[0050] It should be noted that the transaction data in this embodiment may include, but is not limited to, fund flow (including but not limited to: transaction time point, transaction amount, time period length), asset status data.

[0051] In this embodiment, the historical transaction feature value may refer to the integrated feature value of the above historical transaction data. The above historical dynamic feature value set may include data dynamically extracted from the above historical transaction data in the time domain. The above historical trend feature value set and the above historical quantitative feature value set may be obtained from the above historical dynamic feature value set. The above historical static feature value set may refer to transaction data that has not been dynamically extracted. For example, the cumulative value of fund flow for a period of time, or the total amount of funds of a trading object for a period of time. The above specified fund feature may refer to fund data features related to funds. For example, fund flow, asset data of trading objects, etc.

[0052] In the process of constructing the integrated feature model, the historical transaction data and historical transaction feature values can be obtained first. According to the historical transaction data, the historical dynamic feature value set, historical trend feature value set, historical quantitative feature value set, and historical static feature value set of each trading object associated with the specified fund feature can be determined. Using the historical dynamic feature value set, historical trend feature value set, historical quantitative feature value set, and historical static feature value set of each trading object as the input parameters of the initial feature integration model, and using the historical transaction feature value as the output parameter of the initial feature integration model to train the initial feature integration model to obtain the feature integration model.

[0053] After the feature integration model is described, the construction method of the product recommendation model will be described below.

[0054] As an alternative implementation for constructing the product recommendation model in this embodiment, optionally, for the product recommendation model, the training steps adopted include: obtaining the historical integrated feature sets of multiple trading objects and the corresponding historical recommendation parameters, where the historical recommendation parameters at least include: the marketing probability of the recommended financial products; using each historical integrated feature set as the model input and the historical recommendation parameters as the model output to train a pre-constructed initial recommendation model to obtain the product recommendation model.

[0055] It should be noted that the historical recommendation parameters in this embodiment at least include: the marketing probabilities of each financial product recommended in the historical process, and also include the historical recommendation objects and the historical recommended financial products. One can first obtain the historical integrated feature sets of multiple trading objects and the corresponding historical recommendation parameters, use each historical integrated feature set as the model input and the historical recommendation parameters as the model output to train a pre-constructed initial recommendation model to obtain the product recommendation model.

[0056] After constructing and training the feature integration model and the product recommendation model, the following will illustrate how to use these two models for feature integration and product recommendation in real time.

[0057] Step S101: Obtain the feature information of the target object associated with the specified capital feature in the historical time period, where the feature information at least includes: a set of dynamic feature values, a set of trend feature values, a set of quantization feature values, and a set of static feature values. The set of trend feature values and the set of quantization feature values are calculated based on the set of dynamic feature values. The set of dynamic feature values includes: multiple asset fluctuation parameters of the financial products for transactions. The set of static feature values includes: multiple asset flow parameters of the financial products for transactions in the historical time period.

[0058] Optionally, the set of dynamic feature values in this embodiment may include but is not limited to: data dynamically extracted from the above transaction data in the time domain. The dynamic feature values in the set of dynamic feature values may be the moving average capital values over a period of time.

[0059] In this embodiment, the set of trend feature values and the set of quantization feature values can be obtained from the historical set of dynamic feature values. At the same time, the set of static feature values in this embodiment may refer to the transaction data that has not been dynamically extracted, such as: the cumulative value of the capital flow over a period of time, or the total capital of the trading object over a period of time.

[0060] The following will illustrate the calculation methods of the set of dynamic feature values, the set of trend feature values, and the set of quantization feature values.

[0061] Optionally, the step of obtaining the feature information of the target object associated with the specified fund feature in the historical time period includes: slicing the historical time period to obtain a plurality of historical time domain intervals; obtaining the transaction information of the target object generating fund transactions in each historical time domain interval, and calculating the average asset data corresponding to each historical time domain interval; determining the benchmark asset data corresponding to each historical time domain interval based on the transaction information and the pre-selected anchor time; using a pre-designed calculation strategy to calculate the average asset data and the corresponding benchmark asset data in each historical time domain interval to determine the feature information of the specified fund feature.

[0062] The above historical time domain intervals can be generated after slicing. For example, if the historical time period is 1 month and sliced every two dates, for a month with 30 days, it can be sliced into 15 historical time domain intervals. The transaction information of generating fund transactions in each historical time domain interval can be used to determine the eigenvalues in the set of dynamic eigenvalue, the set of trend eigenvalue, and the set of quantization eigenvalue in the feature information of the specified fund feature. Optionally, the interval length of the time domain interval indicated in this embodiment can be adjusted according to different specified fund features, and the interval length of the time domain interval is not limited here.

[0063] The above anchor time can be used to determine the benchmark asset data of the average asset data corresponding to each historical time domain interval, and a pre-designed calculation strategy is used to calculate the average asset data and the corresponding benchmark asset data in each historical time domain interval to determine the set of dynamic eigenvalue, the set of trend eigenvalue, and the set of quantization eigenvalue in the feature information of the specified fund feature.

[0064] An optional implementation manner is that the historical time domain intervals at least include: a plurality of first time domain intervals and a plurality of second time domain intervals, where the interval duration of the first time domain interval is less than the interval duration of the second time domain interval, that is, the first time domain interval is a short time domain interval and the second time domain interval is a long time domain interval.

[0065] Taking the fund flow as an example of the specified fund feature, the above first time domain interval and the second time domain interval are illustrated. The short time domain interval (corresponding to the first time domain interval) can select the moving average fund value of the account in the recent 3, 7, and 15 days (the moving average fund value can correspond to the eigenvalue in the set of dynamic eigenvalue), and the long time domain interval (corresponding to the above second time domain interval) can select the moving average fund value of the account in the recent 30, 60, and 120 days.

[0066] Optionally, the step of calculating the average asset data corresponding to each historical time domain interval includes: calculating the first average asset data of the target object in each first time domain interval based on the transaction information of the target object generating fund transactions in the first time domain interval; calculating the second average asset data of the target object in each second time domain interval based on the transaction information of the target object generating fund transactions in the second time domain interval.

[0067] Optionally, the step of calculating the characteristic information of the specified fund characteristic by using a pre-designed calculation strategy for the average asset data and the corresponding benchmark asset data in each historical time domain interval includes: the average asset data in each historical time domain interval among multiple historical time domain intervals forms a dynamic eigenvalue set of the specified fund characteristic; selecting each average asset data in the dynamic eigenvalue set as the target average asset data and comparing it with the corresponding benchmark asset data; when the target average asset data is greater than the corresponding benchmark asset data, determining that the trend value of the target average asset data is the first trend value; when the target average asset data is less than or equal to the corresponding benchmark asset data, determining that the trend value of the target average asset data is the second trend value; determining the quantization value of the target average asset data as the difference between the target average asset data and the benchmark asset data; determining a trend characteristic value set of the specified fund characteristic based on the trend values of multiple target average asset data; determining a quantization characteristic value set of the specified fund characteristic based on the quantization values of multiple target average asset data.

[0068] The following is a schematic illustration of the calculation method for calculating the characteristic information of the specified fund characteristic by using a pre-designed calculation strategy for the average asset data and the corresponding benchmark asset data in each historical time domain interval.

[0069] If the moving average values of the 3-day moving average (3M), 7-day moving average (7M), 15MA, 30MA,... nMA represent the average asset data in the above historical time domain intervals, take the 3-day moving average (3M), 7-day moving average (7M), 15MA, 30MA,... nMA as the dynamic eigenvalue set, and the benchmark moving average at the anchor time selected by each moving average as the benchmark asset data.

[0070] Trend characteristic value = 1 if yMA - xMA > 0 else 0 (indicating that if the value of the moving average is greater than the value of the corresponding benchmark moving average, the trend characteristic value is 1, otherwise it is 0). It should be noted that the setting of the characteristic value is not limited, and 1 and 0 are used as examples here.

[0071] Among them, the quantization characteristic value of this embodiment = yMA - xMA (indicating the difference between the value of the moving average and the value of the corresponding benchmark moving average).

[0072] In this embodiment, a corresponding trend feature value and a quantization feature value can be calculated for each moving average. The trend feature values of multiple moving averages can form a trend feature value set, and the quantization feature values of multiple moving averages can form a quantization feature value set.

[0073] Step S102: Input the dynamic feature value set, the trend feature value set, the quantization feature value set, and the static feature value set into a feature integration model corresponding to a specified fund feature to obtain an integrated feature value output by the feature integration model, where each specified fund feature corresponds to a feature integration model.

[0074] In this embodiment, a feature integration model can be used to integrate the trend signals and original signals of different types of features. In this way, the trend information of each part of the feature information can be effectively extracted and combined with the original feature information to jointly construct integrated information. Compared with the original features, the features have higher-dimensional feature attributes, improving the quality of data use.

[0075] Step S103: Input the integrated feature values output by all the feature integration models into a product recommendation model to obtain recommendation information output by the product recommendation model, where the recommendation information at least includes: the object to be recommended and the financial product to be recommended.

[0076] In this embodiment, each of the multiple specified fund features can be separately input into the corresponding feature integration model, and each feature integration model inputs an integrated feature value. Inputting all the integrated feature values into the product recommendation model can obtain the recommendation information output by the product recommendation model. Based on the recommendation information, the recommendation probability of the financial product and the recommended object can be determined, improving the recommendation accuracy of the financial product and the customer satisfaction.

[0077] Through the above embodiments of the present invention, the accuracy of the recommendation information output by the product recommendation model can be improved through feature integration, the static feature data information can be fully mined, the quality of data use of effective features can be improved, and multiple data information can be fully compressed to complete data integration, improve the quality of data use, optimize the feature quality to the greatest extent, compress the multi-dimensional space information for feature integration, complete feature integration, reduce the model size, and improve the model performance.

[0078] Next, another alternative embodiment is used to illustrate the present invention. Embodiment

[0079] This embodiment provides a product recommendation method based on a mixed model of multiple integrated eigenvalues. By introducing different dynamic indicators, a feature trend quantization signal is constructed, a static data dynamic change trend feature is constructed, the quality of feature use is expanded, and the integrated learning is used to complete the construction of the integrated feature of different feature groups, so as to improve the effective information volume of the original feature data. Among them, the trend quantization signal of this embodiment includes a trend signal and a quantization signal. The trend signal corresponds to the trend eigenvalue in Embodiment 1, and the quantization signal corresponds to the quantization eigenvalue in Embodiment 1.

[0080] Figure 2 is a flowchart of constructing an integrated feature and a marketing model according to an embodiment of the present invention, as Figure 2 shown, and the specific operation steps are as follows:

[0081] S201: Construct a trend quantization signal of the basic feature. Among them, in the process of selecting the basic feature, features with relatively frequent changes (corresponding to the dynamic features in Embodiment 1) can be selected as the objects to be trend-quantized. And in the process of trend quantization, a long time-domain interval and a short time-domain interval are respectively selected for construction at the same time to ensure the best signal quantization coverage effect.

[0082] In this embodiment, for the trend quantization feature, information such as transaction flow and asset situation can be selected for trend quantization. It should be noted that the interval length of the long time-domain interval and the interval length of the short time-domain interval can be adjusted according to different selected basic features, and no limitation is made here;

[0083] S202: Complete the feature fusion of the static feature and the trend quantization signal through XGBoost integrated learning to construct an integrated feature. The integrated feature is constructed for different basic features to be trend-quantized respectively. The static feature (corresponding to the static feature in Embodiment 1) and the trend quantization signal are jointly used as the input features of the XGBoost trend signal integration model. The labels of each integrated feature model can use the labels of the original business marketing model to ensure that the model refers to the business scenario to learn data information;

[0084] S203: Construct an XGBoost integrated learning model, and use the integrated feature as the input feature to complete the construction of the marketing model. The output of each integrated feature model is used as the input feature of the XGBoost integrated learning model to construct the marketing model. Since the integrated learning is extremely prone to overfitting, it is necessary to strictly control the model depth (max_depth) and the learning rate (eta parameter) to ensure the robustness of the model.

[0085] Figure 3 is a schematic diagram of the specific structure of a hybrid marketing model according to an embodiment of the present invention, as Figure 3As shown, by screening the basic features available for trend quantification (corresponding to the dynamic eigenvalue set, trend eigenvalue set, and quantization eigenvalue set in the first embodiment), and then using the basic features ( Figure 3 illustrated by basic feature_A1, basic feature_A2, basic feature_A3... basic feature_An in Figure 3 ) and static features (corresponding to the static features in the static eigenvalue set in the first embodiment, Figure 3 illustrated by static feature_A1, static feature_A2, static feature_A3... static feature_An in Figure 3 ) as inputs, and inputting them into the XGBoost trend signal integration model ( Figure 3 illustrated by w1, w2, w3... wn in Figure 3 respectively), the XGBoost trend signal integration model outputs integrated features ( Figure 3 illustrated by integrated feature_B1, integrated feature_B2, integrated feature_B3... integrated feature_Bn in Figure 3 respectively). For the XGBoost integrated learning, the integrated features corresponding to this feature are constructed to condense the features to the greatest extent and complete the maximum compression of feature information. The compressed integrated features are used as the input features of the marketing model ( Figure 3 illustrated by the XGBoost hybrid marketing recommendation model in Figure 3 ). The predicted marketing probability is output through the XGBoost hybrid marketing recommendation model, and the marketing object screening is completed according to the predicted marketing probability.

[0086] Figure 4 FIG. is a flowchart of an optional method for analyzing and quantifying trend signals according to an embodiment of the present invention. As Figure 4 shown, it includes a set of features of the trend signal to be quantified (corresponding to the dynamic eigenvalue set in the first embodiment), Figure 4 The set of features of the trend signal to be quantified includes: 3-day moving average (3M), 7-day moving average (7M), 15MA, 30MA,... nMA). The set of features of the trend signal to be quantified on the left is input Figure 4 to the right side in Figure 4 . The anchor time is selected as the basic signal for the trend analysis time point. Basic signal: = n-day moving average (n corresponds to any one of 3, 7, 15, 30... nMA on the left). Trend signal processing: = 1 if yMA - xMA > 0 else 0 (indicating that if the average value of the moving average in the set of features of the trend signal to be quantified is greater than the average value of the corresponding basic signal moving average, the trend signal is 1, otherwise it is 0). Quantization signal processing: = yMA - xMA (indicating the difference between the average value of the moving average in the set of features of the trend signal to be quantified and the average value of the corresponding basic signal moving average). The processed trend signal, feature signal, the set of features of the trend signal to be quantified, and the static features are input into the XGBoost trend signal integration model (constructing integrated features: static + trend).

[0087] The following will be combined with Figure 4 to illustrate the specific steps of trend quantification in this embodiment as follows:

[0088] 1. It is necessary to screen the features available for trend signal quantification. Generally, features with high-frequency changes, low-frequency or fixed-frequency changes are selected as the features to be quantified, such as relevant information on individual account fund flows and personal assets.

[0089] 2. In the quantification process, three-dimensional information corresponding to the feature needs to be generated. Taking the account fund flow as an example, in the short time domain interval, the moving average fund values of the account in the recent 3, 7, and 15 days can be selected. In the long time domain interval, the moving average fund values of the account in the recent 30, 60, and 120 days can be selected. Several anchor time points (the current day, 7 days from now, 30 days from now, etc.) are selected as the reference times for trend analysis and quantification, and the reference average fund values at these reference times are used to calculate the moving average values in the corresponding long and short time domains respectively. The trend signal and quantification signal can be calculated according to Figure 4 the formulas shown.

[0090] 3. The original basic static features and the processed trend quantification signals are used as the input features of the XGBoost trend signal integration model of the XGBoost integration model to construct an integrated feature model. The output of the model is the integrated feature corresponding to the feature. The output integrated feature is used as the input feature of the hybrid marketing recommendation model of the XGBoost integration model, and the predicted marketing probability is output.

[0091] This embodiment can improve the model prediction accuracy, fully explore the static feature data information, effectively use the quality of feature data, and fully compress multiple data information to complete data integration and improve the data usage quality. The key of this embodiment lies in: based on the original sample data, by extracting the data dynamic trend change signals, completing the quantitative analysis of the potential information data of the features, and improving the data usage quality; through the way of integrated learning, optimizing the feature quality to the greatest extent, compressing the multi-dimensional space information for feature integration, completing feature integration, reducing the model size, and improving the model performance.

[0092] The following will illustrate the present invention in combination with another optional embodiment. Embodiment

[0093] This embodiment provides an optional financial product recommendation device. Each implementation unit included in the recommendation device corresponds to each implementation step in the first embodiment above.

[0094] Figure 5 is a schematic diagram of an optional financial product recommendation device according to an embodiment of the present invention. As Figure 5 shown, the recommendation device includes: a first acquisition unit 51, a first processing unit 52, and a second processing unit 53, where

[0095] A first acquisition unit 51 is configured to acquire feature information associated with a specified fund feature of a target object within a historical time period. The feature information at least includes: a dynamic feature value set, a trend feature value set, a quantization feature value set, and a static feature value set. The trend feature value set and the quantization feature value set are calculated based on the dynamic feature value set. The dynamic feature value set includes multiple asset volatility parameters of financial products for transactions. The static feature value set includes multiple asset transaction parameter of financial products for transactions within the historical time period.

[0096] A first processing unit 52 is configured to input the dynamic feature value set, the trend feature value set, the quantization feature value set, and the static feature value set into a feature integration model corresponding to the specified fund feature, and obtain an integrated feature value output by the feature integration model. Each specified fund feature corresponds to a feature integration model respectively.

[0097] A second processing unit 53 is configured to input the integrated feature values output by all the feature integration models into a product recommendation model to obtain recommendation information output by the product recommendation model. The recommendation information at least includes: an object to be recommended and a financial product to be recommended.

[0098] A recommendation device for financial products can acquire, through the first acquisition unit 51, feature information associated with a specified fund feature of a target object within a historical time period. The feature information at least includes: a dynamic feature value set, a trend feature value set, a quantization feature value set, and a static feature value set. The dynamic feature value set, the trend feature value set, the quantization feature value set, and the static feature value set are input into a feature integration model corresponding to the specified fund feature through the first processing unit 52 to obtain an integrated feature value output by the feature integration model. The integrated feature values output by all the feature integration models are input into a product recommendation model through the second processing unit 53 to obtain recommendation information output by the product recommendation model. The recommendation information at least includes: an object to be recommended and a financial product to be recommended. In this embodiment, the integrated feature value can be jointly constructed by extracting the feature information of different specified fund features (at least including: the trend feature value set and the quantization feature value set of each part of the feature information), and combining the original feature information (the dynamic feature value set and the static feature value set), and then the integrated feature value of the specified fund feature is input into the product recommendation model to obtain accurate financial product recommendation information, thereby solving the technical problem in the related art that the loss rate of data information used in the recommendation strategy for financial products is high, resulting in an inaccurate recommendation direction.

[0099] Optionally, the first acquisition unit includes: a slicing subunit, configured to slice a historical time period to obtain a plurality of historical time domain intervals; a first acquisition subunit, configured to acquire transaction information of the target object generating fund transactions within each historical time domain interval, and calculate average asset data corresponding to each historical time domain interval; a first determination subunit, configured to determine benchmark asset data corresponding to each historical time domain interval based on the transaction information and a pre-selected anchor time; a second determination subunit, configured to calculate the average asset data within each historical time domain interval and the corresponding benchmark asset data by using a pre-designed calculation strategy to determine characteristic information of a specified fund characteristic.

[0100] Optionally, the historical time domain intervals at least include: a plurality of first time domain intervals and a plurality of second time domain intervals, where the interval duration of the first time domain intervals is less than the interval duration of the second time domain intervals.

[0101] Optionally, the first acquisition subunit includes: a first calculation module, configured to calculate first average asset data of the target object within each first time domain interval based on the transaction information of the target object generating fund transactions within the first time domain interval; a second calculation module, configured to calculate second average asset data of the target object within each second time domain interval based on the transaction information of the target object generating fund transactions within the second time domain interval.

[0102] Optionally, the second determination subunit includes: a third processing module, configured to form a dynamic eigenvalue set of a specified fund characteristic from the average asset data within each historical time domain interval among the plurality of historical time domain intervals; a comparison module, configured to select any one of the average asset data in the dynamic eigenvalue set as target average asset data and compare it with the corresponding benchmark asset data; a third determination module, configured to determine that the trend value of the target average asset data is a first trend value when the target average asset data is greater than the corresponding benchmark asset data; a fourth determination module, configured to determine that the trend value of the target average asset data is a second trend value when the target average asset data is less than or equal to the corresponding benchmark asset data; a fifth determination module, configured to determine the difference between the target average asset data and the benchmark asset data as the quantization value of the target average asset data; a sixth determination module, configured to determine the trend value and quantization value of each average asset data in the dynamic eigenvalue set of the specified fund characteristic; a seventh determination module, configured to determine a trend eigenvalue set of the specified fund characteristic based on the trend value of each average asset data; an eighth determination module, configured to determine a quantization eigenvalue set of the specified fund characteristic based on the quantization value of each average asset data.

[0103] Optionally, for any feature integration model corresponding to the integrated eigenvalue, the training steps adopted include: a second acquisition unit, configured to acquire historical transaction data and historical transaction eigenvalues of multiple transaction objects; a first determination unit, configured to determine, based on the historical transaction data, a set of historical dynamic eigenvalues, a set of historical trend eigenvalues, a set of historical quantization eigenvalues, and a set of historical static eigenvalues of each transaction object associated with the specified fund feature; a second determination unit, configured to use the set of historical dynamic eigenvalues, the set of historical trend eigenvalues, the set of historical quantization eigenvalues, and the set of historical static eigenvalues of each transaction object as input parameters of the initial feature integration model, and use the historical transaction eigenvalue as the output parameter of the initial feature integration model to train the initial feature integration model to obtain the feature integration model.

[0104] Optionally, for the product recommendation model, the training steps adopted include: a third acquisition unit, configured to acquire a set of historical integration features of multiple transaction objects and corresponding historical recommendation parameters, where the historical recommendation parameters at least include: the marketing probability of the recommended financial product; a third processing unit, configured to use each set of historical integration features as the model input and the historical recommendation parameters as the model output to train a pre-constructed initial recommendation model to obtain the product recommendation model.

[0105] The above-mentioned financial product recommendation device may further include a processor and a memory. The above-mentioned first acquisition unit 51, first processing unit 52, second processing unit 53, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to implement corresponding functions.

[0106] The above-mentioned processor includes a kernel, and the kernel retrieves the corresponding program unit from the memory. One or more kernels can be set. By adjusting the kernel parameters, through multiple eigenvalue sets, input into the feature integration model to obtain the integrated eigenvalue, and input the integrated eigenvalue of the specified fund feature into the product recommendation model to obtain the recommendation information, and implement the recommendation of financial products according to the recommendation information.

[0107] The above-mentioned memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of, for example, read-only memory (ROM) or flash memory (flash RAM), and the memory includes at least one storage chip.

[0108] According to another aspect of the embodiments of the present invention, an electronic device is further provided, including: a processor; and a memory, configured to store executable instructions of the processor; wherein, the processor is configured to execute the above-mentioned financial product recommendation method of any one item by executing the executable instructions.

[0109] According to another aspect of the embodiments of the present invention, there is also provided a computer-readable storage medium, which includes a stored computer program. When the computer program runs, it controls the device where the computer-readable storage medium is located to execute the recommendation method of the financial product in any one of the above.

[0110] Figure 6 is a hardware structure block diagram of an electronic device (or mobile device) for the recommendation method of a financial product according to an embodiment of the present invention. As Figure 6 shown, the electronic device may include one or more processors (processors may include, but are not limited to, processing devices such as microprocessor MCUs or programmable logic device FPGAs, shown as 602a, 602b,..., 602n in the figure) and a memory 604 for storing data. In addition, it may further include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a keyboard, a power supply, and / or a camera. Those of ordinary skill in the art can understand that Figure 6 the structure shown is only schematic and does not limit the structure of the above-mentioned electronic device. For example, the electronic device may further include more or fewer components than Figure 6 shown, or have a different configuration from Figure 6 shown.

[0111] The serial numbers of the above embodiments of the present invention are only for description and do not represent the superiority or inferiority of the embodiments.

[0112] In the above embodiments of the present invention, the descriptions of the respective embodiments have their own emphases. For parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0113] In several embodiments provided in the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only illustrative. For example, the division of the units can be a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of units or modules can be in an electrical or other form.

[0114] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0115] In addition, each functional unit in various embodiments of the present invention may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit.

[0116] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, may be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, read-only memories (ROMs), random access memories (RAMs), mobile hard disks, magnetic disks, or optical discs that can store program codes.

[0117] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can still be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for recommending financial products, characterized in that, it includes: Obtain the feature information related to the specified fund feature of the target object within the historical time period. Among them, the feature information at least includes: a set of dynamic feature values, a set of trend feature values, a set of quantization feature values, and a set of static feature values. The set of trend feature values and the set of quantization feature values are calculated based on the set of dynamic feature values. The set of dynamic feature values includes: multiple asset volatility parameters of the financial products for transactions. The set of static feature values includes: multiple asset transaction parameters of the financial products for transactions within the historical time period; The step of obtaining the feature information related to the specified fund feature of the target object within the historical time period includes: Slice the historical time period to obtain multiple historical time domain intervals; obtain the transaction information of the target object's fund transactions within each historical time domain interval, and calculate the average asset data corresponding to each historical time domain interval; based on the transaction information and the pre-selected anchor time, determine the benchmark asset data corresponding to each historical time domain interval; use a pre-designed calculation strategy to calculate the average asset data within each historical time domain interval and the corresponding benchmark asset data to determine the feature information of the specified fund feature; The step of using a pre-designed calculation strategy to calculate the average asset data within each historical time domain interval and the corresponding benchmark asset data to determine the feature information of the specified fund feature includes: forming the set of dynamic feature values of the specified fund feature from the average asset data within each of the multiple historical time domain intervals; selecting any one of the average asset data in the set of dynamic feature values as the target average asset data and comparing it with the corresponding benchmark asset data; in the case where the target average asset data is greater than the corresponding benchmark asset data, determine the trend feature value of the target average asset data as the first trend value; in the case where the target average asset data is less than or equal to the corresponding benchmark asset data, determine the trend feature value of the target average asset data as the second trend value; determine the quantization feature value of the target average asset data as the difference between the target average asset data and the benchmark asset data; determine the trend feature value and quantization feature value of each average asset data in the set of dynamic feature values of the specified fund feature, and determine the set of trend feature values of the specified fund feature based on the trend feature value of each average asset data; determine the set of quantization feature values of the specified fund feature based on the quantization feature value of each average asset data; Input the set of dynamic feature values, the set of trend feature values, the set of quantization feature values, and the set of static feature values into the feature integration model corresponding to the specified fund feature to obtain the integrated feature value output by the feature integration model, where each specified fund feature corresponds to the feature integration model; Input the integrated feature values output by all the feature integration models into the product recommendation model to obtain the recommendation information output by the product recommendation model, where the recommendation information at least includes: the object to be recommended and the financial product to be recommended.

2. The recommendation method according to claim 1, wherein, the historical time domain interval at least includes: a plurality of first time domain intervals and a plurality of second time domain intervals, wherein the interval duration of the first time domain interval is less than the interval duration of the second time domain interval.

3. The recommendation method according to claim 2, wherein, the step of calculating the average asset data corresponding to each historical time domain interval includes: calculating the first average asset data of the target object in each first time domain interval based on the transaction information of the target object generating fund transactions within the first time domain interval; calculating the second average asset data of the target object in each second time domain interval based on the transaction information of the target object generating fund transactions within the second time domain interval.

4. The recommendation method according to claim 1, wherein, for any feature integration model corresponding to the integrated feature value, the training steps adopted include: obtaining the historical transaction data and historical transaction feature values of a plurality of transaction objects; based on the historical transaction data, determining the historical dynamic feature value set, historical trend feature value set, historical quantization feature value set and historical static feature value set of each transaction object associated with the specified fund feature; using the historical dynamic feature value set, historical trend feature value set, historical quantization feature value set and historical static feature value set of each transaction object as the input parameters of the initial feature integration model, and using the historical transaction feature values as the output parameters of the initial feature integration model to train the initial feature integration model to obtain the feature integration model.

5. The recommendation method according to claim 1, wherein, for the product recommendation model, the training steps adopted include: obtaining the historical integrated feature sets of a plurality of transaction objects and the corresponding historical recommendation parameters, where the historical recommendation parameters at least include: the marketing probability of the recommended financial product; using each historical integrated feature set as the model input and the historical recommendation parameters as the model output to train the pre-constructed initial recommendation model to obtain the product recommendation model.

6. A financial product recommendation device, wherein, it includes: a first acquisition unit, configured to acquire the feature information of the target object associated with the specified fund feature within the historical time period, where the feature information at least includes: a dynamic feature value set, a trend feature value set, a quantization feature value set and a static feature value set, the trend feature value set and the quantization feature value set are calculated based on the dynamic feature value set, the dynamic feature value set includes: a plurality of asset fluctuation parameters of the financial products for transactions, and the static feature value set includes: a plurality of asset flow parameters of the financial products for transactions within the historical time period; The first acquisition unit includes: a slicing subunit, configured to slice the historical time period to obtain a plurality of historical time domain intervals; a first acquisition subunit, configured to acquire transaction information of the target object generating fund transactions within each of the historical time domain intervals, and calculate average asset data corresponding to each of the historical time domain intervals; a first determination subunit, configured to determine benchmark asset data corresponding to each of the historical time domain intervals based on the transaction information and a preselected anchor time; a second determination subunit, configured to calculate the average asset data within each of the historical time domain intervals and the corresponding benchmark asset data by using a preset calculation strategy to determine the feature information of the specified fund feature. The second determination subunit includes: a third processing module, configured to form a dynamic feature value set of the specified fund feature from the average asset data within each of the plurality of historical time domain intervals; a comparison module, configured to select any one of the average asset data in the dynamic feature value set as target average asset data and compare it with the corresponding benchmark asset data; a third determination module, configured to determine that the trend value of the target average asset data is a first trend value when the target average asset data is greater than the corresponding benchmark asset data; a fourth determination module, configured to determine that the trend value of the target average asset data is a second trend value when the target average asset data is less than or equal to the corresponding benchmark asset data; a fifth determination module, configured to determine the difference between the target average asset data and the benchmark asset data as the quantization value of the target average asset data; a sixth determination module, configured to determine the trend value and the quantization value of each of the average asset data in the dynamic feature value set of the specified fund feature; a seventh determination module, configured to determine a trend feature value set of the specified fund feature based on the trend value of each of the average asset data; an eighth determination module, configured to determine a quantization feature value set of the specified fund feature based on the quantization value of each of the average asset data. The first processing unit is configured to input the dynamic feature value set, the trend feature value set, the quantization feature value set, and the static feature value set into a feature integration model corresponding to the specified fund feature to obtain an integrated feature value output by the feature integration model, where each of the specified fund features corresponds to the feature integration model respectively. The second processing unit is configured to input the integrated feature values output by all the feature integration models into a product recommendation model to obtain recommendation information output by the product recommendation model, where the recommendation information at least includes: an object to be recommended and a financial product to be recommended.

7. A computer-readable storage medium, characterized in that the computer-readable storage medium includes a stored computer program, where when the computer program runs, it controls a device where the computer-readable storage medium is located to execute the financial product recommendation method according to any one of claims 1 to 5.

8. An electronic device, characterized in that Comprising one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the method for recommending a financial product according to any one of claims 1 to 5.

Images