Message recommendation method, electronic device, storage medium and program product

Abstract

The present application discloses a message recommendation method, electronic device, storage medium, and program product. The message recommendation method includes the following steps: obtaining message feature data, wherein the message feature data includes at least one of the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step; inputting the message feature data into a conversion qualification cutoff parameter determination model to determine the target conversion qualification cutoff parameter of the customer group at the current time step; obtaining message recommendation sample data corresponding to the customer group at the current time step, filtering the message recommendation sample data of the customer group at the current time step according to the target conversion qualification cutoff parameter to obtain the return conversion sample data of the current time step; and sending the return conversion sample data to the message delivery party. The present application solves the technical problem of poor message recommendation effect in the prior art.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology in financial technology (Fintech), and in particular to a message recommendation method, electronic device, storage medium and program product. Background Art

[0002] With the continuous development of financial technology, especially Internet technology finance, more and more technologies (such as distributed, artificial intelligence, etc.) are applied in the financial field, but the financial industry also puts forward higher requirements for technology, such as higher requirements for the distribution of corresponding to-do items in the financial industry.

[0003] When recommending online news, publishers rely on conversion data sent back by recommenders to build models to accurately identify the target audience for the recommendations, thereby helping recommenders acquire targeted customers. In the media landscape, customer qualifications are long-tailed, with low-qualified customers far outnumbering high-qualified ones. Low-qualified customers can lead to a decline in customer base and a reduction in the input-output ratio. Therefore, recommenders typically set a fixed conversion qualification limit based on experience, filter conversion data, and remove conversion data from low-qualified customers before sending it back. However, many factors influence the conversion returns of news recommendations, and the customer base changes in real time. Fixed conversion qualification limits often don't match the changing customer base. Therefore, filtering the returned samples based on fixed conversion qualification limits is inaccurate, which impacts the effectiveness of news recommendations. Summary of the Invention

[0004] The main purpose of this application is to provide a message recommendation method, electronic device, storage medium and program product, aiming to solve the technical problem of poor message recommendation effect in the prior art.

[0005] To achieve the above objectives, the present application provides a message recommendation method, which is applied to a message recommender and includes the following steps:

[0006] Acquiring message feature data, wherein the message feature data includes at least one of the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step;

[0007] Determining the target conversion qualification cutoff parameter of the customer group at the current time step by inputting the message feature data into a conversion qualification cutoff parameter determination model;

[0008] Obtaining message recommendation sample data corresponding to the customer group at the current time step, filtering the message recommendation sample data for the customer group at the current time step according to the target conversion qualification cutoff parameter, and obtaining the return conversion sample data for the current time step;

[0009] The return conversion sample data is sent to a message delivery party, so that the message delivery party can make message recommendations based on the return conversion sample data.

[0010] The present application also provides a method for determining a conversion qualification truncation parameter, which is applied to a message recommender and includes the following steps:

[0011] Acquiring message feature data, wherein the message feature data includes at least one of the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step;

[0012] By inputting the message feature data of the customer group at the current time step into the conversion qualification prediction model, the conversion qualification distribution data of the customer group at the next time step is predicted, and by inputting the message feature data into the conversion volume prediction model, the conversion volume data of the customer group at the next time step is predicted;

[0013] A target conversion qualification cutoff parameter is determined based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step.

[0014] The present application also provides a message recommendation device, which is applied to a message recommender and includes:

[0015] A first acquisition module is configured to acquire message feature data, wherein the message feature data includes at least one of the following: the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step;

[0016] A first determination module is configured to determine a target conversion qualification cutoff parameter for the customer group at the current time step by inputting the message feature data into a conversion qualification cutoff parameter determination model;

[0017] A filtering module is used to obtain message recommendation sample data corresponding to the customer group at the current time step, and filter the message recommendation sample data of the customer group at the current time step according to the target conversion qualification cutoff parameter to obtain the return conversion sample data of the current time step;

[0018] The sending module is used to send the return conversion sample data to the message delivery party, so that the message delivery party can make message recommendations based on the return conversion sample data.

[0019] The present application also provides a device for determining a conversion qualification truncation parameter, which is applied to a message recommender and includes:

[0020] A second acquisition module is configured to acquire message feature data, wherein the message feature data includes at least one of the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step;

[0021] A prediction module is configured to predict the conversion qualification distribution data of the customer group at the next time step by inputting the message feature data of the customer group at the current time step into a conversion qualification prediction model, and to predict the conversion volume data of the customer group at the next time step by inputting the message feature data into a conversion volume prediction model;

[0022] The second determination module is used to determine the target conversion qualification cutoff parameter according to the conversion volume data of the customer group in the next time step and the conversion qualification distribution data of the customer group in the next time step.

[0023] The present application also provides an electronic device, which is a physical device, and includes: a memory, a processor, and a program of the message recommendation method stored in the memory and runnable on the processor. When the program of the message recommendation method is executed by the processor, the steps of the message recommendation method as described above can be implemented.

[0024] The present application also provides a storage medium, which is a computer-readable storage medium. The computer-readable storage medium stores a program for implementing the message recommendation method. When the program of the message recommendation method is executed by a processor, the steps of the message recommendation method as described above are implemented.

[0025] The present application also provides a computer program product, including a computer program, which implements the steps of the message recommendation method as described above when executed by a processor.

[0026] The present application provides a message recommendation method, electronic device, storage medium and program product. By acquiring message feature data, wherein the message feature data includes at least one of the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step, the acquisition of multi-dimensional message feature data is achieved. Then, by inputting the message feature data into a conversion qualification cutoff parameter determination model, the target conversion qualification cutoff parameter of the customer group at the current time step is determined. The target conversion qualification cutoff parameter of the customer group at the current time step is determined by integrating the multi-dimensional message feature data through the conversion qualification cutoff parameter determination model. Then, by acquiring message recommendation sample data corresponding to the customer group at the current time step, the message recommendation sample data of the customer group at the current time step is filtered according to the target conversion qualification cutoff parameter to obtain the return conversion sample data of the current time step, thereby filtering the message recommendation sample data. Then, by sending the return conversion sample data to the message delivery party, the message delivery party can make message recommendations based on the return conversion sample data. In this way, the present application fully considers the customer group's conversion volume at the current time step, the conversion qualification distribution parameters of the customer group at the current time step, the link conversion characteristic data of the customer group at the current time step, and the conversion qualification truncation parameters of the customer group at the previous time step, which are related to the real-time changes of customers. Based on these customer group time series characteristic information, the conversion qualification truncation parameter determination model can be used to determine a target conversion qualification truncation parameter that is more closely related to the real-time changes of customers, and can improve the matching degree between the target conversion qualification truncation parameter and the real-time changing customers. Therefore, it overcomes the many factors that affect the conversion benefits of message recommendations, and the customer group is also changing in real time. The fixed conversion qualification limit is usually not matched with the real-time changing customers. Therefore, the accuracy of filtering the return samples based on the fixed conversion qualification limit is not high, thereby affecting the effect of message recommendation. Technical defects, thereby improving the effect of message recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0028] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0029] Figure 1 A flowchart of an embodiment of the message recommendation method of the present application;

[0030] Figure 2 A schematic diagram of an example scenario of the message recommendation method of this application;

[0031] Figure 3 A flowchart of another embodiment of the message recommendation method of the present application;

[0032] Figure 4 This is a schematic diagram of the conversion qualification distribution corresponding to different shape parameters and scale parameters in the message recommendation method of this application;

[0033] Figure 5 This is a flow chart of an embodiment of a method for determining truncation parameters for application conversion qualifications;

[0034] Figure 6 This is a structural diagram of a message recommendation device in an embodiment of the present application;

[0035] Figure 7 This is a schematic diagram of the device structure of the hardware operating environment involved in the message recommendation method in the embodiment of the present application.

[0036] The purpose, features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0037] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0038] Example 1

[0039] The present application embodiment provides a message recommendation method. In the first embodiment of the message recommendation method of the present application, referring to Figure 1 The message recommendation method is applied to the message recommender and includes the following steps:

[0040] Step S10: Acquire message feature data, wherein the message feature data includes at least one of the following: the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step;

[0041] In this embodiment, it should be noted that message recommendation involves the message recommender and the message delivery party, and this embodiment is applied to the message recommender. Currently, the message delivery party will establish a customer acquisition model to help the message recommender determine the target population for message recommendation and recommend messages to the determined target population. After the message recommendation, conversion may occur throughout the entire user conversion chain. The user conversion chain is the entire process from the exposure node where the message is first exposed to the user to the completion of user conversion. The user conversion chain can have multiple conversion nodes, and at each conversion node there will be user loss and retention. Retained users will continue to flow to the next conversion node until the conversion is completed or lost on the user conversion chain.

[0042] As an example, the message may be an advertisement, the message recommender may be an advertiser, and the message delivery party may be a media. Figure 2 , Figure 2 A scenario diagram of an example of a message recommendation method, such as Figure 2 As shown, the media helps advertisers acquire customers from the broader market by establishing a customer acquisition model. After acquiring customers, the advertisers transmit back the deep conversion data, that is, the message recommendation sample data, which can only be obtained by the advertiser side. Before the message recommendation sample data is transmitted back to the media side, the message recommendation sample data is filtered by the message recommendation device, and the filtered return conversion sample data is sent to the media side, so that the media side can update the customer acquisition model according to the return conversion sample data, and then continue to help advertisers acquire customers from the broader market based on the updated customer acquisition model. In this way, the target population for message recommendation of the customer acquisition model on the media side can be guided by filtering the message recommendation sample data.

[0043] As an example, in the field of financial loans, the user conversion link can be "exposure-click-leaving information-submission-credit (verification)-withdrawal". In this user conversion link, the message will first be exposed to the user, and the user can choose to click or not click the message. If the user clicks the message, the user can choose to conduct corporate authentication and leave corporate user information (leaving information). If the user conducts corporate authentication, he can choose to register an account. After the account registration is successful, he can choose to verify the loan amount. After the loan amount is verified, he can choose to withdraw the loan, thereby completing the normal amount loan process.

[0044] After a conversion occurs, the customer acquisition model of the message delivery party can update the customer acquisition model based on the generated conversion data to adjust the target population of the message recommendation based on the conversion data. However, conversion data includes shallow conversion data and deep conversion data. The shallow conversion data refers to the user behavior data generated by the exposure and click nodes, such as the number of user clicks, effective viewing time, cursor key point dwell time, etc. The shallow conversion data can be directly collected by the message delivery party; the deep conversion data refers to the user behavior data generated in the node after the click, such as the purchase amount, secondary purchase amount, recharge amount, pre-credit amount, credit amount, withdrawal amount, etc. The deep conversion data cannot be known by the message delivery party and needs to be returned by the message recommender to the message delivery party to improve the accuracy of the customer acquisition model in positioning the target population. The larger the magnitude of the deep conversion data returned to the message delivery party, the larger the sample size used by the message delivery party for modeling, the more fully the customer acquisition model is learned, and the more accurate the estimated conversion of the target population will be. However, in the media market, the qualification distribution of customers is usually long-tailed, with low-qualified customers far outnumbering high-qualified customers. Low-qualified customers bring less benefits to the message recommender. If all conversion data is directly transmitted back to the message publisher, it will cause the customer base to sink and the input-output ratio to decline. For example, if the conversion data transmitted back to the message publisher includes a large amount of conversion data of self-employed individuals, the modeling target of the message publisher will also be biased towards the self-employed population, which will directly affect the input-output ratio of the message recommendation side.

[0045] Currently, message recommenders typically set a fixed conversion qualification threshold based on manual experience, filter conversion data, and remove conversion data from less qualified customers before sending it back to the message distributor to attract more highly qualified customers. However, many factors influence the conversion returns of message recommendations, and customer groups change in real time. Fixed conversion qualification thresholds often don't match these changing customer groups. Therefore, filtering return samples based on fixed conversion qualification thresholds is inaccurate, which impacts the effectiveness of message recommendations. For example, if only the single dimension of conversion qualification is considered, excessive filtering will often occur. Since the filtered conversion data will be used as negative samples in the customer acquisition model, that is, the message publisher will determine that the click is not converted, resulting in the customer acquisition model's low estimate of the click-through conversion rate of the real customer group, thereby lowering the customer group's charge for thousand impressions. Since the customer acquisition model determines that the estimated conversion of this part of the customer group is low, in the actual exposure ranking, the message recommender's message is in a weak competitive position among the many messages to be recommended, which will lead to a decrease in the message exposure and a shrinking customer acquisition volume, which will in turn lead to a decrease in the conversion revenue of the message recommendation and a poor message recommendation effect.

[0046] The message feature data refers to feature data related to message recommendation conversion, including the conversion volume of the customer group in the current time step, the conversion qualification distribution parameters of the customer group in the current time step, the link conversion feature data of the customer group in the current time step and / or the conversion qualification cutoff parameters of the customer group in the previous time step, etc.

[0047] Among them, the conversion volume refers to the number of users who have conversion behaviors of exposure, click and submission after the message is recommended; the conversion qualification distribution parameter refers to the model parameter of the distribution model obeyed by the conversion qualification of the customer group within a period of time. The conversion qualification of the customer group within a period of time may obey the gamma distribution, beta distribution, etc., which can be determined specifically by fitting the distribution model; the link conversion feature data refers to the data of conversion behavior generated at each node in the entire user conversion link after the message is recommended. The link conversion feature data may include the link conversion feature data of this channel and the full-channel link conversion feature data. The link conversion feature data of this channel refers to the conversion of the entire user in this message channel after the message is recommended. The data on conversion behaviors generated by each node of the link, including the credit approval rate, average number of items submitted, average credit value, etc. of this channel, the said conversion characteristic data of this channel link can be used to characterize the actual conversion situation of this channel. The said omni-channel link conversion characteristic data refers to the data on conversion behaviors generated in each message channel other than this message channel in the market after the message recommendation, including the conversion volume, credit approval rate, average number of items submitted, average credit value, etc. of each message recommendation channel in the market. The said omni-channel link conversion characteristic data can be used to characterize the competitive factors of the market environment, thereby allowing the predicted target conversion qualification cutoff parameters to be adjusted based on the actual conversion situation of this channel and the market environment at the same time. Among them, the said conversion volume, conversion qualification distribution parameters and the said conversion characteristic data of this channel link can be obtained by processing and analyzing the message recommendation sample data corresponding to the customer group at the current time step.

[0048] Message recommendation sample data refers to user behavior data that the message recommender can obtain after a message recommendation, such as purchase amount, supplementary purchase amount, top-up amount, pre-credit amount, credit amount, withdrawal amount, etc. This refers to the conversion data that the message recommender needs to transmit back to the message publisher, after filtering or not, to update the customer acquisition model.

[0049] The conversion qualification cutoff parameter is used to determine the target conversion qualification range corresponding to the returned conversion sample data sent back to the message delivery party. The target conversion qualification cutoff parameter can be the lower limit of the target conversion qualification range, or it can include the lower limit and upper limit of the target conversion qualification range. In one implementable method, if the target conversion qualification cutoff parameter only limits the lower limit of the target conversion qualification range, the upper limit of the target conversion qualification range can be unlimited, or it can be determined based on the customer group classification information corresponding to the message recommendation method. For example, the maximum credit limit for small and micro enterprise loans can be determined as the upper limit of the conversion qualification retention range.

[0050] Each of the message feature data has a great impact on the conversion income of the message recommender after the message is recommended. Among them, the higher the conversion volume, the higher the conversion income, the higher the conversion qualification, the higher the conversion income, and the higher the credit average, the higher the conversion income. The conversion qualification cutoff parameter affects the conversion income by affecting the conversion volume and conversion qualification. The larger the target conversion qualification range corresponding to the conversion qualification cutoff parameter, the higher the conversion volume, but the more low-qualification samples may be included, which may lead to the sinking of the customer base and a decrease in the input-output ratio. Therefore, accurately determining the conversion qualification cutoff parameter and matching the real-time changing customer base to balance the impact of conversion volume and conversion qualification on conversion income can maximize the conversion income.

[0051] As an example, step S10 includes: pre-setting a time step for converting message recommendations, and obtaining message feature data of the customer group in the current time step at the end of each time step, wherein the length of the time step can be determined according to actual conditions, for example, 1 day, 3 days or 1 natural week, etc.

[0052] Furthermore, the step of obtaining the conversion qualification distribution parameters of the customer group at the current time step includes:

[0053] Step A10, obtaining user sample data of the customer group at the current time step;

[0054] As an example, step A10 includes: pre-setting a time step for converting message recommendations, and whenever the end time of a time step is reached, obtaining user sample data of all customers in the current time step, wherein the user sample data may be user portrait data.

[0055] Step A20, predicting the credit approval probability of the customer group at the current time step by inputting the sample data of each user into the credit approval model;

[0056] As an example, step A20 includes: inputting the user sample data of each customer in the current time step into a pre-trained credit approval model, and predicting the credit approval probability corresponding to each customer in the current time step through the credit approval model, wherein the credit approval model is a classification model, such as a logistic regression model or a support vector machine model.

[0057] In one practicable manner, before the step of predicting the credit approval probability of the customer group at the current time step by inputting the sample data of each user into the credit approval model, the message recommendation method further includes:

[0058] Step B10: Acquire qualification sample data of customer groups within a preset time range, wherein the qualification sample data includes user sample data and credit tags;

[0059] In this embodiment, before the step of predicting the credit approval probability of the customer group at the current time step by inputting the sample data of each user into the credit approval model, the credit approval model may also be trained.

[0060] As an example, step B10 includes: obtaining qualification sample data of the customer group within a preset time range, wherein the qualification sample data refers to user sample data of each user within the preset time range and the credit label of each user, and the preset time range is a larger time range including multiple time steps.

[0061] Step B20: determining the true value of the credit approval probability of the customer group within a preset time range based on the credit approval label;

[0062] As an example, step B20 includes: performing statistical analysis on the credit tags of the customer group within the preset time range, calculating and determining the true value of the credit approval probability of the customer group within the preset time range. For example, if the credit tag is detected to be a floating-point positive value, then it is determined that the customer corresponding to the credit tag has passed the credit, and the credit tag can also be determined as the credit amount. If it is detected that the credit tag is not a floating-point positive value, such as 0 or an empty value, then it is determined that the customer corresponding to the credit tag has not passed the credit, and the ratio of the number of customers who have passed the credit within the preset time range to the total number of customers within the preset time range is calculated to obtain the true value of the credit approval probability.

[0063] Step B30: training the classification model to be trained based on the true value of the credit approval probability of the customer group within the preset time range and the user sample data of the customer group within the preset time range to obtain a credit approval model.

[0064] As an example, the step B30 includes: obtaining target user sample data of target customers and the target credit approval probability true value of the target customers from the user sample data of the customer group within the preset time range; inputting the target user sample data into the credit approval model to be trained, evaluating the credit approval probability of the target customer, and obtaining a credit approval label, and calculating the model loss corresponding to the credit approval model to be trained based on the difference between the credit approval label and the target credit approval probability true value; judging whether the model loss converges, if the model loss converges, using the credit approval model to be trained as the credit approval model; if the model loss does not converge, updating the credit approval model to be trained based on the model gradient calculated by the model loss, and returning to execute the step of obtaining the target user sample data of the target customer and the target credit approval probability true value of the target customer from the user sample data of the customer group within the preset time range until the calculated model loss converges.

[0065] Step A30, predicting the credit amount of the customer group at the current time step by inputting the sample data of each user into the credit amount model;

[0066] As an example, step A30 includes: inputting user sample data of each customer in the current time step into a pre-trained credit amount model, and predicting the credit amount of each customer in the current time step through the credit amount model, wherein the credit amount model is a regression model.

[0067] In one practicable manner, before the step of predicting the credit amount of the customer group at the current time step by inputting the sample data of each user into the credit amount model, the message recommendation method further includes:

[0068] Step C10: Acquire qualification sample data of customer groups within a preset time range, wherein the qualification sample data includes user sample data and credit tags;

[0069] In this embodiment, before the step of predicting the credit amount of the customer group at the current time step by inputting the sample data of each user into the credit amount model, model training may be performed on the credit amount model.

[0070] As an example, step C10 includes: obtaining qualification sample data of the customer group within a preset time range, wherein the qualification sample data refers to user sample data of each user within the preset time range and the credit label of each user, and the preset time range is a larger time range including multiple time steps.

[0071] Step C20: determining the actual value of the credit amount of the customer group within a preset time range based on the credit label;

[0072] As an example, step C20 includes: performing statistical analysis on the credit tags of the customer group within a preset time range, and determining the credit amount of each customer within the preset time range. For example, if it is detected that the credit tag is a floating-point positive value, it is determined that the credit of the customer corresponding to the credit tag is approved, and then the credit tag is determined as the actual value of the credit amount of the corresponding customer. If it is detected that the credit tag is not a floating-point positive value, for example, it is 0 or a null value, it is determined that the credit of the customer corresponding to the credit tag is not approved, and the credit amount of the corresponding customer can be determined as 0 or a null value.

[0073] Step C30: training the regression model to be trained based on the actual value of the credit amount and the user sample data to obtain a credit amount model.

[0074] As an example, the step C30 includes: obtaining target user sample data of the target customer and the true value of the target credit amount of the target customer from the user sample data of the customer group within the preset time range; inputting the target user sample data into the credit amount model to be trained, evaluating the credit amount of the target customer, obtaining a credit amount label, and calculating the model loss corresponding to the credit amount model to be trained based on the difference between the credit amount label and the true value of the target credit amount; judging whether the model loss converges, if the model loss converges, using the credit amount model to be trained as the credit amount model; if the model loss does not converge, updating the credit amount model to be trained based on the model gradient calculated by the model loss, and returning to execute the step of obtaining the target user sample data of the target customer and the true value of the target credit amount of the target customer from the user sample data of the customer group within the preset time range until the calculated model loss converges.

[0075] Step A40: Determine the conversion qualification mark value of the customer group at the current time step based on the credit approval probability and the credit amount of the customer group at the current time step;

[0076] As an example, step A40 includes: adding, multiplying or weighted summing the credit approval probability and credit amount of each customer in the current time step, determining the calculation result as the conversion qualification labeling value of the customer group in the current time step, and labeling the conversion qualification of each customer in the current time step.

[0077] Step A50 , performing distribution model fitting on the conversion qualification labeling values of the customer group at the current time step to obtain the conversion qualification distribution parameters of the customer group at the current time step.

[0078] As an example, step A50 includes: performing distribution model fitting on the conversion qualification labeling values of all customers in the current time step, obtaining the conversion qualification distribution model of the customer group in the current time step after fitting, obtaining the model parameters of the conversion qualification distribution model, and using the model parameters of the conversion qualification distribution model as the conversion qualification distribution parameters of the customer group in the current time step.

[0079] In one practicable manner, the conversion qualifications of the customer group obey a gamma distribution within a time step, and the conversion qualification distribution data includes a shape parameter and a scale parameter.

[0080] Step S20, determining the target conversion qualification cutoff parameter of the customer group at the current time step by inputting the message feature data into a conversion qualification cutoff parameter determination model;

[0081] As an example, step S20 includes: splicing at least one of the conversion volume of the customer group at the current time step, the conversion qualification distribution parameters of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification truncation parameters of the customer group at the previous time step into a message feature vector, inputting the message feature vector into a pre-trained conversion qualification truncation parameter determination model, and determining the target conversion qualification truncation parameters of the customer group at the current time step through the conversion qualification truncation parameter determination model, wherein the conversion qualification truncation parameter determination model can be a fusion model of a regression model and a time series model, and the conversion qualification truncation parameter determination model can be obtained by iterative optimization in advance based on historical samples before the current time step.

[0082] Step S30: Obtain message recommendation sample data corresponding to the customer group at the current time step, filter the message recommendation sample data of the customer group at the current time step according to the target conversion qualification cutoff parameter, and obtain the return conversion sample data of the current time step;

[0083] As an example, step S30 includes: obtaining message recommendation sample data corresponding to the customer group at the current time step, determining the conversion qualification retention range of the conversion data that needs to be retained in the message recommendation sample data of the customer group at the current time step according to the target conversion qualification cutoff parameter, filtering out the message recommendation sample data that is not within the conversion qualification retention range, and retaining the message recommendation sample data within the conversion qualification retention range as return conversion sample data.

[0084] Step S40: Send the returned conversion sample data to the message delivery party, so that the message delivery party can make message recommendations based on the returned conversion sample data.

[0085] As an example, step S40 includes: sending the feedback conversion sample data retained after filtering to the message delivery party, so that the message delivery party can iteratively update the customer acquisition model based on the feedback conversion sample data, and then adjust the target population of the message recommendation through the updated customer acquisition model.

[0086] In this embodiment, by acquiring message feature data, wherein the message feature data includes at least one of the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification truncation parameter of the customer group at the previous time step, the acquisition of multi-dimensional message feature data is achieved. Then, by inputting the message feature data into the conversion qualification truncation parameter determination model, the target conversion qualification truncation parameter of the customer group at the current time step is determined. This realizes the determination of the target conversion qualification truncation parameter of the customer group at the current time step by integrating the multi-dimensional message feature data through the conversion qualification truncation parameter determination model. Then, by acquiring the message recommendation sample data corresponding to the customer group at the current time step, the message recommendation sample data of the customer group at the current time step is filtered according to the target conversion qualification truncation parameter to obtain the return conversion sample data of the current time step, thereby realizing the filtering of the message recommendation sample data. Then, by sending the return conversion sample data to the message delivery party, the message delivery party can make message recommendations based on the return conversion sample data. In this way, the present application fully considers the customer group's conversion volume at the current time step, the conversion qualification distribution parameters of the customer group at the current time step, the link conversion characteristic data of the customer group at the current time step, and the conversion qualification truncation parameters of the customer group at the previous time step, which are related to the real-time changes of customers. Based on these customer group time series characteristic information, the conversion qualification truncation parameter determination model can be used to determine a target conversion qualification truncation parameter that is more closely related to the real-time changes of customers, and can improve the matching degree between the target conversion qualification truncation parameter and the real-time changing customers. Therefore, it overcomes the many factors that affect the conversion benefits of message recommendations, and the customer group is also changing in real time. The fixed conversion qualification limit is usually not matched with the real-time changing customers. Therefore, the accuracy of filtering the return samples based on the fixed conversion qualification limit is not high, thereby affecting the effect of message recommendation. Technical defects, thereby improving the effect of message recommendation.

[0087] Example 2

[0088] Further, refer to Figure 3Based on the above embodiments of the present application, in the second embodiment of the present application, the same or similar contents as the above embodiments can be referred to the above introduction and will not be repeated hereafter. On this basis, the conversion qualification truncation parameter determination model includes a conversion volume prediction model and a conversion qualification prediction model. The step of inputting the message feature data into the conversion qualification truncation parameter determination model to determine the target conversion qualification truncation parameter for the customer group at the current time step includes:

[0089] Step S21, predicting the conversion qualification distribution data of the customer group at the next time step by inputting the message feature data of the customer group at the current time step into the conversion qualification prediction model;

[0090] In this embodiment, it should be noted that during the internet message recommendation process, the message publisher relies on the conversion data transmitted back by the message recommender to build a model, thereby accurately targeting the target audience for the message recommendation and helping the message recommender acquire customers accurately. The greater the amount of conversion data transmitted back to the message publisher, the larger the sample size used for modeling, the more thoroughly the customer acquisition model is learned, and the more accurate the conversion estimates for the target audience will be. However, in the media landscape, the customer base's qualifications are typically distributed in a long-tail pattern, with low-qualified customers far outnumbering high-qualified ones. Low-qualified customers bring less revenue to the message recommender. If all conversion data is transmitted directly back to the message publisher, the customer base will be shifted downward, reducing the input-output ratio. For example, if the conversion data transmitted back to the message publisher includes a large amount of conversion data for self-employed individuals, the message publisher's modeling target will also be biased towards the self-employed, directly affecting the input-output ratio of the message recommendation side. Therefore, message recommenders typically set a fixed conversion qualification threshold based on manual experience, filter conversion data, and remove conversion data from less qualified customers before sending it back to the message distributor to attract more highly qualified customers. However, many factors influence the conversion returns of message recommendations, and customer groups change in real time. Fixed conversion qualification thresholds often don't match the changing customer base. Therefore, filtering return samples based on fixed conversion qualification thresholds is inaccurate, which impacts the effectiveness of message recommendations. For example, if only the single dimension of conversion qualification is considered, excessive filtering often occurs. This is because the filtered conversion data will be treated as negative samples in the customer acquisition model, that is, the message publisher will determine that the click is not converted, which will cause the customer acquisition model to underestimate the click-through conversion rate of the real customer group, thereby lowering the cost per thousand impressions of the customer group. Because the customer acquisition model determines that the estimated conversion of this part of the customer group is low, in the actual exposure ranking, the message recommender's message is in a weak competitive position among the many messages to be recommended, which will lead to a decrease in the amount of message exposure and a shrinking customer acquisition volume, which will in turn lead to a decrease in the conversion revenue of the message recommendation and poor message recommendation effect. In other words, the current method of filtering conversion data based on a fixed conversion qualification limit has a technical defect of low accuracy in evaluating the conversion volume and the conversion qualification of the customer group after the message recommendation, due to the many factors that affect the conversion revenue of the message recommendation and the real-time changes of the customer group. This leads to poor message recommendation effect and low conversion revenue.

[0091] The conversion qualification cutoff parameter determination model refers to a fusion model obtained by combining a conversion amount prediction model and a conversion qualification prediction model, wherein the conversion amount prediction model is a time series model, for example, a GRU (Gated Recurrent Unit) model, an LSTM (Long Short-Term Memory) model, a Prohpet model (a time series prediction model), etc. The conversion amount has a strong dependence on the previous return data, so the accuracy of prediction through the time series model is higher, and the conversion qualification prediction model is a regression model.

[0092] As an example, the step S21 includes: splicing one or more of the conversion amount of the customer group at the current time step, the conversion qualification distribution parameters of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification truncation parameters of the customer group at the previous time step into a message feature vector, inputting the message feature vector into the pre-trained conversion qualification prediction model, and predicting the conversion qualification distribution data of the customer group at the next time step through the conversion qualification prediction model, wherein the conversion qualification prediction model is a regression model, and the conversion qualification distribution data can be a conversion qualification distribution parameter, that is, the conversion qualification prediction model can be used to predict the conversion qualification distribution data of the customer group at the next time step. The conversion qualification distribution parameters of the customer group; the conversion qualification distribution data can also be used to characterize the correspondence between the conversion qualification distribution parameters of the customer group at the next time step and the conversion qualification cutoff parameters of the customer group at the current time step. In this case, the conversion qualification distribution data can be a function in which the conversion qualification distribution parameters of the customer group at the next time step change with the conversion qualification cutoff parameters of the customer group at the current time step, or a one-to-one correspondence data table consisting of the conversion qualification distribution parameters of the customer group at the next time step and the conversion qualification cutoff parameters of the customer group at the current time step, etc. That is, in this case, when the conversion qualification cutoff parameters of the customer group at the current time step change, the conversion qualification distribution parameters of the customer group at the next time step will also change accordingly. The conversion qualification distribution parameters can be one or more. If there are multiple conversion qualification distribution parameters, each conversion qualification distribution parameter can be predicted separately by the conversion qualification prediction model corresponding to each conversion qualification distribution parameter.

[0093] Furthermore, the conversion qualification distribution data includes a shape parameter and a scale parameter, the conversion qualification prediction model includes a shape parameter prediction model and a scale parameter prediction model, and the step of predicting the conversion qualification distribution data of the customer group at the next time step by inputting the message feature data of the customer group at the current time step into the conversion qualification prediction model includes:

[0094] Step S211: Concatenate the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification truncation parameter of the customer group at the previous time step into a message feature vector at the current time step;

[0095] Step S212, predicting the shape parameters corresponding to the customer group at the next time step by inputting the message feature vector of the customer group at the current time step into the shape parameter prediction model;

[0096] Step S213 , predicting the scale parameter corresponding to the customer group at the next time step by inputting the message feature vector of the customer group at the current time step into the scale parameter prediction model.

[0097] In this embodiment, it should be noted that the conversion qualification distribution data can be a conversion qualification distribution parameter. The conversion qualification is gamma distributed within a certain time range. The gamma model is determined by the shape parameter and the scale parameter. Therefore, the conversion qualification distribution parameter includes the shape parameter and the scale parameter. The shape parameter and the scale parameter are highly abstract of the customer group qualification distribution. Figure 4 , Figure 4 This is a schematic diagram of the transformation qualification distribution corresponding to different shape parameters and scale parameters. Figure 4 In this example, k is the shape parameter and θ is the scale parameter. As shown in the figure, the smaller the shape parameter, the worse the average conversion qualification, while the larger the scale parameter, the flatter the distribution of conversion qualifications and the smaller the magnitude differences between different conversion qualifications. Therefore, we can use the shape parameter prediction model to predict the shape parameter for the next time step, and the scale parameter prediction model to predict the scale parameter for the next time step, thereby determining the distribution of conversion qualifications for the customer group at the next time step.

[0098] As an example, steps S211 to S213 include: splicing the conversion volume of the customer group at the current time step, the conversion qualification distribution parameters of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification truncation parameters of the customer group at the previous time step into a message feature vector, inputting the message feature vector into a pre-trained shape parameter prediction model, and predicting the shape parameters corresponding to the customer group at the next time step through the shape parameter prediction model, wherein the shape parameter prediction model can be a regression model; and inputting the message feature vector into a pre-trained scale parameter prediction model, and predicting the scale parameters corresponding to the customer group at the next time step through the scale parameter prediction model, wherein the scale parameter prediction model can be a regression model.

[0099] In one practicable manner, before the step of inputting the message feature vector of the customer group at the current time step into the shape parameter prediction model to predict the shape parameters corresponding to the customer group at the next time step, the method further includes:

[0100] Obtain message feature data of a historical time step and true value of shape parameters corresponding to the customer group of the next time step of the historical time step, wherein the message feature data of the historical time step at least includes the conversion amount of the customer group of the historical time step, the conversion qualification distribution parameter of the customer group of the historical time step, the link conversion feature data of the customer group of the historical time step, and the conversion qualification cutoff parameter of the customer group of the previous time step of the historical time step. Input the message feature data of the historical time step into the shape parameter prediction model to be trained, evaluate the shape parameters of the customer group of the next time step of the historical time step, and obtain a shape parameter prediction label; calculate the model loss of the shape parameter prediction model to be trained based on the difference between the shape parameter prediction label and the true value of the shape parameter; determine whether the model loss converges. If the model loss converges, use the shape parameter prediction model to be trained as the shape parameter prediction model. If the model loss does not converge, update the shape parameter prediction model to be trained based on the model gradient calculated by the model loss, and return to execute the step of obtaining the message feature data of the historical time step and the true value of the shape parameters of the customer group of the next time step of the historical time step until the calculated model loss converges.

[0101] Furthermore, before the step of predicting the conversion qualification distribution data of the customer group at the next time step by inputting the message feature data of the customer group at the current time step into the conversion qualification prediction model, the message recommendation method further includes:

[0102] Step D10: Determine the historical conversion qualification mark value of the customer group within a preset time range through the credit approval model and the credit amount model;

[0103] As an example, step D10 includes: obtaining historical user sample data within a preset time range, predicting the credit approval probability of the customer group within the preset time range by inputting each of the historical user sample data into a credit approval model, predicting the historical credit amount of each customer within the preset time range by inputting each of the historical user sample data into a credit amount model, marking the conversion qualifications of the customer group within the preset time range according to the historical credit approval probability of each customer within the preset time range, and according to the historical credit amount and the corresponding historical credit approval probability corresponding to each customer, and determining the historical conversion qualification marking value of each customer within the preset time range. For example, the historical credit approval probability corresponding to each customer within the preset time range can be multiplied or added with the corresponding historical credit amount, and the product or sum value can be used as the corresponding historical conversion qualification marking value.

[0104] Step D20, performing distribution model fitting according to the conversion qualification labeling values corresponding to each time step within the preset time range, to obtain the true value of the conversion qualification distribution parameter corresponding to each time step within the preset time range;

[0105] As an example, step D20 includes: dividing the customer groups within the preset time range according to time steps, determining the conversion qualification labeling value of the customer group corresponding to each time step, performing distribution model fitting on the conversion qualification labeling value of the customer group corresponding to each time step, determining the conversion qualification distribution model of the customer group at each time step, and then obtaining the true value of the conversion qualification distribution parameter corresponding to the conversion qualification distribution model of the customer group at each time step. Exemplarily, if the conversion qualification distribution model is a gamma model, then after fitting the gamma model on the conversion qualification labeling value of the customer group corresponding to each time step, the true value of the shape parameter and the true value of the scale parameter of the conversion qualification distribution of the customer group corresponding to each time step can be determined.

[0106] Step D30: Acquire a test set of message feature data for a customer group within a preset time range, wherein the test set of message feature data includes at least one of the following: the conversion volume of the customer group within the preset time range, the link conversion feature data of the customer group within the preset time range, and the conversion qualification cutoff parameter of the customer group within the preset time range;

[0107] As an example, step D30 includes: obtaining a message feature data test set of the customer group within a preset time range, wherein the message feature data test set includes at least one of the conversion volume of the customer group within the preset time range, the link conversion feature data of the customer group within the preset time range, and the conversion qualification truncation parameter of the customer group within the preset time range.

[0108] Step D40 , iteratively optimizing the to-be-trained conversion qualification prediction model based on the message feature data test set of the customer group within a preset time range and the true value of the conversion qualification distribution parameter to obtain a conversion qualification prediction model.

[0109] As an example, the step D40 includes: obtaining historical message feature data of the first time step from the message feature data test set of the customer group within the preset time range, and obtaining the true value of the first conversion qualification distribution parameter of the first time step and the true value of the second conversion qualification distribution parameter of the time step after the first time step; inputting the historical message feature data and the true value of the first conversion qualification distribution parameter into the conversion qualification prediction model to be trained, evaluating the conversion qualification distribution parameter of the time step after the first time step, and obtaining the conversion qualification distribution parameter prediction label; according to the difference between the conversion qualification distribution parameter prediction label and the true value of the second conversion qualification distribution parameter The method further comprises the steps of: calculating the model loss of the conversion qualification prediction model to be trained; judging whether the model loss converges; if the model loss converges, using the conversion qualification prediction model to be trained as the conversion qualification prediction model; if the model loss does not converge, updating the conversion qualification prediction model to be trained according to the model gradient calculated by the model loss, and returning to execute the steps of obtaining the historical message feature data of the first time step from the message feature data test set of the customer group within the preset time range, and obtaining the true value of the first conversion qualification distribution parameter of the first time step and the true value of the second conversion qualification distribution parameter in the next time step of the first time step, until the calculated model loss converges.

[0110] Step S22: inputting the message feature data into the conversion volume prediction model to predict the conversion volume data of the customer group at the next time step, wherein the conversion volume data is used to represent the corresponding relationship between the conversion volume of the customer group at the next time step and the conversion qualification cutoff parameter of the customer group at the current time step;

[0111] As an example, step S22 includes: splicing one or more of the conversion volume of the current time step customer group, the conversion qualification distribution parameters of the current time step customer group, the link conversion feature data of the current time step customer group, and the conversion qualification truncation parameters of the previous time step customer group into a message feature vector, inputting the message feature vector into the pre-trained conversion volume prediction model, and predicting the conversion volume data of the next time step customer group through the conversion volume prediction model, wherein the conversion volume data is used to characterize the correspondence between the conversion volume of the next time step customer group and the conversion qualification truncation parameters of the current time step customer group. The conversion volume data can be a function of the conversion volume of the next time step customer group changing with the conversion qualification truncation parameters of the current time step customer group, or it can be a one-to-one corresponding data table consisting of the conversion volume of the next time step customer group and the conversion qualification truncation parameters of the current time step customer group, etc. That is, when the conversion qualification truncation parameters of the current time step customer group change, the conversion volume of the next time step customer group will also change accordingly.

[0112] Furthermore, the step of predicting the conversion volume data of the customer group at the next time step by inputting the message feature data into the conversion volume prediction model includes:

[0113] Step S221: Concatenate the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification truncation parameter of the customer group at the previous time step into a message feature vector at the current time step;

[0114] Step S222 , predicting the conversion volume data of the customer group in the next time step by inputting the message feature vector of the current time step into the conversion volume prediction model.

[0115] As an example, steps S221 to S222 include: concatenating the conversion volume of the customer group at the current time step, the conversion qualification distribution parameters of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification truncation parameters of the customer group at the previous time step into a message feature vector, inputting the message feature vector into the pre-trained conversion volume prediction model, and predicting the conversion volume data of the customer group at the next time step through the conversion volume prediction model.

[0116] In one practicable manner, before the step of predicting the conversion volume data of the customer group at the next time step by inputting the message feature data into the conversion volume prediction model, the method further includes:

[0117] Obtain message feature data of a historical time step and the true value of the conversion volume of the customer group in the next time step of the historical time step, wherein the message feature data of the historical time step at least includes the conversion volume of the customer group in the historical time step, the conversion qualification distribution parameter of the customer group in the historical time step, the link conversion feature data of the customer group in the historical time step, and one of the conversion qualification cutoff parameters of the customer group in the previous time step of the historical time step. Input the message feature data of the historical time step into the conversion volume prediction model to be trained, evaluate the conversion volume of the customer group in the next time step of the historical time step, and obtain a conversion volume prediction label; calculate the model loss of the conversion volume prediction model to be trained based on the difference between the conversion volume prediction label and the true value of the conversion volume; determine whether the model loss converges. If the model loss converges, use the conversion volume prediction model to be trained as the conversion volume prediction model. If the model loss does not converge, update the conversion volume prediction model to be trained based on the model gradient calculated by the model loss, and return to execute the step of obtaining the message feature data of the historical time step and the true value of the conversion volume of the customer group in the next time step of the historical time step until the calculated model loss converges.

[0118] Step S23: determining the target conversion qualification cutoff parameter based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step.

[0119] As an example, the step S23 includes: for each conversion qualification cutoff parameter, the estimated conversion volume of the corresponding customer group in the next time step can be determined based on the conversion volume data of the customer group in the next time step. According to the conversion qualification distribution data of the customer group in the next time step, if the conversion qualification distribution data is a conversion qualification distribution parameter, there is no need to determine the conversion qualification cutoff parameter, and the total estimated conversion qualification of the customer group in the next time step can be determined. At this time, the conversion qualification cutoff parameter determination model is relatively simple and the prediction efficiency is high; if the conversion qualification distribution data is used to characterize the correspondence between the conversion amount of the customer group in the next time step and the conversion qualification truncation parameter of the customer group in the current time step, then for each conversion qualification truncation parameter, the total estimated conversion qualification of the customer group in the next time step can be determined, that is, the total estimated conversion qualification and the estimated conversion amount may change with the target conversion qualification truncation parameter, and then the conversion amount and conversion qualification can be analyzed and evaluated simultaneously to determine the target conversion qualification truncation parameter when the estimated conversion rate and the total estimated conversion qualification meet the requirements, which can effectively improve the prediction accuracy of the target conversion qualification truncation parameter and improve the conversion revenue after message recommendation.

[0120] Furthermore, the step of determining the target conversion qualification cutoff parameter based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step includes:

[0121] Step S231, calculating the conversion revenue evaluation value of the customer group at the next time step under different estimated conversion qualification cutoff parameters based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step;

[0122] As an example, step S231 includes: according to the conversion qualification distribution parameters of the customer group in the next time step, the distribution of the conversion qualifications of the customer group in the next time step can be determined, and the total estimated conversion qualifications of the customer group in the corresponding next time step can be determined by integration, addition, etc.; for each conversion qualification cutoff parameter, according to the conversion volume data of the customer group in the next time step, the estimated conversion volume of the customer group in the corresponding next time step can be determined. The sum or product of the total estimated conversion qualifications and the estimated conversion rate is determined as the conversion income corresponding to the estimated conversion qualification cutoff parameter. The conversion income obtained at this time is only related to the conversion qualification cutoff parameter. Therefore, for each conversion qualification cutoff parameter, a corresponding conversion income evaluation value can be determined, wherein the conversion income can be a conversion income evaluation function, or a set of conversion income evaluation values corresponding to each conversion qualification cutoff parameter.

[0123] In one practicable manner, the step of calculating the conversion revenue evaluation value of the customer group at the next time step under different estimated conversion qualification cutoff parameters based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step includes:

[0124] The conversion qualification distribution data of the customer group in the next time step under different estimated conversion qualification cutoff parameters are integrated to obtain the total estimated conversion qualification value of the customer group in the next time step under different estimated conversion qualification cutoff parameters. The total estimated conversion qualification value and the estimated conversion rate corresponding to the same estimated conversion qualification cutoff parameter are multiplied to determine the conversion benefit evaluation value of the customer group in the next time step under different estimated conversion qualification cutoff parameters.

[0125] Step S232: Select the estimated conversion qualification cutoff parameter corresponding to the maximum conversion revenue evaluation value as the target conversion qualification cutoff parameter.

[0126] As an example, step S232 includes: calculating the conversion revenue evaluation value corresponding to each estimated conversion qualification truncation parameter, comparing them, and determining the estimated conversion qualification truncation parameter corresponding to the maximum conversion revenue evaluation value as the target conversion qualification truncation parameter; or constructing a conversion revenue function based on the conversion volume data of the customer group in the next time step and the conversion qualification distribution data of the customer group in the next time step, and using the conversion revenue function as the objective function. By maximizing the objective function, the target conversion qualification truncation parameter corresponding to the maximum conversion revenue evaluation value can be obtained.

[0127] In this embodiment, the conversion revenue after the message recommendation is split into two parts, namely conversion volume and conversion qualification, for evaluation. The conversion qualification distribution data of the customer group at the next time step is predicted by the conversion qualification prediction model, thereby realizing an independent prediction of the conversion qualification of the customer group at the next time step. The conversion volume data of the customer group at the next time step is predicted by the conversion volume prediction model, thereby realizing an independent prediction of the conversion volume of the customer group at the next time step. Furthermore, according to the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step, the target conversion qualification cutoff parameter is determined, thereby realizing the determination of the target conversion qualification cutoff parameter based on the comprehensive evaluation of the conversion volume and conversion qualification. Compared with the method of filtering out customers with low conversion qualifications based solely on conversion qualifications, the target conversion qualification cutoff parameters determined after a comprehensive evaluation of conversion volume and conversion qualifications can effectively avoid excessive filtering when filtering out customer groups, and effectively balance the impact of conversion volume and conversion qualifications on overall conversion revenue, thereby increasing conversion revenue and improving message recommendation effects. This overcomes the many factors that affect the conversion revenue of message recommendations, as well as the fact that customer groups are changing in real time and that the accuracy of evaluating the conversion volume and customer group's conversion qualifications after message recommendations is low, leading to poor message recommendation effects and low conversion revenue. This technical defect improves the effect of message recommendations.

[0128] Example 3

[0129] This application embodiment provides a method for determining conversion qualification cutoff parameters, referring to Figure 5 The conversion qualification truncation parameter determination method is applied to the message recommender, and includes the following steps:

[0130] Step E10: Acquire message feature data, wherein the message feature data includes at least one of the following: the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step;

[0131] Step E20, inputting the message feature data of the customer group at the current time step into a conversion qualification prediction model to predict the conversion qualification distribution data of the customer group at the next time step, and inputting the message feature data into a conversion quantity prediction model to predict the conversion quantity data of the customer group at the next time step;

[0132] Step E30 , determining a target conversion qualification cutoff parameter based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step.

[0133] In this embodiment, it should be noted that during the internet message recommendation process, the message publisher relies on the conversion data transmitted back by the message recommender to build a model, thereby accurately targeting the target audience for the message recommendation and helping the message recommender acquire customers accurately. The greater the amount of conversion data transmitted back to the message publisher, the larger the sample size used for modeling, the more thoroughly the customer acquisition model is learned, and the more accurate the conversion estimates for the target audience will be. However, in the media landscape, the customer base's qualifications are typically distributed in a long-tail pattern, with low-qualified customers far outnumbering high-qualified ones. Low-qualified customers bring less revenue to the message recommender. If all conversion data is transmitted directly back to the message publisher, the customer base will be shifted downward, reducing the input-output ratio. For example, if the conversion data transmitted back to the message publisher includes a large amount of conversion data for self-employed individuals, the message publisher's modeling target will also be biased towards the self-employed, directly affecting the input-output ratio of the message recommendation side. Therefore, message recommenders typically set a fixed conversion qualification threshold based on manual experience, filter conversion data, and remove conversion data from less qualified customers before sending it back to the message distributor to attract more highly qualified customers. However, many factors influence the conversion returns of message recommendations, and customer groups change in real time. Fixed conversion qualification thresholds often don't match the changing customer base. Therefore, filtering return samples based on fixed conversion qualification thresholds is inaccurate, which impacts the effectiveness of message recommendations. For example, if only the single dimension of conversion qualification is considered, excessive filtering often occurs. This is because the filtered conversion data will be treated as negative samples in the customer acquisition model, that is, the message publisher will determine that the click is not converted, which will cause the customer acquisition model to underestimate the click-through conversion rate of the real customer group, thereby lowering the cost per thousand impressions of the customer group. Because the customer acquisition model determines that the estimated conversion of this part of the customer group is low, in the actual exposure ranking, the message recommender's message is in a weak competitive position among the many messages to be recommended, which will lead to a decrease in the amount of message exposure and a shrinking customer acquisition volume, which will in turn lead to a decrease in the conversion revenue of the message recommendation and poor message recommendation effect. In other words, the current method of filtering conversion data based on a fixed conversion qualification limit has a technical defect of low accuracy in evaluating the conversion volume and the conversion qualification of the customer group after the message recommendation, due to the many factors that affect the conversion revenue of the message recommendation and the real-time changes of the customer group. This leads to poor message recommendation effect and low conversion revenue.

[0134] In this embodiment, the message recommendation involves the message recommender and the message delivery party, and this embodiment is applied to the message recommender. Currently, the message delivery party will establish a customer acquisition model to help the message recommender determine the target population for message recommendation and recommend messages to the determined target population. After the message recommendation, conversion may occur throughout the entire user conversion chain. The user conversion chain is the entire process from the exposure node where the message is first exposed to the user to the completion of user conversion. The user conversion chain can have multiple conversion nodes, and at each conversion node there will be user loss and retention. Retained users will continue to flow to the next conversion node until the conversion is completed or lost in the user conversion chain.

[0135] As an example, the message may be an advertisement, the message recommender may be an advertiser, and the message delivery party may be a media. Figure 2 , Figure 2 A scenario diagram of an example of a message recommendation method, such as Figure 2 As shown, the media helps advertisers acquire customers from the broader market by establishing a customer acquisition model. After acquiring customers, the advertisers transmit back the deep conversion data, that is, the message recommendation sample data, which can only be obtained by the advertiser side. Before the message recommendation sample data is transmitted back to the media side, the message recommendation sample data is filtered by a message recommendation device. The message recommendation device includes a conversion qualification cutoff parameter determination device, which sends the filtered returned conversion sample data to the media side, so that the media side can update the customer acquisition model according to the returned conversion sample data, and then continue to help advertisers acquire customers from the broader market based on the updated customer acquisition model. In this way, the cycle can be repeated to guide the target population for message recommendation by the customer acquisition model on the media side by filtering the message recommendation sample data.

[0136] As an example, in the field of financial loans, the user conversion link can be "exposure-click-leaving information-submission-credit (verification)-withdrawal". In this user conversion link, the message will first be exposed to the user, and the user can choose to click or not click the message. If the user clicks the message, the user can choose to conduct corporate authentication and leave corporate user information (leaving information). If the user conducts corporate authentication, he can choose to register an account. After the account registration is successful, he can choose to verify the loan amount. After the loan amount is verified, he can choose to withdraw the loan, thereby completing the normal amount loan process.

[0137] After a conversion occurs, the customer acquisition model of the message publisher can update the customer acquisition model based on the generated conversion data to adjust the target population of the message recommendation based on the conversion data. However, conversion data includes shallow conversion data and deep conversion data. The shallow conversion data refers to the user behavior data generated by the exposure and click nodes, such as the number of user clicks, effective viewing time, cursor key point dwell time, etc. The shallow conversion data can be directly collected by the message publisher; the deep conversion data refers to the user behavior data generated in the node after the click, such as the purchase amount, secondary purchase amount, recharge amount, pre-credit amount, credit amount, withdrawal amount, etc. The deep conversion data cannot be known by the message publisher and needs to be transmitted back by the message recommender to the message publisher to improve the accuracy of the customer acquisition model in positioning the target population.

[0138] The message feature data refers to feature data related to message recommendation conversion, including the conversion volume of the customer group in the current time step, the conversion qualification distribution parameters of the customer group in the current time step, the link conversion feature data of the customer group in the current time step and / or the conversion qualification cutoff parameters of the customer group in the previous time step, etc.

[0139] Among them, the conversion volume refers to the number of users who have conversion behaviors of exposure, click and submission after the message is recommended; the conversion qualification distribution parameter refers to the model parameter of the distribution model obeyed by the conversion qualification of the customer group within a period of time. The conversion qualification of the customer group within a period of time may obey the gamma distribution, beta distribution, etc., which can be determined specifically by fitting the distribution model; the link conversion feature data refers to the data of conversion behavior generated at each node in the entire user conversion link after the message is recommended. The link conversion feature data may include the link conversion feature data of this channel and the full-channel link conversion feature data. The link conversion feature data of this channel refers to the conversion of the entire user in this message channel after the message is recommended. The data on conversion behaviors generated by each node of the link, including the credit approval rate, average number of items submitted, average credit value, etc. of this channel, the said conversion characteristic data of this channel link can be used to characterize the actual conversion situation of this channel. The said omni-channel link conversion characteristic data refers to the data on conversion behaviors generated in each message channel other than this message channel in the market after the message recommendation, including the conversion volume, credit approval rate, average number of items submitted, average credit value, etc. of each message recommendation channel in the market. The said omni-channel link conversion characteristic data can be used to characterize the competitive factors of the market environment, thereby allowing the predicted target conversion qualification cutoff parameters to be adjusted based on the actual conversion situation of this channel and the market environment at the same time. Among them, the said conversion volume, conversion qualification distribution parameters and the said conversion characteristic data of this channel link can be obtained by processing and analyzing the message recommendation sample data corresponding to the customer group at the current time step.

[0140] Message recommendation sample data refers to user behavior data that the message recommender can obtain after a message recommendation, such as purchase amount, supplementary purchase amount, top-up amount, pre-credit amount, credit amount, withdrawal amount, etc. This refers to the conversion data that the message recommender needs to transmit back to the message publisher, after filtering or not, to update the customer acquisition model.

[0141] The conversion qualification cutoff parameter is used to determine the target conversion qualification range corresponding to the returned conversion sample data sent back to the message delivery party. The target conversion qualification cutoff parameter can be the lower limit of the target conversion qualification range, or it can include the lower limit and upper limit of the target conversion qualification range. In one implementable method, if the target conversion qualification cutoff parameter only limits the lower limit of the target conversion qualification range, the upper limit of the target conversion qualification range can be unlimited, or it can be determined based on the customer group classification information corresponding to the conversion qualification cutoff parameter determination method. For example, the maximum credit limit for small and micro enterprise loans can be determined as the upper limit of the conversion qualification retention range.

[0142] Each of the message feature data has a great impact on the conversion income of the message recommender after the message is recommended. Among them, the higher the conversion volume, the higher the conversion income, the higher the conversion qualification, the higher the conversion income, and the higher the credit average, the higher the conversion income. The conversion qualification cutoff parameter affects the conversion income by affecting the conversion volume and conversion qualification. The larger the target conversion qualification range corresponding to the conversion qualification cutoff parameter, the higher the conversion volume, but the more low-qualification samples may be included, which may lead to the sinking of the customer base and a decrease in the input-output ratio. Therefore, accurately determining the conversion qualification cutoff parameter and matching the real-time changing customer base to balance the impact of conversion volume and conversion qualification on conversion income can maximize the conversion income.

[0143] As an example, steps E10 to E30 include: presetting a time step for converting message recommendations, and obtaining message feature data for the customer group at the current time step at the end of each time step, wherein the length of the time step can be determined based on actual conditions, for example, 1 day, 3 days, or 1 natural week. Furthermore, at least one of the following is concatenated: the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step into a message feature vector. The message feature vector is input into the pre-trained conversion qualification prediction model and conversion volume prediction model, respectively. The conversion qualification prediction model is used to predict the conversion qualification distribution data of the customer group at the next time step, and the conversion volume prediction model is used to predict the conversion volume data of the customer group at the next time step. Furthermore, for each conversion qualification cutoff parameter, the estimated conversion volume and / or estimated conversion qualification of the corresponding customer group at the next time step can be determined. Specifically, for each conversion qualification cutoff parameter, the estimated conversion amount of the corresponding customer group in the next time step can be determined based on the conversion amount data of the customer group in the next time step. Based on the conversion qualification distribution data of the customer group in the next time step, if the conversion qualification distribution data is a conversion qualification distribution parameter, there is no need to determine the conversion qualification cutoff parameter, and the total estimated conversion qualification of the customer group in the next time step can be determined. At this time, the conversion qualification cutoff parameter determination model is relatively simple and the prediction efficiency is high; if the conversion qualification distribution data is used to characterize the correspondence between the conversion amount of the customer group in the next time step and the conversion qualification cutoff parameter of the customer group in the current time step, then for each conversion qualification cutoff parameter, the total estimated conversion qualification of the customer group in the next time step can be determined, that is, the total estimated conversion qualification and the estimated conversion amount may change with the target conversion qualification cutoff parameter, and then the conversion amount and conversion qualification can be analyzed and evaluated simultaneously to determine the target conversion qualification cutoff parameter when the estimated conversion rate and the total estimated conversion qualification meet the requirements. This can effectively improve the prediction accuracy of the target conversion qualification cutoff parameter and improve the conversion revenue after the message recommendation.

[0144] Wherein, the conversion qualification prediction model is a regression model. The conversion qualification distribution data can be a conversion qualification distribution parameter, that is, the conversion qualification distribution parameter of the customer group at the next time step can be predicted by the qualification prediction model; the conversion qualification distribution data can also be used to characterize the correspondence between the conversion qualification distribution parameter of the customer group at the next time step and the conversion qualification cutoff parameter of the customer group at the current time step. In this case, the conversion qualification distribution data can be a function of the conversion qualification distribution parameter of the customer group at the next time step changing with the conversion qualification cutoff parameter of the customer group at the current time step, or a one-to-one correspondence data table consisting of the conversion qualification distribution parameter of the customer group at the next time step and the conversion qualification cutoff parameter of the customer group at the current time step, etc. That is, in this case, when the conversion qualification cutoff parameter of the customer group at the current time step changes, the conversion qualification distribution parameter of the customer group at the next time step will also change accordingly. The conversion qualification distribution parameter can be one or more. If the conversion qualification distribution parameter is multiple, each conversion qualification distribution parameter can be predicted separately by the conversion qualification prediction model corresponding to each conversion qualification distribution parameter.

[0145] The conversion volume prediction model is a time series model, for example, a GRU (Gated Recurrent Unit) model, an LSTM (Long Short-Term Memory) model, a Prohpet model (a time series prediction model), etc. The conversion volume has a strong dependence on the previous return data, so the accuracy of prediction through the time series model is higher. The conversion volume data is used to characterize the correspondence between the conversion volume of the customer group at the next time step and the conversion qualification cutoff parameter of the customer group at the current time step. The conversion volume data can be a function of the conversion volume of the customer group at the next time step that changes with the conversion qualification cutoff parameter of the customer group at the current time step, or a one-to-one corresponding data table consisting of the conversion volume of the customer group at the next time step and the conversion qualification cutoff parameter of the customer group at the current time step, etc. That is, when the conversion qualification cutoff parameter of the customer group at the current time step changes, the conversion volume of the customer group at the next time step will also change accordingly.

[0146] Furthermore, the step of determining the target conversion qualification cutoff parameter based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step includes:

[0147] Step E31, calculating the conversion revenue evaluation value of the customer group at the next time step under different estimated conversion qualification cutoff parameters based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step;

[0148] Step E32: Select the estimated conversion qualification cutoff parameter corresponding to the maximum conversion benefit evaluation value as the target conversion qualification cutoff parameter.

[0149] In this embodiment, the specific implementation process of steps E31 to E32 can refer to the contents of steps S231 to S232 in the above embodiment, and will not be repeated here.

[0150] Furthermore, before the step of predicting the conversion qualification distribution data of the customer group at the next time step by inputting the message feature data of the customer group at the current time step into the conversion qualification prediction model, the conversion qualification cutoff parameter determination method further includes:

[0151] Step F10: Determine the historical conversion qualification mark value of the customer group within a preset time range through the credit approval model and the credit amount model;

[0152] Step F20, performing distribution model fitting according to the conversion qualification labeling value corresponding to each time step within the preset time range, and obtaining the true value of the conversion qualification distribution parameter corresponding to each time step within the preset time range;

[0153] Step F30: Acquire a test set of message feature data for a customer group within a preset time range, wherein the test set of message feature data includes at least one of the following: the conversion volume of the customer group within the preset time range, the link conversion feature data of the customer group within the preset time range, and the conversion qualification cutoff parameter of the customer group within the preset time range;

[0154] Step F40 , iteratively optimizing the to-be-trained conversion qualification prediction model based on the message feature data test set of the customer group within a preset time range and the true value of the conversion qualification distribution parameter to obtain a conversion qualification prediction model.

[0155] In this embodiment, the specific implementation process of steps F10 to F40 may refer to the contents of steps D10 to D40 in the above embodiment, and will not be repeated here.

[0156] In one practicable manner, the step of predicting the conversion volume data of the customer group at the next time step by inputting the message feature data into the conversion volume prediction model includes:

[0157] Step G10: Concatenate the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification truncation parameter of the customer group at the previous time step into a message feature vector at the current time step;

[0158] Step G20: predicting the conversion volume data of the customer group in the next time step by inputting the message feature vector of the current time step into the conversion volume prediction model.

[0159] In this embodiment, the specific implementation process of steps G10 to G20 can refer to the contents of steps S221 to S222 in the above embodiment, and will not be repeated here.

[0160] In one practicable manner, the conversion qualification distribution data includes a shape parameter and a scale parameter, the conversion qualification prediction model includes a shape parameter prediction model and a scale parameter prediction model, and the step of predicting the conversion qualification distribution data of the customer group at the next time step by inputting the message feature data of the customer group at the current time step into the conversion qualification prediction model includes:

[0161] Step H10: Concatenate the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification truncation parameter of the customer group at the previous time step into a message feature vector at the current time step;

[0162] Step H20, predicting the shape parameters corresponding to the customer group at the next time step by inputting the message feature vector of the customer group at the current time step into the shape parameter prediction model;

[0163] Step H30 , predicting the scale parameter corresponding to the customer group at the next time step by inputting the message feature vector of the customer group at the current time step into the scale parameter prediction model.

[0164] In this embodiment, the specific implementation process of steps H10 to H30 can refer to the contents of steps S211 to S213 in the above embodiment, and will not be repeated here.

[0165] In one practicable manner, the step of obtaining the conversion qualification distribution parameters of the customer group at the current time step includes:

[0166] Step I10, obtaining user sample data of the customer group at the current time step;

[0167] Step I20: inputting the sample data of each user into a credit approval model to predict the credit approval probability of the customer group at the current time step;

[0168] Step I30: inputting the sample data of each user into a credit amount model to predict the credit amount of the customer group at the current time step;

[0169] Step I40: Determine the conversion qualification mark value of the customer group at the current time step based on the credit approval probability and the credit amount of the customer group at the current time step;

[0170] Step I50, performing distribution model fitting on the conversion qualification labeling values of the customer group at the current time step to obtain the conversion qualification distribution parameters of the customer group at the current time step.

[0171] In this embodiment, the specific implementation process of steps I10 to I50 can refer to the contents of steps A10 to A50 in the above embodiment, and will not be repeated here.

[0172] In this embodiment, by acquiring message feature data, wherein the message feature data includes at least one of the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification truncation parameter of the customer group at the previous time step, the acquisition of multi-dimensional message feature data is achieved. Then, by inputting the message feature data of the customer group at the current time step into the conversion qualification prediction model to predict the conversion qualification distribution data of the customer group at the next time step, and by inputting the message feature data into the conversion volume prediction model to predict the conversion volume data of the customer group at the next time step, the independent prediction of the conversion volume data and the conversion qualification distribution data of the customer group at the next time step by comprehensive multi-dimensional message feature data is achieved. Then, by determining the target conversion qualification truncation parameter based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step, the purpose of determining the target conversion qualification truncation parameter based on the comprehensive evaluation of the conversion volume and conversion qualification is achieved. In this way, the present application fully considers the customer group's conversion volume at the current time step, the conversion qualification distribution parameters of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameters of the customer group at the previous time step, which are associated with the real-time changes of the customers. Based on these customer group time series feature information, the conversion volume data and conversion qualification cutoff parameters of the customer group at the next time step are predicted respectively, and the conversion volume data and conversion qualification data of the customer group at the next time step with a higher correlation with the real-time changes of the customers are predicted, and then the target customer group is determined by combining the conversion volume data and conversion qualification cutoff parameters of the customer group at the next time step. The target conversion qualification cutoff parameter can determine a target conversion qualification cutoff parameter that is more closely related to the real-time changes in customers, and can improve the matching degree between the target conversion qualification cutoff parameter and the real-time changing customers. On the other hand, compared with the method of filtering out customers with low conversion qualifications based solely on conversion qualifications, the target conversion qualification cutoff parameter determined after a comprehensive evaluation of the conversion volume and conversion qualifications of the customer group in the next time step has higher accuracy when filtering out the customer group, which can effectively avoid excessive filtering and effectively balance the impact of conversion volume and conversion qualifications on the overall conversion revenue, thereby increasing conversion revenue and improving message recommendation effects. This overcomes the fact that there are many factors that affect the conversion revenue of message recommendations, the customer group is also changing in real time, and there is a technical defect that the accuracy of the conversion volume and conversion qualifications of the customer group after evaluating the message recommendation is low, which leads to poor message recommendation effects and low conversion revenue, thereby improving the effect of message recommendations.

[0173] Example 4

[0174] This embodiment of the present application provides a message recommendation method, which is applied to a message delivery party. The message recommendation method includes:

[0175] Step J10: Receive the returned conversion sample data sent by the message recommender, wherein the returned conversion sample data is obtained by the message recommender by filtering the message recommendation sample data of the current time step customer group according to the target conversion qualification cutoff parameter, and the target conversion qualification cutoff parameter is obtained by the message recommender by inputting message feature data into a conversion qualification cutoff parameter determination model for prediction, and the message feature data includes at least one of the conversion volume of the current time step customer group, the conversion qualification distribution parameter of the current time step customer group, the link conversion feature data of the current time step customer group, and the conversion qualification cutoff parameter of the previous time step customer group;

[0176] Step J20: Determine the target customer group based on the returned conversion sample data, and recommend messages to the target customer group.

[0177] In this embodiment, the target customer group is determined based on the returned conversion sample data by inputting the returned conversion sample data and other special data required by the customer acquisition model into the customer acquisition model to predict the target customer group. The specific implementation process of the message recommender filtering the message recommendation sample data for the customer group at the current time step based on the target conversion qualification cutoff parameter and inputting the message feature data into the conversion qualification cutoff parameter determination model for prediction can be referred to the specific contents of steps S10 to S40 above and will not be repeated here.

[0178] The embodiment of the present application provides a message recommendation method, which determines the target customer group based on the feedback conversion sample data sent by the message publisher, and recommends messages to the target customer group. Since the feedback conversion sample data used to determine the target customer group is the customer group time series feature information associated with the real-time changes of the customer based on the conversion volume of the customer group at the current time step, the conversion qualification distribution parameters of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification truncation parameters of the customer group at the previous time step, the conversion qualification truncation parameters are used to determine the model, which can predict a customer group with a higher correlation with the real-time changes of the customer. The target conversion qualification cutoff parameter improves the matching degree between the target conversion qualification cutoff parameter and the real-time changing customers, and then filters the message recommendation sample data based on the target conversion qualification cutoff parameter. Therefore, the matching degree between the returned conversion sample data and the real-time changing customers will also be higher, so it overcomes many factors that affect the message recommendation conversion revenue, the customer base is also changing in real time, and the fixed conversion qualification limit is usually not matched with the real-time changing customers. Therefore, the accuracy of filtering the returned samples based on the fixed conversion qualification limit is not high, thereby affecting the effect of message recommendation. The technical defect improves the effect of message recommendation.

[0179] Example 5

[0180] Furthermore, the embodiment of the present application also provides a message recommendation device, referring to Figure 6 The message recommendation device is applied to the message recommendation party, including:

[0181] A first acquisition module 10 is configured to acquire message feature data, wherein the message feature data includes at least one of the following: the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step;

[0182] A first determination module 20 is configured to determine a target conversion qualification cutoff parameter for the customer group at the current time step by inputting the message feature data into a conversion qualification cutoff parameter determination model;

[0183] The filtering module 30 is used to obtain the message recommendation sample data corresponding to the customer group at the current time step, and filter the message recommendation sample data of the customer group at the current time step according to the target conversion qualification cutoff parameter to obtain the return conversion sample data of the current time step;

[0184] The sending module 40 is configured to send the returned conversion sample data to a message delivery party, so that the message delivery party can make message recommendations based on the returned conversion sample data.

[0185] Optionally, the first determining module 20 is further configured to:

[0186] By inputting the message feature data of the customer group at the current time step into the conversion qualification prediction model, the conversion qualification distribution data of the customer group at the next time step is predicted;

[0187] The conversion volume data of the customer group at the next time step is predicted by inputting the message feature data into the conversion volume prediction model, wherein the conversion volume data is used to represent the corresponding relationship between the conversion volume of the customer group at the next time step and the conversion qualification cutoff parameter of the customer group at the current time step;

[0188] The target conversion qualification cutoff parameter is determined based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step.

[0189] Optionally, the first determining module 20 is further configured to:

[0190] Calculate the conversion revenue evaluation value of the customer group at the next time step under different estimated conversion qualification cutoff parameters based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step;

[0191] The estimated conversion qualification cutoff parameter corresponding to the maximum conversion benefit evaluation value is selected as the target conversion qualification cutoff parameter.

[0192] Optionally, the first determining module 20 is further configured to:

[0193] The conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification truncation parameter of the customer group at the previous time step are concatenated into a message feature vector at the current time step;

[0194] The message feature vector of the current time step is input into the conversion volume prediction model to predict the conversion volume data of the customer group in the next time step.

[0195] Optionally, the first determining module 20 is further configured to:

[0196] The conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification truncation parameter of the customer group at the previous time step are concatenated into a message feature vector at the current time step;

[0197] Predicting the shape parameters corresponding to the customer group at the next time step by inputting the message feature vector of the customer group at the current time step into the shape parameter prediction model;

[0198] The message feature vector of the customer group at the current time step is input into the scale parameter prediction model to predict the scale parameter corresponding to the customer group at the next time step.

[0199] Optionally, the message recommendation device further includes a training module, wherein the training module is configured to:

[0200] Determine the historical conversion qualification mark value of the customer group within the preset time range through the credit approval model and credit amount model;

[0201] According to the conversion qualification annotation values corresponding to each time step within the preset time range, distribution model fitting is performed respectively to obtain the true value of the conversion qualification distribution parameter corresponding to each time step within the preset time range;

[0202] Obtaining a test set of message feature data for a customer group within a preset time range, wherein the test set of message feature data includes at least one of the following: the conversion volume of the customer group within the preset time range, link conversion feature data of the customer group within the preset time range, and a conversion qualification cutoff parameter of the customer group within the preset time range;

[0203] According to the message feature data test set of the customer group within a preset time range and the true value of the conversion qualification distribution parameter, the conversion qualification prediction model to be trained is iteratively optimized to obtain the conversion qualification prediction model.

[0204] Optionally, the first acquisition module 10 is further configured to:

[0205] Get user sample data of the customer group at the current time step;

[0206] By inputting the sample data of each user into the credit approval model, the credit approval probability of the customer group at the current time step is predicted;

[0207] By inputting the sample data of each user into the credit amount model, the credit amount of the customer group at the current time step is predicted;

[0208] Determine the conversion qualification mark value of the customer group at the current time step based on the credit approval probability and credit amount of the customer group at the current time step;

[0209] A distribution model is fitted for the conversion qualification labeling values of the customer group at the current time step to obtain the conversion qualification distribution parameters of the customer group at the current time step.

[0210] The message recommendation device provided by the present invention utilizes the message recommendation method described in the aforementioned embodiment, resolving the technical issue of poor message recommendation effectiveness in the prior art. Compared to the prior art, the message recommendation device provided by the present invention achieves the same beneficial effects as the message recommendation method described in the aforementioned embodiment. Other technical features of the message recommendation device are the same as those disclosed in the aforementioned embodiment and are not further detailed here.

[0211] Example 6

[0212] Furthermore, an embodiment of the present application further provides a device for determining a conversion qualification truncation parameter, which is applied to a message recommender and includes:

[0213] The second acquisition module 100 is configured to acquire message feature data, wherein the message feature data includes at least one of the following: the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step;

[0214] Prediction module 200, configured to predict the conversion qualification distribution data of the customer group at the next time step by inputting the message feature data of the customer group at the current time step into a conversion qualification prediction model, and to predict the conversion volume data of the customer group at the next time step by inputting the message feature data into a conversion volume prediction model;

[0215] The second determination module 300 is configured to determine a target conversion qualification cutoff parameter based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step.

[0216] Optionally, the second determining module 300 is further configured to:

[0217] Calculate the conversion revenue evaluation value of the customer group at the next time step under different estimated conversion qualification cutoff parameters based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step;

[0218] The estimated conversion qualification cutoff parameter corresponding to the maximum conversion benefit evaluation value is selected as the target conversion qualification cutoff parameter.

[0219] Optionally, the prediction module 200 is further configured to:

[0220] Determine the historical conversion qualification mark value of the customer group within the preset time range through the credit approval model and credit amount model;

[0221] According to the conversion qualification annotation values corresponding to each time step within the preset time range, distribution model fitting is performed respectively to obtain the true value of the conversion qualification distribution parameter corresponding to each time step within the preset time range;

[0222] Obtaining a test set of message feature data for a customer group within a preset time range, wherein the test set of message feature data includes at least one of the following: the conversion volume of the customer group within the preset time range, link conversion feature data of the customer group within the preset time range, and a conversion qualification cutoff parameter of the customer group within the preset time range;

[0223] According to the message feature data test set of the customer group within a preset time range and the true value of the conversion qualification distribution parameter, the conversion qualification prediction model to be trained is iteratively optimized to obtain the conversion qualification prediction model.

[0224] The device for determining conversion qualification truncation parameters provided by the present invention employs the method for determining conversion qualification truncation parameters described in the aforementioned embodiments, resolving the technical issue of poor message recommendation effectiveness in the prior art. Compared to the prior art, the device for determining conversion qualification truncation parameters provided by the present invention achieves the same beneficial effects as the method for determining conversion qualification truncation parameters described in the aforementioned embodiments. Other technical features of the device are the same as those disclosed in the aforementioned embodiments and are not further elaborated upon here.

[0225] Example 7

[0226] Furthermore, an embodiment of the present invention provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the message recommendation method or the conversion qualification truncation parameter determination method in the above-mentioned embodiment.

[0227] Reference below Figure 7, which shows a schematic diagram of the structure of an electronic device suitable for implementing the embodiments of the present disclosure. The electronic devices in the embodiments of the present disclosure may include, but are not limited to, mobile terminals such as Bluetooth headsets, mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (such as in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 7 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0228] like Figure 7 As shown, the electronic device may include a processing device (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) or a program loaded from a storage device into a random access memory (RAM). In the RAM, various programs and arrays required for the operation of the electronic device are also stored. The processing device, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0229] Typically, the following systems can be connected to the I / O interface: input devices including, for example, a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices including, for example, magnetic tape, hard disk, etc.; and communication devices. The communication devices can allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although the figures show electronic devices with various systems, it should be understood that not all of the illustrated systems are required to be implemented or present. More or fewer systems may be implemented or present instead.

[0230] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processing device, the above-mentioned functions defined in the method of the embodiment of the present disclosure are performed.

[0231] The electronic device provided by the present invention employs the message recommendation method or the method for determining conversion qualification truncation parameters in the above-described embodiments, thereby resolving the technical problem of poor message recommendation effectiveness in the prior art. Compared to the prior art, the electronic device provided by the embodiments of the present invention achieves the same beneficial effects as the message recommendation method or the method for determining conversion qualification truncation parameters in the above-described embodiments. Other technical features of the electronic device are the same as those disclosed in the above-described embodiments and are not further described here.

[0232] It should be understood that various parts of the present disclosure can be implemented with hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in an appropriate manner.

[0233] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

[0234] Example 8

[0235] Furthermore, this embodiment provides a computer-readable storage medium having computer-readable program instructions stored thereon, and the computer-readable program instructions are used to execute the message recommendation method or the conversion qualification truncation parameter determination method in the above-mentioned embodiment.

[0236] The computer-readable storage medium provided in the embodiment of the present invention can be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination thereof. More specific examples of computer-readable storage media can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium can be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.

[0237] The computer-readable storage medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0238] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by an electronic device, the electronic device: obtains message feature data, wherein the message feature data at least includes one of the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step; determines the target conversion qualification cutoff parameter of the customer group at the current time step by inputting the message feature data into the conversion qualification cutoff parameter determination model; obtains message recommendation sample data corresponding to the customer group at the current time step, filters the message recommendation sample data of the customer group at the current time step according to the target conversion qualification cutoff parameter, and obtains the return conversion sample data of the current time step; sends the return conversion sample data to the message delivery party, so that the message delivery party can make message recommendations based on the return conversion sample data.

[0239] Alternatively, the computer-readable storage medium carries one or more programs, and when the one or more programs are executed by an electronic device, the electronic device is enabled to: obtain message feature data, wherein the message feature data includes at least one of the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step; predict the conversion qualification distribution data of the customer group at the next time step by inputting the message feature data of the customer group at the current time step into a conversion qualification prediction model, and predict the conversion volume data of the customer group at the next time step by inputting the message feature data into a conversion volume prediction model; and determine the target conversion qualification cutoff parameter based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step.

[0240] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0241] The flow charts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0242] The modules involved in the embodiments described in this disclosure may be implemented in software or hardware, wherein the name of a module does not necessarily limit the unit itself.

[0243] The computer-readable storage medium provided by the present invention stores computer-readable program instructions for executing the aforementioned message recommendation method or conversion qualification truncation parameter determination method, resolving the technical problem of poor message recommendation effectiveness in the prior art. Compared to the prior art, the beneficial effects of the computer-readable storage medium provided by the embodiments of the present invention are similar to those of the message recommendation method or conversion qualification truncation parameter determination method provided by the aforementioned embodiments, and are not further elaborated here.

[0244] Example 9

[0245] Furthermore, the present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned message recommendation method or conversion qualification truncation parameter determination method.

[0246] The computer program product provided in this application solves the technical problem of poor message recommendation effectiveness in the prior art. Compared with the prior art, the beneficial effects of the computer program product provided by the embodiments of the present invention are the same as those of the message recommendation method or the method for determining the conversion qualification cutoff parameter provided in the above embodiments, and are not further described here.

[0247] The above are only preferred embodiments of the present application and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent processing scope of the present application.

Claims

1. A message recommendation method, characterized in that: The message recommendation method is applied to the message recommender, and includes the following steps: Acquiring message feature data, wherein the message feature data includes at least one of the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step; Determining the target conversion qualification cutoff parameter of the customer group at the current time step by inputting the message feature data into a conversion qualification cutoff parameter determination model; Obtaining message recommendation sample data corresponding to the customer group at the current time step, filtering the message recommendation sample data for the customer group at the current time step according to the target conversion qualification cutoff parameter, and obtaining the return conversion sample data for the current time step; Sending the returned conversion sample data to a message delivery party, so that the message delivery party can make message recommendations based on the returned conversion sample data; The conversion qualification truncation parameter determination model includes a conversion volume prediction model and a conversion qualification prediction model. The step of inputting the message feature data into the conversion qualification truncation parameter determination model to determine the target conversion qualification truncation parameter for the customer group at the current time step includes: By inputting the message feature data into the conversion qualification prediction model, the conversion qualification distribution data of the customer group in the next time step is predicted; The conversion volume data of the customer group at the next time step is predicted by inputting the message feature data into the conversion volume prediction model, wherein the conversion volume data is used to represent the corresponding relationship between the conversion volume of the customer group at the next time step and the conversion qualification cutoff parameter of the customer group at the current time step; Determining the target conversion qualification cutoff parameter based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step; Before the step of inputting the message feature data of the customer group at the current time step into the conversion qualification prediction model to predict the conversion qualification distribution data of the customer group at the next time step, the message recommendation method further includes: Determine the historical conversion qualification mark value of the customer group within the preset time range through the credit approval model and credit amount model; According to the conversion qualification annotation values corresponding to each time step within the preset time range, distribution model fitting is performed respectively to obtain the true value of the conversion qualification distribution parameter corresponding to each time step within the preset time range; Obtaining a test set of message feature data for a customer group within a preset time range, wherein the test set of message feature data includes at least one of the following: the conversion volume of the customer group within the preset time range, link conversion feature data of the customer group within the preset time range, and a conversion qualification cutoff parameter of the customer group within the preset time range; According to the message feature data test set of the customer group within a preset time range and the true value of the conversion qualification distribution parameter, the conversion qualification prediction model to be trained is iteratively optimized to obtain the conversion qualification prediction model.

2. The message recommendation method according to claim 1, wherein: The step of determining the target conversion qualification cutoff parameter based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step includes: Calculate the conversion revenue evaluation value of the customer group at the next time step under different estimated conversion qualification cutoff parameters based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step; The estimated conversion qualification cutoff parameter corresponding to the maximum conversion benefit evaluation value is selected as the target conversion qualification cutoff parameter.

3. The message recommendation method according to claim 1, wherein: The step of predicting the conversion volume data of the customer group at the next time step by inputting the message feature data into the conversion volume prediction model includes: The conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification truncation parameter of the customer group at the previous time step are concatenated into a message feature vector at the current time step; The message feature vector of the current time step is input into the conversion volume prediction model to predict the conversion volume data of the customer group in the next time step.

4. The message recommendation method according to claim 1, wherein: The conversion qualification distribution data includes a shape parameter and a scale parameter, the conversion qualification prediction model includes a shape parameter prediction model and a scale parameter prediction model, and the step of predicting the conversion qualification distribution data of the customer group at the next time step by inputting the message feature data of the customer group at the current time step into the conversion qualification prediction model includes: The conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification truncation parameter of the customer group at the previous time step are concatenated into a message feature vector of the customer group at the current time step; Predicting the shape parameters corresponding to the customer group at the next time step by inputting the message feature vector of the customer group at the current time step into the shape parameter prediction model; The message feature vector of the customer group at the current time step is input into the scale parameter prediction model to predict the scale parameter corresponding to the customer group at the next time step.

5. The message recommendation method according to claim 1, wherein: The step of obtaining the conversion qualification distribution parameters of the customer group at the current time step includes: Get user sample data of the customer group at the current time step; By inputting the sample data of each user into the credit approval model, the credit approval probability of the customer group at the current time step is predicted; By inputting the sample data of each user into the credit amount model, the credit amount of the customer group at the current time step is predicted; Determine the conversion qualification mark value of the customer group at the current time step based on the credit approval probability and credit amount of the customer group at the current time step; A distribution model is fitted for the conversion qualification labeling values of the customer group at the current time step to obtain the conversion qualification distribution parameters of the customer group at the current time step.

6. A method for determining conversion qualification cutoff parameters, characterized in that: The method for determining conversion qualification truncation parameters is applied to the message recommender and includes the following steps: Acquiring message feature data, wherein the message feature data includes at least one of the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step; By inputting the message feature data into a conversion qualification prediction model, the conversion qualification distribution data of the customer group in the next time step is predicted, and by inputting the message feature data into a conversion quantity prediction model, the conversion quantity data of the customer group in the next time step is predicted; Determining a target conversion qualification cutoff parameter based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step; Before the step of predicting the conversion qualification distribution data of the customer group at the next time step by inputting the message feature data of the customer group at the current time step into the conversion qualification prediction model, the conversion qualification cutoff parameter determination method further includes: Determine the historical conversion qualification mark value of the customer group within the preset time range through the credit approval model and credit amount model; According to the conversion qualification annotation values corresponding to each time step within the preset time range, distribution model fitting is performed respectively to obtain the true value of the conversion qualification distribution parameter corresponding to each time step within the preset time range; Obtaining a test set of message feature data for a customer group within a preset time range, wherein the test set of message feature data includes at least one of the following: the conversion volume of the customer group within the preset time range, link conversion feature data of the customer group within the preset time range, and a conversion qualification cutoff parameter of the customer group within the preset time range; According to the message feature data test set of the customer group within a preset time range and the true value of the conversion qualification distribution parameter, the conversion qualification prediction model to be trained is iteratively optimized to obtain the conversion qualification prediction model.

7. The method for determining conversion qualification cutoff parameters according to claim 6, wherein: The step of determining the target conversion qualification cutoff parameter based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step includes: Calculate the conversion revenue evaluation value of the customer group at the next time step under different estimated conversion qualification cutoff parameters based on the conversion volume data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step; The estimated conversion qualification cutoff parameter corresponding to the maximum conversion benefit evaluation value is selected as the target conversion qualification cutoff parameter.

8. A message recommendation device, the message recommendation device being applied to a message recommender, comprising: A first acquisition module is configured to acquire message feature data, wherein the message feature data includes at least one of the following: the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step; A training module is used to determine the historical conversion qualification labeling values of a customer group within a preset time range through a credit approval model and a credit amount model; perform distribution model fitting according to the conversion qualification labeling values corresponding to each time step within the preset time range, and obtain the true value of the conversion qualification distribution parameter corresponding to each time step within the preset time range; obtain a message feature data test set of the customer group within the preset time range, wherein the message feature data test set includes at least one of the conversion volume of the customer group within the preset time range, the link conversion feature data of the customer group within the preset time range, and the conversion qualification truncation parameter of the customer group within the preset time range; iteratively optimize the conversion qualification prediction model to be trained according to the message feature data test set of the customer group within the preset time range and the true value of the conversion qualification distribution parameter to obtain a conversion qualification prediction model; A first determination module is configured to determine the target conversion qualification truncation parameter of the customer group at the current time step by inputting the message feature data into a conversion qualification truncation parameter determination model, wherein the conversion qualification truncation parameter determination model includes a conversion quantity prediction model and a conversion qualification prediction model. The first determination module is further configured to predict the conversion qualification distribution data of the customer group at the next time step by inputting the message feature data into the conversion qualification prediction model; predict the conversion quantity data of the customer group at the next time step by inputting the message feature data into the conversion quantity prediction model, wherein the conversion quantity data is used to characterize the correspondence between the conversion quantity of the customer group at the next time step and the conversion qualification truncation parameter of the customer group at the current time step; and determine the target conversion qualification truncation parameter based on the conversion quantity data of the customer group at the next time step and the conversion qualification distribution data of the customer group at the next time step; A filtering module is used to obtain message recommendation sample data corresponding to the customer group at the current time step, and filter the message recommendation sample data of the customer group at the current time step according to the target conversion qualification cutoff parameter to obtain the return conversion sample data of the current time step; The sending module is used to send the return conversion sample data to the message delivery party, so that the message delivery party can make message recommendations based on the return conversion sample data.

9. A device for determining a conversion qualification truncation parameter, the device being applied to a message recommender, comprising: A second acquisition module is configured to acquire message feature data, wherein the message feature data includes at least one of the conversion volume of the customer group at the current time step, the conversion qualification distribution parameter of the customer group at the current time step, the link conversion feature data of the customer group at the current time step, and the conversion qualification cutoff parameter of the customer group at the previous time step; A prediction module is used to predict the conversion qualification distribution data of the customer group at the next time step by inputting the message feature data into a conversion qualification prediction model, and to predict the conversion quantity data of the customer group at the next time step by inputting the message feature data into a conversion quantity prediction model; the prediction module is also used to determine the historical conversion qualification labeling value of the customer group within a preset time range through a credit approval model and a credit amount model; according to the conversion qualification labeling value corresponding to each time step within the preset time range, a distribution model is fitted respectively to obtain the true value of the conversion qualification distribution parameter corresponding to each time step within the preset time range; a message feature data test set of the customer group within the preset time range is obtained, wherein the message feature data test set includes at least one of the conversion quantity of the customer group within the preset time range, the link conversion feature data of the customer group within the preset time range, and the conversion qualification truncation parameter of the customer group within the preset time range; according to the message feature data test set of the customer group within the preset time range and the true value of the conversion qualification distribution parameter, the conversion qualification prediction model to be trained is iteratively optimized to obtain a conversion qualification prediction model; The second determination module is used to determine the target conversion qualification cutoff parameter according to the conversion volume data of the customer group in the next time step and the conversion qualification distribution data of the customer group in the next time step.

10. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the message recommendation method described in any one of claims 1 to 5, or the steps of the conversion qualification truncation parameter determination method described in any one of claims 6 to 7.

11. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, on which is stored a program for implementing the message recommendation method. The program for implementing the message recommendation method is executed by a processor to implement the steps of the message recommendation method as described in any one of claims 1 to 5, or the steps of the conversion qualification truncation parameter determination method as described in any one of claims 6 to 7.

12. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the message recommendation method according to any one of claims 1 to 5 or the steps of the conversion qualification truncation parameter determination method according to any one of claims 6 to 7 are implemented.

Images