Advertisement information pushing method and device, electronic equipment and storage medium

Abstract

The invention provides an advertisement information pushing method and device, electronic equipment and a storage medium, and relates to the technical field of credit loan, and the method comprises the steps: obtaining object credit loan data of a plurality of candidate objects in a target time period; inputting the object credit data into a first value prediction model to obtain a first prediction value of each candidate object, the first prediction value being used for representing an object value of the candidate object at the target time point; inputting the target advertisement information and the object credit data into a second value prediction model to obtain a second prediction value of each candidate object, the second prediction value being used for representing an object value of the candidate object at the target time point after the target advertisement information is pushed to the candidate object; and based on the first predicted value and the second predicted value of each candidate object, determining a target object from the plurality of candidate objects, and pushing target advertisement information to the target object. According to the method, accurate putting of the target advertisement information is realized.

Description

Technical Field

[0001] This application relates to the field of credit technologies, and in particular, to an advertisement information pushing method, apparatus, electronic device, and storage medium. Background Art

[0002] In the credit field, how to formulate a scientific and effective advertisement pushing strategy is a key issue, involving various considerations such as user experience, credit scale, and institutional profitability. The traditional coarse-grained advertisement pushing strategies such as manually formulating rules can no longer meet the requirements, and personalized advertisement pushing needs to consider a series of factors such as customer qualifications, customer needs, risk performance, and borrowing history. Therefore, how to mine high-value users from massive data and formulate corresponding advertisement pushing strategies is an urgent problem to be solved. Summary of the Invention

[0003] This application provides an advertisement information pushing method, apparatus, electronic device, and storage medium. The technical solutions are as follows:

[0004] According to one aspect of this application, an advertisement information pushing method is provided. The method includes:

[0005] Obtain the object credit data of multiple candidate objects within a target time period;

[0006] Input the object credit data into a first value prediction model to obtain a first prediction value for each candidate object, where the first prediction value is used to represent the object value of the candidate object at a target time point;

[0007] Input the target advertisement information and the object credit data into a second value prediction model to obtain a second prediction value for each candidate object, where the second prediction value is used to represent the object value of the candidate object at the target time point after pushing the target advertisement information to the candidate object;

[0008] Based on the first prediction value and the second prediction value of each candidate object, determine a target object from multiple candidate objects, and push the target advertisement information to the target object.

[0009] According to another aspect of this application, an advertisement information pushing apparatus is provided. The apparatus includes:

[0010] A first obtaining module, configured to obtain the object credit data of multiple candidate objects within a target time period;

[0011] A first prediction module, configured to input the object credit data into a first value prediction model to obtain a first prediction value for each candidate object, where the first prediction value is used to represent the object value of the candidate object at a target time point;

[0012] A second prediction module, configured to input the target advertisement information and the object credit data into a second value prediction model to obtain a second prediction value for each candidate object, where the second prediction value is used to characterize the object value of the candidate object at the target time point after pushing the target advertisement information to the candidate object;

[0013] An information pushing module, configured to determine a target object from multiple candidate objects based on the first prediction value and the second prediction value of each candidate object, and push the target advertisement information to the target object.

[0014] According to one aspect of the present application, there is provided an electronic device, including: a processor and a memory storing a program, where the program includes instructions that, when executed by the processor, cause the processor to execute the advertisement information pushing method as described above.

[0015] According to another aspect of the present application, there is provided a non-transitory computer-readable storage medium storing computer instructions, where the computer instructions are used to cause the computer to execute the advertisement information pushing method as described above.

[0016] According to another aspect of the present application, there is provided a computer program product, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above advertisement information pushing method.

[0017] The beneficial effects brought by the technical solution provided by the embodiments of the present application at least include:

[0018] The embodiments of the present application provide an advertisement information pushing method: by respectively predicting the LTV of a user without pushing advertisement information (the first prediction value), and the LTV of the user when pushing the target advertisement information (the second prediction value), to find out high-increment-value users (target objects) by comparing the two LTVs, and push the corresponding target advertisement information to the high-increment-value users. It can accurately mine high-increment-value users based on existing credit data, and push target advertisement information to this part of users, realizing the precise delivery of target advertisement information, thereby reducing the delivery cost of target advertisement information. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In the following description of exemplary embodiments with reference to the accompanying drawings, more details, features, and advantages of the present application are disclosed. In the drawings:

[0020] Figure 1A schematic diagram of an example system in which various methods described herein can be implemented according to an exemplary embodiment of the present application;

[0021] Figure 2 A flowchart of an advertising information push method according to an exemplary embodiment of the present application;

[0022] Figure 3 A flowchart of another advertising information push method according to an exemplary embodiment of the present application;

[0023] Figure 4 A schematic diagram of the architecture of a value prediction model provided by an exemplary embodiment of the present application;

[0024] Figure 5 A flowchart of a first value prediction model training method according to an exemplary embodiment of the present application;

[0025] Figure 6 A flowchart of a second value prediction model training method according to an exemplary embodiment of the present application;

[0026] Figure 7 A schematic diagram of the structure of an advertising information push device provided by an embodiment of the present application;

[0027] Figure 8 A block diagram of an exemplary electronic device that can be used to implement an embodiment of the present application is shown. Detailed implementation manners

[0028] Embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Instead, these embodiments are provided to more thoroughly and completely understand the present application. It should be understood that the drawings and embodiments of the present application are only for exemplary purposes and are not used to limit the protection scope of the present application.

[0029] It should be understood that the steps recited in the method embodiments of the present application can be executed in a different order and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present application is not limited in this regard.

[0030] As used herein, the term "including" and its variations are open-ended, i.e., "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Relevant definitions of other terms will be given in the following description. It should be noted that the concepts such as "first", "second", etc. mentioned in this application are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units. It should be noted that the modifications of "one" and "a plurality of" mentioned in this application are illustrative rather than restrictive. Those skilled in the art should understand that unless clearly specified otherwise in the context, it should be understood as "one or more". The names of the messages or information exchanged between multiple devices in the embodiments of this application are only for illustrative purposes and are not used to limit the scope of these messages or information.

[0031] The solution of the present invention will be described below with reference to the accompanying drawings. The technical solution provided by the embodiments of the present invention will be described in detail through specific embodiments and their application scenarios.

[0032] Currently, intelligent advertisement push based on deep neural networks has been widely applied to the credit business. When formulating different advertisement information push strategies, financial institutions mostly target a single goal. For example, by distributing coupons to improve customer retention, or by increasing the credit limit to improve profitability, etc. However, the goals of different advertisement information push activities may conflict, and even have a negative impact on the long-term benefits of the platform. Therefore, it is particularly important to integrate various advertisement information push means to form a personalized intelligent advertisement placement system with the goal of maximizing long-term value. The industry generally uses LTV (Life Time Value) to represent the long-term life value of users, which represents the total expected value generated by customers during their entire relationship with the enterprise, and usually factors such as usage frequency, retention duration, and order value need to be considered. And for user value prediction, there are currently two major challenges: (1) The proportion of zero-value labels is extremely high, and most users will not have borrowing behaviors within a certain period of time; (2) In the scenarios where borrowing behaviors actually occur, the LTV distribution is extremely skewed, and the borrowing amounts of a small number of users vary greatly, that is, the LTV distribution shows the characteristics of an extremely long-tailed distribution. Currently, the main methods for predicting extremely long-tailed distributions based on deep neural networks are as follows:

[0033] (1) Sampling methods based on class balance: In the case of unbalanced samples, methods such as oversampling and undersampling are used. By increasing the number of samples of the minority class or reducing the number of samples of the majority class to balance the data set, thereby improving the training effect of the model.

[0034] (2) Prediction method based on quantile regression: Quantile regression uses the weighted least absolute deviation sum (WLAD) method for estimation, which is usually not affected by outliers and the results are more robust.

[0035] (3) Two-stage learning method: This type of method generally divides the LTV prediction into two parts. First, a binary classification model is used to predict the customer category (such as purchase propensity, etc.). Second, a regression model is used to predict the LTV of customers who have made a successful purchase or whose total life cycle value is greater than a certain threshold.

[0036] The above three methods have the following defects in practical applications: (1) Although the idea of random undersampling or oversampling is relatively easy to apply in practice. However, both downsampling and upsampling modify the sample distribution, losing the sample randomness, and thus weakening the performance of the model. (2) Quantile regression cannot provide specific LTV values; (3) The two-stage learning method requires maintaining two models simultaneously, which may lead to the gradual accumulation of errors, thereby reducing the overall prediction accuracy.

[0037] In addition, for the LTV values predicted by the above methods, corresponding advertising information push strategies (or marketing strategies) are generally formulated according to the LTV values of users, and thus more advertising resources (or marketing resources) are invested in high-value users (users with high LTV values). However, high-value users are not necessarily equivalent to high-increment-value users. There is a part of high-value users who, even if advertising resources are invested, have basically no impact on their borrowing behavior and can be regarded as users insensitive to advertising (or marketing). Therefore, the existing advertising information push strategies are ineffective in tilting resources towards this part of users.

[0038] In response to the problems of related technologies, the present application provides a new way to predict LTV values and formulate advertising information push strategies (or marketing strategies). Figure 1 The schematic diagram of an example system in which the various methods described herein can be implemented according to an exemplary embodiment of the present application is shown. As Figure 1 shown, the system includes a first device 110 and a second device 120.

[0039] The first device 110 is a training device for training the first value prediction model and the second value prediction model. After the first value prediction model and the second value prediction model are trained, the first device 110 can send the trained first value prediction model and the second value prediction model to the second device 120 for deploying the first value prediction model and the second value prediction model in the second device 120. The second device 120 is a device for predicting LTV values using the first value prediction model and the second value prediction model.

[0040] In the model training stage: The first value prediction model is trained using the first sample credit data and the first sample distribution observations, and the second value prediction model is trained using the second sample credit data, the second sample distribution observations, and the historical advertising information (historical marketing strategies).

[0041] In the model application stage: The object credit data of the candidate objects is input into the first value prediction model to obtain a first predicted value. The object credit data and the target advertising information (target marketing strategies) are input into the second value prediction model to obtain a second predicted value. Based on the first predicted value and the second predicted value, target objects are screened from the candidate objects to push the target advertising information to them (or apply the target marketing strategies to them).

[0042] Optionally, the above-mentioned first device 110 and second device 120 can be computer devices with machine learning capabilities. For example, the computer device can be a terminal or a server.

[0043] Optionally, the above-mentioned first device 110 and second device 120 can be the same computer device, or the first device 110 and the second device 120 can also be different computer devices. Moreover, when the first device 110 and the second device 120 are different devices, the first device 110 and the second device 120 can be the same type of device. For example, the first device 110 and the second device 120 can both be servers; or, the first device 110 and the second device 120 can also be different types of devices. The above-mentioned server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The above-mentioned terminal can be a smart phone, in-vehicle terminal, smart TV, wearable device, tablet computer, laptop computer, desktop computer, smart speaker, smart watch, etc., but is not limited thereto. The terminal and the server can be directly or indirectly connected through wired or wireless communication methods, and this application does not make any restrictions here.

[0044] Please refer to Figure 2 , which shows a flowchart of an advertising information pushing method according to an exemplary embodiment of the present application. Taking this method applied to Figure 1 the second device 120 shown as an example for illustration. As Figure 2 shown, the method includes:

[0045] Step 201, obtain the object credit data of multiple candidate objects within a target time period.

[0046] When formulating an advertising information push strategy, it is more inclined to invest more advertising resources in high-increment-value users. High-increment-value users are those who can generate incremental value through advertisements. For example, some users whose retention rate can be improved by issuing coupons, or some users whose order value can be increased by raising the credit limit, etc. In order to accurately identify high-increment-value users from existing credit data, the embodiments of the present application provide a feasible idea for formulating an advertising information push strategy: First, based on the existing credit data, predict the LTV (life cycle value) of users without pushing advertising information, and then based on the existing credit data and the advertising information to be pushed, predict the LTV of users when advertising information is pushed. Furthermore, find high-increment-value users from the two LTVs to formulate the corresponding advertising information push strategy.

[0047] Based on the above idea of formulating an advertising information push strategy, in order to predict the LTV of users, it is first necessary to obtain the object credit data of multiple candidate objects in the target time period for analyzing and constructing the LTV distribution, and then extract the LTV of each candidate object. Among them, in the credit field, the candidate object is a candidate user, the target object is a target user, the object credit data is the user credit data of the candidate user in the target time period, and the target time period is a defined historical time period. Exemplarily, the object credit data may include the income and expenditure data of the user, the loan amount data of the user, the debt data of the user, the credit investigation data of the user, and so on. It should be noted that the object credit data needs to be obtained after the user's authorization.

[0048] Step 202: Input the object credit data into the first value prediction model to obtain the first prediction value of each candidate object, and the first prediction value is used to represent the object value of the candidate object at the target time point.

[0049] Different from the two-stage LTV prediction method in the related art, the embodiments of the present application construct a first value prediction model, which can directly predict the LTV of each user based on the existing credit data. In one possible implementation, first input the object credit data into the first value prediction model, and the first value prediction model predicts the LTV distribution composed of multiple candidate objects, and then calculate the first prediction value of each candidate object from it (the first prediction value is the LTV of each candidate object), and this first prediction value is used to represent the object value of the candidate object at the target time point without pushing any advertising information. Among them, the target time point is a future time node different from any time point in the target time period. Exemplarily, if the target time period is from January 2001 to December 2002, the target time point can be March 2002.

[0050] Step 203: Input the target advertisement information and the object credit data into the second value prediction model to obtain the second prediction value for each candidate object. The second prediction value is used to represent the object value of the candidate object at the target time point after pushing the target advertisement information to the candidate object.

[0051] Different from directly formulating an advertisement information pushing strategy based on the predicted LTV in the related art, in the embodiment of the present application, a second value model is further constructed. Different from the first value prediction model, the second value prediction model introduces a variable of the target advertisement information and is used to predict the possible LTV of each user when the advertisement information is pushed based on the existing credit data and the advertisement information to be pushed. In a corresponding possible implementation manner, the target advertisement information and the object credit data are jointly input into the second value prediction model, and the second value prediction model predicts the LTV distribution composed of multiple candidate objects, and then calculates the second prediction value for each candidate object. The second prediction value is used to represent the object value of the candidate object at the target time point after pushing the target advertisement information to the candidate object.

[0052] Step 204: Based on the first prediction value and the second prediction value of each candidate object, determine the target object from multiple candidate objects, and push the target advertisement information to the target object.

[0053] Further, by analyzing the first prediction value without pushing the target advertisement information and the second prediction value with the target advertisement information pushed, high-increment-value users, that is, target objects, are determined from multiple candidate objects for pushing the target advertisement information to the target objects. Specifically, if the second prediction value of a candidate object is higher than the first prediction value, it indicates that pushing the target advertisement information can significantly improve the user value, and the candidate object can be used as a high-increment-value user to push the advertisement information to it; on the contrary, if the second prediction value of a candidate object is less than or equal to the first prediction value, it indicates that pushing the target advertisement information does not play a positive role and cannot bring advertising investment returns, and then it is not necessary to push the advertisement information to this part of users, so as to achieve the precise delivery of the target advertisement information and thus reduce the delivery cost of the target advertisement information.

[0054] In summary, the embodiment of the present application provides an advertisement information pushing method: by respectively predicting the LTV (the first prediction value) of a user without pushing advertisement information and the LTV (the second prediction value) of the user with the target advertisement information pushed, high-increment-value users (target objects) are found by comparing the two LTVs, and the corresponding target advertisement information is pushed to the high-increment-value users. High-increment-value users can be accurately mined based on the existing credit data, and the target advertisement information is pushed to this part of users, realizing the precise delivery of the target advertisement information, thereby reducing the delivery cost of the target advertisement information.

[0055] When performing LTV prediction, the first value prediction model and the second value prediction model will first predict the transaction probability, the mean and standard deviation of the LTV distribution of the candidate objects, and then determine the LTV of each candidate object based on these three parameters.

[0056] Please refer to Figure 3 , which shows a flowchart of another advertising information pushing method according to an exemplary embodiment of the present application. This method is described by taking its application to a computer device as an example. As Figure 3 shown, the method includes:

[0057] Step 301, obtain the object credit data of multiple candidate objects within a target time period.

[0058] The implementation manner of step 301 can refer to step 201, and will not be elaborated herein in this embodiment.

[0059] Step 302, input the object credit data into the first value prediction model to obtain the first prediction probability of each candidate object output by the first value prediction model, and the first predicted mean and the first predicted standard deviation of the first object value distribution.

[0060] Please refer to Figure 4 , which is a schematic diagram of the architecture of the value prediction model provided by an exemplary embodiment of the present application. The value prediction model consists of multiple fully connected layers. The first layer 401 is used to extract the data features (or object features) of the input object credit data. The second layer 402 is a hidden layer. The third layer 403 is an activation layer, which includes three pre-activation units. These three pre-activation units respectively use the Sigmoid, Identity, and Softplus activation functions to output the three parameters of the transaction probability, the mean of the LTV distribution, and the standard deviation. The last layer 404 is used to determine the LTV of each candidate object according to the three parameters output by the third layer. It can be seen that the middle layer of this value prediction model is essentially a shared representation of two related tasks, the prediction of the transaction probability (the transaction probability is the first prediction probability) and the actual user value (LTV) after borrowing.

[0061] Among them, the model architectures of the first value prediction model and the second value prediction model are the same, which is the Figure 4 architecture of the value prediction model shown. Based on this architecture schematic diagram, after inputting the object credit data into the first value prediction model, first, the first prediction probability of each candidate object output by the first value prediction model (that is, the probability of predicting whether there will be a transaction behavior of the candidate object at the target time point) can be obtained, as well as the first predicted mean (μ) and the first predicted standard deviation (σ) of the first object value distribution. The first object value distribution is the distribution of predicting the LTV of multiple candidate objects at the target time point.

[0062] Step 303: Determine a first prediction value for each candidate object based on the first prediction mean, the first prediction standard deviation, and the first prediction probability of each candidate object.

[0063] Furthermore, after the first predicted mean (μ) and the first predicted standard deviation (σ) have been predicted, the first predicted value of each candidate object can be determined based on the first object value distribution constructed based on the first predicted mean, the first predicted standard deviation, and the first predicted probability (p) of each candidate object, wherein the first predicted probability is the transaction probability of the candidate object at the target time point, and the first predicted value is the LTV (or LTV value) of the candidate object at the target time point.

[0064] Step 304, input the object credit data and target advertisement information into the second value prediction model, and obtain the second predicted probability of each candidate object output by the second value prediction model, as well as the second predicted mean and second predicted standard deviation of the second object value distribution.

[0065] Similarly, when predicting the user value of the user after the targeted advertising information is pushed, after the object credit data and the target advertising information are input into the second value prediction model, the second predicted probability of each candidate object output by the second value prediction model (that is, the probability of predicting whether the candidate object will have a transaction at the target time point after the targeted advertising information is released) and the second predicted mean (μ) and second predicted standard deviation (σ) of the second object value distribution are obtained. The second object value distribution is the distribution of LTV of multiple candidate objects at the target time point predicted after the targeted advertising information is pushed.

[0066] Step 305: Determine a second predicted value for each candidate object based on the second predicted mean, the second predicted standard deviation, and the second predicted probability of each candidate object.

[0067] Furthermore, after the second predicted mean (μ) and the second predicted standard deviation (σ) have been predicted, the second predicted value of each candidate object can be determined based on the second object value distribution constructed based on the second predicted mean and the second predicted standard deviation, and the second predicted probability of each candidate object. The second predicted probability is the transaction probability of the candidate object at the target time point after the target advertising information is pushed, and the second predicted value is the LTV (or LTV value) of the candidate object at the target time point after the target advertising information is pushed.

[0068] Step 306 , determining a candidate incremental value for each candidate object based on the first predicted value and the second predicted value for each candidate object.

[0069] Among them, incremental value users are users whose value increases after pushing advertising information. In order to accurately screen out high-increment value users from candidate objects, the candidate incremental value of each candidate object after pushing the target advertising information can be determined based on the first prediction value and the second prediction value of each candidate object, and then the target object can be screened based on the candidate incremental value. Exemplarily, the candidate incremental value = the second prediction value - the first prediction value, that is, the difference between the user value after pushing the target advertising information and the user value without pushing the target advertising information.

[0070] Step 307: Determine the candidate object with a candidate incremental value greater than the preset threshold as the target object.

[0071] The larger the candidate incremental value, the greater the user value that may be brought by pushing the target advertising information. When screening high-increment value users, a preset threshold is also set, and the candidate object with a candidate incremental value greater than the preset threshold is determined as the target object. Among them, the preset threshold is a value greater than or equal to 0. When the preset threshold is 0, the target object is the candidate object with a candidate incremental value greater than 0, that is, the candidate object with the second prediction value greater than the first prediction value.

[0072] Step 308: Obtain the candidate incremental value of each target object.

[0073] After screening out the target objects with incremental value, the target advertising information can be directly delivered or applied to this part of the target objects. For example, strategies such as issuing coupons and increasing the limit can be adopted.

[0074] Optionally, among the target objects with incremental value, there are also users with high incremental value and low incremental value. If more advertising resources are tilted towards users with high incremental value, greater value benefits may be brought. Further, the candidate incremental value of each target object can be obtained, and a more refined advertising information push strategy can be formulated based on the level of the candidate incremental value.

[0075] Step 309: Sort the target objects according to the candidate incremental value to obtain a sorting result.

[0076] In a possible implementation manner, the target objects can be sorted according to the level of the candidate incremental value to obtain a sorting result. Among them, when sorting, the candidate incremental value can be sorted from high to low, or it can also be sorted from low to high according to the candidate incremental value.

[0077] Step 310: Push the target advertising information to the target objects based on the sorting result.

[0078] Obviously, allocating more advertising resources to users with high candidate incremental value will yield more value returns. While allocating relatively fewer advertising resources to users with low candidate incremental value can reduce advertising resource expenditure while obtaining value returns. In one possible implementation, target advertising information with different resource amounts can be applied to different target objects based on the order of the sorting results. Specifically, the higher the candidate incremental value, the more resources are applied; the lower the candidate incremental value, the fewer resources are applied. That is, the amount of resources applied is positively correlated with the candidate incremental value.

[0079] Exemplarily, if there are Object 1, Object 2, Object 3, Object 4, and Object 5, and the sorting result in ascending order of candidate incremental value is: Object 2 < Object 4 < Object 3 < Object 5 < Object 1, then the resource amounts of the target advertising information delivered to each object are: Object 2 < Object 4 < Object 3 < Object 5 < Object 1, that is, Object 1 is delivered the most resources, while Object 2 is delivered the fewest resources. Taking the target advertising information as issuing coupons as an example, the relationship of the number of coupons issued to each object is: Object 2 < Object 4 < Object 3 < Object 5 < Object 1.

[0080] Optionally, different incremental intervals can also be divided according to the sorting result, and target advertising information with different resource amounts is delivered to target objects in different incremental intervals.

[0081] In this embodiment, by predicting the LTV distribution and conversion probability of multiple candidate objects to obtain the LTV of each candidate object, the prediction of LTV is realized; moreover, by comparing the LTV of each candidate object with or without applying the target advertising information, the candidate incremental value is obtained, and based on the candidate incremental value, the target objects are screened, and target advertising information is delivered to the target objects based on the high or low candidate incremental value, realizing the delivery of personalized target advertising information.

[0082] In order to enable the above first value prediction model and second value prediction model to have the LTV prediction function, it is necessary to pre-construct the model and conduct targeted training. The following embodiments mainly exemplarily illustrate the training process of the value prediction model.

[0083] Please refer to Figure 5 , which shows a flowchart of a method for training a first value prediction model according to an exemplary embodiment of the present application. This method is described by taking its application to a computer device as an example. As Figure 5 shown, this method includes:

[0084] Step 501, obtain the first sample credit data and the first sample distribution observation values of multiple first sample objects in the historical time period. The first sample credit data is the credit data without pushing historical advertising information.

[0085] Since the first value prediction model is used to predict the LTV without any historical advertising information, the sample data used in training the first value prediction model are: the first sample credit data and the first sample distribution observations of multiple first sample objects in the historical time period, the first sample credit data is the credit data without the historical advertising information, and the first sample distribution observations are the labeled LTV distribution observations generated based on the first sample credit data, which may specifically include the labeled mean, labeled standard deviation and labeled LTV of each first sample object.

[0086] Step 502: training a first value prediction model based on first sample credit data and first sample distribution observations.

[0087] Furthermore, in each round of training, the first sample credit data is input into the initial first value prediction model to obtain three prediction parameters output by the first value prediction model, and then the first value prediction model is trained based on the three prediction parameters and the first sample distribution observations.

[0088] Since the first value prediction model involves two task predictions, the corresponding loss function also includes two parts, one is the classification loss (i.e., the loss of predicting the transaction probability of the candidate object), and the other is the LTV distribution prediction loss. Correspondingly, in an exemplary example, step 502 may also include steps 502A to 502C.

[0089] Step 502A, input the first sample credit data into the first value prediction model to obtain the first sample probability of each first sample object output by the first prediction model, as well as the first sample mean and first sample standard deviation of the first sample value distribution.

[0090] After the first sample credit data is input into the first value prediction model, the first sample probability of each first sample object output by the first value prediction model (i.e., the transaction probability of the first sample object) and the first sample mean and first sample standard deviation of the first sample value distribution can be obtained.

[0091] Step 502B, determining a first prediction loss based on the first sample probability, the first sample mean, the first sample standard deviation, and the first sample distribution observation value.

[0092] Furthermore, based on the first sample probability and the standard sample probability, the classification loss is determined, and based on the first sample mean, the first sample standard deviation and the first sample distribution observation value, the first distribution prediction loss of the first sample value distribution is determined, and the sum of the classification loss and the first distribution prediction loss is determined as the first prediction loss of the first value prediction model to train the first value prediction model.

[0093] Exemplarily, the complete loss function can be as shown in formula (1).

[0094]

[0095] Wherein, L 1 represents the first prediction loss, represents the classification loss, represents the first distribution prediction loss, p 1 represents the first sample probability, x 1 represents the first sample distribution observation value, μ 1 represents the first sample mean, σ 1 represents the first sample standard deviation,

[0096] In addition, considering that although the proportion of ultra-high-value samples (ultra-high-value users) is extremely low, the ultra-high-value users are very important for business contributions. Therefore, the unstable prediction of this part of the samples may cause serious interference to the online effect. To solve this problem, when calculating the loss function during model training, the losses of this part of the first sample objects will be weighted.

[0097] Exemplarily, step 502B may further include steps 502B1 to 502B4.

[0098] Step 502B1, obtaining the first sample observation value corresponding to each first sample object from the first sample distribution observation value.

[0099] Step 502B2, sorting the multiple first sample objects based on the first sample observation value to obtain a first value sorting result.

[0100] Step 502B3, determining the first sample weight corresponding to each first sample object based on the first value sorting result.

[0101] Step 502B4, determining the first prediction loss based on the first sample probability, the first sample mean, the first sample standard deviation, the first sample distribution observation value, and the first sample weight.

[0102] In order to identify high-value users from the first sample objects, it is first necessary to determine the labeled LTV (i.e., the first sample observed value) of each first sample object from the first sample distribution observations, and sort the multiple first sample objects according to the level of the labeled LTV to obtain the first value ranking result; then, based on the front and back of the first value ranking result, different first sample weights are set for different first sample objects. And the higher the first sample observed value, the greater the first sample weight, and the lower the first sample observed value, the smaller the first sample weight, that is, the first sample weight is positively correlated with the first sample observed value. So that when calculating the first prediction loss, different first sample weights can be added to different first sample objects, that is, based on the first sample probability, the first sample mean, the first sample standard deviation, the first sample distribution observations and the first sample weight, jointly determine the first prediction loss.

[0103] Optionally, different value intervals can also be divided according to the first value ranking result, and different first sample weights are set for the first sample objects in different value intervals.

[0104] Step 502C, train the first value prediction model based on the first prediction loss.

[0105] After determining the first prediction loss, the backpropagation algorithm can be used to update the model parameters of the first value prediction model with the first prediction loss; after multiple rounds of training until the loss is less than the preset value, it is determined that the training of the first value prediction model is completed.

[0106] This embodiment provides a training method for the first value prediction model. By integrating the classification loss and the distribution prediction loss to train the first value prediction model, the first value prediction model can be made to have the ability to predict the LTV distribution and the conversion probability, and then realize the output of the LTV; in addition, different sample weights are set for users with different LTVs during training, which can improve the prediction accuracy of the model for ultra-high-value users.

[0107] Similar to the first value prediction model, as Figure 6 shown, the training process of the second value prediction model can include the following steps:

[0108] Step 601, obtain the second sample credit data and the second sample distribution observations of multiple second sample objects in the historical time period, and the second sample credit data is the credit data under the historical advertisement information push.

[0109] Different from the training process of the first value prediction model, since the second value prediction model is used to predict the LTV when pushing historical advertising information, the sample data used in training the second value prediction model are: the second sample credit data and second sample distribution observations of multiple second sample objects in the historical time period. The second sample credit data is the credit data when pushing historical advertising information, and the second sample distribution observations are the labeled LTV distribution observations generated based on the second sample credit data, which may specifically include the labeled mean, labeled standard deviation and labeled LTV of each second sample object.

[0110] Step 602: training a second value prediction model based on the second sample credit data and the second sample distribution observations.

[0111] Furthermore, in each round of training, the second sample credit data and historical advertising information are input into the initial second value prediction model to obtain three prediction parameters output by the second value prediction model, and then the second value prediction model is trained based on the three prediction parameters and the second sample distribution observations.

[0112] Since the second value prediction model involves two task predictions, its corresponding loss function also includes two parts, one is the classification loss (i.e., the loss of predicting the transaction probability of the candidate object), and the other is the LTV distribution prediction loss. Correspondingly, in an exemplary example, step 602 may include steps 602A to 602C.

[0113] Step 602A, input the second sample credit data and historical advertising information into the second value prediction model to obtain the second sample probability of each second sample object output by the second prediction model, as well as the second sample mean and second sample standard deviation of the second sample value distribution.

[0114] After the second sample credit data and historical advertising information are input into the second value prediction model, the second sample probability of each second sample object output by the second value prediction model (i.e., the transaction probability of the second sample object) and the second sample mean and second sample standard deviation of the second sample value distribution can be obtained.

[0115] Step 602B, determining a second prediction loss based on the second sample probability, the second sample mean, the second sample standard deviation, and the second sample distribution observation value.

[0116] Furthermore, based on the second sample probability and the standard sample probability, the classification loss is determined, and based on the second sample mean, the second sample standard deviation and the second sample distribution observation value, the second distribution prediction loss of the second sample value distribution is determined, and the sum of the classification loss and the second distribution prediction loss is determined as the second prediction loss of the second value prediction model to train the second value prediction model.

[0117] Exemplarily, the complete loss function can be as shown in formula (2).

[0118]

[0119] Wherein, L 2 represents the second prediction loss, represents the classification loss, represents the second distribution prediction loss, p 2 represents the second sample probability, x 2 represents the second sample distribution observation value, μ 2 represents the second sample mean, σ 2 represents the second sample standard deviation,

[0120] In addition, considering that although the proportion of ultra-high-value samples (ultra-high-value users) is extremely low, the ultra-high-value users are very important for business contributions. Therefore, the unstable prediction of this part of the samples may cause serious interference to the online effect. To solve this problem, when calculating the loss function during model training, the loss of this part of the second sample objects will be weighted.

[0121] Exemplarily, step 602B may further include steps 602B1 to 602B4.

[0122] Step 602B1, obtaining the corresponding second sample observation value of each second sample object from the second sample distribution observation value.

[0123] Step 602B2, sorting the multiple second sample objects based on the second sample observation value to obtain a second value sorting result.

[0124] Step 602B3, determining the corresponding second sample weight of each second sample object based on the second value sorting result.

[0125] Step 602B4, determining the second distribution prediction loss based on the second sample probability, the second sample mean, the second sample standard deviation, the second sample distribution observation value, and the second sample weight.

[0126] In order to identify high-value users from the second sample objects, it is first necessary to determine the labeled LTV (i.e., the second sample observed value) of each second sample object from the second sample distribution observations, and sort the multiple second sample objects according to the level of the labeled LTV to obtain the second value ranking result; furthermore, based on the front and back of the second value ranking result, different second sample weights are set for different second sample objects. And the higher the second sample observed value, the greater the second sample weight, and the lower the second sample observed value, the smaller the second sample weight, that is, the second sample weight is positively correlated with the second sample observed value. So that when calculating the second prediction loss, different second sample weights can be added to different second sample objects, that is, based on the second sample probability, the second sample mean, the second sample standard deviation, the second sample distribution observations and the second sample weights, the second prediction loss is jointly determined.

[0127] Optionally, different value intervals can also be divided according to the second value ranking result, and different second sample weights are set for the second sample objects in different value intervals.

[0128] Step 602C, train the second value prediction model based on the second prediction loss.

[0129] After determining the second prediction loss, the backpropagation algorithm can be used to update the model parameters of the second value prediction model using the second prediction loss; after multiple rounds of training until the loss is less than the preset value, it is determined that the training of the second value prediction model is completed.

[0130] This embodiment provides a training method for the second value prediction model. By integrating the classification loss and the distribution prediction loss to train the second value prediction model, the second value prediction model can be made to have the ability to predict the LTV distribution and the conversion probability, and then the output of the LTV can be realized; in addition, different sample weights are set for users with different LTVs during training, which can improve the prediction accuracy of the model for ultra-high-value users.

[0131] It should be noted that the advertising information involved in the embodiments of this application specifically refers to the marketing strategies in the credit field.

[0132] Please refer to Figure 7 , which is a schematic structural diagram of an advertising information push device provided by the embodiments of this application. Exemplarily, as Figure 7 shown, the device 700 includes:

[0133] The first acquisition module 701 is used to acquire the object credit data of multiple candidate objects within a target time period;

[0134] The first prediction module 702 is configured to input the object credit data into a first value prediction model to obtain a first prediction value for each candidate object, where the first prediction value is used to characterize the object value of the candidate object at a target time point;

[0135] The second prediction module 703 is configured to input the target advertisement information and the object credit data into a second value prediction model to obtain a second prediction value for each candidate object, where the second prediction value is used to characterize the object value of the candidate object at the target time point after pushing the target advertisement information to the candidate object;

[0136] The information push module 704 is configured to determine a target object from multiple candidate objects based on the first prediction value and the second prediction value of each candidate object, and push the target advertisement information to the target object.

[0137] Optionally, the policy determination module 704 is further configured to:

[0138] Determine a candidate incremental value for each candidate object based on the first prediction value and the second prediction value of each candidate object;

[0139] Determine the candidate objects with the candidate incremental value greater than a preset threshold as the target objects.

[0140] Optionally, the policy determination module 704 is further configured to:

[0141] Obtain the candidate incremental value of each target object;

[0142] Sort the target objects according to the candidate incremental value to obtain a sorting result;

[0143] Push the target advertisement information to the target objects based on the sorting result.

[0144] Optionally, the first prediction module 702 is further configured to:

[0145] Input the object credit data into the first value prediction model to obtain a first prediction probability of each candidate object output by the first value prediction model, as well as a first prediction mean and a first prediction standard deviation of a first object value distribution;

[0146] Determine the first prediction value of each candidate object based on the first prediction mean, the first prediction standard deviation, and the first prediction probability of each candidate object.

[0147] Optionally, the second prediction module 703 is further configured to:

[0148] Input the object credit data and the target advertisement information into the second value prediction model to obtain the second prediction probability of each candidate object output by the second value prediction model, as well as the second prediction mean and the second prediction standard deviation of the second object value distribution;

[0149] Based on the second prediction mean, the second prediction standard deviation, and the second prediction probability of each candidate object, determine the second prediction value of each candidate object.

[0150] Optionally, the device further includes:

[0151] A second acquisition module, configured to acquire the first sample credit data and the first sample distribution observation values of a plurality of first sample objects within a historical time period, where the first sample credit data is credit data without pushing historical advertisement information;

[0152] A first training module, configured to train the first value prediction model based on the first sample credit data and the first sample distribution observation values.

[0153] Optionally, the device further includes:

[0154] A third acquisition module, configured to acquire the second sample credit data and the second sample distribution observation values of a plurality of second sample objects within a historical time period, where the second sample credit data is credit data with pushing historical advertisement information;

[0155] A second training module, configured to train the second value prediction model based on the second sample credit data and the second sample distribution observation values.

[0156] An exemplary embodiment of the present application further provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program that can be executed by the at least one processor, and when the computer program is executed by the at least one processor, it is used to cause the electronic device to execute the method according to the embodiment of the present application.

[0157] An exemplary embodiment of the present application further provides a non-transitory computer-readable storage medium storing a computer program, where when the computer program is executed by a processor of a computer, it is used to cause the computer to execute the method according to the embodiment of the present application.

[0158] An exemplary embodiment of the present application further provides a computer program product, including a computer program, where when the computer program is executed by a processor of a computer, it is used to cause the computer to execute the method according to the embodiment of the present application.

[0159] Reference Figure 8, a structural block diagram of an electronic device 800 that can be used as a server or a client of the present application will now be described. It is an example of a hardware device that can be applied to various aspects of the present application. The electronic device is intended to represent various forms of digital electronic computer devices, such as, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.

[0160] As Figure 8 shown, the electronic device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0161] Multiple components in the electronic device 800 are connected to the I / O interface 805, including: an input unit 806, an output unit 807, a storage unit 808, and a communication unit 809. The input unit 806 can be any type of device that can input information into the electronic device 800. The input unit 806 can receive input digital or character information, and generate key signal inputs related to the user settings and / or function controls of the electronic device. The output unit 807 can be any type of device that can present information, and can include, but is not limited to, a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. The storage unit 808 can include, but is not limited to, magnetic disks, optical disks. The communication unit 809 allows the electronic device 800 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks, and can include, but is not limited to, a modem, a network card, an infrared communication device, a wireless communication transceiver, and / or a chipset, such as a Bluetooth device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0162] The computing unit 801 can be various general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 executes the various methods and processes described above. For example, in some embodiments, Figure 2 , Figure 3 , Figure 5 , Figure 6 The methods shown can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 800 via the ROM 802 and / or the communication unit 809. In some embodiments, the computing unit 801 can be configured to execute Figure 2 , Figure 3 , Figure 5 , Figure 6 The methods shown.

[0163] The program code for implementing the methods of the present application can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the program codes are executed by the processor or controller, the functions / operations specified in the flowchart and / or block diagram are implemented. The program codes can be executed entirely on the machine, partially on the machine, as an independent software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0164] In the context of the present application, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on 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 disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0165] As used in this application, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus, and / or device (e.g., a magnetic disk, an optical disk, a memory, a programmable logic device (PLD)) that provides machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term "machine-readable signal" refers to any signal that provides machine instructions and / or data to a programmable processor.

[0166] For purposes of providing an interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can also be used to provide for interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic, speech, or tactile input).

[0167] The systems and techniques described herein can be implemented in a computing system that includes a back-end component (e.g., as a data server), or a computing system that includes a middleware component (e.g., an application server), or a computing system that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0168] A computer system can include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

Claims

1. A method for pushing advertising information, characterized in that: The method comprises: Obtaining object credit data of multiple candidate objects within a target time period; Inputting the object credit data into a first value prediction model to obtain a first prediction value for each candidate object, wherein the first prediction value is used to characterize the object value of the candidate object at a target time point; Inputting the target advertisement information and the object credit data into a second value prediction model to obtain a second prediction value for each candidate object, wherein the second prediction value is used to characterize the object value of the candidate object at the target time point after the target advertisement information is pushed to the candidate object; Based on the first prediction value and the second prediction value of each of the candidate objects, a target object is determined from the plurality of candidate objects, and the target advertisement information is pushed to the target object.

2. The method according to claim 1, characterized in that The step of determining a target object from a plurality of candidate objects based on the first prediction value and the second prediction value of each candidate object includes: Determining a candidate incremental value for each candidate object based on the first predicted value and the second predicted value for each candidate object; The candidate object whose candidate incremental value is greater than a preset threshold is determined as the target object.

3. The method according to claim 2, characterized in that The pushing the target advertisement information to the target object includes: Obtaining the candidate incremental value of each of the target objects; Sorting the target objects according to the candidate incremental values ​​to obtain a sorting result; The target advertisement information is pushed to the target object based on the ranking result.

4. The method according to any one of claims 1 to 3, characterized in that: The step of inputting the object credit data into a first value prediction model to obtain a first prediction value for each candidate object includes: Inputting the object credit data into the first value prediction model, obtaining a first prediction probability of each candidate object output by the first value prediction model, and a first prediction mean and a first prediction standard deviation of the first object value distribution; The first prediction value of each of the candidate objects is determined based on the first prediction mean, the first prediction standard deviation, and the first prediction probability of each of the candidate objects.

5. The method according to any one of claims 1 to 3, characterized in that: The step of inputting the target advertisement information and the object credit data into a second value prediction model to obtain a second prediction value for each candidate object includes: Inputting the object credit data and the target advertisement information into the second value prediction model, obtaining a second prediction probability of each candidate object output by the second value prediction model, and a second prediction mean and a second prediction standard deviation of the second object value distribution; The second predicted value of each candidate object is determined based on the second predicted mean, the second predicted standard deviation, and the second predicted probability of each candidate object.

6. The method according to any one of claims 1 to 3, characterized in that: The method further comprises: Acquire first sample credit data and first sample distribution observation values ​​of a plurality of first sample objects within a historical time period, wherein the first sample credit data is credit data without pushing historical advertising information; The first value prediction model is trained based on the first sample credit data and the first sample distribution observations.

7. The method according to any one of claims 1 to 3, characterized in that: The method further comprises: Acquire second sample credit data and second sample distribution observation values ​​of multiple second sample objects in a historical time period, wherein the second sample credit data is credit data under the pushed historical advertising information; The second value prediction model is trained based on the second sample credit data and the second sample distribution observations.

8. An advertising information pushing device, characterized in that: The device comprises: A first acquisition module is used to acquire object credit data of multiple candidate objects within a target time period; A first prediction module, used for inputting the object credit data into a first value prediction model to obtain a first prediction value of each candidate object, wherein the first prediction value is used to characterize the object value of the candidate object at a target time point; A second prediction module, used for inputting the target advertisement information and the object credit data into a second value prediction model to obtain a second prediction value of each candidate object, wherein the second prediction value is used to represent the object value of the candidate object at the target time point after the target advertisement information is pushed to the candidate object; An information push module is used to determine a target object from a plurality of the candidate objects based on the first prediction value and the second prediction value of each of the candidate objects, and to push the target advertisement information to the target object.

9. An electronic device, comprising: processor; as well as Memory for storing programs, The program includes instructions, which, when executed by the processor, cause the processor to perform the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-7.

Images