Promotion resource configuration method, system, device and storage medium

Abstract

The present application discloses a promotion resource configuration method, system device and storage medium, which relates to the field of artificial intelligence. The method includes the following steps: obtaining a first data group and a second data group of multiple promotion accounts, inputting the first data group of each promotion account into a first model, and obtaining a first prediction result of each promotion account, the first model is used to predict the promotion message delivery capability of the promotion account; inputting the second data group of each promotion account into a second model, and obtaining a second prediction result, the second model is used to predict the probability that the promotion account is a target account; determining multiple target accounts according to the first prediction result and the second prediction result of each promotion account; improving the ranking of at least some of the target accounts in the advertising delivery sorting. The present application can improve the prediction accuracy of high-potential customers, thereby optimizing the resource configuration of the promotion message delivery system.

Description

Technical Field

[0001] The present application relates to artificial intelligence technology, and in particular to a promotion resource configuration method, system, device and storage medium. Background Art

[0002] In modern promotional message delivery systems, promotional users generally use prepaid payment to place promotions in the promotional message delivery system. Promotional users set certain delivery conditions, such as setting the characteristics of the target group, bidding, etc. The promotional message delivery system performs calculations based on the conditions set by the promotional users, and adjusts the promotional user's promotional message delivery position, appearance time and other parameters based on the results of the calculations, thereby improving the profits of the promotional message delivery system and the promotional user's delivery effect.

[0003] However, due to various reasons, such as the bid set by the customer is too low, the characteristics of the target group are problematic, etc., the promotion effect of the promoted users is not good, and these promoted users gradually disappear or their consumption slows down. Among the massive promoted users, there are many customers with consumption potential. It is necessary to use certain means to tap these promoted users with high consumption potential (hereinafter referred to as high-potential users), so as to achieve targeted marketing or tilt resources, help promoted users improve their delivery strategies and promote their consumption.

[0004] In related technologies, the RFM (Recency of the most recent consumption, Frequency of promotional message delivery, and Monetary abbreviation of the amount of resources consumed by promotional message delivery) model is usually used to measure the potential value of customers. However, the RFM model performs poorly in predicting the potential of promotional users. The reason is that promotional message delivery belongs to the type of recharge first and then consumption, and consumption is lagging information. This model easily leads to the erroneous filtering out of a large number of promotional users who recharge but do not consume. In addition, the traditional RFM model only has three dimensions of data and lacks the mining of deep information of promotional users, resulting in poor recall rate of the model in the task of predicting the consumption potential of promotional users. Due to inaccurate screening results of high-potential promotional users, the promotional message delivery system incorrectly allocates resources, resulting in increased operating costs. Summary of the invention

[0005] In order to solve at least one of the above technical problems, the present application provides a promotion resource configuration method, system, device and storage medium to optimize resource allocation of a promotion message delivery system.

[0006] According to one aspect of the present application, a promotion resource configuration method is provided, comprising the following steps:

[0007] Acquire a first data group and a second data group of multiple promotion accounts, wherein the first data group includes a record of the most recent promotion message delivery of the promotion account, a promotion message delivery frequency of the promotion account within a preset time, a consumption amount of resources for promotion message delivery of the promotion account within the preset time, a resource transfer record of the promotion account, and a resource surplus of the promotion account, and the second data group includes a variety of data for describing the promotion message delivery behavior of the promotion account;

[0008] inputting the first data group of each promotion account into a first model to obtain a first prediction result of each promotion account, wherein the first model is used to predict the promotion message delivery capability of the promotion account;

[0009] Inputting the second data set of each promotion account into a second model to obtain a second prediction result, wherein the second model is used to predict the probability that the promotion account is a target account;

[0010] Determine a plurality of target accounts according to the first prediction result and the second prediction result of each of the promotion accounts;

[0011] Improve the ranking of the promotional messages delivered by the multiple target accounts in the promotional message delivery sequence, wherein the promotional message delivery sequence includes multiple promotional messages arranged in sequence, and the promotional message delivery system delivers the promotional messages to the target terminal in the sequence. According to another aspect of the present application, a promotion resource configuration system is provided, comprising:

[0012] an acquisition unit, configured to acquire a first data group and a second data group of a plurality of promotion accounts, wherein the first data group includes a record of the most recent promotion message delivery of the promotion account, a promotion message delivery frequency of the promotion account within a preset time, a consumption amount of resources for promotion message delivery of the promotion account within the preset time, a resource transfer record of the promotion account, and a resource surplus of the promotion account, and the second data group includes a plurality of data for describing the promotion message delivery behavior of the promotion account;

[0013] a first prediction unit, configured to input a first data group of each promotion account into a first model to obtain a first prediction result of each promotion account, wherein the first model is used to predict the promotion message delivery capability of the promotion account;

[0014] A second prediction unit, used for inputting the second data group of each promotion account into a second model to obtain a second prediction result, wherein the second model is used for predicting the probability that the promotion account is a target account;

[0015] a target account determination unit, configured to determine a plurality of target accounts according to the first prediction result and the second prediction result of each of the promotion accounts;

[0016] The resource configuration unit is used to improve the ranking of the promotional messages delivered by the multiple target accounts in the promotional message delivery sequence, wherein the promotional message delivery sequence includes multiple promotional messages arranged in sequence, and the promotional message delivery system delivers the promotional messages to the target terminal in the sequence.

[0017] According to another aspect of the present application, a promotion resource configuration device is provided, comprising:

[0018] program;

[0019] A memory, used for storing the program;

[0020] A processor is used to load the program to execute the promotion resource configuration method.

[0021] According to another aspect of the present application, a computer-readable storage medium is provided, wherein a program is stored in the computer-readable storage medium, and when the program is executed by a processor, the promotion resource configuration method is implemented.

[0022] The beneficial effect of the embodiments of the present application is that the promotional message delivery capability of the promotional account is predicted by the first data group, and the two dimensions of resource transfer record and balance are added compared to the traditional RFM model. This is more suitable for the application scenario of promotional message delivery, which is a recharge-before-consumption application scenario, and increases the precision and recall of the prediction results in this application scenario. At the same time, the promotional account is predicted by the second model in binary classification to make up for the lack of precision of the first model. Therefore, the target account is determined based on the first prediction result and the second prediction result, so that the prediction result is more accurate. By improving the ranking of the promotional messages delivered by the target account in the promotional message delivery sequence, more or better resources can be automatically allocated to the promotional account with consumption potential, which can optimize the resource allocation of the system and save operating costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flowchart of a promotion resource configuration method provided according to an embodiment of the present application;

[0024] Figure 2 A block diagram of a promotion message delivery system for an application promotion resource configuration method provided according to an embodiment of the present application;

[0025] Figure 3 A schematic diagram of a working principle of a promotional message delivery system provided according to an embodiment of the present application;

[0026] Figure 4 It is another schematic diagram of the working principle of the promotion message delivery system provided according to an embodiment of the present application;

[0027] Figure 5 A schematic diagram of the structure of an RFMRB model provided according to an embodiment of the present application;

[0028] Figure 6 A schematic diagram of the structure of a GBDT model provided according to an embodiment of the present application;

[0029] Figure 7 A schematic diagram of a joint model of an RFMRB model and a GBDT model provided according to an embodiment of the present invention;

[0030] Figure 8 A schematic diagram of sorting promotional messages provided according to an embodiment of the present invention;

[0031] Fig. 9 A schematic diagram of a media terminal interface provided according to an embodiment of the present invention;

[0032] Fig.10 A schematic diagram of advertisement sorting of promotional message display positions on a media terminal provided according to an embodiment of the present invention;

[0033] Fig.11 A flowchart of sub-steps of step 140 provided according to an embodiment of the present invention;

[0034] Fig.12 A flow chart of a training method for a second model provided according to an embodiment of the present invention;

[0035] Fig.13 A flowchart of a method for obtaining the data type of a second data group provided according to an embodiment of the present invention;

[0036] Fig.14 A flowchart of another promotion resource configuration method provided according to an embodiment of the present invention;

[0037] Fig.15 A schematic diagram of the classification of promotion users provided according to an embodiment of the present invention;

[0038] Fig.16 A module block diagram of a promotion resource configuration system provided according to an embodiment of the present invention;

[0039] Fig.17 A module block diagram of a server provided according to an embodiment of the present invention;

[0040] Fig.18 The present invention is a module block diagram of a terminal provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it can be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0042] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0043] Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations.

[0044] GBDT: (Gradient Boosting Decision Tree) Gradient Boosting Iterative Decision Tree, its main idea is that each time a single classifier is established, it is in the gradient descent direction of the loss function of the previously established model. GBDT goes through multiple rounds of iterations, each round of iteration produces a weak classifier, each classifier is trained based on the residual of the previous round of classifiers, and the final total classifier is the weighted sum of the weak classifiers obtained in each round of training.

[0045] Consumption: The amount spent by the promotion user on delivering promotional messages.

[0046] ARPU: (Average Revenue Per User) Average revenue per user.

[0047] The RFM (Recency, Frequency, Monetary) model is an important tool and means to measure customer value and customer profitability. The model measures customer value through three indicators: the most recent consumption (Recency), the frequency of promotional message delivery (Frequency), and the amount of resources consumed by promotional message delivery (Monetary). The RFM model is a common economic forecasting model. The model generally scores applications in the above three dimensions and then weights them to obtain the total score of the customer.

[0048] RFMRB (Recency, Frequency, Monetary, Recharge, Balance) model is a new model in the field of promotional message delivery proposed in this solution. It improves the pay-before-consumption model in the field of promotional message delivery based on the RFM model, and adds two new dimensions, recharge (Recharge) and balance (Balance), to measure the customer's potential promotional message delivery ability. This model takes into account the recharge and balance of the promotion user, so that promotion users with high recharge and balance will not be given low scores by the model because of no recent consumption.

[0049] Correlation coefficient: In this application, it mainly refers to the parameter used to describe the correlation between features and results. In general, if a change in a feature causes a relatively large change in the result, it means that the correlation between this feature and the result is relatively large. For example, in the linear equation z=10000x+y, in this example, it is obvious that the change of feature x has a relatively large impact on the result z, while the impact of feature y on the result z is relatively small. By analyzing the correlation between different feature parameters and results, it can help the model reduce the number of features.

[0050] Recall Ratio: In the field of machine learning, recall is calculated by the following formula: TP / (TP+FN), where TP refers to the number of samples that were originally positive and predicted to be positive by the model; FN refers to the number of samples that were originally positive and predicted to be negative by the model.

[0051] Precision: In the field of machine learning, precision is calculated by the following formula: TP / (TP+FP), where TP refers to the number of samples that were originally positive and predicted to be positive by the model; FP refers to the number of samples that were originally negative and predicted to be positive by the model. Both recall and precision can be used to describe the accuracy of a model.

[0052] OCPX is a bidding delivery model. Generally, users who choose this type of delivery model optimize conversion costs. The key formula that determines whether a promotional message can be exposed is: OCPM = Bid*PCTR*PCVR*β. The higher the OCPM, the greater the probability of exposure. Therefore, there are two possibilities that affect the competitiveness of promotional messages: the bid Bid and the delivery system estimate. That is, the higher the bid or the higher the system estimated conversion rate, the greater the OCPM value, and thus the greater the possibility of promotional messages being exposed.

[0053] CTR (Click Through Rate): refers to the click-through rate of online promotional messages, that is, the actual number of clicks on the promotional message divided by the number of impressions of the promotional message. It is an important indicator to measure the effectiveness of Internet promotional messages.

[0054] Bid: When placing an ad, a user hopes to obtain a single conversion at a certain cost, where a certain bid is defined as a Bid.

[0055] PCTR (Predict Click Through Rate) is the probability of a promotional message being clicked after it is delivered in a certain situation. It is an important component of the sorting model. In the multi-level sorting model, it represents the click-through rate model in the featured sorting, and the model complexity is high.

[0056] CVR (Conversion Rate): It is also an indicator to measure the effectiveness of promotional messages. It refers to the conversion ratio of users clicking on promotional messages to becoming an effectively activated, registered or paying user, that is, the actual number of conversions of the promotional message divided by the number of clicks on the promotional message.

[0057] PCVR (Predict Conversion Rate) is the probability of conversion after the promotional message is clicked in a certain situation. The online promotional message delivery system estimates it. It is an important component of the sorting model. In the multi-level sorting model, it represents the conversion rate model in the selected sorting, and the model complexity is high.

[0058] β is a price adjustment factor, also called weight in this application. In this application, the price adjustment factor is mainly used to intervene in the ranking of the promoted user in the bidding ranking system.

[0059] Successful incubation: refers to the average consumption of promoted users for any consecutive N days in a period of time being greater than M after certain means of operation. For example, it can mean that the average daily consumption for any consecutive 7 days is greater than 5,000 yuan.

[0060] Promotional messages: refers to information used to show the audience about products and services, or to promote the audience to consume or take certain actions. For example, promotional messages can be commercial advertisements, public service advertisements, information calling on the audience to participate in specific activities or implement relevant policies.

[0061] Resources: can be understood as means of production or general equivalents in the real world or the virtual world. For example, currency is a resource in the real world, and the recharge balance or virtual currency in the promotional message delivery system can be understood as resources in the virtual world. The act of using real-world currency to recharge the promotional message delivery system in exchange for resources in the virtual world is a resource transfer behavior. In addition, the act of obtaining virtual resources by completing specific tasks or participating in lotteries is also a resource transfer behavior. Therefore, the resource transfer of this application can be understood as the process of obtaining virtual resources. The resource balance can be understood as the balance in the promotional message delivery system, that is, the balance of virtual currency.

[0062] Blockchain is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0063] The underlying blockchain platform can include processing modules such as user management, basic services, smart contracts, and operation monitoring. Among them, the user management module is responsible for the identity information management of all blockchain participants, including maintaining public and private key generation (account management), key management, and the maintenance of the correspondence between the user's real identity and the blockchain address (authority management), etc., and, under authorization, supervises and audits the transactions of certain real identities and provides risk control rule configuration (risk control audit); the basic service module is deployed on all blockchain node devices to verify the validity of business requests, and records valid requests to storage after consensus is reached. For a new business request, the basic service first adapts the interface for parsing and authentication (interface adaptation), and then encrypts the business information through the consensus algorithm (consensus management). The smart contract module is responsible for the registration and issuance of contracts, contract triggering and contract execution. Developers can define the contract logic in a programming language and publish it to the blockchain (contract registration). According to the logic of the contract terms, the key or other events are called to trigger the execution and complete the contract logic. It also provides the function of contract upgrade and cancellation. The operation monitoring module is mainly responsible for the deployment, configuration modification, contract setting, cloud adaptation and real-time status visualization output of the product during the product release process, such as alarm, monitoring network status, monitoring node equipment health status, etc. The platform product service layer provides the basic capabilities and implementation framework of typical applications. Developers can implement the business logic on the blockchain based on these basic capabilities and superimpose the characteristics of the business. The application service layer provides application services based on the blockchain solution for business participants to use.

[0064] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making. Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level technology and software-level technology. Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0065] This solution mainly involves the application of machine learning / deep learning models in the field of promotional message delivery. In related technologies, the RFM model is mainly used to analyze the consumption potential of promotion users. However, in the promotional message delivery system, the model of recharge first and then consumption is often adopted. Consumption belongs to lagging information. This model is easy to mistakenly filter out a large number of promotion users who recharge but do not consume. In addition, the traditional RFM model only has three dimensions of data and lacks the mining of deep information of promotion users, resulting in poor precision and recall of the model in the task of predicting the consumption potential of promotion users. Due to the inaccurate screening results of high-potential promotion users, the promotional message delivery system incorrectly allocates resources and allocates high-quality promotional message display positions to promotion users without consumption potential, resulting in waste of promotional message display positions. In addition, for high-potential promotion users, they can be converted through certain operational means. If an error occurs in the screening of high-potential promotion users, the system may dispatch operators to follow up with customers, but the success rate of such customer incubation is relatively low. In essence, it is a resource allocation error of the system, which leads to a sharp increase in operating costs and a decrease in the efficiency of the promotional message delivery system.

[0066] Based on the above situation, this solution proposes a promotion resource allocation method, which predicts the promotion message delivery capability of the promotion account through five dimensions: the record of the most recent promotion message delivery of the promotion account, the promotion message delivery frequency of the promotion account within a preset time, the consumption of resources for the promotion message delivery of the promotion account within a preset time, the resource transfer record of the promotion account, and the resource surplus of the promotion account. It also uses a variety of data used to describe the promotion message delivery behavior of the promotion account to predict whether the promotion account is a target account, and combines the results of the two predictions to screen out multiple target accounts, and improve their ranking in the promotion message sorting, so as to allocate superior resources to these customers and promote their consumption. In this way, the resource allocation of the promotion message delivery system can be optimized and the profit of the promotion message delivery system can be improved.

[0067] Reference Figure 1 This embodiment discloses a promotion resource configuration method, which can be applied as follows: Figure 2In the promotional message delivery system shown, the system includes a server 210 of the promotional message delivery system, a terminal 220 of the promotion user, a terminal 230 of the operator of the promotional message delivery system, and a media terminal 240. It should be understood that the terminal can be a device such as a mobile phone, a tablet computer or a personal computer. The server can be an independent physical server or a distributed server. Among them, the terminal 220 is mainly used to send advertising delivery requirements to the server 210. After calculation, the server 210 sends these requirements to the media terminal 240 connected to the server 210. It should be understood that these media terminals 240 can refer to advertising media, such as shopping software, social software, tool software, etc., which can place promotional message display positions. The media is usually presented through hardware carriers, such as mobile phones, computers or outdoor billboards. The media terminal 240 can also be a device itself that only has Internet advertising functions. The audience 250 of the advertisement refers to the group of people who see the advertisements delivered in the media terminal 240 by browsing the screen of the device. When the audience 250 of the advertisement clicks on the advertisement or performs operations such as registration or software download, the server 210 charges the advertisement, and this part of the fee will be deducted from the recharge balance of the promoted user and become consumption. When the server 210 determines that some promoted users need to be manually incubated by the operation team, the relevant information of the promoted users can be pushed to the terminal 230 of the operator, and the operator can be dispatched to follow up.

[0068] Reference Figure 3 In the server 210, an estimate will be made according to the delivery range set by the promotion user. For example, in this embodiment, there are n promotion users who deliver to men aged 20 to 30, and these n promotion users all bid for one audience behavior. The server 210 of the promotion message delivery system will first make an inventory estimate, analyzing how many men aged 20 to 30 will be active in the media connected to the promotion message delivery system in a certain period of time. This estimated behavior is determined based on the user's browsing, operation traces, user tags or user information. For example, on a certain video software, there are about 1 million tags "20 to 30 years old men" active on the video software every day. This means that the media has a daily inventory of 1 million. In the advertising field, the exposure of advertisements to the audience is called "inventory", and the behavior of estimating the number of audiences is called inventory estimation. Of course, according to the OCPX bidding advertising model, the key formula that determines whether an advertisement can be exposed is: OCPM = Bid*PCTR*PCVR*β. In this scenario, OCPM is affected by the customer's bid, estimated conversion rate, estimated click-through rate and weight β. Therefore, when the weight β of each promoted user is not adjusted, the conversion rate and click-through rate need to be estimated, and the click-through rate and conversion rate can be analyzed based on other tags of the audience and the promoted user's promotion user title. Figure 4 In the above example, there are audiences numbered 1 to m, each of which has several tags. Figure 4 In the example shown, the promotional messages sent by the promotional users are ACG, foreign wine, and movies, and the first user's tag happens to hit ACG. The degree of match between the two is generally high, and the model will calculate a relatively high estimated click-through rate and estimated conversion rate. Then, the ACG tag is also related to movies, so in the movie advertising theme, the estimated click-through rate and estimated conversion rate of this audience will also be relatively high. However, if the weights of the promotional users are the same, in addition to considering the estimated click-through rate and estimated conversion rate, the bid of the promotional user must also be considered. For example, Figure 4 In the example, suppose that the product of the estimated conversion rate and the estimated click-through rate of the promotion user 1 corresponding to the audience numbered 1 is 0.8, and the bid of the promotion user 1 is 1.5. Suppose that the product of the estimated conversion rate and the estimated click-through rate of the promotion user n corresponding to the audience numbered 1 is 0.7, and the bid of the promotion user n is 2. If there is only one promotion message display position (such as an ad position), the OCPM value of the promotion user n is larger, so the exposure of the audience numbered 1 will be sold to the promotion user n, that is, the advertisement will be displayed to the audience. Of course, the above is just a popular example. In fact, the system will push different advertisements to different audiences according to more complex conditions, so that more audiences can click or complete tasks. In this way, the advertising platform increases the speed at which the promotion user converts the balance into consumption and obtains greater profits.

[0069] Reference Figure 1 The promotion resource configuration method disclosed in this embodiment can be applied to Figure 2 In the server 210 shown, the method includes steps 110 to 150.

[0070] Step 110, obtaining a first data group and a second data group of multiple promotion accounts, the first data group includes a record of the most recent promotional message delivery of the promotion account (Recency), a promotional message delivery frequency of the promotion account within a preset time (Frequency), an amount of resources consumed by the promotional message delivery of the promotion account within the preset time (Monetary), a resource transfer record of the promotion account (Recharge), and a resource balance (Balance) of the promotion account, and the second data group includes a variety of data for describing the promotional message delivery behavior of the promotion account.

[0071] Among them, for the second data group, it is possible to distinguish between new customers and old customers (existing customers), and take different input parameters for prediction. For example, for new customers, the data in the second data group may include: industry, arpu in the past 30 days, first consumption amount, first recharge amount, total consumption in the past 30 days, total recharge in the past 30 days, balance and total historical consumption. For existing customers, the second data group may include: industry, total consumption in the past year, arpu in the past year, recharge amount in the past 30 days, arpu in the past 30 days, total consumption in the past 30 days, consumption in the past 30-60 days and balance. Among them, consumption refers to the amount consumed by the promotion account in the delivery platform.

[0072] Of course, the data type and the amount of data in the second data group can be determined according to the structure of the model. The above selected parameter type is selected based on the existing training samples, and the correlation between the above parameter type and the training label is analyzed. In this embodiment, a vector can be used to represent the first data group or the second data group. For example, when a = the record of the most recent promotional message delivery (Recency), b = the frequency of promotional message delivery (Frequency), c = the amount of resources consumed by the promotional message delivery (Monetary), d = the resource transfer record (Recharge), and e = the balance (Balance), the first data group can be expressed as x1 = (a, b, c, d, e), and the second data group can be expressed as x2 = (n1, n2, ..., n8). It should be understood that the acquisition of the above data can be achieved by accessing local storage media (such as memory, hard disk, etc.), or by accessing a remotely deployed database, or based on data and files (such as Excel) entered by the user in the interactive interface.

[0073] Step 120: Input the first data group of each promotion account into the first model to obtain the first prediction result of each promotion account. The first model is used to predict the promotion message delivery capability of the promotion account.

[0074] In this embodiment, the first model may adopt the RFMRB model. Figure 5 As shown in the figure, the RFMRB model consists of five sub-functions, which are: f for scoring the record of the most recent promotional message delivery of the promotion account; Recency (a) Function f used to score the promotional message delivery frequency of the promotion account Frequency (b) Function, used to score the amount of resources consumed by the promotional message delivery of the promotional account Monetary (c) Function used to score the resource transfer records of the promotion account Recharge (d) Function f used to score the resource balance of the promotion account Balance(e) Function. It should be understood that for parameters such as a, b, c, d and e, they themselves can also be in the form of data such as a vector, a matrix or a tensor. Taking the resource transfer record as an example, it can be represented by a vector such as d = (recharge frequency, total recharge amount, single recharge amount). It should be understood that the above five sub-functions can be linear functions, such as a linear function, a quadratic function, etc. In this embodiment, the time from the last consumption of the customer to the latest date in the past 30 days, as the time becomes shorter, the score increases from 10 to 1, the total consumption amount of the customer in the past 30 days, as the amount becomes smaller, the score increases from 10 to 1, the number of customer consumption in the past 30 days, as the frequency becomes lower, the score increases from 10 to 1, the total recharge of the customer, as the amount becomes smaller, the score increases from 10 to 1, and the customer balance, as the amount becomes smaller, the score increases from 10 to 1. In this embodiment, the parameters of each sub-function used to score the parameters can be fitted through training samples. The five sub-functions calculate each input separately and output a score of 1 to 10 as the result. Then the total scores of the five sub-functions are added together to get the final function output, i.e., F RFMRB (x1), where F RFMRB (x1) is used to indicate the promotional message delivery capability of the promotion account. The promotional message delivery capability here can indicate the user's future promotional message delivery capability.

[0075] Step 130: Input the second data set of each promotion account into the second model to obtain a second prediction result. The second model is used to predict the probability that the promotion account is the target account.

[0076] It should be understood that the second model can be a binary classification model, and its specific model structure can be a decision tree model, a neural network model, etc. The general model can complete the binary classification task after being trained with a large number of training samples and labeled data. It should be understood that the binary classification model extracts features from multiple input data and classifies them. Compared with the traditional RFM model, the binary classification model can extract the implicit features of the input data, so it has better prediction accuracy. The model selected in this embodiment is the gradient boosting iterative decision tree model, namely the GBDT model. From Figure 6 It can be seen that in this embodiment, the GBDT model is obtained by weighting multiple classifiers (also called weak classifiers). In the learning process of the GBDT model, the weak classifiers created in each round of training will be further trained based on the residuals of the weak classifiers in the previous round until the training reaches the maximum number of training times or the last weak classifier meets specific conditions.

[0077] It is understood that the target account may refer to an account with consumption potential, wherein consumption potential means that it will be successfully hatched within a certain period of time in the future, or the target account may be an account belonging to a positive example of the second model. Among them, the positive example refers to an example having the same attributes as the positive sample in the training sample of the second model. For example, in this embodiment, the positive example has the same attributes as the positive sample in the training sample, that is, it will be successfully hatched within a preset time. It is understood that after the second model is trained, it has learned the implicit features (attributes) in the training sample, and these implicit features are invisible. Before the training starts, the input data will be marked as a positive example (belonging to the target account) or a negative example (not belonging to the target account). After learning, the model can be predicted based on the new input data, so as to determine whether the account corresponding to the data is the target account. Similarly, the target account can also be understood as an account whose second data group has certain explicit or implicit features, such as the value of certain parameters is greater than a certain threshold, or the calculation results of multiple parameters are greater than a certain threshold.

[0078] The reason why the GBDT model is selected in this embodiment is that after a large amount of post-analysis of the promotional message delivery data, it was found that for the task of whether the promotion user has the ability to deliver promotional messages, there is mostly a very linear relationship between the data and the results. Therefore, the tree model has a better accuracy rate than the neural network model.

[0079] Step 140: Determine multiple target accounts based on the first prediction result and the second prediction result of each promotion account.

[0080] Reference Figure 7 ,from Figure 7 It can be seen that the first data set x1 obtains the first prediction result F after passing through the first model. RFMRB (x1), the second data set x2 is passed through the second model to obtain the second prediction result F GBDT (x2), and then F is adjusted in the ratio of 1:σ RFMRB (x1) and F GBDT (x2) is weighted, usually to improve F GBDT The influence of (x2) will set σ to a value greater than 1, so that the high accuracy of the GBDT model can be used to improve the accuracy of the model as a whole, so that the two models can complement each other to improve the accuracy and recall rate. After processing the first prediction result and the second prediction result, the prediction score F of the promoted account can be obtained. score (x). Then, the promoted accounts can be ranked according to their predicted scores. Figure 8As shown, multiple promotion accounts are arranged in descending order according to their predicted scores, and then the first several or the first several proportions of promotion accounts are selected as target accounts. The holders of these target accounts are the high-potential customers described in this application. Providing these target accounts with a tilted resource or incubating them through operations personnel can improve the incubation rate of these customers and make them stable consumption promotion users.

[0081] Step 150: Improve the ranking of the promotional messages delivered by the multiple target accounts in the promotional message delivery sequence, wherein the promotional message delivery sequence includes multiple promotional messages arranged in sequence, and the promotional message delivery system delivers the promotional messages to the target terminal in the sequence.

[0082] It should be understood that in order to tilt resources towards the target promotion user, this embodiment adjusts the ranking of the target account in the advertising placement sequence to increase the exposure of the promotion user, so as to encourage the promotion user to adjust the strategy more actively and increase the placement strength. Figure 2 The media end 240 in the above bidding advertising scenario. In the advertising delivery sorting process, the promotional messages delivered by each advertiser will be sorted according to OCPM to obtain an advertising delivery sequence composed of multiple advertisements arranged in order from large to small, and the top-ranked advertisements in the advertising delivery sequence will be selected and delivered to the advertising traffic side for exposure to users. Therefore, the exposure rate of the promotion user's advertisement is related to OCPM=Bid*PCTR*PCVR*β. When OCPM is low, the promotion user's advertisement is difficult to compete with other promotion users' advertisements. Therefore, the OCPM of the promotional message delivered by the target account can be improved by adjusting the weight β of the promotional message delivered by the target account, thereby improving the ranking and obtaining exposure.

[0083] Reference Fig. 9 In the above example, a media interface 900 is shown. In the interface 900, promotional message display position 910 and promotional message display position 920 can be seen. Both promotional message display positions are promotional message display positions driven by the data of the advertising delivery platform. Among them, promotional message display position 910 is a banner advertisement. In promotional message display position 910, multiple advertisements are played in a loop. It should be understood that the driving data of promotional message display position 910 is provided by the advertising delivery platform, as Fig. 9As shown in area 930 in the figure, the only ads that can be displayed on the media end are the top n ads, which are sorted according to the OCPM values ​​of the ads. When the promotion message delivery system does not adjust the weight coefficient β of the promotion user. The ranking of ads is determined by the product of the bid, the estimated click-through rate, and the estimated conversion rate. In the initial state, the ranking of m ads is shown in area 930. When the system identifies a high-potential promotion user, it can increase the weight β of the high-potential promotion user's ad n+1, for example, from 1 to 1.5. The result of the adjustment is shown in Figure 930. Fig.10 As shown, after adjustment, the OCPM value of ad n+1 is increased to 0.75, which is greater than the OCPM value of ad n of 0.7. Therefore, the ranking changes. Ad n+1 squeezes ad n out of the list of successful bidding rankings, which promotes the exposure of ad n+1.

[0084] Of course, in order to promote the advertising exposure of these target accounts, you can also improve the ranking of these target accounts in the specified promotion message display position and within the specified time. For example, fix the advertising ranking of a certain promoted user to the first place. Or fix the advertising of these target accounts to a certain number of places. These methods can temporarily increase the consumption of promoted users, thereby activating these potential promoted accounts.

[0085] From the above embodiments, it can be seen that by improving the structure of the prediction model of high-potential promotion accounts, the prediction accuracy and recall rate of the model can be improved, and the advertising resources can be adjusted based on the prediction results of the above models, which can reduce the allocation of high-quality resources to the wrong objects and avoid the waste of limited resources. It can be inferred that the above embodiments can reduce the optimization operation cost and promote high-potential promotion users to consume.

[0086] In addition, by allocating resources to the promotional messages delivered to the target account, the delivery effect of the target account can be improved, thereby encouraging the promotional users to increase delivery, so that more consumers' feedback on the promotional messages can be obtained. The increase of these data is conducive to optimizing the prediction model of the promotional message delivery system, making the click-through rate prediction and conversion rate prediction of the promotional message delivery system more accurate, thereby making the promotional message delivery process more stable and controllable.

[0087] Reference Fig.11 This embodiment discloses a promotion resource configuration method. In this embodiment, step 140, determining multiple target accounts according to the first prediction result and the second prediction result of each promotion account, specifically includes:

[0088] Step 1110: Weight the first prediction result and the second prediction result of each promotion account to obtain a prediction score of each promotion account.

[0089] This step is as follows Figure 7As shown in the model results, the first prediction result and the second prediction result can be weighted by 1:σ, so as to adjust the influence of the GBDT model output result on the entire model. It should be understood that since the first prediction result is the accumulation of multiple sub-functions, and each sub-function output is 0 to 10 points, the total score is as high as 50, and the second prediction result is output by the classification model, and its output result is 0 to 1. Therefore, in general, in order to maximize the role of the GBDT model, σ will be adjusted to a larger value, such as 10, 30, etc. After obtaining the predicted score of each promoted account, a certain number, a certain proportion, or promoted accounts that meet certain conditions can be selected as the target accounts for this screening.

[0090] In the step of determining multiple target accounts, it may be step 1121, step 1122 or step 1123:

[0091] Step 1121: Select multiple promotion accounts whose predicted scores are higher than a set threshold as target accounts.

[0092] The step 1121 is selected based on conditions, that is, a promotion account with a certain predicted score is selected as a target account. For example, the predicted score of the target account may be greater than 40 points. The higher the predicted score, the more likely it is that the promotion account is the target account that the system needs to find.

[0093] Step 1122: Arrange the promotion accounts in descending order according to their predicted scores, and select the top several proportions or the top several promotion accounts as target accounts.

[0094] The step 1122 is selected based on the ranking to select the target account. It does not require the predicted score of the promotion account to be high, but only requires to take out the highest batch and adjust the resources for them. This embodiment is mainly applicable to the case where the number of resources that can be allocated is limited. In these embodiments, the top 30 promotion accounts can be selected as the target accounts, or the top 30% can be selected as the target accounts.

[0095] Step 1123: Sort the promotion accounts in ascending order according to their predicted scores, and select a certain proportion or a certain number of promotion accounts at the bottom of the ranking as target accounts. Similar to step 1122, this step is to sort in ascending order, so the last promotion account in the sorting is selected as the target account. It should be understood that when the configurable promotion resources are limited, such as when the operation personnel and the promotion message display positions that can be deployed are limited, these two methods can be used to centrally incubate a certain number of target accounts.

[0096] Reference Fig.12This embodiment discloses a method for obtaining a second model. In this embodiment, the second model is a gradient boosting iterative decision tree model, and the model structure is as follows: Figure 6 As shown, of course Figure 6 Only three weak classifiers are shown in FIG. 1 . In practical applications, the number of weak classifiers may be more or less. The gradient boosting iterative decision tree model is obtained in the following manner:

[0097] Step 1210: Obtain a training sample set and a first label corresponding to each training sample in the training sample set;

[0098] Step 1220: training multiple classifiers based on the training samples in the training sample set and the first labels corresponding to the training samples, wherein the Nth classifier is trained based on the residual of the N-1th classifier, where N is a positive integer greater than 2;

[0099] Step 1230: Weight the multiple trained classifiers to obtain the gradient boosting iterative decision tree model.

[0100] In the above embodiment, each round of learning can create a weak classifier, and then each weak classifier can be learned based on the residual of the previous round of training. The following is a more detailed description of the training process of this embodiment:

[0101] Input: training set T = {(x1,y1),(x2,y2),…,(x N ,y N )}, the loss function is L(y,f(x))=ln(1+exp-2yfx,y=0,1. In this embodiment, x N is the training sample, y N Represents the label corresponding to the training sample. The goal is to output: classification tree F gbdt (x).

[0102] Among them, the first weak classifier is trained in the initialization stage to obtain f0(x):

[0103]

[0104] For m=1,2,…,M (M represents the number of iterations, i.e., the number of weak classifiers), for samples i=1,2,…,N, calculate the pseudo residual.

[0105]

[0106] For the probability residual {(x1,r m1 ),(x2,r m2 ),…,(x N ,r mN)} Fit a classification tree and get the leaf node region R of the mth tree mj , j=1,2,…,J.

[0107] For j=1,2,…,J i=1,2,…,N calculate:

[0108]

[0109] Get the final classification tree:

[0110]

[0111] During the training process, the training process is mainly adjusted by the following two parameters, which will affect the effect of the trained model. The two parameters are: n_estimators: maximum number of iterations of weak classifiers. learning_rate: weight reduction coefficient (step size) of each weak classifier. This embodiment tries multiple parameter combinations and selects the parameter combination with the best performance in the test set. New customers: n_estimators = 20, learning_rate = 0.3. Existing customers: n_estimators = 30, learning_rate = 0.4.

[0112] In some embodiments, it is to be understood that Figure 7 The model shown is obtained through training. When building the model, in order to reduce the complexity of the model, compress the storage volume of the model and reduce the amount of calculation during each execution, features that are more relevant to the prediction results can be selected from a variety of features as input features of the model.

[0113] Reference Fig.13 In this embodiment, the data types in the second data group are selected from the data types in the third data group, wherein when the data types of the second data group are selected from the data types in the third data group, the following steps are included:

[0114] Step 1310: Acquire multiple third data groups and second labels corresponding to each third data group.

[0115] Step 1320: Determine the correlation coefficient between each data type in the third data group and the second label according to each third data group and the second label corresponding to each third data group.

[0116] Step 1330: Select multiple data types whose correlation coefficients meet preset conditions as data types in the second data group.

[0117] In this embodiment, there are a total of 19 original features that can be selected for new customers, namely: industry, number of types of sales, number of types of traffic, number of recharge days in the past 30 days, number of consumption days in the past 30 days, ARPU in the past 30 days, bidding consumption ratio in the past 30 days, contract consumption ratio in the past 30 days, investment promotion consumption ratio in the past 30 days, news video traffic consumption ratio in the past 30 days, information flow QQ traffic consumption ratio in the past 30 days, alliance traffic consumption ratio in the past 30 days, WeChat traffic consumption ratio in the past 30 days, first consumption amount, first recharge amount, total consumption in the past 30 days, total recharge in the past 30 days, balance and historical total consumption.

[0118] For existing customers (customers who have consumed in the past 30 days), there are a total of 22 original features that can be selected, including industry, number of sales types in the past year, number of traffic types in the past year, number of recharge days in the past year, number of consumption days in the past year, total consumption in the past year, arpu in the past year, proportion of bidding consumption in the past year, proportion of contract consumption in the past year, proportion of investment promotion consumption in the past year, proportion of news video traffic consumption in the past year, proportion of information flow QQ traffic consumption in the past year, proportion of alliance traffic consumption in the past year, proportion of WeChat traffic consumption in the past year, total recharge amount in the past year, recharge amount in the past 30 days, arpu in the past 30 days, total consumption in the past 30 days, number of consumption days in the past 30 days, consumption in the past 30-60 days, and the consumption difference and balance between the past 30 days and 60-30 days.

[0119] The above new and existing customer samples can be labeled based on the daily consumption of the subject dimension within a specified period of time, with 1 representing successful incubation and 0 representing unsuccessful incubation, that is, users are classified into two categories. It should be understood that the above labeled data can be used for model training.

[0120] Since the training data must be numeric, it is necessary to encode the Chinese characters representing the industry into numbers and supplement some missing values. For example, for new customers, first charge: missing values ​​are filled with first consumption, total charge: if the balance is also empty, fill with 30-day total consumption, otherwise fill with balance plus total consumption, balance: fill with total charge minus total consumption, historical consumption: fill with 0. For existing customers, charge in the past 30 days: if the balance is also empty, fill with 30-day total consumption, otherwise fill with balance plus total consumption, balance: fill with total charge minus total consumption.

[0121] In this embodiment, the Spearman correlation coefficient may be used to calculate the correlation coefficient in step 1320. It should be understood that the Spearman correlation coefficient may be replaced by the Pearson correlation coefficient or the Kendall correlation coefficient, so the calculation of the correlation coefficient in this embodiment is not limited to the above example.

[0122] The Spearman correlation coefficient is a nonparametric indicator that measures the dependence of two variables. It uses a monotonic equation to evaluate the correlation between two statistical variables. If there are no repeated values ​​in the data and when the two variables are completely monotonically correlated, the Spearman correlation coefficient is +1 or -1. The Spearman correlation coefficient is defined as the Pearson correlation coefficient between ordinal variables. For a sample with a sample size of n, n original data are converted into ordinal data, and the correlation coefficient ρ is:

[0123]

[0124] In the above formula, x i represents the i-th x sample, represents the average y of all x samples i represents the label corresponding to the i-th sample x, Represents the average of all labels.

[0125] According to the analysis of training samples and labels, the features with a high correlation with the labels selected in this embodiment correspond to new customers, namely, industry, ARPU in the past 30 days, first consumption amount, first recharge amount, total consumption in the past 30 days, total recharge in the past 30 days, balance and total historical consumption.

[0126] Corresponding to existing customers are the eight following: industry, total consumption in the past year, arpu in the past year, recharge amount in the past 30 days, arpu in the past 30 days, total consumption in the past 30 days, consumption in the past 30-60 days and balance.

[0127] It is understandable that due to the difference in feature selection between new and old customers, Figure 7 The model construction shown constructs models for new customers and existing customers respectively. It should be understood that by calculating the correlation coefficient between features and labels to select a data type for the second data group from multiple data types, the complexity of the model can be reduced, the speed of training and prediction can be increased, and the hardware requirements for model operation can be reduced.

[0128] In this embodiment, step 150, improving the ranking of the promotional messages delivered by the plurality of target accounts in the promotional message delivery sequence, includes:

[0129] Increase the weight of the promotional messages of all the target accounts in the promotional message delivery system, where the ranking of the promotional messages of the target accounts in the promotional message delivery system is determined by the product of the weight, cost per action, estimated click rate and estimated conversion rate of the target account. That is, increase the β value in OCPM=Bid*PCTR*PCVR*β.

[0130] Increase the weight of promotional messages of all target accounts in the promotional message delivery system, including:

[0131] The weight of the promotional message of each target account in the promotional message delivery system is increased according to the predicted score of each target account, so that the increase in the weight of the promotional message of the target account in the promotional message delivery system is proportional to the predicted score of the target account.

[0132] In this embodiment, target accounts with higher predicted scores are more likely to be incubated, so better resources can be given to target accounts with higher predicted scores. Assuming that the score of target account 1 is 30, the score of target account 2 is 20, and the score of target account 3 is 10, if the weight β of the promotional message of target account 1 is increased by 0.3, the weight β of the promotional message of target account 2 is correspondingly increased by 20, and the weight β of the promotional message of target account 3 is increased by 0.1.

[0133] Reference Figure 5 In this embodiment, the first model includes a first sub-function, a second sub-function, a third sub-function, a fourth sub-function and a fifth sub-function. The first data group of each promotion account is input into the first model, and the step of obtaining the first prediction result of each promotion account includes:

[0134] Input the record of the most recent promotional message delivery of the promotion account into the first sub-function to score the promotion account and obtain a first score;

[0135] Inputting the promotion message delivery frequency of the promotion account within the preset time into the second sub-function to score the promotion account and obtain a second score;

[0136] Inputting the amount of resources consumed by the promotion account for the placement of promotional messages within a preset time into the third sub-function to score the promotion account and obtain a third score;

[0137] Inputting the resource transfer record of the promotion account into the fourth sub-function to score the promotion account, thereby obtaining a fourth score;

[0138] Inputting the resource margin of the promotion account into the fifth sub-function to score the resource margin of the promotion account to obtain a fifth score;

[0139] The first score, the second score, the third score, the fourth score and the fifth score are weighted to obtain a first prediction result.

[0140] In this embodiment, as described in the previous embodiment, the five sub-functions may be linear functions, such as a linear function or a quadratic function. In some examples, the weighted proportions of the five scores are equal. Therefore, it can be known from the above embodiment that the first prediction result F RFMRB (x1) = f Recency (a)+f Frequency (b)+f Monetary (c)+fRecharge (d)+f Balance (e) Of course, it is understandable that, based on the teaching of the above embodiment, the weight ratios of the five sub-functions can also be adjusted to adapt to the characteristics of new customers and existing customers.

[0141] Reference Fig.14 This embodiment discloses a promotion resource configuration method, which further includes the following steps:

[0142] Step 1410: Select push content for each target account according to the first data group of the target account.

[0143] It should be understood that when selecting push content for a target account, the situation of the target account can be determined based on the data in the first data group of the target account, and different content can be pushed to the customer according to different situations. These contents can provide suggestions for customers in certain situations, such as improving the style of advertisements, increasing bids, etc. By pushing content, users can be guided to perform relevant operations, thereby improving the effect of customer delivery and generating more consumption on the advertising platform.

[0144] The classification of promotion users is mainly as follows: Fig.15 As shown, the problems are: poor conversion rate, too narrow targeting range, insufficient competitiveness, pending activation (i.e. there is still balance but it is no longer active) and other situations.

[0145] In some embodiments, in view of the poor conversion rate, step 1410 selects push content for each target account according to the first data group of the target account, specifically including:

[0146] When the third score of the target account is greater than the first threshold, the fourth score of the target account is less than the second threshold, and the fifth score of the promotion account is less than the third threshold, a push message containing a promotion material library link is pushed to the target account.

[0147] Specifically, the first threshold value can be set to 7, the second threshold value can be set to 3, and the third threshold value can be set to 3, that is,:

[0148] f Monetary (c)≥7ANDf Recharge (d)≤3ANDf Balance (e)≤3;

[0149] For example, a target account has consumed a large amount of money in the past 30 days, but the recharge and balance are insufficient. For this type of promotion user, the main problem is that the advertising effect is not good, so they will not recharge again. For this type of situation, the system automatically compares the conversion rate of the advertisements placed by this promotion user with the overall average conversion rate of the industry in which the promotion user is located. If the system determines that the conversion rate of this customer is lower than the industry average conversion rate by a certain threshold, it will push targeted advertisements to the customer on how to improve the advertising conversion rate based on the mobile phone number used by the promotion user to register, or directly push messages to the promotion account. For example, if the advertising copy is "The conversion rate of a certain industry will increase by 50% by using these materials", after clicking on the advertisement, it will jump to the material library of the corresponding industry in the promotion message delivery system, and recommend materials with high conversion rates to customers in the material library, because materials are an important factor affecting the advertising conversion rate, thereby prompting customers to optimize and deliver the materials of their own advertisements.

[0150] In some other examples, step 1410, selecting push content for each target account according to the first data group of the target account, includes:

[0151] When the fifth score of the target account is greater than the fourth threshold, the third score of the target account is less than the fifth threshold, and the fourth score of the target account is less than the sixth threshold, a push message containing a link to a bidding configuration page is pushed to the target account.

[0152] Specifically, the fourth threshold value may be set to 7, the fifth threshold value may be set to 3, and the sixth threshold value may be set to 3, that is,

[0153] f Balance (e)≥7ANDf Monetary (c)≤3ANDf Recharge (d)≤3;

[0154] This type of promotion user has a large balance, but the balance is consumed slowly. The main problem of this type of promotion user is that there is little ad exposure, which leads to low consumption. For this type of situation, the system can automatically detect the reason for the low ad exposure and then take the next step:

[0155] If the targeting range is determined to be narrow: for example, the difference between the number of people covered by the targeting (age, gender, region, etc.) set by the promotion user and the industry targeting average reaches a specific threshold, when the promotion user logs into the delivery interface, the system will automatically pop up a prompt "The advertising targeting coverage is far lower than the average of the same industry, it is recommended to adjust the targeting", and give recommended targeting options. For industries that already have a population package, such as finance, games, etc., the promotion user will also be recommended to choose such a population package. That is, the system pushes information to the target advertising account for suggesting adjustments to the advertising delivery range.

[0156] If it is determined to be insufficiently competitive: For the current mainstream OCPX bidding ads, the key formula that determines whether an ad can be exposed is: OCPM = Bid*PCTR*PCVR*β. The higher the OCPM, the greater the probability of exposure. Therefore, there are two possibilities that affect the competitiveness of an ad, the bid Bid and the estimated value of the delivery system. That is, the higher the bid or the higher the system-estimated conversion rate, the greater the OCPM value, and the greater the possibility of the ad being exposed. Therefore, if the difference between the bid set by the promotion user and the industry average reaches a certain threshold, the system will pop up a prompt "The bid is far below the industry average, it is recommended to increase the bid" and give a recommended value. If the bid is normal, it is a problem of inaccurate delivery system estimation, especially for new ads, the system estimation may have large deviations. Therefore, the system automatically adjusts the β value of the promotion user to enable the promotion user to gain more exposure and help new promotion users to quickly increase their volume.

[0157] In some embodiments, step 1410, selecting push content for each target account according to the first data group of the target account, includes:

[0158] When the fifth score of the target account is greater than the seventh threshold, and the third score of the target account is 0, and the fourth score of the target account is 0, a promotion message is pushed to the target account. The seventh threshold is set to 7, that is:

[0159] f Balance (e)≥7ANDf Monetary (c) = 0 AND f Recharge (d) = 0;

[0160] The promotion account has no recharge or consumption in the past 30 days, but there is a large balance in the account. For such customers, activation actions are required. The system pushes advertisements from the advertising platform to customers based on the mobile phone number used by the promotion user to register. This advertisement is used to promote the delivery effect of the advertising platform. Advertising can bring more business and customers to the promotion user, etc., so as to achieve the purpose of activating this promotion user. The system regularly pushes the mobile phone number of such customers to the customer activation team, that is, pushes the account holder information corresponding to the target account to the preset terminal. The customer activation team contacts the customer to understand the specific demands of the customer, so as to convert the balance of the promotion user into advertising revenue. For other situations, the customer performance is acceptable and relatively lacking in typicality, so the remaining potential customers can be pushed to the operation team for follow-up.

[0161] It should be understood that adjusting the push information of the target account or assigning operators to follow up is also a means of allocating advertising resources. The resource allocation process is executed by the promotional message delivery system.

[0162] Reference Fig.16This embodiment discloses a promotion resource configuration system, including:

[0163] The acquisition unit 1610 is used to acquire a first data group and a second data group of multiple promotion accounts, wherein the first data group includes a record of the most recent promotion message delivery of the promotion account, a promotion message delivery frequency of the promotion account within a preset time, a consumption amount of resources of the promotion message delivery of the promotion account within the preset time, a resource transfer record of the promotion account, and a resource surplus of the promotion account, and the second data group includes a plurality of data for describing the promotion message delivery behavior of the promotion account;

[0164] The first prediction unit 1620 is used to input the first data group of each promotion account into the first model to obtain the first prediction result of each promotion account, and the first model is used to predict the promotion message delivery capability of the promotion account;

[0165] The second prediction unit 1630 is used to input the second data group of each promotion account into the second model to obtain a second prediction result, and the second model is used to predict the probability that the promotion account is a target account;

[0166] A target account determining unit 1640 is used to determine a plurality of target accounts according to the first prediction result and the second prediction result of each promotion account;

[0167] The resource configuration unit 1650 is used to improve the ranking of the promotional messages delivered by the multiple target accounts in the promotional message delivery sequence, wherein the promotional message delivery sequence includes multiple promotional messages arranged in sequence, and the promotional message delivery system delivers the promotional messages to the target terminal in the sequence.

[0168] This embodiment discloses a promotion resource configuration device, including:

[0169] program;

[0170] A memory, used for storing the program;

[0171] The processor is used to load the promotion resource configuration method mentioned above in the program.

[0172] This embodiment discloses a computer-readable storage medium, in which a program is stored. When the program is executed by a processor, the above-mentioned promotion resource configuration method is implemented.

[0173] It should be understood that the above product embodiments have the same or corresponding technical features as the method embodiments, and any same or corresponding technical features can have the same or corresponding technical effects.

[0174] The embodiment of the present application also discloses a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. The processor of the computer device can read the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the above-mentioned promotion resource configuration method.

[0175] Fig.17 The module block diagram of the server 1700 provided in the embodiment of the present application, the server 1700 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPUs) 1722 (for example, one or more processors) and memory 1732, one or more storage media 1730 (for example, one or more mass storage devices) storing application programs 1742 or data 1744. Among them, the memory 1732 and the storage medium 1730 can be short-term storage or permanent storage. The program stored in the storage medium 1730 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Furthermore, the central processing unit 1722 can be configured to communicate with the storage medium 1730, and execute a series of instruction operations in the storage medium 1730 on the server 1700 to implement the above method.

[0176] The server 1700 may also include one or more power supplies 1726, one or more wired or wireless network interfaces 1750, one or more input and output interfaces 1758, and / or, one or more operating systems 1741, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.

[0177] The present application also provides a device that can execute the above-mentioned promotion resource configuration method. The device is described below in conjunction with the accompanying drawings. Fig.18 , the embodiment of the present application provides a device, which may also be a terminal device, which may be any intelligent terminal including a mobile phone, a tablet computer, a personal digital assistant (PDA), a point of sales (POS), a vehicle-mounted computer, etc., taking the mobile phone as an example:

[0178] Fig.18 FIG. 1 is a block diagram showing a partial structure of a mobile phone related to a terminal device provided in an embodiment of the present application. Fig.18The mobile phone includes: a radio frequency (RF) circuit 1810, a memory 1820, an input unit 1830, a display unit 1840, a sensor 1850, an audio circuit 1860, a wireless fidelity (WiFi) module 1870, a processor 1880, and a power supply 1890. Those skilled in the art will understand that Fig.18 The structure of the mobile phone shown in the figure does not constitute a limitation on the mobile phone, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently. Fig.18 The various components of the mobile phone are introduced in detail: the RF circuit 1810 can be used to receive and send signals during the process of sending and receiving information or making calls. In particular, after receiving the downlink information of the base station, it is sent to the processor 1880 for processing; in addition, the designed uplink data is sent to the base station. Generally, the RF circuit 1810 includes but is not limited to an antenna, at least one amplifier, a transceiver, a coupler, a low noise amplifier (LNA), a duplexer, etc. In addition, the RF circuit 1810 can also communicate with the network and other devices through wireless communication.

[0179] The memory 1820 can be used to store software programs and modules. The processor 1880 executes various functional applications and data processing of the mobile phone by running the software programs and modules stored in the memory 1820. The memory 1820 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory 1820 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0180] The input unit 1830 can be used to receive input digital or character information, and to generate key signal input related to the user settings and function control of the mobile phone. Specifically, the input unit 1830 may include a touch panel 1831 and other input devices 1832. The touch panel 1831, also known as a touch screen, can collect the user's touch operation on or near it (such as the user's operation on the touch panel 1831 or near the touch panel 1831 using any suitable object or accessory such as a finger, stylus, etc.), and drive the corresponding connection device according to a pre-set program. Optionally, the touch panel 1831 may include two parts: a touch detection device and a touch controller. Among them, the touch detection device detects the user's touch orientation, detects the signal brought by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device, converts it into contact coordinates, and then sends it to the processor 1880, and can receive and execute commands sent by the processor 1880. In addition, the touch panel 1831 can be implemented in various types such as resistive, capacitive, infrared, and surface acoustic waves. In addition to the touch panel 1831, the input unit 1830 may further include other input devices 1832. Specifically, other input devices 1832 may include but are not limited to one or more of a physical keyboard, function keys (such as volume control keys, switch keys, etc.), a trackball, a mouse, a joystick, and the like.

[0181] The display unit 1840 may be used to display information input by the user or information provided to the user and various menus of the mobile phone. The display unit 1840 may include a display panel 1841. Optionally, the display panel 1841 may be configured in the form of a liquid crystal display (LCD), an organic light-emitting diode (OLED), etc. Further, the touch panel 1831 may cover the display panel 1841. When the touch panel 1831 detects a touch operation on or near it, it is transmitted to the processor 1880 to determine the type of touch event. Subsequently, the processor 1880 provides a corresponding visual output on the display panel 1841 according to the type of touch event. Although in Fig.18 In the embodiment, the touch panel 1831 and the display panel 1841 are used as two independent components to realize the input and output functions of the mobile phone, but in some embodiments, the touch panel 1831 and the display panel 1841 can be integrated to realize the input and output functions of the mobile phone.

[0182] The mobile phone may also include at least one sensor 1850, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor, wherein the ambient light sensor may adjust the brightness of the display panel 1841 according to the brightness of the ambient light, and the proximity sensor may turn off the display panel 1841 and / or the backlight when the mobile phone is moved to the ear.

[0183] The audio circuit 1860, the speaker 1861, and the microphone 1862 can provide an audio interface between the user and the mobile phone. The audio circuit 1860 can transmit the electrical signal converted from the received audio data to the speaker 1861, which converts it into a sound signal for output; on the other hand, the microphone 1862 converts the collected sound signal into an electrical signal, which is received by the audio circuit 1860 and converted into audio data. The audio data is then processed by the output processor 1880 and sent to, for example, another mobile phone through the RF circuit 1810, or the audio data is output to the memory 1820 for further processing. WiFi is a short-range wireless transmission technology. The mobile phone can help users send and receive emails, browse the web, and access streaming media through the WiFi module 1870. It provides users with wireless broadband Internet access. Although Fig.18 A WiFi module 1870 is shown, but it is understandable that it is not an essential component of the mobile phone and can be omitted as needed without changing the essence of the invention.

[0184] The processor 1880 is the control center of the mobile phone, and uses various interfaces and lines to connect various parts of the entire mobile phone, and executes various functions of the mobile phone and processes data by running or executing software programs and / or modules stored in the memory 1820, and calling data stored in the memory 1820, so as to monitor the mobile phone as a whole. Optionally, the processor 1880 may include one or more processing units, and the mobile phone also includes a power supply 1890 for supplying power to various components.

[0185] In this embodiment, the processor 1880 included in the terminal device is capable of executing the promotion resource configuration method of the previous embodiment.

[0186] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0187] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0188] If the integrated unit described in the present application is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store program codes.

[0189] The above is a specific description of the preferred implementation of the present application, but the present application is not limited to the above embodiments. Technical personnel familiar with the field can also make various equivalent modifications or substitutions without violating the spirit of the present application. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present application.

Claims

1. A promotion resource allocation method, characterized in that: The following steps are involved: Acquire a first data group and a second data group of multiple promotion accounts, wherein the first data group includes a record of the most recent promotion message delivery of the promotion account, a promotion message delivery frequency of the promotion account within a preset time, a consumption amount of resources for promotion message delivery of the promotion account within the preset time, a resource transfer record of the promotion account, and a resource surplus of the promotion account, and the second data group includes a variety of data for describing the promotion message delivery behavior of the promotion account; inputting the first data group of each promotion account into a first model to obtain a first prediction result of each promotion account, wherein the first model is used to predict the promotion message delivery capability of the promotion account; Inputting the second data set of each promotion account into a second model to obtain a second prediction result, wherein the second model is used to predict the probability that the promotion account is a target account; Determine a plurality of target accounts according to the first prediction result and the second prediction result of each of the promotion accounts; Improving the ranking of the promotional messages delivered by the plurality of target accounts in a promotional message delivery sequence, wherein the promotional message delivery sequence includes a plurality of promotional messages arranged in sequence, and the promotional message delivery system delivers the promotional messages to the target terminal in the sequence.

2. The method according to claim 1, characterized in that The determining of a plurality of target accounts according to the first prediction result and the second prediction result of each of the promotion accounts includes: weighting the first prediction result and the second prediction result of each promotion account to obtain a prediction score of each promotion account; The step of determining the multiple target accounts includes one of the following: Selecting a plurality of the promotion accounts whose predicted scores are higher than a set threshold as target accounts; or, Arrange the promotion accounts in descending order according to their predicted scores, and select the top several percentages or top several promotion accounts as target accounts; or, The promotion accounts are arranged in ascending order according to their predicted scores, and a certain proportion or a certain number of the promotion accounts ranked last are selected as target accounts.

3. The method according to claim 1, characterized in that The second model is a gradient boosting iterative decision tree model, and the gradient boosting iterative decision tree model is obtained in the following manner: Obtaining a training sample set and a first label corresponding to each training sample in the training sample set; Training multiple classifiers according to the training samples in the training sample set and the first labels corresponding to the training samples, wherein the Nth classifier is trained based on the residual of the N-1th classifier, where N is a positive integer greater than 2; The gradient boosting iterative decision tree model is obtained by weighting multiple trained classifiers.

4. The method according to claim 1, characterized in that: The data types in the second data group are selected from the data types in the third data group, wherein when the data types of the second data group are selected from the data types in the third data group, the following steps are included: Acquire a plurality of the third data groups and a second label corresponding to each of the third data groups; Determine the correlation coefficient between each data type in the third data group and the second label according to each third data group and the second label corresponding to each third data group; A plurality of data types whose correlation coefficients satisfy a preset condition are selected as the data types in the second data group.

5. The method according to claim 2, characterized in that: Improving the ranking of the promotional messages delivered by the plurality of target accounts in the promotional message delivery sequence includes: The weight of the promotional message of the target account in the promotional message delivery system is increased, wherein the ranking of the promotional message delivered by the target account in the promotional message delivery system is determined by the product of the weight of the target account, the cost per action, the estimated click rate and the estimated conversion rate.

6. The method according to claim 5, characterized in that The step of increasing the weight of the promotional message of the target account in the promotional message delivery system includes: The weight of the promotional message of each target account in the promotional message delivery system is increased according to the predicted score of each target account, so that the increase in the weight of the promotional message of the target account in the promotional message delivery system is proportional to the predicted score of the target account.

7. The method according to claim 1, characterized in that The first model includes a first sub-function, a second sub-function, a third sub-function, a fourth sub-function and a fifth sub-function. The step of inputting the first data group of each promotion account into the first model to obtain the first prediction result of each promotion account includes: Inputting the record of the most recent promotional message delivery of the promotion account into the first sub-function to score the promotion account to obtain a first score; Inputting the promotion message delivery frequency of the promotion account within the preset time into the second sub-function to score the promotion account to obtain a second score; Inputting the amount of resources consumed by the promotion account for placing promotion messages within the preset time into the third sub-function to score the promotion account, thereby obtaining a third score; Inputting the resource transfer record of the promotion account into the fourth sub-function to score the promotion account to obtain a fourth score; Inputting the resource margin of the promotion account into the fifth sub-function to score the resource margin of the promotion account to obtain a fifth score; The first score, the second score, the third score, the fourth score and the fifth score are weighted to obtain a first prediction result.

8. The method according to claim 7, characterized in that The method further comprises the following steps: Select push content for each of the target accounts according to the first data group of the target accounts.

9. The method according to claim 8, characterized in that The selecting push content for each target account according to the first data group of the target account includes: When the third score of the target account is greater than the first threshold, the fourth score of the target account is less than the second threshold, and the fifth score of the promotion account is less than the third threshold, a push message containing a promotion material library link is pushed to the target account.

10. The method according to claim 8, characterized in that The selecting push content for each target account according to the first data group of the target account includes: When the fifth score of the target account is greater than the fourth threshold, the third score of the target account is less than the fifth threshold, and the fourth score of the target account is less than the sixth threshold, a push message containing a link to a bidding configuration page is pushed to the target account.

11. The method according to claim 8, characterized in that The selecting push content for each target account according to the first data group of the target account includes: When the fifth score of the target account is greater than a seventh threshold, and the third score of the target account is 0, and the fourth score of the target account is 0, a promotion message is pushed to the target account.

12. The method according to claim 1, characterized in that The method further comprises the following steps: The account holder information corresponding to the target account is pushed to a preset terminal.

13. A promotion resource allocation system, characterized in that: include: an acquisition unit, configured to acquire a first data group and a second data group of a plurality of promotion accounts, wherein the first data group includes a record of the most recent promotion message delivery of the promotion account, a promotion message delivery frequency of the promotion account within a preset time, a consumption amount of resources for promotion message delivery of the promotion account within the preset time, a resource transfer record of the promotion account, and a resource surplus of the promotion account, and the second data group includes a plurality of data for describing the promotion message delivery behavior of the promotion account; a first prediction unit, configured to input a first data group of each promotion account into a first model to obtain a first prediction result of each promotion account, wherein the first model is used to predict the promotion message delivery capability of the promotion account; A second prediction unit, used for inputting the second data group of each promotion account into a second model to obtain a second prediction result, wherein the second model is used for predicting the probability that the promotion account is a target account; a target account determination unit, configured to determine a plurality of target accounts according to the first prediction result and the second prediction result of each of the promotion accounts; The resource configuration unit is used to improve the ranking of the promotional messages delivered by the multiple target accounts in the promotional message delivery sequence, wherein the promotional message delivery sequence includes multiple promotional messages arranged in sequence, and the promotional message delivery system delivers the promotional messages to the target terminal in the sequence.

14. A promotion resource allocation device, characterized in that: include: program; A memory, used for storing the program; A processor is used to load the program to execute the promotion resource configuration method according to any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program, and when the program is executed by a processor, the promotion resource configuration method according to any one of claims 1 to 12 is implemented.

Images