Advertisement putting effect prediction method and system, electronic equipment and storage medium
By dividing the historical time period of advertising delivery into multiple time segments and building an LSTM model, we predict the number of order conversions and returns of fast-moving consumer goods advertising, the problem of uncertainty in the prediction of delivery effect under the CPM investment mode is solved, and the accuracy of delivery effect and merchants' decision-making ability are improved.
Patent Information
- Application Number
- CN202510562877.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-30
- Publication Date
- 2025-06-03
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Under the CPM investment model, there is uncertainty in the prediction of the advertising delivery effect of fast-moving consumer goods, which leads to diminishing marginal benefits of increased investment and may lead to the problem of increased returns and exchanges.
By dividing the ad delivery historical time period into multiple time segments, a prediction system based on the LSTM model is built to predict the number of order conversions and returns, so as to judge the advertising delivery effect.
It improves the accuracy of advertising delivery performance prediction, helps merchants determine whether to continue to place advertisements or adjust their delivery plans in cost and profit accounting, and reduces delivery uncertainty.
Smart Images
Figure CN120088016A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of advertising placement, and particularly to a method, system, electronic device and storage medium for predicting the effect of advertising placement. Background Art
[0002] Among many existing advertising forms, online advertising has advantages such as wide dissemination and low cost compared to offline advertising, so it is chosen by many sellers. The main charging methods for online advertising include CPM (charging per thousand impressions) and CPA (charging per single conversion), etc. The CPM method is mainly applicable to products with a large demand for exposure.
[0003] For fast-moving consumer goods, increasing the investment in CPM to form long-term continuous advertising placement can effectively increase the exposure of products, and thus extend the sales cycle of fast-moving consumer goods, resulting in more sales of goods. However, on the other hand, the relationship between the increase in exposure and the increase in sales volume is not linear. As the investment amount and investment time in CPM increase, the marginal benefit obtained by the product decreases. Moreover, fast-moving consumer goods often have the characteristic of meager profits, so the occurrence of returns and exchanges has a strong negative impact on the overall profit margin of fast-moving consumer goods. Although the increase in exposure and exposure time bring an increase in sales volume, they also bring an increase in the situation of returns and exchanges. Therefore, whether the increase in investment in CPM can bring an increase in actual profit has a high degree of uncertainty. Summary of the Invention
[0004] Based on this, in view of the problem that the investment effect on CPM has a high degree of uncertainty, it is necessary to provide a method, system, electronic device and storage medium for predicting the effect of advertising placement.
[0005] The technical solution provided by the present invention is as follows: The present invention provides a method for predicting the effect of advertising placement, including: Dividing the historical advertising placement time period into multiple time segments, where the length of each time segment is T, and the advertising placement number n corresponding to the Mnads th time segment Mnocvr and the order conversion number Mnocvr corresponding to the order conversion number Mnpr ; Combining at least the advertising placement number Mnads , Mn + mads and the order conversion number Mnocvr , Mn + mocvr to form a first training array, and forming a first training set by all the first training arrays, and constructing a first LSTM model based on the first training set, wherem ≥2; At least the number of order conversions Mnocvr 、 Mn + mocvr and the number of returns Mnpr 、 Mn + mpr Combine to form a second training array, and form a second training set from all the second training arrays, and build a second LSTM model based on the second training set; The first LSTM model receives the number of advertising placement plans and outputs the predicted number of order conversions, and the second LSTM model receives the predicted number of order conversions and outputs the predicted number of returns.
[0006] In one embodiment, the method further includes: judging the advertising placement effect based on at least the number of advertising placement plans, the predicted number of order conversions, and the predicted number of returns.
[0007] In one embodiment, a conversion rate threshold and a return rate threshold are set. Only when the ratio between the predicted number of order conversions and the number of advertising placement plans is greater than the conversion rate threshold, and the ratio between the predicted number of returns and the predicted number of order conversions is less than the return rate threshold, the advertising placement plan is executed.
[0008] In one embodiment, the time segment length T is negatively correlated with m .
[0009] In one embodiment, the number of time segments is N , T = 12h, m= 14, the loss function when building the first LSTM model , where, is the predicted number of order conversions output by the first LSTM model for the number of advertising placements Mi +14 ads
[0010] In one embodiment, the loss function when building the second LSTM model , where, is the predicted return rate output by the second LSTM model for the number of order conversions Mi +14 ocvr
[0011] In one embodiment, the first training array also includes the number of advertising placements Mn+ 1 ads 、 Mn + m -1 ads and the number of order conversions Mn+ 1 ocvr 、 Mn + m -1ocvr 。
[0012] An advertising placement effect prediction system, comprising: A data acquisition module, configured to divide the historical time period of advertising placement into multiple time segments, and retrieve the advertising placement number 、 The order conversion number and the return number corresponding to the order conversion number; A first training module, configured to obtain a first training array composed of the advertising placement number and the order conversion number, and form all the first training arrays into a first training set, and construct a first LSTM model based on the first training set; A second training module, configured to obtain a second training array composed of the order conversion number and the return number, and form all the second training arrays into a second training set, and construct a second LSTM model based on the second training set; A data processing module, configured to input the planned advertising placement number into the first LSTM model, and output the predicted order conversion number output by the first LSTM model to the second LSTM model.
[0013] An electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the advertising placement effect prediction method is implemented.
[0014] A computer-readable storage medium for storing computer instructions, which when executed by a processor, complete the advertising placement effect prediction method.
[0015] The beneficial effects of the present invention are as follows: The main factor determining the value of the order conversion number is the advertising placement number. The present invention combines the advertising placement number Mnads , the advertising placement number Mn + mads , the order conversion number Mnocvr , the order conversion number Mn + mocvr to form a first training array, and form all the first training arrays into a first training set, and construct a first LSTM model based on the first training set, so that the first LSTM model eliminates a large number of parameters irrelevant to the order conversion number when predicting the order conversion number, thereby improving the accuracy of the first LSTM model in predicting the order conversion number.
[0016] Furthermore, based on the relatively large time difference between the n+m th time segment and the n th time segment, the present invention ensures that when the n+m th time segment just ends, the return number Mnocvr corresponding to the order conversion number MnprThe statistics are basically completed, thereby providing the second LSTM model based on the number of order conversions Mnocvr 、the number of order conversions Mn + mocvr 、the number of returns Mnpr for the feasibility of predicting the number of order conversions Mn + mocvr and the corresponding number of returns. In addition, the characteristic that the return rate fluctuates less on the time scale ensures that the second LSTM model has a high accuracy in predicting the return rate.
[0017] Merchants can conduct cost-profit accounting through the number of advertising placement plans, the predicted number of order conversions, and the predicted number of returns to determine whether to continue advertising in the future or adjust the advertising placement plan in a timely manner. Brief Description of the Drawings
[0018] Figure 1 is a flowchart of the advertising placement effect prediction method according to an embodiment of the present invention; Figure 2 is a block diagram of the advertising placement effect prediction system according to an embodiment of the present invention. Detailed Embodiments
[0019] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following detailed description of the specific embodiments of the present invention is made with reference to the accompanying drawings. Many specific details are set forth in the following description to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.
[0020] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, "a plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.
[0021] Embodiment: This embodiment provides an advertising placement effect prediction method, as Figure 1 shown, specifically including the following steps: Step 101: Divide the historical advertising placement time period into multiple time segments. Among them, the number of time segments is N , and the length of each time segment is T.
[0022] There is an advertisement placement behavior in each time segment. Through the advertisement click behavior, users can be converted into commodity orders. It is not difficult to understand that there may be some commodity orders that will eventually result in returns. In other words, each time segment has a corresponding advertisement placement number and order conversion number, and each order conversion number has a corresponding return number.
[0023] Specifically, the advertisement placement number n corresponding to the M n ads th time segment M n ocvr and the order conversion number M n ocvr corresponding to the order conversion number M n pr .
[0024] The advertisement placement number M n ads represents the total exposure times of the advertisement within the n th time segment. Correspondingly, M n ads / T represents the advertisement placement frequency within the n th time segment. The order conversion number M n ocvr represents the number of orders brought by users clicking on advertisements within the n th time segment. Generally, the more the advertisement placement number M n ads , the more the order conversion number M n ocvr . There is a positive correlation between the two.
[0025] Among them, after the user clicks on the advertisement and places an order, based on means such as URL parameter tracking and / or advertisement platform attribution tools, the seller can associate the order with the advertisement from which it originated, so as to achieve the statistics of the order conversion number M n ocvr . The above-mentioned means of tracing the order source are existing technologies and will not be elaborated in this embodiment.
[0026] It is not difficult to understand that based on the real-time nature of online ordering, the occurrence time of the order conversion number Mnocvr corresponding to the ordering behavior is generally also in the nWithin a time segment. Different from the order placement behavior, the return behavior often requires actions such as shipping, inspection, and return first. Therefore, there is often a large time difference between the return behavior and the order placement behavior. Correspondingly, the occurrence time of the return behavior often does not fall within the corresponding time segment. That is, the Mnpr statistical completion time of the return quantity often lags far behind the n th time segment, and the statistical behavior of the return quantity Mnpr cannot be completed within the n th time segment.
[0027] Step 102: Obtain the advertisement placement quantity and order conversion quantity corresponding to all time segments, and at the same time obtain the return quantity corresponding to all order conversion quantities.
[0028] Step 103: At least combine the advertisement placement quantity M n ads , the advertisement placement quantity M n+m ads , the order conversion quantity M n ocvr , the order conversion quantity M n+m ocvr to form a first training array. Correspondingly, the first training array X n = M n ads , M n+m ads , M n ocvr , M n+m ocvr , where m ≥2 and m is an integer.
[0029] Furthermore, combine all the first training arrays to form a first training set, that is, the first training set Z 1 =[X 1 , X 2 ,..., X N-m .
[0030] Step 104: Build a first LSTM model based on the first training set.
[0031] During the training process, the first LSTM model can be based on the advertisement placement quantity M n ads , the advertisement placement quantity M n+mads 、 Number of order conversions M n ocvr Predict the number of order conversions for the n+m th time segment, and the corresponding predicted value is .
[0032] Exemplarily, in this embodiment m= 14. In other words, in this embodiment, the first LSTM model can be based on the number of ad placements M n ads 、 Number of ad placements M n+14 ads 、 Number of order conversions M n ocvr Predict the number of order conversions for the n+ 14th time segment, and the corresponding predicted value is .
[0033] Correspondingly, the loss function constructed for the first LSTM model in this embodiment .
[0034] It is not difficult to understand that there is a strong correlation between the number of order conversions and the number of ad placements. Here, the meaning of strong correlation does not mean that the number of order conversions will change significantly with the change of the number of ad placements, but that the main factor determining the value of the number of order conversions is the number of ad placements. In other words, the first LSTM model mainly retains the number of ad placements during the training process and eliminates a large number of parameters unrelated to the number of order conversions, thereby improving the accuracy of the prediction of the number of order conversions by the final first LSTM model.
[0035] As the overall ad placement period lengthens, the marginal benefit decreases. In other words, the order conversion rate (number of order conversions / number of ad placements) usually gradually decreases over time. Thus, in the process of obtaining M n ads 、 Number of ad placements M n+14 ads and the number of order conversions M n ocvr acquire , the prediction of the change trend of the order conversion rate is implied, thereby further improving the accuracy of the prediction of the number of order conversions.
[0036] Step 105: At least the number of order conversions M n ocvr 、 Number of order conversions Mn+m ocvr and the number of returns M n pr and the number of returns M n+m pr are combined to form a second training array. Correspondingly, the second training array Y n = M n ocvr , M n+m ocvr , M n pr , M n+m pr .
[0037] Further, all the second training arrays are combined to form a second training set, that is, the second training set Z 2 = [Y 1 , Y 2 ,..., Y N-m .
[0038] Step 106: Construct a second LSTM model based on the second training set.
[0039] During the training process, the second LSTM model can be based on the order conversion number M n ocvr and the order conversion number M n+m ocvr and the number of returns M n pr to predict the number of returns corresponding to the order conversion number M n+m ocvr and the corresponding predicted value is .
[0040] Specifically, based on the condition m = 14 in the foregoing embodiment, in this embodiment, the second LSTM model can be based on the order conversion number M n ocvr and the order conversion number M n+14 ocvr and the number of returns M n pr to predict the number of returns corresponding to the order conversion number M n+14 ocvrPredict the corresponding number of returns, and the corresponding predicted value is .
[0041] Correspondingly, the loss function constructed for the second LSTM model in this embodiment .
[0042] Based on the characteristics of the aforementioned return number statistics, at the end of the n+m th time segment, it is impossible to immediately complete the statistics of the return number M n+m pr . If the value of the time segment length T is small, then it is very likely that the order conversion number M n+m-1 ocvr corresponding return number M n+m-1 pr has not been completed yet. The smaller the value of T, the greater the possibility that the return number M n+m-1 pr has not been completed. Based on this characteristic, it is often impossible to utilize the order conversion number M n+m-1 ocvr , order conversion number M n+m ocvr and return number M n+m-1 pr to predict the return number M n+m pr .
[0043] In view of the above situation, in this embodiment, first, the time segment length T is negatively correlated with m , that is, the smaller T is, the m value is larger, so as to ensure that there is enough time difference between the n+m th time segment and the n th time segment. For example, in this embodiment m = 14 is obtained based on T = 12h. Therefore, there is a 7-day difference between the n+m th time segment and the n th time segment. According to the existing logistics capabilities, 7 days is basically enough to complete the return process. Therefore, at the end of the n+m th time segment, the order conversion number M n ocvr corresponding return number M n pr has basically completed the statistics. Therefore, based on the order conversion number M nocvr 、 Number of order conversions M n+14 ocvr 、 Number of returns M n pr For the number of order conversions M n+14 ocvr Predicting the corresponding number of returns is feasible at the practical operation level.
[0044] Secondly, in the non-abnormal state, the return rate (number of returns / number of order conversions) fluctuates less on the time scale. Therefore, even if the time difference between the n+m th time segment and the n th time segment is large, the second LSTM model still has a high accuracy in predicting the return rate.
[0045] Step 107: The merchant can input the number of advertisements planned to be placed within a future time segment (with the same time length of T), for example, the N + m th time segment (i.e., the advertisement placement plan number) into the first LSTM model, so that the first LSTM model outputs the predicted number of order conversions corresponding to the advertisement placement plan number.
[0046] Step 108: The second LSTM model can receive the predicted number of order conversions output by the first LSTM model and output the corresponding predicted number of returns therefrom.
[0047] Step 109: The merchant can at least judge the advertisement placement effect based on the advertisement placement plan number, the predicted number of order conversions, and the predicted number of returns.
[0048] The advertisement placement cost can be calculated through the advertisement placement plan number, the loss cost can be calculated through the predicted number of returns, and the predicted number of order conversions can be used to predict the sales amount. Based on these three types of data, it can provide the merchant with a judgment on whether it is cost-effective to place advertisements within the N +1th time segment (the net profit has a sufficient value). If it is cost-effective, the advertisement placement of the advertisement placement plan number can be executed. If it is not cost-effective, the advertisement placement is no longer carried out, or the advertisement placement plan number is readjusted until the merchant judges that it is cost-effective to place advertisements within the N +1th time segment.
[0049] In some alternative embodiments, a conversion rate threshold and a return rate threshold can be set for the order conversion rate and the return rate respectively. Only when the ratio between the predicted number of order conversions and the advertisement placement plan number is greater than the conversion rate threshold, and the ratio between the predicted number of returns and the predicted number of order conversions is less than the return rate threshold, does it indicate that the subsequent advertisement placement has sufficient benefits, and thus the advertisement placement plan is executed.
[0050] Considering the real-time nature of the order conversion count, preferably, in some other preferred embodiments, the first training array further includes the number of ad placements M n+1 ads 、 M n+m-1 ads and the order conversion count M n+1 ocvr 、 M n+m-1 ocvr 。At this time, X n = M n ads , M n+1 ads , M n+m-1 ads , M n+m ads , M n ocvr , M n+1 ocvr , M n+m-1 ocvr , M n+m ocvr . Thus, the accuracy of the first LSTM model in predicting the order conversion count can be further improved. In this part of the embodiments, those skilled in the art can adaptively modify the loss function of the first LSTM model.
[0051] Figure 2 is a block diagram of an advertising placement effect prediction system shown in an exemplary embodiment. The system includes: A data acquisition module 100 for dividing the advertising placement historical time period into multiple time segments and retrieving the number of ad placements 、 the order conversion count, and the number of returns corresponding to the order conversion count; A first training module 200 for obtaining a first training array composed of the number of ad placements and the order conversion count, and forming a first training set from all the first training arrays, and constructing a first LSTM model based on the first training set; A second training module 300 for obtaining a second training array composed of the order conversion count and the number of returns, and forming a second training set from all the second training arrays, and constructing a second LSTM model based on the second training set; The data processing module 400 is configured to input the number of advertising placement plans into the first LSTM model and output the predicted number of order conversions output by the first LSTM model to the second LSTM model.
[0052] For the device in the embodiment, the specific manners in which each module performs operations have been described in detail in the embodiment related to the method, and will not be elaborated herein.
[0053] This embodiment also provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the advertising placement effect prediction method described above is implemented.
[0054] This embodiment also provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the advertising placement effect prediction method described above is completed.
[0055] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.
[0056] The above embodiments only represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the invention patent should be subject to the appended claims.
Claims
1. A method for predicting the effect of advertising, characterized in that: include: The advertising delivery history period is divided into multiple time segments, where the length of each time segment is T. n Number of ads served for each time segment M n ads and order conversions M n ocvr , order conversion number M n ocvr The corresponding number of returns M n pr ; Place ads at least M n ads , M n+m ads and order conversions M n ocvr , M n+m ocvr Combine to form a first training array, form all the first training arrays into a first training set, and build a first LSTM model based on the first training set, where m ≥2; At least the number of orders converted M n ocvr , M n+m ocvr and the number of returns M n pr , M n+m pr Combining to form a second training array, forming all the second training arrays into a second training set, and building a second LSTM model based on the second training set; The first LSTM model receives the number of advertising delivery plans and outputs the number of order conversion predictions, and the second LSTM model receives the number of order conversion predictions and outputs the number of returns predictions.
2. The method for predicting the effect of advertising delivery according to claim 1, characterized in that: Also includes: The advertising effect is judged based on at least the number of advertising plans, the number of order conversion forecasts, and the number of returns forecasts.
3. The method for predicting the effect of advertising delivery according to claim 2, characterized in that: Set the conversion rate threshold and return rate threshold. The advertising delivery plan will be executed only when the ratio between the predicted number of order conversions and the planned number of advertising is greater than the conversion rate threshold, and the ratio between the predicted number of returns and the predicted number of order conversions is less than the return rate threshold.
4. The method for predicting the effect of advertising delivery according to claim 1, characterized in that: The time segment length T is negatively correlated with m .
5. The method for predicting the effect of advertising delivery according to claim 3, characterized in that: The number of time slices is N , T=12h, m= 14. Loss function when building the first LSTM model ,in, For the first LSTM model, the number of advertisements M i+14 ads Output the predicted number of order conversions.
6. The method for predicting the effect of advertising delivery according to claim 4, characterized in that: Loss function when building the second LSTM model ,in, For the second LSTM model, the order conversion number M i+14 ocvr Output the predicted return rate.
7. The method for predicting the effect of advertising according to claim 1, characterized in that: The first training array also has the number of advertisement delivery M n+1 ads , M n+m-1 ads and order conversions M n+1 ocvr , M n+m-1 ocvr .
8. An advertising effect prediction system, characterized in that: include: The data acquisition module is used to divide the historical advertising time period into multiple time segments and retrieve the advertising data corresponding to each time segment. 、 The number of order conversions and the number of returns corresponding to the order conversions; A first training module is used to obtain the number of advertisements and the number of order conversions to form a first training array, and to form all the first training arrays into a first training set, and to build a first LSTM model based on the first training set; The second training module is used to obtain the number of order conversions and the number of returns to form a second training array, and to form all the second training arrays into a second training set, and to build a second LSTM model based on the second training set; The data processing module is used to input the number of advertising delivery plans into the first LSTM model, and output the order conversion prediction number output by the first LSTM model to the second LSTM model.
9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method for predicting the effect of advertising delivery according to any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the advertising delivery effect prediction method described in any one of claims 1 to 7.