Advertisement delivery effect prediction method, device, computer equipment and storage medium

Abstract

The present application relates to a method, device, computer equipment and storage medium for predicting the effect of advertising, the method comprising: identifying a target population package for advertising to be delivered; obtaining target feature information for predicting the target effect score of the target population package; inputting the target feature information into a prediction score model to obtain a target effect score, the target effect score is used to predict the advertising delivery effect when the advertisement to be delivered is delivered to the target population package; wherein the prediction score model is obtained by training sample feature information and sample effect score, the sample feature information is feature information for predicting the sample effect score of the sample population package, and the sample effect score is the effect score obtained when the advertisement to be delivered is delivered to the sample population package. The above method can predict the effect of advertising delivery before advertising delivery to improve the management efficiency of the population package for advertising delivery.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device, computer equipment and storage medium for predicting the effect of advertising delivery. Background Art

[0002] With the continuous development of the Internet and e-commerce, it is a very common advertising / marketing method to promote e-commerce crowd activities based on Internet big data. At present, the mainstream method is to select crowd packages for activities for advertising, and the actual effect of crowd package delivery can only be obtained after the activity is over. Therefore, the acquisition of advertising effect data is delayed, resulting in low management efficiency of crowd packages for advertising. Summary of the invention

[0003] Based on this, it is necessary to provide an advertising effect prediction method, device, computer equipment and storage medium that can predict the effect of advertising before advertising is released to improve the management efficiency of the advertising population package in response to the above technical problems.

[0004] A method for predicting the effect of advertising, comprising: identifying a target population package for advertising to be delivered; obtaining target feature information for predicting a target effect score for the target population package; inputting the target feature information into a prediction score model to obtain a target effect score, the target effect score being used to predict the advertising delivery effect when the advertisement to be delivered is delivered to the target population package; wherein the prediction score model is obtained by training sample feature information and sample effect score, the sample feature information is feature information for predicting a sample effect score for the sample population package, and the sample effect score is an effect score obtained when the advertisement to be delivered is delivered to the sample population package.

[0005] In one of the embodiments, a method for predicting the effect of advertising also includes: obtaining effect data when the advertisement to be placed is placed in a sample population package; determining a calculation formula for calculating the effect score; determining the sample effect score based on the effect data and the calculation formula; obtaining sample feature information; and training a prediction score model using the sample effect score and sample feature information.

[0006] In one embodiment, the effect data includes order conversion rate, click-through rate, return on investment and 1 / average click cost, and a method for predicting the effect of advertising placement also includes: obtaining a first weight, a second weight, a third weight and a fourth weight, the first weight being the weight corresponding to the order conversion rate, the second weight being the weight corresponding to the click-through rate, the third weight being the weight corresponding to the return on investment, and the fourth weight being the weight corresponding to 1 / average click cost; determining a sample effect score based on the effect data and a calculation formula, including: determining the sample effect score based on the order conversion rate and the first weight, the click-through rate and the second weight, the return on investment and the third weight, 1 / average click cost and the fourth weight, and a calculation formula.

[0007] In one embodiment, a method for predicting the effect of advertising placement further includes: obtaining an industry benchmark value; comparing the order conversion rate, click-through rate, return on investment, and 1 / average click cost with the industry benchmark value respectively; if any indicator value among the order conversion rate, click-through rate, return on investment, and 1 / average click cost is less than or equal to the industry benchmark value, then adopting Calculate the index score corresponding to any index value, where x * represents the indicator score, x represents the current value of any indicator value, min represents the minimum value of any indicator value, max represents the maximum value of any indicator value, and L represents the industry benchmark value; if any indicator value among order conversion rate, click-through rate, return on investment and 1 / average click cost is greater than the industry benchmark value, then the industry benchmark value is adopted. Calculate the indicator score corresponding to any indicator value; determine the sample effect score according to the order conversion rate and the first weight, the click-through rate and the second weight, the return on investment and the third weight, 1 / the average click cost and the fourth weight, and the calculation formula, including: determine the sample effect score according to the indicator score of the order conversion rate and the first weight, the indicator score of the click-through rate and the second weight, the indicator score of the return on investment and the third weight, 1 / the average click cost and the fourth weight, and the calculation formula.

[0008] In one embodiment, the sample characteristic information includes user label characteristic information of each user in the sample population package, behavior characteristic information corresponding to each user under specific parameters, shopping behavior characteristic information of each user on the target product, and shopping behavior characteristic information of each user on the competing products corresponding to the target product. The target product is the product for which the advertisement is to be placed.

[0009] In one embodiment, obtaining target feature information for predicting a target effect score of a target population package includes: identifying a candidate feature set of the target population package, the candidate feature set including multiple candidate features; filtering out a target feature from the multiple candidate features of the candidate feature set based on a feature selection algorithm of mutual information; and obtaining target feature information of the target feature.

[0010] In one of the embodiments, a target feature is selected from multiple candidate features in a candidate feature set according to a feature selection algorithm based on mutual information, including: selecting a candidate feature with the largest average mutual information from multiple candidate features in the candidate feature set according to an average mutual information algorithm in the feature selection algorithm of mutual information, and storing the candidate feature with the largest average mutual information as the first selected feature in the selected feature set; removing the first selected feature from the candidate feature set to obtain a sub-selected set; selecting a second selected feature from the sub-selected set according to a maximum correlation minimum redundancy algorithm in the feature selection algorithm of mutual information, and storing the second selected feature in the selected feature set, wherein the maximum correlation minimum redundancy algorithm is used to calculate the maximum correlation between each candidate feature in the sub-selected set and the sub-selected set and the minimum redundancy between each candidate feature and the selected feature set; and using the first selected feature and the second selected feature in the selected feature set as the target feature.

[0011] An effect prediction device for advertisement placement comprises: an identification module for identifying a target population package for advertisement placement; an acquisition module for acquiring target feature information for predicting a target effect score for the target population package; a prediction module for inputting the target feature information into a prediction score model to obtain a target effect score, wherein the target effect score is used to predict the advertisement placement effect when the advertisement to be placed is placed in the target population package; wherein the prediction score model is obtained by training sample feature information and sample effect score, wherein the sample feature information is feature information for predicting a sample effect score for the sample population package, and the sample effect score is an effect score obtained when the advertisement to be placed is placed in the sample population package.

[0012] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of any of the above-mentioned embodiments are implemented.

[0013] A computer-readable storage medium stores a computer program, which implements the steps of any of the above-mentioned method embodiments when the computer program is executed by a processor.

[0014] The above-mentioned advertising effect prediction method, device, computer equipment and storage medium identify the target population package of the advertisement to be placed, obtain the target feature information used to predict the target effect score of the target population package, input the target feature information into the prediction score model, obtain the target effect score, and the target effect score is used to predict the advertising effect when the advertisement to be placed is placed in the target population package; wherein the prediction score model is obtained by training sample feature information and sample effect score, the sample feature information is feature information used to predict the sample effect score of the sample population package, and the sample effect score is the effect score obtained when the advertisement to be placed is placed in the sample population package. Therefore, the prediction score model can predict the advertising effect when the advertisement to be placed is placed in the target population package, and the advertising effect can be obtained without waiting for the end of the advertising activity, so that the effect of advertising can be predicted before the advertisement is placed to improve the management efficiency of the population package for advertising. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A diagram of an application environment of an advertisement delivery effect prediction method in an embodiment;

[0016] Figure 2 A schematic diagram of a flow chart of a method for predicting the effect of advertisement delivery in one embodiment;

[0017] Figure 3 is a feature details diagram of user tag feature information in one embodiment;

[0018] Figure 4 A feature detail diagram of a user's behavior characteristics under a specific brand in one embodiment;

[0019] Figure 5 It is a flowchart of a method for predicting the effect of advertisement delivery in a specific embodiment;

[0020] Figure 6 It is a structural block diagram of an advertisement delivery effect prediction device in one embodiment;

[0021] Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0023] The present application provides a method for predicting the effect of advertising delivery, which is applied to Figure 1 In the application environment shown. Figure 1As shown, the server cluster 102 is used to implement an effect prediction method for advertising delivery of the present application. Among them, the database 104 stores relevant information of the target population package of the advertisement to be delivered, and the relevant information includes target feature information for predicting the target effect score of the target population package. When the server cluster 102 implements an effect prediction method for advertising delivery, the target population package of the advertisement to be delivered is identified, and the target feature information for predicting the target effect score of the target population package is obtained from the database 104, and the target feature information is input into the prediction score model to obtain the target effect score, which is used to predict the advertising delivery effect when the advertisement to be delivered is delivered to the target population package. Among them, the prediction score model is obtained by training sample feature information and sample effect score, and the sample feature information is feature information for predicting the sample effect score of the sample population package, and the sample effect score is the effect score obtained when the advertisement to be delivered is delivered to the sample population package. That is, the prediction score model is obtained by training the sample data set in advance and stored in the server cluster 102 or in the database 104. Therefore, the prediction score model can predict the advertising effect when the advertisement is delivered to the target population package, and the advertising effect can be obtained without waiting for the end of the advertising campaign. Therefore, the effect of advertising can be predicted before the advertisement is delivered to improve the management efficiency of the population package for advertising delivery.

[0024] In one embodiment, Figure 2 As shown, a method for predicting the effect of advertising delivery is provided, and the method is applied to Figure 1 Taking the server cluster 102 in the example as an example, the following steps are included:

[0025] S202, identifying the target group to which the advertisement is to be delivered.

[0026] In this embodiment, the database stores relevant information of multiple population packages, each of which contains multiple users. When the server cluster receives an instruction for predicting the effect of advertisement delivery of an advertisement to be delivered, it identifies the target population package of the advertisement to be delivered. The identification method may be: receiving identification information of the population package input by the user, obtaining identification information of the target population package matching the identification information of the population package from the database, and then identifying the target population package of the advertisement to be delivered according to the matching result.

[0027] S204, obtaining target feature information for predicting the target effect score of the target population package.

[0028] In this embodiment, the database stores multiple features of the target population package, wherein the multiple features include the target feature. The target feature information of the target feature is used to predict the target effect score of the target population package.

[0029] S206, input the target feature information into the prediction score model to obtain the target effect score, and the target effect score is used to predict the advertising effect when the advertisement to be delivered is delivered to the target population package; wherein the prediction score model is obtained by training the sample feature information and the sample effect score, the sample feature information is the feature information used to predict the sample effect score of the sample population package, and the sample effect score is the effect score obtained when the advertisement to be delivered is delivered to the sample population package.

[0030] In this embodiment, the server cluster pre-trains the prediction score model through sample feature information and sample effect score. The sample feature information is feature information used to predict the sample effect score of the sample population package, and the sample effect score is the effect score obtained when the advertisement to be placed is placed in the sample population package. When predicting the effect of the advertisement to be placed in the target population package, the server cluster obtains the trained prediction score model, and inputs the target feature information into the prediction score model to obtain the target effect score output by the prediction score model. Among them, the target effect score is used to predict the advertising effect when the advertisement to be placed is placed in the target population package. That is, the target effect score can be used to intuitively predict the advertising effect when the advertisement to be placed is placed in the target population package. The prediction score model can be a model of the support vector machine algorithm (SVM).

[0031] The above-mentioned method for predicting the effect of advertising delivery identifies the target population package of the advertisement to be delivered, obtains the target feature information used to predict the target effect score of the target population package, inputs the target feature information into the prediction score model, obtains the target effect score, and the target effect score is used to predict the advertising delivery effect when the advertisement to be delivered is delivered to the target population package; wherein the prediction score model is obtained by training sample feature information and sample effect score, the sample feature information is feature information used to predict the sample effect score of the sample population package, and the sample effect score is the effect score obtained when the advertisement to be delivered is delivered to the sample population package. Therefore, the prediction score model can predict the advertising delivery effect when the advertisement to be delivered is delivered to the target population package, and the advertising delivery effect can be obtained without waiting for the end of the advertising delivery activity, so that the effect of advertising delivery can be predicted before the advertisement is delivered to improve the management efficiency of the population package for advertising delivery.

[0032] In one embodiment, a method for predicting the effect of advertising delivery also includes a model training process for a prediction score model. The model training process for the prediction score model includes: obtaining effect data when the advertisement to be delivered is delivered to a sample population package; determining a calculation formula for calculating the effect score; determining the sample effect score according to the effect data and the calculation formula; obtaining sample feature information; and training the prediction score model according to the sample effect score and the sample feature information.

[0033] In this embodiment, a sample population package and a calculation formula for calculating the effect score are determined. The effect data when the advertisement to be placed is placed in the sample population package is collected from historical data, and the sample effect score of the sample population package is calculated by the calculation formula and the effect data. The sample effect score is used to characterize the advertising effect when the advertisement to be placed is placed in the sample population package in history. In addition, the sample feature information of the sample population package is obtained, and the sample feature information is used to predict the sample effect score of the sample population package. Then, the sample effect score and the sample feature information are input into the prediction score model to train the prediction score model. Therefore, the trained prediction score model can accurately predict the advertising effect of the advertisement to be placed in the target population package.

[0034] In one embodiment, the effect data includes order conversion rate, click-through rate, return on investment, and 1 / average click cost. Before the above step of determining the sample effect score according to the effect data and the calculation formula, it also includes: obtaining a first weight, a second weight, a third weight, and a fourth weight, the first weight being the weight corresponding to the order conversion rate, the second weight being the weight corresponding to the click-through rate, the third weight being the weight corresponding to the return on investment, and the fourth weight being the weight corresponding to 1 / average click cost. The above step of determining the sample effect score according to the effect data and the calculation formula includes: determining the sample effect score according to the order conversion rate and the first weight, the click-through rate and the second weight, the return on investment and the third weight, 1 / average click cost and the fourth weight, and the calculation formula.

[0035] In this embodiment, specific dimensions of the effect data include:

[0036] Order conversion rate = number of buyers / number of clicks;

[0037] Click rate = number of clicks / number of successful sends;

[0038] Return on investment (ROI) = payment amount / marketing cost;

[0039] Average cost per click = marketing cost / number of clicks.

[0040] Each dimension has a corresponding weight. For example, the first weight corresponding to the order conversion rate is 40%, the second weight corresponding to the click rate is 30%, the third weight corresponding to the return on investment ROI is 20%, and the fourth weight corresponding to the average click cost is 10%.

[0041] At this time, the sample effect score is determined according to the order conversion rate and the first weight, the click rate and the second weight, the return on investment and the third weight, 1 / average click cost and the fourth weight, and the calculation formula:

[0042] Sample effect score = order conversion rate score*40% + click-through rate score*30% + return on investment ROI score*20% + (1 / average click cost score)*10%.

[0043] In one embodiment, before the step of determining the sample effect score according to the order conversion rate and the first weight, the click-through rate and the second weight, the return on investment and the third weight, 1 / the average click cost and the fourth weight, and the calculation formula, the step further includes: obtaining the industry benchmark value; comparing the order conversion rate, the click-through rate, the return on investment and 1 / the average click cost with the industry benchmark value respectively; if any indicator value among the order conversion rate, the click-through rate, the return on investment and 1 / the average click cost is less than or equal to the industry benchmark value, then adopting Calculate the index score corresponding to any index value, where x * represents the indicator score, x represents the current value of any indicator value, min represents the minimum value of any indicator value, max represents the maximum value of any indicator value, and L represents the industry benchmark value; if any indicator value among order conversion rate, click-through rate, return on investment and 1 / average click cost is greater than the industry benchmark value, then the industry benchmark value is adopted. Calculate the index score corresponding to any index value. At this time, the step of determining the sample effect score according to the order conversion rate and the first weight, the click-through rate and the second weight, the return on investment and the third weight, 1 / the average click cost and the fourth weight, and the calculation formula includes: determining the sample effect score according to the index score of the order conversion rate and the first weight, the index score of the click-through rate and the second weight, the index score of the return on investment and the third weight, 1 / the average click cost and the fourth weight, and the calculation formula.

[0044] In this embodiment, it is necessary to calculate the indicator scores corresponding to the order conversion rate, click-through rate, return on investment, and 1 / average click cost, and then calculate the sample effect score based on the indicator scores and their corresponding weights and calculation formulas. Specifically, determine the industry benchmark value. The industry benchmark value can be the median of the four dimensional data of order conversion rate, click-through rate, return on investment ROI, and 1 / average click cost. Compare the order conversion rate, click-through rate, return on investment ROI, and 1 / average click cost with the industry benchmark values, respectively, and then determine the calculation method of the corresponding indicator score based on the comparison results. For example: compare the four indicator values ​​of order conversion rate, click-through rate, return on investment ROI, and 1 / average click cost with the industry benchmark values, respectively. If they are less than or equal to the industry benchmark values, the calculation formula for the indicator score of the indicator value is On the contrary, the calculation formula of the indicator score of this indicator value is When calculating the specific indicator score, max can be the maximum value of the indicator value, min can be the minimum value of the indicator value, and x can be the indicator value. When x is less than min, it is calculated according to min, and when x is greater than max, it is calculated according to max. Therefore, the accuracy of the sample effect score can be improved.

[0045] In one embodiment, the sample characteristic information includes user label characteristic information of each user in the sample population package, behavior characteristic information corresponding to each user under specific parameters, shopping behavior characteristic information of each user on the target product, and shopping behavior characteristic information of each user on competing products corresponding to the target product. The target product is the product for which the advertisement is to be placed.

[0046] In this embodiment, the user tag feature information includes fact tag information and rule tag information. The user tag feature information is obtained by tagging the user based on the user's personal basic information and the user's shopping behavior information. Figure 3 shown.

[0047] The behavior characteristic information corresponding to each user under specific parameters is the information of the user's shopping behavior characteristics under specific parameters. The behavior characteristic information corresponding to specific parameters includes characteristic information under the purchase dimension, characteristic information under the add to cart dimension, characteristic information under the collection dimension, characteristic information under the browsing dimension, characteristic information under the evaluation dimension, and characteristic information under the complaint dimension. The characteristic information under each dimension is as follows:

[0048] Purchase Dimensions:

[0049] Shopping interval (the interval between the most recent purchase and the last purchase);

[0050] Purchase days (365 days / 180 days / 90 days / 21 days / 14 days / 7 days);

[0051] Purchase amount (365 days / 180 days / 90 days / 21 days / 14 days / 7 days);

[0052] Purchase frequency (365 days / 180 days / 90 days / 21 days / 14 days / 7 days);

[0053] Average order value (365 days / 180 days / 90 days / 21 days / 14 days / 7 days);

[0054] Frequency of returns (365 days / 180 days / 90 days / 21 days / 14 days / 7 days);

[0055] Add to cart dimensions:

[0056] Number of additional purchases (90 days / 30 days / 15 days / 7 days / 3 days / 1 day);

[0057] Collection Dimensions:

[0058] Number of collections (90 days / 30 days / 15 days / 7 days / 3 days / 1 day);

[0059] Browsing dimensions:

[0060] Number of views (90 days / 30 days / 15 days / 7 days / 3 days / 1 day);

[0061] Number of views (90 days / 30 days / 15 days / 7 days / 3 days / 1 day);

[0062] Number of products viewed (90 days / 30 days / 15 days / 7 days / 3 days / 1 day);

[0063] Evaluation dimensions:

[0064] Average product rating star rating (365 days / 180 days / 90 days / 21 days / 14 days / 7 days);

[0065] Number of product reviews (365 days / 180 days / 90 days / 21 days / 14 days / 7 days);

[0066] Complaint dimensions:

[0067] Number of complaints (365 days / 180 days / 90 days / 21 days / 14 days / 7 days).

[0068] For information on the corresponding behavior characteristics under specific parameters, please refer to Figure 4 shown.

[0069] The shopping behavior characteristic information of each user on the target product and the shopping behavior characteristic information of each user on the target product's corresponding competing products are expressed as characteristic information of the user's shopping behavior on the target product and the target product's corresponding competing products. The shopping behavior characteristic information on the target product includes the shopping behavior characteristic information of each category on the target product, including but not limited to the following categories: air conditioners, refrigerators and washing machines, black electronics, digital products, computers, communications, small appliances, kitchen and bathroom appliances, department stores, personal care and household cleaning, mother and baby products, beauty products, imported health and fresh food, instant drinks, pharmacy and pharmacy divisions, grains, oils and snacks, alcohol, fresh food, home furnishings, extreme goods, home decoration, group purchases, etc. For example, the following are some shopping behavior characteristics on the target product:

[0070] Shopping interval (the time interval between the most recent purchase of a specific brand and the last purchase of the brand’s category);

[0071] Purchase days (90 days / 30 days / 14 days);

[0072] Purchase amount (90 days / 30 days / 14 days);

[0073] Purchase frequency (90 days / 30 days / 14 days).

[0074] The shopping behavior characteristic information on the competing products corresponding to the target product includes the behavior characteristics of the user on the competing brands / categories under the parameters such as the specific brand / category that needs to be predicted. The details are as follows:

[0075] Purchase Dimensions:

[0076] Shopping interval (the time interval between the most recent and last purchase);

[0077] Purchase days (90 days / 30 days / 14 days / 7 days);

[0078] Purchase amount (90 days / 30 days / 14 days / 7 days);

[0079] Purchase frequency (90 days / 30 days / 14 days / 7 days);

[0080] Average order value (90 days / 30 days / 14 days / 7 days);

[0081] Frequency of returns (90 days / 30 days / 14 days / 7 days);

[0082] Add to cart dimensions:

[0083] Number of additional purchases (30 days / 15 days / 7 days / 3 days / 1 day);

[0084] Collection Dimensions:

[0085] Number of collections (30 days / 15 days / 7 days / 3 days / 1 day);

[0086] Browsing dimensions:

[0087] Number of views (30 days / 15 days / 7 days / 3 days / 1 day);

[0088] Number of views (30 days / 15 days / 7 days / 3 days / 1 day);

[0089] Number of products viewed (30 days / 15 days / 7 days / 3 days / 1 day).

[0090] In one embodiment, the above-mentioned acquisition of target feature information for predicting the target effect score of the target population package includes: identifying a set of candidate features for the target population package, the set of candidate features containing multiple candidate features; filtering out the target feature from the multiple candidate features in the set of candidate features according to a feature selection algorithm based on mutual information; and acquiring the target feature information of the target feature.

[0091] In this embodiment, the feature selection algorithm of mutual information is an algorithm for feature selection based on mutual information between features. The feature selection algorithm of mutual information is used to select a target feature from multiple candidate features in a candidate feature set, and then obtain target feature information of the target feature. Therefore, the accuracy of the target feature information of the target feature can be improved.

[0092] In one embodiment, the feature selection algorithm based on mutual information selects a target feature from multiple candidate features in a candidate feature set, including: selecting a candidate feature with the largest average mutual information from multiple candidate features in the candidate feature set according to the average mutual information algorithm in the feature selection algorithm of mutual information, and storing the candidate feature with the largest average mutual information as the first selected feature in the selected feature set; removing the first selected feature from the candidate feature set to obtain a sub-selected set; selecting a second selected feature from the sub-selected set according to the maximum correlation minimum redundancy algorithm in the feature selection algorithm of mutual information, and storing the second selected feature in the selected feature set, wherein the maximum correlation minimum redundancy algorithm is used to calculate the maximum correlation between each candidate feature in the sub-selected set and the sub-selected set and the minimum redundancy between each candidate feature and the selected feature set; and using the first selected feature and the second selected feature in the selected feature set as the target feature.

[0093] In this embodiment, a to-be-selected feature set A and a selected feature set B are set. In the initial state, the selected feature set B is an empty set. The to-be-selected feature set A contains multiple to-be-selected features. The average mutual information of the multiple to-be-selected features is calculated, and the to-be-selected feature b1 with the largest average mutual information is screened out. The to-be-selected feature b1 is stored as the first selected feature in the selected feature set B. The screened to-be-selected feature b1 is eliminated from the to-be-selected feature set A, and a sub-to-be-selected set C is obtained after elimination. Furthermore, the second selected feature b2 is screened from the sub-to-be-selected set C according to the maximum relevance minimum redundancy algorithm in the feature selection algorithm of mutual information, and the second selected feature b2 is stored in the selected feature set B. Finally, the to-be-selected feature b1 and the second selected feature b2 in the selected feature set B are used as target features. Among them, the second selected feature b2 is one or more, and each second selected feature b2 is selected based on the maximum relevance minimum redundancy algorithm in the feature selection algorithm of mutual information.

[0094] The maximum correlation minimum redundancy algorithm is used to calculate the maximum correlation between each candidate feature in the sub-selected set and the sub-selected set, as well as the minimum redundancy between each candidate feature and the selected feature set. For example, the sub-selected set C contains the second selected feature b2, the candidate feature b3, the candidate feature b4, the candidate feature b5 and the candidate feature b6. The selected feature set B contains the candidate feature b1. The method of screening out the second selected feature b2 based on the maximum correlation minimum redundancy algorithm is: calculating the maximum correlation between each candidate feature in the second selected feature b2, the candidate feature b3, the candidate feature b4, the candidate feature b5 and the candidate feature b6 and the sub-selected set C, and calculating the minimum redundancy between each candidate feature in the second selected feature b2, the candidate feature b3, the candidate feature b4, the candidate feature b5 and the candidate feature b6 and the selected feature set B, screening out the candidate feature corresponding to the calculated maximum value, and the screened candidate feature is the second selected feature b2. The following is a specific explanation:

[0095] Encapsulate the feature selection algorithm based on mutual information for feature processing. The processing steps of the algorithm are as follows:

[0096] ① Calculate the average mutual information of each candidate feature in the candidate feature set, and then select the candidate feature with the largest average mutual information and add it to the selected feature set as the first selected feature, that is, the most important feature.

[0097] ② The next step is to select the nth feature (n≥2): the criterion for selecting the selected feature from the set of candidate features is: take the candidate feature corresponding to the maximum value of maximum correlation - minimum redundancy (the correlation between the candidate feature and the set of candidate features - the redundancy between the candidate feature and the set of selected features) as the nth selected feature, where the nth selected feature is the second selected feature mentioned above.

[0098] ③ When the redundancy between the nth feature to be selected and the selected features in the selected feature set is greater than the correlation between the nth feature to be selected and the selected features in the selected feature set (that is, when the maximum correlation-minimum redundancy is a negative number), the selection process of the feature to be selected ends, and at this time, the selected feature set is the optimal feature subset. The selected features in the optimal feature subset can be the selected target features.

[0099] In one embodiment, after the step of using the first selected feature and the second selected feature in the selected feature set as the target feature, the step further includes: determining the weight of each selected feature according to the value calculated based on the maximum correlation minimum redundancy algorithm corresponding to each selected feature in the first selected feature and the second selected feature. The step of inputting the target feature information into the prediction score model includes: inputting the weight of the first selected feature, the feature information of the first selected feature, the weight of the second selected feature, and the feature information of the second selected feature into the prediction score model.

[0100] In this embodiment, after the target features are screened out, the weights of each selected feature can be calculated based on the maximum relevance minimum redundancy algorithm, and then the feature information of each selected feature and its corresponding weight are input into the prediction score model to obtain the target effect score. It should be noted that in the model training process of the prediction score model, the sample feature information and its corresponding weight are processed in the same way, and then the sample feature information and its corresponding weight are used to train the prediction score model.

[0101] For example, the weight value of each selected feature is determined according to the maximum correlation-minimum redundancy value calculated in the above steps for each selected feature in the optimal feature subset. The specific method for determining the weight value is: assuming that there are m selected features in the optimal feature subset, the maximum correlation-minimum redundancy values ​​of the m selected features are F1 to Fm respectively, taking F1 as an example, the weight of F1 is F1 / (F1+…Fm).

[0102] Therefore, the accuracy of the target effect score can be improved, thereby improving the predicted effect of the advertisement to be delivered.

[0103] In one embodiment, before the above step of identifying the set of candidate features of the target population package, it also includes: obtaining multiple feature data of the target population package, performing data cleaning on the multiple feature data according to preset data cleaning rules, performing data conversion on the multiple feature data after data cleaning according to a preset data conversion method, and screening out multiple candidate features of the set of candidate features of the target population package from the multiple feature data after data conversion.

[0104] Specifically, the preset data cleaning rules include business layer data cleaning rules and data layer cleaning rules. The business layer cleaning rules are: eliminating the characteristic data of refund orders, eliminating the characteristic data of fake orders, eliminating the characteristic data of public membership cards, and eliminating the characteristic data of abnormal browsing data. Among them, when a membership code corresponds to a large number of orders, the business system determines it as a public membership card. When a public membership card or device corresponds to a large number of levels, the business system determines it as a suspected crawler user and sets it as the characteristic data of abnormal browsing data. Data layer cleaning rules: Different missing value filling methods need to be adopted according to the specific fields involved in the characteristics. Specific methods include but are not limited to nearest filling, average filling, linear filling, etc.

[0105] The preset data conversion methods include character-type and categorical feature conversion processing methods. For example, character-type and categorical feature conversion processing: character-type variables need to be converted into int and double types, and categorical variables need to be one-hot encoded.

[0106] Therefore, the accuracy of the target effect score can be improved, thereby improving the predicted effect of the advertisement to be delivered.

[0107] A specific embodiment is given below to describe in detail the method for predicting the effect of advertising placement described in the above embodiment. Figure 5 As shown:

[0108] Step 1: Calculate the scores of relevant crowd packages that have been used in the past, including: user input including but not limited to brand, category, product group, store, scene and other input parameters, the system retrieves the detailed data of all crowd packages that have been used with this parameter as input parameter and the usage effect data of the crowd package, and calculates the score of the crowd package that has been used based on the effect data of the crowd package and the score calculation formula.

[0109] Step 2: Define the feature set used to predict the score of a population package, including: the distribution of user label features in a population package, the shopping behavior of all users in a population package under specific parameters, the shopping behavior of all users in a population package on Suning, and the shopping behavior of all users in a population package under competing products. Among them, the user's shopping behavior includes but is not limited to purchase, add to cart, browse, click, search, favorite, evaluate, complain, etc.

[0110] The third step: data cleaning and feature selection. Data cleaning includes: processing of outliers and missing values ​​in the underlying data, conversion and calculation of numerical variables, character variables, and categorical variables. Encapsulate the mutual information-based feature selection algorithm to select features from the unselected feature set one by one to the selected feature set, and finally weight the selected features according to the average mutual information value.

[0111] Step 4: Input the feature values ​​and scores of the used population packages in the first step into the support vector machine algorithm (SVM) for model training and solidification, and then input the feature values ​​of the unused population packages into the solidified model to obtain the predicted score.

[0112] This application provides an advertising effect prediction method, which manages and uses crowd packages in a unified manner by predicting the effect scores of crowd packages. The advantages of this solution are as follows:

[0113] ① Manage the crowd packages in a unified manner according to their prediction results and improve the master data information of the crowd packages;

[0114] ② Improve the experience of the crowd package screening process and lower the screening threshold. Users can improve the actual delivery effect of the crowd package by screening the predicted high-quality crowd packages for activity delivery;

[0115] ③ By predicting the usage effect score of the crowd package, provide technical support for crowd package remarketing, improve the efficiency of crowd package use, and save computing resources.

[0116] ④ This feature selection algorithm selects the optimal feature subset based on the specific data situation, and avoids human factors interfering with the algorithm results. It is more scientific than the traditional method of manually setting the number of features in the optimal feature subset.

[0117] In addition, the present application provides a method for predicting the effect of advertising delivery, which solves the technical problems existing in traditional advertising delivery, such as:

[0118] 1. At present, crowd packages in the industry are mainly produced based on certain logic / algorithms and then used, and the actual use effect is often not good. To address this problem, this application proposes to first predict the effect (order conversion rate, click-through rate, ROI, average click cost) of the produced crowd packages before launching them, thereby improving the use effect of the crowd packages and reducing the use cost of the crowd packages.

[0119] 2. Since the initially extracted features are high-dimensional features, dimensionality reduction is often used in machine learning to solve this problem. The commonly used method in the industry is to use some algorithm logic to sort the features, and then manually determine the number of selected features or the proportion of selected features, which results in the final selection of features being greatly affected by human factors. To this end, this application proposes a maximum relevance-minimum redundancy algorithm based on mutual information for feature screening, and innovatively formulates corresponding feature selection quantity rules to obtain the optimal feature subset, and at the same time proposes to select features and determine feature weight values ​​according to the average mutual information value.

[0120] It should be understood that, although the steps in the flowchart are shown in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0121] The present application also provides an advertisement delivery effect prediction device, such as Figure 6As shown, the device includes an identification module 602, an acquisition module 604 and a prediction module 606. The identification module 602 is used to identify the target population package to be placed the advertisement; the acquisition module 604 is used to obtain the target feature information used to predict the target effect score of the target population package; the prediction module 606 is used to input the target feature information into the prediction score model to obtain the target effect score, and the target effect score is used to predict the advertisement placement effect when the advertisement to be placed is placed in the target population package; wherein the prediction score model is obtained by training the sample feature information and the sample effect score, the sample feature information is the feature information used to predict the sample effect score of the sample population package, and the sample effect score is the effect score obtained when the advertisement to be placed is placed in the sample population package.

[0122] In one embodiment, an effect prediction device for advertising delivery also includes a training module. The training module is used to obtain effect data when the advertisement to be delivered is delivered to a sample population package; determine a calculation formula for calculating the effect score; determine the sample effect score according to the effect data and the calculation formula; obtain sample feature information; and perform model training on the prediction score model through the sample effect score and the sample feature information.

[0123] In one embodiment, the effect data includes order conversion rate, click-through rate, return on investment, and 1 / average click cost. An effect prediction device for advertising placement also includes a weight acquisition module for acquiring a first weight, a second weight, a third weight, and a fourth weight, wherein the first weight is the weight corresponding to the order conversion rate, the second weight is the weight corresponding to the click-through rate, the third weight is the weight corresponding to the return on investment, and the fourth weight is the weight corresponding to the 1 / average click cost; the above-mentioned determination of the sample effect score based on the effect data and the calculation formula includes: determining the sample effect score based on the order conversion rate and the first weight, the click-through rate and the second weight, the return on investment and the third weight, 1 / average click cost and the fourth weight, and the calculation formula.

[0124] In one embodiment, an advertising effect prediction device further includes a calculation module. The calculation module is used to obtain industry benchmark values; respectively compare order conversion rate, click-through rate, return on investment, and 1 / average click cost with the industry benchmark values; if any indicator value among order conversion rate, click-through rate, return on investment, and 1 / average click cost is less than or equal to the industry benchmark value, then adopt Calculate the index score corresponding to any index value, where x * represents the indicator score, x represents the current value of any indicator value, min represents the minimum value of any indicator value, max represents the maximum value of any indicator value, and L represents the industry benchmark value; if any indicator value among order conversion rate, click-through rate, return on investment and 1 / average click cost is greater than the industry benchmark value, then the industry benchmark value is adopted. Calculate the indicator score corresponding to any indicator value. The above-mentioned determination of the sample effect score based on the order conversion rate and the first weight, the click-through rate and the second weight, the return on investment and the third weight, 1 / the average click cost and the fourth weight, and the calculation formula includes: determining the sample effect score based on the indicator score of the order conversion rate and the first weight, the indicator score of the click-through rate and the second weight, the indicator score of the return on investment and the third weight, the indicator score of 1 / the average click cost and the fourth weight, and the calculation formula.

[0125] In one embodiment, the sample characteristic information includes user label characteristic information of each user in the sample population package, behavior characteristic information corresponding to each user under specific parameters, shopping behavior characteristic information of each user on the target product, and shopping behavior characteristic information of each user on the competing products corresponding to the target product. The target product is the product for which the advertisement is to be placed.

[0126] In one of the embodiments, the above-mentioned acquisition of target feature information for predicting the target effect score of the target population package includes: identifying a set of candidate features of the target population package, the set of candidate features containing multiple candidate features; filtering out the target feature from the multiple candidate features of the set of candidate features according to a feature selection algorithm of mutual information; and acquiring the target feature information of the target feature.

[0127] In one embodiment, the feature selection algorithm based on mutual information selects a target feature from multiple candidate features in a candidate feature set, including: selecting a candidate feature with the largest average mutual information from multiple candidate features in the candidate feature set according to the average mutual information algorithm in the feature selection algorithm of mutual information, and storing the candidate feature with the largest average mutual information as the first selected feature in the selected feature set; removing the first selected feature from the candidate feature set to obtain a sub-selected set; selecting a second selected feature from the sub-selected set according to the maximum correlation minimum redundancy algorithm in the feature selection algorithm of mutual information, and storing the second selected feature in the selected feature set, wherein the maximum correlation minimum redundancy algorithm is used to calculate the maximum correlation between each candidate feature in the sub-selected set and the sub-selected set and the minimum redundancy between each candidate feature and the selected feature set; and using the first selected feature and the second selected feature in the selected feature set as the target feature.

[0128] The specific definition of the effect prediction device for advertising placement can be found in the definition of the effect prediction method for advertising placement mentioned above, which will not be repeated here. Each module in the above-mentioned effect prediction device for advertising placement can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0129] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to connect to an external terminal to receive configuration information of the external terminal. When the computer program is executed by the processor, a method for predicting the effect of advertising is implemented.

[0130] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0131] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program: identifying a target population package to be delivered an advertisement; obtaining target feature information for predicting a target effect score for the target population package; inputting the target feature information into a prediction score model to obtain a target effect score, wherein the target effect score is used to predict the effect of delivering the advertisement when the advertisement to be delivered is delivered to the target population package; wherein the prediction score model is obtained by training sample feature information and sample effect score, wherein the sample feature information is feature information for predicting a sample effect score for the sample population package, and the sample effect score is an effect score obtained when the advertisement to be delivered is delivered to the sample population package.

[0132] In one embodiment, the processor implements the following steps when executing the computer program: obtaining effect data when the advertisement to be placed is placed in a sample population package; determining a calculation formula for calculating the effect score; determining the sample effect score based on the effect data and the calculation formula; obtaining sample feature information; and training the prediction score model using the sample effect score and the sample feature information.

[0133] In one embodiment, the effect data includes order conversion rate, click-through rate, return on investment, and 1 / average click cost. When the processor executes the computer program, the following steps are implemented: obtaining the first weight, the second weight, the third weight, and the fourth weight, the first weight being the weight corresponding to the order conversion rate, the second weight being the weight corresponding to the click-through rate, the third weight being the weight corresponding to the return on investment, and the fourth weight being the weight corresponding to the 1 / average click cost; when the processor executes the computer program to implement the above-mentioned step of determining the sample effect score according to the effect data and the calculation formula, the following steps are specifically implemented: determining the sample effect score according to the order conversion rate and the first weight, the click-through rate and the second weight, the return on investment and the third weight, 1 / average click cost and the fourth weight, and the calculation formula.

[0134] In one embodiment, when the processor executes the computer program, the following steps are implemented: obtaining industry benchmark values; comparing the order conversion rate, click-through rate, return on investment, and 1 / average click cost with the industry benchmark values ​​respectively; if any indicator value among the order conversion rate, click-through rate, return on investment, and 1 / average click cost is less than or equal to the industry benchmark value, adopting Calculate the index score corresponding to any index value, where x * represents the indicator score, x represents the current value of any indicator value, min represents the minimum value of any indicator value, max represents the maximum value of any indicator value, and L represents the industry benchmark value; if any indicator value among order conversion rate, click-through rate, return on investment and 1 / average click cost is greater than the industry benchmark value, then the industry benchmark value is adopted. Calculate the index score corresponding to any index value. When the processor executes the computer program to implement the above-mentioned step of determining the sample effect score according to the order conversion rate and the first weight, the click-through rate and the second weight, the return on investment and the third weight, 1 / the average click cost and the fourth weight, and the calculation formula, the following steps are specifically implemented: determine the sample effect score according to the index score and the first weight of the order conversion rate, the index score and the second weight of the click-through rate, the index score and the third weight of the return on investment, the index score and the fourth weight of 1 / the average click cost, and the calculation formula.

[0135] In one embodiment, the sample characteristic information includes user label characteristic information of each user in the sample population package, behavior characteristic information corresponding to each user under specific parameters, shopping behavior characteristic information of each user on the target product, and shopping behavior characteristic information of each user on the competing products corresponding to the target product. The target product is the product for which the advertisement is to be placed.

[0136] In one of the embodiments, when the processor executes a computer program to implement the above-mentioned step of obtaining target feature information for predicting the target effect score of the target population package, the following steps are specifically implemented: identifying a set of candidate features for the target population package, the set of candidate features containing multiple candidate features; filtering out a target feature from the multiple candidate features in the set of candidate features according to a feature selection algorithm of mutual information; and obtaining target feature information of the target feature.

[0137] In one of the embodiments, when the processor executes a computer program to implement the above-mentioned step of selecting a target feature from multiple candidate features in a candidate feature set according to a feature selection algorithm based on mutual information, the following steps are specifically implemented: selecting a candidate feature with the largest average mutual information from multiple candidate features in a candidate feature set according to an average mutual information algorithm in the feature selection algorithm of mutual information, and storing the candidate feature with the largest average mutual information as the first selected feature in the selected feature set; removing the first selected feature from the candidate feature set to obtain a sub-selected set; selecting a second selected feature from the sub-selected set according to a maximum correlation minimum redundancy algorithm in the feature selection algorithm of mutual information, and storing the second selected feature in the selected feature set, wherein the maximum correlation minimum redundancy algorithm is used to calculate the maximum correlation between each candidate feature in the sub-selected set and the sub-selected set and the minimum redundancy between each candidate feature and the selected feature set; and using the first selected feature and the second selected feature in the selected feature set as the target feature.

[0138] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented: identifying a target population package for an advertisement to be delivered; obtaining target feature information for predicting a target effect score for the target population package; inputting the target feature information into a prediction score model to obtain a target effect score, and the target effect score is used to predict the effect of the advertisement delivery when the advertisement to be delivered is delivered to the target population package; wherein the prediction score model is obtained by training sample feature information and sample effect score, the sample feature information is feature information for predicting a sample effect score of the sample population package, and the sample effect score is an effect score obtained when the advertisement to be delivered is delivered to the sample population package.

[0139] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: obtaining effect data when the advertisement to be delivered is delivered to a sample population package; determining a calculation formula for calculating the effect score; determining the sample effect score based on the effect data and the calculation formula; obtaining sample feature information; and training the prediction score model using the sample effect score and the sample feature information.

[0140] In one embodiment, the effect data includes order conversion rate, click-through rate, return on investment, and 1 / average click cost. When the computer program is executed by the processor, the following steps are implemented: obtaining the first weight, the second weight, the third weight, and the fourth weight, the first weight being the weight corresponding to the order conversion rate, the second weight being the weight corresponding to the click-through rate, the third weight being the weight corresponding to the return on investment, and the fourth weight being the weight corresponding to the 1 / average click cost; when the computer program is executed by the processor to implement the above-mentioned step of determining the sample effect score according to the effect data and the calculation formula, the following steps are specifically implemented: determining the sample effect score according to the order conversion rate and the first weight, the click-through rate and the second weight, the return on investment and the third weight, 1 / average click cost and the fourth weight, and the calculation formula.

[0141] In one embodiment, when the computer program is executed by a processor, the following steps are implemented: obtaining an industry benchmark value; comparing the order conversion rate, click-through rate, return on investment, and 1 / average click cost with the industry benchmark value; if any indicator value among the order conversion rate, click-through rate, return on investment, and 1 / average click cost is less than or equal to the industry benchmark value, adopting Calculate the index score corresponding to any index value, where x * represents the indicator score, x represents the current value of any indicator value, min represents the minimum value of any indicator value, max represents the maximum value of any indicator value, and L represents the industry benchmark value; if any indicator value among order conversion rate, click-through rate, return on investment and 1 / average click cost is greater than the industry benchmark value, then the industry benchmark value is adopted. Calculate the index score corresponding to any index value. When the computer program is executed by the processor to implement the above-mentioned step of determining the sample effect score according to the order conversion rate and the first weight, the click-through rate and the second weight, the return on investment and the third weight, 1 / the average click cost and the fourth weight, and the calculation formula, the following steps are specifically implemented: determine the sample effect score according to the index score and the first weight of the order conversion rate, the index score and the second weight of the click-through rate, the index score and the third weight of the return on investment, the index score and the fourth weight of 1 / the average click cost, and the calculation formula.

[0142] In one embodiment, the sample characteristic information includes user label characteristic information of each user in the sample population package, behavior characteristic information corresponding to each user under specific parameters, shopping behavior characteristic information of each user on the target product, and shopping behavior characteristic information of each user on the competing products corresponding to the target product. The target product is the product for which the advertisement is to be placed.

[0143] In one of the embodiments, when the computer program is executed by a processor to implement the above-mentioned step of obtaining target feature information for predicting the target effect score of the target population package, the following steps are specifically implemented: identifying a set of candidate features for the target population package, the set of candidate features containing multiple candidate features; filtering out a target feature from the multiple candidate features in the set of candidate features based on a feature selection algorithm of mutual information; and obtaining target feature information of the target feature.

[0144] In one of the embodiments, when the computer program is executed by the processor to implement the above-mentioned step of selecting the target feature from multiple candidate features in the candidate feature set according to the feature selection algorithm based on mutual information, the following steps are specifically implemented: selecting the candidate feature with the largest average mutual information from multiple candidate features in the candidate feature set according to the average mutual information algorithm in the feature selection algorithm of mutual information, and storing the candidate feature with the largest average mutual information as the first selected feature in the selected feature set; eliminating the first selected feature from the candidate feature set to obtain a sub-selected set; selecting the second selected feature from the sub-selected set according to the maximum correlation minimum redundancy algorithm in the feature selection algorithm of mutual information, and storing the second selected feature in the selected feature set, wherein the maximum correlation minimum redundancy algorithm is used to calculate the maximum correlation between each candidate feature in the sub-selected set and the sub-selected set and the minimum redundancy between each candidate feature and the selected feature set; and using the first selected feature and the second selected feature in the selected feature set as the target features.

[0145] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0146] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, 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, they should be considered to be within the scope of this specification.

[0147] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for predicting the effect of advertising delivery, the method include: Identify target groups for advertising; Acquire target feature information for predicting a target effect score of the target population package; Inputting the target feature information into a prediction score model to obtain the target effect score, wherein the target effect score is used to predict the advertising effect when the advertisement to be placed is placed to the target population package; The prediction score model is obtained by training sample feature information and sample effect score, wherein the sample feature information is feature information used to predict the sample effect score of the sample population package, and the sample effect score is the effect score obtained when the advertisement to be placed is placed in the sample population package; wherein the model training of the prediction score model includes the following steps: Acquire effect data when the advertisement to be placed is placed on the sample population package, the effect data including order conversion rate, click-through rate, return on investment, and 1 / average click cost; Determine the calculation formula used to calculate the effect score; Obtain a first weight, a second weight, a third weight, and a fourth weight, wherein the first weight is a weight corresponding to the order conversion rate, the second weight is a weight corresponding to the click rate, the third weight is a weight corresponding to the return on investment, and the fourth weight is a weight corresponding to 1 / average click cost; Obtain industry benchmark values; Compare the order conversion rate, the click-through rate, the return on investment, and the 1 / average click cost with the industry benchmark value respectively; If any of the order conversion rate, click-through rate, return on investment, and 1 / average click cost is less than or equal to the industry benchmark value, then Calculate the index score corresponding to any of the index values, where x * represents the index score, x represents the current value of any index value, min represents the minimum value of any index value, max represents the maximum value of any index value, and L represents the industry benchmark value; If any of the order conversion rate, click-through rate, return on investment, and 1 / average click cost is greater than the industry benchmark value, then Calculate the index score corresponding to any of the index values; Determine the sample effect score according to the index score of the order conversion rate and the first weight, the index score of the click-through rate and the second weight, the index score of the return on investment and the third weight, the index score of 1 / average click cost and the fourth weight, and the calculation formula; Acquire the sample feature information, the sample feature information including user label feature information of each user in the sample population package, behavior feature information corresponding to each user under specific parameters, shopping behavior feature information of each user on a target product, and shopping behavior feature information of each user on a competing product corresponding to the target product, wherein the target product is a product for which the advertisement to be placed is placed; The prediction score model is trained using the sample effect score and the sample feature information.

2. The method according to claim 1, It is characterized in that The acquiring of target feature information for predicting the target effect score of the target population package includes: Identifying a set of candidate features of the target population package, wherein the set of candidate features includes a plurality of candidate features; Filtering out a target feature from a plurality of candidate features in the candidate feature set according to a feature selection algorithm of mutual information; Obtain target feature information of the target feature.

3. The method according to claim 2, It is characterized in that The feature selection algorithm based on mutual information selects a target feature from a plurality of candidate features in the candidate feature set, including: According to the average mutual information algorithm in the feature selection algorithm of the mutual information, a candidate feature with the largest average mutual information is selected from multiple candidate features in the candidate feature set, and the candidate feature with the largest average mutual information is stored as the first selected feature in the selected feature set; Eliminating the first selected feature from the set of to-be-selected features to obtain a sub-set of to-be-selected features; Selecting a second selected feature from the sub-selected set according to the maximum relevance minimum redundancy algorithm in the feature selection algorithm of the mutual information, and storing the second selected feature in the selected feature set, wherein the maximum relevance minimum redundancy algorithm is used to calculate the maximum relevance of each selected feature in the sub-selected set and the sub-selected set, and the minimum redundancy of each selected feature and the selected feature set; The first selected feature and the second selected feature in the selected feature set are used as the target features.

4. An advertising effect prediction device, It is characterized in that The device comprises: An identification module, used to identify the target group package to be advertised; An acquisition module, used for acquiring target feature information for predicting a target effect score of the target population package; A prediction module, used for inputting the target feature information into a prediction score model to obtain the target effect score, wherein the target effect score is used to predict the advertising effect when the advertisement to be placed is placed to the target population package; The prediction score model is obtained by training sample feature information and sample effect score, wherein the sample feature information is feature information used to predict the sample effect score of the sample population package, and the sample effect score is the effect score obtained when the advertisement to be placed is placed in the sample population package; wherein the model training of the prediction score model includes the following steps: Acquire effect data when the advertisement to be placed is placed on the sample population package, the effect data including order conversion rate, click-through rate, return on investment, and 1 / average click cost; Determine the calculation formula used to calculate the effect score; Obtain a first weight, a second weight, a third weight, and a fourth weight, wherein the first weight is a weight corresponding to the order conversion rate, the second weight is a weight corresponding to the click rate, the third weight is a weight corresponding to the return on investment, and the fourth weight is a weight corresponding to 1 / average click cost; Obtain industry benchmark values; Compare the order conversion rate, the click-through rate, the return on investment, and the 1 / average click cost with the industry benchmark value respectively; If any of the order conversion rate, click-through rate, return on investment, and 1 / average click cost is less than or equal to the industry benchmark value, then Calculate the index score corresponding to any of the index values, where x * represents the index score, x represents the current value of any index value, min represents the minimum value of any index value, max represents the maximum value of any index value, and L represents the industry benchmark value; If any of the order conversion rate, click-through rate, return on investment, and 1 / average click cost is greater than the industry benchmark value, then Calculate the index score corresponding to any of the index values; Determine the sample effect score according to the index score of the order conversion rate and the first weight, the index score of the click-through rate and the second weight, the index score of the return on investment and the third weight, the index score of 1 / average click cost and the fourth weight, and the calculation formula; Acquire the sample feature information, the sample feature information including user label feature information of each user in the sample population package, behavior feature information corresponding to each user under specific parameters, shopping behavior feature information of each user on a target product, and shopping behavior feature information of each user on a competing product corresponding to the target product, wherein the target product is a product for which the advertisement to be placed is placed; The prediction score model is trained using the sample effect score and the sample feature information.

5. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the steps of the method according to any one of claims 1 to 3 are implemented.

6. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

Images