Cost control method, device and equipment for recommendation information, computer readable storage medium and computer program product

Abstract

The invention discloses a recommendation information cost control method, device and equipment, a computer readable storage medium and a computer program product, and the method comprises the steps: determining a classification parameter corresponding to initial to-be-recommended information based on an effective time period corresponding to the initial to-be-recommended information; determining the information category of the initial to-be-recommended information according to the basic classification parameters; based on the information category, the first target feature of the user, the second target feature of the initial to-be-recommended information and the target historical interaction feature of the user and the initial to-be-recommended information, processing the initial to-be-recommended information to obtain processed to-be-recommended information; based on the processed to-be-recommended information and a target conversion rate estimation model, determining an estimated conversion rate of the user for the processed to-be-recommended information; determining a target probability based on the target budget, the estimated conversion rate, the cost of the processed to-be-recommended information and a target variable, wherein the target variable is a defined probability variable for recommending each piece of initial to-be-recommended information to a user; and determining target to-be-recommended information to the user based on the plurality of target probabilities.

Description

Technical Field

[0001] The present application relates to the field of e-commerce technology, and in particular to a cost control method, apparatus, device, computer-readable storage medium, and computer program product for recommendation information. Background Art

[0002] In the field of modern marketing, recommendation systems have become an important tool for improving user experience and increasing sales. The recommendation algorithms used by recommendation systems usually provide users with personalized product or service recommendations by analyzing users' historical behaviors, preferences, and contextual information. However, during the implementation process, the inventors found that there are at least the following problems in the existing technology: Although traditional recommendation system algorithms pay great attention to the relevance of recommendations and their improvement of user satisfaction, they do not consider the cost control inherent in marketing activities. As a result, information promotion may be carried out without appropriate cost optimization measures, resulting in high recommendation costs. Summary of the Invention

[0003] In order to solve the above technical problems, the embodiments of the present application hope to provide a cost control method, device, equipment, computer-readable storage medium and computer program product for recommendation information, which solves the problem that the existing technology does not take into account the cost control inherent in marketing activities, resulting in high recommendation costs.

[0004] The technical solution of this application is achieved as follows:

[0005] Determining a classification parameter corresponding to each piece of initial information to be recommended based on a valid time period corresponding to each piece of initial information to be recommended;

[0006] Determining the information category of each of the initial information to be recommended based on each of the classification parameters;

[0007] Based on the information category, the first target feature of the user corresponding to each of the initial information to be recommended, the second target feature of each of the initial information to be recommended, and the target historical interaction feature between the user and each of the initial information to be recommended, each of the initial information to be recommended is processed to obtain each processed information to be recommended;

[0008] Determining an estimated conversion rate for each user for each processed piece of information to be recommended based on the plurality of processed pieces of information to be recommended and a target conversion rate estimation model;

[0009] Determining a target probability of recommending each of the initial information to be recommended to the user based on the target budget, each of the estimated conversion rates, the cost of each of the processed information to be recommended, and each target variable; wherein the target variable is a defined probability variable for recommending each of the initial information to be recommended to the user;

[0010] Based on the multiple target probabilities, target information to be recommended is determined from the multiple initial information to be recommended and recommended to the user.

[0011] In the above solution, determining the classification parameter corresponding to each initial information to be recommended based on the valid time period corresponding to each initial information to be recommended includes:

[0012] For each piece of initial information to be recommended, if the valid time period is less than or equal to the target time period, determining the valid time in the valid time period as the classification parameter;

[0013] If the effective time period is greater than the target time period, determining the cumulative conversion ratio of the plurality of initial information to be recommended in each sub-time period of the effective time period;

[0014] Based on the multiple cumulative conversion proportions, a target sub-time period is determined from the multiple sub-time periods as the classification parameter.

[0015] In the above solution, determining the information category of each of the initial information to be recommended based on each of the classification parameters includes:

[0016] For each of the initial information to be recommended, if the initial information to be recommended is converted before the classification parameter, determining that the information category is a positive sample;

[0017] If the initial information to be recommended has not been converted before the classification parameter, it is determined that the information category is a negative sample.

[0018] In the above solution, before processing each of the initial information to be recommended to obtain each processed information to be recommended, the method further includes:

[0019] Obtaining a first initial feature of each of the users, a second initial feature of each of the initial information to be recommended, and an initial historical interaction feature between the user and each of the initial information to be recommended;

[0020] Performing feature screening on each of the first initial features, each of the second initial features, and each of the initial historical interaction features to obtain each filtered first initial feature, each filtered second initial feature, and each filtered initial historical interaction feature;

[0021] Feature preprocessing is performed on each of the filtered first initial features, each of the filtered second initial features, and each of the filtered initial historical interaction features to obtain each of the first target features, each of the second target features, and each of the target historical interaction features.

[0022] In the above solution, before determining the target probability of recommending each of the initial to-be-recommended information to the user, the method further includes:

[0023] determining an average estimated conversion rate based on the plurality of estimated conversion rates and the amount of the processed information to be recommended;

[0024] The target budget is determined based on the number of users, the average cost of the processed information to be recommended, and the average estimated conversion rate.

[0025] In the above solution, determining the target probability of recommending each of the initial information to be recommended to the user based on the target budget, each of the estimated conversion rates, the cost of each of the processed information to be recommended, and each target variable includes:

[0026] constructing an objective function for the estimated conversion rate based on each of the estimated conversion rates and each of the target variables;

[0027] constructing a first constraint condition on the cost of the initial information to be recommended based on each of the estimated conversion rates, each of the target variables, each of the costs, and the target budget;

[0028] For any of the users, constructing a second constraint condition on the amount of the initial information to be recommended based on each of the target variables and the target constant;

[0029] Under the constraints of the first constraint condition and the second constraint condition, a target probability of recommending each piece of the initial to-be-recommended information to the user is determined when the objective function reaches a target value.

[0030] In the above solution, determining the target probability of recommending each piece of the initial to-be-recommended information to the user when the objective function reaches a target value under the constraints of the first constraint condition and the second constraint condition includes:

[0031] Obtaining a Lagrangian relaxation function for estimating the conversion rate using a Lagrangian multiplier, the first constraint, and the objective function;

[0032] Under the constraint of the second constraint condition, a target probability of recommending each piece of the initial to-be-recommended information to the user is determined when the Lagrangian relaxation function reaches a target value.

[0033] A device for controlling the cost of recommendation information, comprising:

[0034] A first processing unit is configured to determine a classification parameter corresponding to each piece of initial information to be recommended based on a valid time period corresponding to each piece of initial information to be recommended;

[0035] The first processing unit is further configured to determine an information category of each of the initial information to be recommended based on each of the classification parameters;

[0036] The first processing unit is further configured to process each of the initial information to be recommended to obtain each processed information to be recommended based on the information category, the first target feature of the user corresponding to each of the initial information to be recommended, the second target feature of each of the initial information to be recommended, and the target historical interaction feature between the user and each of the initial information to be recommended;

[0037] a first determining unit, configured to determine, based on the plurality of processed information to be recommended and a target conversion rate estimation model, an estimated conversion rate for each user for each processed information to be recommended;

[0038] a second processing unit, configured to determine a target probability of recommending each of the initial information to be recommended to the user based on a target budget, each of the estimated conversion rates, the cost of each of the processed information to be recommended, and each target variable; wherein the target variable is a defined probability variable for recommending each of the initial information to be recommended to the user;

[0039] The second determining unit is configured to determine target information to be recommended from the plurality of initial information to be recommended based on the plurality of target probabilities and recommend the target information to the user.

[0040] A cost control device for recommendation information, the device comprising: a processor, a memory, and a communication bus;

[0041] The communication bus is used to realize the communication connection between the processor and the memory;

[0042] The processor is used to execute the cost control program of the recommendation information in the memory to implement the steps of the cost control method of the recommendation information.

[0043] A computer-readable storage medium stores one or more programs, which can be executed by one or more processors to implement the steps of the above-mentioned cost control method for recommendation information.

[0044] A computer program product includes a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for cost control of recommendation information.

[0045] Because based on the effective time period corresponding to each initial information to be recommended, the classification parameters corresponding to each initial information to be recommended are determined, and then based on each classification parameter, the information category of each initial information to be recommended is determined, and then based on the information category, the first target feature of the user corresponding to each initial information to be recommended, the second target feature of each initial information to be recommended and the target historical interaction feature between the user and each initial information to be recommended, each initial information to be recommended is processed to obtain each processed information to be recommended, and then based on multiple processed information to be recommended and the target conversion rate estimation model, the estimated conversion rate of each user for each processed information to be recommended is determined. In this way, the information category of each initial information to be recommended is determined by the classification parameters determined by the effective time period, so that the initial information to be recommended determined based on the classification parameters is The information category of the information to be recommended is more accurate, so that an unbiased sample can be obtained, which makes the estimated conversion rate obtained by the unbiased sample and the target estimated conversion rate model more accurate. In addition, based on the target budget, each estimated conversion rate, the cost of each processed information to be recommended and the defined probability variable of recommending each initial information to be recommended to the user, the target probability of recommending each initial information to be recommended to the user is determined. Then, based on multiple target probabilities, the target information to be recommended is determined from multiple initial information to be recommended to the user. In this way, when recommending information to be recommended, the cost of each processed information to be recommended and the target budget are used, that is, the cost of recommendation is taken into account when recommending information, instead of recommending without considering the cost of recommendation as in the related art, thereby overcoming the problem of high recommendation cost in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A flowchart of a cost control method for recommendation information provided in an embodiment of the present application;

[0047] Figure 2 A schematic diagram of sample classification in related art;

[0048] Figure 3 A flowchart of another cost control method for recommendation information provided in an embodiment of the present application;

[0049] Figure 4 A schematic diagram of another cost control method for recommendation information provided in an embodiment of the present application;

[0050] Figure 5 A schematic diagram of the structure of a cost control device for recommendation information provided in an embodiment of the present application;

[0051] Figure 6 A schematic diagram of the structure of a cost control device for recommendation information provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application.

[0053] It should be understood that the “embodiments of the present application” or “the aforementioned embodiments” mentioned throughout the specification mean that the specific features, structures or characteristics related to the embodiments are included in at least one embodiment of the present application. Therefore, “in the embodiments of the present application” or “in the aforementioned embodiments” appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. In the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments.

[0054] Unless otherwise specified, when an electronic device performs any step in the embodiments of the present application, the processor of the electronic device may perform the step. It is also worth noting that the embodiments of the present application do not limit the order in which the electronic device performs the following steps. In addition, the methods used to process data in different embodiments may be the same method or different methods. It should also be noted that any step in the embodiments of the present application can be independently executed by the electronic device, that is, when the electronic device performs any step in the following embodiments, it can be independent of the execution of other steps.

[0055] It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0056] It should be noted that in the field of modern marketing, traditional recommendation system algorithms have the following problems: 1) Insufficient cost control and budget constraints: Current marketing optimization models usually focus on maximizing revenue, with less consideration of cost control and budget constraints. This single revenue orientation may cause the promotion costs of the recommendation system to exceed expectations and fail to achieve the optimal allocation of resources. In the absence of effective cost management, companies may face the problem of budget overruns, affecting the sustainability of the overall marketing activities. 2) Insufficient integration of operations planning and recommendation algorithms: Operations planning can play a significant role in solving the cost efficiency of recommendation systems, and operations planning technology has significant advantages in resource allocation and cost optimization. It is widely used in logistics, production planning, finance and other fields. Combined with mathematical optimization methods such as linear programming and integer programming, operations research provides a series of powerful tools for optimal resource allocation and cost minimization. Effectively combining the advantages of data-driven recommendation algorithms and operational planning can address the cost control of marketing recommendations. However, the existing integration of operational planning and recommendation algorithms is still insufficient. In other words, existing recommendation systems often fail to effectively utilize the powerful tools provided by operational planning to achieve cost minimization and resource optimization. This lack of technical integration limits the overall advantages of both, making it difficult for recommendation systems to simultaneously balance recommendation effectiveness and cost control in practical applications. 3) Delayed feedback problem: In conversion rate estimation modeling, existing technologies often ignore the delayed feedback problem. This neglect can lead to biased sample construction, specifically by incorrectly classifying actual positive samples as negative samples, which in turn affects the accuracy of conversion rate estimation. Although some streaming architectures attempt to introduce models to predict the delay time of conversion and use delayed conversion samples for adjustment, this approach faces several challenges. Specifically, the conversion time prediction model is difficult to train using unbiased samples. Second, delayed samples are often sparse and difficult to obtain sufficient accuracy. These factors together lead to reduced conversion rate estimation efficiency, affecting the overall performance of the recommendation system.

[0057] It's important to note that an effective combination not only improves recommendation accuracy but also controls campaign costs within budget constraints. Furthermore, it can flexibly adapt to and predict user behavior when faced with delayed feedback, allowing for proactive preparation. This effective integration can help companies optimize marketing campaigns in complex market environments, achieve higher ROI, and enhance their flexibility and competitiveness in marketing activities.

[0058] Based on this, the embodiment of the present application provides a cost control method for recommendation information, which can be applied to a cost control device for recommendation information. Figure 1 As shown, the method includes the following steps:

[0059] Step 101: Determine a classification parameter corresponding to each initial information to be recommended based on a valid time period corresponding to each initial information to be recommended.

[0060] In an embodiment of the present application, information to be recommended refers to recommended information to be issued to the user, and the information to be recommended can be represented by item; initial information to be recommended refers to unlabeled information to be recommended, and there can be multiple different initial information to be recommended; the effective time period refers to the validity period corresponding to the initial information to be recommended, and can be in days; the classification parameter can be a parameter used to classify multiple initial information to be recommended, and can specifically be a time parameter (i.e., a waiting time window); for each initial information to be recommended, the effective time period corresponding to the initial information to be recommended can be analyzed first, and different times can be selected as classification parameters based on different analysis results.

[0061] In a feasible implementation, the initial information to be recommended may refer to a coupon to be issued to the user; the validity period may be 5 days, 10 days, 30 days, or other number of days.

[0062] Step 102: Determine the information category of each initial information to be recommended based on each classification parameter.

[0063] In an embodiment of the present application, the information category includes two types: initial information to be recommended is marked as a positive sample and initial information to be recommended is marked as a negative sample, and a certain initial information to be recommended is either marked as a positive sample or marked as a negative sample; for each initial information to be recommended, the classification parameter can be used as a reference to determine whether the initial information to be recommended is converted (i.e., cancelled, i.e., used) before the classification parameter to determine whether the initial information to be recommended is marked as a positive sample or a negative sample.

[0064] It should be noted that if Figure 2The figure shows a schematic diagram of sample classification in the related art. Specifically, 1) samples where the received coupons are converted within the waiting time window are marked as true positives, and the conversion behavior of these samples occurs within the expected time range, which accurately reflects the user's immediate reaction. 2) samples where the conversion of the received coupons occurs outside the waiting window but within the attribution window are marked as false negatives. A false negative means that although the user ultimately converts, the conversion occurs outside the waiting window, resulting in the initial marking as a negative. 3) samples where the received coupons are not converted are marked as true negatives, and true negatives have no conversion behavior within the entire attribution window, accurately reflecting the user's lack of interest in the particular coupon. In summary, the existence of false negatives in the related art (i.e., samples where the coupons are received but not converted but are converted at some point in the future) leads to bias in the classification of positive and negative samples. In contrast, the present application determines the classification parameters corresponding to the initial information to be recommended based on the valid time period corresponding to the initial information to be recommended. Then, based on the classification parameters, the information category of each initial information to be recommended is determined. In other words, positive and negative samples are strictly divided according to the time parameter, ensuring the accuracy and unbiasedness of positive and negative samples.

[0065] Step 103: Based on the information category, the first target feature of the user corresponding to each initial information to be recommended, the second target feature of each initial information to be recommended, and the target historical interaction feature between the user and each initial information to be recommended, each initial information to be recommended is processed to obtain each processed information to be recommended.

[0066] In an embodiment of the present application, the first target feature may refer to the feature of the user corresponding to the initial information to be recommended, and is the feature obtained after feature screening and feature preprocessing of the acquired first initial feature; the second target feature may refer to the feature of the initial information to be recommended, and is the feature obtained after feature screening and feature preprocessing of the acquired second initial feature; the target historical interaction feature may include the historical interaction feature on the user side and the historical interaction information on the information to be recommended side, and the historical interaction feature on the user side involves statistical data of the user's interaction with different initial information to be recommended, and the historical interaction information on the information to be recommended side includes the interaction feature between the initial information to be recommended and the user group; and the target historical interaction feature is also the feature obtained after feature screening and feature preprocessing of the acquired initial historical interaction feature; the processed information to be recommended may refer to the initial recommendation information with a label, and the label may specifically be the first target feature, the second target feature, the target historical interaction feature and the information category.

[0067] In an embodiment of the present application, for each initial information to be recommended, the first target feature of the user corresponding to the initial information to be recommended, the second target feature of the initial information to be recommended itself, the target historical interaction feature between the user corresponding to the initial information to be recommended and the initial information to be recommended, and the information category to which the initial information to be recommended belongs can be used to label the initial information to be recommended, so as to obtain the labeled information to be recommended (i.e., the processed information to be recommended), which can also be understood as a sample composed of the initial information to be recommended, the first target feature, the second target feature, the target historical interaction feature and the information category.

[0068] Step 104 : Based on the plurality of processed information to be recommended and the target conversion rate estimation model, determine the estimated conversion rate of each user for each processed information to be recommended.

[0069] In an embodiment of the present application, the target conversion rate prediction model may refer to a model that can determine the estimated conversion rate of each processed information to be recommended, and may be a trained model obtained by training the initial conversion rate prediction model using a training set; the estimated conversion rate may refer to the conversion rate of each user for each processed information to be recommended; all processed information to be recommended corresponding to all users may be input into the target conversion rate prediction model to obtain the estimated conversion rate of each user for each processed information to be recommended.

[0070] It should be noted that when using the training set to train the initial conversion rate prediction model, the cross entropy loss function (Cross Entropy) shown in the following formula (1) can be used to optimize the model parameters, and the grid search can be used to find the optimal hyperparameters to obtain the target conversion rate prediction model:

[0071]

[0072] in, represents the predicted probability of the i-th sample in the training set output by the conversion rate estimation model; y i represents the true label of the i-th sample, with a value of 0 or 1; i represents the i-th sample; n represents the number of samples; it should be noted that each sample in the training set is also a recommendation information with a label.

[0073] In a feasible implementation method, the target conversion rate prediction model can be a tree model, such as Light Gradient Boosting (LGB), or a deep model, such as Deep Factorization Machine (DeepFM) model; preferably, the target conversion rate prediction model can be an Entire Space Multi-task Model (ESMM) model, etc.

[0074] Step 105 : Determine a target probability of recommending each initial information to be recommended to the user based on the target budget, each estimated conversion rate, the cost of each processed information to be recommended, and each target variable.

[0075] The target variable is the defined probability variable for recommending each initial information to be recommended to the user.

[0076] In an embodiment of the present application, an optimization variable (i.e., target variable) for the probability of recommending each initial information to be recommended to the user can be defined first, each estimated conversion rate and the optimization variable can be processed, and each estimated conversion rate, the optimization variable, the cost of each processed information to be recommended and the target budget can be processed, and the number of initial information to be recommended sent and the optimization variable can be processed, and the target probability of recommending each processed information to be recommended to the user can be determined based on the above multiple processing results.

[0077] Step 106: Based on the multiple target probabilities, determine the target information to be recommended from the multiple initial information to be recommended and recommend it to the user.

[0078] In an embodiment of the present application, the target information to be recommended may refer to the initial information to be recommended that is ultimately recommended to the user; the initial information to be recommended corresponding to the largest target probability is selected from multiple target probabilities and recommended to the user.

[0079] The cost control method for recommended information provided in the embodiment of the present application determines the information category of each initial information to be recommended through the classification parameters determined by the effective time period, so that the information category of the initial information to be recommended determined based on the classification parameters is more accurate, so that an unbiased sample can be obtained, thereby making the estimated conversion rate obtained by the unbiased sample and the target estimated conversion rate model more accurate. At the same time, when recommending information to be recommended, the cost and target budget of each processed information to be recommended are used, that is, the cost of recommendation is taken into account when recommending information, instead of recommending without considering the recommendation cost as in the related art, thereby overcoming the problem of high recommendation cost in the prior art.

[0080] Based on the above embodiments, the present application provides a cost control method for recommendation information. Figure 3 and Figure 4 As shown, the method includes the following steps:

[0081] Step 201: For each initial information to be recommended, if the valid time period corresponding to the initial information to be recommended is less than or equal to the target time period, the cost control device for recommending information determines the valid time in the valid time period as a classification parameter.

[0082] In an embodiment of the present application, the target time period may be a time threshold determined based on historical experimental data; the effective time may refer to the expiration time of the initial information to be recommended; the effective time period and the target time period may be compared. When the effective time period is less than or equal to the target time period, its expiration time may be directly used as a classification parameter, which means that before the expiration time, the user's behavioral characteristics will be fully collected for training the initial conversion rate prediction model.

[0083] In a feasible implementation method, if the valid time period is the five days from January 1, 2025 to January 5, 2025, then the valid time is January 5, 2025; when the target time period is 7 days, then the valid time period is less than or equal to the target time period (i.e., 5 days < 7 days), then January 5, 2025 is determined as the classification time.

[0084] It should be noted that step 204 may be executed after step 201 .

[0085] Step 202: For each initial information to be recommended, if the valid time period corresponding to the initial information to be recommended is greater than the target time period, the cost control device for recommending information determines the cumulative conversion ratio of the multiple initial information to be recommended in each sub-time period of the valid time period.

[0086] In this embodiment of the present application, the cumulative conversion rate may refer to the cumulative conversion rate of multiple initial recommended information within a sub-time period; the target time period may refer to the time period in which the cumulative conversion rate is greater than the target threshold; the effective time period and the target time period may be compared. When the effective time period is greater than the target time period, it may indicate that the effective period of the initial recommended information is still quite long. In this case, the cumulative conversion rate rate within each sub-time period may be calculated. In one feasible implementation, the target threshold may be set to 95%.

[0087] Step 203: The cost control device for recommendation information determines a target sub-time as a classification parameter from among multiple sub-time periods based on multiple cumulative conversion proportions.

[0088] In an embodiment of the present application, the target sub-time may refer to the last time in a sub-time period; after obtaining the cumulative conversion ratio in each sub-time period as described above, each cumulative conversion ratio may be compared with the target threshold. When a certain cumulative conversion ratio is greater than or equal to the target threshold, the sub-time period corresponding to the target estimated conversion rate may be determined as a classification parameter; it should be noted that this method can ensure that most conversion behaviors are captured, thereby providing sufficient time to update the label.

[0089] In a feasible implementation method, when the valid time period corresponding to the initial recommended information is 31 days from January 1, 2025 to January 31, 2025, and the target time period is 7 days, the valid time period is greater than the target time period. Then, when the cumulative conversion ratio of the sub-time period from January 1, 2025 to January 5, 2025 is greater than the target threshold, the five days from January 1 to January 5 are determined as a classification parameter.

[0090] It should be noted that step 204 may also be executed after step 203 .

[0091] Step 204: The cost control device for recommending information determines the information category of each initial information to be recommended based on the classification parameter corresponding to each initial information to be recommended.

[0092] Among them, information categories include positive samples and negative samples.

[0093] In an embodiment of the present application, different classification parameters can be used to determine the information category for different initial information to be recommended. Specifically, when the valid time period corresponding to the initial information to be recommended is less than or equal to the target time period, the classification parameters determined by the method in step 201 can be used to determine the information category of the initial information to be recommended; when the valid time period corresponding to the initial information to be recommended is greater than the target time period, the classification parameters determined by the methods in steps 202 to 204 can be used to determine the information category of the initial information to be recommended.

[0094] It should be noted that step 204 can be implemented in the following ways:

[0095] Step 204A1: For each piece of initial information to be recommended, if the initial information to be recommended is converted before the classification parameter, the cost control device for recommending information determines that the information category of the initial information to be recommended is a positive sample.

[0096] Step 204A2: For each piece of initial information to be recommended, if the initial information to be recommended has not been converted before the classification parameter is determined, the cost control device for recommending information determines that the information category of the initial information to be recommended is a negative sample.

[0097] In an embodiment of the present application, the initial information to be recommended is converted before the classification parameters, which means that the initial information to be recommended is verified (i.e., used) before the classification parameters; that is, before the classification parameters, all the initial information to be recommended that is received and successfully converted is marked as a positive sample; all the initial information to be recommended that is received but not converted is marked as a negative sample.

[0098] Step 205: The cost control device for recommending information obtains a first initial feature of each user, a second initial feature of each initial information to be recommended, and an initial historical interaction feature between the user and each initial information to be recommended.

[0099] In an embodiment of the present application, the first initial feature may refer to the feature of the user corresponding to the initial information to be recommended, and the first initial feature may include but is not limited to the user's demographic information and the user's behavioral features, and the demographic information may include multiple sub-features, and the user's behavioral features may also include multiple sub-features; the second initial feature may refer to the feature of the initial information to be recommended, and the second initial feature may include but is not limited to the basic attributes of the initial information to be recommended and statistical data on the initial information to be recommended, and the basic attributes may include multiple sub-features, and the statistical data may also include multiple sub-features; the initial historical interaction feature may include historical interaction features on the user side and historical interaction information on the recommendation information side, and the historical interaction features on the user side involve statistical data on the user's interaction with different initial information to be recommended, and the historical interaction information on the recommendation information side includes the interaction features between the initial information to be recommended and the user group.

[0100] In a feasible implementation, the multiple sub-features included in the demographic information may specifically be the user's gender, the user's age, and the user's city; the multiple sub-features included in the user's behavioral characteristics may specifically be the user's consumption preferences, the user's consumption level, and the user's activity on various e-commerce platforms; the multiple sub-features included in the basic attributes of the initial information to be recommended may specifically be the cost of the initial information to be recommended, the pricing of the initial information to be recommended, etc.; the multiple sub-features included in the statistical data of the initial information to be recommended may specifically be the number of times the initial information to be recommended is received, the number of conversions of the initial information to be recommended, and the conversion rate of the initial information to be recommended.

[0101] Step 206: The cost control device for recommendation information performs feature screening on each first initial feature, each second initial feature, and each initial historical interaction feature to obtain each screened first initial feature, each screened second initial feature, and each screened initial historical interaction feature.

[0102] In an embodiment of the present application, since the first initial feature, each second initial feature, and each initial historical interaction feature may include multiple sub-features, feature screening is performed on each first initial feature, each second initial feature, and each initial historical interaction feature. Specifically, feature screening can be performed on the sub-features in the first initial feature, the second initial feature, and the initial historical interaction feature, that is, feature screening can be performed on multiple sub-features in each first initial feature to obtain each filtered first initial feature, and feature screening can be performed on multiple sub-features in each second initial feature to obtain each filtered second initial feature, and feature screening can be performed on multiple sub-features in each initial historical interaction feature to obtain each filtered initial historical interaction feature; specifically, feature screening can be performed by eliminating sub-features with zero importance and / or eliminating sub-features with coverage less than 10% and / or similar sub-features, selecting one of them to be retained based on domain knowledge, and / or eliminating sub-features whose correlation with the target label (i.e., whether the recommended information corresponding to the sub-feature is converted) is less than a set threshold; it should be noted that the set threshold here can be determined by ablation experiments to determine the optimal configuration.

[0103] Step 207 : The cost control device for recommendation information performs feature preprocessing on each filtered first initial feature, each filtered second initial feature, and each filtered initial historical interaction feature to obtain each first target feature, each second target feature, and each target historical interaction feature.

[0104] In an embodiment of the present application, each sub-feature of the first initial feature, the second initial feature, and the initial historical interaction feature can be divided into a discrete feature and a continuous feature, wherein the discrete feature is usually categorical data, and the continuous feature is numerical data, and hash coding (such as MD5 message digest algorithm (MD5 Message-Digest Algorithm, MD5) is used for feature preprocessing of the discrete feature, and standardization and / or normalization can be used for feature preprocessing of the continuous feature. In a feasible implementation method, when the first initial feature includes demographic information, and the demographic information specifically includes multiple sub-features such as the user's gender, the user's age, and the user's city, wherein the user's age can be numerical data such as 12 years old, 13 years old, and 14 years old, then the user's age is a continuous feature; the user's gender is represented by 0 for female and 1 for male, which belongs to categorical data, so the user's gender is a discrete feature.

[0105] In the embodiment of the present application, for the sub-features of the continuous feature category, firstly, the abnormal data is removed and the default value is filled, and then the sub-feature is standardized and / or normalized. Specifically, the standardization processing formula shown in the following formula (2) can be used for processing, and the normalization processing formula shown in the following formula (3) can be used for processing:

[0106]

[0107] Among them, Feature dense Represents the sub-feature itself; Feature mean Represents the mean of sub-features; Feature std Represents the standard deviation of the sub-feature.

[0108]

[0109] Among them, standardization is used to eliminate the dimensional differences between features, while normalization is applicable to models that are sensitive to feature scales to ensure that the model responds consistently to different features; Feature min Indicates the minimum value of the sub-feature, Feature max Indicates the maximum value of the sub-feature.

[0110] In a feasible implementation, when the user's gender is represented by 0 for female and 1 for male, if 3 appears in the gender, it means that the sub-feature is abnormal, and the abnormal feature is removed and filled with the default value.

[0111] Step 208: The cost control device for recommending information processes each initial information to be recommended based on the information category, the first target feature of the user corresponding to each initial information to be recommended, the second target feature of each initial information to be recommended, and the target historical interaction feature between the user and each initial information to be recommended, to obtain each processed information to be recommended.

[0112] Step 209 : The cost control device for recommending information determines an estimated conversion rate for each user for each processed piece of information to be recommended based on the plurality of processed pieces of information to be recommended and the target conversion rate estimation model.

[0113] In an embodiment of the present application, before determining the estimated conversion rate based on multiple processed information to be recommended and the target conversion rate estimation model, the initial conversion rate estimation model can be trained to obtain the target conversion rate estimation model. Specifically, the data set can be divided into a training set, a validation set, and a test set according to the time when the user receives the initial sample information to be recommended (i.e., the data set), and for each receipt record, the user's status data within the classification parameters is associated, including conversion within the classification parameters and non-conversion within the classification parameters, and this association method ensures that the model can accurately capture the full picture of user behavior.

[0114] In an embodiment of the present application, when using a training set to train an initial conversion rate prediction model, it is necessary to balance the ratio of positive and negative samples in the data set. For example, when the ratio of positive and negative samples is unbalanced and the proportion of negative samples is large, the optimal sampling ratio can be used for negative sampling; in a feasible implementation method, when the ratio of positive and negative samples is 1:100, it means that the ratio of positive and negative samples is unbalanced. At this time, the negative samples can be randomly sampled so that the ratio of positive and negative samples reaches 1:10.

[0115] In an embodiment of the present application, the sampled positive and negative samples are integrated to form a final model input data set, which is input into the initial conversion rate prediction model for model training. It should be noted that the classification parameter selection of training, validation, and test samples should be adjusted according to the specific business scale and needs to ensure the representativeness of the samples and the generalization ability of the model.

[0116] It should be noted that this solution ensures the accuracy and unbiasedness of positive and negative samples, and avoids the interference of samples that have not been converted but will be converted at some point in the future on model training. Through precise sample construction and reasonable time window design, the model can better capture user behavior patterns and improve the predictive performance of the recommendation system.

[0117] Step 210: The cost control device for recommending information determines an average estimated conversion rate based on the multiple estimated conversion rates and the amount of processed information to be recommended.

[0118] In the embodiment of the present application, a plurality of estimated conversion rates may be summed, and the summed result and the number of processed information to be recommended may be divided to obtain an average estimated conversion rate.

[0119] Step 211 : The cost control device for recommending information determines a target budget based on the number of users, the average cost of the processed information to be recommended, and the average estimated conversion rate.

[0120] In an embodiment of the present application, the average cost can be determined based on multiple costs and the number of processed information to be recommended. Specifically, the costs of multiple processed information to be recommended can be summed up, and the sum result and the number of processed information to be recommended can be divided; the average cost can also be directly determined based on business requirements; after determining the average cost, the number of users, the average cost and the average estimated conversion rate can be arithmetic operations to obtain the target budget (represented by B), which is specifically expressed by the formula B = number of users * estimated average conversion rate * average cost.

[0121] It should be noted that the present application may use operations planning methods to implement cost control solutions, where methods for solving operations planning problems can be divided into two categories. One category is exact algorithms, such as dynamic programming, which aim to find exact solutions to optimization problems. Specifically, dynamic programming is very effective for smaller-scale problems, but its time and space complexity often increase exponentially with the growth of the problem size, which makes it unsuitable for processing large data sets, especially in marketing recommendation systems. Since the number of users and products to be processed is usually large, dynamic programming may become difficult to implement. The other category is inexact algorithms, which can provide approximate solutions and are often used for large-scale data sets, and obtain satisfactory results within a reasonable time. Inexact algorithms include many types. In this application, the Lagrange dual algorithm is used for solution. By introducing Lagrange multipliers to transform the original problem, it is decomposed into more tractable sub-problems, and the duality property is used to accelerate the optimization process in the update iteration. In practical applications, Lagrange dual and other inexact algorithms are usually more suitable for processing large-scale complex problems, especially in cases with high-dimensional feature spaces and the need to meet multiple constraints, as follows:

[0122] Step 212: The cost control device for recommendation information constructs an objective function for the estimated conversion rate based on each estimated conversion rate and each target variable.

[0123] In the embodiment of the present application, the target variable for recommending each processed information to be recommended to each user can be a defined decision optimization variable; the objective function can be a function of the total estimated conversion rate after the information to be recommended is released; based on each estimated conversion rate and each target variable, an objective function for the estimated conversion rate can be constructed as Σ u Σ i p ui *X ui ; Among them, p ui It represents the estimated conversion rate of user u for the i-th initial recommended information (i.e. item i), X ui is the target variable, and represents the probability X of recommending item i to user u ui .

[0124] Step 213. Based on each estimated conversion rate, each target variable, each cost, and the target budget, the cost control device for recommended information constructs a first constraint condition for the cost of the initial to-be-recommended information.

[0125] In the embodiment of the present application, the first constraint condition may be a budget constraint, which may require that the cost of releasing the initial to-be-recommended information is less than the target budget. Then, the cost of releasing all the initial to-be-recommended information can be calculated first according to each estimated conversion rate, each target variable, and each cost, and then make it less than the target budget. Specifically, it may be Σ u Σ i p ui ×X ui ×b i <B, where b i represents the cost of the i-th initial to-be-recommended information.

[0126] Step 214. For any user, the cost control device for recommended information constructs a second constraint condition for the quantity of the initial to-be-recommended information based on each target variable and the target constant.

[0127] In the embodiment of the present application, the target constant can be determined based on the actual situation; the second constraint condition can be a constraint on the quantity of the initial to-be-recommended information released; when the quantity of the initial to-be-recommended information sent to and only sent to each user is 1 (i.e., the target constant), then for any user u, the second constraint condition that can be constructed is Σ i X ui =1.

[0128] It should be noted that according to Steps 213 to 214, the objective function and constraints that can be constructed are shown in the following formula (4):

[0129]

[0130] Step 215. Under the constraints of the first constraint condition and the second constraint condition, the cost control device for recommended information determines the target probability of recommending each initial to-be-recommended information to the user when the objective function reaches the target value.

[0131] In the embodiment of the present application, the Lagrange multiplier can be introduced, and the objective function (i.e., the original problem) can be transformed into a relaxation problem, which only contains some relatively simple constraint conditions, while the objective function and constraint conditions of the original problem remain unchanged. In this way, through the optimal solution of the relaxation problem, the lower bound solution (i.e., the target probability) of the original problem can be obtained.

[0132] Step 215B1. The cost control device for recommended information uses the Lagrange multiplier, the first constraint condition, and the objective function to obtain the Lagrange relaxation function for the estimated conversion rate.

[0133] In the embodiments of the present application, by introducing the Lagrange multiplier λ and constructing the Lagrangian relaxation function of the original problem based on the Lagrange multiplier, the first constraint condition, and the objective function, it is specifically shown as the following formula (5):

[0134]

[0135] It should be noted that due to the Lagrangian dual decomposition, the first constraint condition Σ u Σ i p ui ×X ui ×b i <B is removed from the constraints, and for a fixed λ, the optimization problem can be transformed into a set of subproblems.

[0136] Step 215B2: The cost control device for recommended information determines the target probability of recommending each initial recommended information to the user when the Lagrangian relaxation function reaches the target value under the constraint of the second constraint condition.

[0137] Step 216: The cost control device for recommended information determines the target recommended information from multiple initial recommended information based on multiple target probabilities and recommends it to the user.

[0138] In the embodiments of the present application, after obtaining the Lagrangian relaxation function, the Lagrangian relaxation function can be deformed as shown in the following formula (6): [[ID=二十九]]

[0139] [[ID=三十]] [[ID=三十一]] [[ID=三十二]]

[0140] [[ID=三十三]]It should be noted that the above formula (6) can be further deformed into the following formula (7):[[ID=三十四]] [[ID=三十五]]

[0141] [[ID=三十六]] [[ID=三十七]] [[ID=三十八]]

[0142] [[ID=三十九]]It should be noted that compared with the original problem (i.e., formula (4)), the original problem has O(MN) decision variables, but solving the subproblem only requires traversing M items, reducing the computational complexity to O(M). And according to formula (7), the optimal allocation strategy is expressed as shown in the following formula (8):[[ID=四十]] [[ID=四十一]]

[0143] [[ID=四十二]] [[ID=四十三]] [[ID=四十四]]

[0144] [[ID=四十五]]To solve the Lagrangian dual problem, the following steps can be alternately executed: (1) Fix λ and traverse to calculate p[[ID=四十六]] ij [[ID=四十七]]-λp[[ID=四十八]] ij [[ID=四十九]]b[[ID=五十]] jThe subsidy with the largest value is taken as the optimal decision X; (2) Based on X, iterate λ to meet the optimality condition, and in order to improve the efficiency of the λ iteration process, a binary search method can be used. Finally, the hyperparameter λ that has been calculated is used. For each user's request for issuing recommendation information, each item can be traversed to calculate the value that makes p ij -λp ij b j The item with the largest value (i.e. the target information to be recommended) is sent to the user.

[0145] It's important to note that this application combines the strengths of recommendation systems and operations planning, maximizing metrics like scenario conversion rate within cost constraints. While meeting business needs, it flexibly adjusts recommendation strategies to optimize cost-effectiveness. This not only improves recommendation accuracy but also enhances companies' ability to control costs in marketing activities. Furthermore, it uses the Lagrange duality method to simplify and solve operations planning problems, significantly reducing system resource overhead, accelerating the solution process, and improving model iteration efficiency, ultimately delivering tangible business value.

[0146] It should be noted that, for the description of the same steps and contents in this embodiment as those in other embodiments, reference can be made to the description in other embodiments and will not be repeated here.

[0147] The cost control method for recommended information provided in the embodiment of the present application determines the information category of each initial information to be recommended through the classification parameters determined by the effective time period, so that the information category of the initial information to be recommended determined based on the classification parameters is more accurate, so that an unbiased sample can be obtained, thereby making the estimated conversion rate obtained by the unbiased sample and the target estimated conversion rate model more accurate. At the same time, when recommending information to be recommended, the cost and target budget of each processed information to be recommended are used, that is, the cost of recommendation is taken into account when recommending information, instead of recommending without considering the recommendation cost as in the related art, thereby overcoming the problem of high recommendation cost in the prior art.

[0148] Based on the above embodiments, the present invention provides a cost control device for recommendation information. The cost control device for recommendation information can be applied to Figure 1 and Figure 3 In the cost control method of recommendation information provided in the corresponding embodiment, refer to Figure 5 As shown, the cost control device 3 for recommendation information may include: a first processing unit 31, a first determining unit 32, a second processing unit 33 and a second determining unit 33, wherein:

[0149] The first processing unit 31 is configured to determine a classification parameter corresponding to each piece of initial information to be recommended based on a valid time period corresponding to each piece of initial information to be recommended;

[0150] The first processing unit 31 is further configured to determine an information category of each initial information to be recommended based on each classification parameter;

[0151] The first processing unit 31 is further configured to process each initial information to be recommended to obtain processed information to be recommended based on the information category, the first target feature of the user corresponding to each initial information to be recommended, the second target feature of each initial information to be recommended, and the target historical interaction feature between the user and each initial information to be recommended;

[0152] A first determining unit 32 is configured to determine an estimated conversion rate for each user for each processed piece of information to be recommended based on the plurality of processed pieces of information to be recommended and a target conversion rate estimation model;

[0153] The second processing unit 33 is configured to determine a target probability of recommending each initial piece of information to be recommended to the user based on the target budget, each estimated conversion rate, the cost of each processed piece of information to be recommended, and each target variable; wherein the target variable is a defined probability variable for recommending each initial piece of information to be recommended to the user;

[0154] The second determining unit 34 is configured to determine target information to be recommended from the multiple initial information to be recommended based on multiple target probabilities and recommend it to the user.

[0155] In other embodiments of the present application, the first processing unit 31 is specifically configured to perform the following steps:

[0156] For each initial information to be recommended, if the valid time period is less than or equal to the target time period, the valid time in the valid time period is determined as a classification parameter;

[0157] For each initial information to be recommended, if the effective time period is greater than the target time period, determine the cumulative conversion ratio of the multiple initial information to be recommended in each sub-time period of the effective time period;

[0158] Based on multiple cumulative conversion proportions, a target sub-time period is determined from multiple sub-time periods as a classification parameter.

[0159] In other embodiments of the present application, the first processing unit 31 is specifically configured to perform the following steps:

[0160] For each initial information to be recommended, if the initial information to be recommended is transformed before the classification parameter, the information category is determined to be a positive sample;

[0161] If the initial information to be recommended is not transformed before the classification parameters, the information category is determined to be a negative sample.

[0162] In other embodiments of the present application, the first determining unit 32 is specifically configured to perform the following steps:

[0163] Obtaining a first initial feature of each user, a second initial feature of each initial information to be recommended, and an initial historical interaction feature between the user and each initial information to be recommended;

[0164] Performing feature screening on each first initial feature, each second initial feature, and each initial historical interaction feature to obtain each filtered first initial feature, each filtered second initial feature, and each filtered initial historical interaction feature;

[0165] Feature preprocessing is performed on each filtered first initial feature, each filtered second initial feature, and each filtered initial historical interaction feature to obtain each first target feature, each second target feature, and each target historical interaction feature.

[0166] In other embodiments of the present application, the second processing unit 33 is specifically configured to perform the following steps:

[0167] Determine an average estimated conversion rate based on the multiple estimated conversion rates and the number of processed information to be recommended;

[0168] Determine the target budget based on the number of users, the average cost of processed recommendations, and the average estimated conversion rate.

[0169] In other embodiments of the present application, the second processing unit 33 is specifically configured to perform the following steps:

[0170] Based on each estimated conversion rate and each target variable, construct an objective function for the estimated conversion rate;

[0171] Constructing a first constraint condition on the cost of the initial information to be recommended based on each estimated conversion rate, each target variable, each cost, and the target budget;

[0172] For any user, based on each target variable and target constant, construct a second constraint condition on the amount of initial information to be recommended;

[0173] Under the constraints of the first constraint and the second constraint, a target probability of recommending each initial to-be-recommended information to the user is determined when the objective function reaches a target value.

[0174] In other embodiments of the present application, the second processing unit 33 is specifically configured to perform the following steps:

[0175] Using the Lagrange multiplier, the first constraint condition and the objective function, a Lagrange relaxation function for estimating the conversion rate is obtained;

[0176] Under the constraint of the second constraint, the target probability of recommending each initial to-be-recommended information to the user is determined when the Lagrangian relaxation function reaches the target value.

[0177] It should be noted that the specific description of the steps performed by each unit can be referred to Figure 1 and Figure 3 The cost control method for recommendation information provided in the corresponding embodiment will not be described in detail here.

[0178] The cost control device for recommended information provided in the embodiment of the present application determines the information category of each initial information to be recommended through the classification parameters determined by the effective time period, so that the information category of the initial information to be recommended determined based on the classification parameters is more accurate, so that an unbiased sample can be obtained, thereby making the estimated conversion rate obtained by the unbiased sample and the target estimated conversion rate model more accurate. At the same time, when recommending information to be recommended, the cost and target budget of each processed information to be recommended are used, that is, the cost of recommendation is taken into account when recommending information, instead of recommending without considering the recommendation cost as in the related art, thereby overcoming the problem of high recommendation cost in the prior art.

[0179] Based on the above embodiments, the embodiments of the present application provide a cost control device for recommendation information, which can be applied to Figure 1 and Figure 3 In the cost control method of recommendation information provided in the corresponding embodiment, refer to Figure 6 As shown, the cost control device 4 for the recommendation information may include: a processor 41, a memory 42 and a communication bus 43, wherein:

[0180] The communication bus 43 is used to realize the communication connection between the processor 41 and the memory 42;

[0181] The processor 41 is configured to execute the cost control program of the recommendation information in the memory 42 to implement the following steps:

[0182] Determining a classification parameter corresponding to each initial information to be recommended based on a valid time period corresponding to each initial information to be recommended;

[0183] Based on each classification parameter, determining the information category of each initial information to be recommended;

[0184] Based on the information category, the first target feature of the user corresponding to each initial information to be recommended, the second target feature of each initial information to be recommended, and the target historical interaction feature between the user and each initial information to be recommended, each initial information to be recommended is processed to obtain each processed information to be recommended;

[0185] Determining an estimated conversion rate for each user for each processed piece of information to be recommended based on the plurality of processed pieces of information to be recommended and a target conversion rate estimation model;

[0186] Determine a target probability of recommending each initial piece of information to the user based on the target budget, each estimated conversion rate, the cost of each processed piece of information to be recommended, and each target variable; wherein the target variable is a defined probability variable for recommending each initial piece of information to the user;

[0187] Based on multiple target probabilities, target information to be recommended is determined from multiple initial information to be recommended and recommended to the user.

[0188] In other embodiments of the present application, the processor 41 is configured to execute the cost control program for recommendation information in the memory 42, and determine the classification parameter corresponding to each initial information to be recommended based on the valid time period corresponding to each initial information to be recommended, so as to implement the following steps:

[0189] For each initial information to be recommended, if the valid time period is less than or equal to the target time period, the valid time in the valid time period is determined as a classification parameter;

[0190] For each initial information to be recommended, if the effective time period is greater than the target time period, determine the cumulative conversion ratio of multiple information to be recommended on the fourth day of the Lunar New Year in each sub-time period within the effective time period;

[0191] Based on multiple cumulative conversion proportions, a target sub-time period is determined from multiple sub-time periods as a classification parameter.

[0192] In other embodiments of the present application, the processor 41 is configured to execute the cost control program for recommendation information in the memory 42 to determine the information category of each initial information to be recommended based on each classification parameter, so as to implement the following steps:

[0193] For each initial information to be recommended, if the initial information to be recommended is transformed before the classification parameter, the information category is determined to be a positive sample;

[0194] If the initial information to be recommended is not transformed before the classification parameters, the information category is determined to be a negative sample.

[0195] In other embodiments of the present application, the processor 41 is configured to execute a cost control method for recommendation information of a cost control program for recommendation information in the memory 42 to implement the following steps:

[0196] Obtaining a first initial feature of each user, a second initial feature of each initial information to be recommended, and an initial historical interaction feature between the user and each initial information to be recommended;

[0197] Performing feature screening on each first initial feature, each second initial feature, and each initial historical interaction feature to obtain each filtered first initial feature, each filtered second initial feature, and each filtered initial historical interaction feature;

[0198] Feature preprocessing is performed on each filtered first initial feature, each filtered second initial feature, and each filtered initial historical interaction feature to obtain each first target feature, each second target feature, and each target historical interaction feature.

[0199] In other embodiments of the present application, the processor 41 is configured to execute a cost control method for recommendation information of a cost control program for recommendation information in the memory 42 to implement the following steps:

[0200] Determine an average estimated conversion rate based on the multiple estimated conversion rates and the number of processed information to be recommended;

[0201] Determine the target budget based on the number of users, the average cost of processed recommendations, and the average estimated conversion rate.

[0202] In other embodiments of the present application, the processor 41 is configured to execute the cost control program for recommendation information in the memory 42 to determine a target probability of recommending each initial information to be recommended to the user based on the target budget, each estimated conversion rate, the cost of each processed information to be recommended, and each target variable, so as to implement the following steps:

[0203] Based on each estimated conversion rate and each target variable, construct an objective function for the estimated conversion rate;

[0204] Constructing a first constraint condition on the cost of the initial information to be recommended based on each estimated conversion rate, each target variable, each cost, and the target budget;

[0205] For any user, based on each target variable and target constant, construct a second constraint condition on the amount of initial information to be recommended;

[0206] Under the constraints of the first constraint and the second constraint, a target probability of recommending each initial to-be-recommended information to the user is determined when the objective function reaches a target value.

[0207] In other embodiments of the present application, the processor 41 is configured to execute the cost control program for recommendation information in the memory 42, under the constraints of the first constraint and the second constraint, to determine the target probability of recommending each initial to-be-recommended information to the user when the objective function reaches the target value, so as to implement the following steps:

[0208] Using the Lagrange multiplier, the first constraint condition and the objective function, a Lagrange relaxation function for estimating the conversion rate is obtained;

[0209] Under the constraint of the second constraint, the target probability of recommending each initial to-be-recommended information to the user is determined when the Lagrangian relaxation function reaches the target value.

[0210] It should be noted that the specific description of the steps performed by the processor can be referred to Figure 1 and Figure 3 The cost control method for recommendation information provided in the corresponding embodiment will not be described in detail here.

[0211] The cost control device for recommended information provided in the embodiment of the present application determines the information category of each initial information to be recommended through the classification parameters determined by the effective time period, so that the information category of the initial information to be recommended determined based on the classification parameters is more accurate, so that an unbiased sample can be obtained, thereby making the estimated conversion rate obtained by the unbiased sample and the target estimated conversion rate model more accurate. At the same time, when recommending information to be recommended, the cost and target budget of each processed information to be recommended are used, that is, the cost of recommendation is taken into account when recommending information, instead of recommending without considering the cost of recommendation as in the related art, thereby overcoming the problem of high recommendation cost in the prior art.

[0212] Based on the above embodiments, the embodiments of the present application provide a computer-readable storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement Figure 1 and Figure 3 The corresponding embodiment provides steps of a method for controlling the cost of recommendation information.

[0213] Based on the above embodiment, the embodiment of the present application further provides a computer program product, including a computer program, which can be executed by the processor 41 of the cost control device 4 of the recommendation information to achieve Figure 1 and Figure 3 The corresponding embodiment provides steps of a method for controlling the cost of recommendation information.

[0214] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0215] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0216] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0217] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0218] The above description is merely a preferred embodiment of the present application and is not intended to limit the scope of protection of the present application.

Claims

1. A cost control method for recommendation information, characterized in that: The method comprises: Determining a classification parameter corresponding to each piece of initial information to be recommended based on a valid time period corresponding to each piece of initial information to be recommended; Determining the information category of each of the initial information to be recommended based on each of the classification parameters; Based on the information category, the first target feature of the user corresponding to each of the initial information to be recommended, the second target feature of each of the initial information to be recommended, and the target historical interaction feature between the user and each of the initial information to be recommended, each of the initial information to be recommended is processed to obtain each processed information to be recommended; Determining an estimated conversion rate for each user for each processed piece of information to be recommended based on the plurality of processed pieces of information to be recommended and a target conversion rate estimation model; Determining a target probability of recommending each of the initial information to be recommended to the user based on the target budget, each of the estimated conversion rates, the cost of each of the processed information to be recommended, and each target variable; wherein the target variable is a defined probability variable for recommending each of the initial information to be recommended to the user; Based on the multiple target probabilities, target information to be recommended is determined from the multiple initial information to be recommended and recommended to the user.

2. The method according to claim 1, characterized in that The determining, based on the valid time period corresponding to each piece of initial information to be recommended, a classification parameter corresponding to each piece of initial information to be recommended, includes: For each piece of initial information to be recommended, if the valid time period is less than or equal to the target time period, determining the valid time in the valid time period as the classification parameter; If the effective time period is greater than the target time period, determining the cumulative conversion ratio of the plurality of initial information to be recommended in each sub-time period of the effective time period; Based on the multiple cumulative conversion proportions, a target sub-time period is determined from the multiple sub-time periods as the classification parameter.

3. The method according to claim 2, characterized in that Determining the information category of each of the initial information to be recommended based on each of the classification parameters includes: For each of the initial information to be recommended, if the initial information to be recommended is converted before the classification parameter, determining that the information category is a positive sample; If the initial information to be recommended has not been converted before the classification parameter, it is determined that the information category is a negative sample.

4. The method according to claim 1, wherein Before processing each of the initial information to be recommended to obtain each processed information to be recommended, the method further includes: Obtaining a first initial feature of each of the users, a second initial feature of each of the initial information to be recommended, and an initial historical interaction feature between the user and each of the initial information to be recommended; Performing feature screening on each of the first initial features, each of the second initial features, and each of the initial historical interaction features to obtain each filtered first initial feature, each filtered second initial feature, and each filtered initial historical interaction feature; Feature preprocessing is performed on each of the filtered first initial features, each of the filtered second initial features, and each of the filtered initial historical interaction features to obtain each of the first target features, each of the second target features, and each of the target historical interaction features.

5. The method according to claim 1, wherein Before determining the target probability of recommending each of the initial to-be-recommended information to the user, the method further includes: determining an average estimated conversion rate based on the plurality of estimated conversion rates and the amount of processed information to be recommended; The target budget is determined based on the number of users, the average cost of the processed information to be recommended, and the average estimated conversion rate.

6. The method according to claim 5, characterized in that Determining a target probability of recommending each of the initial to-be-recommended information to the user based on the target budget, each of the estimated conversion rates, and each target variable includes: constructing an objective function for the estimated conversion rate based on each of the estimated conversion rates and each of the target variables; constructing a first constraint condition on the cost of the initial information to be recommended based on each of the estimated conversion rates, each of the target variables, each of the costs, and the target budget; For any of the users, constructing a second constraint condition on the amount of the initial information to be recommended based on each of the target variables and the target constant; Under the constraints of the first constraint condition and the second constraint condition, a target probability of recommending each piece of the initial to-be-recommended information to the user is determined when the objective function reaches a target value.

7. The method according to claim 6, characterized in that The determining, under the constraints of the first constraint condition and the second constraint condition, a target probability of recommending each piece of the initial to-be-recommended information to the user when the objective function reaches a target value includes: Obtaining a Lagrangian relaxation function for estimating the conversion rate using a Lagrangian multiplier, the first constraint, and the objective function; Under the constraint of the second constraint condition, a target probability of recommending each piece of the initial to-be-recommended information to the user is determined when the Lagrangian relaxation function reaches a target value.

8. A cost control device for recommendation information, characterized in that: The device comprises: A first processing unit is configured to determine a classification parameter corresponding to each piece of initial information to be recommended based on a valid time period corresponding to each piece of initial information to be recommended; The first processing unit is further configured to determine an information category of each of the initial information to be recommended based on each of the classification parameters; The first processing unit is further configured to process each of the initial information to be recommended to obtain each processed information to be recommended based on the information category, the first target feature of the user corresponding to each of the initial information to be recommended, the second target feature of each of the initial information to be recommended, and the target historical interaction feature between the user and each of the initial information to be recommended; a first determining unit, configured to determine, based on the plurality of processed information to be recommended and a target conversion rate estimation model, an estimated conversion rate for each user for each processed information to be recommended; a second processing unit, configured to determine a target probability of recommending each of the initial information to be recommended to the user based on a target budget, each of the estimated conversion rates, the cost of each of the processed information to be recommended, and each target variable; wherein the target variable is a defined probability variable for recommending each of the initial information to be recommended to the user; The second determining unit is configured to determine target information to be recommended from the plurality of initial information to be recommended based on the plurality of target probabilities and recommend the target information to the user.

9. A cost control device for recommendation information, characterized in that: The device includes: a processor, a memory and a communication bus; The communication bus is used to realize the communication connection between the processor and the memory; The processor is configured to execute the cost control program for recommendation information in the memory to implement the steps of the cost control method for recommendation information according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the cost control method for recommendation information according to any one of claims 1 to 7.

11. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the computer program implements the steps of the cost control method of recommendation information according to any one of claims 1 to 7.

Images