Online marketing method and device based on deep lifting network, and terminal equipment

Abstract

The invention is suitable for the technical field of intelligent marketing, and provides an online marketing method and device based on a deep lifting network, and terminal equipment, and the method comprises the steps: obtaining user features of users in a user group; inputting the user features and the incentive features into a target deep lifting network model, processing the user features through the target deep lifting network model to obtain user key features, determining the importance degrees of the user key features under different incentive features, and determining the target depth lifting network model according to the user key features and the importance degrees. Predicting prediction response results of the user under different excitation characteristics; and according to the prediction response results of the users under the different excitation characteristics, determining the promotion values of the users, determining target users in the user group based on the promotion values, and carrying out online marketing on the target users. By adopting the method and the device, the response change of the user to different incentive means can be effectively identified, and online marketing is performed according to the response change of the user, so that the effectiveness and the accuracy of marketing recommendation are improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent marketing technology, and in particular to an online marketing method, apparatus and terminal device based on deep enhancement network. Background Art

[0002] In order to increase platform revenue and user stickiness, various online platforms usually formulate some online marketing strategies with different incentives for users from time to time, such as coupons, discounts and bonuses. Different users react differently to each incentive, and improper delivery may waste marketing costs and user disgust.

[0003] In actual scenarios, there is a difference between the user's natural reaction when they receive incentives and when they do not receive incentives, and different reactions will also divide user groups.

[0004] In view of this, how to effectively identify the changes in users' responses to different incentive measures and conduct online marketing based on the changes in users' responses in order to improve the effectiveness and accuracy of marketing recommendations is a problem that needs to be considered currently. Summary of the invention

[0005] The embodiments of the present application provide an online marketing method, apparatus and terminal device based on a deep boosting network, which can effectively identify changes in user responses to different incentives, and perform online marketing based on the user's response changes, thereby improving the effectiveness and accuracy of marketing recommendations.

[0006] In a first aspect, an embodiment of the present application provides an online marketing method based on a deep boosting network, comprising:

[0007] Obtain user characteristics of users in the user group;

[0008] Inputting the user features and the incentive features into a pre-trained target deep boosting network model, processing the user features through the target deep boosting network model to obtain user key features, determining the importance of the user key features under different incentive features, and predicting the predicted response results of the user under different incentive features based on the user key features and the importance; determining the user's boost value based on the user's predicted response results under different incentive features, and the boost value is used to evaluate the response changes of the user after receiving the different incentive features;

[0009] Target users in the user group are determined based on the lift value, and online marketing is performed to the target users.

[0010] In a possible implementation of the first aspect, the target deep boosting network model includes a group-shared feature gating, a feature embedding layer, a group-specific embedding gating, and a main network module;

[0011] The shared feature gating is used to screen the user features based on a weighting mechanism to obtain the user key features;

[0012] The feature embedding layer is used to process the user key features to obtain user key embedded features;

[0013] The group-specific embedding gating is used to determine the importance of the user key embedding features under different stimulus features;

[0014] The main network module is used to predict the predicted response results of the user under different incentive features according to the key embedded features of the user and the importance, and determine the improvement value of the user according to the predicted response results of the user under different incentive features.

[0015] In a possible implementation manner of the first aspect, before the step of inputting the user features and the incentive features into a pre-trained target deep boosting network model, the step includes:

[0016] Acquire an original sample set, wherein the original sample set includes sample features, incentive labels, and response labels of sample users;

[0017] Constructing a deep boosting network model, the deep boosting network model comprising a group-shared feature gating, a feature embedding layer, a group-specific embedding gating, and a main network module;

[0018] Inputting the sample feature and the stimulus label into the group shared feature gating, processing the sample feature through the group shared feature gating, and obtaining the sample key feature;

[0019] Processing the user key features through the feature embedding layer to obtain sample key embedding features;

[0020] Determining the importance of the key embedding features of the sample under different response labels through the group-specific embedding gating;

[0021] Predicting the response results of the sample user under different incentive features according to the sample key embedding features and the importance through the main network module; determining the lift value of the sample user according to the response results of the sample user under different incentive features;

[0022] Inputting the incentive label, the response label and the boost value of the sample user into a preset evaluation mechanism for evaluation to obtain the boost effect of the deep boost network model;

[0023] The model parameters of the deep boosting network model are optimized based on a preset loss function, and iterations are repeated. When the boosting effect does not improve continuously within a preset number of times, a trained target deep boosting network model is obtained.

[0024] In a possible implementation manner of the first aspect, the step of processing the sample features through the group shared feature gating to obtain the sample key features includes:

[0025] The sample features are screened based on a weighted mechanism through the group shared feature gating to obtain the sample key features.

[0026] In a possible implementation manner of the first aspect, different response tags correspond to different group-specific embedding gatings, and the step of determining the importance of the key feature of the sample under different response tags by using the group-specific embedding gating includes:

[0027] Inputting the sample key embedding features into the group-specific embedding gates corresponding to the response labels respectively;

[0028] The importance of the key embedding features of the samples under each of the response tags is determined by the group-specific embedding gating corresponding to each of the response tags.

[0029] In a possible implementation manner of the first aspect, the optimization target of the deep boosting network model is shown in the following calculation formula:

[0030]

[0031] in, represents the preset loss function, g f represents the group shared feature gating, g e represents group-specific embedding gating, represents the cross entropy loss, represents the L1 regularization loss of group-shared feature gating, represents the L1 regularized loss for group-specific embedding gating, represents the proxy loss, and λ1 and λ2 represent the optimization weights.

[0032] In a second aspect, the embodiment of the present application provides an online marketing device based on a deep boosting network, comprising:

[0033] A feature acquisition unit, used to acquire user features of users in a user group;

[0034] A lift value determination unit, used for inputting the user features and incentive features into a pre-trained target deep lift network model, processing the user features through the target deep lift network model to obtain user key features, determining the importance of the user key features under different incentive features, and predicting the predicted response results of the user under different incentive features according to the user key features and the importance; determining the user's lift value according to the predicted response results of the user under different incentive features, wherein the lift value is used to evaluate the response change of the user after receiving the different incentive features;

[0035] An online marketing unit is used to determine target users in the user group based on the boost value and perform online marketing to the target users.

[0036] In a third aspect, an embodiment of the present application provides a terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the online marketing method based on the deep boosting network as described in the first aspect above is implemented.

[0037] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the online marketing method based on the deep boosting network as described in the first aspect above is implemented.

[0038] In a fifth aspect, an embodiment of the present application provides a computer program product, which, when executed on a terminal device, enables the terminal device to execute the online marketing method based on a deep boosting network as described in the first aspect above.

[0039] In an embodiment of the present application, user characteristics of users in a user group are first obtained, and the user characteristics and incentive characteristics are input into a pre-trained target deep boosting network model. The user characteristics are processed by the target deep boosting network model to obtain user key characteristics, and the importance of the user key characteristics under different incentive characteristics is determined. Based on the user key characteristics and the importance, the predicted response results of the user under different incentive characteristics are predicted, which can effectively improve the accuracy of the predicted response results. Then, based on the predicted response results of the user under different incentive characteristics, the boost value of the user is determined, and the boost value is used to evaluate the response changes of the user after accepting the different incentive characteristics, which is conducive to screening the user group based on the boost value, and then effectively determining the target user from the user group, and then conducting precise online marketing to the target user, which can improve the effectiveness and accuracy of marketing recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0041] Figure 1 It is a flowchart of an implementation of an online marketing method based on a deep network enhancement provided in an embodiment of the present application;

[0042] Figure 2 It is a specific implementation flow chart of constructing and training a target deep boosting network model in the online marketing method based on the deep boosting network provided in the embodiment of the present application;

[0043] Figure 3 It is a schematic diagram of the architecture of a deep boosting network model in the online marketing method based on the deep boosting network provided in the embodiment of the present application;

[0044] Figure 4 It is a schematic diagram of the architecture of another deep boosting network model in the online marketing method based on the deep boosting network provided in the embodiment of the present application;

[0045] Figure 5 It is a structural block diagram of an online marketing device based on a deep boosting network provided in an embodiment of the present application;

[0046] Figure 6 It is a schematic diagram of a terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0047] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0048] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0049] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0050] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0051] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0052] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0053] As an example but not a limitation, the online marketing method based on the deep boosting network provided in the embodiment of the present application can be applied to terminal devices and servers of various types of data that need to perform marketing recommendations, and the terminal devices can specifically include vehicle-mounted terminals, mobile phones, tablet computers, laptop computers, ultra-mobile personal computers (UMPCs), desktop computers, etc. The embodiment of the present application does not impose any restrictions on the specific types of terminal devices.

[0054] Figure 1 The implementation process of the online marketing method based on deep network enhancement provided by the embodiment of the present application is shown, and the method flow includes steps S101 to S103. The specific implementation principle of each step is as follows:

[0055] Step S101: Obtain user characteristics of users in a user group.

[0056] In this embodiment, user characteristics may include one or more of the user's basic information, behavior characteristics, preference characteristics, social characteristics, and technical characteristics. Among them, basic information includes age, gender, geographic location, income level, etc.; behavior characteristics include historical browsing records, historical purchase records, etc.; preference characteristics include user preferences, etc.; social characteristics include information such as the user's friends, followed users, or followed users on social media; technical characteristics include the type of device used by the user, operating system, browser, network connection type, speed, etc.

[0057] As one of the input information of the target depth improvement network model, user characteristics play a vital role in the model's prediction results and subsequent target user determination. User characteristics usually cover multiple dimensions, aiming to fully reflect the user's attributes, behaviors, preferences, etc., so as to conduct more accurate personalized marketing.

[0058] Step S102: Input the user features and incentive features into a pre-trained target deep boosting network model, process the user features through the target deep boosting network model to obtain user key features, determine the importance of the user key features under different incentive features, and predict the predicted response results of the user under different incentive features based on the user key features and the importance; determine the user's boost value based on the predicted response results of the user under different incentive features.

[0059] The boost value is used to evaluate the change in the user's response after receiving the different incentive features.

[0060] In actual marketing scenarios, users can usually only be in one state: accepting a specific incentive (i.e., the treatment group) or not accepting the incentive (i.e., the control group). This means that only one type of user response can be observed, and it is difficult to directly and accurately estimate the difference between the user's response to each incentive and the natural response (i.e., no incentive received). In the embodiment of the present application, user features and incentive features are input into a pre-trained target deep improvement network model, and the target deep improvement network model processes the user features, removes redundant information, obtains the user's key features, determines the importance of the user's key features under different incentive features, and predicts the user's predicted response results under different incentive features based on the user's key features and their importance. Then, based on the difference in the predicted response results of the user under different incentive features, the user's improvement value can be quickly determined, and the response changes of the user after accepting the different incentive features can be effectively evaluated.

[0061] The above-mentioned excitation feature is a binary feature, such as 1 and 0. If the excitation feature is 1, it is excitation, and if the excitation feature is 0, it is not excitation. If the excitation feature is 1, the lift value is the difference between the predicted response result of the model under the excitation condition and the predicted response result of the model under the non-excitation condition); if the excitation feature is 0, the lift value is the difference between the predicted response result of the model under the excitation condition and the predicted response result of the model under the non-excitation condition.

[0062] In an embodiment of the present application, the target deep boosting network model includes a group shared feature gating, a feature embedding layer, a group specific embedding gating and a main network module. The group shared feature gating and the group specific embedding gating are both model components. The shared feature gating is used to screen the user features based on a weighted mechanism, remove redundant features, and obtain the user key features; the feature embedding layer is used to process the user key features to obtain the user key embedded features; the group specific embedding gating is used to determine the importance of the user key embedded features under different incentive features; the main network module is used to predict the predicted response results of the user under different incentive features based on the user key embedded features and the importance, and determine the user's boost value based on the predicted response results of the user under different incentive features.

[0063] As an embodiment of the present application, Figure 2 A specific implementation process of constructing and training a target deep boosting network model in the online marketing method based on the deep boosting network provided in the embodiment of the present application is shown, and is described in detail as follows:

[0064] Step S201: Obtain an original sample set, which includes sample features, incentive labels and response labels of sample users. The original sample set is divided into a treatment group (an incentive group) and a control group (a group that does not receive incentives).

[0065] Step S202: construct a deep boosting network model, wherein the deep boosting network model includes a group-shared feature gating, a feature embedding layer, a group-specific embedding gating, and a main network module.

[0066] Step S203: inputting the sample feature and the excitation label into the group shared feature gating, processing the sample feature through the group shared feature gating, and obtaining the sample key feature.

[0067] The group shared feature gating is based on a weighted mechanism to screen the sample features from the group shared layer to obtain the sample key features.

[0068] Step S204: Process the user key features through the feature embedding layer to obtain sample key embedding features.

[0069] Figure 3 An architecture diagram of a deep boost network model is shown. In an embodiment of the present application, the deep boost network model is a new deep boost network framework with group-aware component selection (GCS), which uses a cascade gating mechanism to select components for modeling each group. In a marketing recommendation system, the choice of components is directly related to the quality of the recommendation effect. Different components may be based on different algorithms, data sources, or user behavior patterns, so their performance in different scenarios will also be different. Selecting appropriate components can significantly improve the accuracy, diversity, and user satisfaction of the recommendation system, thereby increasing sales, improving user stickiness, etc.

[0070] In this embodiment, the process of converting the key features of the sample into the low-dimensional representation feature embedding layer by the feature embedding layer can be expressed as follows:

[0071] e ij =E×x ij , E∈R m×d (1)

[0072] where x ij is the j-th dimension feature of the i-th instance, m is the number of feature values, d is the hyperparameter of the embedding dimension, E represents the embedding table shared across scenarios, and R is the dataset.

[0073] In addition, the shared latent embedding h i It is obtained by feeding the feature embedding into a group of group-shared layers, expressed as follows:

[0074]

[0075] in W l and b l represents the learnable weight matrix and bias vector, σ(·) represents the activation function, and ; represents the number of layers. The goal of GCS is to obtain the group-aware optimal selection result of the feature embedding process and the shared potential embedding, which is expressed as follows (3):

[0076]

[0077] Among them, g f is the group-shared feature gating, {g e,t＝0 ,g e,t＝1} is the group-specific embedding gating. Since there is an obvious instance selection bias in online marketing and group-aware feature gating can easily amplify this problem, using shared gating instead of group-aware gating for feature selection can effectively reduce instance selection bias.

[0078] In the embodiment of the present application, a cascade gating mechanism is introduced in the model training stage, which has a scheme of learning through continuous training to obtain a universal and effective group shared feature gating g f and group-specific embedding gating {g e,t＝0 ,g e ,t＝1}.

[0079] The first part of the cascade gating mechanism aims to train the group-shared feature gates from a global perspective by leveraging instances in all groups to capture the importance of features between groups. In order to effectively optimize the group-shared feature gating vector , a continuous group shared feature gating is introduced in the embodiment of the present application, which is expressed as follows (4):

[0080]

[0081] in, is the initial value of the continuous group shared feature gate, σ(·) is the sigmoid function, k is the current training round number, K is the total number of training rounds, β is a hyperparameter that changes with the number of training rounds, and γ is the final value of β after K rounds of training.

[0082] Group shared feature gating dynamically selects and utilizes shared features across groups (or across tasks, or across scenarios) through a gating mechanism, effectively obtaining the feature importance of all groups, that is, filtering out redundant information from the group shared layer from a global perspective.

[0083] Step S205: Determine the importance of the key features of the sample under different response labels through the group-specific embedding gating.

[0084] In a possible implementation, different response tags correspond to different group-specific embedding gates, and the sample key embedding features are respectively input into the group-specific embedding gates corresponding to the response tags; the importance of the sample key embedding features under each response tag is determined by the group-specific embedding gates corresponding to the response tags. For example, the treatment group corresponds to the first embedding gate, and the control group corresponds to the second embedding gate.

[0085] For example, Figure 4 As shown, the deep boosting network model includes a first set of specific embedding gating and a second set of specific embedding gating.

[0086] The second part of the cascade gating mechanism aims to utilize the instances in each group from a local perspective to train a group-specific embedding gate to capture the importance of the shared latent embedding dimension with respect to each group. In order to effectively optimize the group-specific embedding gate {g e,t＝0 ,g e,t＝1}, introduced and The specific process is as follows:

[0087]

[0088] Among them, g e,t＝0 represents the first set of specific embedding gates with the stimulus label 0, is the initial value of the first set of specific embedding gates in succession, g e,t＝1 The second set of specific embedding gates representing the stimulus label is 1, is the initial value of the second set of continuous specific embedding gates, k is the current training round number, K is the total number of training rounds, β is a hyperparameter that changes with the number of training rounds, and γ is the final value of β after K rounds of training.

[0089] In this embodiment, the importance of the embedding dimension of the sample features in the sample feature subset is identified from a local perspective using group-specific embedding gating to identify the source from each specific group layer (ie, the potential embedding output of the group shared layer).

[0090] Step S206: predicting the response results of the sample user under different incentive features according to the sample key embedded features and the importance through the main network module; determining the improvement value of the sample user according to the response results of the sample user under different incentive features.

[0091] In this embodiment, the main network module is used to predict the response results of sample users under different incentive feature labels.

[0092] Step S207: input the incentive label, the response label and the boost value of the sample user into a preset evaluation mechanism for evaluation to obtain the boost effect of the deep boosting network model.

[0093] The above-mentioned preset evaluation mechanism is used to evaluate the network performance of the deep improvement network model according to the evaluation index, and determine the improvement effect of the deep improvement network model. The larger the value of the evaluation index, the better the method improvement value ranking effect, and the more accurate the prediction of the target population.

[0094] Step S208: Optimizing the model parameters of the deep boosting network model based on a preset loss function, repeating the iteration, and when the boosting effect does not improve within a preset number of times, obtaining a trained target deep boosting network model.

[0095] The preset number of times can be set according to the needs. For example, the preset number of times is set to 5 times. When the improvement effect is not improved within 5 consecutive times, overfitting is prevented, the model training is completed, and the trained target deep improvement network model is obtained.

[0096] Exemplarily, the QINI evaluation mechanism is used for evaluation, and the maximum value of the QINI evaluation is the optimal value. The current network performance value of the deep boosting network model is updated according to the maximum evaluation value, and the early stopping mechanism is used to prevent the model from overfitting. If the current optimal performance value is not less than the current network performance value, the early stopping counter is increased by 1. If the current optimal performance value is less than the current network performance value, the current network performance value is updated. If the iteration stop condition is met and the early stopping counter is greater than 5, the optimized target deep boosting network model is obtained, otherwise the current optimal performance value is continued to be compared to be not less than the current network performance value.

[0097] In a possible implementation, the optimization target of the deep boosting network model is as follows:

[0098]

[0099] in, represents the preset loss function, g f represents the group shared feature gating, g e represents group-specific embedding gating, represents the cross entropy loss, represents the L1 regularization loss of group-shared feature gating, represents the L1 regularized loss for group-specific embedding gating, represents the proxy loss, and λ1 and λ2 represent the optimization weights.

[0100] To ensure that the model can reconstruct marketing instances, this embodiment introduces a cross entropy loss for the corresponding response labels. The calculation of is as follows (8):

[0101]

[0102] Among them, y i is the response label of the i-th sample feature, is the prediction result of the i-th sample feature output by the model, n is the total number of sample features, and the model will use the original group-specific layer to obtain the prediction result corresponding to each instance after obtaining the shared potential embedding, as shown in the following formula (9):

[0103]

[0104] in, and Represents different groups of learnable weight matrices and bias vectors, t i Represents the incentive characteristic variable.

[0105] In some implementations, in order to encourage the sparsity of the group-shared feature gating vector and the group-specific embedding gating vector, L1 regularization is also introduced in this embodiment. The L1 regularization loss of the group-specific embedding gating is As shown in equation (10), the L1 regularization loss of group shared feature gating is As shown in the following formula ((11)

[0106]

[0107] Here, ||·||1 represents the L1 norm.

[0108] In this embodiment, considering that the goal of online marketing is different from that of traditional supervised learning, the above optimization goal constraint model alone may not necessarily bring better marketing results. The preset loss function of the embodiment of the present application also includes a proxy loss related to the online marketing goal. Specifically, according to the response label y i and the incentive characteristic variable t i The relationship between y and y is used to infer the potential improvement of an instance, where positive improvement only exists when y i =t i There is negative lift y when i ≠t i Therefore, without loss of generality, we assume that a batch of instance pairs is constructed by random sampling, where x1 is an instance whose improvement may be positive and x2 is an instance whose improvement may be negative. The proxy loss is expressed as follows (12):

[0109]

[0110] where τ diff =τ(x1)-τ(x2), τ(x1) represents a positive lift value, and τ(x2) ​​represents a negative lift value.

[0111] Step S103: determining target users in the user group based on the boost value, and performing online marketing to the target users.

[0112] By comparing the predicted response results of users under different incentive features, we can calculate the lift value of each user. Users in the user group are sorted from high to low by lift value. The higher the lift value, the better the marketing effect. The lift value helps us effectively segment the user group and identify those users who respond most actively to the incentive features, that is, the target users. When conducting online marketing to the target users, we can improve the accuracy and effectiveness of online marketing recommendations.

[0113] As can be seen from the above, in the embodiment of the present application, the user characteristics of the users in the user group are first obtained, and the user characteristics and incentive characteristics are input into the pre-trained target deep improvement network model, and the user characteristics are processed by the target deep improvement network model to obtain the user's key characteristics, and the importance of the user's key characteristics under different incentive characteristics is determined. According to the user's key characteristics and the importance, the predicted response results of the user under different incentive characteristics are predicted, which can effectively improve the accuracy of the predicted response results, and then according to the predicted response results of the user under different incentive characteristics, the user's improvement value is determined, and the improvement value is used to evaluate the response changes of the user after accepting the different incentive characteristics, which is conducive to screening the user group based on the improvement value, and then effectively determining the target user from the user group, and then conducting precise online marketing to the target user, which can improve the effectiveness and accuracy of marketing recommendations.

[0114] It should be understood that the size of the serial numbers of the steps in the above embodiments 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.

[0115] Corresponding to the online marketing method based on deep network enhancement described in the above embodiment, Figure 4 A structural block diagram of an online marketing device based on a deep boosting network provided in an embodiment of the present application is shown. For ease of explanation, only the parts related to the embodiment of the present application are shown.

[0116] Reference Figure 5 The online marketing device based on the deep boosting network includes: a feature acquisition unit 51, a boosting value determination unit 52, and an online marketing unit 53, wherein:

[0117] A feature acquisition unit 51 is used to acquire user features of users in a user group;

[0118] A lift value determination unit 52 is used to input the user features and incentive features into a pre-trained target deep lift network model, process the user features through the target deep lift network model, obtain user key features, determine the importance of the user key features under different incentive features, and predict the predicted response results of the user under different incentive features based on the user key features and the importance; determine the user's lift value based on the predicted response results of the user under different incentive features, and the lift value is used to evaluate the response changes of the user after receiving the different incentive features;

[0119] The online marketing unit 53 is used to determine the target users in the user group based on the boost value and perform online marketing to the target users.

[0120] As a possible implementation of the present application, the target depth boosting network model includes a group-shared feature gating, a feature embedding layer, a group-specific embedding gating, and a main network module;

[0121] The shared feature gating is used to screen the user features based on a weighting mechanism to obtain the user key features;

[0122] The feature embedding layer is used to process the user key features to obtain user key embedded features;

[0123] The group-specific embedding gating is used to determine the importance of the user key embedding features under different stimulus features;

[0124] The main network module is used to predict the predicted response results of the user under different incentive features according to the key embedded features of the user and the importance, and determine the improvement value of the user according to the predicted response results of the user under different incentive features.

[0125] As a possible implementation manner of the present application, the above-mentioned device further includes a first model building training unit, which is used to:

[0126] Acquire an original sample set, wherein the original sample set includes sample features, incentive labels, and response labels of sample users;

[0127] Constructing a deep boosting network model, the deep boosting network model comprising a group-shared feature gating, a feature embedding layer, a group-specific embedding gating, and a main network module;

[0128] Inputting the sample feature and the stimulus label into the group shared feature gating, processing the sample feature through the group shared feature gating, and obtaining the sample key feature;

[0129] Processing the user key features through the feature embedding layer to obtain sample key embedding features;

[0130] Determining the importance of the key features of the sample under different response labels through the group-specific embedding gating;

[0131] Predicting the response results of the sample user under different incentive features according to the sample key embedding features and the importance through the main network module; determining the lift value of the sample user according to the response results of the sample user under different incentive features;

[0132] Inputting the incentive label, the response label and the boost value of the sample user into a preset evaluation mechanism for evaluation to obtain the boost effect of the deep boost network model;

[0133] The model parameters of the deep boosting network model are optimized based on a preset loss function, and iterations are repeated. When the boosting effect does not improve continuously within a preset number of times, a trained target deep boosting network model is obtained.

[0134] As a possible implementation manner of the present application, the step of processing the sample features through the group shared feature gating to obtain the sample key features includes:

[0135] The sample features are screened based on a weighted mechanism through the group shared feature gating to obtain the sample key features.

[0136] As a possible implementation manner of the present application, different response tags correspond to different group-specific embedding gating, and the step of determining the importance of the key features of the sample under different response tags by the group-specific embedding gating includes:

[0137] Inputting the sample key embedding features into the group-specific embedding gates corresponding to the response labels respectively;

[0138] The importance of the key embedding features of the samples under each of the response tags is determined by the group-specific embedding gating corresponding to each of the response tags.

[0139] As a possible implementation of the present application, the optimization target of the deep boosting network model is shown in the following calculation formula:

[0140]

[0141] in, represents the preset loss function, g f represents the group shared feature gating, g e represents group-specific embedding gating, represents the cross entropy loss, represents the L1 regularization loss of group-shared feature gating, represents the L1 regularized loss for group-specific embedding gating, represents the proxy loss, and λ1 and λ2 represent the optimization weights.

[0142] As can be seen from the above, in the embodiment of the present application, the user characteristics of the users in the user group are first obtained, and the user characteristics and incentive characteristics are input into the pre-trained target deep improvement network model, and the user characteristics are processed by the target deep improvement network model to obtain the user's key characteristics, and the importance of the user's key characteristics under different incentive characteristics is determined. According to the user's key characteristics and the importance, the predicted response results of the user under different incentive characteristics are predicted, which can effectively improve the accuracy of the predicted response results, and then according to the predicted response results of the user under different incentive characteristics, the user's improvement value is determined, and the improvement value is used to evaluate the response changes of the user after accepting the different incentive characteristics, which is conducive to screening the user group based on the improvement value, and then effectively determining the target user from the user group, and then conducting precise online marketing to the target user, which can improve the effectiveness and accuracy of marketing recommendations.

[0143] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0144] The present application also provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, Figures 1 to 4 Represents the steps of any online marketing method based on deep network promotion.

[0145] The embodiment of the present application also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, Figures 1 to 4 Represents the steps of any online marketing method based on deep network promotion.

[0146] The embodiment of the present application also provides a computer program product, when the computer program product is run on a terminal device, the terminal device executes the following Figures 1 to 4 Represents the steps of any online marketing method based on deep network promotion.

[0147] Figure 6 Schematic diagram of a terminal device provided by an embodiment of the present application. Figure 6 As shown, the terminal device 6 of this embodiment includes: a processor 60, a memory 61, and a computer program 62 stored in the memory 61 and executable on the processor 60. When the processor 60 executes the computer program 62, the steps in the above-mentioned embodiments of the online marketing method based on the deep boosting network are implemented, for example Figure 1Alternatively, when the processor 60 executes the computer program 62, the functions of each module / unit in the above-mentioned device embodiments are realized, for example Figure 5 The functions of units 51 to 53 are shown.

[0148] Exemplarily, the computer program 62 may be divided into one or more modules / units, which are stored in the memory 61 and executed by the processor 60 to complete the present application. The one or more modules / units may be a series of computer-readable instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program 62 in the terminal device 6.

[0149] The terminal device 6 may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art will appreciate that Figure 6 It is only an example of the terminal device 6 and does not constitute a limitation of the terminal device 6. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the terminal device 6 may also include input and output devices, network access devices, buses, etc.

[0150] The processor 60 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0151] The memory 61 may be an internal storage unit of the terminal device 6, such as a hard disk or memory of the terminal device 6. The memory 61 may also be an external storage device of the terminal device 6, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device 6. Further, the memory 61 may also include both an internal storage unit and an external storage device of the terminal device 6. The memory 61 is used to store the computer program and other programs and data required by the terminal device. The memory 61 may also be used to temporarily store data that has been output or is to be output.

[0152] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0153] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0154] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to the device / terminal device, a recording medium, a computer memory, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), an electric carrier signal, a telecommunication signal, and a software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0155] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0156] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. An online marketing method based on deep boosting network, characterized in that: include: Obtain user characteristics of users in the user group; Inputting the user features and the incentive features into a pre-trained target deep boosting network model, processing the user features through the target deep boosting network model to obtain user key features, determining the importance of the user key features under different incentive features, and predicting the predicted response results of the user under different incentive features based on the user key features and the importance; determining the user's boost value based on the user's predicted response results under different incentive features, and the boost value is used to evaluate the response changes of the user after receiving the different incentive features; Target users in the user group are determined based on the lift value, and online marketing is performed to the target users.

2. The method according to claim 1, characterized in that The target deep boosting network model includes a group-shared feature gating, a feature embedding layer, a group-specific embedding gating, and a main network module; The shared feature gating is used to screen the user features based on a weighting mechanism to obtain the user key features; The feature embedding layer is used to process the user key features to obtain user key embedded features; The group-specific embedding gating is used to determine the importance of the user key embedding features under different stimulus features; The main network module is used to predict the predicted response results of the user under different incentive features according to the key embedded features of the user and the importance, and determine the improvement value of the user according to the predicted response results of the user under different incentive features.

3. The method according to claim 1, characterized in that Before the step of inputting the user features and the incentive features into the pre-trained target deep boosting network model, the method includes: Acquire an original sample set, wherein the original sample set includes sample features, incentive labels, and response labels of sample users; Constructing a deep boosting network model, the deep boosting network model comprising a group-shared feature gating, a feature embedding layer, a group-specific embedding gating, and a main network module; Inputting the sample feature and the stimulus label into the group shared feature gating, processing the sample feature through the group shared feature gating, and obtaining the sample key feature; Processing the user key features through the feature embedding layer to obtain sample key embedding features; Determining the importance of the key embedding features of the sample under different response labels through the group-specific embedding gating; Predicting the response results of the sample user under different incentive features according to the sample key embedding features and the importance through the main network module; determining the lift value of the sample user according to the response results of the sample user under different incentive features; Inputting the incentive label, the response label and the boost value of the sample user into a preset evaluation mechanism for evaluation to obtain the boost effect of the deep boost network model; The model parameters of the deep boosting network model are optimized based on a preset loss function, and iterations are repeated. When the boosting effect does not improve continuously within a preset number of times, a trained target deep boosting network model is obtained.

4. The method according to claim 3, characterized in that The step of processing the sample features through the group shared feature gating to obtain the sample key features includes: The sample features are screened based on a weighted mechanism through the group shared feature gating to obtain the sample key features.

5. The method according to claim 3, characterized in that: Different response labels correspond to different group-specific embedding gating, and the step of determining the importance of the key features of the sample under different response labels by using the group-specific embedding gating includes: Inputting the sample key embedding features into the group-specific embedding gates corresponding to the response labels respectively; The importance of the key embedding features of the samples under each of the response tags is determined by the group-specific embedding gating corresponding to each of the response tags.

6. The method according to claim 3, characterized in that: The optimization goal of the deep boosting network model is shown in the following calculation formula: in, represents the preset loss function, g f represents the group shared feature gating, g e represents group-specific embedding gating, represents the cross entropy loss, represents the L1 regularization loss of group-shared feature gating, represents the L1 regularized loss for group-specific embedding gating, represents the proxy loss, and λ1 and λ2 represent the optimization weights.

7. An online marketing device based on a deep boosting network, characterized in that: include: A feature acquisition unit, used to acquire user features of users in a user group; A lift value determination unit, used for inputting the user features and incentive features into a pre-trained target deep lift network model, processing the user features through the target deep lift network model to obtain user key features, determining the importance of the user key features under different incentive features, and predicting the predicted response results of the user under different incentive features according to the user key features and the importance; determining the user's lift value according to the predicted response results of the user under different incentive features, wherein the lift value is used to evaluate the response change of the user after receiving the different incentive features; An online marketing unit is used to determine target users in the user group based on the boost value and perform online marketing to the target users.

8. The device according to claim 7, characterized in that The target deep boosting network model includes a group-shared feature gating, a feature embedding layer, a group-specific embedding gating, and a main network module; The shared feature gating is used to screen the user features based on a weighting mechanism to obtain the user key features; The feature embedding layer is used to process the user key features to obtain user key embedded features; The group-specific embedding gating is used to determine the importance of the user key embedding features under different stimulus features; The main network module is used to predict the predicted response results of the user under different incentive features according to the key embedded features of the user and the importance, and determine the improvement value of the user according to the predicted response results of the user under different incentive features.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the online marketing method based on the deep boosting network as described in any one of claims 1 to 6 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the online marketing method based on a deep boosting network as described in any one of claims 1 to 6 is implemented.

Images