User contact mode determination method and device, and electronic device

Abstract

The purpose of the embodiments of the present disclosure is to provide a method and device for determining a user reach mode, an electronic device, and a computer-readable storage medium, which relates to the field of artificial intelligence. The method for determining a user reach mode includes: determining the user attribute characteristics of a target user, and determining the reach mode characteristics corresponding to the target user; inputting the reach mode characteristics and the user attribute characteristics into an incremental improvement value prediction system respectively to obtain an incremental prediction value corresponding to the target user; wherein the incremental improvement value prediction system is obtained by adjusting the parameters of a first machine learning model based on sample characteristics and sample incremental values; and determining a target reach mode for the target user based on the incremental prediction value corresponding to the target user. The present disclosure can improve the accuracy of reach mode allocation and the overall reach effect.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular, to a method for determining a user reach mode, a device for determining a user reach mode, an electronic device, and a computer-readable storage medium. Background Art

[0002] With the development of Internet technology, Internet business is also increasing. In order to promote the reach of users, it is usually achieved through SMS, email, application push and application service notification.

[0003] Under the existing technology, a response model is constructed, and the incremental improvement is modeled based on the response model to obtain the incremental improvement model, so as to select the reach method based on the incremental improvement model.

[0004] However, in the above technical solution, the model cannot be optimized according to the gain, so that there is room for improvement in the accuracy of reach method allocation and the overall reach effect.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0006] The purpose of the embodiments of the present disclosure is to provide a method for determining a user reach mode, a device for determining a user reach mode, an electronic device, and a computer-readable storage medium, thereby improving the accuracy of reach mode allocation and the overall reach effect at least to a certain extent.

[0007] According to one aspect of the present disclosure, a method for determining a user contact mode is provided, comprising:

[0008] Determine the user attribute characteristics of the target user, and determine the corresponding contact method characteristics of the target user;

[0009] The contact mode feature and the user attribute feature are respectively input into the incremental lift value prediction system to obtain the incremental prediction value corresponding to the target user; wherein the incremental lift value prediction system is obtained by adjusting the parameters of the first machine learning model based on the sample features and the sample incremental value, the first machine learning model is constructed based on the second machine learning model and the sample incremental value, the sample incremental value is calculated based on the predicted conversion data, and the predicted conversion data is obtained by inputting the sample features into the second machine learning model;

[0010] Determine a target reaching method for the target user according to the incremental prediction value corresponding to the target user;

[0011] Select the access method corresponding to the incremental improvement value when the incremental improvement value reaches the preset value for access; and / or

[0012] Select the reaching method corresponding to the maximum incremental improvement value of the incremental improvement value for reaching;

[0013] The user attribute characteristics include the user's basic attribute characteristics and the user's behavioral attribute characteristics;

[0014] Obtaining sample features and annotated conversion data corresponding to the sample features; wherein the sample features include sample user features and sample contact method features;

[0015] Inputting the sample user features and the sample contact mode features into the first decision tree in the second machine learning model;

[0016] Iteratively generate multiple decision trees in a residual descending direction according to the output result of the previous decision tree and the labeled transformation data;

[0017] When the number of the generated decision trees meets a preset condition, the output result of the last decision tree is used as the predicted conversion data output by the second machine learning model;

[0018] Calculating the matching degree between the sample feature and each feature path;

[0019] Calculate the predicted conversion data of each of the contact modes corresponding to each characteristic path according to the matching degree;

[0020] Calculate the loss function according to the sample increment value;

[0021] The parameters of the first machine learning model are adjusted according to the loss function, and the adjusted first machine learning model is used as an incremental improvement value prediction system.

[0022] According to one aspect of the present disclosure, a method for acquiring an incremental improvement value prediction system is provided, comprising:

[0023] Acquire sample features and annotated conversion data corresponding to the sample features; wherein the sample features include sample user features and sample contact method features, and the user attribute features also include user basic attribute features and user behavior attribute features;

[0024] Inputting the sample user features and the sample contact method features into a second machine learning model to output corresponding predicted conversion data;

[0025] Calculate the sample increment value corresponding to the sample user feature according to the predicted conversion data;

[0026] The parameters of the first machine learning model are adjusted according to the sample characteristics and the sample incremental value to obtain an incremental improvement value prediction system; wherein the first machine learning model is constructed based on the second machine learning model and the sample incremental value.

[0027] According to one aspect of the present disclosure, a device for determining a user access mode is provided, comprising:

[0028] A user feature acquisition module is used to determine the user attribute characteristics of the target user and determine the contact method characteristics corresponding to the target user;

[0029] An incremental prediction value acquisition module, used to input the reach mode feature and the user attribute feature into the incremental lift value prediction system respectively, to obtain the incremental prediction value corresponding to the target user;

[0030] A module for determining a reaching method is used to determine a target reaching method for the target user based on the incremental prediction value corresponding to the target user.

[0031] According to one aspect of the present disclosure, there is provided an incremental improvement value prediction system acquisition device, comprising:

[0032] A module for obtaining sample features and annotated conversion data is used to obtain sample features and annotated conversion data corresponding to the sample features; wherein the sample features include sample user features and sample contact method features;

[0033] A predicted conversion data output module, used to input the sample user characteristics and the sample contact mode characteristics into a second machine learning model to output corresponding predicted conversion data;

[0034] An incremental improvement value calculation module is used to calculate the sample incremental value corresponding to the sample user feature according to the predicted conversion data;

[0035] An incremental improvement prediction system construction module is used to adjust the parameters of the first machine learning model according to the sample characteristics and the incremental improvement value to obtain an incremental improvement value prediction system; wherein, the first machine learning model is constructed based on the second machine learning model and the sample incremental value.

[0036] According to one aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute any one of the above-mentioned methods by executing the executable instructions.

[0037] According to one aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, any one of the above methods is implemented.

[0038] The exemplary embodiments of the present disclosure may have some or all of the following beneficial effects:

[0039] In the method for determining the user reach mode provided in the exemplary implementation of the present disclosure, the user attribute characteristics of the target user can be determined, and the reach mode characteristics corresponding to the target user can be determined; the reach mode characteristics and the user attribute characteristics are respectively input into the incremental lift value prediction system to obtain the incremental prediction value corresponding to the target user; wherein the incremental lift value prediction system is obtained by adjusting the parameters of the first machine learning model based on the sample characteristics and the sample incremental value, the first machine learning model is constructed based on the second machine learning model and the sample incremental value, the sample incremental value is calculated based on the predicted conversion data, and the predicted conversion data is obtained by inputting the sample characteristics into the second machine learning model; according to the incremental prediction value corresponding to the target user, the target reach mode is determined for the target user. On the one hand, by obtaining the incremental prediction value of the target user under different reach modes through the incremental lift value prediction system, the error accumulation caused by multiple models can be effectively avoided, thereby reducing the amount of model calculation, which is conducive to improving the accuracy of reach mode allocation. On the other hand, since the accuracy of reach mode allocation is improved, the efficiency of user reach can be improved, thereby improving the overall reach effect.

[0040] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0042] Figure 1 A schematic diagram showing an exemplary model architecture of a user reach determination device and an incremental lift value prediction system acquisition device to which the embodiments of the present disclosure can be applied;

[0043] Figure 2 A flowchart of a method for determining a user contact mode according to an embodiment of the present disclosure is schematically shown;

[0044] Figure 3 A flowchart of a decision tree determination process according to an embodiment of the present disclosure is schematically shown;

[0045] Figure 4A flowchart for obtaining predicted conversion data according to an embodiment of the present disclosure is schematically shown;

[0046] Figure 5 A flowchart of a process for adjusting parameters of an incremental improvement value prediction system according to an embodiment of the present disclosure is schematically shown;

[0047] Figure 6 A flowchart of a method for obtaining an incremental improvement value prediction system according to an embodiment of the present disclosure is schematically shown;

[0048] Figure 7 A flowchart schematically shows a process of applying a method for obtaining an incremental improvement value prediction system and a method for determining a user reach mode to a terminal device according to an embodiment of the present disclosure;

[0049] Figure 8 A block diagram of a device for determining a user access mode according to an embodiment of the present disclosure is schematically shown;

[0050] Fig. 9 A block diagram of an incremental improvement value prediction system acquisition device according to an embodiment of the present disclosure is schematically shown;

[0051] Fig.10 A schematic diagram of the structure of a computer model of an electronic device suitable for implementing an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0052] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concepts of the example embodiments are fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0053] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0054] Figure 1 A schematic diagram of a model architecture of an exemplary application environment in which a method and device for determining a user contact mode according to an embodiment of the present disclosure can be applied is shown. In addition, Figure 1 It can also be used to represent a schematic diagram of a model architecture of an exemplary application environment of a method and apparatus for determining user reach.

[0055] like Figure 1 As shown, the model architecture 100 may include one or more of terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc. The terminal devices 101, 102, 103 may be various electronic devices with display screens, including but not limited to desktop computers, portable computers, smart phones, tablet computers, etc. It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. According to the implementation requirements, there may be any number of terminal devices, networks and servers. For example, the server 105 may be a server cluster composed of multiple servers.

[0056] The user reach mode determination method provided in the embodiment of the present disclosure can be executed in the server 105, and accordingly, the user reach mode determination device is generally set in the server 105. The user reach mode determination method provided in the embodiment of the present disclosure can also be executed by the terminal devices 101, 102, and 103, and accordingly, the user reach mode determination device can also be set in the terminal devices 101, 102, and 103, which is not particularly limited in this exemplary embodiment.

[0057] For example, in an exemplary embodiment, the server 105 can obtain sample features and annotated conversion data corresponding to the sample features; input the sample user features and the sample reach method features into the second machine learning model to output corresponding predicted conversion data; calculate the sample incremental value corresponding to the sample user features according to the predicted conversion data; adjust the parameters of the first machine learning model according to the sample features and the sample incremental value to obtain an incremental improvement value prediction system; determine the user attribute features of the target user, and determine the reach method features corresponding to the target user; input the reach method features and the user attribute features into the incremental improvement value prediction system respectively to obtain the incremental prediction value corresponding to the target user; finally, based on the incremental prediction value corresponding to the target user, determine the target reach method for the target user and send it to the terminal devices 101, 102, 103; the terminal devices 101, 102, 103 send the reach method to the user, but it is easy for those skilled in the art to understand that the above application scenarios are only for example purposes and are not limited to this in this exemplary embodiment.

[0058] The implementation details of the technical solution of the embodiment of the present invention are described in detail below:

[0059] Figure 2 A flowchart of a method for determining a user access mode according to an embodiment of the present invention is schematically shown. The method for determining a user access mode is applicable to the electronic device described in the above embodiment. Figure 2 As shown, the method for determining the user contact mode includes at least steps S210 to S230, which are described in detail as follows:

[0060] Step S210, determining the user attribute characteristics of the target user, and determining the contact method characteristics corresponding to the target user;

[0061] Step S220, respectively inputting the contact mode feature and the user attribute feature into the incremental lift value prediction system to obtain the incremental prediction value corresponding to the target user; wherein the incremental lift value prediction system is obtained by adjusting the parameters of the first machine learning model based on the sample feature and the sample incremental value, the first machine learning model is constructed based on the second machine learning model and the sample incremental value, the sample incremental value is calculated based on the predicted conversion data, and the predicted conversion data is obtained by inputting the sample feature into the second machine learning model;

[0062] Step S230: Determine a target reaching method for the target user based on the incremental prediction value corresponding to the target user.

[0063] In the user reach mode determination method provided in this example implementation, on the one hand, the incremental prediction value of the target user under different reach modes is obtained through the incremental lift value prediction system, which can effectively avoid the error accumulation caused by multiple models, thereby reducing the amount of model calculation, which is beneficial to improving the accuracy of reach mode allocation. On the other hand, since the accuracy of reach mode allocation is improved, the efficiency of user reach can be improved, thereby improving the overall reach effect.

[0064] Next, in another embodiment, the above steps are described in more detail.

[0065] refer to Figure 2 As shown, the user contact method can be selected through steps S210 to S230:

[0066] In step S210, the user attribute characteristics of the target user are determined, and the contact method characteristics corresponding to the target user are determined; wherein the target user may refer to a user with contact experience, for example, the target user may refer to a user who has been contacted using contact methods such as text messages and phone calls, and the user attribute characteristics include the basic attribute characteristics of the user, and the behavioral attribute characteristics of the user; the contact method characteristics may be telemarketing, text messages, push (application push) and IVR (Interactive Voice Response, interactive voice response), in this embodiment, the contact method characteristics may also be application service notifications, emails, etc., but this is not limited to this in this embodiment. It should be noted that in the embodiment of the present disclosure, the contact method used when performing the target operation on the target user may be used as the contact method characteristic corresponding to the target user. For example, when the recommendation information of product B is pushed to user A via text message, it can be determined that when the product is recommended to user A, the corresponding contact method characteristic is text message push.

[0067] In step S220, the reach mode feature and the user attribute feature are respectively input into the incremental lift value prediction system to obtain the incremental prediction value corresponding to the target user; wherein the incremental lift value prediction system is obtained by adjusting the parameters of the first machine learning model based on the sample features and the sample incremental value, the first machine learning model is constructed based on the second machine learning model and the sample incremental value, the sample incremental value is calculated based on the predicted conversion data, and the predicted conversion data is obtained by inputting the sample features into the second machine learning model;

[0068] Among them, the incremental lift value prediction system can be constructed through sample features and conversion data, and can also be obtained by adjusting the parameters of the first machine learning model. For example, multiple user attribute features can be obtained to form sample features, and the conversion results corresponding to the user attribute features can be used as labels. The sample features and labels are brought into the initial model for training to obtain predicted conversion data, adjust the parameter values ​​of the initial model and the features input into the initial model to obtain the first machine learning model, adjust the parameters of the first machine learning model according to the predicted conversion data and sample features to obtain the incremental lift value prediction system, and finally select the user reach method based on the incremental lift value prediction system. For example, in this embodiment, the reach method features and the user feature features can be input into the already constructed incremental lift value prediction system to obtain the incremental prediction value corresponding to the target user;

[0069] The sample features include sample user features and sample contact mode features. In this example implementation, the user attribute features may include the user's basic attribute features and the user's behavior attribute features. The contact mode features may be phone calls, text messages, push (application push) and ivr (interactive voice response). In this embodiment, the contact mode features may also be application service notifications, emails, etc., but are not limited to this in this embodiment.

[0070] The sample features and the labeled transformation data corresponding to the sample features can be obtained. The labeled transformation data can be represented by 0 and 1, where 1 represents transformation and 0 represents non-transformation. For example, in this embodiment, y i represents the labeled transformation data of the i-th sample (1 means transformation, 0 means no transformation), and can also be represented by x i represents the characteristics of the i-th sample; wherein the sample user characteristics and the sample contact method characteristics are expressed as [X, T], X represents the sample user characteristics, T represents the sample contact method characteristics, and the annotated conversion data is used as the sample label. The sample label Y can refer to whether to apply for a certain product. Of course, other expressions can also be used, which is not limited to this in this embodiment;

[0071] The first machine learning model may be a pre-trained model, and the sample increment value may be brought into the second machine learning model training as a label, and the parameters of the second machine learning model may be adjusted to obtain the first machine learning model;

[0072] The second machine learning model may be an initial model. By inputting sample features and labels into the initial model, predicted conversion data is obtained. For example, the second machine learning model may be a random forest model or a neural network model. In this embodiment, the second machine learning model may be a gradient boosted tree model (XGBoost). The boosted tree model includes multiple tree models, which can be used represents the sum of the prediction results of the first t-1 trees, f t (x i ) represents the t-th tree for sample x i The prediction results; the above sample user characteristics and the above sample contact method characteristics can be input into the XGBoost model to output the corresponding predicted conversion data;

[0073] In this embodiment, the objective function of the XGBoost model can be expressed as:

[0074]

[0075] Where n represents a total of n samples (i, n and t in the above objective function are all integers); Ω(f i ) is a regular term to prevent overfitting.

[0076] For example, in this embodiment, you can also refer to Figure 3 As shown, the predicted conversion data is obtained through the following steps S310 to S340:

[0077] In step S310, sample features and annotated conversion data corresponding to the sample features are obtained; wherein the sample features include sample user features and sample contact method features; wherein the annotated conversion data can be represented by 0 and 1, 1 represents conversion, and 0 represents non-conversion. For example, in this embodiment, y i represents the labeled transformation data of the i-th sample (1 means transformation, 0 means no transformation), and can also be represented by x i Represents the characteristics of the i-th sample; wherein the sample user characteristics and the sample contact method characteristics are expressed as [X, T], X represents the sample user attribute characteristics, T represents the sample contact method characteristics, and the annotated conversion data is used as the sample label. The sample label Y can refer to whether to apply for a certain product. Of course, other expressions can also be used, which is not limited to this in this embodiment.

[0078] In step S320, the sample user features and the sample contact mode features are input into the first decision tree in the second machine learning model, and the objective function of the first decision tree can be obtained according to the above objective function. In this embodiment, the objective function generated by the first decision tree can be expressed as:

[0079]

[0080] For example, in this embodiment, the decision tree may include multiple feature paths, each of which corresponds to a contact method, which may be telephone contact or SMS contact; wherein the multiple feature paths also include multiple feature nodes. In this embodiment, for example, each of the feature paths corresponds to different sample contact method characteristics. For example, the individual feature nodes may include contact method characteristics such as telephone, SMS, application notification, IVR, push, etc.

[0081] The decision tree can obtain output results through steps S410 to S420:

[0082] In step S410, the matching degree between the sample feature and each feature path is calculated; the matching degree can be calculated based on the relationship between the objective functions of adjacent feature nodes. For example, the larger the difference between the conversion values ​​corresponding to the objective functions, the higher the matching degree. For example, when the model structure is known, the conversion value of each feature node can be obtained through the objective function. In a feature path, the matching degree is judged by the absolute difference between the upper and lower adjacent feature nodes. For example, the larger the absolute difference between the upper and lower adjacent feature nodes, the higher the matching degree, because the conversion value of the previous feature node is known. The larger the absolute difference, the smaller the conversion value of the next feature node, which proves that the conversion result is more accurate.

[0083] In step S420, the predicted conversion data of each contact method corresponding to each feature path is calculated according to the matching degree; for example, in this embodiment, assuming that the user is contacted by method 1 (telephone) in the training sample, then the feature is transformed into 0 (no contact), 2 (SMS), 3 (PUSH), 4 (IVR), and the second machine learning model is used to make predictions respectively, so as to obtain the predicted conversion data of the target user under each contact method.

[0084] In step S330, multiple decision trees are iteratively generated in the residual descending direction according to the output result of the previous decision tree and the labeled transformation data; wherein, multiple decision trees can be constructed by the predicted transformation data of the previous decision tree. For example, in this embodiment, the difference between the output predicted transformation data and the labeled transformation data can be calculated, and the difference can be used to move forward in the direction of gradient descent to generate multiple decision trees, but this is not limited to this in this embodiment.

[0085] In step S340, when the number of decision trees generated meets the preset conditions, the output result of the last decision tree is used as the predicted conversion data output by the second machine learning model. Among them, satisfying the preset conditions can refer to the number of decision trees that meet the output of the final predicted conversion data, or it can refer to satisfying the preset accuracy rate. For example, in this embodiment, when the generated decision tree reaches the mth (m is an integer) and meets the preset accuracy rate, the predicted conversion data of the last tree can be output; it can be understood that in this embodiment, the purpose of satisfying the preset conditions is to make the output predicted conversion data more accurate, so it is not limited to this in this embodiment.

[0086] In step S230, a target reaching method is determined for the target user according to the incremental predicted value corresponding to the target user; wherein, a reaching method suitable for the user can be selected according to the incremental predicted value. For example, in this embodiment, i can represent the i-th user, j represents the j-th reaching method, wherein j=0 represents no reaching; in this embodiment, the incremental predicted value can be calculated according to the following objective function:

[0087] uplift(i,j)=G(Y i |X i ,T=j)-G(Y i |X i ,T=0)sti∈(0,n),j∈[1,N]

[0088] Among them, G(Y i |X i ,T=j) represents the model prediction conversion data of the i-th user under the j-th contact method. If the training sample contains the j-th contact method of the i-th user, then G(Y i |X i ,T=j)=Y(i,j); uplift(i,j) represents the incremental prediction value of user i in the jth contact method. For example, the "5" and "2" in uplift(5,2) correspond to the "i" and "j", where "5" can be understood as the fifth user and "2" can be understood as the second contact method, which can be SMS contact. G(Y 5 |X 5 ,T=2) can refer to the conversion probability predicted by the model for the fifth user under the second contact method. If the training sample contains the second contact method for the fifth user, then G(Y 5 |X 5 ,T=2) can be expressed as Y(5,2);

[0089] Wherein, the first machine learning model is constructed based on the second machine learning model and the sample incremental value; wherein the incremental improvement value prediction system may include multiple tree models, which can be used represents the sum of the incremental prediction results of the first t-1 trees, f t (x i ) represents the t-th tree for sample x i The incremental prediction value result of the first machine learning model can be an objective function parameter or a parameter on a feature node. For example, in this embodiment, the output result can be made more accurate by adjusting the parameters of the feature nodes on the tree model. The incremental improvement value prediction system objective function can be expressed as:

[0090]

[0091] Among them, x i represents the characteristics of the i-th sample, y i represents the sample increment value of the i-th sample, n represents a total of n samples (n, i and t are all integers), Ω(f i ) is a regular term to prevent overfitting.

[0092] In this embodiment, reference Figure 5 As shown, the incremental lift value prediction system can be further optimized through steps S510 to S520, which can be by taking the sample incremental value as label y, and the user attribute characteristics and contact method characteristics as features x into the model and training the model, or by optimizing the model according to the loss function of the incremental lift value prediction system, which is not limited to this in this embodiment.

[0093] In step S510, a loss function is calculated according to the sample increment value; wherein the loss function can be used to represent the degree of inconsistency between the sample increment value and the increment prediction value of the i-th tree. For example, the loss function can be calculated by the difference between the sample increment value and the increment prediction value, the mean square error, etc. In this embodiment, the loss function can be expressed as Among them, y i Can represent sample increment value, It can represent the incremental prediction result of the i-th tree.

[0094] In step S520, the parameters of the first machine learning model are adjusted according to the loss function, and the adjusted first machine learning model is used as an incremental improvement value prediction system; wherein the parameters may be parameters in the objective function of the first machine learning model or parameters on feature nodes. In this embodiment, by adjusting the parameters on the feature nodes of the first machine learning model, the model output results can be made more accurate, and the adjusted model can be used as an incremental improvement value prediction system.

[0095] In this embodiment, a target reaching method is determined for the target user based on the incremental prediction value corresponding to the target user, including: selecting the reaching method corresponding to the incremental improvement value whose incremental improvement value reaches a preset value for reaching the target user; and / or selecting the reaching method corresponding to the maximum incremental improvement value of the incremental improvement value for reaching the target user.

[0096] For example, in this embodiment, the preset value may be 0. If the maximum incremental prediction value is greater than 0, the reaching method corresponding to the maximum incremental prediction value is selected to reach the user. For example, multiple incremental prediction values ​​are output through the incremental improvement value prediction system. Assume that the multiple incremental prediction values ​​output are 0.000, -1.017, 0.022, -0.011, and 0.007, respectively. Then 0.022 is the maximum incremental prediction value. In addition, the maximum incremental prediction value may be multiple. For example, a set of incremental prediction values ​​obtained includes 0.000, -1.017, 0.022, 0.022, and 0.007, including two maximum incremental prediction values ​​of 0.022. Then, the reaching methods corresponding to the two maximum incremental prediction values ​​may be selected at the same time for reaching. In another embodiment, for a new user, N samples {[X,T 1 ],[X,T 2 ],…,[X,T N ]}, wherein N represents the number of reaching methods. The trained incremental boosting model is used to predict N samples respectively to obtain N incremental prediction values. If the maximum gain value is greater than 0, the reaching method corresponding to the maximum gain value is selected to reach the user; otherwise, the user does not need to be reached. This is not limited to the present embodiment.

[0097] This embodiment also proposes a method for obtaining an incremental improvement value prediction system, wherein the incremental improvement value prediction system can be used to determine the way to reach the user, referring to Figure 6 As shown, the incremental improvement value prediction system acquisition method includes steps S610 to S640:

[0098] In step S610, sample features and annotated conversion data corresponding to the sample features are obtained; wherein the sample features include sample user features and sample contact mode features; in this example implementation, the sample user features may include basic attribute features of the user and behavioral attribute features of the user; the contact mode features may be phone calls, text messages, push (application push) and ivr (interactive voice response). In this embodiment, the contact mode features may also be application service notifications, emails, etc., but are not limited thereto in this embodiment;

[0099] The labeled conversion data can be represented by 0 and 1, 1 represents conversion, and 0 represents non-conversion. For example, in this embodiment, y i represents the labeled transformation data of the i-th sample (1 means transformation, 0 means no transformation), and can also be represented by x i Represents the characteristics of the i-th sample; wherein the sample user characteristics and the sample reach characteristics are expressed as [X, T], X represents the sample user characteristics, T represents the sample reach characteristics, and the annotated conversion data is used as the sample label. The sample label Y can refer to whether to apply for a certain product. Of course, other expressions can also be used, which is not limited to this in this embodiment.

[0100] In step S620, the sample user features and the sample contact mode features are input into a second machine learning model to output corresponding predicted conversion data; wherein the second machine learning model may be a random forest model or a neural network model, etc. In this embodiment, the second machine learning model may be a gradient boosting tree model (XGBoost), and the boosting tree model includes multiple tree models, which can be used represents the sum of the prediction results of the first t-1 trees, f t (x i ) represents the t-th tree for sample x i The prediction results; the above sample user characteristics and the above sample contact method characteristics can be input into the XGB oost model to output the corresponding predicted conversion data;

[0101] In this embodiment, the objective function of the XGBoost model can be expressed as:

[0102]

[0103] Where n represents a total of n samples (i, n and t in the above objective function are all integers); Ω(f i ) is a regular term to prevent overfitting.

[0104] In step S630, the sample increment value corresponding to the sample user feature is calculated according to the predicted conversion data; wherein, according to the sample increment value, a suitable contact method for the user can be selected. For example, in this embodiment, i can represent the i-th user, j represents the j-th contact method, wherein j=0 represents no contact; in this embodiment, the sample increment value can be calculated according to the following objective function:

[0105] uplift(i,j)=G(Y i |X i ,T=j)-G(Y i |X i,T=0)sti∈(0,n),j∈[1,N]

[0106] Among them, G(Y i |X i ,T=j) represents the model prediction conversion data of the i-th user under the j-th contact method. If the training sample contains the j-th contact method of the i-th user, then G(Y i |X i ,T=j)=Y(i,j); uplift(i,j) represents the increment of user i in the jth contact method. For example, the "5" and "2" in uplift(5,2) correspond to the "i" and "j", where "5" can be understood as the fifth user and "2" can be understood as the second contact method, which can be SMS contact. G(Y 5 |X 5 ,T=2) can refer to the conversion probability predicted by the model for the fifth user under the second contact method. If the training sample contains the second contact method for the fifth user, then G(Y 5 |X 5 ,T=2) can be expressed as Y(5,2).

[0107] In step S640, the parameters of the first machine learning model are adjusted according to the sample characteristics and the sample incremental value to obtain an incremental improvement value prediction system; wherein the first machine learning model is constructed based on the second machine learning model and the sample incremental value; wherein the incremental improvement value prediction system may include multiple tree models, which may be used represents the sum of the predicted sample increment values ​​of the first t-1 trees, f t (x i ) represents the t-th tree for sample x i The predicted sample incremental value result; the parameters of the first machine learning model can be objective function parameters or parameters on feature nodes. For example, in this embodiment, the output result can be made more accurate by adjusting the parameters of the feature nodes on the tree model. The incremental improvement value prediction system objective function can be expressed as:

[0108]

[0109] Among them, x i represents the characteristics of the i-th sample, y i represents the sample increment value of the i-th sample, n represents a total of n samples (n, i and t are all integers); Ω(f i ) is a regular term to prevent overfitting.

[0110] based on Figures 2 to 5 The method for determining the user reach mode shown, and Figure 6The incremental improvement value prediction system acquisition method shown in FIG. 1 may, in one embodiment of the present invention, Figures 2 to 6 The technical solutions shown in the figure are combined and applied to the terminal device, as shown in the figure. Figure 7 As shown, the following steps are included:

[0111] Step S710, obtaining sample features and annotated conversion data corresponding to the sample features; wherein the sample features include sample user features and sample contact method features;

[0112] Step S720, inputting the sample user features and the sample contact mode features into a second machine learning model to output corresponding predicted conversion data, and calculating the sample increment value through the predicted conversion data;

[0113] Step S730, adjusting the parameters of the first machine learning model according to the sample characteristics and the sample incremental value to obtain an incremental improvement value prediction system;

[0114] Step S740, determining the user attribute characteristics of the target user, and determining the contact method characteristics corresponding to the target user;

[0115] Step S750, inputting the reach mode feature and the user attribute feature into an incremental lift value prediction system respectively to obtain an incremental prediction value corresponding to the target user;

[0116] Step S760, determining whether the maximum increment value among the N increment values ​​is greater than 0;

[0117] Step S770: if the maximum increment value is greater than 0, output the touch mode corresponding to the maximum increment value;

[0118] Step S780: If the maximum increment value is less than 0, there is no need to output the access mode.

[0119] In summary, in the user reach determination method provided in the exemplary implementation of the present disclosure, the user attribute characteristics of the target user can be determined, and the reach characteristics corresponding to the target user can be determined; the reach characteristics and the user attribute characteristics are respectively input into the incremental lift value prediction system to obtain the incremental prediction value corresponding to the target user; wherein the incremental lift value prediction system is obtained by adjusting the parameters of the first machine learning model based on the sample characteristics and the sample incremental value, the first machine learning model is constructed based on the second machine learning model and the sample incremental value, the sample incremental value is calculated based on the predicted conversion data, and the predicted conversion data is obtained by inputting the sample characteristics into the second machine learning model; according to the incremental prediction value corresponding to the target user, the target reach method is determined for the target user. On the one hand, by obtaining the incremental prediction value of the target user under different reach methods through the incremental lift value prediction system, the error accumulation caused by multiple models can be effectively avoided, thereby reducing the amount of model calculation, which is conducive to improving the accuracy of reach method allocation. On the other hand, since the accuracy of reach method allocation is improved, the efficiency of user reach can be improved, thereby improving the overall reach effect.

[0120] Figure 8 A block diagram of a device for determining a user access mode according to another embodiment of the present invention is schematically shown.

[0121] Reference Figure 8 As shown, a device 800 for determining a user reach mode according to an embodiment of the present invention includes a user feature acquisition module 810, an incremental prediction value acquisition module 820, and a reach mode determination module 830.

[0122] The user feature acquisition module 810 is used to determine the user attribute characteristics of the target user and determine the contact method characteristics corresponding to the target user;

[0123] The incremental prediction value acquisition module 820 is used to input the contact mode feature and the user attribute feature into the incremental lift value prediction system respectively to obtain the incremental prediction value corresponding to the target user; wherein the incremental lift value prediction system is obtained by adjusting the parameters of the first machine learning model based on the sample features and the sample incremental value, the first machine learning model is constructed based on the second machine learning model and the sample incremental value, the sample incremental value is calculated based on the predicted conversion data, and the predicted conversion data is obtained by inputting the sample features into the second machine learning model;

[0124] The reaching method determination module 830 is used to determine a target reaching method for the target user based on the incremental prediction value corresponding to the target user.

[0125] Fig. 9 A block diagram of an incremental improvement value prediction system acquisition device according to an embodiment of the present invention is schematically shown.

[0126] Reference Fig. 9 As shown, an incremental lift value prediction system acquisition device 900 according to an embodiment of the present invention includes a sample feature acquisition and annotation conversion data module 910, a prediction conversion data output module 920, an incremental lift value calculation module 930, and an incremental lift prediction system construction module 940.

[0127] The module 910 for obtaining sample features and annotated conversion data is used to obtain sample features and annotated conversion data corresponding to the sample features; wherein the sample features include sample user features and sample contact method features;

[0128] A predicted conversion data output module 920, configured to input the sample user features and the sample contact mode features into a second machine learning model to output corresponding predicted conversion data;

[0129] An incremental improvement value calculation module 930 is used to calculate the sample incremental value corresponding to the sample user feature according to the predicted conversion data;

[0130] The incremental improvement prediction system construction module 940 is used to adjust the parameters of the first machine learning model according to the sample characteristics and the sample incremental value to obtain an incremental improvement value prediction system; wherein, the first machine learning model is constructed based on the second machine learning model and the sample incremental value.

[0131] In an exemplary embodiment of the present disclosure, the predicted conversion data output module 920 includes:

[0132] A decision tree generation unit: used to input the sample user features and the sample contact mode features into the first decision tree in the second machine learning model; and iteratively generate multiple decision trees in the direction of residual descent according to the output result of the previous decision tree and the labeled transformation data;

[0133] Prediction conversion data output unit: used for taking the output result of the last decision tree as the prediction conversion data output by the second machine learning model when the number of the generated decision trees meets a preset condition;

[0134] Calculation unit: used to calculate the matching degree between the sample features and each feature path, and to calculate the predicted conversion data of each access method corresponding to each feature path according to the matching degree.

[0135] In an exemplary embodiment of the present disclosure, the incremental improvement prediction system building module 940 includes:

[0136] A loss function calculation unit: used to calculate the loss function according to the sample increment value;

[0137] A model parameter adjustment unit is used to adjust the parameters of the first machine learning model according to the loss function, and use the adjusted first machine learning model as an incremental improvement value prediction system.

[0138] Fig.10 A schematic diagram of the structure of a computer model of an electronic device suitable for implementing an embodiment of the present disclosure is shown.

[0139] It should be noted that Fig.10 The computer model 1000 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0140] like Fig.10 As shown, the computer model 1000 includes a central processing unit (CPU) 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage part 1008 into a random access memory (RAM) 1003. Various programs and data required for model operation are also stored in the RAM 1003. The CPU 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0141] The following components are connected to the I / O interface 1005: an input section 1006 including a keyboard, a mouse, etc.; an output section 1007 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN card, a modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I / O interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1010 as needed, so that a computer program read therefrom is installed into the storage section 1008 as needed.

[0142] In particular, according to an embodiment of the present disclosure, the process described below with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 1009, and / or installed from the removable medium 1011. When the computer program is executed by the central processing unit (CPU) 1001, various functions defined in the method and apparatus of the present application are executed. In some embodiments, the computer model 1000 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0143] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the implementation of the present invention.

[0144] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for determining a user contact method, It is characterized in that include: Determine user attribute characteristics of a target user, and determine the contact method used by the target user when performing a target operation as a contact method characteristic corresponding to the target user, wherein the user attribute characteristics include basic attribute characteristics of the user and behavioral attribute characteristics of the user; The reach mode feature and the user attribute feature are respectively input into the incremental lift value prediction system to obtain the incremental prediction value corresponding to the target user; wherein the incremental lift value prediction system is obtained by adjusting the parameters of the first machine learning model based on the sample features and the sample incremental value, the first machine learning model is obtained by adjusting the parameters of the second machine learning model based on the sample incremental value, the sample incremental value is calculated based on the difference between the predicted conversion data of the target user under the jth reach mode and the predicted conversion data of the target user under the no-reach mode, and the predicted conversion data is obtained by inputting the sample features into the second machine learning model; A target reaching method is determined for the target user based on the incremental prediction value corresponding to the target user.

2. The method for determining a user contact mode according to claim 1, It is characterized in that The determining a target reaching method for the target user according to the incremental prediction value corresponding to the target user includes: Select the access method corresponding to the incremental prediction value whose incremental prediction value reaches the preset value for access; and / or The reaching method corresponding to the maximum incremental prediction value of the incremental prediction value is selected for reaching.

3. The method for determining a user contact mode according to claim 1, It is characterized in that The second machine learning model is a gradient boosting tree model, and the method further includes: Obtaining sample features and annotated conversion data corresponding to the sample features; wherein the sample features include sample user features and sample contact method features; Inputting the sample user features and the sample contact mode features into the first decision tree in the second machine learning model; Iteratively generate multiple decision trees in a residual descending direction according to the output result of the previous decision tree and the labeled transformation data; When the number of decision trees generated meets the preset conditions, the output result of the last decision tree is used as the predicted conversion data output by the second machine learning model.

4. The method for determining a user contact mode according to claim 3, It is characterized in that The decision tree includes multiple characteristic paths, each of which corresponds to a contact mode; the decision tree obtains the output result in the following way: Calculating the matching degree between the sample feature and each feature path; The predicted conversion data of each of the access methods corresponding to each characteristic path is calculated based on the matching degree.

5. The method for determining a user contact mode according to claim 1, It is characterized in that The method further comprises: Calculate the loss function according to the sample increment value; The parameters of the first machine learning model are adjusted according to the loss function, and the adjusted first machine learning model is used as an incremental improvement value prediction system.

6. A method for obtaining an incremental improvement value prediction system, It is characterized in that include: Obtaining sample features and annotated conversion data corresponding to the sample features; wherein the sample features include sample user features and sample contact method features; Inputting the sample user features and the sample contact method features into a second machine learning model to output corresponding predicted conversion data; Calculate the sample increment value corresponding to the sample user feature according to the predicted conversion data; the sample increment value is calculated based on the difference between the predicted conversion data of the target user among the sample users under the jth contact method and the predicted conversion data of the target user under the non-contact method; The parameters of the first machine learning model are adjusted according to the sample characteristics and the sample incremental value to obtain an incremental improvement value prediction system; wherein the first machine learning model is obtained by adjusting the parameters of the second machine learning model based on the sample incremental value.

7. A device for determining a user contact mode, It is characterized in that include: A user feature acquisition module is used to determine user attribute features of a target user, and determine the access method used by the target user when performing a target operation as the access method feature corresponding to the target user, wherein the user attribute features include basic attribute features of the user and behavioral attribute features of the user; An incremental prediction value acquisition module, for inputting the reach mode feature and the user attribute feature into an incremental lift value prediction system respectively, to obtain an incremental prediction value corresponding to the target user; wherein the incremental lift value prediction system is obtained by adjusting the parameters of a first machine learning model based on sample features and sample incremental values, the first machine learning model is obtained by adjusting the parameters of a second machine learning model based on sample incremental values, the sample incremental value is calculated based on the difference between the predicted conversion data of the target user under the jth reach mode and the predicted conversion data of the target user under a no-reach mode, and the predicted conversion data is obtained by inputting the sample features into the second machine learning model; A module for determining a reaching method is used to determine a target reaching method for the target user based on the incremental prediction value corresponding to the target user.

8. An incremental improvement value prediction system acquisition device, It is characterized in that include: A module for obtaining sample features and annotated conversion data is used to obtain sample features and annotated conversion data corresponding to the sample features; wherein the sample features include sample user features and sample contact method features; A predicted conversion data output module, used to input the sample user characteristics and the sample contact mode characteristics into a second machine learning model to output corresponding predicted conversion data; An incremental improvement value calculation module is used to calculate the sample incremental value corresponding to the sample user feature according to the predicted conversion data; the sample incremental value is calculated based on the difference between the predicted conversion data of the target user among the sample users under the jth contact method and the predicted conversion data of the target user under the non-contact method; An incremental improvement prediction system building module is used to adjust the parameters of the first machine learning model according to the sample characteristics and the incremental improvement value to obtain an incremental improvement value prediction system; wherein the first machine learning model is obtained by adjusting the parameters of the second machine learning model based on the sample incremental value.

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

10. An electronic device, It is characterized in that include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to perform the method of any one of claims 1 to 6 by executing the executable instructions.

Images