Marketing model construction method and device, electronic equipment and storage medium

Abstract

The invention discloses a marketing model construction method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring business attribute characteristics of users in a historical marketing channel; obtaining business participation characteristics of users in the historical marketing channels; splicing the service attribute features and the service participation features to obtain marketing sorting features; obtaining marketing prediction parameters based on the marketing sorting features, and performing marketing putting based on the marketing prediction parameters to obtain an expected business behavior sample; and performing model fine tuning based on the expected business behavior sample to obtain a target marketing model. According to the method, the marketing prediction parameters in the new channel can be obtained by using the service attribute features and the service participation features of the users in the existing channel, and then the marketing model is constructed, so that the problem that the new marketing channel is difficult to model due to lack of modeling samples can be effectively avoided.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and more specifically, to a marketing model building method, device, electronic device and storage medium. Background Art

[0002] Marketing models refer to marketing model tools that are selected and operated in different scenarios of an enterprise. In the current marketing model field, user marketing models are usually built based on the business behavior data of existing users. The marketing models built in this way can predict the probability of users' expected business behaviors, thereby providing a decision-making basis for formulating accurate marketing strategies. However, for newly developed marketing channels, due to the lack of business tags, it is difficult to achieve effective model construction even if users already have a large amount of behavior data in the channel. Summary of the invention

[0003] This application proposes a marketing model construction method, device, electronic device and storage medium to improve the above-mentioned problems.

[0004] In a first aspect, an embodiment of the present application provides a marketing model construction method that can be applied to electronic devices, the method comprising: obtaining business attribute characteristics of users in historical marketing channels; obtaining business participation characteristics of users in the historical marketing channels; splicing the business attribute characteristics with the business participation characteristics to obtain marketing ranking characteristics; obtaining marketing prediction parameters based on the marketing ranking characteristics, and conducting marketing delivery based on the marketing prediction parameters to obtain expected business behavior samples; and fine-tuning the model based on the expected business behavior samples to obtain a target marketing model.

[0005] In a second aspect, an embodiment of the present application provides a marketing model building device that can run on an electronic device, the device comprising: a first acquisition module, used to acquire business attribute characteristics of users in historical marketing channels; a second acquisition module, used to acquire business participation characteristics of users in the historical marketing channels; a splicing module, used to splice the business attribute characteristics with the business participation characteristics to obtain marketing ranking characteristics; a parameter acquisition module, used to acquire marketing prediction parameters based on the marketing ranking characteristics, and perform marketing delivery based on the marketing prediction parameters to obtain expected business behavior samples; a model building module, used to fine-tune the model based on the expected business behavior samples to obtain a target marketing model.

[0006] In a third aspect, the present application provides an electronic device comprising one or more processors and a memory; one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method of the first aspect above.

[0007] In a fourth aspect, the present application provides a computer-readable storage medium having program code stored therein, wherein the method of the first aspect is executed when the program code is run.

[0008] The present application provides a method, device, electronic device and storage medium for constructing a marketing model. The method obtains the business attribute characteristics of users in historical marketing channels; obtains the business participation characteristics of users in the historical marketing channels; splices the business attribute characteristics with the business participation characteristics to obtain marketing ranking characteristics; obtains marketing prediction parameters based on the marketing ranking characteristics, and conducts marketing delivery based on the marketing prediction parameters to obtain expected business behavior samples; and fine-tunes the model based on the expected business behavior samples to obtain a target marketing model. Thus, the above method can achieve the construction of a marketing model by using the business attribute characteristics and business participation characteristics of users in existing channels to obtain marketing prediction parameters in new channels, which can effectively avoid the problem that new marketing channels are difficult to model due to lack of modeling samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A flowchart of a marketing model building method provided by an embodiment of the present application is shown.

[0011] Figure 2 A method flow chart of step S110 in an embodiment of the present application is shown.

[0012] Figure 3 A method flow chart of step S120 in an embodiment of the present application is shown.

[0013] Figure 4 An example flowchart of a marketing model building method provided in an embodiment of the present application is shown.

[0014] Figure 5 A structural block diagram of a marketing model building device provided in an embodiment of the present application is shown.

[0015] Figure 6 A structural block diagram of an electronic device provided in an embodiment of the present application is shown.

[0016] Figure 7A storage unit for storing or carrying program codes for implementing the marketing model building method according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0017] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0018] In the current marketing model field, many technical solutions are dedicated to building user marketing models based on existing user business behavior data. Specifically, this type of technical solution collects user business behavior information in specific marketing channels, including but not limited to clicks, registrations, real names, product purchases, etc., and then builds marketing models through data mining and machine learning. This model can predict the probability of users' expected business behaviors, thereby providing a decision-making basis for formulating accurate marketing strategies. However, for newly developed marketing channels, due to the lack of business tags, it is difficult to achieve effective model construction even if users already have a large amount of behavior data in the channel.

[0019] After long-term research, the inventor found that the business attribute characteristics of users in historical marketing channels can be obtained; the business participation characteristics of users in the historical marketing channels can be obtained; the business attribute characteristics and the business participation characteristics can be spliced ​​to obtain marketing ranking characteristics; marketing prediction parameters can be obtained based on the marketing ranking characteristics, and marketing can be launched based on the marketing prediction parameters to obtain expected business behavior samples; model fine-tuning can be performed based on the expected business behavior samples to obtain the target marketing model. Thus, the above method can achieve the construction of a marketing model by using the business attribute characteristics and business participation characteristics of users in existing channels to obtain marketing prediction parameters in new channels, which can effectively avoid the problem that new marketing channels are difficult to model due to lack of modeling samples.

[0020] Therefore, in order to improve the above-mentioned problems, the inventor proposed the marketing model construction method, device, electronic device and computer-readable storage medium provided in the present application, which can obtain marketing prediction parameters in new channels by utilizing the business attribute characteristics and business participation characteristics of users in existing channels, and then build a marketing model, which can effectively avoid the problem that new marketing channels are difficult to model due to lack of modeling samples.

[0021] In order to facilitate a better understanding of the solution described in the embodiments of the present application, the following briefly describes the terms involved in the description of the embodiments of the present application:

[0022] User Attribute Module: refers to the module in the model that is used to characterize the user's business attributes and determine whether they are target customers.

[0023] Channel Attribute Module: refers to the module in the model used to measure the user's behavioral intention in the current channel.

[0024] Mixture-of-Experts Module: A module that decomposes the predictive modeling task into several subtasks, trains an expert model on each subtask, develops a gating model that learns which expert to trust based on the input to be predicted, and combines the prediction results.

[0025] Cold-Start: It refers to the process of building a business model from 0 to 1 when there is no or very little positive data samples.

[0026] The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0027] See also Figure 1 , shows a flowchart of a marketing model construction method provided by an embodiment of the present application. This embodiment provides a marketing model construction method, which can be applied to electronic devices, and the electronic devices can be mobile phones, computers, tablets or wearable electronic devices. The specific type of electronic devices may not be limited. The method includes:

[0028] Step S110: Obtaining business attribute characteristics of users in historical marketing channels.

[0029] The historical marketing channel may include one marketing channel or multiple marketing channels. When the historical marketing channel includes multiple marketing channels, the multiple marketing channels may be different types of marketing channels. In this way, the marketing model of the same business field may be applicable to the different types of marketing channels.

[0030] Generally speaking, historical marketing channels develop early and have accumulated a large number of data samples, which can be used to build marketing models for any expected business behavior. However, newly expanded marketing channels lack sufficient modeling samples, resulting in low traffic utilization efficiency in the early stages of development of new marketing channels.

[0031] In order to optimize the above problems, this implementation method constructs a marketing model that can be flexibly spliced ​​and combined, and the marketing model includes a user attribute module, a channel attribute module, and an expert gating module. Among them, the user attribute module can select some models built in the existing earlier developed marketing channels; the channel attribute module can model the channel user behavior alone, or draw on the models of other business fields of this channel to evaluate indicators such as user interaction, participation, and conversion rate in the current channel; the expert gating module can combine the outputs of the user attribute module and the channel attribute module to obtain the final output result, and perform marketing ranking based on the final output result (that is, measure the probability that the user will respond to the marketing activity and rank the probability).

[0032] As an implementation method, the service attribute characteristics of the user may include user attribute data. The user attribute data may be some objective user information, for example, the user attribute data may include the user's age, gender, occupation, income, assets, etc. The specific content of the user attribute data may not be limited.

[0033] As an implementation, see Figure 2 Step S110 may specifically include:

[0034] Step S111: Acquire user attribute data in historical marketing channels, wherein the user attribute data includes first numerical feature data, first category feature data, and first sequence feature data.

[0035] In order to better obtain the user attribute data in the business attribute characteristics, the user attribute data can be divided into first numerical feature data, first category feature data and first sequence feature data. It should be noted that the first here does not represent the order. Among them, the first numerical feature data represents the user attribute data of the numerical type. For example, the first numerical feature data can be the user's age data. The first category feature data represents the user attribute data of the category type. For example, the first category feature data can be the user's gender data. The first sequence feature data represents the user attribute data of the sequence type. For example, whether the user starts the mobile phone every day within a month, assuming that 0 means not started and 1 means started, then a string of 30 "......01010......." can be formed in one month, and the string is the user attribute data of the sequence type.

[0036] As an implementation mode, user attribute data may be directly obtained from historical marketing channels to obtain first numerical feature data, first category feature data, and first sequence feature data.

[0037] Step S112: splicing the first numerical feature data, the first category feature data and the first sequence feature data and inputting them into a user attribute model of an existing channel, wherein the user attribute model is used to evaluate the business attributes of the user.

[0038] As an implementation method, after obtaining the first numerical feature data, the first category feature data and the first sequence feature data, the first numerical feature data, the first category feature data and the first sequence feature data can be spliced, and then the spliced ​​vector can be input into the user attribute model of the existing channel to evaluate the user's business attributes in the current channel (which can be understood as a newly constructed channel) through the user attribute model.

[0039] As an implementation method, before the first numerical feature data, the first category feature data and the first sequence feature data are concatenated and input into the user attribute model of the existing channel, the first numerical feature data can also be input into the fully connected layer, and the first feature vector output by the fully connected layer is obtained; the first category feature data is input into the embedding layer, and the second feature vector output by the embedding layer is obtained, where the input of the embedding layer is the number of category features, and the output is the set embedding dimension. The role of the embedding layer is to map each category feature to a continuous vector space to better capture the similarity between category features; and the first sequence feature data is passed through the embedding layer (the embedding layer is not the same as the embedding layer passed by the first category feature data) and the attention layer, and the third feature vector output by the attention layer is obtained, where the attention layer uses an attention mechanism to process the embedded sequence features, and the role of the attention mechanism is to calculate the contribution of each element in the sequence to the final result, so that the model can pay more attention to those important elements.

[0040] In the process of passing the first sequence feature data through the embedding layer and the attention layer respectively, after the first sequence feature data passes through the embedding layer, the output of the embedding layer can be passed through a fully connected layer, and then the output of the fully connected layer can be passed through the attention layer to realize the automatic generation of the weight of each element in the sequence feature through the fully connected layer. It should be noted that when the output of the embedding layer passes through the fully connected layer, it can also pass through the softmax function at the same time. The role of the softmax function here is to make all weight values ​​add up to 1. Optionally, it can also be another activation function, which is not specifically limited. The attention layer is used to weight all elements in the sequence according to these weight values ​​to obtain the final output of the sequence feature (i.e., the third feature vector mentioned above).

[0041] In the above manner, the first numerical feature data, the first category feature data and the first sequence feature data can be spliced ​​and input into the user attribute model of the existing channel by splicing the first feature vector, the second feature vector and the third feature vector into the user attribute model of the existing channel.

[0042] In this embodiment, the number of user attribute models of the existing channel may be at least one. In this manner, the splicing of the first feature vector, the second feature vector, and the third feature vector into the user attribute model of the existing channel may include: splicing the first feature vector, the second feature vector, and the third feature vector into at least one user attribute model of the existing channel. At this time, if the number of user attribute models of the existing channel is one, then the business attribute features output by the user attribute model of the existing channel may be directly obtained; and if the user attribute model of the existing channel includes multiple (the specific values ​​of the multiple may not be limited, and the user may flexibly select the number as needed), then the business attribute features output by the user attribute model may include: obtaining the user attribute vectors output by each of the multiple user attribute models of the existing channel to obtain multiple user attribute vectors; respectively inputting the multiple user attribute vectors into the attention layer, obtaining the business attribute features output by the attention layer corresponding to each of the multiple user attribute vectors, and then splicing the business attribute features corresponding to each of the multiple user attribute vectors to obtain the final business attribute features.

[0043] Step S113: Acquire the business attribute features output by the user attribute model.

[0044] Among them, the user attribute model of the existing channel represents the user attribute model of the previously developed mature channel model. If there is only one user attribute model of the existing channel, the business attribute characteristics of the output of the user attribute model can be directly obtained; if there are multiple user attribute models of the existing channel, the outputs of multiple user attribute models can be input into the attention layer, and these multiple outputs are weighted by the attention layer to obtain a weighted output, and then the weighted output is used as the final output of the user attribute module of the new channel model.

[0045] Step S120: Acquire the business participation characteristics of users in the historical marketing channel.

[0046] Among them, the business participation feature can be understood as the business behavior data of the user in the historical marketing channel. The business behavior data represents the specific business behavior of the user. For example, the business behavior data may include information such as "whether the user activates the mobile phone every day within a month" and "whether the time period when the user activates the mobile phone is the same each time". The specific content of the business behavior data may not be limited.

[0047] As an implementation, see Figure 3 , step S120 may include:

[0048] Step S121: Acquire channel attribute data in historical marketing channels, wherein the channel attribute data includes second numerical feature data, second category feature data, and second sequence feature data.

[0049] Among them, channel attribute data can be understood as the business behavior data of users in a specific channel. Similar to user attribute data, channel attribute data can also be divided into second numerical characteristic data, second category characteristic data and second sequence characteristic data. It should be noted that the second here does not represent the order. Among them, the second numerical characteristic data represents the business behavior data of the numerical type, the second category characteristic data represents the business behavior data of the category type, and the second sequence characteristic data represents the business behavior data of the sequence type.

[0050] As an implementation method, the user's business behavior data may be directly obtained from historical marketing channels to obtain the second numerical feature data, the second category feature data, and the second sequence feature data.

[0051] Step S122: inputting the second numerical feature data, the second category feature data and the second sequence feature data into a channel attribute model of a mature channel, wherein the channel attribute model is used to evaluate the user's behavioral intention in the current channel.

[0052] As an implementation mode, after obtaining the second numerical feature data, the second category feature data and the second sequence feature data, the second numerical feature data, the second category feature data and the second sequence feature data can be directly spliced ​​and input into the channel attribute model of the mature channel, so as to evaluate the user's behavioral intention in the current new marketing channel through the channel attribute model of the mature channel. The number of channel attribute models here is not limited, and can be one or more. This implementation mode is described by taking the number of channel attribute models as one as an example.

[0053] It should be noted that the implementation method of the overall structure of the channel attribute module is similar to that of the user attribute module. The difference lies in the source of the model structure of the channel attribute module, which can be derived from the channel attribute module of the mature channel model in other business areas, or channel activity can be used as a sample label for modeling (channel activity can be understood as the attribute of the user in this channel, and the specific definition of channel activity can be determined based on business practice).

[0054] Then, as an implementation method, if there is no channel attribute model of a mature channel, the inputting of the second numerical feature data, the second category feature data and the second sequence feature data into the channel attribute model of a mature channel may include: inputting the second numerical feature data, the second category feature data and the second sequence feature data into a newly created model, the newly created model is modeled with channel activity as the Y label, and the channel activity represents a customized user behavior data selection rule.

[0055] Step S123: Acquire the business participation characteristics output by the channel attribute model.

[0056] As an implementation method, if there is a channel attribute model of a mature channel in the channel attribute module, then the business participation characteristics output by the channel attribute model of the mature channel can be obtained as a reference for the user's business behavior in the current marketing channel; if there is no channel attribute model of a mature channel in the channel attribute module, then the business participation characteristics output by the newly created channel attribute model can be obtained, and the business participation characteristics can be used as the user's business behavior characteristics in the current marketing channel.

[0057] Step S130: concatenate the business attribute feature with the business participation feature to obtain a marketing ranking feature.

[0058] As an implementation method, the output dimension of the business attribute feature finally output by the user attribute model should be the same as the output dimension of the business participation feature finally output by the channel attribute model, so that the business attribute feature and the business participation feature can be directly spliced ​​to obtain the marketing ranking feature. The marketing ranking feature will serve as the input of the expert gating module.

[0059] Step S140: obtaining marketing prediction parameters based on the marketing ranking features, and performing marketing delivery based on the marketing prediction parameters to obtain expected business behavior samples.

[0060] In this embodiment, the expert gating module (type) includes multiple expert models and a gating model, and each expert model and the gating model is a fully connected layer. As an embodiment, the marketing ranking features obtained above can be input into the expert gating model, and then the prediction parameters output by each of the multiple expert models can be obtained, and the weight prediction parameters for each expert model output by the gating model can be obtained, and then the prediction parameters are combined with the weight prediction parameters to obtain the marketing prediction parameters. The outputs of each expert are combined according to the weight of the gating network through the gating model, so that the model can automatically select and combine the predictions of different experts, thereby improving the accuracy and generalization ability of the model.

[0061] It should be noted that in the initial stage of obtaining marketing prediction parameters based on the marketing ranking features, since the user attribute module and the channel attribute module are extracted from the models of historical mature channels, the specific parameters of the expert gating module cannot be obtained through training. Therefore, the following two methods can be used to set the initial weight of the expert gating module:

[0062] (1) Historical model weight initialization: Based on the weight parameters of the historical model in the expert gating module of this business field, these historical weights can be used as the initial weights of the expert gating module. By using these historical weights, the expert gating model can produce reasonable outputs in the initial stage and be optimized in the subsequent fine-tuning process. This method is suitable for situations where there are a large number of historical models available for reference in the same or similar tasks. By using the weights of the historical models, the model can have a certain degree of prediction accuracy in the initial stage, thereby accelerating the convergence of the model.

[0063] (2) Equal weight initialization: When setting the initial weights for each expert, they can be initialized to equal weights, that is, all experts have equal weights. It can be understood that each expert contributes equally to the output. This method is suitable for situations where there is no historical model to refer to, or the historical model performs poorly. Although this method may have lower prediction accuracy in the initial stage, it can ensure that each expert has equal opportunities in the initial stage, thereby preventing the model from being too dependent on a specific expert. As the model is fine-tuned, these weights will be dynamically adjusted according to the actual data samples, so that the prediction accuracy of the model will gradually improve.

[0064] It should be noted that when the expert gating module generates output for the first time, the weight of the expert gating module can be calculated using any of the above two methods. Starting from the second time the output is generated, the weight of each expert model will be calculated based on the actual data through the gating model of the expert gating module to improve the reliability and accuracy of the results.

[0065] After obtaining the marketing prediction parameters, marketing delivery can be carried out based on the marketing prediction parameters to obtain samples of expected business behavior.

[0066] Step S150: fine-tune the model based on the expected business behavior sample to obtain a target marketing model.

[0067] As an implementation method, the weights of the gating model can be adjusted by fine-tuning the model based on the expected business behavior samples, and the model obtained after the weight adjustment can be used as the target marketing model. By continuously optimizing the weights, the gating model will be able to better adapt to the attributes and characteristics of the new marketing channels and produce more accurate output results.

[0068] In a specific application scenario, see Figure 4 , shows an example flow chart of the marketing model building method provided by the embodiment of the present application. Figure 4 As shown, in the user attribute module, the numerical feature can be input into the fully connected layer (not shown in the figure), and then the numerical feature vector output by the fully connected layer is spliced ​​to obtain the first feature vector; the category feature is input into the embedding layer, and then the second feature vector is output; the sequence feature is input into the embedding layer, the fully connected layer (not shown in the figure) and the attention layer in turn, and then the third feature vector output by the attention layer is obtained. Furthermore, the first feature vector, the second feature vector and the third feature vector can be spliced, and the vector obtained after splicing is input into the user attribute model of at least one mature channel model previously developed, so as to obtain the business attribute features output by the user attribute model.

[0069] The implementation process and principle of the channel attribute module are generally similar to those of the user attribute module. The difference is that after concatenating the first, second and third feature vectors of the channel attribute module, the concatenated vectors can be input into the channel attribute model of mature channel models in other business fields. If there are no mature channel models in other business fields, the concatenated vectors will be input into a new model modeled with channel activity as the Y-label, and then the features output by the channel attribute model of mature channel models in other business fields will be used as business participation features, or the features output by the new model will be used as business participation features.

[0070] Furthermore, the business attribute features and business participation features can be spliced ​​and input into the expert gating module. First, each expert model (i.e., expert 1, expert 2...expert n in the figure) will output its own prediction results, and the gating model will also initialize the weights in combination with the weights of the historical model, or directly initialize the weights according to equal weights. Then, the weighted values ​​of the prediction results of each expert are calculated according to the initialized weights to obtain the final output (this final output represents the final output of the large model composed of the user attribute module, the channel attribute module and the expert gating module, which can also be understood as the model result). In this way, marketing can be launched based on the model results to obtain expected business behavior samples. Optionally, the model can be fine-tuned based on the expected business behavior samples to continuously optimize the model weights and improve the accuracy of model predictions.

[0071] The marketing model construction method provided in this embodiment obtains the business attribute characteristics of users in historical marketing channels; obtains the business participation characteristics of users in the historical marketing channels; splices the business attribute characteristics with the business participation characteristics to obtain marketing ranking characteristics; obtains marketing prediction parameters based on the marketing ranking characteristics, and conducts marketing delivery based on the marketing prediction parameters to obtain expected business behavior samples; and fine-tunes the model based on the expected business behavior samples to obtain a target marketing model. Thus, the above method realizes the acquisition of marketing prediction parameters in new channels by utilizing the business attribute characteristics and business participation characteristics of users in existing channels, and then constructs a marketing model, which can effectively avoid the problem that new marketing channels are difficult to model due to lack of modeling samples.

[0072] At the same time, through modular structural design, user attribute modules and channel attribute modules can be flexibly spliced ​​and combined to maximize the reusability of the model while reducing repetitive workload, thereby accelerating the development cycle of new channel marketing models.

[0073] See also Figure 5 , is a structural block diagram of a marketing model building device provided in an embodiment of the present application. This embodiment provides a marketing model building device 200, which can be run on an electronic device. The device 200 includes a first acquisition module 210, a second acquisition module 220, a splicing module 230, a parameter acquisition module 240 and a model building module 250:

[0074] The first acquisition module 210 is used to acquire business attribute characteristics of users in historical marketing channels.

[0075] As an implementation mode, the first acquisition module 210 can be used to acquire user attribute data in historical marketing channels, wherein the user attribute data includes first numerical feature data, first category feature data, and first sequence feature data; the first numerical feature data, the first category feature data, and the first sequence feature data are concatenated and input into a user attribute model of an existing channel, wherein the user attribute model is used to evaluate the business attributes of the user; and the business attribute features output by the user attribute model are acquired.

[0076] Among them, before splicing the first numerical feature data, the first category feature data and the first sequence feature data and inputting them into the user attribute model of the existing channel, it can also include: inputting the first numerical feature data into the fully connected layer, and obtaining the first feature vector output by the fully connected layer; inputting the first category feature data into the embedding layer, and obtaining the second feature vector output by the embedding layer; passing the first sequence feature data through the embedding layer and the attention layer respectively, and obtaining the third feature vector output by the attention layer; in this way, splicing the first numerical feature data, the first category feature data and the first sequence feature data and inputting them into the user attribute model of the existing channel can include: splicing the first feature vector, the second feature vector and the third feature vector into the user attribute model of the existing channel.

[0077] As an implementation mode, the concatenating the first feature vector, the second feature vector and the third feature vector and inputting them into the user attribute model of an existing channel may include: concatenating the first feature vector, the second feature vector and the third feature vector and inputting them into at least one user attribute model of an existing channel; if the existing channel includes multiple user attribute models, the obtaining of business attribute features output by the user attribute model may include: obtaining user attribute vectors output by each of the multiple user attribute models of the existing channel to obtain multiple user attribute vectors; inputting the multiple user attribute vectors into an attention layer to obtain business attribute features output by the attention layer and corresponding to the multiple user attribute vectors.

[0078] The second acquisition module 220 is used to acquire the business participation characteristics of users in the historical marketing channel.

[0079] As an implementation mode, the second acquisition module 220 can be used to acquire channel attribute data in historical marketing channels, wherein the channel attribute data includes second numerical feature data, second category feature data, and second sequence feature data; input the second numerical feature data, second category feature data, and second sequence feature data into a channel attribute model of a mature channel, wherein the channel attribute model is used to evaluate the user's behavioral intention in the current channel; and acquire the business participation characteristics output by the channel attribute model.

[0080] As an implementation manner, the inputting of the second numerical feature data, the second category feature data and the second sequence feature data into a channel attribute model of a mature channel may include: if a channel attribute model of a mature channel does not exist, the second numerical feature data, the second category feature data and the second sequence feature data are input into a newly created model, wherein the newly created model is modeled using channel activity as a Y label, and the channel activity represents a customized user behavior data selection rule.

[0081] The splicing module 230 is used to splice the business attribute characteristics with the business participation characteristics to obtain marketing ranking characteristics.

[0082] The parameter acquisition module 240 is used to acquire marketing prediction parameters based on the marketing ranking characteristics, and to perform marketing delivery based on the marketing prediction parameters to obtain expected business behavior samples.

[0083] As an implementation method, the parameter acquisition module 240 can be used to input the marketing ranking feature into an expert gating model, wherein the expert gating model includes multiple expert models and a gating model; obtain the prediction parameters output by each of the multiple expert models; obtain the weight prediction parameters for each expert model output by the gating model; combine the prediction parameters with the weight prediction parameters to obtain the marketing prediction parameters.

[0084] The model building module 250 is used to fine-tune the model based on the expected business behavior sample to obtain a target marketing model.

[0085] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and modules can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here.

[0086] In several embodiments provided in the present application, the coupling between modules may be electrical, mechanical or other forms of coupling.

[0087] In addition, each functional module in each embodiment of the present application can be integrated into a processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or software functional modules.

[0088] See also Figure 6 Based on the above-mentioned marketing model construction method and device, the embodiment of the present application also provides an electronic device 100 that can execute the above-mentioned marketing model construction method. The electronic device 100 includes a memory 102 and one or more (only one is shown in the figure) processors 104 coupled to each other, and the memory 102 and the processor 104 are connected by a communication line. The memory 102 stores a program that can execute the content of the above-mentioned embodiment, and the processor 104 can execute the program stored in the memory 102.

[0089] Among them, the processor 104 may include one or more processing cores. The processor 104 uses various interfaces and lines to connect various parts of the entire electronic device 100, and executes various functions and processes data of the electronic device 100 by running or executing instructions, programs, code sets or instruction sets stored in the memory 102, and calling data stored in the memory 102. Optionally, the processor 104 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor 104 can integrate one or a combination of a central processing unit (Central Processing Unit, CPU), a graphics processing unit (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing display content; and the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 104, but may be implemented separately through a communication chip.

[0090] The memory 102 may include a random access memory (RAM) or a read-only memory (ROM). The memory 102 may be used to store instructions, programs, codes, code sets or instruction sets. The memory 102 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the aforementioned embodiments, etc. The data storage area may also store data (such as a phone book, audio and video data, chat record data) created by the electronic device 100 during use.

[0091] Please refer to Figure 7 , which shows a structural block diagram of a computer-readable storage medium provided in an embodiment of the present application. The computer-readable medium 300 stores program codes, which can be called by a processor to execute the method described in the above method embodiment.

[0092] The computer readable storage medium 300 may be an electronic memory such as a flash memory, an EEPROM (electrically erasable programmable read-only memory), an EPROM, a hard disk, or a ROM. Optionally, the computer readable storage medium 300 includes a non-transitory computer-readable storage medium. The computer readable storage medium 300 has storage space for program code 310 that performs any method steps in the above method. These program codes can be read from or written to one or more computer program products. The program code 310 can be compressed, for example, in an appropriate form.

[0093] In summary, the embodiments of the present application provide a method, device, electronic device and storage medium for constructing a marketing model. The method can be applied to electronic devices. The method obtains the business attribute characteristics of users in historical marketing channels; obtains the business participation characteristics of users in the historical marketing channels; splices the business attribute characteristics with the business participation characteristics to obtain marketing ranking characteristics; obtains marketing prediction parameters based on the marketing ranking characteristics, and conducts marketing delivery based on the marketing prediction parameters to obtain expected business behavior samples; and fine-tunes the model based on the expected business behavior samples to obtain a target marketing model. Thus, the above-mentioned method can achieve the construction of a marketing model by utilizing the business attribute characteristics and business participation characteristics of users in existing channels to obtain marketing prediction parameters in new channels, which can effectively avoid the problem that new marketing channels are difficult to model due to lack of modeling samples.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A marketing model construction method, characterized in that: The method comprises: Obtain business attribute characteristics of users in historical marketing channels; Obtaining business participation characteristics of users in the historical marketing channel; The business attribute feature is combined with the business participation feature to obtain a marketing ranking feature; Acquiring marketing prediction parameters based on the marketing ranking features, and conducting marketing delivery based on the marketing prediction parameters to obtain expected business behavior samples; The model is fine-tuned based on the expected business behavior samples to obtain a target marketing model.

2. The method according to claim 1, characterized in that The obtaining of business attribute characteristics of users in historical marketing channels includes: Acquire user attribute data in historical marketing channels, wherein the user attribute data includes first numerical feature data, first category feature data, and first sequence feature data; splicing the first numerical feature data, the first category feature data, and the first sequence feature data and inputting them into a user attribute model of an existing channel, wherein the user attribute model is used to evaluate the business attributes of the user; Obtain the business attribute characteristics output by the user attribute model.

3. The method according to claim 2, characterized in that Before the first numerical feature data, the first category feature data and the first sequence feature data are spliced ​​and input into the user attribute model of the existing channel, the method further includes: Inputting the first numerical feature data into a fully connected layer, and obtaining a first feature vector output by the fully connected layer; Inputting the first category feature data into an embedding layer, and obtaining a second feature vector output by the embedding layer; Pass the first sequence feature data through an embedding layer and an attention layer respectively, and obtain a third feature vector output by the attention layer; The step of splicing the first numerical feature data, the first category feature data, and the first sequence feature data and inputting the spliced ​​data into the user attribute model of the existing channel includes: The first feature vector, the second feature vector and the third feature vector are concatenated and input into the user attribute model of the existing channel.

4. The method according to claim 3, characterized in that The step of splicing the first feature vector, the second feature vector, and the third feature vector and inputting the spliced ​​features into the user attribute model of the existing channel includes: splicing the first feature vector, the second feature vector, and the third feature vector and inputting the concatenated features into a user attribute model of at least one existing channel; If the user attribute models of the existing channels include multiple ones, the step of obtaining the business attribute features output by the user attribute models includes: Obtaining user attribute vectors output by respective user attribute models of existing channels to obtain multiple user attribute vectors; The multiple user attribute vectors are input into the attention layer, and business attribute features corresponding to the multiple user attribute vectors output by the attention layer are obtained.

5. The method according to claim 2, characterized in that: The obtaining of the business participation characteristics of the user in the historical marketing channel includes: Acquire channel attribute data in historical marketing channels, wherein the channel attribute data includes second numerical feature data, second category feature data, and second sequence feature data; inputting the second numerical feature data, the second category feature data and the second sequence feature data into a channel attribute model of a mature channel, wherein the channel attribute model is used to evaluate the user's behavioral intention in the current channel; Obtain business participation characteristics output by the channel attribute model.

6. The method according to claim 5, characterized in that The step of inputting the second numerical characteristic data, the second category characteristic data and the second sequence characteristic data into the channel attribute model of the mature channel includes: If there is no channel attribute model for a mature channel, the second numerical feature data, the second category feature data and the second sequence feature data are input into a newly created model. The newly created model is modeled using channel activity as the Y label. The channel activity represents a customized user behavior data selection rule.

7. The method according to any one of claims 1 to 6, characterized in that: The acquiring of marketing prediction parameters based on the marketing ranking features includes: Inputting the marketing ranking features into an expert gating model, wherein the expert gating model includes a plurality of expert models and a gating model; Obtaining prediction parameters output by each of the multiple expert models; Obtaining weight prediction parameters for each expert model output by the gating model; The prediction parameter is combined with the weight prediction parameter to obtain a marketing prediction parameter.

8. A marketing model building device, characterized in that: The device comprises: The first acquisition module is used to acquire the business attribute characteristics of users in historical marketing channels; A second acquisition module is used to acquire business participation characteristics of users in the historical marketing channel; A splicing module, used to splice the business attribute feature with the business participation feature to obtain a marketing ranking feature; A parameter acquisition module, used to acquire marketing prediction parameters based on the marketing ranking features, and to perform marketing delivery based on the marketing prediction parameters to acquire expected business behavior samples; The model building module is used to fine-tune the model based on the expected business behavior sample to obtain a target marketing model.

9. An electronic device, characterized in that: including one or more processors and memory; One or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program code, wherein when the program code is executed by a processor, the method according to any one of claims 1 to 7 is executed.

Images