CN116843371B

CN116843371B - Marketing promotion method, marketing promotion device, marketing promotion equipment and computer-readable storage medium

Info

Publication number: CN116843371B
Application number: CN202311123293.8A
Authority: CN
Inventors: 杨良志; 白琳; 汪志新; 方跃涵; 周光辉; 杜炜铃
Original assignee: Guangzhou Caixun Digital Technology Co ltd; Richinfo Technology Co ltd
Current assignee: Guangzhou Caixun Digital Technology Co ltd; Richinfo Technology Co ltd
Priority date: 2023-09-01
Filing date: 2023-09-01
Publication date: 2023-12-05
Anticipated expiration: 2043-09-01
Also published as: CN116843371A

Abstract

The embodiment of the application provides a marketing promotion method, a marketing promotion device, marketing promotion equipment and a computer readable storage medium. The method comprises the steps of obtaining original characteristic data of a user; performing feature derivation on the original feature data to obtain first feature data based on data distribution and second feature data with self-supervision learning characteristics; summarizing the first characteristic data and the second characteristic data to obtain final characteristic data; processing the final characteristic data through a trained prediction model and preset rewarding parameters to obtain a value prediction value and a strategy probability prediction value; and determining a final marketing popularization mode based on the value predicted value and the strategy probability predicted value. In this way, the effectiveness and efficiency of marketing is improved.

Description

Marketing promotion method, marketing promotion device, marketing promotion equipment and computer-readable storage medium

Technical Field

Embodiments of the present application relate to the field of marketing, and more particularly, to marketing promotion methods, apparatuses, devices, and computer-readable storage media.

Background

The conventional marketing popularization algorithm comprises a collaborative filtering algorithm, a lingering semantic model, a decision tree-based classification algorithm, a deep learning algorithm and the like, and the algorithms can be applied to a feature-derived marketing popularization method, so that the effect and efficiency of personalized recommendation and marketing popularization are improved.

However, when the algorithm is adopted for marketing promotion, the following disadvantages are caused:

first, feature representation capability is limited. There may be certain limitations in feature representation, not reflecting the interests and preferences of the user well, and not being robust enough to be susceptible to noise and interference.

Second, the prediction and planning capabilities are inadequate. Certain shortages exist in the aspects of prediction and planning, future marketing strategies cannot be well predicted, and different marketing environments and requirements cannot be well adapted.

Third, the model interpretation ability is weak. The problem that a certain model has weaker interpretation capability is solved, recommendation results generated by the model are difficult to interpret, and certain opacity and untrustiness are brought to users.

Disclosure of Invention

According to the embodiment of the application, a marketing popularization scheme is provided, the independent component analysis and the self-supervision learning algorithm are adopted for feature derivation, the original features can be subjected to feature extraction and dimension reduction, and the potential value of the data is better mined, so that the more independent and robust feature representation is obtained, and the accuracy and effect of the model are improved. Meanwhile, the model algorithm disclosed by the application has strong prediction and planning capabilities on future behaviors, can simulate and predict future marketing strategies, can better adapt to the change and uncertainty of user behaviors, and is suitable for different marketing environments and demands.

In a first aspect of the present application, a marketing promotion method is provided. The method comprises the following steps:

acquiring original characteristic data of a user;

performing feature derivation on the original feature data to obtain first feature data based on data distribution and second feature data with self-supervision learning characteristics;

summarizing the first characteristic data and the second characteristic data to obtain final characteristic data;

processing the final characteristic data through a trained prediction model and preset rewarding parameters to obtain a value prediction value and a strategy probability prediction value;

and determining a final marketing popularization mode based on the value predicted value and the strategy probability predicted value.

Further, the performing feature derivation on the original feature data to obtain first feature data based on data distribution includes:

performing feature derivation on the original feature data through the following algorithm to obtain first feature data based on data distribution:

；

wherein A is a mixing matrix;

is an observation variable;

is an independent component.

Further, the performing feature derivation on the original feature data to obtain second feature data with self-supervision learning characteristics includes:

inputting the original characteristic data into a trained characteristic model to obtain second characteristic data with self-supervision learning characteristics;

wherein the feature model comprises a primary network and a target network; the main network and the target network have the same structure.

Further, the inputting the original feature data into the trained feature model to obtain second feature data with self-supervision learning characteristics includes:

performing data enhancement on the original characteristic data to generate first enhancement data and second enhancement data;

processing the first enhancement data and the second enhancement data through an encoder in the feature model to obtain a first feature vector and a second feature vector;

processing the first feature vector and the second feature vector through a decoder in the feature model to obtain a first reconstruction output and a second reconstruction output;

forward propagation is carried out on the first reconstruction output and the second reconstruction output through a main network, so that a first characteristic representation is obtained;

forward propagation is carried out on the first reconstruction output and the second reconstruction output through a target network, so that second characteristic representation is obtained;

calculating a loss function of the feature model based on the first feature representation and the second feature representation;

and obtaining second characteristic data with self-supervision learning characteristics based on the loss function.

Further, the calculating a loss function of the feature model based on the first feature representation and the second feature representation comprises:

based on the first and second feature representations, a loss function of the feature model is calculated by the following formula:

；

wherein N is the batch size;

and->The first characteristic vector and the second characteristic vector are respectively output by the encoder;

and->The first reconstruction output and the second reconstruction output are respectively.

Further, the method further comprises the following steps:

based on the loss function, optimizing strategy gradients and network parameters in the prediction model through a back propagation algorithm:

wherein optimizing the policy gradient in the predictive model includes:

optimizing a strategy gradient in the prediction model based on a strategy value target;

the policy value target is calculated by the following method:

；

wherein,a weighted sum of future rewards starting from time step t;

is a discount factor;

a value predicted value for time step t+1;

as a dominance function;

c1 and c2 are weight parameters;

the strategy probability is output for the neural network;

l2 regularization term that is a parameter of the neural network.

Further, optimizing the policy gradient in the predictive model based on the policy value objective includes:

based on the policy value target, optimizing the policy gradient in the prediction model by the following algorithm:

；

wherein,gradient of the neural network parameters for the loss function;

gradient of the dominance function to the neural network parameters;

the gradient of the strategy probability to the neural network parameters;

weight parameters regularized for L2;

is a parameter of the neural network.

In a second aspect of the present application, a marketing promotion device is provided. The device comprises:

the acquisition module is used for acquiring the original characteristic data of the user;

the deriving module is used for respectively carrying out feature derivation on the original feature data to obtain first feature data based on data distribution and second feature data with self-supervision learning characteristics;

the summarizing module is used for summarizing the first characteristic data and the second characteristic data to obtain final characteristic data; the final feature data includes status data and action data;

the prediction module is used for processing the state data and the action data through a trained prediction model and preset rewarding parameters to obtain a value predicted value and a strategy probability predicted value of the current state;

and the promotion module is used for determining a final marketing promotion mode based on the value predicted value and the strategy probability predicted value.

In a third aspect of the application, an electronic device is provided. The electronic device includes: a memory and a processor, the memory having stored thereon a computer program, the processor implementing the method as described above when executing the program.

In a fourth aspect of the application, there is provided a computer readable storage medium having stored thereon a computer program which when executed by a processor implements a method as according to the first aspect of the application.

According to the marketing popularization method provided by the embodiment of the application, the independent component analysis and the self-supervision learning algorithm are adopted to conduct feature derivation, the original features can be subjected to feature extraction and dimension reduction, and the potential value of the data is better mined, so that more independent and robust feature representation is obtained, and the accuracy and effect of the model are improved. Meanwhile, the model algorithm disclosed by the application has strong prediction and planning capabilities on future behaviors, can simulate and predict future marketing strategies, can better adapt to the change and uncertainty of user behaviors, and is suitable for different marketing environments and demands.

It should be understood that the description in this summary is not intended to limit the critical or essential features of the embodiments of the application, nor is it intended to limit the scope of the application. Other features of the present application will become apparent from the description that follows.

Drawings

The above and other features, advantages and aspects of embodiments of the present application will become more apparent by reference to the following detailed description when taken in conjunction with the accompanying drawings. In the drawings, wherein like or similar reference numerals denote like or similar elements, in which:

FIG. 1 is a flow chart of a marketing promotion method according to an embodiment of the present application;

FIG. 2 is a block diagram of a marketing promotion device according to an embodiment of the present application;

fig. 3 is a schematic diagram of a structure of a terminal device or a server suitable for implementing an embodiment of the present application.

Detailed Description

For the purposes of making the objects, technical solutions and advantages of the embodiments of the present disclosure more apparent, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure, and it is apparent that the described embodiments are some embodiments of the present disclosure, but not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments in this disclosure without inventive faculty, are intended to be within the scope of this disclosure.

In addition, the term "and/or" herein is merely an association relationship describing an association object, and means that three relationships may exist, for example, a and/or B may mean: a exists alone, A and B exist together, and B exists alone. In addition, the character "/" herein generally indicates that the front and rear associated objects are an "or" relationship.

Fig. 1 shows a flowchart of a marketing promotion method according to an embodiment of the present disclosure.

S110, acquiring original characteristic data of a user.

In this embodiment, the execution body for the marketing promotion method may acquire the original feature data in a wired manner or a wireless connection manner.

Further, the execution body may acquire the original feature data transmitted from the electronic device connected to the execution body in communication, or may be the original feature data stored locally in advance.

In some embodiments, the raw feature data comprises: user ID, age, gender, marital status, hobbies and interests, purchasing power, purchasing history of the customer, browsing history, search history, and/or user t-time actions, etc.

And S120, carrying out feature derivation on the original feature data to obtain first feature data based on data distribution and second feature data with self-supervision learning characteristics.

In some embodiments, to ensure the quality and availability of the data, the raw feature data acquired in step S110 may be optimized. The raw feature data is optimized, for example, by means of data cleaning, normalization, numerical coding, and/or dummy coding.

In some embodiments, the raw feature data (optimized) may be feature derived by the following algorithm to obtain first feature data based on data distribution:

；

wherein A is a mixing matrix;

for observing the variables, n independent components +.>Is obtained by linear combination of (a);

is an independent component.

I.e. each individual componentAs a new feature, these new features are then used instead of the original observed variable x, resulting in a new set of features (first feature).

In some embodiments, inputting the original feature data into a trained feature model to obtain second feature data with self-supervised learning characteristics; the feature model comprises an encoder, a decoder, a main network and a target network, wherein the main network and the target network have the same structure;

definition of raw dataThe feature vector of the encoder output is expressed as +.>The reconstructed output of the decoder is expressed as. For each input sample data +.>Randomly performing data enhancement, e.g. random scaling, sampling, noise addition, random insertion, random deletion, etc., generating two different enhancement data, i.e. first enhancement data +.>And second enhancement dataWill->And->Respectively through encodersObtain the corresponding vector->And->. Will beAnd->As input, the first reconstruction output +.>And a second reconstruction output。

Further, input samples [ ]、/>) Forward propagation is carried out through a main network, and a first characteristic representation is obtained;

and inputting the first characteristic representation to a target network for forward propagation to obtain a second characteristic representation.

The similarity of the negative cosine of the two characteristic representations is calculated as a loss function L:

；

wherein N is the batch size;

the parameters of the main network are updated by a back propagation algorithm so that the cosine similarity is more and more close to 1.

Solving according to the loss function to obtain the final productAnd->Taking the average value to obtain the final derivative variable +.>。

And respectively processing the original characteristic data through the derivative variables to obtain a plurality of derivative characteristics (second characteristic data).

And S130, summarizing the first characteristic data and the second characteristic data to obtain final characteristic data.

In some embodiments, the first feature data and the second feature data are added to obtain final feature data. The final feature data includes a status feature and an action feature;

wherein the status features are characteristic of the current market environment and the customer. For example, the market environment includes historic behavior of a customer's purchase history, browsing history, search history, etc., and the characteristics of the customer include age, gender, etc. The state is expressed asThe state at t is shown.

An action feature is a marketing strategy that can be taken. For example, we can send an email, a short message, a push notification, etc. to the customer. The actions may be expressed asRepresenting the action selected at time t.

And S140, processing the final characteristic data through the trained prediction model and preset rewarding parameters to obtain a value prediction value and a strategy probability prediction value.

The reward parameter is used to represent the effect of the marketing strategy, and may be preset according to the actual application scenario. If the customer accepts the marketing strategy and purchases the product, a reward is awarded. The rewards may be expressed asIndicating a prize at time tAnd (5) excitation. The rewards are given by the operator, defining the next action by the customer as positive rewards, defaulting to 1, and conversely negative rewards-1.

The predictive models include an action model and a value model. The action model is used to predict the next action and the value model is used to predict the value of the current state.

In some embodiments, the following algorithm is employed to simulate the behavior of the user and update the value estimates for the various states. A policy network is used to select actions at each state and a dynamic model is used to update the state. A search tree is generated in which each node of the search tree species represents a state and each edge represents an action to be taken from one state to another. At leaf nodes of the search tree, a value network is employed to evaluate the value of the current state.

For each time step t, an action is selectedObserving user state->And rewarding->：

；

Wherein,rewards simulated for the ith time of searching tree species;

n is the number of simulations;

is a discount factor;

status for simulation using dynamic model +.>Is of value (1);

weight parameters for control exploration and development;

a policy probability value output for the neural network;

the number of accesses to take action b in state s;

the number of accesses to take action a in state s;

an action selected for time t;

selecting a value predicted value of the action a for the current t time and the state s output by the neural network;

is the confidence upper bound (exploration term) for action a, which is used to explore unknown actions.

In some embodiments, the neural network model f receives the current stateAnd historical action->As input, value predictive value +.>Policy probability predictor +.>. The value is the long-term value of the current state, and the policy probability represents the probability of selecting each action:

；

wherein,a value predicted value for the current state;

is the policy probability predictor for the current state.

In some embodiments, for each time step t, the policy value target is calculated using the data generated by the tree search. The policy value objective is a weighted average where the weights are derived by training a neural network to balance the relative importance of policy and value. The calculation formula of the policy value target is as follows:

；

wherein,a weighted sum of future rewards starting from time step t;

gamma is a discount factor;

is the value predicted value of time step t+1;

the method is a dominance function and is used for measuring the quality of the current state;

c1 and c2 are weight parameters;

the strategy probability is output for the neural network;

l2 regularization term that is a parameter of the neural network.

In some embodiments, policy gradients and network parameter updates are performed as follows:

for each time step t, the policy function of the neural network is updated as follows. The policy gradient method uses the gradient of the policy value target to update the parameters of the policy function (maximizing the policy value), and the calculation formula of the policy gradient is as follows:

；

wherein,gradient of the neural network parameters for the loss function;

for neural networks as dominant functionsGradient of the parameter;

the gradient of the strategy probability to the neural network parameters;

weight parameters regularized for L2;

is a parameter of the neural network.

Further, the network parameters are updated by:

and for each time step t, updating parameters of a dynamic model and a cost function of the neural network by adopting data generated by tree search in an experience recovery mode so as to improve accuracy. The prediction error of the dynamic model and the cost function is minimized by adopting a gradient descent mode.

Specifically, the update formula of the dynamic model and the cost function is as follows:

；

wherein,gradients of parameters of the dynamic model or the cost function for the loss function;

z is a weighted sum of future rewards starting from time t;

the value predicted value of the time step t;

gradients of parameters of the dynamic model or the cost function for the policy probability;

c2 is the L2 regularized weight parameter;

w is a parameter of the dynamic model or the cost function.

Further, according to the simulation result, the strategy is updated to continuously optimize the effects of accurate marketing and advertisement delivery. The probability of each action is evaluated using the policy network and the action with the highest probability is selected. Finally, the effect of the strategy is evaluated using a reward function and used to optimize the predictive model.

And S150, determining a final marketing popularization mode based on the value predicted value and the strategy probability predicted value.

In some embodiments, a final marketing promotion is determined based on the value predictors and the policy probability predictors. That is, according to the characteristics and the demands of the users, the advertisement and the recommended products are purposefully put in, so as to improve the conversion rate and the satisfaction of the users

A specific embodiment based on the present disclosure is given below:

obtaining original feature data, wherein the original feature data comprises 12 original features, and the features of each user comprise: user ID, age, gender, marital status, interests and purchasing power, purchasing history of the customer, browsing history, search history, and user t-time actions, refer to table 1.

TABLE 1

ID

Age of

Sex (sex)

Marital status

Hobbies and interests

Purchasing power

Purchase history

Browsing history

Search history

Period t of user Action

1

32

Man's body

Wedding

Film, sound Happy, travelling, Food for delicacies

High height

Mobile phone and tablet Computer, intelligent Watch with a watch body

Mobile phone accessory Piece and flat plate Computer, intelligence Wearable and wearable

Mobile phone and tablet A board computer, Intelligent watch

Registration

2

45

Female

Wedding

Stock, base Gold, reading, Travel and swimming

Very high

Television and ice Box and washing machine

Household electric appliance Device and kitchen Electric appliance, personal Protecting health, Mother and infant articles

Television, sky Regulating refrigerator, Washing machine

Clicking

3

25

Man's body

Not married

Game, electricity Shadow and music

Medium and medium

Snack and drink Material and wine

Fresh fruit Vegetable, grain and oil Seasoning and rest Idle food, Household daily use

Beer and beverage Material and snack

Login

...

99999

50

Man's body

Wedding

Cooking, home Administrative, travel, and, Music

Low and low

Cosmetic and cosmetic product Skin care and personal care Personal care

For mother and infant Household and article Electrical appliance and kitchen Room electric appliance, Fresh fruits and vegetables

Cosmetic product, Beauty care Skin and person Nursing device

Purchasing

100000

27

Female

Not married

Tourist and beauty Food and film, Fashion style

Medium and medium

Toy, educational Toy, children Book with book cover

Infant and child use Article and children's garment Shoes for children and parents Sub-dress, pregnancy Women's dress

Toy, benefit Intelligent toy, Book for children

Registration

Preprocessing data such as historical behaviors and attributes of a user, including steps of data cleaning, data normalization, numerical coding, dummy coding and the like, and finally processing the data into 30-column characteristics so as to ensure the quality and usability of the data.

And carrying out feature derivation on the preprocessed data to obtain independent features based on data distribution and features with self-supervision learning characteristics. Taking 10000 observation variables as an example,the variable consists of 10 independent componentsIs obtained by linear combination of:

；

by the above wayRecovering each individual componentThe 10 independent components are taken as new characteristics after derivative variables.

For each input user data, randomly enhancing the data to obtain two enhanced dataAnd->. Will->Andthe corresponding feature vectors are obtained by the encoder respectively>And->. Will->And->As input, the reconstructed outputs +.>And->。

The minimum solution is carried out through the loss function, and the obtained result isAnd->Averaging to obtain the final derivative variable +.>. The 30 original features are processed separately (one-to-one processing) to obtain 30 derivative features.

Further, summarizing derived features obtained by the two algorithms to obtain 70 features; the 70 features include a status feature and an action feature.

And establishing a prediction model comprising an action model and a value model, wherein the action model is used for predicting the next action, and the value model is used for predicting the value of the current state. Current state ofAnd historical action->As input, predicting by using neural network to obtain value predicted value of current state +.>And policy probability predictor ++>：

；

Further, an action is selected using the policy probability predictors to obtain an immediate rewardAnd observe subsequent rewards. In the model training process, a deep reinforcement learning method is adopted, and based on a loss function, the strategy gradient and network parameters of the model are optimized through a back propagation algorithm, so that a better strategy probability predictive value of a user is gradually learned>Sum value prediction value->Providing reference for the next decision of delivery channel.

Policy probability predictors: all possible behavioral accesses of the user are predicted and a probability is assigned to each mode. For example, the probability of predicting that a certain user views a product in a self-media mode, a short message mode, an outbound mode and the like is 0.2, 0.3 and 0.5 respectively;

value predictive value: and predicting the final life cycle value LTV of the user according to the current behavior. For example, a predicted user lifecycle value of 300 yuan represents the general life value of the user.

Through the predicted value, the method can help companies to more accurately determine the most effective marketing channels in the time period and put emphasis on the channels, so that the conversion rate of users and the life cycle value are improved. For example, users with high LTV are preferentially extracted, corresponding channels with high probability are selected for delivery, and clicking and investment of the users are improved.

Further, the delivery monitoring and optimization is performed by:

generating actions according to the strategy network, determining a channel for putting advertisements, executing the actions and observing environmental feedback, including the actions of real clicking, purchasing and the like of a user at the time t+1, and then, as new user actions and states, re-predicting a strategy model to obtain a new strategy probability predicted valueSum value prediction value->And carrying out a new round of throwing. The accuracy and the efficiency of the decision strategy are gradually improved, and the optimization and the growth of marketing popularization are realized. To improve advertising effectiveness and ROI to improve decision accuracyAnd efficiency.

The method of the disclosure realizes optimization of decision strategy through self-training and searching, and the click rate (CTR) and conversion rate of each channel of the user are improved by 60%. The Return On Investment (ROI) is 1.5 times of the original ROI, has wide application prospect, and can be widely applied to the complex decision problems in the field of advertisement delivery and the like.

According to the embodiment of the disclosure, the following technical effects are achieved:

the potential value of the data can be better mined through the feature derivation of the original data, and the accuracy and effect of the model are improved;

through the reinforcement learning algorithm adopted by the method, future marketing strategies can be predicted and planned better, the change and uncertainty of user behaviors can be adapted better, and the effect and efficiency of personalized recommendation and marketing promotion are improved;

compared with the existing marketing algorithm, the method disclosed by the application emphasizes the importance of feature derivation, model prediction and model interpretation capability, can better adapt to different marketing requirements and scenes, and improves the marketing effect and efficiency.

That is, by the algorithm of the present disclosure, a richer feature extraction and dimension reduction capability can be provided. Through linear change, each component of the transformed data is mutually independent, independent features in the data can be extracted, redundant information and noise are removed, so that efficient feature extraction and dimension reduction are realized, original data are mapped into a low-dimension representation space through an encoder, a low-dimension vector is output, and important feature information of the original data is contained in the vector, so that feature dimension reduction and feature extraction are realized;

the independence and the robustness of the characteristics are improved. The raw data is decomposed into separate components that are independent of each other and contain most of the information of the raw data. Compared with the traditional feature extraction method, the method has the advantages that the independence and the robustness of the features are improved through the algorithm adopted by the method, and therefore the accuracy and the generalization capability of the model are improved.

Future marketing strategies can be better predicted and planned. The algorithm disclosed by the application has stronger prediction and planning capabilities, and utilizes the historical data and the user behavior model to predict, so that the algorithm is better suitable for the change and uncertainty of the user behavior, and the behavior and preference of future users are predicted. The method can help enterprises to better know the demands and interests of users, better predict and plan future marketing strategies, improve the marketing effect and efficiency, and adapt to different marketing scenes and demands.

Better optimizing iterative ability and learning better strategies. The algorithm can learn a better marketing strategy and a better decision rule through continuous trial and error and feedback, helps enterprises to continuously optimize the marketing strategy, and improves the effect and efficiency of accurate marketing.

It should be noted that, for simplicity of description, the foregoing method embodiments are all described as a series of acts, but it should be understood by those skilled in the art that the present application is not limited by the order of acts described, as some steps may be performed in other orders or concurrently in accordance with the present application. Further, those skilled in the art will also appreciate that the embodiments described in the specification are alternative embodiments, and that the acts and modules referred to are not necessarily required for the present application.

The above description of the method embodiments further describes the solution of the present application by means of device embodiments.

Fig. 2 shows a marketing promotion device 200 according to an embodiment of the present application, including, as shown in fig. 2:

an obtaining module 210, configured to obtain original feature data of a user;

the deriving module 220 is configured to perform feature derivation on the original feature data, so as to obtain first feature data based on data distribution and second feature data with self-supervised learning characteristics;

a summarizing module 230, configured to summarize the first feature data and the second feature data to obtain final feature data; the final feature data includes status data and action data;

the prediction module 240 is configured to process the state data and the action data through a trained prediction model and a preset reward parameter, so as to obtain a value predicted value and a policy probability predicted value of the current state;

and the promotion module 250 is used for determining a final marketing promotion mode based on the value predicted value and the strategy probability predicted value.

It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the described modules may refer to corresponding procedures in the foregoing method embodiments, which are not described herein again.

Fig. 3 shows a schematic diagram of a structure of a terminal device or server suitable for implementing an embodiment of the application.

As shown in fig. 3, the terminal device or the server includes a Central Processing Unit (CPU) 301 that can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM) 302 or a program loaded from a storage section 308 into a Random Access Memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the terminal device or the server are also stored. The CPU 301, ROM302, and RAM 303 are connected to each other through a bus 304. An input/output (I/O) interface 305 is also connected to bus 304.

The following components are connected to the I/O interface 305: an input section 306 including a keyboard, a mouse, and the like; an output portion 307 including a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and the like, a speaker, and the like; a storage section 308 including a hard disk or the like; and a communication section 309 including a network interface card such as a LAN card, a modem, or the like. The communication section 309 performs communication processing via a network such as the internet. The drive 310 is also connected to the I/O interface 305 as needed. A removable medium 311 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is installed on the drive 310 as needed, so that a computer program read therefrom is installed into the storage section 308 as needed.

In particular, the above method flow steps may be implemented as a computer software program according to an embodiment of the application. For example, embodiments of the application include a computer program product comprising a computer program embodied on a machine-readable medium, the computer program comprising program code for performing the method shown in the flowcharts. In such an embodiment, the computer program may be downloaded and installed from a network via the communication portion 309, and/or installed from the removable medium 311. The above-described functions defined in the system of the present application are performed when the computer program is executed by a Central Processing Unit (CPU) 301.

The computer readable medium shown in the present application may be a computer readable signal medium or a computer readable storage medium, or any combination of the two. The computer readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples of the computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In the present application, however, the computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, with the computer-readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, fiber optic cable, RF, etc., or any suitable combination of the foregoing.

The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The units or modules involved in the embodiments of the present application may be implemented in software or in hardware. The described units or modules may also be provided in a processor. Wherein the names of the units or modules do not in some cases constitute a limitation of the units or modules themselves.

As another aspect, the present application also provides a computer-readable storage medium that may be contained in the electronic device described in the above embodiment; or may be present alone without being incorporated into the electronic device. The computer-readable storage medium stores one or more programs that when executed by one or more processors perform the methods described herein.

The above description is only illustrative of the preferred embodiments of the present application and of the principles of the technology employed. It will be appreciated by persons skilled in the art that the scope of the application is not limited to the specific combinations of the features described above, but also covers other embodiments which may be formed by any combination of the features described above or their equivalents without departing from the spirit of the application. Such as the above-mentioned features and the technical features having similar functions (but not limited to) applied for in the present application are replaced with each other.

Claims

1. A marketing method, comprising:

acquiring original characteristic data of a user;

determining a final marketing popularization mode based on the value predicted value and the strategy probability predicted value;

based on the loss function, optimizing the strategy gradient and the network parameters in the prediction model through a back propagation algorithm:

wherein optimizing the policy gradient in the predictive model includes:

the policy value target is calculated by the following method:

；

wherein,a weighted sum of future rewards starting from time step t;

is a discount factor;

a value predicted value for time step t+1;

as a dominance function;

c1 and c2 are weight parameters;

the strategy probability is output for the neural network;

l2 regularization term that is a parameter of the neural network.

2. The method of claim 1, wherein the deriving the raw feature data features first feature data based on a data distribution comprises:

；

wherein A is a mixing matrix;

is an observation variable;

is an independent component.

3. The method of claim 1, wherein the feature deriving the raw feature data to obtain second feature data having self-supervised learning features comprises:

4. A method according to claim 3, wherein said inputting said raw feature data into a trained feature model to obtain second feature data having self-supervised learning features comprises:

5. The method of claim 4, wherein the calculating a loss function of the feature model based on the first feature representation and the second feature representation comprises:

；

wherein N is the batch size;

6. The method of claim 1, wherein optimizing the policy gradient in the predictive model based on the policy value objective comprises:

；

wherein,gradient of the neural network parameters for the loss function;

gradient of the dominance function to the neural network parameters;

the gradient of the strategy probability to the neural network parameters;

weight parameters regularized for L2;

is a parameter of the neural network.

7. Marketing popularization device, characterized by including:

the promotion module is used for determining a final marketing promotion mode based on the value predicted value and the strategy probability predicted value;

wherein optimizing the policy gradient in the predictive model includes:

the policy value target is calculated by the following method:

；

wherein,a weighted sum of future rewards starting from time step t;

is a discount factor;

a value predicted value for time step t+1;

as a dominance function;

c1 and c2 are weight parameters;

the strategy probability is output for the neural network;

l2 regularization term that is a parameter of the neural network.

8. An electronic device comprising a memory and a processor, the memory having stored thereon a computer program, characterized in that the processor, when executing the computer program, implements the method according to any of claims 1-6.

9. A computer readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, implements the method according to any one of claims 1-6.