Marketing prediction method and device and computer equipment

Abstract

The invention discloses a marketing prediction method and device and computer equipment. The method comprises the steps that portrait features of a to-be-marketed account are acquired, and the portrait features at least comprise the service use condition of the to-be-marketed account; a preset model is adopted to analyze the portrait features of the to-be-marketed account, the probability that the to-be-marketed account handles the target business is determined, the preset model is obtained through training according to conditional noise and a training data set, the conditional noise is determined according to the training data set, and the training data set at least comprises historical marketing data, historical marketing data and historical marketing data; the conditional noise is used for generating false samples by a generator in the preset model; and outputting a prediction result of the to-be-marketed account. According to the invention, the technical problem of low prediction accuracy in related technologies is solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to a marketing forecasting method, apparatus, and computer equipment. Background Art

[0002] Time series prediction has a wide range of applications in many fields, such as traffic flow prediction, weather forecast, disease transmission, power equipment status monitoring, etc. In the field of transportation, accurate traffic flow prediction can effectively alleviate traffic congestion and optimize traffic management and planning; in the field of meteorology, accurate weather forecast can help people take preventive measures in advance and reduce the losses caused by natural disasters; in the field of public health, disease transmission prediction can provide early warning for health departments and formulate effective prevention and control measures; in the field of power, power equipment status monitoring can help to detect faults in advance and ensure the stable operation of the power system. In the related technology, time series prediction methods include but are not limited to ARIMA (Autoregressive Integrated Moving Average Model), machine learning methods, neural network-based methods and generative adversarial networks (Generative Adversarial Networks, GAN). The main goal of most GAN-based methods is to generate simulated data from noise and then predict marketing results. However, since the current noise is random noise, the correlation between the false samples generated by random noise and the real samples is very small, which leads to a very low prediction accuracy. Summary of the invention

[0003] The embodiments of the present application provide a marketing forecasting method, apparatus, and computer device to at least solve the technical problem of low forecasting accuracy in the related art.

[0004] According to one aspect of an embodiment of the present application, a marketing prediction method is provided, comprising: obtaining an account to be marketed; using a preset model to analyze the account to be marketed, and determining the probability that the account to be marketed handles a target business, wherein the preset model is trained based on conditional noise and a training data set, wherein the conditional noise is determined based on the training data set, and the training data set at least includes: historical marketing data, and the conditional noise is used for a generator in the preset model to generate false samples; and outputting a prediction result for the account to be marketed.

[0005] Optionally, the preset model is trained in the following manner, including: generating an initial noise embedding vector, wherein the initial noise embedding vector conforms to a Gaussian distribution; performing feature extraction on the historical marketing data to obtain a feature embedding vector, and inputting the feature embedding vector and the initial noise embedding vector into a conditional noise generator for pre-training to obtain the conditional noise, wherein the conditional noise generator includes a linear layer; and training the preset model based on the conditional noise and the training data set.

[0006] Optionally, the preset model is obtained based on the conditional noise and the historical marketing data training, including: obtaining a preset mask matrix, which is a lower triangular matrix; multiplying the training data set with the preset mask matrix to obtain a masked training data set; splicing the conditional noise after the masked training data set to obtain the input data of the preset model; and obtaining the generator in the preset model based on the input data of the preset model training.

[0007] Optionally, the generator in the preset model is obtained by training based on the input data of the preset model, including: obtaining the total number of iterations of the generator training and the number of iterations of the current training; determining the cosine function according to the total number of iterations of the generator training and the number of iterations of the current training; determining the dynamic weight according to the cosine function, and determining the training loss of the generator in the preset model based on the dynamic weight; training the generator in the preset model based on the training loss of the generator in the preset model.

[0008] Optionally, the preset model is obtained based on the conditional noise and the historical marketing data training, including: concatenating the conditional noise and the historical marketing data to obtain the input data of the preset model; using the input data of the preset model to train the discriminator in the preset model, wherein the discriminator includes a twin neural network, each neural network including multiple layers of one-dimensional convolution and fully connected layers.

[0009] Optionally, the discriminator in the preset model is trained using the input data of the preset model, including: forming a data set with the input data and the samples generated by the generator; selecting an anchor point, a positive sample probability and a negative sample probability from the data set, wherein the anchor point belongs to a preset standard quantity, the positive sample probability is real data, and the negative sample probability is a false sample generated by the generator; obtaining the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability; and training the discriminator in the preset model according to the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability.

[0010] Optionally, the discriminator in the preset model is trained according to the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability, including: constructing a ternary loss function according to the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability; and training the discriminator according to the ternary loss function.

[0011] According to another aspect of an embodiment of the present application, a marketing prediction device is also provided, including: an acquisition module, used to acquire portrait features of a to-be-marketed account, wherein the portrait features include at least: business usage of the to-be-marketed account; an analysis module, used to analyze the portrait features of the to-be-marketed account using a preset model, and determine the probability that the to-be-marketed account handles a target business, wherein the training data set of the preset model includes at least: conditional noise and historical marketing data, wherein the conditional noise is determined based on the training data set, and the conditional noise is used by the generator in the preset model to generate false samples; an output module, used to output the prediction result of the to-be-marketed account.

[0012] According to another aspect of the embodiment of the present application, a computer device is also provided, including: a memory and a processor, wherein the memory is used to store program instructions; and the processor is connected to the memory and is used to execute the above-mentioned marketing forecasting method.

[0013] According to another aspect of the embodiments of the present application, a non-volatile storage medium is provided, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned marketing forecasting method by running the computer program.

[0014] According to another aspect of the embodiments of the present application, a computer program product is provided, including computer instructions, which implement the above-mentioned marketing prediction method when executed by a processor.

[0015] In an embodiment of the present application, by obtaining the portrait features of the account to be marketed, the portrait features include at least: the business usage of the account to be marketed; using a preset model to analyze the portrait features of the account to be marketed, and determining the probability that the account to be marketed handles the target business, wherein the preset model is trained based on conditional noise and a training data set, wherein the conditional noise is determined based on the training data set, and the training data set includes at least: historical marketing data, and the conditional noise is used for the generator in the preset model to generate false samples; outputting the prediction result of the account to be marketed, thereby achieving the purpose of improving the correlation between noise and the training data set, and then improving the model performance, thereby achieving the technical effect of improving the prediction accuracy, and then solving the technical problem of low prediction accuracy in related technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0017] Figure 1 is a hardware structure block diagram of a computer terminal for implementing a marketing prediction method according to an embodiment of the present application;

[0018] Figure 2 is a flow chart of a marketing forecasting method according to an embodiment of the present application;

[0019] Figure 3 is a schematic diagram of a preset model structure according to an embodiment of the present application;

[0020] Figure 4 is a schematic diagram of comparing the correlation between noise and real data according to an embodiment of the present application;

[0021] Figure 5 is a schematic diagram of a conditional noise generation process according to an embodiment of the present application;

[0022] Figure 6 is a schematic diagram of a preset mask matrix and training data set processing flow according to an embodiment of the present application;

[0023] Figure 7 is a schematic diagram of a ternary loss training process according to an embodiment of the present application;

[0024] Figure 8 It is a structural diagram of a marketing forecasting device according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] 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. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] The information collected in the embodiments of the present application is information and data authorized by the account or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for the account to choose to authorize or reject the automated decision-making results; if the account chooses to reject, it enters the expert decision-making process.

[0028] In order to solve the problems existing in the related art, the present application embodiment provides a marketing prediction method, which can be run on Figure 1 In the computer terminal shown, the computer terminal is explained below.

[0029] The marketing prediction method embodiment provided in the embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal for implementing a marketing forecasting method. Figure 1As shown, the computer terminal 10 may include one or more (102a, 102b, ..., 102n are used to illustrate) processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission module 106 for communication functions connected via a wired and / or wireless network. In addition, it may also include: a display, a keyboard, a cursor control device, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, and a BUS bus. Those skilled in the art can understand that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0030] It should be noted that the one or more processors and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10. As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0031] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the marketing forecasting method in the embodiment of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, realizing the above-mentioned marketing forecasting method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0032] The transmission module 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission module 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission module 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0033] The display may be, for example, a touch screen liquid crystal display (LCD), which may enable the account to interact with the account interface of the computer terminal 10 .

[0034] It should be noted that, in some optional embodiments, the above Figure 1 The computer terminal shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. It should be noted that Figure 1 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computer terminal described above.

[0035] In the above operating environment, an embodiment of the present application provides an embodiment of a marketing forecasting method. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0036] Figure 2 is a flowchart of a marketing forecasting method according to an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0037] Step S202, obtaining the image characteristics of the account to be marketed, wherein the image characteristics at least include: the business usage of the account to be marketed;

[0038] In the above step S202, the account to be marketed may be an existing account in the database, and the service usage includes: service usage duration and type of service used.

[0039] Step S204, using a preset model to analyze the image features of the to-be-marketed account to determine the probability of the to-be-marketed account handling the target business, wherein the preset model is obtained by training based on conditional noise and a training data set, wherein the conditional noise is determined based on the training data set, and the training data set at least includes: historical marketing data, and the conditional noise is used by the generator in the preset model to generate false samples;

[0040] In the above step S204, there are multiple target services, such as broadband rate upgrade services, etc. Figure 3 As shown, the overall architecture of the preset model proposed in this application consists of three parts: 1. Conditional noise generator; 2. Discriminator; 3. Generator. The training process is divided into two stages. In the first stage, the conditional noise is embedded into the generator for pre-training in a supervised manner. In the second stage, the generative adversarial network is trained with a joint loss of supervised and unsupervised learning. This structure can contain more valuable additional information, including encoding the complete input sequence information and unsupervised adversarial information in the latent space.

[0041] For the training data set, it is expressed as: L represents the total step length, represents the feature vector at time t. The prediction task is to fit the conditional distribution p(X t:t+T-1 |X t-h:t-1 ). T represents the prediction step length, and h represents the length of the historical sequence. When T=1, it represents a single-step prediction, and when T>1, it represents a multi-step prediction, which is divided into short-term prediction and long-term prediction. Generally speaking, the prediction task uses Mean Squared Error (MSE) and Mean Absolute Error (MAE) as evaluation indicators of model performance.

[0042] Step S206, outputting the prediction result of the account to be marketed.

[0043] Through the above steps S202 to S206, the image features of the account to be marketed are obtained, and the image features at least include: the business usage of the account to be marketed; the image features of the account to be marketed are analyzed using a preset model to determine the probability of the account to be marketed handling the target business, wherein the preset model is trained based on conditional noise and a training data set, wherein the conditional noise is determined based on the training data set, and the training data set at least includes: historical marketing data, and the conditional noise is used for the generator in the preset model to generate false samples; the prediction result of the account to be marketed is output, and the purpose of improving the correlation between noise and the training data set and then improving the model performance is achieved, thereby achieving the technical effect of improving the prediction accuracy, and then solving the technical problem of low prediction accuracy in related technologies. The following is a detailed description.

[0044] In step S204 of the above marketing prediction method, the preset model is trained in the following manner, including: generating an initial noise embedding vector, the initial noise embedding vector conforming to the Gaussian distribution; extracting features from the historical marketing data to obtain a feature embedding vector, and inputting the feature embedding vector and the initial noise embedding vector into a conditional noise generator for pre-training to obtain the conditional noise, wherein the conditional noise generator includes a linear layer; and training the preset model based on the conditional noise and the training data set. In this way, the model can learn more complex account features and behavior patterns, improve the accuracy and reliability of predictions, and the effect is more significant, especially when dealing with large amounts of data and complex business scenarios.

[0045] In some embodiments of the present application, the linear correlation between conditional noise, random noise and real data can be calculated in the following manner, as shown in the following formula:

[0046] CM(i,j)=r(Z i ,Y j ,L)

[0047]

[0048] Z i = Z(:,i)

[0049] Y j =Y(:,j)

[0050] In the above formula, CM represents the linear correlation matrix, L represents the length of the data sequence, and Z i represents the noise vector in column i, Y j is the true value of the jth column, k represents the time point index of the sequence, and Z i,k represents the noise vector of the i-th column at the k-th time point, Yj,k represents the true value of the jth column at the kth time point, represents the average value of the noise vector in column i, represents the average value of the true value of the jth column. This method measures the linear correlation between the i-th variable of the noise and the j-th variable of the true value. On different data sets, there is almost no correlation between random noise and true data. Figure 3 The learning process of a preset model is shown as follows: Figure 4 As shown, in the first data set, conditional noise has a high correlation except for variable 1, and in the second data set, all variables of conditional noise have a significant correlation. The first data set and the second data set are different data sets. It can be understood that the colors in the data set area increase from black to yellow, and the correlation increases in sequence.

[0051] Figure 5 A schematic diagram of the conditional noise generation process is shown in FIG. Figure 5 As shown, a linear layer is used as a conditional noise generator, and a Gaussian distribution is used to embed the initial noise vector Initiate(Z)=p(Z)~N(μ,σ 2 ), and optimize the feature embedding vector simultaneously during pre-training and conditional noise generator The following formula shows how conditional noise is generated:

[0052] Conditional Noise=p(Z|X,Y,Θ)

[0053] Where Conditional Noise represents conditional noise, and p(Z|X,Y,Θ) is the probability distribution of conditional noise Z given the known feature embedding vector X, the initial noise embedding vector Y, and the linear layer parameter Θ.

[0054] In some embodiments of the present application, the preset model is obtained based on the conditional noise and the historical marketing data training, including: obtaining a preset mask matrix, which is a lower triangular matrix; multiplying the training data set with the preset mask matrix to obtain a masked training data set; splicing the conditional noise after the masked training data set to obtain the input data of the preset model; and obtaining the generator in the preset model by training based on the input data of the preset model.

[0055] like Figure 6 As shown in the left figure, the mask matrix is ​​a lower triangular matrix, each row of which is broadcasted with the data in the fixed-length training dataset Multiply to get the masked input data B represents the batch size of the input, such as Figure 6As shown in the right figure, each masking can obtain 11 sets of masking results to achieve the purpose of different input sequence lengths, and the masked sequence data will be filled with conditional noise with posterior information. Therefore, the final model input is the patch-like splicing result of the data in the training dataset and the conditional noise.

[0056] In some embodiments of the present application, the generator in the preset model is obtained based on the input data training of the preset model, including: obtaining the total number of iterations of the generator training and the number of iterations of the current training; determining the cosine function according to the total number of iterations of the generator training and the number of iterations of the current training; determining the dynamic weight according to the cosine function, and determining the training loss of the generator in the preset model based on the dynamic weight; training the generator in the preset model based on the training loss of the generator in the preset model.

[0057] Specifically, when training the generator, since it is necessary to use two losses, supervised and unsupervised, for multi-task learning, the present invention designs a cosine weight that changes dynamically with the training process to dynamically adjust the loss weights of the two tasks, so the training loss of the generator is G The calculation method is as follows:

[0058] loss G =(1-α)g(Ω)+α(Y-[H; Z] T Ω) 2

[0059]

[0060] In the formula, α represents the dynamic weight, g(Ω) represents the adversarial loss between the discriminator and the generator, Ω represents the parameters of the generator, H represents the input data, Y represents the real data, epoch represents the number of iterations of the current training, and train_epochs represents the total number of iterations of the generator training.

[0061] In some embodiments of the present application, the preset model is obtained based on the conditional noise and the historical marketing data training, including: concatenating the conditional noise and the historical marketing data to obtain the input data of the preset model; using the input data of the preset model to train the discriminator in the preset model, wherein the discriminator includes a twin neural network, and each neural network in the twin neural network includes multiple layers of one-dimensional convolution and fully connected layers.

[0062] Among them, one-dimensional convolution can effectively process time series data. In the sequence dimension, the multi-layer convolution structure can capture the time series features at different resolutions. In the variable dimension, the fully connected layer fuses these time series features to distinguish true from false. This is particularly important for processing time series data. It can be understood that the input data is time series data, and the forward process of the discriminator network can be expressed as:

[0063] f(X,W1,W2)=Sigmoid(FC(CNN(X,W1),W2))

[0064] Where f(X,W1,W2) represents the output data, X represents the input data, W1 represents the weight parameter of the convolutional neural network (CNN), W2 represents the weight parameter of the fully connected layer (FC), CNN(X,W1) represents the convolutional neural network part, receives the input data X and the weight W1 of the convolutional layer, extracts features through a series of convolutional layers, activation functions (usually nonlinear, such as ReLU) and pooling layers, FC(CNN(X,W1),W2 represents the fully connected layer, receives the output of CNN and its own weight W2, and maps the features extracted by CNN to the final decision or classification result. The fully connected layer is usually used after CNN to integrate features and perform classification.

[0065] In some embodiments of the present application, the discriminator in the preset model is trained using the input data of the preset model, including: forming a data set with the input data and the samples generated by the generator; selecting an anchor point, a positive sample probability and a negative sample probability from the data set, wherein the anchor point belongs to a preset standard quantity, the positive sample probability is real data, and the negative sample probability is a false sample generated by the generator; obtaining the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability; training the discriminator in the preset model according to the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability.

[0066] It should be noted that the anchor point can be the statistical center point in the training data set, such as the data mean or median; the anchor point can also be a representative sample in the training data set, for example: a sample located in the middle of the real data set in the feature space; the anchor point can also be dynamically selected, that is, adjusted according to the current generator output and discriminator state during network training. For example, the sample closest to the real data in the current generator output can be used as the anchor point for the next training. In the time series prediction task of the preset model, the anchor point can also be the time series characteristics that need to reflect the data, which can be the observation value at a specific time point in the data set, or the observation value when a specific event occurs, for example: the anchor point is 1.

[0067] In some embodiments of the present application, the discriminator in the preset model is trained according to the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability, including: constructing a ternary loss function according to the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability; and training the discriminator according to the ternary loss function.

[0068] In an optional approach, the goal of the ternary loss function is to minimize the distance between anchor point A and the probability of true samples, while maximizing the distance between anchor point A and the probability of false samples.

[0069] like Figure 7 As shown, in the embodiment of the present application, pairs of real data and fake data are input into the SNN discriminator, and the ternary loss relative to anchor point A (anchor point) is calculated. The task of the discriminator is to minimize the distance of the real data relative to anchor point A and maximize the distance of the fake data relative to anchor A. For example, in Figure 7 In the real sample 2 and fake sample 2 in a, the goal of the SNN discriminator is to minimize the positive distance (Positive Distance) while maximizing the negative distance (Negative Distance). This goal can be restated as minimizing the difference between these two distances, where the minimum distance difference between the real sample and the fake sample is Margin, and finally a discriminator that can effectively distinguish between real samples and fake samples is produced. The triple loss enables the network to learn the relative relationship between real samples and fake samples, rather than focusing on absolute classification information.

[0070] Based on the existing account group, the preset model is used to accurately predict their intention to apply for gigabit broadband services in the future. In essence, it is a time series prediction problem of multivariate prediction of single variable, but according to business needs, this problem can be transformed into a regression problem of probability estimation, and further quantitative evaluation can be performed according to binary classification. In order to accurately evaluate the classification results, the confusion matrix is ​​used for analysis, and the confusion matrix is ​​shown in Table 4.

[0071] Table 4 Confusion matrix

[0072]

[0073] Among them, TP represents the statistical number of account samples for which it is predicted that the account has handled business and the actual situation is that the account has handled business; FP represents the statistical number of account samples for which it is predicted that the account has not handled business but the actual situation is that the account has handled business; TN represents the statistical number of account samples for which it is predicted that the account has not handled business and the actual situation is that the account has not handled business; FN represents the statistical number of account samples for which it is predicted that the account has handled business but the actual situation is that the account has not handled business.

[0074] This invention uses precision, recall, F1 score, and AUC as quantitative evaluation indicators. Precision is used to count the percentage of marketing successes when all predicted results are marketing successes; while recall is the percentage of marketing successes when successful predictions are marketing successes among all actual marketing success accounts.

[0075] Specifically, the precision, recall, and F1-score can be determined by the following formula:

[0076]

[0077] The performance of the prediction model is evaluated according to the above evaluation indicators.

[0078] In the embodiment of the present application, in the account marketing scenario, the traditional generative adversarial network does not use prior knowledge (including historical marketing data) when generating "false" samples. They usually generate data such as images from random noise. In the account marketing scenario, the present application adds prior knowledge to the random noise, proposes parameterized conditional noise, and optimizes its parameters through supervised pre-training, converting it from random noise to conditional noise with posterior information, thereby improving the generated prediction quality; in addition, the embodiment of the present application proposes the use of a generative adversarial network CNGAN based on conditional noise, and explores the combination of SNN and GAN. By utilizing the characteristics of SNN and GANs, the "real" samples and the "false" samples for generating predictions are input into the SNN to calculate the ternary loss, which naturally constructs the adversarial nature of the GAN, thereby generating more robust predictions and effectively improving the prediction ability of the model.

[0079] Figure 8 is a structural diagram of a marketing prediction device according to an embodiment of the present application, such as Figure 8 As shown, the device comprises:

[0080] The acquisition module 80 is used to acquire the image characteristics of the account to be marketed, wherein the image characteristics at least include: the business usage of the account to be marketed;

[0081] An analysis module 82 is used to analyze the image features of the to-be-marketed account using a preset model to determine the probability of the to-be-marketed account handling a target business, wherein the training data set of the preset model at least includes: conditional noise and historical marketing data, wherein the conditional noise is determined based on the training data set, and the conditional noise is used by the generator in the preset model to generate false samples;

[0082] The output module 84 is used to output the prediction result of the account to be marketed.

[0083] Through the above-mentioned marketing prediction device, it is achieved by obtaining the portrait features of the account to be marketed, wherein the portrait features at least include: the business usage of the account to be marketed; using a preset model to analyze the portrait features of the account to be marketed, and determining the probability that the account to be marketed handles the target business, wherein the preset model is trained based on conditional noise and a training data set, wherein the conditional noise is determined based on the training data set, and the training data set at least includes: historical marketing data, and the conditional noise is used for the generator in the preset model to generate false samples; outputting the prediction result of the account to be marketed, achieving the purpose of improving the correlation between noise and the training data set, and then improving the model performance, thereby achieving the technical effect of improving the prediction accuracy, and then solving the technical problem of low prediction accuracy in related technologies.

[0084] The analysis module 82 in the above-mentioned marketing prediction device includes: a training submodule, which is used to train the preset model in the following manner, including: generating an initial noise embedding vector, which conforms to a Gaussian distribution; performing feature extraction on the historical marketing data to obtain a feature embedding vector, and inputting the feature embedding vector and the initial noise embedding vector into a conditional noise generator for pre-training to obtain the conditional noise, wherein the conditional noise generator includes a linear layer; and training the preset model based on the conditional noise and the training data set.

[0085] The training submodule includes: a first training unit, which is used to train the preset model based on the conditional noise and the historical marketing data, including: obtaining a preset mask matrix, which is a lower triangular matrix; multiplying the training data set with the preset mask matrix to obtain a masked training data set; splicing the conditional noise after the masked training data set to obtain the input data of the preset model; and training the generator in the preset model based on the input data of the preset model.

[0086] The first training unit includes: a first training subunit, which is used to train the generator in the preset model based on the input data of the preset model, including: obtaining the total number of iterations of the generator training and the number of iterations of the current training; determining the cosine function according to the total number of iterations of the generator training and the number of iterations of the current training; determining the dynamic weight according to the cosine function, and determining the training loss of the generator in the preset model based on the dynamic weight; training the generator in the preset model based on the training loss of the generator in the preset model.

[0087] The training submodule also includes: a second training unit, used to train the preset model based on the conditional noise and the historical marketing data, including: concatenating the conditional noise and the historical marketing data to obtain the input data of the preset model; using the input data of the preset model to train the discriminator in the preset model, wherein the discriminator includes a twin neural network, each neural network including multiple layers of one-dimensional convolution and fully connected layers.

[0088] The second training unit includes: a second training subunit, which is used to train the discriminator in the preset model using the input data of the preset model, including: forming a data set with the input data and the samples generated by the generator; selecting an anchor point, a positive sample probability and a negative sample probability from the data set, wherein the anchor point belongs to a preset standard quantity, the positive sample probability is real data, and the negative sample probability is a false sample generated by the generator; obtaining the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability; training the discriminator in the preset model according to the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability.

[0089] Among them, the discriminator in the preset model is trained according to the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability, including: constructing a ternary loss function according to the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability; training the discriminator according to the ternary loss function.

[0090] It should be noted that Figure 8 The marketing forecasting device shown is used to perform Figure 2 The marketing forecasting method shown, therefore the relevant explanations in the above marketing forecasting method are also applicable to the marketing forecasting device, and will not be repeated here.

[0091] An embodiment of the present application also provides a computer device, including: a memory and a processor, wherein the memory is used to store program instructions; the processor is connected to the memory and is used to execute the above-mentioned marketing forecasting method.

[0092] An embodiment of the present application also provides a non-volatile storage medium, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned marketing forecasting method by running the computer program.

[0093] An embodiment of the present application also provides a computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the marketing forecasting method in the present application are implemented.

[0094] An embodiment of the present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the marketing forecasting method in the present application.

[0095] The embodiment of the present application also provides a computer program, which, when executed by a processor, implements the steps of the marketing forecasting method in the present application.

[0096] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0097] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0098] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0099] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0100] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0101] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0102] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A marketing forecasting method, It is characterized in that include: Acquire the image features of the account to be marketed, wherein the image features at least include: the business usage of the account to be marketed; A preset model is used to analyze the image features of the to-be-marketed account to determine the probability of the to-be-marketed account handling the target business, wherein the preset model is obtained by training based on conditional noise and a training data set, wherein the conditional noise is determined based on the training data set, and the training data set at least includes: historical marketing data, and the conditional noise is used by a generator in the preset model to generate false samples; Output the prediction result of the account to be marketed.

2. The method according to claim 1, It is characterized in that The preset model is trained by the following methods, including: Generate an initial noise embedding vector, where the initial noise embedding vector conforms to a Gaussian distribution; Extracting features from the historical marketing data to obtain a feature embedding vector, and inputting the feature embedding vector and the initial noise embedding vector into a conditional noise generator for pre-training to obtain the conditional noise, wherein the conditional noise generator includes a linear layer; The preset model is obtained by training based on the conditional noise and the training data set.

3. The method according to claim 2, It is characterized in that The preset model is obtained by training based on the conditional noise and the historical marketing data, including: Obtain a preset mask matrix, where the preset mask matrix is ​​a lower triangular matrix; Multiplying the training data set by the preset mask matrix to obtain a masked training data set; splicing the conditional noise after the masked training data set to obtain input data of the preset model; The generator in the preset model is obtained by training based on the input data of the preset model.

4. The method according to claim 3, It is characterized in that The generator in the preset model is obtained by training the input data of the preset model, including: Obtain the total number of iterations of the generator training and the number of iterations of the current training; Determine a cosine function according to the total number of iterations of the generator training and the number of iterations of the current training; Determining a dynamic weight according to the cosine function, and determining a training loss of a generator in the preset model based on the dynamic weight; The generator in the preset model is trained based on the training loss of the generator in the preset model.

5. The method according to claim 2, It is characterized in that The preset model is obtained by training based on the conditional noise and the historical marketing data, including: splicing the conditional noise and the historical marketing data to obtain input data of the preset model; The discriminator in the preset model is trained using the input data of the preset model, wherein the discriminator includes a twin neural network, each of which includes multiple layers of one-dimensional convolution and fully connected layers.

6. The method according to claim 5, It is characterized in that Using the input data of the preset model to train the discriminator in the preset model includes: Combining the input data and the samples generated by the generator into a data set; Selecting an anchor point, a positive sample probability, and a negative sample probability from the data set, wherein the anchor point belongs to a preset standard quantity, the positive sample probability is real data, and the negative sample probability is a false sample generated by the generator; Obtaining the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability; The discriminator in the preset model is trained according to the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability.

7. The method according to claim 6, It is characterized in that The discriminator in the preset model is trained according to the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability, including: Constructing a ternary loss function according to the distance between the anchor point and the positive sample probability, the distance between the anchor point and the negative sample probability, and the distance between the positive sample probability and the negative sample probability; The discriminator is trained according to the ternary loss function.

8. A marketing prediction device, It is characterized in that include: Acquisition module, used to acquire accounts to be marketed; An analysis module, used for analyzing the to-be-marketed account by using a preset model to determine the probability of the to-be-marketed account handling a target business, wherein the preset model is obtained by training based on conditional noise and a training data set, wherein the conditional noise is determined based on the training data set, and the training data set at least includes: historical marketing data, and the conditional noise is used by a generator in the preset model to generate false samples; The output module is used to output the prediction results of the account to be marketed.

9. A computer device, It is characterized in that include: A memory and a processor, wherein the memory is used to store program instructions; The processor is connected to the memory, and is used to execute the marketing forecasting method described in any one of claims 1 to 7.

10. A computer program product comprising computer instructions, It is characterized in that When the computer instructions are executed by a processor, the marketing forecasting method according to any one of claims 1 to 7 is implemented.

Images