A marketing method, system, device and medium based on user portrait

Abstract

The present invention discloses a marketing method, system, device and medium based on user portrait, which belongs to the field of marketing technology, including obtaining behavioral data of multiple users, and performing preprocessing operations to obtain preprocessed behavioral data, performing attribute feature extraction and probability prediction based on a user attribute model, constructing user portraits of multiple users, and grouping them to obtain multiple user groups; constructing a knowledge graph, extracting high-order neighbor node features based on a graph algorithm, using a marketing plan generation model to obtain multiple marketing plans, inputting the knowledge graph and multiple marketing plans into a recommendation model, and then selecting the marketing plan corresponding to the highest recommendation degree and pushing it to the corresponding user. The present invention fully processes the multidimensional characteristics of users through a user attribute model, wherein the user portrait includes multiple attribute features, solves the problem of single user portrait data, and selects the marketing plan with the highest recommendation degree to push to the customer, greatly improving the accuracy of marketing plan recommendations.

Description

Technical Field

[0001] The present invention belongs to the field of marketing technology, and specifically relates to a marketing method, system, device and medium based on user portraits. Background Art

[0002] With the rapid development of internet technology and big data, the amount of modern information has exploded, greatly enriching people's lives. However, it has also made it more time-consuming and difficult for people to obtain the information they need from data. As the number of users and interactive data increases, traditional data processing tools struggle to efficiently handle massive, high-dimensional user data. Furthermore, various user information is fragmented and distributed across different platforms or systems, which easily leads to data silos. As the number of users and interactive data increases, traditional data processing tools struggle to efficiently handle massive, high-dimensional user data, making it difficult to quickly conduct in-depth analysis.

[0003] Currently, common methods for constructing user profiles based on user data include statistical methods, machine learning methods, and deep learning methods. However, for more complex data distributions, the accuracy of feature extraction using statistical methods will affect the results. Machine learning and deep learning methods can only mine features from a single dimension when processing user data, ignoring features from other dimensions, resulting in an inability to fully utilize the information in the user data. Existing marketing systems generate and push marketing plans based on personalized user data. Since personalized user data is relatively simple, user profiles generated based on personalized user data may have the problem of single data, resulting in low accuracy in marketing plan recommendations. Therefore, there is an urgent need to provide a marketing method based on user profile solutions to address the problems of insufficient utilization of multi-dimensional features and low recommendation accuracy. Summary of the Invention

[0004] The purpose of the present invention is to provide a marketing method, system, device and medium based on user portraits to solve the above-mentioned problems existing in the prior art.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, the present invention provides a marketing method based on user portraits, comprising:

[0007] Obtaining behavioral data of multiple users, the behavioral data including cross-domain platform behavioral log data, UGC content, and social relationship graphs, and performing preprocessing operations on the behavioral data to obtain preprocessed behavioral data;

[0008] Input the preprocessed behavior data into the user attribute model to extract and predict attribute features, obtain user attribute features and user attribute probabilities, build user profiles of multiple users based on the user attribute features and user attribute probabilities, and group the user profiles of multiple users to obtain multiple user groups;

[0009] Obtain project portraits and the interaction between users and projects, and build a knowledge graph based on multiple user groups, project portraits, and the interaction between users and projects. The user portraits and project portraits in the user groups are used as entity nodes of the knowledge graph, and the interaction between users and projects is used as the edge of the knowledge graph.

[0010] Search for adjacent nodes of each entity node in the knowledge graph, extract features of adjacent nodes based on graph algorithms, obtain high-order neighbor node features, input user portraits and high-order neighbor node features into the marketing plan generation model to obtain multiple marketing plans, input the knowledge graph and multiple marketing plans into the recommendation model to obtain a recommendation degree sequence, select the marketing plan corresponding to the first recommendation degree in the recommendation degree sequence and push it to the corresponding user.

[0011] In a possible design, a preprocessing operation is performed on the behavior data to obtain preprocessed behavior data, including:

[0012] Perform data cleaning on the behavioral data, and fill in missing values ​​in the cleaned behavioral data to obtain complete behavioral data;

[0013] The complete behavioral data were denoised and normalized using the moving average method to obtain preprocessed behavioral data.

[0014] In one possible design, the user attribute model includes a pre-trained encoding layer, a DPCNN channel layer, a GRU and attention mechanism channel layer, and an output layer. The pre-processed behavior data is input into the user attribute model to extract and predict attribute features, thereby obtaining user attribute features and user attribute probabilities, including:

[0015] Inputting the preprocessed behavior data into the pretrained encoding layer to obtain data deep semantic information, and representing the data deep semantic information in a vectorized form to obtain a first data feature vector;

[0016] Input the first data feature vector into the DPCNN channel layer for feature extraction to obtain a second data feature vector;

[0017] Input the second data feature vector into the GRU and attention mechanism channel layer to obtain the first feature data;

[0018] The output layer concatenates and fuses the first data feature vector, the second data feature vector, and the first feature data to obtain the user's attribute features, and uses the sigmoid function to predict the user's attribute features to obtain the user's attribute probability.

[0019] In one possible design, the second data feature vector is input into the GRU and attention mechanism channel layer to obtain the first feature data, including:

[0020] Capturing the long-distance dependency in the second data feature vector based on the GRU model;

[0021] Use the attention mechanism layer to calculate the weights of the captured long-distance dependencies and obtain long-distance dependencies with different weights;

[0022] Features of long-distance dependencies with different weights are extracted to obtain first feature data.

[0023] In a possible design, the calculation expression of the attribute characteristics of the user and the attribute probability of the user is:

[0024] ;

[0025] In the above formula, represents the fusion feature data, represents the first data eigenvector, represents the second data eigenvector, represents the first feature data, represents the attribute probability of the user, represents the sigmoid function, represents the activation function, represents the weight, Represents the bias parameter.

[0026] In one possible design, user profiles of multiple users are grouped to obtain multiple user groups, including:

[0027] Randomly select a user's user profile as the initial cluster center;

[0028] Calculate the distance between each user's user profile and the initial cluster center, and assign each user's user profile to the cluster whose distance is less than the preset distance;

[0029] Calculate the mean of the user portraits in each cluster, use each mean as the cluster center of each cluster, calculate the distance from each user's user portrait to the cluster center again, and redistribute the user portraits to clusters with a distance less than the preset distance until the preset termination condition is reached, and obtain multiple user groups.

[0030] In one possible design, the recommendation model includes an embedding layer, an attention embedding propagation layer, and a prediction layer; the knowledge graph and multiple marketing plans are input into the recommendation model to obtain a recommendation degree sequence, including:

[0031] The embedding layer embeds the knowledge graph and multiple marketing plans into a low-dimensional vector space to obtain the initial embedding vector;

[0032] Input the initial embedding vector into the attention embedding propagation layer for calculation to obtain the marketing plan weight coefficient;

[0033] In the prediction layer, a graph neural network is used to calculate the knowledge graph and the weight coefficients of multiple marketing plans to obtain multiple recommendation degrees, which are then sorted to obtain a recommendation degree sequence.

[0034] In a second aspect, the present invention provides a marketing system based on user portraits, comprising:

[0035] An acquisition module is used to acquire behavioral data of multiple users, the behavioral data including behavioral log data, UGC content, and social relationship graphs across cross-domain platforms, and perform preprocessing operations on the behavioral data to obtain preprocessed behavioral data;

[0036] A feature module is used to input the pre-processed behavior data into the user attribute model to extract and predict attribute features, obtain user attribute features and user attribute probabilities, construct user profiles of multiple users based on the user attribute features and user attribute probabilities, and group the user profiles of multiple users to obtain multiple user groups;

[0037] A construction module is used to obtain project portraits and the interaction between users and projects, and to construct a knowledge graph based on multiple user groups, project portraits, and the interaction between users and projects. The user portraits and project portraits in the user groups are used as entity nodes of the knowledge graph, and the interaction between users and projects is used as the edge of the knowledge graph.

[0038] The recommendation module is used to find the adjacent nodes of each entity node in the knowledge graph, extract the features of the adjacent nodes based on the graph algorithm, obtain the features of the high-order neighbor nodes, input the user portrait and the features of the high-order neighbor nodes into the marketing plan generation model, obtain multiple marketing plans, input the knowledge graph and multiple marketing plans into the recommendation model, obtain the recommendation degree sequence, and select the marketing plan corresponding to the first recommendation degree in the recommendation degree sequence and push it to the corresponding user.

[0039] In a third aspect, the present invention provides a computer device comprising a memory, a processor and a transceiver communicatively connected in sequence, wherein the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the user portrait-based marketing method as described in the first aspect.

[0040] In a fourth aspect, the present invention provides a computer-readable storage medium having instructions stored thereon. When the instructions are run on a computer, the user portrait-based marketing method as described in the first aspect is executed.

[0041] The beneficial effects of the present invention are as follows:

[0042] The present invention discloses a marketing method, system, device and medium based on user portraits. First, the behavioral data of multiple users are acquired and preprocessed to obtain preprocessed behavioral data, attribute features are extracted and predicted using a user attribute model, user portraits of multiple users are constructed based on the user attribute features and attribute probabilities, and the user portraits are grouped to obtain multiple user groups, project portraits and the interaction relationship between users and projects are acquired, a knowledge graph is constructed based on the multiple user groups, project portraits and the interaction relationship between users and projects, high-order neighbor node features are obtained according to the knowledge graph, a marketing plan generation model is generated based on the user portraits and high-order neighbor node features to obtain multiple marketing plans, a recommendation model is used to generate and sort recommendation degrees to obtain a recommendation degree sequence, and a marketing plan corresponding to the first recommendation degree is selected and pushed to the corresponding user. The user attribute model fully processes the multi-dimensional characteristics of users and constructs a user portrait. The user portrait includes multiple user attribute features, which solves the problem of single user portrait data. A knowledge graph is constructed based on the user portrait and project portrait to intuitively reflect the preference relationship between users and projects. After generating the marketing plan, the marketing plans are sorted and the marketing plan with the highest recommendation is selected and pushed to customers, which greatly improves the accuracy of marketing plan recommendations, maximizes marketing effects, reduces marketing costs, improves return on investment, and facilitates practical application and promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A flowchart of the user portrait-based marketing method provided in the first aspect of this embodiment;

[0044] Figure 2 This is a module block diagram of the user portrait-based marketing system provided in the second aspect of this embodiment. DETAILED DESCRIPTION

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the present invention will be briefly introduced below in conjunction with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0046] It should be understood that although the terms "first," "second," etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first element can be referred to as a second element, and similarly, a second element can be referred to as a first element without departing from the scope of the exemplary embodiments of the present invention.

[0047] Example:

[0048] like Figure 1 As shown, the first aspect of this embodiment provides a marketing method based on user portraits, including but not limited to the following steps:

[0049] S1. Obtaining behavioral data of multiple users, including cross-domain platform behavioral log data, UGC (User-Generated Content), and social relationship graphs, and preprocessing the behavioral data to obtain preprocessed behavioral data;

[0050] Specifically, in step S1, the behavior data is preprocessed to obtain preprocessed behavior data, including:

[0051] S11. Clean the behavioral data and fill missing values ​​in the cleaned behavioral data to obtain complete behavioral data;

[0052] S12. Perform moving average denoising and normalization on the complete behavioral data to obtain preprocessed behavioral data.

[0053] In this embodiment, the behavioral data is cleaned and the missing parts are filled to prevent outliers and missing values ​​from affecting the subsequent data processing process. The moving average method is used to denoise and normalize the complete behavioral data to reduce the impact of noise and improve the robustness of the model. The principle of moving average denoising is to remove noise in the data by calculating the local average of each data point in the data set.

[0054] S2. Input the preprocessed behavior data into the user attribute model to extract and predict attribute features, obtain user attribute features and user attribute probabilities, construct user profiles for multiple users based on the user attribute features and user attribute probabilities, and group the multiple user profiles to obtain multiple user groups;

[0055] Furthermore, the user attribute model includes a pre-trained encoding layer, a DPCNN (Deep Pyramid Convolutional Neural Networks for Text Categorization) channel layer, a GRU (Gated Recurrent Unit) and attention mechanism channel layer, and an output layer.

[0056] Specifically, in step S2, the pre-processed behavior data is input into the user attribute model to extract and predict attribute features, thereby obtaining the user's attribute features and the user's attribute probability, including:

[0057] S21. Input the pre-processed behavior data into the pre-trained encoding layer to obtain data deep semantic information, and express the data deep semantic information in a vectorized form to obtain a first data feature vector;

[0058] S22. Input the first data feature vector into the DPCNN channel layer for feature extraction to obtain a second data feature vector;

[0059] S23. Input the second data feature vector into the GRU and attention mechanism channel layer to obtain the first feature data;

[0060] Specifically, in step S23, the second data feature vector is input into the GRU and attention mechanism channel layer to obtain the first feature data, including:

[0061] S231. Capturing long-distance dependencies in the second data feature vector based on the GRU model;

[0062] Specifically, in step S231, the GRU model mainly uses two gating units, which are the update gate and the reset gate. The update gate is used to control how much historical information needs to be retained in the current state and how much new information needs to be received. The reset gate is used to determine how much information about the network state at the previous moment is written. The specific calculation formulas for the update gate and the reset gate are as follows:

[0063] ;

[0064] In the above formula, Indicates the current input, Indicates the state at the previous moment. Indicates the current state. represents the activation function, Indicates the current learnable connection weight matrix of the update gate, Represents the reset gate's current learnable connection weight matrix, Indicates that the connection weight matrix can be learned at the moment before the update gate, Indicates that the connection weight matrix can be learned at the moment before resetting the gate, represents the learnable offset vector of the update gate, Represents the learnable offset vector for the reset gate.

[0065] S232. Use the attention mechanism layer to calculate the weights of the captured long-distance dependencies to obtain long-distance dependencies with different weights;

[0066] Generally, the attention mechanism can be divided into hard attention mechanism, soft attention mechanism and self-attention mechanism. Preferably, in this embodiment, the attention mechanism is a self-attention mechanism, which can efficiently calculate the attention weights of all positions in parallel, and learn the relative importance of each position to other positions. It can effectively capture the relationship between different positions in the text sequence, so that the neural network pays more attention to important features and suppresses the influence of irrelevant features.

[0067] S233. Extract features of long-distance dependencies with different weights to obtain first feature data.

[0068] S24. The output layer concatenates and fuses the first data feature vector, the second data feature vector, and the first feature data to obtain the user's attribute features, and uses the sigmoid function to predict the user's attribute features to obtain the user's attribute probability.

[0069] Furthermore, the calculation expression of the attribute characteristics of the user and the attribute probability of the user is:

[0070] ;

[0071] In the above formula, represents the fusion feature data, represents the first data eigenvector, represents the second data eigenvector, represents the first feature data, represents the attribute probability of the user, represents the sigmoid function, represents the activation function, represents the weight, Represents the bias parameter.

[0072] Specifically, in step S2, the user profiles of multiple users are grouped to obtain multiple user groups, including:

[0073] S25. Randomly select a user profile as the initial cluster center;

[0074] S26. Calculate the distance between each user's user profile and the initial cluster center, and assign each user's user profile to a cluster whose distance is less than a preset distance;

[0075] S27. Calculate the mean of the user portraits in each cluster, use each mean as the cluster center of each cluster, recalculate the distance from each user's user portrait to the cluster center, and redistribute the user portraits to clusters with a distance less than a preset distance until the preset termination condition is reached, thereby obtaining multiple user groups.

[0076] S3. Obtain project profiles and user-project interactions, and construct a knowledge graph based on multiple user groups, project profiles, and user-project interactions. The user profiles and project profiles in the user groups serve as entity nodes in the knowledge graph, and the user-project interactions serve as edges in the knowledge graph.

[0077] S4. Search for the adjacent nodes of each entity node in the knowledge graph, extract the features of the adjacent nodes based on the graph algorithm, obtain the features of the high-order neighbor nodes, input the user portrait and the features of the high-order neighbor nodes into the marketing plan generation model to obtain multiple marketing plans, input the knowledge graph and the multiple marketing plans into the recommendation model to obtain the recommendation degree sequence, select the marketing plan corresponding to the first recommendation degree in the recommendation degree sequence and push it to the corresponding user.

[0078] In step S4, the marketing plan generation model is an existing technology, and a recurrent neural network (RNN) and a transformer can be selected to generate a marketing plan, or a generative adversarial network can be used to generate a marketing plan.

[0079] Furthermore, the recommendation model includes an embedding layer, an attention embedding propagation layer and a prediction layer.

[0080] Specifically, in step S4, the knowledge graph and multiple marketing plans are input into the recommendation model to obtain a recommendation degree sequence, including:

[0081] S41. The embedding layer embeds the knowledge graph and multiple marketing plans into a low-dimensional vector space to obtain an initial embedding vector.

[0082] Among them, the representation of the knowledge graph is a triple (h, r, t), where h and t are entity nodes in the knowledge graph, and r is the connection relationship in the knowledge graph (that is, the edge in the knowledge graph). Multiple triplets can share the same entity, and the entity acts as the head or tail in different triplets, serving as the association medium for different triplets.

[0083] S42. Input the initial embedding vector into the attention embedding propagation layer for calculation to obtain the marketing plan weight coefficient;

[0084] S43. Use graph neural networks in the prediction layer to calculate the knowledge graph and the weight coefficients of multiple marketing plans to obtain multiple recommendation degrees, and sort the multiple recommendation degrees to obtain a recommendation degree sequence.

[0085] This embodiment provides a marketing method based on user portraits, including obtaining behavioral data of multiple users, wherein the behavioral data includes behavioral log data of cross-domain platforms, UGC content and social relationship graphs, performing preprocessing operations on the behavioral data to obtain preprocessed behavioral data; inputting the preprocessed behavioral data into a user attribute model to extract and predict attribute features to obtain user attribute features and user attribute probabilities, constructing user portraits of multiple users based on the user attribute features and user attribute probabilities, grouping the user portraits of multiple users to obtain multiple user groups; obtaining project portraits and the interactive relationship between users and projects, and A knowledge graph is constructed based on user groups, project portraits, and the interactions between users and projects. User portraits and project portraits in user groups are used as entity nodes in the knowledge graph, and the interactions between users and projects are used as edges in the knowledge graph. Adjacent nodes of each entity node are searched in the knowledge graph, and features of adjacent nodes are extracted based on graph algorithms to obtain high-order neighbor node features. These user portraits and high-order neighbor node features are then input into a marketing plan generation model to obtain multiple marketing plans. The knowledge graph and multiple marketing plans are then input into a recommendation model to obtain a recommendation degree sequence. The marketing plan corresponding to the first recommendation degree in the recommendation degree sequence is selected and pushed to the corresponding user. User attribute features and attribute probabilities are extracted from the user attribute model to construct user portraits. A clustering algorithm is then used to group multiple user portraits. A knowledge graph is constructed based on user groups, projects, and the relationships between them. This effectively integrates multiple types of data, comprehensively captures user and project features, and enhances the richness of the knowledge graph to improve recommendation accuracy. The marketing plan generation model and recommendation model are used in conjunction to maximize marketing effectiveness, while improving user experience and user stickiness.

[0086] like Figure 2 As shown, the second aspect of this embodiment provides a marketing system based on user portraits, including:

[0087] An acquisition module is used to acquire behavioral data of multiple users, the behavioral data including behavioral log data, UGC content, and social relationship graphs across cross-domain platforms, and perform preprocessing operations on the behavioral data to obtain preprocessed behavioral data;

[0088] A feature module is used to input the preprocessed behavior data into the user attribute model to extract and predict attribute features, obtain user attribute features and user attribute probabilities, construct user profiles of multiple users based on the user attribute features and user attribute probabilities, and group the user profiles of multiple users to obtain multiple user groups;

[0089] A construction module is used to obtain project portraits and the interaction between users and projects, and to construct a knowledge graph based on multiple user groups, project portraits, and the interaction between users and projects. The user portraits and project portraits in the user groups are used as entity nodes of the knowledge graph, and the interaction between users and projects is used as the edge of the knowledge graph.

[0090] The recommendation module is used to find the adjacent nodes of each entity node in the knowledge graph, extract the features of the adjacent nodes based on the graph algorithm, obtain the features of the high-order neighbor nodes, input the user portrait and the features of the high-order neighbor nodes into the marketing plan generation model, obtain multiple marketing plans, input the knowledge graph and multiple marketing plans into the recommendation model, obtain the recommendation degree sequence, and select the marketing plan corresponding to the first recommendation degree in the recommendation degree sequence and push it to the corresponding user.

[0091] The working process, working details and technical effects of the aforementioned system provided in the second aspect of this embodiment can be found in the marketing method described in the first aspect and will not be described in detail here.

[0092] A third aspect of this embodiment provides a computer device for executing the marketing method described in the first aspect, comprising a memory, a processor, and a transceiver communicatively connected in sequence. The memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the marketing method described in the first aspect. Specifically, the memory may include, but is not limited to, random-access memory (RAM), read-only memory (ROM), flash memory, first-input first-output (FIFO), and / or first-input last-output (FILO) memory, among others. The processor may include, but is not limited to, a microprocessor from the STM32F105 series. Furthermore, the computer device may include, but is not limited to, a power supply module, a display screen, and other necessary components.

[0093] The working process, working details and technical effects of the aforementioned computer equipment provided in the third aspect of this embodiment can be found in the marketing method described in the first aspect and will not be described in detail here.

[0094] A fourth aspect of this embodiment provides a computer-readable storage medium storing instructions containing the marketing method described in the first aspect, that is, the computer-readable storage medium stores instructions that, when executed on a computer, execute the marketing method described in the first aspect. The computer-readable storage medium refers to a carrier for storing data, and may include, but is not limited to, computer-readable storage media such as a floppy disk, an optical disk, a hard disk, a flash memory, a USB flash drive, and / or a memory stick. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0095] The working process, working details and technical effects of the aforementioned computer-readable storage medium provided in the fourth aspect of this embodiment can be referred to the marketing method described in the first aspect and will not be repeated here.

[0096] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the scope of protection of the present invention.

Claims

1. A marketing method based on user portrait, characterized in that: include: Obtaining behavioral data of multiple users, the behavioral data including cross-domain platform behavioral log data, UGC content, and social relationship graphs, and performing preprocessing operations on the behavioral data to obtain preprocessed behavioral data; Input the preprocessed behavior data into the user attribute model to extract and predict attribute features, obtain user attribute features and user attribute probabilities, build user profiles of multiple users based on the user attribute features and user attribute probabilities, and group the user profiles of multiple users to obtain multiple user groups; Obtain project portraits and the interaction between users and projects, and build a knowledge graph based on multiple user groups, project portraits, and the interaction between users and projects. The user portraits and project portraits in the user groups are used as entity nodes of the knowledge graph, and the interaction between users and projects is used as the edge of the knowledge graph. Search for adjacent nodes of each entity node in the knowledge graph, extract features of adjacent nodes based on graph algorithms, obtain high-order neighbor node features, input user portraits and high-order neighbor node features into the marketing plan generation model to obtain multiple marketing plans, input the knowledge graph and multiple marketing plans into the recommendation model to obtain a recommendation degree sequence, select the marketing plan corresponding to the first recommendation degree in the recommendation degree sequence and push it to the corresponding user.

2. A marketing method based on user portrait according to claim 1, characterized in that: Perform preprocessing operations on the behavior data to obtain preprocessed behavior data, including: Perform data cleaning on the behavioral data, and fill in missing values ​​in the cleaned behavioral data to obtain complete behavioral data; The complete behavioral data were denoised and normalized using the moving average method to obtain preprocessed behavioral data.

3. The marketing method based on user portrait according to claim 1, characterized in that: The user attribute model includes a pre-trained encoding layer, a DPCNN channel layer, a GRU and attention mechanism channel layer, and an output layer; The pre-processed behavior data is input into the user attribute model to extract and predict attribute features, thereby obtaining the user's attribute features and the user's attribute probability, including: Inputting the preprocessed behavior data into the pretrained encoding layer to obtain data deep semantic information, and representing the data deep semantic information in a vectorized form to obtain a first data feature vector; Input the first data feature vector into the DPCNN channel layer for feature extraction to obtain a second data feature vector; Input the second data feature vector into the GRU and attention mechanism channel layer to obtain the first feature data; The output layer concatenates and fuses the first data feature vector, the second data feature vector, and the first feature data to obtain the user's attribute characteristics, and uses the sigmoid function to predict the user's attribute characteristics to obtain the user's attribute probability.

4. A marketing method based on user portrait according to claim 3, characterized in that: Input the second data feature vector into the GRU and attention mechanism channel layer to obtain the first feature data, including: Capturing the long-distance dependency in the second data feature vector based on the GRU model; Use the attention mechanism layer to calculate the weights of the captured long-distance dependencies and obtain long-distance dependencies with different weights; Features of long-distance dependencies with different weights are extracted to obtain first feature data.

5. The marketing method based on user portrait according to claim 3, characterized in that: The calculation expressions of the user's attribute characteristics and the user's attribute probability are as follows: ； In the above formula, represents the fusion feature data, represents the first data eigenvector, represents the second data eigenvector, represents the first feature data, represents the attribute probability of the user, represents the sigmoid function, represents the activation function, represents the weight, Represents the bias parameter.

6. The marketing method based on user portrait according to claim 1, characterized in that: Group the user portraits of multiple users to obtain multiple user groups, including: Randomly select a user's user profile as the initial cluster center; Calculate the distance between each user's user profile and the initial cluster center, and assign each user's user profile to the cluster whose distance is less than the preset distance; Calculate the mean of the user portraits in each cluster, use each mean as the cluster center of each cluster, calculate the distance from each user's user portrait to the cluster center again, and redistribute the user portraits to clusters with a distance less than the preset distance until the preset termination condition is reached, and obtain multiple user groups.

7. The marketing method based on user portrait according to claim 1, characterized in that: The recommendation model includes an embedding layer, an attention embedding propagation layer, and a prediction layer. The knowledge graph and multiple marketing plans are input into the recommendation model to obtain a recommendation degree sequence, including: The embedding layer embeds the knowledge graph and multiple marketing plans into a low-dimensional vector space to obtain the initial embedding vector; Input the initial embedding vector into the attention embedding propagation layer for calculation to obtain the marketing plan weight coefficient; In the prediction layer, a graph neural network is used to calculate the knowledge graph and the weight coefficients of multiple marketing plans to obtain multiple recommendation degrees, which are then sorted to obtain a recommendation degree sequence.

8. A marketing system based on user portraits, characterized in that: include: An acquisition module is used to acquire behavioral data of multiple users, the behavioral data including behavioral log data, UGC content, and social relationship graphs across cross-domain platforms, and perform preprocessing operations on the behavioral data to obtain preprocessed behavioral data; A feature module is used to input the pre-processed behavior data into the user attribute model to extract and predict attribute features, obtain user attribute features and user attribute probabilities, construct user profiles of multiple users based on the user attribute features and user attribute probabilities, and group the user profiles of multiple users to obtain multiple user groups; A construction module is used to obtain project portraits and the interaction between users and projects, and to construct a knowledge graph based on multiple user groups, project portraits, and the interaction between users and projects. The user portraits and project portraits in the user groups are used as entity nodes of the knowledge graph, and the interaction between users and projects is used as the edge of the knowledge graph. The recommendation module is used to find the adjacent nodes of each entity node in the knowledge graph, extract the features of the adjacent nodes based on the graph algorithm, obtain the features of the high-order neighbor nodes, input the user portrait and the features of the high-order neighbor nodes into the marketing plan generation model, obtain multiple marketing plans, input the knowledge graph and multiple marketing plans into the recommendation model, obtain the recommendation degree sequence, and select the marketing plan corresponding to the first recommendation degree in the recommendation degree sequence and push it to the corresponding user.

9. A computer device, characterized in that: It includes a memory, a processor and a transceiver that are communicatively connected in sequence, wherein the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the user portrait-based marketing method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, and when the instructions are executed on a computer, the user portrait-based marketing method according to any one of claims 1 to 7 is executed.

Images