A method and system for constructing user portraits and delivering precise advertisements in digital marketing

Abstract

The present invention provides a method and system for constructing user portraits and precise advertising in digital marketing, and relates to the field of digital marketing technology. The method includes: S1, data collection and preprocessing, obtaining user data from social media, e-commerce platforms and mobile applications, and performing spatiotemporal normalization processing; S2, extracting features of behavioral data, hobbies and social network relationships respectively; S3, weighted fusion of different features through self-attention mechanism to construct user portraits; S4, using improved cosine similarity algorithm, combined with context-aware similarity calculation method to match advertising content; S5, optimizing advertising delivery decisions based on deep reinforcement learning; S6, continuously optimizing model parameters through real-time feedback data. The present invention can improve the accuracy of user portraits and the accuracy of advertising delivery, and significantly improve the effect of digital marketing.

Description

Technical Field

[0001] The present invention relates to the field of digital marketing technology, and in particular to a method and system for constructing user portraits and delivering precise advertisements in digital marketing. Background Art

[0002] In the current digital marketing field, accurate advertising and user portrait construction are two crucial links. However, existing technologies still have many technical problems in these aspects, which are specifically manifested in the following aspects:

[0003] Existing user portrait construction methods usually rely on a single or limited data source, such as users' browsing history, click behavior, and basic social network data. These methods lack comprehensive analysis of multi-dimensional information about users and cannot fully and accurately depict users' interests, behaviors, and preferences. As a result, the constructed user portraits often lack depth and accuracy, affecting the effectiveness of subsequent advertising.

[0004] In the process of multi-source data fusion, existing technologies face great challenges in data preprocessing and cleaning. The inconsistent data formats, quality, and update frequency of different data sources make the data fusion process complex and prone to errors. In addition, existing methods lack effective normalization and outlier processing mechanisms, resulting in low quality of fused data and affecting the accuracy of subsequent user portrait construction.

[0005] Existing technologies mostly use traditional statistical analysis methods for feature extraction and selection, such as principal component analysis (PCA) and linear discriminant analysis (LDA). Although these methods can extract the main features of data to a certain extent, they are difficult to capture the complex nonlinear relationships and deep information in the data. In addition, these methods have high computational complexity when processing large-scale high-dimensional data, making them difficult to be efficiently applied in practical systems.

[0006] Existing ad content matching methods mainly rely on simple keyword matching and basic similarity calculations, which fail to fully consider the user's contextual information and dynamically changing interests and hobbies. This matching method often fails to accurately reflect the user's current needs and preferences, resulting in a mismatch between ad content and user interests, and reducing the click-through rate and conversion rate of ads.

[0007] Most existing advertising delivery systems are based on rule engines or simple machine learning models, which lack the ability of adaptive learning and optimization. Traditional decision-making models make it difficult to dynamically adjust advertising delivery strategies and continuously optimize advertising effects based on real-time feedback data. In addition, existing technologies lack in-depth analysis of user behavior data during the advertising delivery decision-making process, resulting in poor accuracy of advertising delivery and poor user experience.

[0008] At present, most advertising delivery systems lack effective real-time feedback and optimization mechanisms, and are unable to adjust and optimize based on users' real-time behavior data. This static delivery strategy leads to large fluctuations in advertising effects, making it difficult to achieve the optimal configuration of advertising resources. Existing optimization methods mostly use linear models, fail to fully utilize advanced technologies such as deep learning and reinforcement learning, and have limited optimization effects.

[0009] In summary, existing technologies have many technical problems in user portrait construction and precise advertising, which makes it difficult to meet the current digital marketing needs for efficiency, accuracy and intelligence. These technical problems urgently need to be solved through innovative technical means to improve the overall effect of digital marketing. Summary of the invention

[0010] In order to solve the technical problems in the prior art of insufficient accuracy of user portraits, overly complex data preprocessing and fusion, limitations of feature extraction and selection methods, insufficient accuracy of advertising content matching, insufficient intelligence of advertising delivery decisions, and lack of real-time feedback and optimization mechanisms, the present invention provides a method and system for user portrait construction and precise advertising delivery in digital marketing.

[0011] The technical solution provided by the present invention is as follows:

[0012] First aspect:

[0013] The present invention provides a method for constructing user portraits and delivering precise advertisements in digital marketing, comprising:

[0014] S1. Data collection and preprocessing: obtaining user data from social media, e-commerce platforms and mobile applications, wherein the user data includes user behavior data, interests and hobbies, and social network relationships, and performing spatiotemporal normalization on the user data through a multi-level normalization method;

[0015] S2. Feature extraction is performed on behavioral data, hobbies and interests, and social network relationships respectively. Principal component analysis (PCA), linear discriminant analysis (LDA), deep neural network (DNN) and variable autoencoder (VAE) are used to extract features. Finally, the features of behavioral data, hobbies and interests, and social network relationships are expressed as, and;

[0016] S3, user portrait construction, weighted fusion of different features through the self-attention mechanism;

[0017] S4, advertising content matching, using an improved cosine similarity algorithm and combining it with a context-aware similarity calculation method to match advertising content;

[0018] S5, advertising placement decisions are optimized based on Deep Reinforcement Learning and use an improved Deep Q Network (DQN) algorithm;

[0019] S6. Continuously optimize model parameters through real-time feedback data after mobile phone advertising is launched.

[0020] Second aspect:

[0021] The present invention provides a user portrait construction and precise advertising delivery system in digital marketing, comprising:

[0022] processor;

[0023] A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method for constructing user portraits and delivering precise advertisements in digital marketing as described in the first aspect is implemented.

[0024] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0025] (1) In the present invention, by introducing multi-source data fusion and deep learning technology, multi-dimensional features are extracted from user behavior data, interest data and social network relationship data, and the user's interests, behaviors and preferences are comprehensively analyzed. By using improved principal component analysis (PCA), linear discriminant analysis (LDA), deep neural network (DNN) and variational autoencoder (VAE) and other methods, accurate extraction and fusion of user multi-dimensional information are achieved, thereby constructing a more comprehensive and accurate user portrait;

[0026] (2) In the present invention, an improved cosine similarity algorithm is used in combination with a context-aware similarity calculation method to match the advertising content with the user portrait. By introducing a self-attention mechanism, different features are weighted and fused, fully considering the user's contextual information and dynamically changing interests and hobbies, so that the advertising content is highly matched with the user's current needs, significantly improving the click-through rate and conversion rate of the advertisement;

[0027] (3) In the present invention, deep reinforcement learning technology is introduced, and the adaptive learning and optimization of advertising delivery strategies are realized through the improved deep Q network (DQN) algorithm. The system can dynamically adjust the advertising delivery strategy according to real-time feedback data (such as click-through rate and conversion rate), continuously optimize the advertising effect, optimize the advertising delivery decision through experience playback and fixed Q target strategy, ensure the optimal configuration of advertising resources, and significantly improve the accuracy of advertising delivery and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0029] Figure 1 A flowchart of a method for constructing a user portrait and delivering precise advertisements in digital marketing provided by an embodiment of the present invention;

[0030] Figure 2 A structural diagram of a user portrait construction and precise advertising delivery system in digital marketing provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0032] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0033] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0034] In the embodiments of the present invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are consistent.

[0035] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0036] Reference Manual Attached Figure 1 , showing a flow chart of a method and system for constructing user portraits and delivering precise advertisements in digital marketing provided by an embodiment of the present invention.

[0037] The embodiment of the present invention provides a method and system for constructing user portraits and delivering precise advertisements in digital marketing. The processing flow may include the following steps:

[0038] S1. Data collection and preprocessing: Obtain user data from social media, e-commerce platforms and mobile applications. User data includes user behavior data, interests and hobbies, and social network relationships. User data is normalized in time and space through a multi-level normalization method.

[0039] It should be noted that user data is obtained from social media, e-commerce platforms and mobile applications to build a rich user information foundation. User data includes user behavior data, interests and hobbies, and social network relationships. This step uses a multi-level normalization method to perform spatiotemporal normalization on user data to ensure data consistency and reliability, laying the foundation for subsequent feature extraction and user portrait construction.

[0040] S2. Feature extraction is performed on behavioral data, hobbies and interests, and social network relationships respectively. Principal component analysis (PCA), linear discriminant analysis (LDA), deep neural network (DNN) and variable autoencoder (VAE) are used for feature extraction. The final feature representation of behavioral data, hobbies and interests, and social network relationships is F b 、F i and F s .

[0041] It should be noted that feature extraction is performed on different types of data to extract key features that can represent user behavior, interests and social relationships. Advanced technical methods such as principal component analysis (PCA), linear discriminant analysis (LDA), deep neural network (DNN) and variational autoencoder (VAE) are used to deeply mine behavioral data, interests and hobbies, and social network relationships, and extract corresponding feature representations (F b 、F i 、F s ), providing high-quality feature data for user portrait construction.

[0042] S3: User portrait construction, weighted fusion of different features through the self-attention mechanism.

[0043] It should be noted that the self-attention mechanism is used to perform weighted fusion of different features, integrating behavioral data, hobbies and social network relationship features into a comprehensive user profile. The self-attention mechanism can dynamically adjust the weight of features so that the fused user profile can more accurately reflect the overall situation of the user, improving the accuracy and representativeness of the profile.

[0044] S4. Advertising content matching: an improved cosine similarity algorithm is used in combination with a context-aware similarity calculation method to perform advertising content matching.

[0045] It should be noted that the improved cosine similarity algorithm is used in combination with the context-aware similarity calculation method to match the advertising content with the user portrait. By introducing context information, the advertising content matching not only considers the basic portrait of the user, but also dynamically adapts to the user's current needs and environment, improves the matching degree between the advertising content and the user's interests, and increases the click-through rate and conversion rate of the advertisement.

[0046] S5. Advertising placement decisions are optimized based on deep reinforcement learning (DRL) and use an improved deep Q network (DQN) algorithm.

[0047] It should be noted that the advertising delivery strategy is optimized based on deep reinforcement learning technology. The improved deep Q network (DQN) algorithm is used to achieve adaptive learning and optimization of advertising delivery strategies. The system can continuously adjust the advertising delivery strategy based on real-time feedback data, optimize advertising effects, and improve the utilization efficiency and delivery accuracy of advertising resources.

[0048] S6. Continuously optimize model parameters through real-time feedback data after mobile phone advertising is launched.

[0049] It should be noted that by collecting real-time feedback data (such as click-through rate, conversion rate, etc.) after advertising, the model parameters are continuously optimized. Using real-time feedback data, the system can continuously learn and adjust, gradually improve the advertising effect, ensure that the advertising strategy is always in the best state, and significantly improve the overall advertising effect and user experience.

[0050] In a possible implementation, S1 specifically includes:

[0051] Assume that the user behavior dataset is D b ={d b1 , d b2 , ..., d bn}, the interest and hobby data set is D i ={d i1 , d i2 , ..., d im}, the social network relationship dataset is D s ={d s1 , d s2 , ..., d sk};

[0052] Let the time dimension normalization function be f time , the spatial dimension normalization function is f space , the preprocessed data is:

[0053] D″ b =f space (f time (D b )) D″ i =f space (f time (D i )) D″ s =f space (f time (D s ))

[0054] Among them, the time normalization function f time Defined as:

[0055]

[0056] Among them, t min and t max are the minimum and maximum timestamps in the dataset, respectively.

[0057] The spatial normalization function is defined as:

[0058]

[0059] Among them, s min and max They are the minimum and maximum geographic coordinate values ​​in the dataset, respectively.

[0060] It is understandable that by acquiring user behavior data, hobby data, and social network relationship data from social media, e-commerce platforms, and mobile applications, a rich user information foundation is established. The spatiotemporal normalization process ensures the consistency of the data in time and space, providing high-quality preprocessed data for subsequent feature extraction.

[0061] In a possible implementation, the behavior data feature extraction in S2 specifically includes:

[0062] The improved PCA algorithm is used to extract the principal components. The improved PCA function is assumed to be f ipca , extract the principal components:

[0063] F b1 =f ipca (D″ b )

[0064] The improved PCA algorithm introduces sparsity constraints and is defined as:

[0065]

[0066] Where W is the principal component matrix, and λ is the sparsity constraint parameter;

[0067] Use deep neural network to extract deep features of behavior data, and set the deep feature extraction function as f dnn :

[0068] F b2 =f dnn (D″ b )

[0069] The structure of a deep neural network is a multi-layer perceptron, defined as:

[0070] Input: D″ b

[0071] Layer 1: h1 = σ(W1D″ b +b1)

[0072] Layer 2: h2 = σ(W2D″ b +b2)

[0073] Output: F b2 =W3h2+b3

[0074] Among them, σ represents the activation function, W and b are weight and bias parameters, and the final feature set is:

[0075]

[0076] in, Represents a join operation on a feature.

[0077] In a possible implementation, the interest feature extraction in S2 specifically includes:

[0078] Use LDA for feature selection, assuming that the LDA function is f lda :

[0079] F i1 =f lda (D″ i )

[0080] The LDA algorithm performs feature selection by maximizing the ratio of between-class variance to within-class variance, which is defined as:

[0081]

[0082] Among them, S bis the inter-class scatter matrix, S w is the intra-class distribution matrix;

[0083] Use the flat powder autoencoder to extract the variational features of interests and hobbies, and set the variational feature extraction function as f vae :

[0084] F i2 =f vae (D″ i )

[0085] The variational autoencoder consists of an encoder and a decoder, defined as:

[0086] Encoder: z=μ+σ⊙∈,∈~N(0,1)

[0087]

[0088] Among them, μ and σ are the mean and standard deviation of the encoder output, g is the decoder function, and the final feature set is:

[0089]

[0090] in, Represents a join operation on a feature.

[0091] In a possible implementation, the social network relationship feature extraction in S2 specifically includes:

[0092] Use graph embedding method to extract features, and set the graph embedding function as f ge :

[0093] F s1 =f ge (D″ s )

[0094] The graph embedding method is based on random walks and node similarity, defined as:

[0095]

[0096] Where E is the edge set of the graph, and W is the node embedding matrix;

[0097] Use graph neural network to extract deep features of social network. Let the graph neural network feature extraction function be f gnn :

[0098] F s2 =f gnn (D″ s )

[0099] The propagation rule of graph neural network is:

[0100]

[0101] in, is the feature representation of node v at the kth layer, N(v) is the neighbor set of node v, cvu is the normalization coefficient, and the final feature set is:

[0102]

[0103] in, Represents a join operation on a feature.

[0104] In a possible implementation, S3 specifically includes:

[0105] Let the self-attention function be f att , then the fused features are expressed as:

[0106] F u =f att (F b , F i , F s )

[0107] The self-attention mechanism is defined as:

[0108]

[0109] Among them, W Q , W K and W V is the weight matrix of the attention mechanism, d k is the scaling factor;

[0110] The fusion features are processed through the deep learning model to obtain the final user portrait P u :

[0111] P u =MLP(F u )

[0112] The specific multi-layer perceptron model structure is:

[0113] Input: F u

[0114] Layer 1: h1=σ(W1F u +b1)

[0115] Layer 2: h2=σ(W2h1+b2)

[0116] Output: P u =W3h2+b3

[0117] Among them, σ represents the activation function, W and b are weight and bias parameters.

[0118] In a possible implementation, S4 specifically includes:

[0119] Let the user portrait vector be P u , the advertising content vector is A v , then the basic similarity is S basic :

[0120]

[0121] Define the context vector as C u , the context vector includes the current time, place and device, then the context-aware similarity is:

[0122]

[0123] The final similarity is S:

[0124] S=αS basic +(1-α)S context

[0125] Among them, α is the equilibrium parameter.

[0126] In a possible implementation, S5 specifically includes:

[0127] Define states t is the current user profile and context information, action a t For ad selection, reward r t is the effect of advertising after delivery, including click-through rate and conversion rate, and the state-action value function Q(s t , a t ) is approximated by a neural network:

[0128]

[0129] Among them, γ is the discount factor, θ and θ′ are the parameters of the current and target networks respectively;

[0130] Using Experience Replay and Fixed Q-Target strategies, the optimization goal is to minimize the loss function:

[0131]

[0132] Parameter update is performed via gradient descent:

[0133] In a possible implementation manner, S6 specifically includes:

[0134] Real-time feedback data includes click-through rate and conversion rate. Let the feedback data be F d The optimization goal is the long-term cumulative reward of drunken talk. The policy gradient method is used for optimization. The policy π(a|s) is defined as the probability of selecting an action under a given state. The optimization objective function J(θ) is:

[0135]

[0136] Update the policy parameters by gradient ascent:

[0137]

[0138] The specific gradient calculation is:

[0139]

[0140] in, is the discounted sum of future rewards.

[0141] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0142] (1) In the present invention, by introducing multi-source data fusion and deep learning technology, multi-dimensional features are extracted from user behavior data, interest data and social network relationship data, and the user's interests, behaviors and preferences are comprehensively analyzed. By using improved principal component analysis (PCA), linear discriminant analysis (LDA), deep neural network (DNN) and variational autoencoder (VAE) and other methods, accurate extraction and fusion of user multi-dimensional information are achieved, thereby constructing a more comprehensive and accurate user portrait;

[0143] (2) In the present invention, an improved cosine similarity algorithm is used in combination with a context-aware similarity calculation method to match the advertising content with the user portrait. By introducing a self-attention mechanism, different features are weighted and fused, fully considering the user's contextual information and dynamically changing interests and hobbies, so that the advertising content is highly matched with the user's current needs, significantly improving the click-through rate and conversion rate of the advertisement;

[0144] (3) In the present invention, deep reinforcement learning technology is introduced, and the adaptive learning and optimization of advertising delivery strategies are realized through the improved deep Q network (DQN) algorithm. The system can dynamically adjust the advertising delivery strategy according to real-time feedback data (such as click-through rate and conversion rate), continuously optimize the advertising effect, optimize the advertising delivery decision through experience playback and fixed Q target strategy, ensure the optimal configuration of advertising resources, and significantly improve the accuracy of advertising delivery and user experience.

[0145] Reference Manual Attached Figure 2 , showing a structural diagram of a user portrait construction and precise advertising delivery system in digital marketing provided by the present invention.

[0146] The present invention also provides a system for constructing user portraits and delivering precise advertisements in digital marketing, which is applied to the above-mentioned method for constructing user portraits and delivering precise advertisements in digital marketing, and includes:

[0147] processor.

[0148] A memory having computer-readable instructions stored thereon, which, when executed by a processor, implements a method for constructing user portraits and delivering precise advertisements in digital marketing as described in a method embodiment.

[0149] The system for constructing user portraits and delivering precise advertisements in digital marketing provided by the present invention can execute the above-mentioned method for constructing user portraits and delivering precise advertisements in digital marketing, and achieve the same or similar technical effects. To avoid repetition, the present invention will not go into details.

[0150] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0151] (1) In the present invention, by introducing multi-source data fusion and deep learning technology, multi-dimensional features are extracted from user behavior data, interest data and social network relationship data, and the user's interests, behaviors and preferences are comprehensively analyzed. By using improved principal component analysis (PCA), linear discriminant analysis (LDA), deep neural network (DNN) and variational autoencoder (VAE) and other methods, accurate extraction and fusion of user multi-dimensional information are achieved, thereby constructing a more comprehensive and accurate user portrait;

[0152] (2) In the present invention, an improved cosine similarity algorithm is used in combination with a context-aware similarity calculation method to match the advertising content with the user portrait. By introducing a self-attention mechanism, different features are weighted and fused, fully considering the user's contextual information and dynamically changing interests and hobbies, so that the advertising content is highly matched with the user's current needs, significantly improving the click-through rate and conversion rate of the advertisement;

[0153] (3) In the present invention, deep reinforcement learning technology is introduced, and the adaptive learning and optimization of advertising delivery strategies are realized through the improved deep Q network (DQN) algorithm. The system can dynamically adjust the advertising delivery strategy according to real-time feedback data (such as click-through rate and conversion rate), continuously optimize the advertising effect, optimize the advertising delivery decision through experience playback and fixed Q target strategy, ensure the optimal configuration of advertising resources, and significantly improve the accuracy of advertising delivery and user experience.

[0154] In summary, the embodiments of the present invention comprehensively improve the accuracy of user portrait construction, the precision of advertising content matching, and the intelligence level of advertising delivery decisions through advanced technical means such as multi-source data fusion, deep learning, and reinforcement learning, solve many technical problems in the existing technology, and greatly improve the overall effect of digital marketing.

[0155] The above contents are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0156] There are a few points to note:

[0157] (1) The drawings of the embodiments of the present invention only relate to the structures related to the embodiments of the present invention, and other structures may refer to the general design.

[0158] (2) For the sake of clarity, in the drawings used to describe the embodiments of the present invention, the thickness of the layers or regions is exaggerated or reduced, that is, these drawings are not drawn according to the actual scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or there may be intermediate elements.

[0159] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to obtain new embodiments.

[0160] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for constructing user portraits and delivering precise advertisements in digital marketing, characterized in that: include: S1. Data collection and preprocessing: obtaining user data from social media, e-commerce platforms and mobile applications, wherein the user data includes user behavior data, interests and hobbies, and social network relationships, and performing spatiotemporal normalization on the user data through a multi-level normalization method; S2. Feature extraction is performed on behavioral data, hobbies and interests, and social network relationships respectively. Principal component analysis (PCA), linear discriminant analysis (LDA), deep neural network (DNN) and variable autoencoder (VAE) are used for feature extraction. The final feature representation of behavioral data, hobbies and interests, and social network relationships is F b 、F i and F s ; S3, user portrait construction, weighted fusion of different features through the self-attention mechanism; S4, advertising content matching, using an improved cosine similarity algorithm and combining it with a context-aware similarity calculation method to match advertising content; S5, advertising placement decisions are optimized based on Deep Reinforcement Learning and use an improved Deep Q Network (DQN) algorithm; S6. Continuously optimize model parameters through real-time feedback data after mobile phone advertising is launched.

2. A method and system for constructing user portraits and delivering precise advertisements in digital marketing according to claim 1, characterized in that: The S1 specifically includes: Assume that the user behavior dataset is D b ={d b1 , d b2 , ..., d bn }, the interest and hobby data set is D i ={d i1， d i2 , ..., d im }, the social network relationship dataset is D s ={d s1 , d s2 , ..., d sk }; Let the time dimension normalization function be f time , the spatial dimension normalization function is f space , the preprocessed data is: D″ b =f space (f time (D b ))D″ i =f space (f time (D i ))D″ s =f space (f time (D s )) Among them, the time normalization function f time Defined as: Among them, t min and t max are the minimum and maximum timestamps in the dataset, respectively. The spatial normalization function is defined as: Among them, s min and max They are the minimum and maximum geographic coordinate values ​​in the dataset, respectively.

3. A method and system for constructing user portraits and delivering precise advertisements in digital marketing according to claim 1, characterized in that: The behavior data feature extraction in S2 specifically includes: The improved PCA algorithm is used to extract the principal components. The improved PCA function is assumed to be f ipca , extract the principal components: F b1 =f ipca (D″ b ) The improved PCA algorithm introduces sparsity constraints and is defined as: Where W is the principal component matrix, and λ is the sparsity constraint parameter; Use deep neural network to extract deep features of behavior data, and set the deep feature extraction function as f dnn : F b2 =f dnn (D″ b ) The structure of a deep neural network is a multi-layer perceptron, defined as: Input:D″ b Layer 1: h1=σ(W1D″ b +b1) Layer 2:h2=σ(W2D″ b +b2) Output:F b2 =W3h2+b3 Among them, σ represents the activation function, W and b are weight and bias parameters, and the final feature set is: in, Represents a join operation on a feature.

4. A method and system for constructing user portraits and delivering precise advertisements in digital marketing according to claim 1, characterized in that: The interest feature extraction in S2 specifically includes: Use LDA for feature selection, assuming that the LDA function is f lda : F i1 =f lda (D″ i ) The LDA algorithm performs feature selection by maximizing the ratio of between-class variance to within-class variance, which is defined as: Among them, S b is the inter-class scatter matrix, S w is the intra-class distribution matrix; Use the flat powder autoencoder to extract the variational features of interests and hobbies, and set the variational feature extraction function as f vae : F i2 =f vae (D″ i ) The variational autoencoder consists of an encoder and a decoder, defined as: Encoder: z=μ+σ⊙∈,∈~N(0,1) Among them, μ and σ are the mean and standard deviation of the encoder output, g is the decoder function, and the final feature set is: in, Represents a join operation on a feature.

5. A method and system for constructing user portraits and delivering precise advertisements in digital marketing according to claim 1, characterized in that: The social network relationship feature extraction in S2 specifically includes: Use graph embedding method to extract features, and set the graph embedding function as f ge : F s1 =f ge (D″ s ) The graph embedding method is based on random walks and node similarity, defined as: Where E is the edge set of the graph, and W is the node embedding matrix; Use graph neural network to extract deep features of social network. Let the graph neural network feature extraction function be f gnn : F s2 =f gnn (D″ s ) The propagation rule of graph neural network is: in, is the feature representation of node v at the kth layer, N(v) is the neighbor set of node v, cvu is the normalization coefficient, and the final feature set is: in, Represents a join operation on a feature.

6. A method and system for constructing user portraits and delivering precise advertisements in digital marketing according to claim 1, characterized in that: The S3 specifically includes: Let the self-attention function be f att , then the fused features are expressed as: F u =f att (F b ,F i ,F s ) The self-attention mechanism is defined as: Among them, W Q , W K and W V is the weight matrix of the attention mechanism, d k is the scaling factor; The fusion features are processed through the deep learning model to obtain the final user portrait P u : P u =MLP(F u ) The specific multi-layer perceptron model structure is: Input:F u Layer 1: h1=σ(W1F u +b1) Layer 2: h2=σ(W2h1+b2) Output:P u =W3h2+b3 Among them, σ represents the activation function, W and b are weight and bias parameters.

7. A method and system for constructing user portraits and delivering precise advertisements in digital marketing according to claim 1, characterized in that: The S4 specifically includes: Let the user portrait vector be P u , the advertising content vector is A v , then the basic similarity is S basic : Define the context vector as C u , the context vector includes the current time, place and device, then the context-aware similarity is: The final similarity is S: S=αS basic +(1-α)S context Among them, α is the equilibrium parameter.

8. A method and system for constructing user portraits and delivering precise advertisements in digital marketing according to claim 1, characterized in that: The S5 specifically includes: Define states t is the current user profile and context information, action a t For ad selection, reward r t is the effect of advertising after delivery, including click-through rate and conversion rate, and the state-action value function Q(s t , a t ) is approximated by a neural network: Among them, γ is the discount factor, θ and θ′ are the parameters of the current and target networks respectively; Using Experience Replay and Fixed Q-Target strategies, the optimization goal is to minimize the loss function: Parameter update is performed via gradient descent:

9. A method and system for constructing user portraits and delivering precise advertisements in digital marketing according to claim 1, characterized in that: The S6 specific package includes: The real-time feedback data includes click-through rate and conversion rate. Let the feedback data be F d The optimization goal is the long-term cumulative reward of drunken talk. The policy gradient method is used for optimization. The policy π(a|s) is defined as the probability of selecting an action under a given state. The optimization objective function J(θ) is: Update the policy parameters by gradient ascent: The specific gradient calculation is: in, is the discounted sum of future rewards.

10. A user portrait construction and precise advertising delivery system in digital marketing, characterized in that: include: processor; A memory having computer-readable instructions stored thereon, which, when executed by the processor, implements the method for constructing user portraits and delivering precise advertisements in digital marketing as described in any one of claims 1 to 9.

Images