Method and device for constructing social media ELMENT model

Abstract

The embodiment of the invention provides a method and device for constructing a social media ELMENT model, and belongs to the technical field of data processing. The method comprises the steps that user historical behavior data DH is analyzed to obtain a behavior feature set KF; determining grouping information FG based on the KF and the current behavior state ZC; determining a first push content P1 according to the FG, the behavior stage information JE, the behavior feedback KR and the environment state HS; using the user feedback data DF and the hyper-parameters theta g and theta h of the optimization functions g and h to optimize P1 to obtain P2 and a push strategy TS; and determining a push frequency Fr and a content form Fc according to the P2, the TS and the behavior characteristics KB of the user on different platforms. According to the invention, the accuracy of the pushed content is improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, and electronic device for constructing a social media ELEMENT model. Background Technology

[0002] In today's social media marketing landscape, traditional marketing models suffer from numerous limitations. Most rely solely on simple user information for promotion, failing to accurately grasp dynamic changes in user behavior and differences in needs. There's a lack of in-depth analysis and integration of complex user behavioral data across different platforms, such as operating habits, emotional tendencies, and interaction patterns. Simultaneously, content delivery strategies are often simplistic and static, failing to dynamically adjust based on real-time user feedback, resulting in poor marketing effectiveness, a subpar user experience, and an inability to achieve precise and personalized marketing, ultimately impacting the effectiveness and conversion rates of marketing campaigns.

[0003] The aforementioned issues have become technical problems that need to be solved. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a method, apparatus, and electronic device for constructing a social media ELEMENT model, which at least partially solves the problems existing in the prior art.

[0005] In a first aspect, embodiments of the present invention provide a method for constructing a social media ELEMENT model, comprising:

[0006] Analyze the acquired historical user behavior data DH to obtain the user behavior feature set KF;

[0007] Based on behavioral characteristics KF and the user's current behavioral state ZC, the user's grouping information FG is determined;

[0008] Based on the user's group information FG, the user's behavioral stage information JE, the user's behavioral feedback KR, and the environmental status HS, determine the first push content P1;

[0009] Based on user feedback data DF and the hyperparameters θg and θh of optimization functions g and h, the first push content P1 is optimized to obtain the second push content P2 and the push strategy TS corresponding to the second push content P2.

[0010] Based on the second push content P2, the push strategy TS, and the user's behavioral characteristics KB on different social media platforms, the push frequency Fr and content format Fc are determined.

[0011] According to a specific implementation of an embodiment of the present invention, the step of analyzing the acquired historical behavior data DH of the user to obtain the user's behavior feature set KF includes:

[0012] Data cleaning algorithms are used to perform preliminary screening of users' historical behavior data (DH) to remove erroneous or invalid data.

[0013] Extract different types of features from the cleaned data;

[0014] Feature selection algorithms are used to filter out features that are highly relevant to user behavior from the original feature set;

[0015] The filtered features are transformed to have a uniform scale;

[0016] Different types of features are combined to form a user's behavioral feature set KF.

[0017] According to a specific implementation of an embodiment of the present invention, determining the user's grouping information FG based on behavioral features KF and the user's current behavioral state ZC includes:

[0018] The user's current behavioral state ZC is quantified and represented.

[0019] The user feature vector is constructed by integrating the behavioral feature set KF with the quantified behavioral state vector.

[0020] Set the clustering algorithm and its initial parameters, execute the clustering algorithm, and divide the user feature vectors into different groups.

[0021] Determine the user's group information FG.

[0022] According to a specific implementation of an embodiment of the present invention, determining the first push content P1 based on the user's group information FG, the user's current behavioral stage information JE, the user's behavioral feedback KR, and the environmental state HS includes:

[0023] Convert the user's grouping information FG, behavioral stage information JE, behavioral feedback KR, and environmental state HS into an input format that the large language model can accept;

[0024] Based on the user's behavior stage information JE, a preset template is selected from the predefined content template library;

[0025] Perform sentiment analysis on user behavioral feedback (KR) to identify users' sentiment tendencies and intensity.

[0026] The encoded information is input into a large language model, and initial content is generated based on the selected template and sentiment.

[0027] Fine-tune the content generated by the large language model to ensure that the first push content P1 generated meets the user's expectations and the context.

[0028] According to a specific implementation of an embodiment of the present invention, the step of optimizing the first push content P1 based on user feedback data DF and the hyperparameters θg and θh of optimization functions g and h to obtain the second push content P2 and the push strategy TS corresponding to the second push content P2 includes:

[0029] Collect comprehensive user feedback data (DF) to form a complete feedback dataset;

[0030] The hyperparameters θg and θh of the optimization functions g and h are optimized using optimization algorithms.

[0031] The first push content P1 is fine-tuned using the optimization function g and the optimized hyperparameter θg.

[0032] Based on user feedback data DF, the push strategy TS is adjusted using the optimization function h in combination with the hyperparameter θh;

[0033] The optimized second push content P2 and push strategy TS are evaluated, and the optimization parameters and content are further adjusted based on the evaluation results.

[0034] According to a specific implementation of an embodiment of the present invention, determining the push frequency Fr and content format Fc of the push content based on the second push content P2, the push strategy TS, and the user's behavioral characteristics KB on different social media platforms includes:

[0035] Analyze user behavior characteristics (KB) on different social media platforms to determine the relationships between user behavior characteristics on different platforms;

[0036] Based on user behavior characteristics (KB) and push strategy (TS), a predictive model is used to predict the optimal push frequency on different platforms.

[0037] The content format (Fc) is determined based on user behavior characteristics (KB) and platform features;

[0038] Convert the second push content P2 into a specific content format Fc;

[0039] Based on the platform's interaction mechanisms and user preferences, the content format Fc will be further adjusted, and the push frequency Fr will be adjusted accordingly.

[0040] According to a specific implementation of an embodiment of the present invention, optimizing the first push content P1 based on user feedback data DF and the hyperparameters θg and θh of optimization functions g and h includes:

[0041] Determine the optimization objectives for the optimization functions g and h, and transform the optimization objectives into quantifiable indicators;

[0042] Use machine learning to build the basic models for optimizing functions g and h;

[0043] User feedback data (DF) and corresponding evaluation metrics are used as training data, and the dataset is divided into training set, validation set, and test set.

[0044] The model is trained using the training data and the hyperparameters of the optimization functions g and h are tuned using the validation set.

[0045] The trained optimization functions g and h are evaluated using a test set, and the model structure or training parameters are adjusted based on the evaluation results.

[0046] According to a specific implementation of an embodiment of the present invention, the method further includes:

[0047] Collect user behavior data from different social media platforms, extract behavioral features from the collected data, and construct a user behavior feature set;

[0048] The extracted behavioral features are standardized, and the behavioral features from different platforms are merged to construct a unified user behavior feature KB across different social media platforms;

[0049] Based on real-time changes in user behavior, we regularly update the user behavior characteristics (KB) across different social media platforms.

[0050] Secondly, embodiments of the present invention also provide an apparatus for constructing a social media ELEMENT model, comprising:

[0051] The analysis module analyzes the acquired historical user behavior data DH to obtain the user's behavioral feature set KF;

[0052] The grouping module determines the user's grouping information FG based on behavioral characteristics KF and the user's current behavioral state ZC.

[0053] The determination module determines the first push content P1 based on the user's group information FG, the user's current behavioral stage information JE, the user's behavioral feedback KR, and the environmental state HS.

[0054] The optimization module optimizes the first push content P1 based on user feedback data DF and the hyperparameters θg and θh of the optimization functions g and h, to obtain the second push content P2 and the push strategy TS corresponding to the second push content P2.

[0055] The push module determines the push frequency Fr and content format Fc of the push content based on the second push content P2, the push strategy TS, and the user's behavioral characteristics KB on different social media platforms.

[0056] Thirdly, embodiments of the present invention also provide an electronic device, the electronic device comprising:

[0057] At least one processor; and,

[0058] The memory is communicatively connected to the at least one processor; wherein,

[0059] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in any of the first aspects or any implementations thereof.

[0060] Fourthly, embodiments of the present invention also provide a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the method described in the first aspect or any implementation thereof.

[0061] Fifthly, embodiments of the present invention also provide a computer program product, the computer program product including a computing program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the method described in the first aspect or any implementation thereof.

[0062] The solution in this embodiment of the invention includes: analyzing the acquired historical behavior data DH of the user to obtain the user's behavior feature set KF; determining the user's grouping information FG based on the behavior features KF and the user's current behavior state ZC; determining the first push content P1 based on the user's grouping information FG, the user's behavior stage information JE, the user's behavior feedback KR, and the environmental state HS; optimizing the first push content P1 based on the user feedback data DF and the hyperparameters θg and θh of the optimization functions g and h to obtain the second push content P2 and the push strategy TS corresponding to the second push content P2; and determining the push frequency Fr and content format Fc of the push content according to the second push content P2, the push strategy TS, and the user's behavior features KB on different social media platforms.

[0063] This solution has the following beneficial effects:

[0064] By deeply analyzing users' historical behavior data and extracting multi-dimensional behavioral features, we can accurately depict user profiles, achieve more precise user segmentation, and improve the accuracy of user positioning.

[0065] By leveraging user grouping information, behavioral stages, feedback, and environmental states, and combining this with a large language model to generate and optimize content, we ensure that the generated first push content and the optimized second push content are more in line with users' actual needs and interests, thereby improving user engagement and satisfaction.

[0066] Based on user feedback data and optimization functions, content and push strategies can be optimized and adjusted in real time to adapt to dynamic changes in users and improve the adaptability of content push strategies.

[0067] By considering users' behavioral characteristics across different social media platforms, the frequency and format of push notifications can be determined in a targeted manner, enabling optimized push notifications across platforms. Attached Figure Description

[0068] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0069] Figure 1 This is a schematic diagram of a method for constructing a social media ELEMENT model according to an embodiment of the present invention;

[0070] Figure 2 A schematic diagram illustrating another method for constructing a social media ELEMENT model provided in an embodiment of the present invention;

[0071] Figure 3 A schematic diagram of the structure of the social media ELEMENT model construction device provided in an embodiment of the present invention;

[0072] Figure 4 A schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0073] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0074] See Figure 1 and Figure 2 This invention provides a method for constructing a social media ELEMENT model, comprising:

[0075] S101, Analyze the acquired historical user behavior data DH to obtain the user behavior feature set KF.

[0076] DH obtains users' historical behavior data from multiple social media platforms. This data covers various user activity records on the platforms, such as likes, comments, shares, reposts, browsing time, and following lists.

[0077] The acquired data is cleaned to remove noise, such as duplicate records and abnormal data (e.g., unreasonable data with negative browsing duration or comment length exceeding a certain threshold).

[0078] Different feature extraction methods are used for different types of data. For numerical data, such as browsing time and number of likes, statistical features are calculated, including mean, median, and variance. For textual data, such as comments and post content, natural language processing techniques (such as bag-of-words model or TF-IDF) are used to extract textual features such as keywords and phrases.

[0079] The extracted features are combined and organized to form a comprehensive set of user behavior features (KF), which will enable further in-depth analysis and modeling of user behavior.

[0080] Historical behavioral data (DH) can be represented as a matrix. Where i represents the i-th user, and j represents different behavioral metrics, the correlation between different behavioral metrics can be calculated. :

[0081]

[0082] This indicates the correlation between the k-th and l-th indicators. and Let the values ​​of the k-th and l-th metrics for the i-th user be given respectively. and is the mean of the kth and lth behavioral metrics, and n is the number of users.

[0083] Through correlation The data is differentiated to obtain the user's behavioral feature set KF.

[0084] S102, Based on the behavioral features KF and the user's current behavioral state ZC, determine the user's grouping information FG.

[0085] It can collect the user's current behavioral status (ZC) in real time, which may include the user's current operation behavior, activity level indicators (such as the operation frequency in the past hour), and other information.

[0086] The user's current behavioral state ZC is merged with the previously extracted behavioral feature set KF to form a comprehensive user behavior description vector. .

[0087] The objective function for grouping users using clustering algorithms is:

[0088]

[0089] Where k is the number of clusters, It is the i-th cluster. It is the centroid of the i-th cluster. This is a user behavior description vector. Users are divided into different groups by minimizing this objective function.

[0090] The clustering results are evaluated and optimized using metrics such as the silhouette coefficient s(i) to assess the quality of the clustering. The silhouette coefficient formula is as follows:

[0091]

[0092] Where a(i) is the average distance from sample i to other samples in the same cluster, and b(i) is the average distance from sample i to all samples in the nearest different clusters. If the clustering effect is not ideal, adjust the parameters of the clustering algorithm until reasonable user grouping information FG is obtained, so that users in the same group have similar behavioral characteristics and current behavioral states.

[0093] Furthermore, a kernel function is introduced. Clustering maps the samples to a high-dimensional space, where and Given a user behavior description vector, the objective function can be expressed as:

[0094]

[0095] in It is a mapping function. It is the centroid of the i-th cluster after mapping.

[0096] S103, based on the user's group information FG, the user's behavioral stage information JE, the user's behavioral feedback KR, and the environmental state HS, determine the first push content P1.

[0097] Based on user group information FG, analyze the commonalities and differences between different user groups in order to generate personalized content for different user groups.

[0098] Determine the user's current behavioral stage information (JE). This information can be determined through time-series analysis of the user's historical behavior or based on specific business rules. For example, new users may be in the exploration stage, while old users may be in the consolidation or churn risk stage.

[0099] Collect user behavior feedback (KR), including user evaluations of previously received content, click behavior, conversion behavior, etc., and use sentiment analysis tools to analyze the user's sentiment tendency (positive, neutral, negative) and sentiment intensity. Assuming sentiment tendency is represented by E and sentiment intensity by I, the user's sentiment state can be represented as a vector. .

[0100] Combine the current environmental status HS, which may include current hot topics, market trends, seasonal information, etc., and use this information as input.

[0101] The first push notification content P1 is generated using a large language model based on the above information. When inputting the large language model, the above information is encoded and organized appropriately to guide the model in generating content that aligns with user grouping, behavioral stage, user feedback, and environmental state, ensuring the content is targeted and relevant. Assume the generation process of the large language model can be represented by a function f, whose input is a user information vector. The first push content Meanwhile, to ensure the diversity and quality of generated content, the concept of entropy is introduced. Assuming the probability distribution of generated content is p(x), the entropy H(P1) can be expressed as:

[0102]

[0103] The diversity of generated content can be adjusted by controlling the magnitude of entropy.

[0104] To better adjust the generated content based on user behavior stages, a weight matrix W is introduced to represent the impact of different behavior stages on the generated content. Assuming there are m0 dimensions to user behavior stages and n0 dimensions to generated content, the generated content can be represented as follows:

[0105]

[0106] in This is the initially generated content vector, where W is an m0×n0 weight matrix with elements... This represents the weight of the influence of the i-th behavior stage on the j-th content dimension.

[0107] S104, based on user feedback data DF and the hyperparameters θg and θh of optimization functions g and h, optimize the first push content P1 to obtain the second push content P2 and the push strategy TS corresponding to the second push content P2.

[0108] Comprehensive data collection of user feedback (DF) on the first push content (P1) is conducted, covering both direct feedback (such as ratings and comments) and indirect feedback (such as page dwell time and bounce rate).

[0109] For optimization functions g and h, determine the initial values ​​of their hyperparameters θg and θh. These hyperparameters can control the degree and direction of optimization. For example, θg may affect the extent of content modification, and θh may affect the extent of push strategy adjustment.

[0110] Hyperparameter optimization methods are used to optimize hyperparameters θg and θh based on user feedback data (DF). The optimization process involves constructing a probabilistic surrogate model; for the optimization function g, its predicted mean is... and variance It can be represented as:

[0111]

[0112]

[0113] in Let θg and θh represent the hyperparameters. It is a kernel matrix. It is the objective function value. It is the noise variance. It is a kernel function that finds the optimal combination of hyperparameters to optimize the function's performance by continuously updating these parameters.

[0114] Using the optimized hyperparameters, the first push content P1 is optimized through the optimization function g to obtain the second push content P2. This optimization can focus on aspects such as the content's structure, language expression, and information depth. Let's assume the modification of the content by the optimization function g can be expressed as:

[0115]

[0116] Where P1' is the new content generated based on user feedback, and α is the weight determined by θg.

[0117] Simultaneously, the push strategy TS is adjusted using an optimization function h, such as adjusting the push time, channel, and frequency, to improve the effectiveness of the push and user engagement. Assuming the push strategy TS includes three dimensions: time t, channel c, and frequency f, the optimization function h can be expressed as:

[0118]

[0119] in This is the initial push strategy. TS is the adjustment amount determined by θh.

[0120] To more accurately adjust the push strategy, an adjustment matrix M based on user feedback is introduced. Assuming user feedback has p dimensions and the push strategy has q dimensions, the adjustment amount ΔTS can be expressed as:

[0121]

[0122] Where KR is the user feedback vector, and MJ is a q×p matrix with elements This represents the sensitivity of the i-th push strategy dimension to the j-th user feedback dimension.

[0123] S105, based on the second push content P2, the push strategy TS, and the user's behavioral characteristics KB on different social media platforms, determine the push frequency Fr and content format Fc of the push content.

[0124] Analyze user behavior characteristics (KB) across different social media platforms, including the distribution of active time on each platform, preferred content types (e.g., text, images, videos), and engagement with different types of information (e.g., preference for long articles or short videos). Assume the distribution of user active time on platform i is as follows: Participation rate is A comprehensive indicator can be used. To represent the comprehensive behavioral characteristics of users on this platform:

[0125]

[0126] This represents a comprehensive behavioral characteristic indicator of a user on the i-th social media platform. It is a comprehensive metric calculated by combining the distribution of users' active time and engagement on that platform, and is used for subsequent quantitative evaluation and analysis of user behavior on that platform.

[0127] Let represent the time distribution function of a user's activity on the i-th social media platform. Here, t represents the time variable, describing the user's level of activity or probability of activity on the platform at different points in time.

[0128] This represents a user's engagement on the i-th social media platform at different points in time t. Engagement can be measured in various ways, such as the number of clicks, viewing time, comments, or shares on different content. It can reflect users' participation in platform content at different points in time.

[0129] Based on user behavior characteristics KB and push strategy TS, a machine learning algorithm is used to predict the most suitable push frequency Fr, and the prediction function is:

[0130]

[0131] Fr represents the push frequency. 2 is the feature vector of user behavior characteristics and push strategy. Here, b is the weight vector, 'b' is the bias term, and 'T' is the transpose. This model is trained to learn the optimal values ​​based on users' historical and feedback data. And b, to predict push frequency.

[0132] Based on user behavior characteristics (KB) and the nature of the second push content (P2), determine the most suitable content format (Fc). For example, if users have high engagement with images on a certain platform, convert the second push content (P2) into an image or a format containing more image elements; for platforms that prefer long texts, adjust the content to a more in-depth and detailed article format.

[0133] To further refine the adjustment of content format, a similarity metric function S(P2,Fc) is introduced to measure the degree of matching between the second push content P2 and the final determined content format Fc:

[0134]

[0135] The goal is to maximize this similarity so that content format Fc can better showcase the information in the second push content P2.

[0136] When determining the content format Fc and push frequency Fr, the rules and restrictions of different platforms should be considered to ensure that the content can be well displayed and disseminated on different platforms, while meeting user expectations and platform characteristics, and improving user experience.

[0137] Through steps S101 to S105, this social media ELEMENT model construction method forms a complete closed loop, from user behavior data analysis and user grouping to content generation and optimization, and finally to determining the push frequency and content format. This aims to achieve precision, personalization, and dynamism in social media marketing, thereby improving marketing effectiveness and user satisfaction. The complex mathematical formulas introduced in each step enhance the scientific rigor and accuracy of the method, facilitating better quantitative analysis and optimization of user behavior and marketing content. These formulas not only aid in theoretical analysis but also provide a deeper quantitative foundation for algorithm implementation and optimization.

[0138] According to a specific implementation of an embodiment of the present invention, the step of analyzing the acquired historical behavior data DH of the user to obtain the user's behavior feature set KF includes:

[0139] Data cleaning algorithms are used to perform preliminary screening of users' historical behavior data (DH) to remove erroneous or invalid data.

[0140] Extract different types of features from the cleaned data;

[0141] Feature selection algorithms are used to filter out features that are highly relevant to user behavior from the original feature set;

[0142] The filtered features are transformed to have a uniform scale;

[0143] Different types of features are combined to form a user's behavioral feature set KF.

[0144] According to a specific implementation of an embodiment of the present invention, determining the user's grouping information FG based on behavioral features KF and the user's current behavioral state ZC includes:

[0145] The user's current behavioral state ZC is quantified and represented.

[0146] The user feature vector is constructed by integrating the behavioral feature set KF with the quantified behavioral state vector.

[0147] Set the clustering algorithm and its initial parameters, execute the clustering algorithm, and divide the user feature vectors into different groups.

[0148] Determine the user's group information FG.

[0149] According to a specific implementation of an embodiment of the present invention, determining the first push content P1 based on the user's group information FG, the user's current behavioral stage information JE, the user's behavioral feedback KR, and the environmental state HS includes:

[0150] Convert the user's grouping information FG, behavioral stage information JE, behavioral feedback KR, and environmental state HS into an input format that the large language model can accept;

[0151] Based on the user's behavior stage information JE, a preset template is selected from the predefined content template library;

[0152] Perform sentiment analysis on user behavioral feedback (KR) to identify users' sentiment tendencies and intensity.

[0153] The encoded information is input into a large language model, and initial content is generated based on the selected template and sentiment.

[0154] Fine-tune the content generated by the large language model to ensure that the first push content P1 generated meets the user's expectations and the context.

[0155] According to a specific implementation of an embodiment of the present invention, the step of optimizing the first push content P1 based on user feedback data DF and the hyperparameters θg and θh of optimization functions g and h to obtain the second push content P2 and the push strategy TS corresponding to the second push content P2 includes:

[0156] Collect comprehensive user feedback data (DF) to form a complete feedback dataset;

[0157] The hyperparameters θg and θh of the optimization functions g and h are optimized using optimization algorithms.

[0158] The first push content P1 is fine-tuned using the optimization function g and the optimized hyperparameter θg.

[0159] Based on user feedback data DF, the push strategy TS is adjusted using the optimization function h in combination with the hyperparameter θh;

[0160] The optimized second push content P2 and push strategy TS are evaluated, and the optimization parameters and content are further adjusted based on the evaluation results.

[0161] According to a specific implementation of an embodiment of the present invention, determining the push frequency Fr and content format Fc of the push content based on the second push content P2, the push strategy TS, and the user's behavioral characteristics KB on different social media platforms includes:

[0162] Analyze user behavior characteristics (KB) on different social media platforms to determine the relationships between user behavior characteristics on different platforms;

[0163] Based on user behavior characteristics (KB) and push strategy (TS), a predictive model is used to predict the optimal push frequency on different platforms.

[0164] The content format (Fc) is determined based on user behavior characteristics (KB) and platform features;

[0165] Convert the second push content P2 into a specific content format Fc;

[0166] Based on the platform's interaction mechanisms and user preferences, the content format Fc will be further adjusted, and the push frequency Fr will be adjusted accordingly.

[0167] According to a specific implementation of an embodiment of the present invention, optimizing the first push content P1 based on user feedback data DF and the hyperparameters θg and θh of optimization functions g and h includes:

[0168] Determine the optimization objectives for the optimization functions g and h, and transform the optimization objectives into quantifiable indicators;

[0169] Use machine learning to build the basic models for optimizing functions g and h;

[0170] User feedback data (DF) and corresponding evaluation metrics are used as training data, and the dataset is divided into training set, validation set, and test set.

[0171] The model is trained using the training data and the hyperparameters of the optimization functions g and h are tuned using the validation set.

[0172] The trained optimization functions g and h are evaluated using a test set, and the model structure or training parameters are adjusted based on the evaluation results.

[0173] According to a specific implementation of an embodiment of the present invention, the method further includes:

[0174] Collect user behavior data from different social media platforms, extract behavioral features from the collected data, and construct a user behavior feature set;

[0175] The extracted behavioral features are standardized, and the behavioral features from different platforms are merged to construct a unified user behavior feature KB across different social media platforms;

[0176] Based on real-time changes in user behavior, we regularly update the user behavior characteristics (KB) across different social media platforms.

[0177] For a corresponding method embodiment, see [link to relevant documentation]. Figure 3 This invention also discloses a device 30 for constructing a social media ELEMENT model, comprising:

[0178] Analysis module 301 analyzes the acquired historical user behavior data DH to obtain the user behavior feature set KF;

[0179] Grouping module 302 determines user grouping information FG based on behavioral features KF and the user's current behavioral state ZC;

[0180] The determination module 303 determines the first push content P1 based on the user's group information FG, the user's behavior stage information JE, the user's behavior feedback KR, and the environmental status HS.

[0181] The optimization module 304 optimizes the first push content P1 based on the user feedback data DF and the hyperparameters θg and θh of the optimization functions g and h to obtain the second push content P2 and the push strategy TS corresponding to the second push content P2.

[0182] The push module 305 determines the push frequency Fr and content format Fc of the push content based on the second push content P2, the push strategy TS, and the user's behavioral characteristics KB on different social media platforms.

[0183] See Figure 4 This invention also provides an electronic device 60, which includes:

[0184] At least one processor; and,

[0185] The memory is communicatively connected to the at least one processor; wherein,

[0186] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method for constructing the social media ELEMENT model in the foregoing method embodiments.

[0187] This invention also provides a non-transitory computer-readable storage medium that stores computer instructions for causing the computer to perform the aforementioned method embodiments.

[0188] This invention also provides a computer program product, which includes a computing program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions that, when executed by a computer, cause the computer to perform the method for constructing the social media ELEMENT model in the foregoing method embodiments.

[0189] like Figure 4 As shown, electronic device 60 may include a processing device (e.g., a central processing unit, a graphics processor, etc.) 601, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 602 or a program loaded from storage device 608 into random access memory (RAM) 603. RAM 603 also stores various programs and data required for the operation of electronic device 60. Processing device 601, ROM 602, and RAM 603 are interconnected via bus 604. Input / output (I / O) interface 605 is also connected to bus 604.

[0190] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 608 including, for example, magnetic tapes, hard disks, etc.; and communication devices 609. Communication device 609 allows electronic device 60 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 An electronic device 60 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0191] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 609, or installed from a storage device 608, or installed from a ROM 602. When the computer program is executed by the processing device 601, it performs the functions defined in the methods of the embodiments of the present invention.

[0192] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for constructing a social media ELEMENT model, characterized in that, include: Analyze the acquired historical user behavior data DH to obtain the user behavior feature set KF; Based on behavioral characteristics KF and the user's current behavioral state ZC, the user's grouping information FG is determined; Based on the user's group information FG, the user's behavioral stage information JE, the user's behavioral feedback KR, and the environmental status HS, determine the first push content P1; Based on user feedback data DF and the hyperparameters θg and θh of optimization functions g and h, the first push content P1 is optimized to obtain the second push content P2 and the push strategy TS corresponding to the second push content P2. Based on the second push content P2, the push strategy TS, and the user's behavioral characteristics KB on different social media platforms, the push frequency Fr and content format Fc are determined.

2. The method according to claim 1, characterized in that, The process of analyzing the acquired historical user behavior data DH yields a user behavior feature set KF, including: Data cleaning algorithms are used to perform preliminary screening of users' historical behavior data (DH) to remove erroneous or invalid data. Extract different types of features from the cleaned data; Feature selection algorithms are used to filter out features that are highly relevant to user behavior from the original feature set; The filtered features are transformed to have a uniform scale; Different types of features are combined to form a user's behavioral feature set KF.

3. The method according to claim 2, characterized in that, The process of determining user grouping information FG based on behavioral features KF and the user's current behavioral state ZC includes: The user's current behavioral state ZC is quantified and represented. The user feature vector is constructed by integrating the behavioral feature set KF with the quantified behavioral state vector. Set the clustering algorithm and its initial parameters, execute the clustering algorithm, and divide the user feature vectors into different groups. Determine the user's group information FG.

4. The method according to claim 3, characterized in that, The first push content P1 is determined based on the user's group information FG, the user's behavioral stage information JE, the user's behavioral feedback KR, and the environmental state HS, including: Convert the user's grouping information FG, behavioral stage information JE, behavioral feedback KR, and environmental state HS into an input format that the large language model can accept; Based on the user's behavior stage information JE, a preset template is selected from the predefined content template library; Perform sentiment analysis on user behavioral feedback (KR) to identify users' sentiment tendencies and intensity. The encoded information is input into a large language model, and initial content is generated based on the selected template and sentiment. Fine-tune the content generated by the large language model to ensure that the first push content P1 generated meets the user's expectations and the context.

5. The method according to claim 4, characterized in that, The optimization of the first push content P1 based on user feedback data DF and the hyperparameters θg and θh of the optimization functions g and h, to obtain the second push content P2 and the push strategy TS corresponding to the second push content P2, includes: Collect comprehensive user feedback data (DF) to form a complete feedback dataset; The hyperparameters θg and θh of the optimization functions g and h are optimized using optimization algorithms. The first push content P1 is fine-tuned using the optimization function g and the optimized hyperparameter θg. Based on user feedback data DF, the push strategy TS is adjusted using the optimization function h in combination with the hyperparameter θh; The optimized second push content P2 and push strategy TS are evaluated, and the optimization parameters and content are further adjusted based on the evaluation results.

6. The method according to claim 5, characterized in that, The step of determining the push frequency Fr and content format Fc of the push content based on the second push content P2, the push strategy TS, and the user's behavioral characteristics KB on different social media platforms includes: Analyze user behavior characteristics (KB) on different social media platforms to determine the relationships between user behavior characteristics on different platforms; Based on user behavior characteristics (KB) and push strategy (TS), a predictive model is used to predict the optimal push frequency on different platforms. The content format (Fc) is determined based on user behavior characteristics (KB) and platform features; Convert the second push content P2 into a specific content format Fc; Based on the platform's interaction mechanisms and user preferences, the content format Fc will be further adjusted, and the push frequency Fr will be adjusted accordingly.

7. The method according to claim 6, characterized in that, The optimization of the first push content P1 based on user feedback data DF and the hyperparameters θg and θh of optimization functions g and h includes: Determine the optimization objectives for the optimization functions g and h, and transform the optimization objectives into quantifiable indicators; Use machine learning to build the basic models for optimizing functions g and h; User feedback data (DF) and corresponding evaluation metrics are used as training data, and the dataset is divided into training set, validation set, and test set. The model is trained using the training data and the hyperparameters of the optimization functions g and h are tuned using the validation set. The trained optimization functions g and h are evaluated using a test set, and the model structure or training parameters are adjusted based on the evaluation results.

8. The method according to claim 7, characterized in that, The method further includes: Collect user behavior data from different social media platforms, extract behavioral features from the collected data, and construct a user behavior feature set; The extracted behavioral features are standardized, and the behavioral features from different platforms are merged to construct a unified user behavior feature KB across different social media platforms; Based on real-time changes in user behavior, we regularly update the user behavior characteristics (KB) across different social media platforms.

9. A device for constructing a social media ELEMENT model, characterized in that, include: The analysis module analyzes the acquired historical user behavior data DH to obtain the user's behavioral feature set KF; The grouping module determines the user's grouping information FG based on behavioral characteristics KF and the user's current behavioral state ZC. The determination module determines the first push content P1 based on the user's group information FG, the user's current behavioral stage information JE, the user's behavioral feedback KR, and the environmental state HS. The optimization module optimizes the first push content P1 based on user feedback data DF and the hyperparameters θg and θh of the optimization functions g and h, to obtain the second push content P2 and the push strategy TS corresponding to the second push content P2. The push module determines the push frequency Fr and content format Fc of the push content based on the second push content P2, the push strategy TS, and the user's behavioral characteristics KB on different social media platforms.

10. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method for constructing the social media ELEMENT model as described in any one of claims 1 to 8.

Images