Intelligent copywriting design and generation system oriented to enterprises and social brands

Abstract

The invention discloses an intelligent copywriting design and generation system oriented to enterprises and social brands, and relates to the technical field of natural language processing and brand marketing intelligent decision making. According to the method, interaction data of brand assets and users are collected, a brand-user-scene three-dimensional demand perception matrix is constructed, three-dimensional core indexes are quantified, and high-potential quadrants are divided; a directed weighted conduction network is constructed based on a user cross-scene behavior, and a GraphSAGE algorithm is adopted to identify a key hub and a bridging scene; screening candidate copywriting from the three types of scenes to form a first set in combination with a dual-demand analysis result; then, a preference and conduction optimization vector is constructed based on user feedback, and a second set is generated through optimization; according to the scheme, accurate matching and dynamic iteration of full-link copywriting are realized, the pertinence of the copywriting, the link coherence and the conversion efficiency are improved, and the multi-channel collaborative marketing requirements of brands are met.

Description

Technical Field

[0001] This invention belongs to the field of natural language processing and intelligent decision-making technology for brand marketing, specifically involving an intelligent copywriting design and generation system for enterprise and social brands. Background Technology

[0002] With the deepening development of the digital economy, brand marketing has entered a new stage of multi-channel collaboration and personalized user needs. As the core carrier of communication between brands and users, copywriting directly affects brand exposure, user interaction, and conversion efficiency. From short social media posts and e-commerce product detail page copywriting to private domain traffic generation scripts, copywriting in different scenarios needs to simultaneously consider brand consistency, user preference suitability, and scenario conversion characteristics. This places higher demands on the accuracy and end-to-end adaptability of copywriting generation.

[0003] Existing copywriting generation technology has evolved from template-based output to intelligently driven processes. Early template-based generation methods relied on manually preset copywriting frameworks, generating content simply by replacing keywords. While efficient, this lacked personalization and scenario adaptability, making it difficult to meet the marketing needs of different channels. With the development of artificial intelligence technology, machine learning-based copywriting generation models have been gradually applied. By learning from historical copywriting data and user feedback, they can generate content that matches user preferences or brand tone, improving the personalization level of copywriting to some extent. In recent years, some solutions have begun to attempt to build a demand analysis system combining brand and user dimensions, optimizing the matching accuracy of copywriting by quantifying the brand's core appeals and user preferences.

[0004] However, existing technologies still have many bottlenecks and are difficult to adapt to the complex needs of full-link marketing: demand analysis is flat and lacks three-dimensional collaboration. Most solutions remain at the level of two-dimensional demand analysis between the brand and the user. Even those that introduce the scenario dimension only use static indicators for quantification and fail to build a collaborative analysis framework involving the brand, the user, and the scenario. For example, focusing only on the matching of copy with brand tone and user preferences while ignoring the differences in scenario conversion efficiency leads to low adaptability between copy and scenario characteristics and poor conversion results.

[0005] Static scenario clustering ignores cross-scenario demand transmission: Traditional solutions categorize scenarios in isolation based on channel or marketing objectives, selecting high-potential scenario clusters by the intensity of a single demand, without considering the demand transmission logic of user behavior across scenarios. In actual marketing, user needs deepen through multiple scenarios, and static clustering leads to a lack of coordination between core and related scenario copywriting, broken links, and high user churn rates.

[0006] The optimization mechanism lacks dynamism and end-to-end feedback: optimization relies heavily on feedback from single scenarios, failing to incorporate cross-scenario transmission efficiency data. User preferences and transmission paths change dynamically with the marketing environment and platform rules, making single-scenario feedback unsuitable for end-to-end needs. For example, when bridging scenario copy has high click-through rates in a single scenario but low subsequent conversion rates, it fails to identify guidance defects.

[0007] Lack of link adaptability and insufficient collaboration in bridging scenarios: Focusing on copywriting for core traffic scenarios while neglecting bridging scenarios that require connecting different clusters, copywriting for these scenarios often uses generic templates, which cannot connect the demands and styles of preceding and following scenarios, resulting in high costs for users to adapt across scenarios and low link completion rates.

[0008] In summary, existing technologies have significant shortcomings in terms of the comprehensiveness of demand analysis, the dynamism of scenario clusters, the full-link nature of optimization mechanisms, and the adaptability and collaboration of the entire process, making it difficult to meet the full-link marketing needs of brands. Therefore, developing intelligent copywriting technology based on comprehensive demand analysis and cross-scenario transmission logic has become crucial for improving marketing effectiveness. Summary of the Invention

[0009] To overcome the shortcomings and deficiencies of the existing technology, the present invention adopts the following technical solution: An intelligent copywriting design and generation system for enterprise and social brands, comprising: The data acquisition module is used to acquire brand asset data and user copywriting interaction behavior data during the brand marketing process. Brand asset data includes the brand's core tone, product / service selling point matrix, target audience profile tags, historical high-quality copywriting library, and brand value weight information. User copywriting interaction behavior data includes cross-scenario behavior trajectory data, copywriting browsing time, deep click ratio, sharing and forwarding rate, conversion completion rate, and comment keyword sentiment data. The 3D demand perception matrix construction module is used to combine brand asset data and user copywriting interaction behavior data to construct a brand-user-scenario 3D demand perception matrix, identify high-potential quadrants and form a visual demand distribution model. The cross-scenario demand transmission network construction module is used to build a directed weighted cross-scenario demand transmission network based on user cross-scenario behavior trajectory data. The network is trained by graph neural network algorithm and key hub scenarios and demand bridging scenarios are identified. The copywriting generation and recommendation module is used to extract core hub scenarios and demand bridging scenarios based on the high-potential quadrant results of the three-dimensional demand perception matrix and the key node scenarios of the cross-scenario demand transmission network. Combined with the scenarios, candidate copywriting is filtered through a preset copywriting generation and recommendation model to obtain the first candidate copywriting set. The copywriting optimization module is used to optimize the candidate copywriting based on user interaction feedback on the first candidate copywriting set, and obtain the second candidate copywriting set through a preset copywriting optimization model.

[0010] Preferably, the three-dimensional demand perception matrix construction module is specifically used for: Brand copywriting application scenarios are categorized into sub-scenarios based on three dimensions: communication channels, marketing objectives, and user lifecycle. The three-dimensional indicators of brand value intensity, user preference concentration, and scenario conversion efficiency for each segmented scenario are quantified through a pre-set demand analysis model. Each subdivided scenario is mapped to a vector point in a three-dimensional demand space. Based on the three-dimensional index values ​​and preset thresholds, high-potential quadrants are divided to construct a visualized three-dimensional demand perception matrix.

[0011] Preferably, the cross-scenario demand transmission network construction module is specifically used for: Clean and extract links from user cross-scenario behavior trajectory data, and extract valid behavior links after removing abnormal trajectories; Network initialization is completed by using subdivided scenarios as network nodes and scenario flow relationships as directed edges; Calculate the jump frequency ratio, jump conversion rate, and link contribution of each directed edge to quantify the edge weight; The GraphSAGE algorithm is used to train a graph neural network, calculate the importance score of nodes, and identify key hub scenarios and demand bridging scenarios.

[0012] Preferably, the key hub scenario is a scenario where the node importance score is higher than a preset high threshold and the connectivity is significantly high, and the required bridging scenario is a scenario where the node importance score meets the bridging scenario judgment condition and the total edge weight is prominent.

[0013] Preferably, the copywriting generation and recommendation module is specifically used for: Extract core hub scenarios and demand bridging scenarios to form a set of key node scenarios; in the core hub scenarios, analyze the triple matching degree of brand value, user preferences and scenario effectiveness through copywriting generation and recommendation models to screen core candidate copywriting; in the demand bridging scenarios, analyze the link guidance and appeal coherence of the copywriting based on cross-scenario transmission logic to screen bridging candidate copywriting. By analyzing the vector similarity between the core hub scenario and the potential expansion quadrant scenario, as well as the indirect correlation in the demand transmission network, potential candidate copywriting is screened in the potential expansion quadrant; the candidate copywriting sets of the hub scenario, the bridging scenario, and the potential scenario are merged to obtain the first candidate copywriting set.

[0014] Preferably, the copywriting optimization module is specifically used for: Based on user interaction feedback data on the first candidate copy set, we analyze changes in user copy preferences, the dynamic adjustment direction of three-dimensional demand indicators, and optimization points for cross-scenario transmission efficiency, and obtain the preference dynamic vector and the transmission optimization vector. By combining the preference dynamic vector and the transmission optimization vector, the candidate copy is optimized through the copy optimization model to obtain the second candidate copy set.

[0015] Preferably, the selection of bridging candidate copy in a demand bridging scenario specifically includes: Extract the preceding and following related scenarios of the demand bridging scenario, and analyze the core copywriting appeals and stylistic features of the preceding and following scenarios. A multi-dimensional analysis of the attributes of the target bridging copy was conducted to obtain the copy attribute feature vectors of appeal direction, guidance method, style type, and keyword distribution; Calculate the consistency score between the bridging copy and the preceding scenario copy, the style compatibility score with the following scenario copy, and combine the copy's own guidance and conversion score to obtain the comprehensive link connection score. Based on the comprehensive score of link connectivity, a predetermined number of bridging candidate texts are selected from high to low.

[0016] Preferably, the copywriting optimization model is specifically used to perform four types of optimization operations: Retain copy that achieves a total triple match score that meets a preset high match threshold, a link connection score that meets a preset connection threshold, and that receives positive user feedback; Adjust the conversion efficiency and core selling points of the copywriting for hub scenarios, and the guiding language and appeal coherence of the copywriting for bridging scenarios; Replace text with a triple match total score lower than the preset minimum match threshold or a link connection score lower than the preset minimum connection threshold; Based on the transmission optimization vector, highly guiding copy is added to the weak links of demand transmission.

[0017] Preferably, the edge weights of the cross-scenario demand transmission network are obtained by weighted summation of three indicators: jump frequency ratio, jump conversion rate, and link contribution. The link contribution is determined by calculating the contribution coefficient to the complete conversion link using a survival analysis model.

[0018] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention uses a three-dimensional demand perception matrix of brand-user-scenario to quantify three core indicators: brand value intensity, user preference concentration, and scenario conversion efficiency, forming a high-potential quadrant visualization model. This design breaks through the limitations of traditional planar demand analysis, achieving the synergistic capture of brand core appeals, user real preferences, and scenario conversion characteristics. By positioning core marketing scenarios through the core high-potential quadrant, the accuracy of demand analysis and the scientific nature of decision-making are improved.

[0019] 2. This invention constructs a directed weighted network based on user cross-scenario behavior trajectories and uses a GraphSAGE graph neural network for training to identify key hub scenarios and demand bridging scenarios. This network dynamically reflects the flow logic of demands between scenarios, strengthens link coherence, reduces cross-scenario churn rate, and improves the efficiency of the entire marketing chain.

[0020] 3. This invention constructs dynamic preference vectors and transmission optimization vectors through user interaction feedback data, driving the iteration of copywriting generation, recommendation, and optimization models. It enhances conversion efficiency for copywriting in pivotal scenarios, optimizes guiding language for copywriting in bridging scenarios, and supplements high-potential copywriting for potential scenarios, dynamically adapting to brand consistency requirements and changes in user needs, thereby improving copywriting targeting and end-to-end conversion effectiveness. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0022] Figure 1 The diagram illustrates a module diagram of an intelligent copywriting design and generation system for enterprise and social brands according to the present invention. Figure 2 The flowchart of the intelligent copywriting design and generation system for enterprise and social brands according to the present invention is shown. Figure 3 The flowchart illustrating the generation of a set of candidate texts for bridging scenarios according to the present invention is shown. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more exemplary embodiments. Numerous specific details are provided in the following description to give a full understanding of exemplary embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, steps, etc., can be employed. In other instances, well-known structures, methods, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0025] Example 1: See Figure 1 As shown, this embodiment provides an intelligent copywriting design and generation system for enterprise and social brands, including: Data acquisition module: Acquires brand asset data and user copywriting interaction data during the brand marketing process; 3D Demand Perception Matrix Construction Module: Combining brand asset data and user copywriting interaction behavior data, constructing a brand-user-scenario 3D demand perception matrix, identifying high-potential quadrants, and forming a visualized demand distribution model; The cross-scenario demand transmission network construction module: Based on user cross-scenario behavior trajectory data, a directed weighted cross-scenario demand transmission network is constructed. The network is trained using a graph neural network algorithm and key hub scenarios and demand bridging scenarios are identified. This includes: a trajectory data processing submodule: cleaning, deduplicating, and extracting links from user cross-scenario behavior trajectory data to generate a valid behavior link dataset; a network initialization submodule: initializing the cross-scenario demand transmission network using subdivided scenarios as nodes and scenario flow relationships as directed edges; and an edge weight quantification submodule: calculating the jump frequency ratio, jump conversion rate, and link contribution of each directed edge to quantify the edge weight. The GNN training and node identification submodule uses the GraphSAGE algorithm to train the graph neural network, calculates the importance score of nodes, identifies key hub scenarios and demand bridging scenarios, and outputs a set of key node scenarios and a demand transmission path analysis report.

[0026] Copywriting generation and recommendation module: Based on the high-potential quadrant results of the three-dimensional demand perception matrix and the key node scenarios of the cross-scenario demand transmission network, core hub scenarios and demand bridging scenarios are extracted. Combined with the scenarios, candidate copywriting is selected through a preset copywriting generation and recommendation model to obtain the first candidate copywriting set. Copywriting optimization module: Based on user feedback on the first set of candidate copywriting, the module optimizes the candidate copywriting using a pre-set copywriting optimization model to obtain a second set of candidate copywriting, thereby achieving intelligent copywriting design and generation for corporate and social brands.

[0027] The beneficial effects of this embodiment are as follows: it enables three-dimensional demand analysis through a three-dimensional demand perception matrix, accurately capturing the collaborative needs of brands, users, and scenarios; it identifies key nodes through a cross-scenario transmission network, strengthening the continuity of the link; and it uses a dynamic optimization mechanism to adapt to changes in preferences and transmission, improving the targeting and conversion efficiency of copywriting and adapting to multi-channel marketing needs.

[0028] Example 2: See Figure 2 As shown, this embodiment provides a method for intelligent copywriting design and generation for enterprise and social brands, used to implement the aforementioned intelligent copywriting design and generation system for enterprise and social brands. The method specifically includes: S101. Obtain brand asset data and user copywriting interaction behavior data during the brand marketing process; S102. Combining brand asset data and user copywriting interaction behavior data, construct a three-dimensional demand perception matrix of brand-user-scenario, identify high-potential quadrants, and form a visual demand distribution model; construct a cross-scenario demand transmission network based on user cross-scenario behavior trajectory, and identify key node scenarios. S103. Based on the high-potential quadrant results of the three-dimensional demand perception matrix and the key node scenarios of the cross-scenario demand transmission network, extract the core hub scenarios with a three-dimensional comprehensive score greater than a preset threshold. Combine the core hub scenarios and demand bridging scenarios to filter candidate texts through a preset text generation recommendation model to obtain the first candidate text set. S104. Based on user feedback on the first candidate copy set, optimize the candidate copy using a preset copy optimization model to obtain the second candidate copy set, thereby achieving intelligent copy design and generation for corporate and social brands.

[0029] With the diversification of brand marketing channels and the increasing demand for personalized copywriting from users, existing copywriting generation methods struggle to meet the dual requirements of brand consistency and dynamic changes in user needs. Furthermore, traditional scenario clustering relies solely on static demand intensity merging, neglecting the demand transmission effect brought about by user behavior across scenarios. This technical solution constructs a three-dimensional demand perception matrix of brand-user-scenario, breaking through the limitations of traditional planar demand analysis. Simultaneously, it constructs a directed weighted cross-scenario demand transmission network based on user cross-scenario behavior trajectories, such as "social media ad exposure → searching brand keywords → browsing e-commerce platform detail page → completing order placement." It utilizes graph neural networks to identify key hub scenarios and demand bridging scenarios, replacing simple high-demand scenario clusters. By quantifying three-dimensional core indicators through multi-source data fusion and combining the node importance scores of the demand transmission network, it accurately locates key scenarios in the core marketing chain. Then, a copywriting generation recommendation model filters candidate copywriting from core hub scenarios, demand bridging scenarios, and potential scenarios, dynamically optimizing based on user interaction feedback, ultimately achieving intelligent copywriting generation that adapts to the entire chain.

[0030] In this embodiment, firstly, brand asset data and user copywriting interaction behavior data are acquired during the brand marketing process. The brand asset data records information such as the brand's core tone, product / service selling point matrix, target audience profile tags, historical high-quality copywriting library, and brand value weight. In addition to including the copywriting browsing time, deep click percentage, sharing and forwarding rate, conversion completion rate, and comment keyword sentiment, the user copywriting interaction behavior data also focuses on adding cross-scenario behavior trajectory data, including the user's access sequence in different channels / scenarios, behavior flow path, and jump trigger points between different scenarios. The behavior flow path is, for example, Douyin short video → WeChat official account → Tmall store, and the jump trigger points between different scenarios are, for example, clicking on jump links in the copywriting or searching for brand keywords.

[0031] By integrating multi-source data, we can not only build a foundation for the construction of a three-dimensional demand perception matrix, but also provide data support for the edge weight calculation and node identification of the cross-scenario demand transmission network.

[0032] For example, the brand marketing management system can be connected to multiple data collection ports, including social media platforms, brand websites, e-commerce platform detail pages, advertising systems, private domain traffic operation tools, etc., to collect brand asset data and user copywriting interaction data in real time. Brand asset data includes, but is not limited to, brand VI specifications, core communication slogans, product function parameters, and brand value hierarchy.

[0033] The system leverages unique user identifiers, such as device IDs and member IDs, to connect cross-scenario behavioral data. For instance, after a user sees a brand's short video on Douyin and clicks a link to access the brand's official website details page, is guided to add the company's WeChat account through the website, and completes a purchase after receiving private domain promotional copy, the system records the complete behavioral trajectory, interaction data for each scenario's copy, and the jump conversion relationship. When a user shares a brand's promotional copy on social media, the platform interface synchronously provides feedback on the secondary dissemination trajectory resulting from the sharing, such as friends clicking the shared link to enter the e-commerce scenario.

[0034] For example, a data collection module can be deployed on the copywriting display page using tracking technology. This module includes a page dwell time statistics plugin, a copywriting click hotspot detection tool, a user behavior trajectory tracking component, a copywriting interaction depth rating module, and a cross-scenario jump monitoring component. The cross-scenario jump monitoring component can record the triggering behaviors of users jumping from the current copywriting scenario to other scenarios, such as clicking on product links in the copywriting, scanning QR codes, copying keyword searches, and marking the jump source scenario, target scenario, and jump conversion rate. Combined with historical copywriting campaign data, user tagging system, cross-channel attribution data, etc., recorded in the brand's backend, this data is integrated into a complete user copywriting interaction behavior dataset that includes static interaction data and dynamic link data.

[0035] Specifically, a three-dimensional demand perception matrix of brand-user-scenario is constructed, dividing the application scenarios of brand copywriting into multiple sub-scenarios according to communication channels, marketing goals, and user lifecycle. Based on multi-source data, the three-dimensional core indicators of each sub-scenarios are quantified: Brand value strength V1, user preference concentration V2, and scenario conversion efficiency V3 are used to form three-dimensional vector points (V1, V2, V3) for each sub-scenario, and mapped to a three-dimensional demand space. By setting weight coefficients for the three-dimensional indicators, the three-dimensional comprehensive score of each vector point is calculated. Based on the comprehensive score and the thresholds of each dimension, a high-potential quadrant is divided (core high-potential quadrant: V1≥70 and V2≥65 and V3≥60; potential expansion quadrant: two items meet the standard and one item is close to the threshold; undeveloped quadrant: one item meets the standard; low demand quadrant: none meet the standard), forming a visualized three-dimensional demand distribution model.

[0036] The steps for determining the weights of the three-dimensional indicators in the three-dimensional demand perception matrix are as follows: Objective weight calculation—a combination of entropy weight method and analytic hierarchy process Data input: Collect historical quantitative data in three core dimensions: V1 Data: Matching score between brand copywriting and brand VI guidelines, reuse rate of core communication phrases, frequency of occurrence of brand value keywords, etc. V2 data includes: user copy browsing time percentile, click hotspot concentration, comment sentiment score, sharing and forwarding rate, etc. V3 data includes: conversion completion rate of copywriting within a scene, retention rate when jumping to the next scene, and target behavior achievement rate, etc. Entropy weight method for calculating objective weights: The initial objective weights are obtained by calculating the information entropy of each indicator.

[0037] Analytic Hierarchy Process (AHP) Correction: Construct a hierarchical structure of overall goal - criteria layer (V1 / V2 / V3) - indicator layer, invite an expert group to conduct pairwise comparisons and scores of the criteria layer to obtain a judgment matrix, calculate the weights and pass the consistency test (CR<0.1), and obtain the objective weights after correction.

[0038] Expert experience calibration Calibration Logic: Adjust weights based on the brand's phased marketing goals—if the brand is in the new brand promotion phase, it needs to strengthen the transmission of brand value and increase the weight of V1; if it is in the mature brand cultivation phase, the weight of V1 can be decreased and the weight of V3 increased.

[0039] This embodiment uses a brand in its growth stage as the benchmark scenario, with the final weight determined by expert calibration.

[0040] Core Upgrade: Construction of a Cross-Scenario Demand Transmission Network and Identification of Key Nodes Based on user cross-scenario behavior trajectory data, a directed weighted cross-scenario demand transmission network is constructed: Network nodes: Each sub-scenario in the 3D demand perception matrix is ​​used as a network node, and the node weight is the 3D comprehensive score (0-100 points) of that scenario. Directed edges in a network: The flow of a user's behavior from scenario A to scenario B is represented by a directed edge (A→B), which represents the transmission relationship of demand from scenario A to scenario B; Edge weight calculation: Taking into account cross-scene jump frequency, jump conversion rate, and link contribution, the edge weight (0-1 point) is quantified. The higher the weight, the stronger the demand transmission intensity. The demand transmission network is trained by the GraphSAGE algorithm in the graph neural network to learn the neighbor structure and edge weight features of each node and output the node importance score. The steps for determining the weights of the three metrics for the edge weights in the cross-scenario demand transmission network are as follows: Objective weight calculation—mutual information method + random forest regression Data input: Collect cross-scenario behavior trajectory data, including the complete conversion link (including the entire path from the initial scenario to the final conversion scenario), the jump frequency of each link, the conversion result after the jump, the link completion rate, and other data.

[0041] The mutual information method calculates the correlation of indicators: the mutual information values ​​of three indicators, namely the jump frequency ratio, jump conversion rate, and link contribution, with the final conversion success rate are calculated to obtain the initial correlation coefficients: jump frequency ratio, jump conversion rate, and link contribution. After normalization, the objective weights are obtained.

[0042] Random forest regression validation: A random forest model is constructed with the final conversion success rate as the dependent variable and three indicators as independent variables. The model outputs the importance scores of the indicators and cross-validates them with the results of the mutual information method to ensure the stability of the weights and ultimately obtain objective weights.

[0043] Expert experience calibration Calibration Logic: Considering that link contribution is a core indicator for ensuring the integrity of the entire link (especially for long-link marketing scenarios, such as social media → private domain → e-commerce), but its sample size in historical data is relatively small, the expert group increased its weight while decreasing the weight of the jump frequency percentage. The final determinations were made regarding the percentage of redirect frequency, redirect conversion rate, and link contribution.

[0044] Based on node importance scores and network structure features, two types of key node scenarios are identified: Key hub scenarios: Nodes with an importance score of ≥80 and a sum of out-degree / in-degree ranking in the top 10% are traffic aggregation centers for users' cross-scenario behaviors (such as e-commerce platform detail pages and core social media promotion positions), and have strong demand radiation capabilities; Demand bridging scenario: Nodes with importance scores ≥ set thresholds and located between two or more scenario clusters, with total edge weight ranking less than set thresholds, are key link nodes connecting different scenario clusters and serve as bridges for demand transmission. This network replaces the traditional static scene clusters, accurately reflecting the dynamic flow logic of requirements between scenes and avoiding the link breakage problem caused by isolated analysis of a single scene.

[0045] Specifically, based on the high-potential quadrant results of the three-dimensional demand perception matrix and the key node identification results of the cross-scenario demand transmission network, scenarios with core high-potential quadrants and node importance scores ≥ a set threshold are extracted as core hub scenarios. Combined with demand bridging scenarios, candidate texts are selected from the core hub scenarios, demand bridging scenarios, and potential expansion quadrant scenarios through a preset text generation recommendation model to form the first candidate text set. First, analyze the copywriting characteristics in core hub scenarios that match brand value, align with user preferences, and demonstrate excellent scenario conversion efficiency, and then select core candidate copywriting that meets these characteristics. For demand bridging scenarios, we focus on selecting transitional copy that can strengthen link connection and reduce drop-off; by calculating the vector similarity between core hub scenarios and potential expansion quadrant scenarios and the indirect correlation in the demand transmission network, we select candidate copy with transmission potential from the potential expansion quadrant. The candidate copy is combined from the three types of scenarios to form the first candidate copy set; the candidate copy is selected through the core hub scenario, which can accurately hit the core traffic and conversion of brand marketing; the candidate copy is selected through the demand bridging scenario, which can strengthen the coherence of cross-scenario links and reduce user churn; the candidate copy is selected through the potential expansion scenario, which can expand the application scope of the copy and achieve synergistic coverage of core scenario, bridging scenario and potential scenario.

[0046] Preferably, based on user feedback on the copywriting in the first candidate copywriting set, the candidate copywriting is optimized using a pre-set copywriting optimization model to obtain a second candidate copywriting set. By analyzing user feedback data such as click-through rate, sharing rate, comment keywords, interaction depth score, and cross-scenario jump rate for the candidate copywriting, the changes in user copywriting preferences, the dynamic adjustment direction of the three-dimensional demand indicators, and the optimization points for cross-scenario demand transmission efficiency are identified. Based on these trends, the copywriting in the first candidate copywriting set is adjusted and optimized: the conversion efficiency of copywriting in core hub scenarios is strengthened, the link guidance of copywriting in demand bridging scenarios is optimized, and copywriting with high transmission potential in potential scenarios is supplemented to obtain the second candidate copywriting set. Driven by user interaction feedback, copywriting optimization can dynamically adapt to changes in cross-scenario marketing links and the migration of user preferences, ensuring that the copywriting not only meets the demand characteristics of individual scenarios but also satisfies the transmission logic of the entire link, thereby improving the market adaptability of the copywriting and the overall conversion effect.

[0047] This application constructs a dual demand analysis system consisting of a three-dimensional demand perception matrix and a cross-scenario demand transmission network. It quantifies three-dimensional core indicators through multi-source data fusion and utilizes graph neural networks to mine dynamic transmission relationships between scenarios, accurately identifying key node scenarios. Based on this system, it extracts core hub scenarios, demand bridging scenarios, and potential scenarios, and combines user feedback to achieve dynamic optimization of copywriting. This solution comprehensively captures the synergistic relationship between the brand's core appeal, users' real preferences, and scenario conversion characteristics, while fully considering the transmission logic and link correlation of cross-scenario demands. This significantly improves the efficiency, accuracy, and end-to-end adaptability of copywriting generation. Simultaneously, through a dynamic optimization mechanism, it adapts to the dynamic changes in brand marketing, ensuring the timeliness and competitiveness of the copywriting.

[0048] Furthermore, a three-dimensional demand perception matrix of brand-user-scenario is constructed to identify high-potential quadrants and form a visualized demand distribution model; based on user cross-scenario behavior trajectories, a cross-scenario demand transmission network is built to identify key node scenarios, including: S201. The application scenarios of brand copywriting are divided according to the three dimensions of communication channels, marketing objectives and user lifecycle, resulting in multiple sub-scenarios; S202. Combining brand asset data and user copywriting interaction behavior data, the brand value intensity, user preference concentration, and scenario conversion efficiency of each sub-scenario are quantified through a pre-set demand analysis model to obtain the three-dimensional indicator values ​​of each sub-scenario. S203. Map each subdivided scenario to a vector point in the three-dimensional demand space, divide the high-potential quadrant based on the three-dimensional index value and the preset threshold, and construct a visualized three-dimensional demand perception matrix. S204. Based on user cross-scenario behavior trajectory data, construct a directed weighted cross-scenario demand transmission network, train the network through graph neural network algorithm, and identify key hub scenarios and demand bridging scenarios.

[0049] S204 specifically includes: a. Cross-scenario behavior trajectory cleaning and link extraction: Clean the cross-scenario behavior trajectories in user copywriting interaction data, and remove abnormal trajectories (such as accidental clicks and invalid redirects with a dwell time of less than 3 seconds); based on the user's unique identifier, extract complete and valid behavior links (such as "Douyin short video scenario → WeChat official account scenario → e-commerce details page scenario → payment success page scenario"), and record the scenario flow order, dwell time, interaction behavior and conversion results of each link; b. Demand transmission network initialization: All subdivided scenarios are network nodes, and the initial weight of the nodes is set to the three-dimensional comprehensive score of the scenario; the scenario flow relationship in the effective behavior link is the directed edge, for example, the link scenario A→scenario B→scenario C corresponds to the directed edges A→B and B→C; c. Edge weight quantization calculation: For each directed edge A→B, calculate three core metrics and sum them using a weighted average to obtain the edge weight: Jump frequency percentage: The percentage of times a user jumps from scenario A to scenario B in all links containing scenario A (weight 0.4); Jump conversion rate: The percentage of users who complete the target interaction (such as click or conversion) in scenario B after jumping from scenario A to scenario B (weight 0.3). Link Contribution (CDL): The contribution coefficient of this jump to the complete conversion link is calculated using a survival analysis model (weight 0.3). The higher the contribution coefficient, the more indispensable the jump is. ; in, : The link completion probability (survival function value) of a sample set of links that jump from A to B; Sample set selection: All complete links that include A→B jumps, with the starting point being the initial scenario and the ending point being the target conversion scenario (such as successful payment or form submission); Calculation logic: Using the Kaplan-Meier method, the completion of the link is transformed into an event (completed = 1, not completed = 0). The time of transition from the initial scenario to the final scenario is used as the time variable. The survival function value of the sample set at the conversion deadline T (e.g., the user behavior window of 24 hours) is calculated, which is the probability that the link successfully completes the conversion.

[0050] : The link completion probability (survival function value) of the sample set of links that exclude jumps from A to B; Sample set selection: with The original link structure is the same, but the A→B jump is deleted (replaced with a subsequent scenario where A jumps directly to B, or other flow paths of A are retained), and the complete link is satisfied with the starting point being the initial scenario and the ending point being the target transformation scenario. Computational logic: Same The Kaplan-Meier method was used to calculate the survival function values ​​for the same transformation cutoff time T.

[0051] ε: Minimum value, used to avoid the abnormal case where the denominator is 0.

[0052] : Truncation function, ensuring that when When the jump makes no positive contribution to the completion of the link, the contribution is set to 0.

[0053] T: Conversion deadline, preset according to the characteristics of brand marketing scenarios, covering the typical window period of user cross-scenario flow, which can be adjusted according to industry.

[0054] d. Graph Neural Network Training and Node Importance Evaluation: A graph neural network model is constructed using the GraphSAGE algorithm. Node weights, edge weights, and the three-dimensional index values ​​of nodes are used as input features. The embedding vector of each node is updated by aggregating information from neighboring nodes. Based on the embedding vector, a node importance score (0-100 points) is calculated. The higher the score, the more significant the core role of the scenario in the demand transmission network. e. Key Node Scene Identification: Set a threshold for node importance scores and filter key nodes based on network structure features: Key hub scenario: The node importance score is ≥80 points, and the sum of out-degree (number of edges that jump outward) + in-degree (number of edges that receive jumps) ranks in the top 10%, while also meeting the requirement of a three-dimensional comprehensive score of ≥75 points (core high potential quadrant). Requirement bridging scenario: Node importance score ≥ 65 points, and located between two non-directly connected scenario clusters (bridging role), with the total edge weight (the sum of the weights of all related edges) ranking in the top 15%, and a three-dimensional comprehensive score ≥ 60 points (potential expansion quadrant and above).

[0055] The cross-scenario demand transmission network built through this process can intuitively show the flow path and intensity of demand in different scenarios. The identification of key node scenarios provides a dual guide for copywriting generation: "core traffic scenarios + link connection scenarios". Compared with traditional static clusters, it is more dynamic, adaptable and collaborative.

[0056] Furthermore, based on the high-potential quadrant results of the three-dimensional demand perception matrix and the key node scenarios of the cross-scenario demand transmission network, core hub scenarios are extracted. Combining these core hub scenarios and demand bridging scenarios, candidate texts are selected using a pre-defined text generation and recommendation model to obtain a first candidate text set, including: S501. Based on the results of the three-dimensional demand perception matrix and the cross-scenario demand transmission network, extract the core hub scenario and demand bridging scenario to form a set of key node scenarios. S502. In each core hub scenario, the triple matching degree of brand value, user preference and scenario effectiveness is analyzed through the preset copywriting generation and recommendation model to filter core candidate copywriting and obtain a set of candidate copywriting for hub scenarios. S503. In each demand bridging scenario, based on the cross-scenario transmission logic analysis of the copy's link guidance and appeal coherence, filter bridging candidate copy to obtain a bridging scenario candidate copy set. S504. By analyzing the vector similarity between the core hub scenario and the potential expansion quadrant scenario, as well as the indirect correlation in the demand transmission network, potential candidate copywriting is screened within the potential expansion quadrant to obtain a set of potential scenario candidate copywriting. S505. By combining the candidate copy sets for hub scenarios, bridging scenarios, and potential scenarios, the first candidate copy set is obtained.

[0057] Based on the results of the three-dimensional demand perception matrix and the cross-scenario demand transmission network, this embodiment extracts the core hub scenario and demand bridging scenario to form a set of key node scenarios; In each core hub scenario, a pre-set copywriting generation and recommendation model is used to filter core candidate copywriting, forming a hub scenario candidate copywriting set; The copywriting generation and recommendation model is as follows: Input layer: Receives the 3D demand perception matrix and key node scene feature vectors (V1 / V2 / V3 weights + node importance scores) output by the cross-scene transmission network, as well as brand asset data and user behavior features. Feature encoding layer: Brand feature encoder + User preference encoder + Scene feature encoder → Fusion encoding vector Brand Feature Encoder: Based on the BERT architecture, extracts brand tone, core selling points, and value keywords. User Preference Encoder: Analyzes user interaction data (browsing duration, click hotspots, comment sentiment) to construct user interest vectors. Scenario Feature Encoder: Encodes scenario types (communication channels - marketing objectives - user lifecycle) into scenario embeddings. Fusion Strategy: Employs an attention mechanism to achieve trimodal feature fusion and calculates weight coefficients.

[0058] Scene adaptation layer: Scene classifier → Template retriever → Generation parameter adjuster Scene Classifier: Identifies the current scene type (e.g., social media, e-commerce product detail page) and outputs scene tags. Template Retrieval Unit: Retrieves matching template structures (title / body / CTA layout) from a pre-defined copywriting template library based on scene tags. Generation Parameter Adjuster: Adjusts the generation strategy by combining the edge weights of the demand transmission network.

[0059] Generative decision layer: Based on the autoregressive generative capabilities of a large language model, and generative strategies.

[0060] Output layer: Generates initial draft copy that matches the characteristics of the scenario, including title, body text, and call to action.

[0061] In the demand bridging scenario, the focus is on selecting copy that can connect the preceding and following links and strengthen the guidance for jumps, forming a candidate copy set for the bridging scenario; the vector similarity (≥0.6) between the core hub scenario and the potential expansion quadrant scenario and the indirect correlation in the demand transmission network (transmission is achieved through 1-2 intermediate nodes) are analyzed, and potential candidate copy is selected in the potential expansion quadrant to obtain a candidate copy set for potential scenarios; the three sets are deduplicated and merged to obtain the first candidate copy set.

[0062] Specifically, in S502, the core hub scenario serves as a traffic aggregation center, and the copywriting must simultaneously meet the requirements of high brand alignment, high user preference, and high conversion efficiency, with a total score of ≥85 points for the triple match. During the selection process, the focus is on the prominence of the core selling points, the clarity of the call to action (such as "Buy Now" and "Learn More"), and the visual adaptability (such as the concise and fast characteristics of copywriting for short video scenarios and the detailed and professional characteristics of copywriting for e-commerce scenarios), to ensure that the copywriting can maximize the flow of traffic from the hub scenario and promote conversion.

[0063] In S503, the demand bridging scenario serves as a link connection node, and the copy must meet the core requirements of "consistent appeal + strong guidance". The selection criteria include: consistency with the core appeal of the preceding scenario copy (e.g., if the preceding scenario emphasizes "new product launch", the bridging scenario must continue this appeal and guide further understanding), the naturalness of the jump guidance (avoiding abrupt link insertion), and cross-scenario adaptability (e.g., when jumping from a social media scenario to an official website scenario, the copy must smoothly transition from "eye-catching" to "detailed explanation"). By calculating the link connection score of the copy (0-100 points), copy with a score ≥70 points constitutes the candidate copy set for the bridging scenario.

[0064] In S504, the connection between potential expansion quadrant scenarios and core hub scenarios is achieved through indirect links in the demand transmission network. During the screening process, the following conditions must be met simultaneously: vector similarity ≥ 0.6 and indirect correlation in the demand transmission network ≥ 0.4 (the sum of transmission strength through intermediate nodes). In potential scenarios that meet the conditions, copywriting is screened according to the standard of triple matching total score ≥ 70 points to form a set of candidate copywriting for potential scenarios, supplementing the potential needs not covered by core hub scenarios and bridging scenarios.

[0065] In S505, the three candidate copy sets are deduplicated and merged. The first candidate copy set is obtained by weighting the sorting rules as follows: "hub scenario copy (weight 0.5) + bridging scenario copy (weight 0.3) + potential scenario copy (weight 0.2)". This set includes high-conversion copy for core traffic scenarios, guiding copy for link connection, and extended copy for potential demand scenarios.

[0066] The steps for determining the scenario weights of the first set of candidate copy are as follows: Objective weight calculation—multiple linear regression + ROC curve analysis Data input: Collect copywriting conversion performance data from the past 3 months, including core metrics such as click-through rate (CTR), link completion rate, and repeat purchase guidance rate for copywriting in various scenarios. Select 100 valid copywriting samples for each scenario.

[0067] Multiple linear regression modeling: Using the total conversion contribution value as the dependent variable and the conversion contribution of the hub scenario copywriting, the conversion contribution of the bridging scenario copywriting, and the conversion contribution of the potential scenario copywriting as independent variables, a regression model is constructed, and the standardized regression coefficients of hub, bridging, and potential are output.

[0068] ROC curve validation: The area under the ROC curve (AUC) of the copywriting in the three scenarios was calculated to measure its ability to distinguish between high-conversion and low-conversion copywriting: hub scenario AUC=0.82, bridging scenario AUC=0.75, potential scenario AUC=0.68, to verify the rationality of the regression coefficients.

[0069] Expert experience calibration Calibration Logic: Combining the dual goals of short-term conversion and long-term growth in brand marketing, the expert group believes that although the current conversion contribution of potential scenario copywriting is low, it has important value for the long-term accumulation of brand users, so its weight is increased; at the same time, considering the indirect impact of the linking role of bridging scenario copywriting on core conversion, its weight is increased, while the weight of hub scenario copywriting is decreased. The weights of hub scenarios, bridging scenarios, and potential scenarios were finally determined.

[0070] Furthermore, in each demand bridging scenario, based on the cross-scenario transmission logic analysis of the copy's guiding nature and appeal coherence, bridging candidate copy is selected to obtain a bridging scenario candidate copy set. (See [link / reference]). Figure 3 As shown, it specifically includes: S5031. Extract the preceding and following related scenarios of the demand bridging scenario, and analyze the core copywriting appeals and style characteristics of the preceding and following scenarios. S5032. Conduct multi-dimensional analysis of the attributes of the target bridging copy to obtain the copy attribute feature vector, focusing on the appeal direction, guidance method, style type, and keyword distribution; S5033. Calculate the consistency score between the bridging copy and the preceding scenario copy, the style compatibility score with the following scenario copy, and combine the copy's own guidance and conversion score to obtain the comprehensive link connection score. S5034. Based on the comprehensive score of link connection from high to low, select a preset number of bridging candidate texts to obtain a set of bridging scenario candidate texts.

[0071] Specifically, in S5031, the preceding related scenarios (scenarios that jump to the bridging scenario) and the following related scenarios (scenarios that jump from the bridging scenario) of the demand bridging scenario are extracted through the cross-scenario demand transmission network. For example, the preceding scenario of the bridging scenario "brand official website landing page" may be "Douyin short video scenario", and the following scenario may be "e-commerce details page scenario". The core appeal (such as "new product exposure" and "discount traffic generation") and style characteristics (such as "lively and eye-catching" and "concise and powerful") of the preceding scenario copywriting are analyzed, as well as the core appeal (such as "product details" and "conversion and sales") and style characteristics (such as "professional and detailed" and "rational product recommendation") of the following scenario copywriting.

[0072] In S5032, the attributes of the target bridging copy are analyzed from multiple dimensions, focusing on the attributes related to link connection: appeal direction (such as "continuing the previous appeal", "guiding the subsequent appeal", "connecting the two appeals"), guidance method (such as "link jump", "keyword-guided search", "private domain traffic"), style type (such as "transitional type" and "connecting the preceding and following"), and keyword distribution (whether it contains the core keywords of the preceding and following scenarios), forming a structured copy attribute feature vector.

[0073] In S5033, the three core scores are calculated and weighted to obtain the comprehensive link connectivity score (0-100 points): Consistency score: The degree of overlap between the core appeal of the bridging copy and the preceding scenario copy, calculated through keyword matching and semantic similarity algorithms; Style Adaptability Score: The degree of fit between the bridging copy and the copy style of the subsequent scene, ensuring that the style adaptation cost for users when jumping from the bridging scene to the subsequent scene is minimized; Conversion score: The effectiveness of the bridging copy's guidance method, including click-through rate and completion rate; for example, a certain bridging copy has a consistency score of 85, a style suitability score of 82, and a conversion score of 78. The overall link connection score = 85×0.3+82×0.3+78×0.4=80.7, which meets the candidate criteria.

[0074] In S5034, the links are sorted from high to low according to the comprehensive score of the link connection. The preset number is set according to the number of related links in the bridging scenario. The top-ranked copywriting is selected to form a candidate copywriting set for the bridging scenario. The copywriting in this set can effectively connect the demands and styles of the preceding and following scenarios, reducing the user churn rate when switching between scenarios.

[0075] Furthermore, based on user feedback regarding the first set of candidate copy, the candidate copy is optimized using a pre-defined copy optimization model to obtain a second set of candidate copy, including: S801. Based on user interaction feedback data on the copywriting in the first candidate copywriting set, analyze the changes in user copywriting preferences, the dynamic adjustment direction of the three-dimensional demand indicators, and the optimization points of cross-scenario transmission efficiency to obtain the preference dynamic vector and the transmission optimization vector. S802. Combining the preference dynamic vector and the transmission optimization vector, the candidate copy is optimized through the preset copy optimization model to obtain the second candidate copy set.

[0076] The copywriting optimization model is as follows: Feedback collection layer: user behavior monitoring + conversion data analysis + copywriting interaction metrics.

[0077] User behavior monitoring: Records micro-level behaviors such as click location, dwell time, and redirect path. Conversion data analysis: Statistics on macro-level indicators such as click-through rate, conversion rate, and link completion rate. Copywriting interaction indicators: Comment keywords, sharing rate, and interaction depth score.

[0078] Performance Evaluation Layer: Construct a multi-dimensional evaluation index system and calculate optimization priority. Optimization Priority = (1 - Current Conversion Rate / Target Conversion Rate) × Weight + (1 - Copywriting Matching Degree) × Weight + Transmission Efficiency Loss × Weight.

[0079] Problem Diagnosis Layer: Copywriting Defect Detector + Scene Mismatch Analyzer + Guided Link Breakpoint Identification. Copywriting Defect Detector: Analyzes grammatical errors, logical coherence, and sentiment consistency based on NLP. Scene Mismatch Analysis: Compares copywriting features with scene requirements to identify style / structure / tone deviations. Guided Link Breakpoint Identification: Analyzes cross-scene navigation data to pinpoint key nodes in the copywriting that lead to user drop-off.

[0080] Copywriting optimization layer: Adopts a modular optimization strategy, applying different optimization operators for different problem types.

[0081] Specific optimization methods: Reinforcement learning-based copy rewriting: Reward function = Expected conversion rate improvement + Brand matching degree + Scenario adaptability - Rewriting complexity. Use the PPO algorithm to optimize the copy generation strategy and directly modify the original copy.

[0082] Effect verification layer: Deploy the optimized copy, monitor changes in key indicators, and form a closed-loop feedback.

[0083] In this embodiment, the construction of a new transmission optimization vector is added in S801: Based on indicators such as cross-scene jump rate, jump conversion rate, and link completion rate in user feedback data, the transmission efficiency optimization points between key node scenarios are analyzed, and the dimensions of the transmission optimization vector are set (such as the transmission efficiency from the hub scenario to the bridging scenario, the transmission efficiency from the bridging scenario to the potential scenario, the direction of copywriting guidance optimization, etc.). The transmission optimization vector is quantified, and the corresponding dimensions are: Douyin → official website transmission efficiency optimization, official website → e-commerce transmission efficiency optimization, and guidance text optimization requirements. The preference dynamic vector and the transmission optimization vector together constitute the optimization guide to ensure that the copywriting not only adapts to changes in user preferences but also improves cross-scene transmission efficiency.

[0084] In S802, the copywriting optimization model comprehensively considers factors such as dynamic preference vectors, transmission optimization vectors, brand tone constraints, and scenario application norms, and performs four types of optimization operations on candidate copywriting: High-match, high-transmission copywriting: Copywriting with a total score of ≥85 points for triple match, a link connection score of ≥75 points, and good user feedback will be retained directly; Adjust medium-match / transmission copy: For copy used in pivotal scenarios, enhance conversion efficiency and highlight core selling points; for copy used in bridging scenarios, optimize guiding language and the coherence of appeals to improve transmission efficiency. Replace low-match / low-transmission copy: Replace copy with a new generated copy that conforms to the optimization vector if the total triple match score is <65 or the link connection score is <60. Supplement with high-potential copywriting: Based on the optimized transmission vector, supplement with highly guiding copywriting in the weak links of demand transmission to fill the gaps in the links; the optimized second candidate copywriting set not only meets the demand characteristics of a single scenario, but also adapts to cross-scenario transmission logic.

[0085] The beneficial effects of this embodiment are as follows: the three-dimensional demand perception matrix breaks through the limitations of planar analysis and accurately captures the collaborative needs of brands, users and scenarios; the cross-scenario transmission network identifies key nodes and avoids link breakage; multi-scenario screening achieves collaborative coverage, dynamically optimizes adaptation preferences and transmission changes, and improves the accuracy of copywriting and the conversion efficiency of the entire link.

[0086] All formulas in this invention are dimensionless and calculated numerically. The preset parameters in the formulas can be set by those skilled in the art according to the actual situation.

[0087] The weighting coefficients of this invention are used to measure the degree of influence of different factors or variables on a certain outcome or decision. The weighting coefficient is defined as the numerical value assigned to each factor when comparing and evaluating multiple factors, reflecting their importance or priority. These weighting coefficients can be determined according to specific circumstances and needs, and are usually jointly formulated and confirmed by professionals or relevant stakeholders. By reasonably setting the weighting coefficients, programs or systems can be helped to make decisions or predictions more accurately.

[0088] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0089] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent copywriting design and generation system for enterprise and social brands, characterized in that, include: The data acquisition module is used to acquire brand asset data and user copywriting interaction behavior data during the brand marketing process; The three-dimensional demand perception matrix construction module is used to combine the brand asset data and user copywriting interaction behavior data to construct a brand-user-scenario three-dimensional demand perception matrix, identify high-potential quadrants and form a visual demand distribution model. The cross-scenario demand transmission network construction module is used to build a directed weighted cross-scenario demand transmission network based on user cross-scenario behavior trajectory data. The network is trained by graph neural network algorithm and key hub scenarios and demand bridging scenarios are identified. The copywriting generation and recommendation module is used to extract core hub scenarios and demand bridging scenarios based on the high-potential quadrant results of the three-dimensional demand perception matrix and the key node scenarios of the cross-scenario demand transmission network. Combined with the scenarios, candidate copywriting is filtered through a preset copywriting generation and recommendation model to obtain the first candidate copywriting set. The copywriting optimization module is used to optimize the candidate copywriting based on user interaction feedback on the first candidate copywriting set, and obtain the second candidate copywriting set through a preset copywriting optimization model.

2. The intelligent copywriting design and generation system for enterprise and social brands according to claim 1, characterized in that, The brand asset data includes the brand's core tone, product / service selling point matrix, target audience profile tags, historical high-quality copywriting library, and brand value weight information; the user copywriting interaction behavior data includes cross-scenario behavior trajectory data, copywriting browsing time, deep click percentage, sharing and forwarding rate, conversion completion rate, and comment keyword sentiment data. The three-dimensional demand perception matrix construction module is specifically used for: Brand copywriting application scenarios are categorized into sub-scenarios based on three dimensions: communication channels, marketing objectives, and user lifecycle. The three-dimensional indicators of brand value intensity, user preference concentration, and scenario conversion efficiency for each segmented scenario are quantified through a pre-set demand analysis model. Each subdivided scenario is mapped to a vector point in a three-dimensional demand space. Based on the three-dimensional index values ​​and preset thresholds, high-potential quadrants are divided to construct a visualized three-dimensional demand perception matrix.

3. The intelligent copywriting design and generation system for enterprise and social brands according to claim 1, characterized in that, The cross-scenario demand transmission network construction module is specifically used for: Clean and extract links from user cross-scenario behavior trajectory data, and extract valid behavior links after removing abnormal trajectories; Network initialization is completed by using subdivided scenarios as network nodes and scenario flow relationships as directed edges; Calculate the jump frequency ratio, jump conversion rate, and link contribution of each directed edge to quantify the edge weight; The GraphSAGE algorithm is used to train a graph neural network, calculate the importance score of nodes, and identify key hub scenarios and demand bridging scenarios.

4. The intelligent copywriting design and generation system for enterprise and social brands according to claim 3, characterized in that, The critical hub scenario is a scenario where the node importance score is higher than a preset high threshold and the connectivity is significantly high. The required bridging scenario is a scenario where the node importance score meets the bridging scenario judgment condition and the total edge weight is prominent.

5. The intelligent copywriting design and generation system for enterprise and social brands according to claim 1, characterized in that, The copywriting generation and recommendation module is specifically used for: Extract core hub scenarios and demand bridging scenarios to form a set of key node scenarios; in the core hub scenarios, analyze the triple matching degree of brand value, user preferences and scenario effectiveness through copywriting generation and recommendation models to screen core candidate copywriting; In demand bridging scenarios, the guiding nature and coherence of the copywriting are analyzed based on the cross-scenario transmission logic to select bridging candidate copywriting. By analyzing the vector similarity between the core hub scenario and the potential expansion quadrant scenario, as well as the indirect correlation in the demand transmission network, potential candidate copywriting is screened in the potential expansion quadrant; the candidate copywriting sets of the hub scenario, the bridging scenario, and the potential scenario are merged to obtain the first candidate copywriting set.

6. The intelligent copywriting design and generation system for enterprise and social brands according to claim 1, characterized in that, The copywriting optimization module is specifically used for: Based on user interaction feedback data on the first candidate copy set, we analyze changes in user copy preferences, the dynamic adjustment direction of three-dimensional demand indicators, and optimization points for cross-scenario transmission efficiency, and obtain the preference dynamic vector and the transmission optimization vector. By combining the aforementioned preference dynamic vector and the transmission optimization vector, the candidate copywriting is optimized through the copywriting optimization model to obtain a second set of candidate copywriting.

7. The intelligent copywriting design and generation system for enterprise and social brands according to claim 5, characterized in that, The specific steps for selecting bridging candidate copy in demand bridging scenarios include: Extract the preceding and following related scenarios of the demand bridging scenario, and analyze the core copywriting appeals and stylistic features of the preceding and following scenarios. A multi-dimensional analysis of the attributes of the target bridging copy was conducted to obtain the copy attribute feature vectors of appeal direction, guidance method, style type, and keyword distribution; Calculate the consistency score between the bridging copy and the preceding scenario copy, the style compatibility score with the following scenario copy, and combine the copy's own guidance and conversion score to obtain the comprehensive link connection score. Based on the comprehensive score of link connectivity, a predetermined number of bridging candidate texts are selected from high to low.

8. The intelligent copywriting design and generation system for enterprise and social brands according to claim 6, characterized in that, The copywriting optimization model is specifically used to perform four types of optimization operations: Retain copy that achieves a total triple match score that meets a preset high match threshold, a link connection score that meets a preset connection threshold, and that receives positive user feedback; Adjust the conversion efficiency and core selling points of the copywriting for hub scenarios, and the guiding language and appeal coherence of the copywriting for bridging scenarios; Replace text with a triple match total score lower than the preset minimum match threshold or a link connection score lower than the preset minimum connection threshold; Based on the transmission optimization vector, highly guiding copy is added to the weak links of demand transmission.

9. The intelligent copywriting design and generation system for enterprise and social brands according to claim 1, characterized in that, The edge weights of the cross-scenario demand transmission network are obtained by weighted summation of three indicators: jump frequency ratio, jump conversion rate, and link contribution. The link contribution is determined by calculating the contribution coefficient to the complete conversion link using a survival analysis model.

Images