A multi-platform content planning generation method, system, electronic device, storage medium and program product

Abstract

The embodiment of the application provides a multi-platform content planning generation method, system, electronic equipment, storage medium and program product, the method comprises the following steps: obtaining original material and task parameters; information deconstruction is carried out on the original material, and the deconstruction result about the verifiable fact item, information missing item and content risk item in the original material is determined; according to the deconstruction result, associated information is acquired from an internal knowledge base and / or an external information source; the associated information is integrated into evidence context used for constraining content generation; according to the verifiable fact item, the content risk item and the evidence context, a content construction strategy corresponding to the target platform information is generated; according to the content construction strategy, corresponding publishable manuscripts are generated for each target platform. The general large model is prone to fabricate facts, platform style confusion and content compliance out of control in the process of generating multi-platform publishing manuscripts.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to a multi-platform content planning and generation method, system, electronic device, storage medium, and program product. Background Technology

[0002] With the development of AI-generated content, natural language processing, retrieval-enhanced generation, knowledge base construction, and intelligent agent collaborative workflows, content production methods are gradually shifting from traditional manual planning, retrieval, and rewriting towards automation, intelligence, and multi-platform adaptability. Especially in scenarios such as brand promotion, government and enterprise communication, education enrollment, course promotion, and new media operations, users often need to process the same original material into content formats suitable for different platforms such as Xiaohongshu, WeChat official accounts, short video platforms, and Douyin.

[0003] However, due to the significant differences between different platforms in terms of user groups, content structure, title style, interaction methods, dissemination logic and conversion paths, it is difficult to simultaneously ensure content efficiency, factual accuracy, platform compatibility and dissemination effect by simply relying on manual processing or general text generation tools. Summary of the Invention

[0004] The purpose of this application is to provide a multi-platform content planning and generation method, system, electronic device, storage medium, and program product to improve the accuracy, compliance, and platform adaptability of multi-platform publishable manuscripts.

[0005] The first aspect of this application provides a multi-platform content planning and generation method, the method comprising: Obtain raw materials and task parameters, wherein the task parameters include at least one target platform information; The original material is deconstructed to determine the deconstruction results of verifiable fact items, missing information items, and content risk items in the original material; the verifiable fact items are used to characterize factual information in the original material that can be used as the basis for content generation; the missing information items are used to characterize key information in the original material that is not provided but affects the integrity of the content; and the content risk items are used to characterize prompt information that affects the compliance of the content or the security of its dissemination. Based on the deconstruction results, relevant information is obtained from internal knowledge bases and / or external information sources; The associated information is integrated into an evidentiary context for constraining content generation; Based on the verifiable fact items, the content risk items, and the evidence context, a content construction strategy corresponding to the target platform information is generated; the content construction strategy includes at least one of the following: content structure strategy, title strategy, language style strategy, interaction method strategy, and conversion path strategy. Based on the content building strategy, corresponding publishable articles are generated for each target platform. The content building strategy corresponds one-to-one with each target platform, and the content building strategies for different target platforms differ in at least one aspect of content structure, title strategy, language style, interaction method, and conversion path.

[0006] In the above implementation process, the factual boundaries and compliance restrictions of content generation are defined by deconstructing the material information, and the evidence context is constructed by combining internal historical experience and external real-time information. Then, under this strict constraint, differentiated and structured multi-platform adaptation strategies are generated, and finally, multi-platform drafts that can be directly published are produced, which solves the pain points of general large models being prone to fabrication of facts, confusion of platform styles, and loss of control over content compliance.

[0007] Furthermore, the step of obtaining relevant information from an internal knowledge base and / or external information sources based on the deconstruction results includes: Based on the verifiable fact items and / or the missing information items, historical data is retrieved from the internal knowledge base, including at least one of historical competitor analysis results, historical hot keywords, historical platform strategies, and historical task results; and / or, Based on the verifiable fact items and / or the missing information items and / or the target platform information, retrieve recent information from the external information source. The recent information includes at least one of publicly available online information, platform content summaries, competitor titles, and trending information.

[0008] In the above implementation process, precise targeted retrieval of internal knowledge base and external information sources was achieved. Specifically, verifiable fact items are used to match historical experiences and competitor dynamics with similar characteristics, missing information items are used to find complete solutions or alternative expressions, and target platform information ensures that the external search direction is consistent with the final distribution channel.

[0009] Furthermore, the deconstruction result is also obtained in the following way: Extract at least one of the following from the original material: theme, summary, core viewpoint, selling point, application scenario, case study, and keywords.

[0010] In the above implementation process, by extracting at least one of the following from the original materials: theme, summary, core viewpoint, selling point, application scenario, case study, and keywords, the unstructured original materials are transformed into a multi-dimensional, hierarchical, structured content asset library. The theme and core viewpoint provide directional anchors for subsequent knowledge retrieval and copywriting conception, the summary and selling point provide meso-level information carriers for strategy formulation and content generation, and the keywords and case studies provide micro-level retrieval units and credibility guarantees for information retrieval and evidence support. This ensures that the original materials are fully deconstructed and tagged before entering the subsequent retrieval, strategy, and generation stages, guaranteeing the semantic consistency between the generated content and the original materials.

[0011] Further, the step of generating a content construction strategy corresponding to the target platform information based on the verifiable fact item, the content risk item, and the evidence context includes: Based on the deconstruction results and the evidence context, a competitive analysis result is generated, which includes at least one of the following: title structure, content entry angle, interaction method, dissemination strategy, and risk points of similar content. Based on the verifiable fact items, the content risk items, the evidence context, and the competitor analysis results, a content building strategy corresponding to the target platform information is generated.

[0012] In the aforementioned implementation process, firstly, the title structure, content entry angle, interaction methods, and dissemination strategies identified in the competitive analysis results provide empirically based references for subsequent platform strategy formulation. This ensures that content construction strategies are not only constrained by the factual boundaries of the source material itself but also guided by the experience of similar successful cases. This shifts the focus from a source material-centric approach to a source material plus successful market experience-centric approach, resulting in generated content with stronger dissemination adaptability and competitive advantage. Secondly, the risk point extraction from the competitive analysis results provides negative warnings for strategy formulation. Which expressions have already exposed compliance issues in similar content can be proactively avoided during the generation process, thereby reducing the compliance risk of the generated content. Finally, by using the competitive analysis results as input for strategy generation, the strategy formulation process is simultaneously subject to four comprehensive constraints: factual constraints (i.e., verifiable facts), compliance constraints (i.e., content risk factors), knowledge support (i.e., evidentiary context), and experience guidance (i.e., competitive analysis results). This avoids the problems of a single perspective and homogenized strategies that easily occur when relying solely on source material facts and searched materials.

[0013] Further, the step of generating a content construction strategy corresponding to the target platform information based on the verifiable fact items, the content risk items, the evidence context, and the competitor analysis results includes: Based on the deconstruction results and the evidence context, a hot topic recommendation result is generated, which includes at least one of the following: hot topic keywords, applicable platforms, combination methods, and risk levels. Based on the verifiable fact items, the content risk items, the evidence context, the competitor analysis results, and the hot topic recommendation results, a content construction strategy corresponding to the target platform information is generated.

[0014] In the above implementation process, firstly, the generation of hot topic recommendation results is based on a comprehensive judgment of the deconstruction results and the evidence context. The recommendation results simultaneously include information from four dimensions: hot topic keywords, applicable platform labels, combination method suggestions, and risk level assessment. This allows the system to not only know what hot topics are available, but also which platform to use them on, how to use them, and whether they are safe to use. Secondly, the hot topic recommendation results and competitor analysis results serve as inputs for strategy generation. This allows the strategy generation to be guided by both historical successful experience (i.e., competitor analysis) and current dissemination trends (i.e., hot topic recommendations), avoiding the lag of strategies that rely solely on historical experience and the lack of depth that relies solely on real-time hot topics. In addition, the content risk item is a compliance warning line identified from the material itself, while the risk level is a dissemination risk warning identified from external hot topics. Both protect the compliance and security of the generated content from two directions: internal material and external environment.

[0015] Furthermore, the method also includes: The publishable manuscript is subjected to a multi-dimensional quality score to obtain the score results; the multi-dimensional quality score includes at least one of the following: information accuracy score, platform compatibility score, title attractiveness score, content completeness score, language naturalness score, conversion guidance score, material consistency score, and risk control score. When the scoring result does not reach the preset quality threshold, the publishable manuscript is modified, and the step of performing multi-dimensional quality scoring on the publishable manuscript is repeated. When the rating result reaches the preset quality threshold, the publishable manuscript is output.

[0016] In the above implementation process, adaptive control of content quality was achieved, with low scores triggering targeted modifications and high scores allowing output, ensuring that the delivered draft always meets quality standards.

[0017] A second aspect of this application provides a multi-platform content planning and generation system, the system comprising: The input module is used to acquire raw materials and task parameters, wherein the task parameters include at least one target platform information; The information deconstruction module is used to deconstruct the original material and determine the deconstruction results of verifiable fact items, missing information items, and content risk items in the original material. The verifiable fact items are used to characterize factual information in the original material that can be used as the basis for content generation. The missing information items are used to characterize key information in the original material that is not provided but affects the integrity of the content. The content risk items are used to characterize prompt information that affects the compliance of the content or the security of its dissemination. The retrieval module is used to obtain related information from an internal knowledge base and / or external information sources based on the deconstruction results; An evidence construction module is used to integrate the associated information into an evidence context for constraining content generation; The strategy generation module is used to generate a content construction strategy corresponding to the target platform information based on the verifiable fact items, the content risk items, and the evidence context; the content construction strategy includes at least one of the following: content structure strategy, title strategy, language style strategy, interaction method strategy, and conversion path strategy. The copywriting generation module is used to generate corresponding publishable drafts for each target platform according to the content construction strategy. The content construction strategy corresponds one-to-one with each target platform, and the content construction strategies for different target platforms differ in at least one aspect of content structure, title strategy, language style, interaction method, and conversion path.

[0018] In this embodiment, the factual boundaries and compliance restrictions of content generation are deconstructed by deconstructing material information, and the evidentiary context is constructed by combining internal historical experience and external real-time information. Under this strict constraint, differentiated and structured multi-platform adaptation strategies are generated, and finally, multi-platform drafts that can be directly published are produced, solving the pain points of easy fabrication of facts, confusion of platform styles, and loss of control of content compliance in general large models.

[0019] A third aspect of this application provides an electronic device, the electronic device comprising: processor; Memory used to store processor-executable instructions; Wherein, when the processor invokes the executable instructions, it implements any of the methods described in the first aspect.

[0020] A fourth aspect of this application provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of any of the methods described in the first aspect.

[0021] A fifth aspect of this application provides a computer program product, the computer program product including a computer program, which, when executed by a processor, implements any of the methods described in the first aspect. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating a multi-platform content planning and generation method provided in this application embodiment; Figure 2 An overall flowchart provided for an embodiment of this application; Figure 3 A schematic diagram of data flow provided in an embodiment of this application; Figure 4 A system module structure diagram provided for an embodiment of this application; Figure 5 This is a structural block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0024] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0025] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0026] The following defines the relevant abbreviations and key terms: AIGC: Artificial Intelligence Generated Content, refers to the technology of automatically generating text, images, audio, video, scripts, titles, tags, summaries, and other content using artificial intelligence models. In this application, it primarily refers to generating multi-platform promotional copy, short video scripts, WeChat official account articles, Xiaohongshu copy, and Douyin (TikTok) scripts based on large language models.

[0027] LLM: Large Language Model, refers to an artificial intelligence model trained on a large-scale corpus that can understand and generate natural language text. In this application, the large language model can serve as the foundational capability source for intelligent agents such as material review, competitor analysis, trend recommendation, platform strategy generation, copywriting generation, quality assessment, and copywriting modification.

[0028] Agent: An intelligent agent is a software unit capable of automatically performing analysis, decision-making, generation, or evaluation operations based on task objectives, input information, and contextual state. The material review intelligent agent, competitor analysis intelligent agent, hot topic recommendation intelligent agent, platform strategy intelligent agent, copywriting generation intelligent agent, quality assessment intelligent agent, and copywriting modification intelligent agent in this application all fall under the category of intelligent agents.

[0029] Multi-Agent: Multi-Agent Collaboration refers to a technical mechanism in which multiple intelligent agents collaborate according to a preset process or task state to jointly complete complex tasks. In this application, multi-agent collaboration is used to realize a closed-loop process from material review, knowledge base retrieval, external retrieval, competitor analysis, hot topic recommendation, platform strategy generation, copywriting generation to quality assessment and automatic modification.

[0030] RAG: Retrieval-Augmented Generation, refers to retrieving relevant information from knowledge bases, external search results, or structured materials before generating content, and then using the search results as context input to the generation model to improve the factual consistency, timeliness, and traceability of the generated content. In this application, RAG is used to combine local knowledge bases and external near-time-sensitive search data to generate multi-platform content.

[0031] Knowledge Base / KB: A knowledge base refers to a database or collection of documents used to store, manage, and retrieve data such as historical competitor analysis, trending keywords, platform strategies, task results, content patterns, observation time, and citation sources. In this application, the knowledge base is used both for information enhancement before content generation and for writing back and storing competitor analysis, trending topic recommendations, platform strategies, and task results.

[0032] Embedding: Vectorized representation refers to converting text, keywords, titles, abstracts, platform strategies, or historical task results into vector form to calculate semantic similarity and perform retrieval, clustering, or matching. In this application, vectorized representation can be used for semantic retrieval and relevant data recall in a local knowledge base.

[0033] Vector retrieval refers to the retrieval of information from a knowledge base based on the similarity between text vectors. This retrieval method is most relevant to the current material's theme, industry keywords, or target platform. In this application, vector retrieval can be used to obtain historical competitor analysis, trending keywords, platform strategies, and patterns in existing content.

[0034] Prompt: A prompt word refers to the task instructions, contextual constraints, output format requirements, and generation rules input to the large language model. In this application, prompt words can be used to constrain the output format of various agents, for example, requiring the copywriting generation agent to output a ready-to-publish draft for multiple platforms, rather than providing general suggestions.

[0035] Workflow: A workflow refers to the process of organizing multiple modules or agents to complete a task according to a certain execution order. In this application, the workflow includes steps such as material input, material review, knowledge base retrieval, external retrieval, competitor analysis, hot topic recommendation, platform strategy, copy generation, quality assessment, automatic modification, and final solution output.

[0036] CTA: Call To Action, refers to statements or designs within content that guide users to perform specific actions, such as "Contact Us Now," "Register Now," "Save for Future Use," "Leave a Comment," or "Receive Materials via Private Message." In this application, CTA serves as an important component of multi-platform copywriting generation and quality assessment.

[0037] External Search: External retrieval refers to the system calling search interfaces or public network data sources to obtain recently available publicly available information related to the subject matter, industry keywords, target platform, or trending topics. In this application, external retrieval is used to supplement missing or outdated information in the local knowledge base.

[0038] Material Review: Material review refers to the process of performing structured analysis on the original materials input by the user before copywriting is generated, extracting themes, summaries, core viewpoints, user pain points, selling points, scenarios, case studies, key phrases, keywords, available facts, missing information, and risk warnings. This application implements this step through a material review intelligent agent to reduce the risk of factual fabrication.

[0039] Available Facts: These are information extracted from original materials or credible sources that can serve as the basis for generating subsequent content, such as time, place, people, organizations, products, courses, activities, data, case studies, qualifications, honors, prices, and registration methods. This application requires that the copywriting process prioritize the generation of content based on available facts.

[0040] Missing Information: This refers to key information not provided in the original materials but affecting the accuracy or completeness of the content, such as event dates, registration portal, course price, organizer, target audience, data source, and case studies. This application uses the missing information to prevent the system from automatically fabricating information without supporting evidence.

[0041] Risk Notes: Risk warnings refer to information identified by the system that may affect the compliance, authenticity, or security of content dissemination, such as unverified rankings, exaggerated claims, absolute terms, false honors, sensitive expressions, and data from unknown sources. This application uses risk warnings to constrain subsequent copywriting generation and quality assessment.

[0042] Competitor Analysis: Competitor analysis refers to the process of analyzing similar content, competitor accounts, similar topics, or platform dissemination cases to extract headline structure, content entry angles, interaction methods, dissemination strategies, and risk points. In this application, the results of competitor analysis can serve as a reference for platform strategies and copywriting generation.

[0043] Hotspot Recommendation: This refers to the process of identifying trending keywords, platform tags, dissemination trends, and usage risk levels related to the current material's topic by combining local knowledge bases and recent external search results. In this application, hotspot recommendation is used to improve the timeliness of content selection and title strategies.

[0044] Platform Strategy: This refers to differentiated planning schemes generated based on the user habits, content structure, language style, and dissemination mechanisms of different content platforms. For example, Xiaohongshu focuses on product seeding, collection, and emotional resonance; WeChat Official Accounts focus on in-depth analysis and professional credibility; short video scripts focus on storyboarding, narration, and pacing; and Douyin focuses on strong hooks, short sentences, and interactive guidance.

[0045] Quality Evaluation: This refers to the process by which the system scores and diagnoses generated content based on preset evaluation dimensions. The evaluation dimensions in this application may include information accuracy, platform compatibility, title attractiveness, content completeness, language naturalness, conversion guidance, and material consistency.

[0046] Quality Threshold: This refers to the minimum acceptable score preset by the system. When the overall score of the generated content falls below this threshold, the system automatically triggers a copywriting modification agent to optimize it. In this application, the quality threshold is used to control the stability and publishability of the generated content.

[0047] Iterative optimization refers to the process by which a system, after generating content, identifies problems through quality assessment, and then a copywriting modification agent makes partial or overall modifications, followed by reassessment, until a preset quality threshold or the maximum number of modifications is reached.

[0048] Knowledge Write-back refers to the process by which the system writes information such as competitor analysis, trending topics, platform strategies, task results, quality scores, and optimization suggestions into its local knowledge base. This application utilizes a knowledge write-back mechanism to enable the system to continuously accumulate reusable data, providing a foundation for subsequent tasks, model optimization, and operational performance evaluation.

[0049] Hallucination refers to the generation of seemingly reasonable but actually untrue, inaccurate, or unverifiable information by a large language model in the absence of factual basis. This application mitigates the risk of hallucination through material review, extraction of available facts, annotation of missing information, risk warnings, knowledge base retrieval, and quality assessment mechanisms.

[0050] Evidence Context: This refers to the basis for content generation, comprised of original materials, available facts, knowledge base data, external search results, competitor analysis summaries, and trending recommendation results. In this application, the copywriting generation agent generates multi-platform content based on the evidence context to improve the reliability and traceability of the content.

[0051] Publishable Draft: This refers to complete content that can be directly used for platform publishing, filming, or further manual adjustments, rather than merely providing writing suggestions, methodologies, or content frameworks. The copywriting generation agent in this application outputs publishable drafts such as Xiaohongshu copywriting, WeChat official account articles, short video scripts, and Douyin (TikTok) scripts.

[0052] In related technologies, common implementation methods are: 1. AI-powered planning approach: Operations staff manually extract themes, selling points, user pain points, and communication angles from raw materials, and then write copy, headlines, scripts, and action prompts for different platforms. This method relies on human experience, has a long process, involves a lot of repetitive work, and the output quality varies depending on the personnel.

[0053] 2. Single-round generation method based on a general large language model: Users input raw materials and simple prompts into the large language model, which then generates copy or content suggestions in one go. While this method can improve generation efficiency, it lacks material review, factual constraints, external data retrieval, competitor analysis, platform strategy generation, and quality assessment, making it prone to problems such as content generalization, factual fabrication, and platform style incompatibility.

[0054] 3. Template- or rule-based content generation: The system pre-sets title templates, text structures, script formats, or platform publishing rules, and then fills in or rewrites them based on user input. This method is simple to implement and the results are relatively controllable, but it lacks flexibility and is difficult to dynamically adjust based on industry trends, competitor content, user goals, and different platform contexts.

[0055] 4. Knowledge-based or external search-assisted writing methods: Some systems retrieve relevant information from knowledge bases or search engines to assist in content generation. This method can enhance the information source to some extent, but without factual verification of materials, screening of search results, quality scoring, and automatic modification mechanisms, it is still difficult to guarantee the accuracy, completeness, and publishability of the generated content.

[0056] The relevant key technologies are: 1. Artificial Intelligence Content Generation Technology: Artificial intelligence content generation technology refers to the technology of automatically generating text, titles, scripts, summaries, tags, and other content using artificial intelligence models. The content generation involved in this application mainly includes Xiaohongshu copywriting, WeChat official account articles, short video scripts, Douyin (TikTok) scripts, and multi-platform content solutions.

[0057] 2. Natural Language Processing (NLP) Technology: NLP technology is used to perform semantic understanding, topic identification, summary generation, keyword extraction, content classification, risk identification, and text rewriting on raw materials input by users, providing a foundation for subsequent content planning and generation.

[0058] 3. Retrieval Enhancement Generation Technology: Retrieval enhancement generation technology refers to retrieving relevant information from local knowledge bases or external public information sources before generating content, and using the retrieval results as the contextual basis for the generation model, thereby improving the factual consistency, timeliness and traceability of the generated content.

[0059] 4. Knowledge Base Construction and Semantic Retrieval Technology: The knowledge base stores information such as historical competitor analysis, trending keywords, platform strategies, task results, content patterns, and observation timeframes. Through keyword retrieval, semantic retrieval, or time-range retrieval, the system can reuse historical experience and provide a reference for subsequent content generation.

[0060] 5. Multi-agent collaborative technology: Multi-agent collaborative technology refers to breaking down complex tasks into multiple software agents with different responsibilities. Different agents perform tasks such as material review, knowledge base retrieval, external retrieval, competitor analysis, hot topic recommendation, platform strategy generation, copywriting generation, quality assessment, and automatic modification, thereby forming a complete automated workflow.

[0061] 6. Content Quality Assessment and Iterative Optimization Technology: Content quality assessment technology scores generated content based on dimensions such as information accuracy, platform compatibility, title attractiveness, content completeness, language naturalness, conversion guidance, and material consistency. When the score falls below a preset threshold, the system can trigger an automatic modification process to optimize the generated content locally or overall.

[0062] The specific technical principles / algorithm schemes / interaction methods of the related technologies are: 1. Manual Multi-Platform Content Planning Solution: A common solution involves content planning done manually by operations staff. The basic process is as follows: users or businesses provide raw materials, operations staff manually read the materials, extract themes, selling points, target audience, communication angles, and action guidelines, and then write content for different platforms such as Xiaohongshu (Little Red Book), WeChat official accounts, short videos, and Douyin (TikTok) promotional text, based on their respective content styles. This solution typically involves: material collection, manual retrieval of relevant trending topics or competitor content, manual summarization of communication angles, manual rewriting of text for different platforms, manual review, and manual modification. This method does not rely on complex algorithms but primarily on the experience and judgment of operations staff. Its advantage is high controllability, but it has lower processing efficiency, involves more repetitive work, and the quality varies significantly between different personnel.

[0063] 2. Template- or Rule-Based Content Generation Solutions: Another related technology is template- or rule-based content generation. This solution typically pre-sets title templates, body structure templates, short video script templates, action guidance templates, and hashtag templates. When a user inputs raw material, the system fills the content into the preset templates based on parameters such as keywords, industry category, and platform type, generating copy for the corresponding platform. Its technical principles include: keyword matching, rule judgment, template filling, field replacement, and formatted output. For example, the system can preset a copy structure of "pain point introduction—solution—case study—action guidance," or a short video script structure of "opening hook—product selling points—target audience—registration method." This type of solution is simple to implement and has a relatively stable output format, but it usually lacks semantic understanding capabilities, dynamic retrieval capabilities, and cross-platform strategy generation capabilities, making it difficult to flexibly adjust according to different materials, trending topics, and competitor situations.

[0064] 3. Single-round content generation technology based on generalized large language models: With the development of large language models, there are also solutions that directly utilize large language models to generate content. This solution typically involves the user inputting raw materials, the target platform, and writing requirements into the model. The model then generates copy, a title, script, or content suggestions in one go based on prompts. The interaction is usually as follows: the user inputs prompts and materials, and the model directly outputs the results. For example, if the user inputs "Please generate a Xiaohongshu copy and a WeChat official account article based on the following course introduction," the model generates the corresponding text based on the context. This type of solution can improve content generation efficiency, but it lacks a closed loop for pre-generation material fact-checking, missing information identification, risk information labeling, external retrieval, competitor analysis, platform strategy generation, and quality scoring. Therefore, when the raw materials are incomplete or the prompt constraints are insufficient, the generated content may suffer from problems such as generalized expressions, factual fabrication, platform style mismatch, or inability to be directly published.

[0065] 4. Retrieval-Enhanced Generation-Based Assisted Writing Technology: Retrieval-enhanced generation retrieves information related to the current question from knowledge bases, document libraries, or external resources before text generation. The retrieval results are then provided as context to the generative model to improve the factual accuracy and knowledge coverage of the generated results. Research on RAGs typically combines pre-trained generative models with non-parametric knowledge sources. A retrieval engine recalls relevant documents from external text sets, and the generative model generates answers or text based on the recalled content. The technical process of this approach includes: document segmentation, vectorization, and indexing; keyword or vector retrieval based on user input; recalling relevant text fragments; concatenating the recalled content into prompts; and calling a large language model to generate answers or text. While this approach can reduce errors caused by models relying solely on their own parameter knowledge, it still struggles to guarantee that the final content will meet the requirements of multi-platform publishing if it lacks material review, retrieval result credibility assessment, platform differentiation strategies, quality evaluation, and automatic modification mechanisms.

[0066] 5. Task Execution Technology Solutions Based on Agents and Tool Invocation: Related technologies also include agent-based solutions that combine large language models with external tools. These solutions typically enable the model not only to generate text but also to invoke search tools, database tools, computing tools, or other interfaces as needed for the task. The ReAct method proposes alternating between reasoning and action processes, allowing the language model to reason while simultaneously accessing information from the external environment or knowledge sources, thereby completing tasks such as question answering, fact verification, and interactive decision-making. In engineering implementation, agent workflows generally include steps such as task input, model judgment, tool selection, tool invocation, result return, and continued model generation. Some frameworks also distinguish between workflow and agents: workflows usually have preset execution paths, while agents can dynamically determine tool usage and execution processes based on context. The tool invocation mechanism typically manifests as follows: the application provides callable tools to the model, the model generates a tool invocation request, the application executes the corresponding tool, returns the tool result to the model, and finally, the model continues to generate results. Such solutions can enhance the external information acquisition and task execution capabilities of large language models, but related general intelligent agent solutions are usually biased towards general task processing and may not necessarily be designed for multi-platform content planning scenarios with dedicated material review, competitor analysis, hot topic recommendation, platform strategy, copywriting generation, quality assessment and automatic modification closed loop.

[0067] 6. Assisted Optimization Technology Solution Based on Content Scoring and Manual Revision: The basic approach involves generating an initial draft, which is then evaluated by humans or models based on metrics such as readability, title appeal, language fluency, and keyword coverage, providing suggestions for revision. Some systems can regenerate the text based on user feedback or model evaluation results. The technical principles of this solution include: pre-setting evaluation dimensions, scoring the generated text, outputting revision suggestions, and regenerating or manually revising based on these suggestions. While this type of solution can improve text quality, if the scoring process is not linked to original factual materials, knowledge base evidence, platform policies, and risk warnings, it may easily lead to optimization of only language expression, failing to effectively guarantee factual consistency, platform compatibility, and final publishability.

[0068] The related technologies have the following shortcomings: 1. The lack of a pre-generation fact-checking mechanism for source materials easily leads to inconsistencies in the generated content: Single-round or template-based generation schemes typically generate text directly based on user input without conducting a structured review of the original materials before generation. This makes it impossible to effectively distinguish between usable facts, missing information, and risky information. When key information such as time, location, entity qualifications, data sources, activity entry points, and case evidence is missing from the original materials, the system may automatically complete or expand upon this information, resulting in inconsistencies between the generated content and the original materials, or even fabrication of facts.

[0069] 2. Lack of missing information identification and risk warning mechanisms makes it difficult to constrain the subsequent generation process: The system focuses on "what content is generated," but lacks constraints on "what content should not be generated." For example, existing systems often fail to label unproven honors, rankings, performance promises, authoritative endorsements, and sensitive expressions before generation, potentially leading to exaggerated claims, absolute statements, or information from unknown sources in subsequent copy.

[0070] 3. Insufficient integration of data retrieval and generation processes, resulting in weak timeliness and traceability of generated content: While some knowledge base- or search engine-based writing assistance solutions can retrieve data, they typically just append the search results to prompts without filtering the time frame, platform source, topic relevance, and credibility of the retrieved data. Especially in scenarios involving trending topics, competitor dissemination, and rapidly changing platform trends, outdated, irrelevant, or low-quality materials are easily cited, thus affecting the accuracy and timeliness of the generated content.

[0071] 4. Lack of a continuous accumulation and reuse mechanism for local knowledge base: Traditional content generation solutions are mostly task-oriented. After the task is completed, the generated results, competitor analysis, hot keywords, platform strategies, and quality assessment results are not structurally saved. When encountering similar topics or similar industry tasks in the future, the system finds it difficult to reuse historical experience, resulting in repeated retrieval, repeated analysis, and repeated generation, and failing to form a sustainable content planning capability.

[0072] 5. Insufficient multi-platform adaptability, easily leading to simplistic paraphrasing of content across different platforms: A single source material may be rewritten for multiple platforms, but there's a lack of differentiated modeling for the content structure, user habits, headline strategies, interaction methods, and conversion paths of each platform. For example, Xiaohongshu requires a sense of product recommendation, collectability, and hashtags; WeChat official accounts require complete arguments and credible expression; short video scripts need storyboards, subtitles, narration, and pacing; and Douyin copywriting needs strong hooks and interactive guidance. Without platform strategy constraints, the generated results are easily just linguistic rewriting, rather than structural and logical adaptation.

[0073] 6. Lack of competitor dissemination structure analysis hinders effective content planning: Existing content generation systems typically generate text directly, with little analysis of similar content, competitor accounts, competitor topics, headline structure, interaction methods, and dissemination angles. Therefore, generated content tends to remain at a generic level, failing to develop more targeted topic angles, headline structures, and interaction strategies based on competitor dissemination patterns.

[0074] 7. Lack of trending topic recommendations and risk level assessment, leading to inaccurate trending topic integration: Even when existing solutions can incorporate trending topics, they often rely solely on mechanical keyword matching, lacking a comprehensive assessment of the relevance of the trending topic to the material's theme, applicable platforms, usage scenarios, and dissemination risks. This can result in awkward use of trending topics, mismatch with the theme, or the introduction of sensitive topics unsuitable for public promotion.

[0075] 8. Lack of multi-agent collaboration mechanism, resulting in high coupling in complex task processing: Single-model or single-module solutions often mix material understanding, data retrieval, strategy generation, copywriting generation, and quality judgment in the same generation process, leading to unclear task boundaries. When the input materials are complex, there are many platforms, or the output requirements are high, problems such as context confusion, missing steps, and unstable output formats are likely to occur.

[0076] 9. Lack of quality assessment and automatic revision loop, resulting in insufficient stability of generated results: Generative writing solutions typically output the first draft directly or only provide manual revision suggestions, lacking an automatic assessment and iteration mechanism based on scoring thresholds. When the generated content fails to meet standards in terms of information accuracy, platform compatibility, title attractiveness, content completeness, language naturalness, conversion guidance, or material consistency, the system cannot automatically locate the problem and trigger revisions, leading to significant fluctuations in the quality of the final result.

[0077] 10. Lack of a precise modification mechanism for specific issues, leading to information drift due to complete rewriting: Some systems regenerate the entire document when users request modifications, instead of targeting specific issues such as the title, opening, action guidance, risk statements, factual errors, or platform formatting. Complete rewriting may introduce new factual biases or disrupt the original content structure that already meets requirements.

[0078] 11. Insufficient publishability of output content, often remaining at the suggestion or framework level: Existing large language model generation solutions sometimes output writing suggestions, planning ideas, or content frameworks, rather than finished drafts ready for direct publication or filming. For example, short video scripts lack time periods, visuals, subtitles, narration, and pacing; Xiaohongshu copywriting lacks titles, body text, reasons for saving, and hashtags; and WeChat official account copywriting lacks a complete article structure. Therefore, manual secondary processing is still required.

[0079] 12. The task process lacks records and traceability information, hindering subsequent optimization: Typically, only the final generated text is returned, lacking unified storage of retrieval sources, knowledge base references, competitor analysis summaries, hot topic recommendation criteria, quality scores, modification suggestions, and iteration records. This makes it difficult for users to review the generation basis and also hinders subsequent system strategy optimization, dataset construction, or model fine-tuning.

[0080] In other words, this application needs to solve the following technical problems: 1. Addressing the issue of lack of factual constraints on source material before content generation on multiple platforms: This application aims to provide a technical solution for structured review of original source material before copywriting is generated. By extracting available facts, identifying missing information, and marking risk warnings, the subsequent content generation process can be based on clear factual evidence, reducing the risk of inconsistencies between the generated content and the original source material or fabrication of facts.

[0081] 2. To address the problem of unfounded expansion of generated content when original material information is incomplete: This application aims to identify key information missing in the original material, such as time, location, subject, qualifications, data source, action entry point, and case basis, and to use this type of information as a constraint in the subsequent generation process, so as to prevent the system from automatically filling in or fabricating content when there is a lack of evidence.

[0082] 3. Addressing the problem of timely identification and control of risky information in promotional content: This application aims to identify and label risky content such as unproven honors, rankings, performance promises, authoritative endorsements, absolute expressions, and sensitive statements, so that the copywriting generation, quality assessment, and automatic modification processes can avoid or weaken related risky expressions, thereby improving the compliance and security of the generated content.

[0083] 4. Addressing the issue of insufficient recent data support for generated content: This application aims to combine local knowledge bases and external search interfaces to obtain recent data related to the subject matter, industry keywords, target platforms, and hot trends. This will enable the generated content to not only rely on the model's own knowledge but also incorporate recent hot topics, competitor content, and platform dissemination information, thereby improving the timeliness of the content.

[0084] 5. Addressing the issue of insufficient integration between search results and content generation: This application aims to summarize, screen, categorize, and structure the retrieved data, and use it as contextual basis for competitor analysis, hot topic recommendations, platform strategies, and copywriting generation, avoiding information noise and inaccurate citations caused by simply splicing search results into prompts.

[0085] 6. Addressing the issue of the inability to continuously accumulate and reuse historical task data: This application aims to build a continuously updated local knowledge base that will structure and store data such as historical competitor analysis, hot keywords, platform strategies, task results, quality scores, and optimization suggestions, enabling the system to reuse historical experience in subsequent similar tasks and reduce redundant searches and analyses.

[0086] 7. Addressing the issue of insufficient content adaptability across different platforms: This application aims to generate differentiated content strategies for different platforms such as Xiaohongshu, WeChat Official Accounts, short video scripts, and Douyin, so that the generated results can adapt to the content structure, language style, title style, interaction method, and conversion path of different platforms, rather than simply rewriting the same copy with synonyms.

[0087] 8. Addressing the issue of insufficient planning basis due to the lack of competitor communication structure analysis: This application aims to analyze similar content, peer accounts, similar topics, and platform communication cases through a competitor analysis intelligence agent, extracting title structure, content entry angle, interaction methods, communication strategies, and risk points, providing a reference for subsequent platform strategy generation and copywriting generation.

[0088] 9. Addressing the issue of inaccurate integration of trending topics or the risk of dissemination: This application aims to use a trending topic recommendation agent to comprehensively analyze local knowledge bases and external search results to identify trending keywords, applicable platforms, and usage risk levels related to the current material topic, so that trending topics can be reasonably integrated with the content topic, avoiding the forced application of trending topics or the introduction of topics that are not suitable for public dissemination.

[0089] 10. Addressing the issues of high coupling and unstable processes in complex content planning tasks: This application aims to adopt a multi-agent collaborative workflow to modularize and divide tasks such as material review, knowledge base retrieval, external retrieval, competitor analysis, hot topic recommendation, platform strategy generation, copywriting generation, quality assessment, and copywriting modification, so that complex content planning tasks can be executed stably according to a clear process.

[0090] 11. Addressing the lack of automatic evaluation and closed-loop optimization for generated content quality: This application aims to provide a closed-loop system for quality evaluation and automatic modification. Generated content is scored based on dimensions such as information accuracy, platform compatibility, title attractiveness, content completeness, language naturalness, conversion guidance, and material consistency. When the score falls below a preset threshold, an intelligent copywriting modification agent is automatically triggered for optimization.

[0091] 12. Addressing the issue of information drift caused by complete rewriting: This application aims to pinpoint specific problems based on the quality assessment results and make partial repairs to content such as titles, introductions, body structure, action guidance, risk expression, platform formatting, or factual errors, so as to avoid introducing new factual biases or destroying the original qualified content by rewriting the entire content with each modification.

[0092] 13. Addressing the issue that output results cannot be directly used for publishing or filming: This application aims to output publishable content for various platforms, including Xiaohongshu copywriting, WeChat official account articles, short video scripts, and Douyin (TikTok) scripts. Short video scripts may include time periods, visuals, subtitles, narration, and pacing; Xiaohongshu copywriting may include titles, body text, reasons for saving, and hashtags; and official account copywriting can form a complete article structure.

[0093] 14. Addressing the lack of traceable records for task processes and generation basis: This application aims to record task input, material review results, knowledge base references, external search summaries, competitor analysis results, hot topic recommendation results, platform strategies, quality scores, modification opinions, and iteration records, enabling users to review the generation basis and facilitating subsequent strategy optimization, dataset construction, and model improvement by the system.

[0094] 15. Solving the problems of low efficiency and repetitive operations in multi-platform content planning: This application aims to automate the processes that originally required manual work, such as material breakdown, data retrieval, competitor analysis, hot topic screening, platform strategy formulation, multi-platform rewriting, quality review, and secondary modification, thereby improving the processing efficiency and delivery stability of multi-platform content planning.

[0095] To address any of the aforementioned issues, embodiments of this application provide a multi-platform content planning and generation method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating a multi-platform content planning and generation method provided in an embodiment of this application.

[0096] In this embodiment, the method includes: Step S10: Obtain the original materials and task parameters, wherein the task parameters include at least one target platform information; It should be noted that the raw materials are the content materials to be processed (such as product introductions and event plans). The task parameters are the set of instructions that drive subsequent differentiated processing.

[0097] The target platform information covers different content distribution channels such as social media platforms, news portals, and short video applications.

[0098] Step S20: Deconstruct the original material to determine the deconstruction results of verifiable fact items, missing information items, and content risk items in the original material; the verifiable fact items are used to characterize factual information in the original material that can be used as the basis for content generation, the missing information items are used to characterize key information in the original material that is not provided but affects the integrity of the content, and the content risk items are used to characterize prompt information that affects the compliance of the content or the security of dissemination; It should be noted that, unlike related technologies that directly feed the material to the model, this step forcibly transforms the unstructured raw material into structured generation boundary conditions.

[0099] Verifiable facts refer to hard information extracted from the material that has clear evidence and can be verified (such as time, place, product parameters, specific names / organizations), which are used as an inviolable basis for the generation of subsequent content.

[0100] Missing information refers to key points that were not provided in the original materials but affect the credibility or completeness of the content (such as price, registration link, data source). Its significance lies in preventing the system from fabricating information without basis and guiding the generation of statements that require confirmation or are open-ended.

[0101] Content risk items refer to risks such as potential violations of advertising laws, exaggerated claims, sensitive statements, or unknown sources.

[0102] Step S30: Based on the deconstruction results, obtain relevant information from the internal knowledge base and / or external information sources; It should be noted that the search behavior in this step is limited to deconstruction results (i.e., finding evidence based on "what is missing" and "what is it"), rather than general keyword matching.

[0103] Internal knowledge base: Represents the accumulated experience and knowledge within the system (such as the structure of historical success cases and past strategies).

[0104] External information sources: These represent real-time publicly available online data (such as recent hot topics and competitors' latest developments) to compensate for the lack of timeliness in the internal database and ensure that planning keeps up with current trends.

[0105] Step S40: Integrate the associated information into an evidentiary context for constraining content generation; Specifically, the retrieved fragmented data is cleaned, summarized, and structured so that it is no longer a pile of raw data, but a package of factual arguments that can be directly used to guide subsequent writing, serving as an intermediate verification layer between the original materials and the final copy.

[0106] Step S50: Based on the verifiable fact item, the content risk item, and the evidence context, generate a content construction strategy corresponding to the target platform information; the content construction strategy includes at least one of the following: content structure strategy, title strategy, language style strategy, interaction method strategy, and conversion path strategy; It should be noted that strategy generation is constrained by factual basis and risk restrictions, and also incorporates external evidence, thus preventing the model from making unfounded arbitrary decisions.

[0107] Content building strategy: The outline developed for a specific platform is further broken down into several sub-strategies: content structure strategy (the logical framework of the main text), title strategy (such as the suspenseful style of Xiaohongshu and the numerical comparison style of Douyin), language style strategy (such as the professional and steady style of WeChat official accounts and the conversational style of Douyin), interaction method strategy (the specific wording design to encourage likes and comments), and conversion path strategy (the design of the inducement link to encourage orders and joining groups).

[0108] Step S60: Based on the content building strategy, generate corresponding publishable drafts for each target platform; the content building strategy corresponds one-to-one with each target platform, and the content building strategies for different target platforms differ in at least one aspect of content structure, title strategy, language style, interaction method, and conversion path.

[0109] It should be noted that a publishable draft is a final draft that can be directly used for publication or filming.

[0110] Publishable articles for different platforms are not simply paraphrased; they differ in at least one of the five dimensions of structure, title, language, interaction, and conversion to ensure deep adaptation to multiple platforms.

[0111] In this embodiment, the factual boundaries and compliance restrictions of content generation are deconstructed by deconstructing material information, and the evidentiary context is constructed by combining internal historical experience and external real-time information. Under this strict constraint, differentiated and structured multi-platform adaptation strategies are generated, and finally, multi-platform drafts that can be directly published are produced, solving the pain points of easy fabrication of facts, confusion of platform styles, and loss of control of content compliance in general large models.

[0112] Based on any of the above embodiments, the step of obtaining related information from an internal knowledge base and / or an external information source according to the deconstruction result includes: Based on the verifiable fact items and / or the missing information items, historical data is retrieved from the internal knowledge base, including at least one of historical competitor analysis results, historical hot keywords, historical platform strategies, and historical task results; and / or, Based on the verifiable fact items and / or the missing information items and / or the target platform information, retrieve recent information from the external information source. The recent information includes at least one of publicly available online information, platform content summaries, competitor titles, and trending information.

[0113] It should be noted that in this embodiment, complementary related information is obtained from two different sources before generating the evidence context. Either of the two retrieval paths can be used, or both can be used simultaneously.

[0114] The purpose of the retrieval based on the verifiable fact items is to find similar experiences with known facts, that is, to use the clear facts in the current material (such as theme, industry, target audience) as indexes to match successful cases with similar fact characteristics in the historical database; The purpose of searching for missing information is to find a solution to complete the missing information, that is, to identify what is missing in the current material, and then to search in the historical database for strategies or templates that have processed similar missing information.

[0115] An internal knowledge base refers to a structured data storage system that is pre-built and continuously maintained within the system. Its characteristics include: Historical: It stores data accumulated and refined during the system's past operation, rather than data captured in real time; Reusability: After being structured, the data can be directly used by subsequent tasks; Continuous updates: knowledge is continuously expanded and optimized through knowledge rewriting.

[0116] Historical competitor analysis results refer to the system's conclusive output from past tasks, analyzing the dissemination structure of content from competitors or similar platforms. For example, in a previous task related to children's education, three structural patterns of highly interactive titles on Xiaohongshu (Little Red Book) were summarized. These extracted dissemination patterns can be directly reused, eliminating the need for re-analysis each time.

[0117] Historical trending keywords refer to trending terms and their associated information (such as applicable time, applicable platform, and user feedback) identified and recorded by the system in previous tasks. This helps determine the lifecycle and actual dissemination effect of a trending topic.

[0118] Historical platform strategies refer to content building strategies that the system has generated and validated for specific platforms in previous tasks. These strategies can be directly used as reference templates or initial frameworks for generating strategies for the current task.

[0119] Historical task results refer to the final publishable drafts generated by the system in previous tasks and their corresponding quality scores. When encountering highly similar tasks, historical drafts can be directly referenced or modified.

[0120] Optionally, internal knowledge base retrieval can employ one or more of the following methods: keyword retrieval, semantic vector retrieval, time range filtering, platform filtering, and industry category filtering, to achieve recall across different dimensions.

[0121] Using target platform information as an independent retrieval criterion is because external information retrieval requires targeted crawling of specific platforms (such as searching for recent popular product recommendation notes within the Xiaohongshu platform, rather than a general search across the entire network) to obtain more accurate platform ecosystem data. This ensures that the retrieved materials are highly relevant to the theme, information gaps, and target platforms of the current task.

[0122] External information sources refer to publicly available information resources outside the system, accessed through network retrieval interfaces. Their characteristics include: Recent Timeliness: It can obtain currently or recently released public information, making up for the lack of timeliness in the internal knowledge base; Extensive coverage: It includes publicly available content from multiple platforms and dimensions, with a rich variety of information sources.

[0123] Publicly available online materials refer to various web page content, news reports, industry information, etc., obtained through search engines or public interfaces, used to obtain background information related to the theme of the material.

[0124] Platform content summaries refer to a list of recent content retrieved from a specific target platform (such as Xiaohongshu, Douyin, or WeChat Official Accounts), along with a summary of the title and key points of each piece of content. This allows for a quick understanding of recent content trends and changes in user preferences on a particular platform.

[0125] Competitor titles refer to a collection of popular titles recently published by similar content or competitor accounts. By analyzing the structure and keyword patterns of highly interactive titles, real-time references are provided for generating title strategies for the current task.

[0126] Trending information refers to popular topics, keywords, and tags that are currently widely disseminated across various platforms.

[0127] Optionally, external searches can be targeted at Xiaohongshu, WeChat official accounts, Douyin, short video platforms, and trending topics across the internet, and structured fields such as titles, abstracts, links, platform sources, keywords, and observation time can be extracted.

[0128] Specifically, based on the deconstruction results, historical data can be retrieved only from the internal knowledge base to generate related information; or, based on the deconstruction results, recently relevant information can be retrieved only from external information sources to generate related information; or, based on the deconstruction results, both types of information can be retrieved from the internal knowledge base and external information sources simultaneously to generate related information.

[0129] In some embodiments, obtaining related information from an internal knowledge base includes: converting the topics and / or keywords in the deconstruction results into vector representations; and performing semantic similarity retrieval in the internal knowledge base based on the vector representations to obtain historical data semantically related to the original material.

[0130] In this embodiment, precise targeted retrieval of internal knowledge base and external information sources is achieved. Verifiable fact items are used to match historical experiences and competitor activities with similar characteristics, missing information items are used to find complete solutions or alternative expressions, and target platform information ensures that the external search direction is consistent with the final distribution channel.

[0131] Based on any of the above embodiments, the deconstruction result is also obtained in the following manner: Extract at least one of the following from the original material: theme, summary, core viewpoint, selling point, application scenario, case study, and keywords.

[0132] It's important to note that the deconstruction process involves not only judgment—identifying facts, omissions, and risks—but also extraction—extracting structured material fragments from the source material that can be used for subsequent content generation. These two aspects together constitute a complete material deconstruction capability.

[0133] Theme: The core topic or central idea around which the original materials revolve. The theme is a high-level summary of the material content. For example, the theme of a course introduction might be enrollment for a children's dance summer program, while the theme of a product introduction might be the multifunctional design of a new smart home product. The theme is the primary indexing condition for subsequent knowledge base retrieval, external information retrieval, and competitor analysis; it determines the direction of the entire planning process. The system uses the theme as the core identifier to retrieve historical data and real-time information related to the same or similar themes from the internal knowledge base and external information sources. The quality of theme extraction directly affects the accuracy of all subsequent retrieval and analysis; if the theme extraction is flawed, all subsequent steps will proceed in the wrong direction.

[0134] Abstract: A summary is a condensed summary of the core content of the original material, typically using a single paragraph to cover the key points. An abstract differs from a concise summary of the topic and also from a complete presentation of the full text. In practice, the abstract is used internally by the system to quickly convey the core information of the material. It can serve as input context for the competitive analysis agent and as a basis for the hot topic recommendation agent to determine the relevance of the material to hot topics. Abstract extraction enables the efficient transfer of the core information of the material between agents, avoiding the need for repeated reading and parsing of long texts in subsequent stages.

[0135] Core argument: The central argument or core claim that the original material aims to convey; that is, the one sentence the material most wants to tell the audience. The core argument generally has a clear stance or assertive tone, such as "This dance course can effectively improve children's self-confidence and physical coordination" or "This product's battery life is superior to competing products in the same price range." The core argument is the source of the main theme in subsequent copywriting and also a potential direction for content entry points in platform strategies. When generating drafts for various platforms, the core argument determines the main theme around which the entire copy will be discussed.

[0136] Selling points: The most attractive differentiating features of a product, service, or content in the original materials—the core advantages that can stimulate the target audience's interest or purchase motivation. For example, "small class teaching," "taught by renowned teachers," "free trial lessons," and "beginner-friendly" are all possible selling point expressions. Selling points are the source of content focus in copywriting, determining the information to be emphasized in the main body of the copy. At the platform strategy level, the expression of selling points will be adjusted differently for different platforms. For example, Xiaohongshu emphasizes a more product-centric, product-driven approach, WeChat official accounts emphasize professional and credible selling points, and Douyin emphasizes impactful selling points. Selling point extraction makes the advantageous information in the original materials explicit, ensuring that core attractive elements are not overlooked during copywriting generation, while also providing a unified content library for differentiated expression across different platforms.

[0137] Application scenarios: The applicable context or user environment for the product, service, or content described in the original material. For example, the scenario for dance course material could be "parents looking for summer interest classes for their children," or it could be more specific, such as "children lack exercise after school" or "children with introverted personalities need to build confidence." Application scenarios provide an entry point for copywriting to create audience engagement and contextual resonance. Strategically, application scenarios determine how the copy evokes the target audience's needs, such as using a scenario-based description in the opening to resonate with users. Extracting application scenarios allows subsequent content generation to develop a narrative around the user's actual usage context, improving the copy's appeal and conversion rate.

[0138] Case studies: Specific examples, user testimonials, success stories, or practical evidence provided in the original materials. Case studies are highly narrative and specific, such as "a student's growth story from zero experience to performing on stage" or "a user's feedback on the product's effects." Case studies are a source of credibility support in copywriting generation. In platform strategies, case studies can be placed in the supporting paragraphs of the main text to corroborate the authenticity of core viewpoints or selling points, enhancing the credibility and persuasiveness of the content. Case study extraction ensures that the copy has supporting evidence when needed, preventing the system from fabricating user stories or success stories in the absence of case studies.

[0139] Keywords: Core terms with identifying and retrieval value extracted from original materials, typically nouns or noun phrases, such as "children's dance," "summer classes," "interest cultivation," and "coordination training." Keywords serve as the foundational retrieval units for subsequent knowledge base searches, external information retrieval, hot topic recommendations, and topic tag generation. Compared to topics, keywords are more specific, diverse, and have a broader coverage, enabling the description of materials from multiple dimensions and improving retrieval recall. Keyword extraction allows the system to semantically characterize materials from multiple dimensions, improving knowledge retrieval recall and serving as direct candidate materials for platform copywriting topic tags (such as Xiaohongshu topic tags).

[0140] In this embodiment, by extracting at least one of the following from the original material: theme, summary, core viewpoint, selling point, application scenario, case study, and keywords, the unstructured original material is transformed into a multi-dimensional, hierarchical, structured content asset library. The theme and core viewpoint provide directional anchors for subsequent knowledge retrieval and copywriting conception, the summary and selling point provide meso-level information carriers for strategy formulation and content generation, and the keywords and case studies provide micro-level retrieval units and credibility guarantees for information retrieval and evidence support. This ensures that the original material is fully deconstructed and tagged before entering the subsequent retrieval, strategy, and generation stages, guaranteeing the semantic consistency between the generated content and the original material.

[0141] Based on any of the above embodiments, the step of generating a content construction strategy corresponding to the target platform information according to the verifiable fact item, the content risk item, and the evidence context includes: Based on the deconstruction results and the evidence context, a competitive analysis result is generated, which includes at least one of the following: title structure, content entry angle, interaction method, dissemination strategy, and risk points of similar content. Based on the verifiable fact items, the content risk items, the evidence context, and the competitor analysis results, a content building strategy corresponding to the target platform information is generated.

[0142] It should be noted that the analysis results provide the thematic direction, factual boundaries, and compliance warning lines of the current material, while the evidentiary context provides historical similar cases and real-time competitor dynamics. The combination of these two elements ensures that competitor analysis is both directionally constrained, staying on track, and data-driven, providing a solid foundation.

[0143] The essence of competitive analysis lies in transforming scattered, raw competitor information into structured knowledge of communication patterns that can be used for planning and decision-making through summarization, comparison, and refinement. Specifically, it includes the following five analytical outputs: Title Structure for Similar Content: This section identifies common patterns and structural rules in the title creation of similar, highly viral content. For example, popular titles on Xiaohongshu (Little Red Book) often use a number + result structure, such as "Swan Neck in 3 Months," or a "Pain Point Question + Solution" structure; popular titles on Douyin (TikTok) often use a suspenseful introduction + conflict contrast structure. This title structure analysis provides a template library for generating title strategies for different platforms. The system can directly reference or combine these successful title structure patterns.

[0144] Content approach angle for similar content: From what perspective or standpoint does similar content develop its argument? For example, on the same topic of children's dance courses, some content approaches it from the perspective of children's physical health, some from the perspective of college entrance exam scores, and some from the perspective of cultivating interest. Analyzing the content approach angle provides a library of topic selection perspectives for the current task, helping the system choose the approach that best matches the characteristics of the current material and has been proven effective in the current communication environment.

[0145] Interaction methods for similar content: Design elements used in highly interactive content to guide user participation, including but not limited to comment prompts (e.g., "How old was your child when they started learning to dance? Let's chat in the comments"), interactive designs (e.g., "Next episode updated if this gets over 100 likes"), and setting emotional triggers. This analysis of interaction methods provides an interactive template for designing interactive strategies in subsequent copywriting across various platforms, helping to improve content dissemination and user engagement.

[0146] Dissemination strategies for similar content: Common arrangements in the overall dissemination path and rhythm of similar content, including but not limited to content structure patterns (such as the overall layout of pain point introduction, solution, case study, and action guidance), expression hierarchy, information density control, and platform feature adaptation patterns. Dissemination strategy analysis provides a holistic writing outline for the subsequent generation of platform strategies, equivalent to the framework knowledge obtained after dissecting successful similar content.

[0147] Risk points in similar content: Expressions or content elements that may raise compliance issues are identified from similar content, including but not limited to overly promising statements (such as "guaranteed to learn"), absolute terms (such as "best" or "first"), data citations from unknown sources, and inappropriate expressions on sensitive topics. Risk point analysis provides a negative list for subsequent strategy and copy generation. During the generation process, the system proactively avoids identified high-risk expressions to reduce content compliance risks.

[0148] This embodiment adds competitor analysis results as a planning basis, meaning that strategy generation is not only limited by internal constraints (such as factual boundaries and compliance warnings) and knowledge support (such as evidentiary context), but also guided by external experience (such as competitor communication patterns). Together, they constitute the constraint and reference system for strategy generation, forming generation conditions based on facts, with compliance as the bottom line, knowledge as support, and competitors as a reference.

[0149] In this embodiment, firstly, the title structure, content approach, interaction methods, and dissemination strategies identified in the competitive analysis provide empirically based references for subsequent platform strategy development. This ensures that content creation strategies are not only constrained by the factual boundaries of the source material itself but also guided by the experience of similar successful cases. This shifts the focus from source material-centric to a combination of source material and successful market experience, resulting in more adaptable and competitive content. Secondly, the risk points extracted from the competitive analysis provide negative warnings for strategy development. Which expressions have already exposed compliance issues in similar content can be proactively avoided during the generation process, thus reducing compliance risks. Finally, by using the competitive analysis results as input for strategy generation, the strategy development process is simultaneously constrained by four comprehensive constraints: factual constraints (verifiable facts), compliance constraints (content risks), knowledge support (evidence context), and experience guidance (competitive analysis results). This avoids the problems of a single perspective and homogenized strategies that easily occur when relying solely on source material facts and searched data.

[0150] Based on any of the above embodiments, the step of generating a content construction strategy corresponding to the target platform information according to the verifiable fact item, the content risk item, the evidence context, and the competitor analysis results includes: Based on the deconstruction results and the evidence context, a hot topic recommendation result is generated, which includes at least one of the following: hot topic keywords, applicable platforms, combination methods, and risk levels. Based on the verifiable fact items, the content risk items, the evidence context, the competitor analysis results, and the hot topic recommendation results, a content construction strategy corresponding to the target platform information is generated.

[0151] It should be noted that, based on the deconstruction results and the context of the evidence, the purpose of generating hotspot recommendation results is to identify usable hotspots from the search data and material features, and to assess their applicability and risks.

[0152] The deconstruction results provide the current material's thematic direction, core keywords, and factual boundaries, while the evidentiary context provides historical hot topic records from the internal knowledge base and real-time hot topic information from external information sources. The combination of these two allows hot topic recommendations to be based on both relevance judgments (the material's theme determines which hot topics are relevant) and timeliness judgments (the retrieved data determines which hot topics are currently available).

[0153] The essence of trending topic recommendation lies in: filtering from massive amounts of dynamically changing trending information to identify those relevant to the current theme, suitable for the target platform, and with controllable dissemination risks, and then transforming them into structured trending topic usage plans that can be directly used for strategy formulation. Specifically, this includes the following outputs: Hot keywords: Popular terms or phrases related to the current material's theme and highly relevant to the current media landscape. Hot keywords differ from the keywords of the material itself. Material keywords reflect the content's content, while hot keywords reflect what users are currently searching for and discussing. For example, a children's dance course might have keywords like "children's dance," "interest classes," and "dance training," while hot keywords might include widely searched and discussed topics such as "good places to take kids during summer vacation," "interest cultivation during the back-to-school season," and "children's physical health." Embedding hot keywords in title formulation, body text, and topic tagging can increase the content's weight in the platform's recommendation algorithm and boost its natural exposure. By linking hot keywords to the deconstruction results, it ensures that hot topic recommendations revolve around the material's theme, avoiding the problem of deviating from the theme by chasing trends for the sake of chasing trends.

[0154] Applicable Platforms: The most suitable content distribution platform for each recommended trending topic to be featured or combined with. The dissemination effect and applicable methods of the same trending topic vary on different platforms. Some trending topics are suitable for Xiaohongshu's product seeding scenario (such as "summer vacation with kids"), some are suitable for Douyin short video scenarios (such as "challenge topics"), and some are suitable for WeChat official account in-depth interpretation scenarios (such as "changes in education policies"). Applicable platforms enable the system to select and combine trending topics in a targeted manner according to the characteristics of different platforms, rather than applying the same trending topic to all platforms.

[0155] Integration Methods: How should trending topics be integrated with the content? Different trending topics require different integration methods. Some are suitable as headline hooks (e.g., "The viral trend of xxx, surprisingly…"), some are suitable as entry points for discussion in the main text (e.g., "Everyone's been discussing xxx lately, actually…"), some are suitable as hashtags (e.g., "#SummerInterestClassChoiceGuide"), and some are suitable as background context for content creation. The recommended integration methods provide trending topic embedding solutions for subsequent copywriting, making the use of trending topics no longer a rigid keyword stuffing, but an organic integration of content. Choosing the appropriate integration method based on the type of trending topic and platform characteristics can improve the naturalness of trending topic usage and user acceptance.

[0156] Risk Level: This refers to the quantitative level of content compliance or dissemination risks that may arise from using this trending topic. The criteria for determining the risk level include, but are not limited to: whether the trending topic involves sensitive social issues, whether it is experiencing aesthetic fatigue or negative public opinion after excessive dissemination, whether it has been criticized or controlled by official media, and whether the combination of the trending topic and related materials is likely to lead to misleading or exaggerated interpretations. For high-risk trending topics, the system can automatically refuse use, suggest alternative methods, or add warnings and explanations when using them.

[0157] Based on the verifiable facts, content risk items, evidence context, competitor analysis results, and trending topic recommendation results, a content construction strategy corresponding to the target platform information is generated. This forms a set of generation conditions that are based on facts, adhere to compliance, are supported by knowledge, refer to competitors, and leverage trending topics. This ensures that the final generated content is factual, does not cross red lines, is supported by experience, and can be disseminated by leveraging trends, thus improving the timeliness, competitiveness, and security in the current communication environment.

[0158] In this embodiment, firstly, the generation of hot topic recommendation results is based on a comprehensive judgment of the deconstruction results and the evidentiary context. The recommendation results simultaneously include information from four dimensions: hot topic keywords, applicable platform annotations, suggested combination methods, and risk level assessment. This allows the system to not only know what hot topics are available, but also which platform to use them on, how to use them, and whether they are safe to use. Secondly, the hot topic recommendation results and competitor analysis results serve as inputs for strategy generation. This allows the strategy generation to be guided by both historical success experience (i.e., competitor analysis) and current dissemination trends (i.e., hot topic recommendations), avoiding the lag of strategies that rely solely on historical experience and the lack of depth that relies solely on real-time hot topics. In addition, the content risk item is a compliance warning line identified from the material itself, while the risk level is a dissemination risk warning identified from external hot topics. Both protect the compliance and security of the generated content from two directions: internal material and external environment.

[0159] Based on any of the above embodiments, the method further includes: The publishable manuscript is subjected to a multi-dimensional quality score to obtain the score results; the multi-dimensional quality score includes at least one of the following: information accuracy score, platform compatibility score, title attractiveness score, content completeness score, language naturalness score, conversion guidance score, material consistency score, and risk control score. When the scoring result does not reach the preset quality threshold, the publishable manuscript is modified, and the step of performing multi-dimensional quality scoring on the publishable manuscript is repeated. When the rating result reaches the preset quality threshold, the publishable manuscript is output.

[0160] It should be noted that the information accuracy score assesses whether the factual information in the generated content is consistent with verifiable facts and whether there is any unfounded fabrication or alteration. For example, if the source material states that the course is for children aged 6-12, but the generated text states that it is for children aged 3-15, this score will be low.

[0161] Platform Compatibility Score: This assesses whether the generated content aligns with the target platform's content style, structural guidelines, and user expectations. Different platforms have different content ecosystems. Xiaohongshu requires a sense of product recommendation, hashtags, and collectability; WeChat Official Accounts require a complete argumentative structure, professional credibility, and guiding intro; Douyin requires a strong hook in the first three seconds, concise sentences, and interactive comment guidance. This score verifies whether the copywriting generation process adhered to the platform's strategy regarding content structure, headline strategy, language style, interaction methods, and conversion paths.

[0162] Title Attractiveness Score: This score assesses whether the title is sufficiently attractive and evokes clicks. Evaluation criteria include whether the title creates suspense or conflict, highlights the core selling point or user pain point, and uses a proven title structure (such as a title structure pattern extracted from competitor analysis). This score verifies whether the copywriting process effectively utilizes title structure patterns from competitor analysis and trending keywords from trending recommendations.

[0163] Content Completeness Score: This assesses whether the generated content covers the core information in the source material and whether it forms a complete narrative structure. For example, a qualified WeChat official account article should include a title, introduction, body sections, case studies, and concluding remarks; a qualified short video script should include time periods, visual descriptions, subtitles, and spoken text. This score checks for any missing information. This score indirectly verifies whether the text generation process is appropriately constrained by missing information. For marked missing information, open-ended expressions are allowed, and for available facts already provided in the source material, coverage should be ensured.

[0164] Language naturalness score: assesses whether the language of the generated content is fluent, natural, and conforms to Chinese expression habits, and whether there are obvious machine translation traces, grammatical errors, inappropriate word choice, or awkward expression.

[0165] Conversion Guidance Scoring: This score assesses the clarity and effectiveness of action guidance design within the generated content. For example, for enrollment-related content, the design quality of conversion guidance such as "Schedule a free trial lesson," "Contact customer service immediately," and "Leave a comment to get materials" significantly impacts the content's commercial conversion rate. This score evaluates the appropriateness of the action guidance's placement, the attractiveness of its wording, and the clarity of the conversion path.

[0166] Material Consistency Score: This assesses whether the generated content maintains overall consistency with the core viewpoints, brand tone, and expressive stance of the original material. Unlike the information accuracy score, which verifies each fact individually, this score judges the consistency of the generated content from an overall semantic and expressive perspective, ensuring that the generated copy does not distort or deviate from the original meaning of the material.

[0167] Risk Control Score: This score assesses whether the generated content contains any of the prohibited expressions highlighted in the risk warnings, such as unverified claims of honor, absolute terms (e.g., first, best), citation of false data, or improper handling of sensitive topics. This score verifies each generated content against the risk items to check whether compliance risks have been mitigated.

[0168] A preset quality threshold refers to a pre-set minimum passing score standard. It can be a numerical threshold for a comprehensive score (such as setting 85 points as the passing score in a 100-point system) or a combination of conditions for various scoring dimensions (such as achieving the comprehensive score without any individual dimension falling below a certain minimum score). The preset quality threshold can be dynamically set or manually adjusted based on factors such as task type, risk level, and platform characteristics.

[0169] In some embodiments, the modification includes: targeted correction of at least one of the title, opening, body structure, platform format, action guidelines, risk statements, and factual errors of the publishable manuscript based on the issue location information and / or modification suggestions generated by the quality assessment steps.

[0170] Specifically, targeted adjustments and optimizations are made to content that does not meet the standards. Modifications may include restructuring the beginning of the title, reorganizing the body structure, adjusting platform formatting, optimizing the wording of action guidance, deleting or weakening risk statements, and correcting factual errors. The specific methods and extent of modifications depend on the specific issues identified in the quality scoring. After modification, the system does not directly output the revised content but instead re-enters the quality assessment stage to re-evaluate it. When the score reaches or exceeds the preset quality threshold, it indicates that the generated content has met the publishable standard. At this point, iterative iteration stops, and the final publishable draft is output.

[0171] In some embodiments, it can simultaneously monitor whether the current number of modifications has reached the preset maximum number of iterations. If the maximum number of iterations has been reached but the target has not been met, the current optimal version is output and a manual review is requested to avoid an infinite loop due to the score failing to meet the target for an extended period.

[0172] Optionally, the multi-platform content planning and generation method further includes: writing back at least one of the competitor analysis results, the hot topic recommendation results, the content construction strategy, and the publishable draft to the internal knowledge base for knowledge retrieval in subsequent tasks.

[0173] The original material is deconstructed to determine the deconstruction result, including: deconstructing the original material through a material review agent to extract the verifiable fact items, the missing information items, and the content risk items.

[0174] Generating competitor analysis results and trending topic recommendation results includes: generating competitor analysis results using a competitor analysis agent based on the deconstruction results and the evidence context, wherein the competitor analysis results include at least one of the following: title structure, content entry angle, interaction method, dissemination strategy, and risk points of similar content; and generating trending topic recommendation results using a trending topic recommendation agent based on the deconstruction results and the evidence context, wherein the trending topic recommendation results include at least one of the following: trending keywords, applicable platforms, combination methods, and risk levels.

[0175] Generating a content construction strategy corresponding to the target platform information includes: generating the content construction strategy through a platform strategy agent based on the verifiable fact items, the content risk items, the evidence context, the competitor analysis results, and the hot topic recommendation results.

[0176] The multi-dimensional quality scoring of the publishable draft is performed by a quality assessment agent; the modification of the publishable draft is performed by a copywriting modification agent.

[0177] In the specific implementation, see Figures 2-3 , Figure 2 An overall flowchart is provided for an embodiment of this application. Figure 3 This is a data flow diagram provided for an embodiment of this application. The original data generation / acquisition method of the multi-platform content planning and generation method is as follows: Raw data can be generated or obtained through front-end pages, application programming interfaces (APIs), file upload interfaces, or third-party system interfaces.

[0178] Raw data includes at least the following categories: 1. User-input raw material data: including but not limited to brand introduction, course introduction, product introduction, event plan, enrollment brochure, promotional materials, press releases, interview transcripts, case materials, image and text descriptions, short video shooting requirements, etc.

[0179] 2. User-input task parameter data: including but not limited to industry keywords, target audience, target platform, content purpose, action guidance, tone and style, output length, model selection parameters, whether to incorporate hot topics, and whether to conduct competitor analysis.

[0180] 3. Local knowledge base data: including historical competitor analysis results, hot keywords, platform strategies, historical task results, content patterns, quality scores, modification records, observation time, and citation sources.

[0181] 4. External search data: This includes recent online materials, platform content summaries, titles, links, keywords, publication time, platform source, hot trends, competitor content, etc., obtained through publicly available external search interfaces.

[0182] For example, a user inputs original materials about enrolling children's dance classes, specifying the target audience as parents, target platforms as Xiaohongshu, WeChat official accounts, and Douyin, a friendly and trustworthy tone, and the action guide as scheduling a trial class. The system encapsulates this information into a content planning task object and initiates the subsequent processing flow. Optionally, the storage structure / data structure of the original data is as follows: The system encapsulates user input data into task objects, which may include the following fields: task identifier, original material text, uploaded file information, industry keywords, target audience, target platform, action guidance, tone and style, model selection parameters, task creation time, and user supplementary requirements.

[0183] After the material review agent processes the data, it generates structured content asset objects: theme, summary, core viewpoint, user pain points, selling points, application scenarios, case studies, key quotes, keywords, available facts, missing information, risk warnings, and whether to continue generating.

[0184] Knowledge base retrieval and external retrieval results can be uniformly encapsulated into evidence context objects: local knowledge base retrieval results, external retrieval results, competitor analysis results, hot topic recommendation results, platform strategy results, evidence source, and observation time.

[0185] The generated copy can be packaged into content objects for multiple platforms: Xiaohongshu copy, WeChat official account articles, short video scripts, Douyin (TikTok) scripts, simplified versions, traditional versions, and mind map data.

[0186] The quality assessment results can be packaged into scoring objects: information accuracy score, platform compatibility score, title attractiveness score, content completeness score, language naturalness score, conversion guidance score, material consistency score, comprehensive score, modification suggestions, and risk warning.

[0187] The data structures described above are interdependent: the task object serves as the initial input, generating the material review results; the material review results serve as factual constraints for knowledge base retrieval, external retrieval, and copywriting generation; the retrieval results and analysis results constitute the evidentiary context; the evidentiary context and platform strategies jointly guide the copywriting generation across multiple platforms; and the quality assessment results determine whether automatic modification is triggered.

[0188] Regarding the algorithm model logic / rules: Step 1: Receive user input and generate a task object: Execution subject: Input and Material Management Module.

[0189] Detailed actions: Receive information such as original materials, industry keywords, target audience, target platform, action guidance, tone style, and model selection parameters input by the user; if the user uploads Word, PDF, image, or other files, perform text parsing, format cleaning, and field extraction on the uploaded files.

[0190] Objects: The raw materials and task parameters input by the user.

[0191] Result: Generates a TaskObject with a uniform format, including fields such as task number, original material, target platform, industry keywords, target audience, tone and style, action guidance, and creation time.

[0192] Relationship between data and steps: The original materials and task parameters are the data source for all subsequent steps. This step unifies and encapsulates the scattered input data, making it easier for the intelligent agent to read and process it later.

[0193] The relationship between data and the actual processing object: For example, when a user inputs enrollment information for a children's dance course and sets the target platform as Xiaohongshu, WeChat official account, and Douyin, the system will organize information such as course introduction, target audience, course features, and registration methods into a task object.

[0194] Relationship between steps: This step is the starting step of the entire process, and its output task object serves as the input for the material review agent in step 2.

[0195] Step 2: The intelligent agent for material review extracts structured content assets: Execution entity: Material review intelligent agent module.

[0196] Detailed actions: Read the original materials from the task object, perform semantic understanding and structured analysis on the materials, and extract the theme, summary, core viewpoints, user pain points, selling points, application scenarios, cases, key sentences, keywords and available facts; at the same time, identify missing information and risk warnings.

[0197] Object: The TaskObject generated in step 1, especially the fields such as original materials, industry keywords, target audience, and action guidance.

[0198] Results: Generate MaterialReviewResult, which includes the topic, summary, core viewpoints, user pain points, selling points, scenarios, case studies, key quotes, keywords, available facts, missing information, risk warnings, and a decision on whether to continue generating materials.

[0199] The relationship between data and steps: The original materials in the task object are transformed into structured content assets, and subsequent knowledge base retrieval, external retrieval, platform strategy generation and copywriting generation are all based on these structured content assets.

[0200] Relationship between data and actual processing objects: For example, if the material states that this course is for children aged 6-12 and mainly trains rhythm, coordination and stage performance, the system will extract it as usable facts; if the material does not provide course prices or registration links, the system will mark it as missing information; if the material contains statements such as guaranteeing that a child will become a dance star in one month, the system will mark it as a risk warning.

[0201] Relationship between steps: This step follows the task object of step 1, and its output serves as the factual basis for knowledge base retrieval in step 3, external retrieval in step 4, and text generation in step 8.

[0202] Step 3: Search the local knowledge base based on the material review results: Execution entity: Knowledge base module.

[0203] Detailed actions: Based on the theme, keywords, target platform, industry category, and time range in the material review results, conduct a search in the local knowledge base; the search methods include one or more of the following: keyword search, semantic vector search, platform filtering, category filtering, and time filtering.

[0204] Objects: MaterialReviewResult generated in step 2, and target platform and industry keywords from step 1.

[0205] Results: Generate localKnowledgeResults, which includes information such as historical competitor analysis, hot keywords, platform strategies, historical task results, quality scores, modification records, and observation time.

[0206] The relationship between data and steps: Keywords and themes in the material review results are used as search conditions, historical data in the knowledge base are used as the recall objects, and the search results provide a reference for subsequent competitor analysis, hot topic recommendations and platform strategy generation.

[0207] The relationship between data and the actual objects being processed: For example, if the current topic is enrollment for children's dance courses, the system can retrieve historical tasks and strategy records related to children's dance, enrollment for interest classes, Xiaohongshu (Little Red Book) product recommendations, and Douyin (TikTok) enrollment short videos from the knowledge base.

[0208] Relationship between steps: This step follows the structured content assets from step 2, and its output goes into the competitive analysis agent in step 5 and the hot topic recommendation agent in step 6.

[0209] Step 4: Call the external search interface to obtain the most up-to-date data: Execution entity: External retrieval module.

[0210] Detailed actions: Construct external search requests based on the theme, keywords, target platform, and time range of the materials, and call public network search interfaces to obtain recent data; the system can search for Xiaohongshu, WeChat official accounts, Douyin, short video platforms, and trending topics across the entire network, and extract titles, abstracts, links, platform sources, keywords, and observation time.

[0211] Targets: The topics, keywords, and target platforms generated in step 2, as well as the industry keywords from step 1.

[0212] Results: Generate externalSearchResults, which include recently published information, platform content summaries, competitor titles, trending keywords, platform sources, and links.

[0213] The relationship between data and steps: The results of material review determine the topics and keywords for external retrieval, and the external retrieval results are used to supplement missing or outdated information in the local knowledge base.

[0214] The relationship between data and the actual objects being processed: For example, the system searches for recent popular content related to summer children's dance enrollment, and obtains the title structure, user focus, hot tags, and competitor expression methods on relevant platforms.

[0215] Relationship between steps: This step is executed in parallel or sequentially with step 3. Its output, along with the local knowledge base retrieval results, enters step 5 (competitive analysis agent) and step 6 (hot topic recommendation agent).

[0216] Step 5: Competitive analysis of intelligent agents to extract propagation structure patterns: Execution entity: Competitive analysis intelligent agent module.

[0217] Detailed steps: Analyze the local knowledge base search results from step 3 and the external search results from step 4 to extract the title structure, content entry angle, interaction methods, dissemination strategies, common expressions, and risk points of competitor content.

[0218] Objects: localKnowledgeResults and externalSearchResults.

[0219] Results: CompetitorInsights analysis results were generated, including highly interactive title structures, content perspectives, platform expression patterns, interactive design, conversion methods, and risk warnings.

[0220] The relationship between data and steps: Knowledge base results and external search results are the data foundation for competitive analysis. Competitive analysis results transform the original search data into dissemination patterns that can be used for content planning.

[0221] The relationship between data and actual processing objects: For example, system analysis found that similar Xiaohongshu content often adopts the structure of introducing parents' pain points - explaining the value of the course - reasons for saving - topic tags; similar Douyin content often adopts the structure of a question hook in the first three seconds - course scene display - comment section interaction.

[0222] Relationship between steps: This step follows the retrieval results of steps 3 and 4. Its output enters the platform strategy agent in step 7 and can be written back to the knowledge base for reuse in subsequent tasks.

[0223] Step 6: Hot Topic Recommendation Agent generates hot keywords and risk levels: Execution entity: Hotspot recommendation intelligent agent module.

[0224] Detailed actions: Comprehensive analysis of material themes, local knowledge base results, and external search results to identify trending keywords, platform tags, and dissemination trends that can be combined with the current material, and to generate applicable platforms, recommended usage methods, and risk levels for each trending topic.

[0225] Objects: MaterialReviewResult, localKnowledgeResults, and externalSearchResults.

[0226] Results: Generates hotspot recommendations, including hot keywords, recommendation platforms, combination methods, risk levels, and usage suggestions.

[0227] The relationship between data and steps: The subject matter is used to determine the relevance of the hot topic, the search results are used to determine the timeliness of the hot topic, and the platform information is used to determine the suitability of the hot topic.

[0228] The relationship between data and the actual objects being processed: For example, when recruiting students for children's dance courses, the system may recommend popular topics such as summer interest classes, children's posture management, and back-to-school planning, and mark them as suitable for use on Xiaohongshu or Douyin, while reminding users not to use expressions that exaggerate the training effect.

[0229] Relationship between steps: This step, together with step 5, provides the basis for the generation of the platform strategy in step 7, and its output can also be written back to the knowledge base.

[0230] Step 7: Platform Strategy Agent Generates Differentiated Platform Strategies: Execution entity: Platform strategy intelligent agent module.

[0231] Detailed actions: Based on the results of material review, competitor analysis, trending topic recommendations, and target platforms, generate content structures, title strategies, language styles, interaction methods, and conversion paths for different platforms.

[0232] Objects: MaterialReviewResult, competitorInsights, hotspotRecommendations, and target platform parameters.

[0233] Result: Generates platformStrategies objects, including Xiaohongshu strategy, WeChat Official Account strategy, short video script strategy, and Douyin strategy.

[0234] The relationship between data and steps: Material review results provide the factual basis, competitor analysis results provide a reference for the dissemination structure, hot topic recommendation results provide timely entry points, and target platform parameters determine the final strategy type.

[0235] The relationship between data and the actual objects being processed: For example, Xiaohongshu strategies can focus on product seeding, collection, emotional resonance, and hashtags; WeChat official account strategies can focus on complete arguments, professional credibility, and case studies; short video script strategies can focus on storyboarding, subtitles, narration, and pacing; and Douyin strategies can focus on strong hooks, short sentences, and interactive guidance.

[0236] Relationship between steps: This step follows steps 2, 5 and 6, and its output serves as a direct constraint on the copywriting generation agent in step 8.

[0237] Step 8: The copywriting AI generates a finished draft that can be published on multiple platforms. Execution entity: Copywriting generation intelligent agent module.

[0238] Detailed actions: Based on the material review results, evidence context, and platform policies, generate specific drafts applicable to different platforms; during the generation process, prioritize the use of information from available facts, do not fill in missing information without basis, and avoid, weaken, or prompt users for confirmation regarding the content marked in the risk warnings.

[0239] Objects: MaterialReviewResult, EvidenceContext, and platformStrategies.

[0240] Result: Generates a multi-platform content object, MultiPlatformDraft, which includes Xiaohongshu copywriting, WeChat official account articles, short video scripts, Douyin voiceover copywriting, simplified version, traditional version, and mind map data.

[0241] The relationship between data and steps: The copywriting generation results are jointly constrained by source facts, search evidence, competitor insights, trending recommendations, and platform strategies, avoiding reliance solely on the model for free generation.

[0242] The relationship between data and the actual target audience: For example, if the course price is not provided in the material, the generated Xiaohongshu copy and Douyin text copy cannot fabricate prices, but can only state that the specific course schedule and registration method can be consulted with the teacher; if the material clearly states that it is suitable for children aged 6-12, then this information can be used to express the target audience.

[0243] Relationship between steps: This step follows the platform strategy in step 7, and its output enters the quality assessment agent in step 9.

[0244] Step 9: The quality assessment agent performs multi-dimensional scoring: Execution entity: Quality assessment intelligent agent module.

[0245] Detailed actions: Score and diagnose the multi-platform content generated in step 8. The scoring dimensions include information accuracy, platform adaptability, title attractiveness, content completeness, language naturalness, conversion guidance, material consistency, and risk control.

[0246] Objects: MultiPlatformDraft (a content object for multiple platforms), MaterialReviewResult (a result of material review), and EvidenceContext (a context for evidence).

[0247] Results: Generates a quality assessment result, QualityEvaluationResult, which includes scores for each dimension, overall score, problem identification, modification suggestions, and risk warnings.

[0248] The relationship between data and steps: Quality assessment does not evaluate language fluency alone, but compares the generated content with the original source facts, search evidence, and platform strategies.

[0249] The relationship between data and the actual processing object: For example, if the generated copy contains "ranked first in the country" which is not provided in the material, the information accuracy and risk control scores will decrease; if the Xiaohongshu copy lacks topic tags and reasons for collection, the platform suitability score will decrease.

[0250] Relationship between steps: This step follows the content generated in step 8 and determines whether to proceed to step 10 for automatic modification.

[0251] Step 10: Determine if the quality score has reached the preset threshold: Execution entity: Workflow control module.

[0252] Detailed actions: Read the overall score and scores of each dimension in the quality assessment results, and determine whether the overall score has reached the preset quality threshold; at the same time, determine whether the current number of modifications has reached the maximum number of iterations.

[0253] Objects: QualityEvaluationResult, preset quality threshold, and current iteration number.

[0254] Results: If the overall score reaches the preset threshold, the final output will be generated; if the overall score is lower than the preset threshold and the maximum number of iterations has not been reached, the copywriting modification agent will be triggered; if the maximum number of iterations has been reached but the target is still not met, the current optimal version will be output and a manual review will be prompted.

[0255] The relationship between data and steps: Quality scores determine the flow of the process, and the number of modifications is used to control the number of closed-loop iterations to prevent infinite loops.

[0256] Relationship between data and actual processing objects: For example, if the system sets the quality threshold to 85 points and the current Xiaohongshu copywriting comprehensive score is 78 points, the system will trigger modifications based on the dimension with the lower score; if the score is 90 points, it will directly enter the final output.

[0257] Relationship between steps: This step is located between quality assessment and document revision, and is the decision point for whether to continue the control loop.

[0258] Step 11: The copywriting modification agent performs targeted optimization: Executing entity: Copywriting modification intelligent agent module.

[0259] Detailed actions: Based on the problem identification and modification suggestions in the quality assessment results, make partial or overall modifications to the generated content; modifications can be made to specific issues such as the title, opening, body structure, platform format, action guidance, risk expression, and factual errors.

[0260] Objects: MultiPlatformDraft (a content object for multiple platforms) and QualityEvaluationResult (a quality evaluation result).

[0261] Result: A revised draft document object is generated and resubmitted to the quality assessment agent.

[0262] The relationship between data and steps: The low-scoring dimensions and modification suggestions in the quality assessment results determine the modification goals, and the original material review results continue to serve as the factual boundary to prevent the introduction of new factual errors during the modification process.

[0263] The relationship between data and the actual processing objects: For example, if the quality assessment results indicate that the hook in the first three seconds of a Douyin copy is not strong, then the copy modification agent will only optimize the beginning of the Douyin copy; if it indicates that the content of a public account lacks case studies, then it will supplement the case studies in the existing materials, rather than fabricating new case studies.

[0264] Relationship between steps: This step follows the judgment result of step 10, and after modification, it returns to step 9 for re-evaluation, forming a closed loop of generation-evaluation-modification-re-evaluation.

[0265] Step 12: Output the final content plan and record the task results: Execution subject: Front-end display and task recording module.

[0266] Detailed Actions: Once the generated content reaches the quality threshold or the maximum number of iterations, the system will uniformly return the final copy, scoring results, modification suggestions, knowledge base references, external search summaries, competitor analysis, hot topic recommendations, and platform strategies to the front end for display, and generate a task record.

[0267] Objects: MultiPlatformDraft (final version of multi-platform content), QualityEvaluationResult, EvidenceContext, and TaskObject.

[0268] Results: Output the final multi-platform content solutions, including Xiaohongshu copywriting, WeChat official account articles, short video scripts, Douyin voiceover copywriting, simplified version, traditional version, quality score, optimization suggestions, and task records.

[0269] Relationship between data and steps: The final output data is generated by all the aforementioned steps, including both the content results and the basis for generation and evaluation records.

[0270] The relationship between data and the actual objects being processed: For example, users can ultimately obtain a complete course promotion plan, including promotional copy suitable for Xiaohongshu, in-depth articles suitable for WeChat official accounts, scripts for Douyin (TikTok) videos, and quality scores and modification instructions for each platform.

[0271] Relationship between steps: This step is the final output node of the current task, and at the same time provides a data source for the knowledge write-back in step 13.

[0272] Step 13: Write the valid results back to the knowledge base: Execution entity: Knowledge base module.

[0273] Detailed actions: Write the competitor analysis results, hot topic recommendation results, platform strategies, final copy, quality score, modification history, external search summary and task observation time into the local knowledge base, and label them according to topic, keywords, platform, industry category and time.

[0274] Objects: CompetitorInsights, hotspotRecommendations, platformStrategies, MultiPlatformDraft (final version of multi-platform content), and QualityEvaluationResult.

[0275] Result: A new knowledge base record, KnowledgeRecord, is generated for retrieval and reuse in subsequent similar tasks.

[0276] The relationship between data and steps: The effective data of the current task is accumulated into knowledge base data, so that subsequent tasks can call on historical experience and reduce repeated retrieval and analysis.

[0277] The relationship between data and the actual objects being processed: For example, the content generated in this task, such as the Xiaohongshu title structure for children's dance enrollment, the usage of hot topics for summer interest classes, and the hook template for the first three seconds of Douyin, can be retrieved and used again in subsequent similar enrollment tasks.

[0278] Relationship between steps: This step follows the final output of step 12 and forms the basis for the continuous optimization and reuse capabilities of the entire system. The output of the current task will become the data source for knowledge base retrieval in step 3 of subsequent tasks.

[0279] Regarding model output: The agents in this application can be implemented based on large language models, rule engines, retrieval models, scoring models, or combinations of the above models. Each agent is not limited to a specific model, but is divided according to its functional responsibilities.

[0280] In one implementation, each agent includes the following structure: 1. Input parsing unit: Used to receive the data object output from the previous step and extract the fields required by the current agent.

[0281] 2. Task prompting construction unit: used to construct task instructions, constraints and output formats according to the responsibilities of the agent.

[0282] 3. Model Inference Unit: Used to call large language models or other algorithm models for semantic analysis, generation, scoring, or modification.

[0283] 4. Structured Output Unit: Used to convert model output into preset data structures, such as material review results, competitor analysis results, hot topic recommendation results, platform strategy objects, copywriting objects, or rating objects.

[0284] 5. Verification and Write-back Unit: Used to check whether the output fields are complete and write valid results into the task record or knowledge base.

[0285] The agents are connected through a unified data object, with the output of one agent serving as the input of the next. This allows the system to form a stable multi-agent collaborative workflow.

[0286] Regarding the output data: After the above processing, the original data, initially unstructured or semi-structured material, is sequentially transformed into TaskObject, MaterialReviewResult, EvidenceContext, PlatformStrategies, MultiPlatformDraft, QualityEvaluationResult, and KnowledgeRecord. During this data flow, the data undergoes changes from raw material to structured factual assets, from scattered retrieval data to evidence context, from platform requirements to differentiated platform strategies, from strategy and factual constraints to publishable drafts, from generated results to scoring and modification records, and from single task results to reusable knowledge base records.

[0287] Through the aforementioned data flow, this application achieves the following technical effects: First, by reviewing materials and extracting available facts, it improves the consistency between generated content and original materials; second, by identifying missing information and providing risk warnings, it reduces unfounded expansion and risky expressions; third, by using local knowledge bases and external near-real-time retrieval, it improves the timeliness and traceability of content generation; fourth, by using competitor analysis and hot topic recommendations, it improves the completeness of content planning basis; fifth, by using platform strategy intelligence, it improves the adaptability of content structures and expression methods across different platforms; sixth, by using quality assessment and automatic modification closed loop, it improves the stability of the final output results; and seventh, by using a knowledge base write-back mechanism, it enables the system to continuously accumulate historical experience and support subsequent reuse.

[0288] Optionally, 1. Regarding the input method of materials, in addition to inputting text and uploading Word, PDF, and image files through the front-end page, raw materials and task parameters can also be obtained through application programming interfaces (APIs), third-party content management systems, enterprise internal material libraries, web link scraping, speech-to-text conversion, image text recognition, or database synchronization. Any method that can obtain the raw content to be processed and task parameters such as target platform, audience, tone style, and action guidance can be used as the data input method for this application.

[0289] 2. Regarding the material review process, the intelligent agent for material review can be implemented using a large language model, or it can be implemented using a rule engine, keyword matching model, information extraction model, named entity recognition model, text classification model, or a combination of the above models. Its core purpose is to extract themes, summaries, keywords, available facts, missing information, and risk warnings from the original materials, and therefore it is not limited to a specific model structure.

[0290] 3. Regarding available methods for extracting facts, natural language processing models can be used for automatic extraction, or preset field templates, manual confirmation forms, entity recognition models, event extraction models, or rule bases can be used for extraction. For example, for course enrollment materials, fields such as course name, applicable age, teaching time, registration method, and course features can be preset; for product promotion materials, fields such as product name, functions, price, applicable scenarios, and purchase methods can be preset.

[0291] 4. Regarding the identification of missing information, fixed field validation rules or dynamic validation rules based on industry categories can be used. For example, tasks related to event promotion need to validate the event time, location, registration portal, and organizer; tasks related to course promotion need to validate the target audience, course time, course features, and contact information; and tasks related to product promotion need to validate the product specifications, price, target audience, and purchase path.

[0292] 5. Regarding risk warning methods, large language models can be used for identification, or sensitive word databases, compliance rule databases, absolute term dictionaries, industry-specific prohibited word lists, manual review rules, or classification models can be employed. Risk warnings may include exaggerated claims, absolute expressions, unverified rankings, false honors, data from unknown sources, sensitive topics, and expressions unsuitable for public dissemination.

[0293] 6. Regarding the implementation method of the knowledge base, a local knowledge base can be implemented using relational databases, document databases, vector databases, graph databases, search engine indexes, or file systems, or a combination of multiple databases. The knowledge base can store information such as historical competitor analysis, trending keywords, platform strategies, task results, quality scores, modification records, observation time, and citation sources.

[0294] 7. Regarding knowledge base retrieval methods, keyword retrieval, full-text retrieval, semantic vector retrieval, hybrid retrieval, tag filtering, time range filtering, platform filtering, industry category filtering, or similar task recall can be used. Any method that can retrieve relevant historical data based on the current material's theme, keywords, target platform, and industry information is a viable alternative implementation method in this application.

[0295] 8. Regarding external retrieval methods, external materials can come from search engines, public web pages, platform open interfaces, third-party data interfaces, industry news websites, public content on social media, enterprise-owned data sources, or public content databases. External retrieval interfaces are not limited to a specific service; as long as they can obtain recent information related to the material's theme, target platform, or trending topics, the same purpose can be achieved.

[0296] 9. Regarding competitive analysis methods, the competitive analysis agent can directly analyze external search results, or it can first perform deduplication, sorting, clustering, summarizing, and credibility assessment on the search results, and then extract title structure, content angles, interaction methods, dissemination strategies, and risk points. Competitive analysis can be completed using large language models, or it can be assisted by text clustering models, keyword statistical models, rule analysis models, or manually labeled results.

[0297] 10. Regarding hot topic recommendation methods, a recommendation method based on keyword popularity can be adopted, or a comprehensive recommendation method based on semantic relevance, time decay weight, platform adaptability, risk level, and user goals can be used. For example, the recommendation priority of hot topics can be calculated based on the relevance of the hot topic to the material theme, the release time of the hot topic, the characteristics of the target platform, and the risk level.

[0298] 11. Regarding the platform strategy generation method, it can be automatically generated by the platform strategy intelligent agent, or it can be generated based on a preset platform strategy template. For example, the Xiaohongshu platform can be configured with a structure of "title - pain point - reason for recommendation - collection value - topic tags"; WeChat official accounts can be configured with a structure of "title - introduction - body section - case description - closing guidance"; short video scripts can be configured with a structure of "time period - visuals - subtitles - narration - rhythm points"; and Douyin can be configured with a structure of "first three seconds hook - main expression - interactive questions - action guidance".

[0299] 12. Regarding target platforms, this application is not limited to Xiaohongshu, WeChat official accounts, short video platforms, and Douyin, but can also be extended to other content distribution channels such as Weibo, Zhihu, Bilibili, Kuaishou, WeChat Video Accounts, Toutiao, official website news, email marketing, community copywriting, and advertising copywriting. As long as different channels have differences in content structure, language style, and dissemination methods, the multi-platform content planning scheme of this application can be applied.

[0300] 13. Regarding copy generation methods, a single large language model can be used, or multiple different models can be used to generate content for different platforms; a unified model combined with different prompt words can be used, or different platforms can use different dedicated generation models; a combination of template generation and model generation can also be used. The core is to generate specific drafts for different platforms based on the material review results, evidence context, and platform policies.

[0301] 14. Regarding quality assessment methods, a large language model scoring system can be used, or rule-based scoring, classification model scoring, multi-model voting scoring, manual review scoring, or a combination of the above methods can be employed. Scoring dimensions may include information accuracy, platform compatibility, title attractiveness, content completeness, language naturalness, conversion guidance, material consistency, risk control, format completeness, and readability.

[0302] 15. Regarding quality threshold settings, either fixed or dynamic thresholds can be used. For example, the system can uniformly set an output threshold after a comprehensive score reaches 85 points, or different thresholds can be set according to different platforms, task types, and risk levels. For high-risk promotional content, the weight of information accuracy and risk control dimensions can be increased.

[0303] 16. Regarding copywriting revision methods, either a complete rewrite or partial modifications can be used. Partial modifications can target specific issues such as the title, opening hooks, body structure, platform formatting, action guidance, risk expression, missing information prompts, and factual errors. After revision, the copy can be re-entered for quality assessment or manually reviewed before being output.

[0304] 17. Regarding the closed-loop iteration method, an automatic closed loop of "generation—evaluation—modification—re-evaluation" can be adopted, or a semi-automatic closed loop of "generation—evaluation—manual confirmation—modification" can be adopted. The system can set a maximum number of iterations to avoid infinite loops; it can also automatically prompt the user to supplement materials or enter manual review when there is no significant improvement in scores after multiple consecutive tests.

[0305] 18. Regarding knowledge base write-back methods, all task results can be written back to the knowledge base, or only results with scores reaching a preset threshold can be written back. Complete text can be written back, or only competitor analysis summaries, trending recommendation results, platform strategies, quality scores, and modification records can be written back. Knowledge write-back can be tagged and stored according to topic, platform, industry, time, score, and task type.

[0306] 19. Regarding system deployment, this application can be deployed on local servers, cloud servers, private environments, enterprise intranets, edge computing devices, or software-as-a-service platforms. Modules can be centrally deployed on the same server, or they can be distributed according to front-end, task scheduling, agent services, knowledge base services, external retrieval services, and model inference services.

[0307] 20. Regarding model invocation methods, you can invoke the locally deployed large language model or the cloud model interface; you can use a general large language model or a dedicated model that has been fine-tuned with industry data; you can use one model to complete multiple intelligent agent tasks, or you can configure different models for material review, competitor analysis, copy generation, quality assessment, and copy modification.

[0308] 21. Regarding system interaction methods, users can submit tasks once and wait for the complete solution output, or they can manually confirm during the material review, platform strategy, copy generation, or quality assessment stages. For example, when the system identifies missing information, it can pause generation and request the user to supplement it; alternatively, it can continue generation but mark the content to be confirmed in the output.

[0309] 22. Regarding the output format, the output of this application is not limited to text page display, but may also include Word documents, PDF files, tables, mind maps, Markdown files, short video storyboards, content calendars, topic selection forms, title libraries, or interface data. As long as the final output can reflect the multi-platform content scheme, generation basis, quality score, and task record, the purpose of this application can be achieved.

[0310] In this embodiment, adaptive control of content quality is implemented, with low scores triggering targeted modifications and high scores allowing output, ensuring that the delivered draft always meets quality standards.

[0311] Furthermore, embodiments of this application also provide a multi-platform content planning and generation system, the system comprising: The input module is used to acquire raw materials and task parameters, wherein the task parameters include at least one target platform information; The information deconstruction module is used to deconstruct the original material and determine the deconstruction results of verifiable fact items, missing information items, and content risk items in the original material. The verifiable fact items are used to characterize factual information in the original material that can be used as the basis for content generation. The missing information items are used to characterize key information in the original material that is not provided but affects the integrity of the content. The content risk items are used to characterize prompt information that affects the compliance of the content or the security of its dissemination. The retrieval module is used to obtain related information from an internal knowledge base and / or external information sources based on the deconstruction results; An evidence construction module is used to integrate the associated information into an evidence context for constraining content generation; The strategy generation module is used to generate a content construction strategy corresponding to the target platform information based on the verifiable fact items, the content risk items, and the evidence context; the content construction strategy includes at least one of the following: content structure strategy, title strategy, language style strategy, interaction method strategy, and conversion path strategy. The copywriting generation module is used to generate corresponding publishable drafts for each target platform according to the content construction strategy. The content construction strategy corresponds one-to-one with each target platform, and the content construction strategies for different target platforms differ in at least one aspect of content structure, title strategy, language style, interaction method, and conversion path.

[0312] In the specific implementation, see Figure 4 , Figure 4 This is a system module structure diagram provided for an embodiment of this application. Figure 4 The multi-platform content planning system mentioned above is the multi-platform content planning and generation system. The system architecture of this application includes a user terminal, a front-end interaction layer, a task scheduling layer, an intelligent agent processing layer, a data storage layer, an external data access layer, and a result display layer. The system comprises the following components: The user interface is used to input original materials, upload files, and set task parameters; the front-end interaction layer receives user input, displays task progress, and outputs results; the task scheduling layer controls the execution order, data transmission, threshold judgment, and iteration count of each module and agent; the agent processing layer includes agents for material review, competitor analysis, hot topic recommendation, platform strategy, copywriting generation, quality assessment, and copywriting modification, which respectively perform material parsing, search enhancement, strategy generation, content generation, quality assessment, and automatic modification; the data storage layer includes a task record library, knowledge base, vector index library, and result storage library, used to store original tasks, structured materials, historical competitor analysis, hot keywords, platform strategies, generated results, scoring results, and modification records; the external data access layer calls external public search interfaces to obtain recently available public information related to the current material theme and target platform; and the result display layer returns the final multi-platform content solution, quality score, optimization suggestions, knowledge base references, and task records to the user.

[0313] Specifically, the functions of each component are as follows: (1) User side: Used for users to input text materials, upload Word, PDF, pictures and other files, and set industry keywords, target audience, target platform, action guidance, tone style and model selection parameters.

[0314] (2) Front-end interaction module: used to receive user input, display task status, display finished drafts on various platforms, display quality scores and export results.

[0315] (3) Input and Material Management Module: Used to parse, clean, extract fields and encapsulate task objects for user-inputted text or uploaded files.

[0316] (4) Workflow control module: used to manage the execution order of material review, knowledge base retrieval, external retrieval, competitor analysis, hot topic recommendation, platform strategy, copywriting generation, quality assessment, automatic modification and final output.

[0317] (5) Material review intelligent agent module: used to extract themes, abstracts, core viewpoints, user pain points, selling points, scenarios, cases, keywords, available facts, missing information and risk warnings from the original materials.

[0318] (6) Knowledge Base Module: Used to store and retrieve historical competitor analysis, hot keywords, platform strategies, task results, quality scores, modification records and observation time.

[0319] (7) External search module: used to call external public search interfaces based on material theme, keywords and target platform to obtain recent online materials, platform content summaries, competitor titles and hot information.

[0320] (8) Competitive analysis intelligent agent module: used to analyze the title structure, content entry angle, interaction method, dissemination strategy and risk points of similar content or competitor content.

[0321] (9) Hot Topic Recommendation Intelligent Agent Module: Used to recommend hot keywords, applicable platforms, combination methods and risk levels based on material topics, local knowledge base and external search results.

[0322] (10) Platform Strategy Intelligent Agent Module: Used to generate differentiated content strategies for different platforms such as Xiaohongshu, WeChat Official Accounts, short video scripts, and Douyin.

[0323] (11) Copywriting generation intelligent agent module: used to generate multi-platform publishable drafts based on material review results, evidence context and platform strategy.

[0324] (12) Quality assessment agent module: used to score the generated results from dimensions such as information accuracy, platform adaptability, title attractiveness, content completeness, language naturalness, conversion guidance, material consistency and risk control.

[0325] (13) Copywriting modification agent module: used to make partial or overall modifications to the title, body text, platform format, action guidance, risk expression or factual errors based on the quality assessment results.

[0326] (14) Front-end display and task record module: used to display the final solution, score, modification suggestions, search summary, knowledge base reference and historical task record.

[0327] The relationships between the components are as follows: The user interface connects to the front-end interaction module, which in turn connects to the input and material management module. The input and material management module encapsulates user input into task objects and passes them to the workflow control module. The workflow control module schedules the following intelligent modules: material review module, knowledge base module, external retrieval module, competitor analysis module, trending topic recommendation module, platform strategy module, copywriting generation module, quality assessment module, and copywriting modification module. The output of the material review module serves as the factual basis for knowledge base retrieval, external retrieval, platform strategy generation, and copywriting generation. The outputs of the knowledge base module and the external retrieval module jointly serve as inputs to the competitor analysis module and the trending topic recommendation module. The competitor analysis results and trending topic recommendation results serve as inputs to the platform strategy module. The platform strategy results and material review results jointly serve as inputs to the copywriting generation module. The copywriting generation results serve as inputs to the quality assessment module. If the quality assessment result reaches a preset quality threshold, it is passed to the front-end display and task recording module; if it does not reach the preset quality threshold, it is passed to the copywriting modification module for modification and then re-enters the quality assessment module. The final generated competitor analysis, trending topic recommendations, platform strategies, copywriting results, quality scores, and modification records are written back to the knowledge base module for reuse in subsequent tasks.

[0328] From a data flow perspective, the system forms a structure of user input—task encapsulation—material review—knowledge base retrieval—external retrieval—competitive analysis—hot topic recommendation—platform strategy—copywriting generation—quality assessment—automatic modification—final output—knowledge write-back. From a control perspective, the workflow control module is responsible for scheduling the execution order and loop judgments of each agent. From a storage perspective, the knowledge base module provides historical data for the current task and also receives new data generated by the current task, thus forming a continuously updated data accumulation mechanism.

[0329] Through the aforementioned system architecture and interaction process, the system ultimately outputs multi-platform content solutions, quality scoring results, modification suggestions, knowledge base citations, external search summaries, competitor analysis results, hot topic recommendation results, platform strategy results, and task records. After processing, the original materials are transformed from unstructured or semi-structured input into structured factual assets, evidentiary context, platform strategy objects, multi-platform drafts, quality assessment results, and knowledge base records.

[0330] This system achieves the following technical effects: First, by transforming raw materials into structured factual assets through the input and material management module and the material review intelligent agent module, it improves the consistency between the generated content and the original materials. Second, by introducing historical knowledge and recently available public information into the content generation process through the knowledge base module and the external retrieval module, it improves the timeliness and traceability of the generated content. Third, by transforming retrieved information into dissemination structure patterns and hot topic usage suggestions through the competitor analysis intelligent agent module and the hot topic recommendation intelligent agent module, it improves the completeness of the content planning basis. Fourth, by enabling different platforms such as Xiaohongshu, WeChat official accounts, short video scripts, and Douyin to have differentiated content structures and expression methods through the platform strategy intelligent agent module. Fifth, by forming an automatic optimization closed loop based on scoring thresholds through the quality assessment intelligent agent module, the copywriting modification intelligent agent module, and the workflow control module, it improves the stability and publishability of the final generated content. Sixth, by using the knowledge base write-back mechanism, it saves the effective results of the current task into reusable data for subsequent tasks, improving the system's continuous optimization capabilities.

[0331] In this embodiment, the factual boundaries and compliance restrictions of content generation are deconstructed by deconstructing material information, and the evidentiary context is constructed by combining internal historical experience and external real-time information. Under this strict constraint, differentiated and structured multi-platform adaptation strategies are generated, and finally, multi-platform drafts that can be directly published are produced, solving the pain points of easy fabrication of facts, confusion of platform styles, and loss of control of content compliance in general large models.

[0332] Based on the methods described in any of the above embodiments, this application also provides, as follows: Figure 5 The diagram shows the structure of an electronic device. Figure 5 At the hardware level, the electronic device includes a processor, an internal bus, a network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to implement the methods described in any of the above embodiments.

[0333] Based on the methods described in any of the above embodiments, this application also provides a computer storage medium storing a computer program, which, when executed by a processor, can be used to perform the methods described in any of the above embodiments.

[0334] Based on the methods described in any of the above embodiments, this application also provides a computer program product, which includes one or more computer programs or instructions. The computer program or instructions may be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. When executed by a processor, the computer program implements the methods described in any of the above embodiments.

[0335] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0336] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0337] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0338] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0339] The above description is merely a specific embodiment of this application, but the scope of protection of this application 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 this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0340] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A method for content planning and generation across multiple platforms, characterized in that, The method includes: Obtain raw materials and task parameters, wherein the task parameters include at least one target platform information; The original material is deconstructed to determine the deconstruction results of verifiable fact items, missing information items, and content risk items in the original material; the verifiable fact items are used to characterize factual information in the original material that can be used as the basis for content generation; the missing information items are used to characterize key information in the original material that is not provided but affects the integrity of the content; and the content risk items are used to characterize prompt information that affects the compliance of the content or the security of its dissemination. Based on the deconstruction results, relevant information is obtained from internal knowledge bases and / or external information sources; The associated information is integrated into an evidentiary context for constraining content generation; Based on the verifiable fact items, the content risk items, and the evidence context, a content construction strategy corresponding to the target platform information is generated; the content construction strategy includes at least one of the following: content structure strategy, title strategy, language style strategy, interaction method strategy, and conversion path strategy. Based on the content building strategy, corresponding publishable articles are generated for each target platform. The content building strategy corresponds one-to-one with each target platform, and the content building strategies for different target platforms differ in at least one aspect of content structure, title strategy, language style, interaction method, and conversion path.

2. The method according to claim 1, characterized in that, The step of obtaining relevant information from an internal knowledge base and / or external information sources based on the deconstruction results includes: Based on the verifiable fact items and / or the missing information items, historical data is retrieved from the internal knowledge base, including at least one of historical competitor analysis results, historical hot keywords, historical platform strategies, and historical task results; and / or, Based on the verifiable fact items and / or the missing information items and / or the target platform information, retrieve recent information from the external information source. The recent information includes at least one of publicly available online information, platform content summaries, competitor titles, and trending information.

3. The method according to claim 1, characterized in that, The deconstruction results were also obtained in the following manner: Extract at least one of the following from the original material: theme, summary, core viewpoint, selling point, application scenario, case study, and keywords.

4. The method according to claim 1, characterized in that, The step of generating a content construction strategy corresponding to the target platform information based on the verifiable fact items, the content risk items, and the evidence context includes: Based on the deconstruction results and the evidence context, a competitive analysis result is generated, which includes at least one of the following: title structure, content entry angle, interaction method, dissemination strategy, and risk points of similar content. Based on the verifiable fact items, the content risk items, the evidence context, and the competitor analysis results, a content building strategy corresponding to the target platform information is generated.

5. The method according to claim 4, characterized in that, The step of generating a content construction strategy corresponding to the target platform information based on the verifiable fact items, the content risk items, the evidence context, and the competitor analysis results includes: Based on the deconstruction results and the evidence context, a hot topic recommendation result is generated, which includes at least one of the following: hot topic keywords, applicable platforms, combination methods, and risk levels. Based on the verifiable fact items, the content risk items, the evidence context, the competitor analysis results, and the hot topic recommendation results, a content construction strategy corresponding to the target platform information is generated.

6. The method according to claim 1, characterized in that, The method further includes: The publishable manuscript is subjected to a multi-dimensional quality score to obtain the score results; the multi-dimensional quality score includes at least one of the following: information accuracy score, platform compatibility score, title attractiveness score, content completeness score, language naturalness score, conversion guidance score, material consistency score, and risk control score. When the scoring result does not reach the preset quality threshold, the publishable manuscript is modified, and the step of performing multi-dimensional quality scoring on the publishable manuscript is repeated. When the rating result reaches the preset quality threshold, the publishable manuscript is output.

7. A multi-platform content planning and generation system, characterized in that, The system includes: The input module is used to acquire raw materials and task parameters, wherein the task parameters include at least one target platform information; The information deconstruction module is used to deconstruct the original material and determine the deconstruction results of verifiable fact items, missing information items, and content risk items in the original material. The verifiable fact items are used to characterize factual information in the original material that can be used as the basis for content generation. The missing information items are used to characterize key information in the original material that is not provided but affects the integrity of the content. The content risk items are used to characterize prompt information that affects the compliance of the content or the security of its dissemination. The retrieval module is used to obtain related information from an internal knowledge base and / or external information sources based on the deconstruction results; An evidence construction module is used to integrate the associated information into an evidence context for constraining content generation; The strategy generation module is used to generate a content construction strategy corresponding to the target platform information based on the verifiable fact items, the content risk items, and the evidence context; the content construction strategy includes at least one of the following: content structure strategy, title strategy, language style strategy, interaction method strategy, and conversion path strategy. The copywriting generation module is used to generate corresponding publishable drafts for each target platform according to the content construction strategy. The content construction strategy corresponds one-to-one with each target platform, and the content construction strategies for different target platforms differ in at least one aspect of content structure, title strategy, language style, interaction method, and conversion path.

8. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store processor-executable instructions; Wherein, when the processor invokes the executable instructions, it implements the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, It stores computer instructions that, when executed by a processor, implement the method described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method of any one of claims 1-6.

Images