A content creation system and method based on multi-agent collaboration

Abstract

This application discloses a content creation system and method based on multi-agent collaboration. The content creation system includes an interaction subsystem, an agent subsystem, a knowledge subsystem, and a model subsystem. The agent subsystem includes a search and recommendation agent cluster, a creation agent cluster, and a review agent cluster. The knowledge subsystem is used to construct a basic knowledge base including business knowledge. The interaction subsystem is used to receive user-input creation request information. The search and recommendation agent cluster is used to perform intent recognition on the creation request information to obtain task intent information, and to retrieve knowledge context information matching the task intent information from the basic knowledge base. The creation agent cluster is used to generate initial draft text information based on the knowledge context information and task intent information. The review agent cluster is used to perform compliance correction operations on the initial draft text information to obtain optimized content results. This system can improve the quality and efficiency of content creation.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a content creation system and method based on multi-agent collaboration. Background Technology

[0002] With the rapid development of the insurance industry, the business volume of various insurance products, medical and elderly care services, and insurance financial products, such as life insurance, health insurance, senior living communities, and high-end customer services, is increasing daily. In empowering insurance financial service marketing, the existing technical solutions for content creation and personal IP empowerment in the insurance and financial industry can be mainly divided into two categories: one is a fragmented tool combination solution, composed of multiple independent tools, lacking intelligent connections, requiring agents to spend a lot of time manually searching and filtering, resulting in low efficiency.

[0003] To improve automation, vertical solutions have emerged in related technologies, using SaaS tools focused on a specific stage, such as enterprise content management systems (CMS), social media management platforms (like Hootsuite), or standalone AI marketing copy generators. However, these approaches still have significant shortcomings: fragmented processes, inability to form a closed loop, disconnected data and user experience across stages, and agents needing to switch repeatedly between different platforms, resulting in cumbersome processes.

[0004] Therefore, the content creation process in related technologies suffers from problems such as low copywriting quality and low generation efficiency. Summary of the Invention

[0005] This application provides a content creation system and method based on multi-agent collaboration to improve the quality of content creation copy and generation efficiency.

[0006] In a first aspect, embodiments of this application provide a content creation system based on multi-agent collaboration, comprising: an interaction subsystem, an agent subsystem, a knowledge subsystem, and a model subsystem; the agent subsystem includes a search and recommendation agent cluster, a creation agent cluster, and a review agent cluster; the model subsystem includes a large language model corresponding to the agents of the agent subsystem. The knowledge subsystem is used to build a basic knowledge base that includes business knowledge; The interactive subsystem is used to receive user input of creative requirements information; The search and recommendation intelligent agent cluster is used to perform intent recognition on the creation demand information to obtain task intent information; and to retrieve knowledge context information that matches the task intent information from the basic knowledge base. The creative intelligent agent cluster is used to generate initial draft text information based on the knowledge context information and the task intent information; The intelligent review agent cluster is used to perform compliance correction operations on the initial draft text information to obtain optimized content results.

[0007] Secondly, embodiments of this application provide a content creation method based on multi-agent collaboration, applied to a content creation system; the content creation system includes an interaction subsystem, an agent subsystem, a knowledge subsystem, and a model subsystem; the agent subsystem includes a search and recommendation agent cluster, a creation agent cluster, and a review agent cluster; the model subsystem includes a large language model corresponding to the agents of the agent subsystem; the method includes: A basic knowledge base, including business knowledge, is constructed through the knowledge subsystem. The interactive subsystem receives user input regarding creative requirements. The search and recommendation intelligent agent cluster performs intent recognition on the creation demand information to obtain task intent information; The search and push intelligent agent cluster retrieves knowledge context information that matches the task intent information from the basic knowledge base; The creative intelligent agent cluster generates initial draft text information based on the knowledge context information and the task intent information; The intelligent review agent cluster performs compliance corrections on the initial draft text information to obtain optimized content results.

[0008] Thirdly, embodiments of this application provide an electronic device, including: at least one processor, and a memory communicatively connected to the at least one processor, wherein: The memory stores a computer program that can be executed by at least one processor, which enables the at least one processor to perform the second aspect of the multi-agent collaborative content creation method.

[0009] Fourthly, embodiments of this application provide a storage medium in which, when a computer program in the storage medium is executed by a processor of an electronic device, the electronic device is able to execute the content creation method based on multi-agent collaboration of the second aspect.

[0010] The beneficial effects of this application are as follows: The content creation system based on multi-agent collaboration provided in this application includes an interaction subsystem, an intelligent agent subsystem, a knowledge subsystem, and a model subsystem. The intelligent agent subsystem includes a search and recommendation intelligent agent cluster, a creation intelligent agent cluster, and a review intelligent agent cluster. The model subsystem includes a large language model corresponding to the intelligent agents in the intelligent agent subsystem. The knowledge subsystem is used to construct a basic knowledge base including business knowledge. The interaction subsystem is used to receive creation request information input by the user. The search and recommendation intelligent agent cluster is used to perform intent recognition on the creation request information to obtain task intent information, and to retrieve knowledge context information matching the task intent information from the basic knowledge base. The creation intelligent agent cluster is used to generate initial draft text information based on the knowledge context information and the task intent information. The review intelligent agent cluster is used to perform compliance correction operations on the initial draft text information to obtain optimized content results. In this embodiment, a four-layer content creation system is constructed, comprising an interaction subsystem, an intelligent agent subsystem, a knowledge subsystem, and a model subsystem. The intelligent agent subsystem includes a search and recommendation intelligent agent cluster, a creation intelligent agent cluster, and a review intelligent agent cluster. The knowledge subsystem constructs a basic knowledge base including business knowledge. The interaction subsystem receives user-inputted creation request information. The search and recommendation intelligent agent cluster performs intent recognition on the creation request information to obtain task intent information. The search and recommendation intelligent agent cluster retrieves knowledge context information matching the task intent information from the basic knowledge base. The creation intelligent agent cluster generates initial draft text information based on the knowledge context information and the task intent information. The review intelligent agent cluster performs compliance correction operations on the initial draft text information to obtain optimized content results. This achieves precision and intelligence in the content creation process, significantly improving the quality and efficiency of content creation. Furthermore, by constructing an independent "knowledge subsystem," responsible for deep integration and structured processing of internal and external knowledge, high-purity, scenario-based knowledge fuel is provided to the upper-layer intelligent agents, thereby ensuring the professionalism of the final output content. Furthermore, by coordinating the search and recommendation intelligent agent cluster with the creation intelligent agent cluster, the underlying model capabilities of the knowledge subsystem and model subsystem are efficiently invoked to process materials and generate copy that meets business requirements, thereby automating and streamlining the creation process. Additionally, the review intelligent agent cluster performs compliance corrections on the initial draft copy information, resulting in optimized content. This provides an AI-driven risk management and data calibration mechanism embedded in the creation process, enabling real-time control of compliance risks and efficient correction of data accuracy, thus improving the quality and efficiency of content creation.

[0011] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0012] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A schematic diagram of the structure of a content creation system based on multi-agent collaboration provided in this application embodiment; Figure 2 A flowchart illustrating a content creation method based on multi-agent collaboration provided in this application embodiment; Figure 3 A schematic diagram of the structure of a content creation system based on multi-agent collaboration provided in this application embodiment; Figure 4 A schematic diagram of the interface of an interactive subsystem for implementing a content creation method based on multi-agent collaboration, provided in an embodiment of this application; Figure 5 This is a schematic diagram of the hardware structure of an electronic device for implementing a content creation method based on multi-agent collaboration, provided in an embodiment of this application. Detailed Implementation

[0013] To improve the quality and efficiency of content creation, this application provides a content creation system and method based on multi-agent collaboration.

[0014] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0015] It should be noted that the terms "first," "second," etc., used in the description of the embodiments of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0016] The acquisition, transmission, storage, and use of data in this application all comply with the requirements of relevant national laws and regulations.

[0017] The following description, in conjunction with the accompanying drawings, illustrates some preferred embodiments of this application.

[0018] See Figure 1 , Figure 1 This is a schematic diagram of the structure of a content creation system based on multi-agent collaboration provided in an embodiment of this application. The content creation system 100 includes: an interaction subsystem 110, an agent subsystem 120, a knowledge subsystem 130, and a model subsystem 140. The agent subsystem 120 includes a search and recommendation agent cluster, a creation agent cluster, and a review agent cluster. The model subsystem 140 includes a large language model corresponding to the agents of the agent subsystem 120.

[0019] In embodiments of this application, the model subsystem 140 includes a large language model for invocation by agents of the agent subsystem 120. The model subsystem provides basic AI capabilities to the agent subsystem.

[0020] In the embodiments of this application, the basic AI capability is the data processing capability of the intelligent agent in the intelligent agent subsystem.

[0021] In some embodiments, the interaction subsystem 110 is used to deeply integrate and structure internal and external knowledge, providing high-purity, scenario-based knowledge raw materials for the intelligent agents of the upper-layer intelligent agent subsystem 120, thereby ensuring the professionalism of the final output content.

[0022] In some embodiments, the large language model in the model subsystem corresponding to the agent in the agent subsystem includes a general understanding and search recommendation LLM, a compliance review LLM, and a content generation LLM.

[0023] This application does not limit the specific number of large language models in the model subsystem.

[0024] In some embodiments, the compliance audit LLM is a dedicated audit big model.

[0025] In some embodiments, content generation LLMs include graph generation LLMs and document generation LLMs.

[0026] In some embodiments, the content creation system adopts a layered and decoupled cloud-native architecture and is deployed on a cloud server cluster. Control operations are initiated by users (field agents and all office staff). For example, user User1 accesses the content creation system through the interaction subsystem 110, where the interaction subsystem can be a Web, App, or API.

[0027] In some embodiments, the intelligent agent subsystem 120 serves as the control center of the content creation system, which can be considered as the "heart"; the intelligent agent subsystem 120 is a multi-agent collaborative engine responsible for receiving, decomposing, and processing user requests; the knowledge subsystem 130 serves as the wisdom center of the content creation system, which can be considered as the "brain," and provides precise and structured business knowledge support to the agents in the intelligent agent subsystem 120; the model subsystem 140 serves as the functional unit of the content creation system, which can be considered as the "power source," providing the basic AI capabilities of the intelligent agent subsystem 120.

[0028] In some embodiments, the model subsystem includes the qwen-plus model, the DeepSeek model, and the kimi-k2-instruct model. User requests enter through the interaction subsystem, where the search and recommendation agent cluster within the agent subsystem uses the qwen-plus model as the corresponding large language model to perform intent recognition and retrieve relevant content from the knowledge subsystem. The creation agent cluster uses the DeepSeek model as the corresponding large language model and generates an initial draft based on the retrieval results and the capabilities of the model subsystem. The review agent cluster uses the kimi-k2-instruct model as the corresponding large language model to perform compliance checks and clarification on the initial draft and returns the results to the interaction subsystem for presentation to the user. Throughout the process, the various subsystems of the content creation system exchange data efficiently and systematically through predefined APIs, providing a new platform for scenario-based experiences.

[0029] In some embodiments, the intelligent agent subsystem 120 is a multi-agent cluster architecture, wherein the creative intelligent agent cluster is responsible for calling the underlying model capabilities of the knowledge subsystem information and model subsystem to process materials and generate copy that meets business requirements, thereby automating and business-oriented the creation process and effectively promoting the insurance and financial industry to achieve intelligent sales based on IT technology.

[0030] In some embodiments, in the intelligent agent subsystem 120, the review intelligent agent cluster cooperates with the search and push intelligent agent cluster and the creation intelligent agent cluster to achieve real-time control of compliance risks and automatic correction of data standards, deeply embedding the review process into the creation chain, and transforming post-event inspection into in-process control.

[0031] In some embodiments of this application, a content creation method based on multi-agent collaboration is achieved through a four-layer architecture consisting of an interaction subsystem, an intelligent agent subsystem, a knowledge subsystem, and a model subsystem. Specifically, the interaction subsystem provides users with input and selection entry points; the model subsystem provides different Large Language Models (LLMs); the knowledge subsystem provides precise knowledge and information on business scenarios; and the intelligent agent subsystem implements various logics through intelligent agents. This content creation system fundamentally overcomes three major business pain points: "difficulty in finding standard business content," "high creation threshold," and "difficulty in ensuring compliance." It provides a comprehensive, low-threshold, and highly compliant content empowerment service for a large number of frontline marketing personnel, integrating "precise retrieval, intelligent creation, and instant review," thereby activating organizational content productivity, achieving a dual value enhancement of brand communication and business conversion, and more effectively promoting the development of experiential sales.

[0032] The content creation system based on multi-agent collaboration in some embodiments of this application can generate insurance-related articles; the content creation system based on multi-agent collaboration in other embodiments of this application can generate financial articles; and the content creation system based on multi-agent collaboration in other embodiments of this application can also generate articles with cross-attributes of insurance and finance.

[0033] It should be noted that the above application scenarios are shown only to facilitate understanding of the spirit and principles of this application, and the implementation methods of this application are not limited in any way. On the contrary, the implementation methods of this application can be applied to any applicable scenario.

[0034] See Figure 2 , Figure 2The flowchart provided in this application embodiment illustrates a content creation method based on multi-agent collaboration. This method can be applied to a content creation system, such as the content creation system 100 described above. The content creation system 100 includes an interaction subsystem, an agent subsystem, a knowledge subsystem, and a model subsystem. The agent subsystem includes a search and recommendation agent cluster, a creation agent cluster, and a review agent cluster. The model subsystem includes a large language model corresponding to the agents in the agent subsystem. Specifically, the method may include the following steps: S201, builds a basic knowledge base including business knowledge through the knowledge subsystem.

[0035] In some embodiments, the process of constructing a basic knowledge base including business knowledge through a knowledge subsystem in step S201 can be implemented through the following steps: Step A01: Obtain internal business information through the knowledge subsystem.

[0036] Step A02 involves structurally integrating the acquired internal business information with external media information to obtain a basic knowledge base that includes business knowledge.

[0037] The method in this embodiment constructs an independent "knowledge subsystem" responsible for the deep integration and structuring of internal and external knowledge, providing high-purity, scenario-based knowledge fuel for upper-level intelligent agents, thereby ensuring the professionalism of the final output content, significantly improving the copywriting quality and generation efficiency of content creation, and facilitating the wider application of experiential sales.

[0038] In some embodiments, the process of constructing a basic knowledge base including business knowledge through the knowledge subsystem in step S201 specifically involves: constructing a basic knowledge base including business knowledge through the knowledge subsystem based on a preset period.

[0039] In some embodiments, the process of constructing a basic knowledge base including business knowledge through the knowledge subsystem in step S201 specifically involves: in response to receiving a knowledge construction instruction, constructing a basic knowledge base including business knowledge through the knowledge subsystem.

[0040] In some embodiments, a knowledge subsystem stores and manages business knowledge; and a knowledge subsystem implements vector retrieval and semantic association to provide knowledge supply to the various agents of the agent subsystem.

[0041] In practical implementation, the knowledge subsystem serves as the central hub for structured business knowledge; it is used to store and manage business knowledge, support vector retrieval and semantic association, and provide knowledge supply for various intelligent agents.

[0042] S202 receives the user's creative request information through the interaction subsystem.

[0043] In practice, the interaction subsystem receives the creation requirements information input by the user. The creation requirements information includes the content type and natural language instructions.

[0044] In some embodiments, content types include short video text, text and image notes, and social media text.

[0045] For example, the interactive subsystem can receive user input on creative requirements, including content types such as "short video scripts" and natural language instructions such as: "Topic keywords: retirement community; core viewpoint: planning a retirement community + annuity insurance for a 30-year-old customer".

[0046] In some embodiments, the process of constructing a basic knowledge base including business knowledge through the knowledge subsystem can be implemented after receiving user input of creative requirement information through the interaction subsystem. That is, the user input of creative requirement information is received first through the interaction subsystem, and then the basic knowledge base including business knowledge is constructed through the knowledge subsystem.

[0047] S203 uses a cluster of search and push intelligent agents to identify the intent of the creation requirements and obtain the task intent information.

[0048] In practice, the search and recommendation intelligent agent cluster performs deep semantic understanding of user commands, identifies core keywords, and expands information according to business definitions as task intent information. This task intent information is used to provide the large model as basic information for subsequent content matching and search and recommendation.

[0049] In some embodiments, the search and recommendation intelligent agent cluster includes an intent understanding intelligent agent and a task decomposition intelligent agent; the creation requirement information includes content type information and natural language instruction information; the process in step S203, which involves performing intent recognition on the creation requirement information through the search and recommendation intelligent agent cluster to obtain task intent information, can be implemented through the following steps: Step B01: The intent-understanding agent performs deep semantic understanding of the natural language instruction information to obtain the core keywords.

[0050] Step B02 involves decomposing the task into an intelligent agent and constructing a structured intent description based on core keywords and content type information to obtain task intent information.

[0051] For example, task intent information can be: “1” Themes (“retirement communities”, “annuity insurance”) and implied conditions (“30-year-old customers”); 2) Automatically match text or materials. In some embodiments, the structured intent description is passed to the retrieval module and the creation agent cluster of the search and push agent cluster. This embodiment can transform vague user needs into structured tasks containing business semantics that can be accurately operated by machines. It is the first technical hurdle to overcome the problem of "high creation threshold" and realizes the transformation from "people searching for knowledge / functions" to "services understanding people".

[0052] In some embodiments, business semantics can be integrated into a key-value format based on employees' understanding of the company's business, where the key is the business scenario and the value is the business semantics. Specifically, this can be achieved by using an intent classification agent based on the Qwen-Plus large model to find specific business semantics as the standard definition of the business scenario.

[0053] In some embodiments, the retrieval module of the search and recommendation agent cluster is a content recommendation agent.

[0054] S204, retrieves knowledge context information that matches the task intent information from the basic knowledge base through a cluster of search and push intelligent agents.

[0055] In some embodiments, the process of retrieving knowledge context information matching the task intent information from the basic knowledge base through the search and push agent cluster in step S204 can be implemented through the following steps: Step C01: Obtain the structure fields of task intent information through the content search and push intelligent agent.

[0056] Step C02: Select search results from the basic knowledge base that meet the preset relevance conditions of the structure fields.

[0057] Step C03: Integrate the retrieval results into a structured knowledge context, which serves as the knowledge context information to match the task intent information.

[0058] In practice, the content creation system retrieves highly relevant product terms, introductions to senior living communities, sales pitches for the corresponding age groups, compliance cases, and related image materials from the vectorized knowledge base based on the task intent information; it then packages the search results (such as knowledge fragments, materials, and data) into a "knowledge context" and outputs it to the creation intelligence agent cluster and the review intelligence agent cluster.

[0059] The existence of the knowledge subsystem means that the creation and review process no longer relies on general knowledge, but is built on a verified internal enterprise knowledge graph, achieving deep coupling between business and AI.

[0060] In some embodiments, product terms may be obtained from the product library included in the base knowledge base; compliance cases may be obtained from the case library included in the base knowledge base.

[0061] In some embodiments, users can select from recommended content and search the knowledge base, and can also customize materials to obtain external information and complete the input of custom materials.

[0062] S205 generates initial draft text information through a cluster of creative intelligent agents based on knowledge context information and task intent information.

[0063] In practice, the creative agent cluster acts like a "chief writer," calling upon the LLM of the model subsystem (either DeepSeek-R1 or Qwen-Plus) and combining knowledge context information and task intent information to generate draft copy that conforms to the business scenario, uses accurate professional terminology, and has an appropriate style. For example, annuity insurance benefit data and the real-world advantages of senior living communities can be naturally integrated into an engaging short video script.

[0064] In some embodiments, knowledge context information includes: data definitions, narrative logic, material links, and reference cases.

[0065] In practice, the knowledge context information can include accurate data definitions, compliant narrative logic, available material links, and reference cases.

[0066] In some embodiments, the tone and word count of the copy can be freely defined to achieve stylized content creation.

[0067] The method in this embodiment, through the cooperation of the search and push intelligent agent cluster and the creation intelligent agent cluster, enables efficient access to the underlying model capabilities of the knowledge subsystem information and model subsystem, performs material processing and generates copy that meets business requirements, realizes the automation and businessization of the creation process, and significantly improves the copy quality and generation efficiency of content creation.

[0068] S206, the intelligent review agent cluster performs compliance correction on the initial draft copy information to obtain the optimized content result.

[0069] In practice, the auditing intelligent agent cluster can simultaneously call the compliance audit LLM (or rule engine) of the model subsystem and re-associate the compliance language library and risk term library in the knowledge subsystem to conduct a double check on the content of the initial draft copy information: first, risk review (such as whether there is exaggeration, promise, or misleading), and second, calibration (such as whether the revenue data and product name are completely accurate).

[0070] In some embodiments, a cluster of intelligent review agents performs compliance corrections on the initial draft text information, outputting instructions such as "Pass - No Risk," "Low Risk," and "High Risk" to prompt users regarding the compliance of their created content, the level of content violation risk, and prohibited words. It also automatically corrects inconsistencies in data usage within the generated content, highlighting these inconsistencies. This embodiment is a key technology for addressing the pain point of "difficulty in ensuring compliance," transforming traditional, delayed, and manual compliance review into a real-time, automated "safety barrier" seamlessly integrated with the creation process, fundamentally controlling the risks of large-scale content production.

[0071] In some embodiments, the review agent cluster scans the initial draft copy information generated by the creation agent cluster; during the scanning process, the compliance rules of the knowledge subsystem are invoked to verify the scanned information: check for sensitive words and their corresponding risk levels; check whether the data conforms to the group's external standards; at the same time, the dedicated review model of the model subsystem is invoked to identify risks in the scanned information: determine whether there are absolute or promise-like terms, and whether they may lead to sales misrepresentation.

[0072] In some embodiments, the process of performing compliance correction on the initial draft text information through a cluster of auditing intelligent agents to obtain the optimized content result in step S206 can be achieved through the following steps: Step D01: The initial draft text information is checked for compliance and corrected for ambiguity by the review intelligent agent cluster to obtain the compliance correction result.

[0073] In some embodiments, the large language model includes a dedicated large review model; the basic knowledge base includes a content review sub-knowledge base; the process in step D01, which involves performing compliance checks and clarification corrections on the initial draft text information through a cluster of review intelligent agents to obtain compliance correction results, can be implemented through the following steps: Step E01 involves calling a dedicated large-scale review model through the review intelligence agent cluster to conduct a risk review of the initial draft text information and obtain the first compliance correction result.

[0074] Step E02: Through the review intelligent agent cluster, based on the compliance context information obtained from the content review sub-knowledge base from the knowledge context information, the initial draft text information is reviewed for compliance, and a second compliance correction result is obtained.

[0075] In practice, the auditing intelligent agent cluster can conduct a compliance review of the initial draft text information based on the compliance context information obtained from the content auditing sub-knowledge base, and obtain a second compliance correction result.

[0076] In some embodiments, the content moderation sub-knowledge base includes compliance rule information.

[0077] In some embodiments, compliance rule information is derived from compliance language.

[0078] In some embodiments, the process of conducting compliance review of the initial draft copy information may specifically involve: conducting compliance review of the initial draft copy information based on the compliance rule information included in the content review sub-knowledge base through a cluster of review intelligent agents, and obtaining a second compliance correction result.

[0079] Step E03: Through the review intelligent agent cluster, based on the standard context information obtained from the content review sub-knowledge base from the knowledge context information, the initial draft text information is reviewed to obtain the third compliance correction result.

[0080] Step E04: Obtain the compliance correction result based on the first compliance correction result, the second compliance correction result, and the third compliance correction result.

[0081] Step D02: Based on the compliance correction results, fine-tune the content of the initial draft text information to generate the optimized content result.

[0082] The method in this embodiment performs compliance correction operations on the initial draft copy information through a cluster of intelligent review agents to obtain optimized content results. It provides an AI-driven risk management and data calibration mechanism embedded in the creation process, realizing real-time management of compliance risks and efficient correction of data standards, thereby improving the copy quality and generation efficiency of content creation.

[0083] In some embodiments, the interaction subsystem sends the creation requirement information to the intelligent agent subsystem, so that the intelligent agent subsystem can generate the optimized content result through intent understanding, information search and recommendation, content creation, and content review.

[0084] In some embodiments, the creation requirement information includes first creation requirement information and second creation requirement information; the interaction subsystem receives the first creation requirement information selected by the user and sends it to the search and push intelligent agent cluster so that the search and push intelligent agent cluster can perform intent recognition; the first creation requirement information includes creation type and core viewpoint; The interactive subsystem receives the second creative requirement information input by the user and sends it to the creative intelligence agent cluster so that the creative intelligence agent cluster can control the copywriting creation to meet business requirements; the second creative requirement information includes copywriting style and copywriting length.

[0085] In some embodiments, the initial draft copy information is fine-tuned based on the compliance correction results to generate optimized content results. Specifically, the creative intelligent agent cluster fine-tunes the initial draft copy information based on the compliance correction results to generate optimized content results.

[0086] In some embodiments, it also includes: presenting the results of content optimization through an interactive subsystem.

[0087] In practice, the intelligent agent subsystem can package the final content, along with relevant materials recommended by the knowledge subsystem, and return it to the interaction subsystem.

[0088] In some embodiments, the interaction subsystem presents the user with a complete, compliant, ready-to-use, or fine-tuned content creation package in the form of rich text, structured scripts, or poster previews.

[0089] The content creation method based on multi-agent collaboration provided in this application is applied to a content creation system. The content creation system includes an interaction subsystem, an agent subsystem, a knowledge subsystem, and a model subsystem. The agent subsystem includes a search and recommendation agent cluster, a creation agent cluster, and a review agent cluster. The model subsystem includes a large language model corresponding to the agents in the agent subsystem. The method includes: constructing a basic knowledge base including business knowledge through the knowledge subsystem; receiving user-inputted creation request information through the interaction subsystem; performing intent recognition on the creation request information through the search and recommendation agent cluster to obtain task intent information; retrieving knowledge context information matching the task intent information from the basic knowledge base through the search and recommendation agent cluster; generating initial draft text information through the creation agent cluster based on the knowledge context information and task intent information; and performing compliance correction operations on the initial draft text information through the review agent cluster to obtain optimized content results. This method achieves precision and intelligence in the content creation process, significantly improving the quality and efficiency of content creation.

[0090] Based on the same technical concept, this application also provides a content creation system based on multi-agent collaboration. The principle of this content creation system based on multi-agent collaboration is similar to that of the above-mentioned content creation method based on multi-agent collaboration. Therefore, the implementation of the content creation system based on multi-agent collaboration can refer to the implementation of the content creation method based on multi-agent collaboration, and the repeated parts will not be described again.

[0091] Figure 3 This is a schematic diagram of a content creation system based on multi-agent collaboration, provided as an embodiment of this application. The content creation system 300 based on multi-agent collaboration, as shown... Figure 3 As shown, it includes: an interaction subsystem 301, an intelligent agent subsystem 302, a knowledge subsystem 303, and a model subsystem 304; the intelligent agent subsystem 302 includes a search and recommendation intelligent agent cluster 3021, a creation intelligent agent cluster 3022, and a review intelligent agent cluster 3023; the model subsystem 304 includes a large language model corresponding to the intelligent agents of the intelligent agent subsystem 302. Knowledge Subsystem 303 is used to build a basic knowledge base that includes business knowledge. Interaction subsystem 301 is used to receive user input of creative request information; The search and push intelligent agent cluster 3021 is used for: performing intent recognition on creation demand information to obtain task intent information; and retrieving knowledge context information that matches the task intent information from the basic knowledge base. The creative intelligent agent cluster 3022 is used to generate initial draft copy information based on knowledge context information and task intent information; The review agent cluster 3023 is used to perform compliance corrections on the initial draft of the text and obtain the optimized content results.

[0092] In the embodiments of this application, each intelligent agent cluster may include one or more intelligent agents.

[0093] An intelligent agent cluster includes multiple intelligent agents, which can provide parallel processing capabilities for content creation, improving the quality of copywriting and the efficiency of content generation.

[0094] In one possible implementation, the knowledge subsystem 303 is specifically used for: Obtain internal business information; The acquired internal business information and external media information are structurally integrated to obtain a basic knowledge base that includes business knowledge.

[0095] In one possible implementation, the search and push intelligent agent cluster 3021 includes an intent understanding intelligent agent and a task decomposition intelligent agent; the creation requirement information includes content type information and natural language instruction information; the search and push intelligent agent cluster 3021 is specifically used for: By performing deep semantic understanding of natural language instructions through an intent-understanding intelligent agent, core keywords can be obtained. By decomposing the task into an intelligent agent, a structured intent description is constructed based on core keywords and content type information to obtain task intent information.

[0096] In one possible implementation, the search and recommendation agent cluster 3021 includes content search and recommendation agents; the search and recommendation agent cluster 3021 is specifically used for: The structured fields of task intent information are obtained through the content search and recommendation intelligent agent; Select search results from the basic knowledge base that meet the preset relevance conditions of the structure fields; The search results are integrated into a structured knowledge context, which serves as the knowledge context information that matches the task intent information.

[0097] In one possible implementation, the audit agent cluster 3023 is specifically used for: The initial draft of the text was checked for compliance and its wording was corrected to obtain compliance correction results; Based on the compliance correction results, the initial draft of the text was fine-tuned to generate optimized content.

[0098] In one possible implementation, the large language model includes a dedicated content moderation model; the basic knowledge base includes a content moderation sub-knowledge base; and the moderation agent cluster 3023 is specifically used for: A dedicated review model is invoked to conduct a risk assessment of the initial draft of the document, resulting in the first compliance correction result. Based on compliance context information obtained from the content review sub-knowledge base from the knowledge context information, the initial draft text information is reviewed for compliance, and a second compliance correction result is obtained. Based on the standard context information obtained from the content review sub-knowledge base from the knowledge context information, the initial draft copy information is reviewed with a data standard to obtain the third compliance correction result; The compliance correction result is obtained based on the first compliance correction result, the second compliance correction result, and the third compliance correction result.

[0099] In one possible implementation, the interaction subsystem 301 is also used to: send creation requirement information to the intelligent agent subsystem 302, so that the intelligent agent subsystem 302 can generate optimized content results through intent understanding, information search and recommendation, content creation and content review.

[0100] In one possible implementation, the creation requirement information includes first creation requirement information and second creation requirement information; the interaction subsystem 301 is specifically used for: The system receives the user's first creative requirement information and sends it to the Search & Push Intelligent Agent Cluster 3021 so that the Search & Push Intelligent Agent Cluster 3021 can perform intent recognition. The first creative requirement information includes the creative type and core viewpoint. The system receives the second creative requirement information input by the user and sends it to the creative intelligence agent cluster 3022 so that the creative intelligence agent cluster 3022 can control the copywriting creation to meet the business requirements; the second creative requirement information includes the copywriting style and the copywriting length.

[0101] In one possible implementation, the interaction subsystem 301 is also used for: presenting the results of optimizing the content.

[0102] In some embodiments of this application, the content creation system includes an interaction subsystem, an intelligent agent subsystem, a knowledge subsystem, and a model subsystem, which can be respectively the interaction layer, the intelligent agent layer, the knowledge layer, and the model layer.

[0103] In one embodiment of this application, the content creation system's interaction layer, intelligent agent layer, knowledge layer, and model layer adopt a layered and decoupled cloud-native architecture, deployed on a cloud server cluster. The executing entity is the user (field agents and all office staff), who accesses the system through the interaction layer (Web / App / API). The core of the system is the intelligent agent layer, a multi-agent collaborative engine responsible for receiving, decomposing, and processing user requests. The knowledge layer, as the system's "brain," provides accurate and structured business knowledge support to the intelligent agents. The model layer, as the "power source," provides basic AI capabilities.

[0104] In one embodiment of this application, the interaction logic of the content creation system includes: user requests entering from the interaction layer, intent recognition being performed by the search and recommendation agent in the intelligent agent layer using qwen-plus as an LLM model, and relevant content being retrieved from the knowledge layer; the creation agent in the intelligent agent layer generating a draft based on the search results and model layer capabilities using deeppsik; and the review agent in the intelligent agent layer performing compliance checks and clarification on the draft using kimi-k2-instruct as an LLM model, and returning the results to the interaction layer for presentation to the user. Throughout the process, each layer exchanges data efficiently and systematically through defined APIs.

[0105] In one embodiment of this application, the business process of the content creation system follows the main line of "demand input → intelligent parsing and knowledge matching → content creation → intelligent review → result output". The key steps of the content creation system's business process are described in detail below, in conjunction with the data flow: Step F01: Requirement input and intent understanding.

[0106] Data flow: Interaction layer → Intelligent agent layer.

[0107] Input source: The interaction layer receives the content type of user input ("short video copy", "text and image notes", "poster creation") and natural language instructions (such as the topic keyword: retirement community; expressing the core viewpoint "30-year-old customer planning retirement community + annuity insurance").

[0108] 1) Themes ("retirement community", "annuity insurance") and implied conditions ("30-year-old customers");

[0109] 2) Automatically match text or materials.

[0110] Execution Entity and Content: Search and Recommendation Agent execution. It performs deep semantic understanding of user commands, identifies core keywords (the specific information table contains branch information of some target companies, including institutions, branches, and key introductions or concepts of some branches), and expands the information according to business definitions, providing it to the large model as basic information for subsequent content matching and search recommendations.

[0111] Output Objects and Functions: The structured intent description is passed to its own retrieval module and creation agent. This step transforms vague user needs into structured tasks containing business semantics that machines can precisely operate on. It is the first technical hurdle to overcome the "high barrier to entry for creation," realizing the shift from "people searching for knowledge / functions" to "services understanding people." The business semantics need to be distilled into a key-value format based on staff's understanding of the company's business, where the key is the business scenario and the value is the business scenario. The specific implementation requires an intent classification agent based on the Qwen-Plus large model to find specific business semantics as the standard definition of the business scenario.

[0112] Step F02: Precise knowledge retrieval and recommendation.

[0113] Data flow: Intelligent agent layer → Knowledge layer.

[0114] Input source: Structured intent description provided by SouTui Agent.

[0115] Execution Entity and Content: Knowledge Layer Execution. Based on intent, the system retrieves highly relevant product terms, introductions to senior living communities, sales scripts for corresponding age groups, compliance cases, and related image materials from the vectorized knowledge base.

[0116] Output Objects and Functions: The search results (knowledge fragments, materials, data) are packaged into a "knowledge context" and output to the creation agent and review agent. This step is crucial for solving the problem of "difficulty in finding standard business content" and ensuring "professionalism." The existence of the knowledge layer means that the creation and review processes no longer rely on general knowledge, but are built on a validated internal enterprise knowledge graph, achieving deep coupling between business and AI. This step not only allows users to select from recommended content and search the knowledge base, but also allows them to customize materials to obtain external information; custom material input can be completed simply by copying and pasting.

[0117] Step F03: Professional content generation.

[0118] Data flow: Intelligent agent layer.

[0119] Input sources: the task description transmitted by the Search Agent, and the "knowledge context" provided by the knowledge layer.

[0120] Execution Entity and Content: Execution by the Creation Agent. The Creation Agent acts as the "chief writer," invoking the LLM model at the model layer (optionally LLMdeepseek-R1 or Qwen-Plus) and, combined with the knowledge context, generating a draft copy that conforms to the business scenario, uses accurate professional terminology, and has an appropriate style. For example, it can naturally integrate annuity insurance return data and the real-world advantages of senior living communities into an engaging short video script.

[0121] Stylization allows for free definition of the copywriting tone and word count.

[0122] Output Object and Function: The generated preliminary content is output to the review Agent. In this step, the creation Agent is not an isolated text generator, but a coordinator that can "understand the task, utilize knowledge, and invoke the model." It ensures that the final output is business-driven and knowledge-enhanced, rather than just grammatically correct text.

[0123] This step can provide two types of large model generation results: 1) content generated after deep thinking with Deepseek R1; 2) results generated quickly with Deepseek V3.

[0124] Step F04: Real-time compliance audit and correction.

[0125] Data flow: Intelligent agent layer.

[0126] Input source: Initial content generated by the creation agent.

[0127] Execution Entity and Content: Execution by the Audit Agent. The Audit Agent calls the dedicated audit LLM (or rule engine) in the model layer and re-links it with the compliance terminology library and risk terminology library in the knowledge layer to perform a double check on the content: first, risk review (such as whether there is exaggeration, promise, or misleading information), and second, calibration (such as whether the revenue data and product names are completely accurate).

[0128] Output Objects and Functions: This function outputs "Pass - No Risk," "Low Risk," and "High Risk" prompts to users regarding the compliance of their content, the level of content violation risk, and prohibited keywords. It also automatically corrects inconsistencies in data usage within the generated content and highlights these inconsistencies. This step is a key technology for addressing the pain point of "difficulty in ensuring compliance." It revolutionizes traditional, delayed, and manual compliance review into a real-time, automated "safety barrier" seamlessly integrated with the creation process, fundamentally controlling the risks of large-scale content production.

[0129] Step F05: Result Delivery and Presentation.

[0130] Data flow: Agent layer → Interaction layer.

[0131] Input source: Content that has been finally approved by the auditing agent.

[0132] Execution subject and content: The intelligent agent layer packages the final content, along with relevant materials recommended from the knowledge layer, and returns it to the interaction layer.

[0133] Output and Function: The interactive layer presents users with a complete, compliant, and ready-to-use or fine-tuned content creation package in the form of rich text, structured scripts, or poster previews. This concludes a complete, closed-loop, and automated creation process from idea to compliant finished product.

[0134] The content creation system in this embodiment, through a clear four-layer architecture and a data flow centered on multi-agent collaboration, deeply integrates the fragmented knowledge, creation, and review processes into an organic whole. It is not only a tool, but also a "digital content colleague" with business awareness, continuous learning, and real-time risk control capabilities. It is an efficient path to solve current industry technical pain points and achieve business empowerment goals.

[0135] For example, the content creation system will be further explained by taking the social media promotion content of agent Xiao Li for a middle-aged client's "retirement community + annuity insurance" combination plan.

[0136] The business scenario and needs for this social media promotion can be specifically described as follows: Agent Li is following up with a 45-year-old high-net-worth client who is concerned about future retirement planning. To enhance client engagement and convey professional value, Li plans to release a short video on his personal video account introducing the "advantages of the linkage between retirement communities and smart elderly care." However, he faces the following difficulties: he is unsure how to accurately and vividly introduce the specific matching advantages of elderly care service products with communities such as XX Garden, and the quality and efficiency of content creation are not high.

[0137] Xiao Li can implement the following process using a content creation system based on multi-agent collaboration.

[0138] Figure 4 This is a schematic diagram of the interface of an interactive subsystem for implementing a content creation method based on multi-agent collaboration, provided in an embodiment of this application.

[0139] Step G01: Input your requirements through the interaction layer.

[0140] In practice, the interactive layer receives the user's input of creative requirements.

[0141] Xiao Li opened the "IP Content Assistant" APP, and in the... Figure 4 In the content creation interface shown, select the "Short Video Script" template and enter your creation requirements. For example, creation requirements include:

[0142] Natural language keywords: "retirement community"; Key takeaway: "Short video scripts designed for customers around 45 years old, introducing how senior living communities and smart healthcare products can be integrated and planned, should highlight the technological aspects and quality of senior care, with a friendly and professional style." Copywriting style: Professional; Copy length: 159 words.

[0143] Step G02: Based on the cooperation of the agent layer and the knowledge layer, perform intent understanding and knowledge retrieval.

[0144] Search Agent's analysis commands identified the following: "Creation type: short video copywriting", "Target customer: 45 years old", "Core theme: senior living community + annuity insurance linkage", "Core selling points: certain returns, quality retirement", and "Style: friendly and professional".

[0145] Subsequently, the Search Agent initiated a precise query at the knowledge layer: Search for relevant materials on annuity insurance products for elderly care services and knowledge bases for elderly care communities such as "XX Garden," and retrieve graphic and textual materials and introductory texts on community environment, service features, eligibility for admission, etc.

[0146] Search the "Compliance Scripts Library" to retrieve examples of compliant and legal statements regarding the Video Account platform.

[0147] Search the "Hot-Selling Stories and Material Case Studies" database to find similar customer stories (after anonymization) as creative references.

[0148] The knowledge layer integrates the above information into a structured "knowledge context package," which includes accurate data definitions, compliant narrative logic, available material links, and reference cases.

[0149] Step G03: Generate specialized content based on the intelligent agent layer.

[0150] The creation agent receives the task instructions and the "knowledge context packet".

[0151] The agent invokes the Large Language Model (LLM) at the model layer and embeds a "knowledge context package" as core instructions and constraints to generate preliminary copy. For example, it might generate the following script snippet: Automatically generated title: "Smart Elderly Care: The Technological Sense of Security with One-Click Call"; Preview of generated text: "Did you know? Our senior living community boasts such cutting-edge technology! 82-year-old Aunt Guo carries an IoT card; in emergencies, a simple press—a 3-second response, three-way coordination, and simultaneous arrival of doctors, concierges, and security personnel! Behind this is our company's self-developed smart operation and maintenance system, an IoT platform supported by 39 patented technologies, monitoring the operation of over 3000 devices in real time. A smart concierge is online 24 / 7: video consultations resolve health concerns, intelligent meal ordering recommends nutritious recipes, and voice commands are all it takes to schedule outings. Data dashboards monitor indoor temperature and humidity in real time, automatically adjusting air quality, ensuring technological warmth is hidden in every detail. For families aged 45 and above, we've created not just a senior living community, but a safety net woven with technology—because a quality retirement should be this comfortable."

[0152] In some embodiments of this application, the content creation system can perform operations such as previewing the generated text, copying the text, saving the work, and re-editing the text.

[0153] Step G04: Based on the intelligent agent layer, conduct real-time compliance audits and corrections.

[0154] The review agent can immediately scan the initial draft generated by the creation agent.

[0155] It first calls the compliance rules of the knowledge layer for verification: checking for sensitive words and their corresponding risk levels; and whether the data conforms to the group's external standards.

[0156] At the same time, it calls the dedicated audit model of the model layer to identify risks: determine whether there are "absolute" or "promised" terms, and whether they may lead to sales misrepresentation.

[0157] If the review agent discovers that the initial draft contains false advertising, such as "Our senior living community is the best in the world, with over 100 parks operating globally," it will immediately alert and correct the content: "Our senior living community is a leading senior living community in China, with 47 projects in 37 cities nationwide, operating 27 communities in 24 cities, and over 20,000 residents," highlighting the correction.

[0158] Based on the correction suggestions, the creative agent quickly fine-tunes the copy.

[0159] In some embodiments of this application, the content creation system can display the results of compliance review. The review results include: risk level, prohibited words, and violation category.

[0160] In some embodiments of this application, the content creation system can also display proofread text.

[0161] In some embodiments of this application, the content creation system can also review the proofread text again.

[0162] In some embodiments of this application, the content creation system can also copy the proofread text.

[0163] Step G05: Deliver and present results based on the interaction layer.

[0164] Finally, a complete short video script that has passed review, is accurate in data, complies with wording regulations, and meets styling requirements is presented to Xiao Li. Xiao Li can use the script directly, or make personalized adjustments to it before recording a spoken version and publishing it on the platform or directly. The content creation system can simultaneously record all knowledge references and review logs for this creation process, for future reference.

[0165] This embodiment, through multi-agent collaboration (search and recommendation, creation, and review) and deep support from the knowledge layer, transforms the agent's vague creative needs into a precise, professional, compliant, and readily available content product. It not only solves Xiao Li's business pain points of "not knowing how to speak," "not being able to speak accurately," and "being afraid of saying the wrong thing," but also transforms content creation from a high-threshold, high-risk personal skill into a standardized, efficient, and controlled enterprise-level productivity process, effectively improving the quality of content copy and the efficiency of content generation.

[0166] After introducing the content creation method and system based on multi-agent collaboration according to exemplary embodiments of this application, the electronic device according to another exemplary embodiment of this application will be introduced next.

[0167] The following reference Figure 5 This application describes an electronic device 500 implemented according to this embodiment. Figure 5 The electronic device 500 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0168] like Figure 5 As shown, the electronic device 500 is presented in the form of a general-purpose electronic device. The components of the electronic device 500 may include, but are not limited to: at least one processor 501, at least one memory 502, and a bus 503 connecting different system components (including memory 502 and processor 501).

[0169] Bus 503 represents one or more of several bus structures, including a memory bus or memory controller, peripheral bus, processor, or local bus using any of the various bus structures.

[0170] The memory 502 may include a readable medium in the form of volatile memory, such as random access memory (RAM) 5021 and / or cache memory 5022, and may further include read-only memory (ROM) 5023.

[0171] The memory 502 may also include a program / utility 5025 having a set (at least one) of program modules 5024, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0172] Electronic device 500 can also communicate with one or more external devices 504 (e.g., keyboard, pointing device, etc.), and with one or more devices that enable a user to interact with electronic device 500, and / or with any device that enables electronic device 500 to communicate with one or more other electronic devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 505. Furthermore, electronic device 500 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 506. As shown, network adapter 506 communicates with other modules used in electronic device 500 via bus 503. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 500, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0173] In an exemplary embodiment, a storage medium is also provided, which, when executed by a processor of an electronic device, enables the electronic device to perform the aforementioned content creation method based on multi-agent collaboration. Optionally, the storage medium may be a non-transitory computer-readable storage medium, such as a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device.

[0174] In an exemplary embodiment, the electronic device of this application may include at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores a computer program that can be executed by the at least one processor, and when the computer program is executed by the at least one processor, it can cause the at least one processor to perform the steps of any of the multi-agent collaborative content creation methods provided in the embodiments of this application.

[0175] In an exemplary embodiment, a computer program product is also provided, which, when executed by an electronic device, enables the electronic device to implement any of the exemplary methods provided in this application.

[0176] Furthermore, computer program products may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, RAM, ROM, erasable programmable read-only memory (EPROM), flash memory, optical fiber, compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0177] The program product for content creation based on multi-agent collaboration in this application embodiment can be a CD-ROM and include program code, and can run on a computing device. However, the program product of this application is not limited to this. In this document, the readable storage medium can be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0178] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying readable program code. This propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0179] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, radio frequency (RF), or any suitable combination thereof.

[0180] Program code for performing the operations of this application can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, such as a Local Area Network (LAN) or a Wide Area Network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0181] It should be noted that although several units or sub-units of the device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units.

[0182] Furthermore, although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0183] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0184] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0185] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0186] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0187] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0188] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A content creation system based on multi-agent collaboration, characterized in that, include: The system comprises an interaction subsystem, an intelligent agent subsystem, a knowledge subsystem, and a model subsystem; the intelligent agent subsystem includes a search and recommendation intelligent agent cluster, a creation intelligent agent cluster, and a review intelligent agent cluster; the model subsystem includes a large language model corresponding to the intelligent agents in the intelligent agent subsystem. The knowledge subsystem is used to build a basic knowledge base that includes business knowledge; The interactive subsystem is used to receive user input of creative requirements information; The search and recommendation intelligent agent cluster is used to perform intent recognition on the creation demand information to obtain task intent information, and to retrieve knowledge context information that matches the task intent information from the basic knowledge base; The creative intelligent agent cluster is used to generate initial draft text information based on the knowledge context information and the task intent information; The intelligent review agent cluster is used to perform compliance correction operations on the initial draft text information to obtain optimized content results.

2. The system as described in claim 1, characterized in that, The knowledge subsystem is specifically used for: Obtain internal business information; The acquired internal business information and external media information are structurally integrated to obtain the basic knowledge base, which includes business knowledge.

3. The system as described in claim 1, characterized in that, The search and recommendation intelligent agent cluster includes an intent understanding intelligent agent and a task decomposition intelligent agent; the creation requirement information includes content type information and natural language instruction information; the search and recommendation intelligent agent cluster is specifically used for: The intent-understanding agent performs deep semantic understanding of the natural language instruction information to obtain core keywords. By decomposing the task into an intelligent agent, a structured intent description is constructed based on the core keywords and the content type information to obtain task intent information.

4. The system as described in claim 1, characterized in that, The search and recommendation intelligent agent cluster includes content search and recommendation intelligent agents; the search and recommendation intelligent agent cluster is specifically used for: The content search and recommendation agent obtains the structure fields of the task intent information; Select retrieval results from the basic knowledge base that satisfy the preset relevance conditions of the structure field; The search results are integrated into a structured knowledge context, which serves as the knowledge context information that matches the task intent information.

5. The system as described in claim 1, characterized in that, The auditing intelligent agent cluster is specifically used for: The initial draft of the text was subjected to a compliance check and clarification, resulting in a compliance correction. Based on the compliance correction results, the initial draft text information is fine-tuned to generate the optimized content results.

6. The system as described in claim 5, characterized in that, The large language model includes a dedicated content moderation model; the basic knowledge base includes a content moderation sub-knowledge base; the moderation intelligent agent cluster is specifically used for: The dedicated review model is invoked to conduct a risk review of the initial draft text information, and the first compliance correction result is obtained. Based on the compliance context information obtained from the content review sub-knowledge base from the knowledge context information, the initial draft text information is subjected to compliance review to obtain a second compliance correction result; Based on the standard context information obtained from the content review sub-knowledge base from the knowledge context information, the initial draft text information is reviewed using data standards to obtain a third compliance correction result. The compliance correction result is obtained based on the first compliance correction result, the second compliance correction result, and the third compliance correction result.

7. The system as described in claim 1, characterized in that, The interaction subsystem is further configured to: send the creation requirement information to the intelligent agent subsystem, so that the intelligent agent subsystem can generate the optimized content result through intent understanding, information search and recommendation, content creation and content review.

8. The system as described in claim 1, characterized in that, The creation requirement information includes first creation requirement information and second creation requirement information; the interaction subsystem is specifically used for: The system receives the first creative requirement information selected by the user and sends it to the search and push intelligent agent cluster so that the search and push intelligent agent cluster can perform intent recognition; the first creative requirement information includes creative type and core viewpoint. The system receives the second creative requirement information input by the user and sends it to the creative intelligence agent cluster so that the creative intelligence agent cluster can control the copywriting creation to meet the business requirements; the second creative requirement information includes copywriting style and copywriting length.

9. The system as described in claim 1, characterized in that, The interactive subsystem is also used for: presenting the results of the optimization content.

10. A content creation method based on multi-agent collaboration, characterized in that, It is applied to a content creation system; the content creation system includes an interaction subsystem, an intelligent agent subsystem, a knowledge subsystem, and a model subsystem; the intelligent agent subsystem includes a search and recommendation intelligent agent cluster, a creation intelligent agent cluster, and a review intelligent agent cluster; The model subsystem includes a large language model corresponding to the agents in the agent subsystem; the method includes: A basic knowledge base, including business knowledge, is constructed through the knowledge subsystem. The interactive subsystem receives user input regarding creative requirements. The search and recommendation intelligent agent cluster performs intent recognition on the creation demand information to obtain task intent information; The search and push intelligent agent cluster retrieves knowledge context information that matches the task intent information from the basic knowledge base; The creative intelligent agent cluster generates initial draft text information based on the knowledge context information and the task intent information; The intelligent review agent cluster performs compliance corrections on the initial draft text information to obtain optimized content results.

Images