Demand processing method and system based on ontology knowledge driving and agent cooperation

Abstract

This invention discloses a requirement processing method and system based on ontology knowledge-driven and agent-based collaboration. The method includes: acquiring user requests and performing intent recognition based on a unified knowledge center to determine the target task type and business category; activating the target workflow from multiple pre-set scenario-based state machine workflows according to the target task type; the main agent scheduling the corresponding business sub-agents, retrieving relevant knowledge from the unified knowledge center according to the knowledge query rules of the current task node, and generating structured node execution results through reasoning analysis using a large language model; obtaining the node execution results through the main agent, determining whether the state transition conditions are met based on user feedback, driving the workflow transition until all nodes are executed, and then summarizing to generate standardized requirement processing results. This invention solves the problems of low automation and poor processing consistency caused by knowledge dispersion, complex processes, and inefficient collaboration in traditional requirement writing methods.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and software requirements engineering, specifically relating to a requirements processing method and system based on ontology knowledge-driven and intelligent agent collaboration. Background Technology

[0002] In the field of software requirements development, business personnel typically need to translate colloquial business requests into standardized requirements documents, which then undergo multiple rounds of clarification, current situation analysis, solution design, and approval before entering the development phase. Business rules, project status, requirement structures, and historical case studies are scattered across various heterogeneous systems, documents, and personal experience, lacking a unified knowledge center. This results in inefficient information retrieval and difficulties in knowledge reuse. The processing flows for different business scenarios vary significantly and are complex. Coupled with repeated communication involving multiple roles such as business, product, and technology, the collaboration cycle is long, manual operations are prone to missing key steps, and manually written requirements documents are difficult to guarantee in terms of formatting standards and content consistency.

[0003] Existing technical solutions are mostly based on single question-answering robots or fixed-form systems. Question-answering robots can typically only conduct single or shallow conversations, making it difficult to support complex development tasks that rely on multi-source knowledge and multi-round confirmation. While fixed-form systems can standardize input formats, they cannot dynamically recognize intent based on users' natural language descriptions, nor can they adaptively adjust information collection strategies according to contextual changes. Knowledge is scattered across different storage systems, lacking a unified retrieval mechanism, resulting in low information retrieval efficiency.

[0004] Furthermore, existing solutions typically employ only one executor, failing to dynamically schedule multiple specialized sub-agents to collaboratively complete complex tasks based on task type. Requirements documentation still heavily relies on manual compilation and template filling, unable to automatically generate structured requirements specifications based on analysis results. Therefore, there is an urgent need for a method that can uniformly manage knowledge, dynamically schedule agents, and automatically complete the entire requirements writing process. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a requirement processing method and system based on ontology knowledge-driven and agent-based collaboration. The core idea of ​​this invention is to abstract the requirement writing process into a programmable and reusable agent-based collaborative workflow. This is achieved by driving a state machine workflow through a unified knowledge center, and executing this workflow through hierarchical collaboration between the main agent and sub-agents, ultimately realizing the automation and intelligence of the requirement writing process.

[0006] A requirement processing method based on ontology knowledge-driven and agent-cooperative approaches includes the following steps:

[0007] Obtain user requests and perform intent recognition on the user requests based on the unified knowledge center to determine the target task type and the business category to which it belongs; Based on the target task type, a target workflow is activated from multiple pre-set scenario-based state machine workflows. Each node of the state machine workflow defines the knowledge query rules and state transition conditions between nodes. Different target task types correspond to different state machine workflows, and each workflow defines differentiated node execution rules and output rules. The main intelligent agent schedules the corresponding business sub-intelligent agents from the pre-configured business intelligent agent layer according to the current task node of the target workflow. The business sub-intelligent agents call the unified knowledge center to retrieve relevant knowledge according to the knowledge query rules of the current task node, and combine the large language model to perform reasoning analysis to generate structured node execution results. The main intelligent agent obtains the execution result of the node and combines it with user feedback to determine whether the state transition conditions of the current task node are met. If so, it drives the target workflow to migrate to the next node. After all task nodes of the target workflow have been executed, the final standardized requirements processing results are generated.

[0008] As one possible implementation, obtaining a user request involves performing intent recognition on the user request based on a unified knowledge center to determine the target task type and its business category, including the following steps: Extract keywords from user requests and match them with the pre-built business category library in the unified knowledge center; Calculate the semantic similarity between the user request and the descriptions of each category in the business category library; Based on a four-level hierarchical structure of business domain, business category, business scenario, and business subcategory, the business subcategory to which the user request belongs is determined by gradually converging from coarse to fine. The type of the target task is determined by identifying the trigger word in the user's request and determining the category of the request.

[0009] As one possible implementation, activating a target workflow from a set of multiple pre-defined scenario-based state machine workflows based on the target task type includes the following steps: Obtain the identifier of the target task type and map the identifier to a pre-stored workflow index table; The corresponding target workflow is uniquely matched and activated from the workflow index table. The target workflow is one of the following: product development optimization workflow, third-party system integration workflow, new business scenario workflow, special business activation workflow, or business consultation workflow.

[0010] As one possible implementation, the main intelligent agent schedules corresponding business sub-intelligent agents from the pre-configured business intelligent agent layer according to the current task node of the target workflow, including the following steps: The main intelligent agent selects a business sub-intelligent agent that matches the current business scenario from the pre-configured business intelligent agent layer based on the current task node determined by the target workflow. Construct a context package, the context including: user original input and supplementary information, identified task type and business category, key conclusions and confirmed slots in the session history, relevant knowledge retrieved from the unified knowledge center, and role definition and behavioral constraints. The context packet is passed to the business sub-agent for parsing and execution; The business sub-agent retrieves relevant knowledge from the unified knowledge center using at least one of the following methods: precise query, semantic retrieval, hybrid retrieval, or field retrieval, based on the knowledge query rules of the current task node.

[0011] As one possible implementation, obtaining the node execution result through the main intelligent agent and determining whether the state transition conditions of the current task node are met based on user feedback includes the following steps: Obtain structured data of the node execution results and make a judgment based on the user's feedback type on the node execution results; The user feedback types include at least: confirmation, modification, supplementation, and continuation; When the user feedback type is "confirmation", the state transition condition is determined to be met; When the user feedback type is "modify", "supplement" or "continue", the system updates or reverts to the relevant node according to preset rules and retrieves user feedback again.

[0012] As one possible implementation, the target workflow includes at least three stages: requirement clarification, requirement design, and requirement generation; the process of summarizing and generating the final standardized requirement processing result includes the following steps: Based on different target task types, assemble the functional chapters, business process chapters, and related system analysis chapters of the requirements document according to differentiated rules; The assembled chapter content is then filled back into the standard requirements template to generate the final standardized requirements processing result.

[0013] As one possible implementation, when the target workflow is a product development optimization workflow, the generation of structured node execution results includes the following steps: Extract the main business process as the analysis framework, and identify the process change nodes in the main business process by comparing them with the transformation points; Output a list of modification points, which shall include at least the object of change, the method of change, and the impact assessment information.

[0014] As one possible implementation, when the target workflow is a third-party system interface workflow, the generation of structured node execution results includes the following steps: Use the built-in standard process to identify the interaction points with third-party systems and output a list of interface points; Match the business semantics in the interface materials with the functional points in the requirement structure library to identify new functional points.

[0015] As one possible implementation, when the target workflow is a new business scenario workflow, the generation of structured node execution results includes the following steps: Based on the requirements structure library, determine the scope of functional levels that need to be fully restored; For missing information items, fill or mark them according to preset rules to generate a function list.

[0016] A requirement processing system based on ontology knowledge-driven and agent-cooperative approaches is provided to implement a requirement processing method based on ontology knowledge-driven and agent-cooperative approaches. The system includes: The intent recognition module is used to acquire user requests and perform intent recognition on the user requests based on the unified knowledge center to determine the target task type and the business category to which it belongs; The workflow activation module is used to activate a target workflow from a set of multiple scenario-based state machine workflows according to the target task type. Each node of the state machine workflow defines the knowledge query rules and state transition conditions between nodes. Different target task types correspond to different state machine workflows, and each workflow defines differentiated node execution rules and output rules. The sub-agent scheduling module is used to schedule corresponding business sub-agents from the pre-configured business agent layer according to the current task node of the target workflow through the main agent. The business sub-agents call the unified knowledge center to retrieve relevant knowledge according to the knowledge query rules of the current task node, and perform reasoning analysis in combination with the large language model to generate structured node execution results. The state judgment module is used to obtain the execution result of the node through the main intelligent agent and, in combination with user feedback, determine whether the state transition conditions of the current task node are met. The result generation module is used to drive the target workflow to the next node when the state transition conditions are met through the main intelligent agent; after all task nodes of the target workflow have been executed, the final standardized requirement processing result is generated.

[0017] The present invention has at least the following beneficial effects: This invention achieves layered decoupling and automated processing of complex tasks in requirement writing scenarios through a knowledge-driven and agent-collaborative approach. It unifies scattered business knowledge, project information, and requirement structures into a knowledge center, supporting accurate intent recognition and knowledge retrieval through a four-level category hierarchy and multiple retrieval methods. Furthermore, through a collaborative mechanism between the main agent and sub-agents, complex tasks are decomposed and scheduled to specialized business sub-agents for execution. Combined with state machine workflow standardization of processing steps in various scenarios, it ultimately generates standardized requirement processing results, comprehensively improving the efficiency and quality of requirement writing.

[0018] Compared to existing technologies that employ single agents or fixed pipelines for demand processing, the core differences of this invention are as follows: First, it adopts a master-slave hierarchical architecture. The master agent, as the central control node, is responsible for task decomposition and workflow state transition decisions. Multiple business sub-agents focus on specialized processing of specific scenarios in their respective independent execution environments. The two achieve loosely coupled collaboration through context packet passing, rather than the single agent or multi-model parallel collaboration mode in existing solutions. Second, it deeply binds the knowledge query rules defined by each node of the state machine workflow with the state transition conditions. This ensures that the unified knowledge center not only provides knowledge materials for demand generation but also serves as the decision basis for driving workflow state transitions. The knowledge query results of the current node are directly used to determine whether the transition requirements to the next node are met, thus forming a closed loop of knowledge-driven process decision-making. Third, it distinguishes user feedback types into four categories: confirmation, modification, supplementation, and continuation. Differentiated feedback processing rules are configured for different workflow stages, enabling the demand processing flow to dynamically adjust its path according to user intent, rather than mechanically advancing in a fixed order. The three differentiated technical features mentioned above work together to effectively overcome the technical shortcomings of existing solutions, such as a single processing logic and a lack of specificity when facing different types of needs.

[0019] This invention effectively improves the automation and consistency of requirement writing, enabling intelligent processing in stages such as requirement clarification, current situation analysis, solution generation, and document writing. This reduces repetitive manual labor and shortens the cycle from requirement submission to completion. Simultaneously, by standardizing processing steps in different scenarios through state machines and workflows, it reduces discrepancies in human judgment, ensuring consistent processing of similar requirements initiated by different users at different times. The layered architecture and modular capability design allow for low-cost integration of new business scenarios and capability units, and new business categories can be quickly deployed simply by supplementing existing knowledge. Through multi-round dialogues, candidate confirmation, and pending confirmation item marking, business personnel can naturally participate in the requirement development process, improving the human-machine collaboration experience.

[0020] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0021] Figure 1 is a flowchart illustrating the method of the present invention; Figure 2 is a functional block diagram of the system of the present invention. Detailed Implementation

[0022] To clearly illustrate the present invention and make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The technical solutions of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] Example 1: A requirement processing method based on ontology knowledge-driven and agent-cooperative approaches, as shown in Figure 1, includes the following steps: S100. Obtain user request, and perform intent recognition on the user request based on the unified knowledge center to determine the target task type and the business category to which it belongs; S200. According to the target task type, activate a target workflow from multiple preset scenario-based state machine workflows. Each node of the state machine workflow defines the knowledge query rules and state transition conditions between nodes. Different target task types correspond to different state machine workflows, and each workflow defines differentiated node execution rules and output rules. S300: The main intelligent agent schedules the corresponding business sub-intelligent agent from the pre-configured business intelligent agent layer according to the current task node of the target workflow. The business sub-intelligent agent calls the unified knowledge center to retrieve relevant knowledge according to the knowledge query rules of the current task node, and performs reasoning analysis in combination with the large language model to generate structured node execution results. S400. The main intelligent agent obtains the execution result of the node, and determines whether the state transition conditions of the current task node are met by combining the user feedback. If so, the target workflow is driven to migrate to the next node. S500. After all task nodes of the target workflow have been executed, the final standardized requirements processing results are generated.

[0024] S100. Obtain the user request, and perform intent recognition on the user request based on the unified knowledge center to determine the target task type and its business category, including the following steps: S1001: Receive the user's input of natural language requirements and related context information, including the user's identity, affiliated organization, conversation history, and collected slot information; In this embodiment of the invention, the user inputs business requests through a natural language interaction interface, and the system simultaneously obtains the user's identity identifier, the organization identifier, historical dialogue records, and confirmed slot data for the current session to form complete contextual information.

[0025] S1002: Extract keywords from the user request and match them with the pre-built business category library in the unified knowledge center; In this embodiment of the invention, the unified knowledge center is pre-built with a business category library, which is modeled using a four-level hierarchical structure of business domain, business category, business scenario, and business subcategory. After the system performs word segmentation on the user request, it extracts key business terms and matches them one by one with the keywords of each level of the business category library.

[0026] S1003: Calculate the semantic similarity between the user request and the descriptions of various categories in the business category library using a large language model; In this embodiment of the invention, for scenarios where keyword matching cannot uniquely determine the identity, a large language model is invoked to calculate the similarity score between the semantic vector of the user's request and the semantic vector of the category description text.

[0027] S1004: Based on the four-level hierarchical structure of business domain, business category, business scenario, and business sub-category, gradually converge from coarse to fine to determine the business category to which the user request belongs; In this embodiment of the invention, starting from the highest level of business domain, the candidate range is converged layer by layer downwards until the finest-grained business category unit is located, thereby achieving precise matching of user intent.

[0028] S1005: Determine the demand category based on the trigger word of the user request to determine the target task type; In this embodiment of the invention, a demand category library is pre-stored, including categories such as special service activation, product development (optimization and modification, third-party integration, and new scenarios), and business consultation. The system matches the trigger words in the user's request with the demand category library to determine the target task type. When multiple candidate results exist, the system outputs the candidate options to the user and requests confirmation.

[0029] S200. Based on the target task type, activate a target workflow from multiple pre-set scenario-based state machine workflows. Each node of the state machine workflow defines knowledge query rules and state transition conditions between nodes. Different target task types correspond to different state machine workflows, and each workflow defines differentiated node execution rules and output rules, including the following steps: S2001: Obtain the identifier of the target task type and map the identifier to a pre-stored workflow index table; In this embodiment of the invention, each target task type corresponds to a unique fixed identifier, and the workflow index table pre-stores the mapping relationship between each task type identifier and the corresponding workflow.

[0030] S2002: Uniquely match and activate the corresponding target workflow from the workflow index table; In this embodiment of the invention, the target workflow is one of the following: product development optimization workflow, third-party system integration workflow, new business scenario workflow, special business activation workflow, or business consultation workflow. Each workflow consists of multiple state nodes, and each state node defines input conditions, recognition rules, processing logic, output results, knowledge query, state transition, and confirmation rules.

[0031] S300. The main intelligent agent, based on the current task node of the target workflow, schedules the corresponding business sub-intelligent agent from the pre-configured business intelligent agent layer. The business sub-intelligent agent, based on the knowledge query rules of the current task node, calls the unified knowledge center to retrieve relevant knowledge, and performs reasoning analysis in conjunction with the large language model to generate structured node execution results, including the following steps: S3001: The main intelligent agent selects a business sub-intelligent agent that matches the current business scenario from the pre-configured business intelligent agent layer based on the current task node determined by the target workflow, and constructs a context package; In this embodiment of the invention, the master agent serves as the unified entry point and central control node for demand processing tasks, responsible for preparing the context package required for execution by the sub-agents. The context includes: the user's original input and supplementary information for the current round, the identified task type and business category, key conclusions and confirmed slots in the session history, relevant project information and demand structure retrieved from the unified knowledge center, workspace location, and role definitions and behavioral constraints.

[0032] S3002: Pass the context packet to the business sub-agent for parsing and execution; In this embodiment of the invention, the main intelligent agent selects a matching sub-intelligent agent from the business intelligent agent layer according to the task type, and passes the constructed context package to the sub-intelligent agent. The business intelligent agent layer includes sub-intelligent agents oriented towards different business scenarios, such as sub-intelligent agents for special business activation, product development and optimization, third-party system integration, new business scenario addition, and business consultation. Each sub-intelligent agent focuses on the processing logic of a specific scenario.

[0033] S3003: The business sub-agent calls the unified knowledge center to retrieve relevant knowledge according to the knowledge query rules of the current task node; In this embodiment of the invention, a unified knowledge center stores and manages project repositories, requirement structure repositories, business knowledge bases, business category repositories, requirement category repositories, and capability component templates. Knowledge retrieval employs at least one of the following methods: precise query, semantic retrieval, hybrid retrieval, or field retrieval. Precise query performs equality or range queries based on structured fields, suitable for structured data such as project repositories and requirement structure repositories; semantic retrieval performs similarity searches on unstructured documents based on vector embedding technology, suitable for business knowledge bases, policy documents, etc.; hybrid retrieval combines keyword retrieval, vector retrieval, and structured filtering conditions, comprehensively ranking and returning the most relevant knowledge entries; field retrieval specifies the returned fields, extracting only specific attributes from the knowledge entries.

[0034] S3004: The business sub-agent combines the large language model to perform reasoning analysis and generate structured node execution results; In this embodiment of the invention, the sub-agent, in an independent execution environment, combines the reasoning capabilities of a large language model to analyze and judge the retrieved knowledge, and organizes the analysis results into structured outputs, such as requirement plans, process change lists, function lists, requirement documents, etc.

[0035] S3005: When the target workflow is a product development optimization workflow, the generation of structured node execution results includes the following steps: Extract the main business process as the analysis framework, identify the process change nodes in the main business process by comparing them with the improvement points, and output a list of improvement points. The list of improvement points includes at least the object of change, the method of change, and the impact assessment information.

[0036] In this embodiment of the invention, the business sub-agent extracts the main business process under the target business category from the unified knowledge center as the analysis framework. Then, based on the user's proposed modification requests, it identifies the types of nodes in the main business process that need to be changed, including added nodes, modified nodes, and deleted nodes. For each changed node, the business sub-agent further analyzes its changed object, change method, and impact assessment information on other related modules, and finally summarizes and outputs a structured list of modification points.

[0037] S3006: When the target workflow is a third-party system interface workflow, generating structured node execution results includes the following steps: The built-in standard process is used to identify the interaction links with third-party systems and output a list of interface points; the business semantics in the interface materials are matched with the functional points in the requirement structure library to identify new functional points.

[0038] In this embodiment of the invention, the business sub-agent loads a built-in standard banking business process model, identifies the interaction links in the standard process that require data exchange with third-party systems, such as identity verification result query interfaces and payment status callback interfaces, and outputs a list of connection points including the name of the connection link, the name of the local system, the name of the third-party system, the interaction method, and the data content. Simultaneously, the business sub-agent performs semantic matching between the business semantics in the interface documents provided by the third-party systems and the functional points already stored in the requirement structure library in the unified knowledge center. Existing functional points are reused first, and functions that cannot be matched are added as a group of functional points according to the interface business capabilities, ultimately forming a complete set of connection functional requirements.

[0039] S3007: When the target workflow is a new business scenario workflow, the generation of structured node execution results includes the following steps: Based on the requirement structure library, determine the scope of functional levels that need to be fully restored; for missing information items, fill or mark them according to preset rules to generate a functional list.

[0040] In this embodiment of the invention, the business sub-agent first queries the requirement structure library of the unified knowledge center for the functional structure of the existing scenario most similar to the new business scenario. Based on this structure, it determines the scope of the first-level and second-level functions that need to be fully restored. For fields or functional items within this scope but whose specific content cannot be determined from user input and the knowledge base at present, the business sub-agent fills them in according to preset rules, such as using the same default configuration as the existing scenario, or marking them as "to be supplemented" fields for subsequent manual confirmation. After filling and marking are completed, the complete functional structure is organized into a functional list as the node execution result output.

[0041] S400. The main intelligent agent obtains the execution result of the node, and determines whether the state transition conditions of the current task node are met based on user feedback. If so, the target workflow is driven to migrate to the next node, including the following steps: S4001: Obtain the structured data of the execution result of the node; In this embodiment of the invention, the main agent receives the structured execution result returned by the sub-agent, and the structured execution result contains all the information required to be output by the current task node.

[0042] S4002: Determine the state transition conditions based on the user's feedback type on the node's execution result; In this embodiment of the invention, user feedback types include four types: confirmation, modification, supplementation, and continuation. Confirmation indicates that the user approves of the current output, the system determines that the state transition conditions are met, and transitions to the next node; modification indicates that the user requests to modify the current output, the system updates the result and re-enters the confirmation phase of the current node; supplementation indicates that the user adds new information, the system reverts to the relevant collection node for reprocessing; continuation indicates that the user agrees to continue processing but retains some items pending confirmation, the system marks the content to be supplemented and then continues.

[0043] S500. After all task nodes of the target workflow have been executed, the final standardized requirements processing results are generated, including the following steps: S5001: The target workflow sequentially passes through three stages: requirement clarification, requirement solution, and requirement generation. In this embodiment of the invention, the requirement clarification stage is used to identify the scope of user requirements, confirm business scenarios, query activation status, and collect necessary information; the requirement solution stage is used to analyze the current situation and gaps, generate solutions, conduct process analysis and functional decomposition; and the requirement generation stage is used to assemble the solutions into standard requirement documents, present them to users for confirmation, and submit them to the requirement management platform.

[0044] S5002: Based on different target task types, assemble the functional chapters, business process chapters, and related system analysis chapters of the requirements document according to differentiated rules; In this embodiment of the invention, for optimization and transformation tasks, the change type is marked for each functional point, and the transformation point is written as a comparison from before to after the transformation. If the process has not changed, it is clearly stated that the original process has not changed. For three-party interaction tasks, the functional chapters are supplemented with the three-party interaction objectives, and the business process uses the standard flowchart marked with the three-party interaction points. For new scenario tasks, the functional chapters are fully restored according to the requirement structure library to restore the first-level and second-level functions, and the business process is restored based on the structure library and the standard flowchart is displayed.

[0045] S5003: Fill the assembled chapter content back into the standard requirements template, check it before release, and then show it to the user for confirmation. In this embodiment of the invention, the standard requirements template includes sections such as requirements overview, business process, functional requirements, related system analysis, and risks and attachments. Pre-release checks include completeness checks, consistency checks, format checks, and marking of items awaiting confirmation.

[0046] S5004: After user confirmation, the requirements are submitted to the requirements management platform to generate the final standardized requirements processing results.

[0047] Example 2: A demand processing system based on ontology knowledge-driven and agent-cooperative operation, as shown in Figure 2, includes: The intent recognition module 100 is used to acquire user requests and perform intent recognition on the user requests based on the unified knowledge center in order to determine the target task type and the business category to which it belongs. Workflow activation module 200 is used to activate a target workflow from a plurality of preset scenario-based state machine workflows according to the target task type. Each node of the state machine workflow defines the knowledge query rules and state transition conditions between nodes. Different target task types correspond to different state machine workflows, and each workflow defines differentiated node execution rules and output rules. The sub-agent scheduling module 300 is used to schedule corresponding business sub-agents from the pre-configured business agent layer according to the current task node of the target workflow through the main agent. The business sub-agents call the unified knowledge center to retrieve relevant knowledge according to the knowledge query rules of the current task node, and perform reasoning analysis in combination with the large language model to generate structured node execution results. The state judgment module 400 is used to obtain the execution result of the node through the main intelligent agent and determine whether the state transition conditions of the current task node are met in combination with user feedback. The result generation module 500 is used to drive the target workflow to the next node when the state transition conditions are met through the main intelligent agent; after all task nodes of the target workflow have been executed, the final standardized requirement processing result is generated.

[0048] The above description of the embodiments is provided to enable those skilled in the art to understand and apply the present invention. It will be apparent to those skilled in the art that various modifications can be made to the above embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made to the present invention by those skilled in the art based on the disclosure thereof should be within the scope of protection of the present invention.

Claims

1. A requirement processing method based on ontology knowledge-driven and agent-based collaboration, characterized in that, Includes the following steps: Obtain user requests and perform intent recognition on the user requests based on the unified knowledge center to determine the target task type and the business category to which it belongs; Based on the target task type, a target workflow is activated from multiple pre-set scenario-based state machine workflows. Each node of the state machine workflow defines the knowledge query rules and state transition conditions between nodes. Different target task types correspond to different state machine workflows, and each workflow defines differentiated node execution rules and output rules. The main intelligent agent schedules the corresponding business sub-intelligent agents from the pre-configured business intelligent agent layer according to the current task node of the target workflow. The business sub-intelligent agents call the unified knowledge center to retrieve relevant knowledge according to the knowledge query rules of the current task node, and combine the large language model to perform reasoning analysis to generate structured node execution results. The main intelligent agent obtains the execution results of the nodes and, in conjunction with user feedback, determines whether the state transition conditions of the current task node are met. If so, it drives the target workflow to migrate to the next node. After all task nodes of the target workflow have been executed, the final standardized requirement processing results are generated.

2. The requirement processing method based on ontology knowledge-driven and agent-cooperative approaches according to claim 1, characterized in that, The process of obtaining user requests involves performing intent recognition on the user requests based on a unified knowledge center to determine the target task type and its business category, including the following steps: Extract keywords from user requests and match them with the pre-built business category library in the unified knowledge center; Calculate the semantic similarity between the user request and the descriptions of each category in the business category library; Based on a four-level hierarchical structure of business domain, business category, business scenario, and business subcategory, the business subcategory to which the user request belongs is determined by gradually converging from coarse to fine. The type of the target task is determined by identifying the trigger word in the user's request and determining the category of the request.

3. The requirement processing method based on ontology knowledge-driven and agent-cooperative approaches according to claim 1, characterized in that, Activating a target workflow from a set of pre-defined scenario-based state machine workflows based on the target task type includes the following steps: Obtain the identifier of the target task type and map the identifier to a pre-stored workflow index table; The corresponding target workflow is uniquely matched and activated from the workflow index table. The target workflow is one of the following: product development optimization workflow, third-party system integration workflow, new business scenario workflow, special business activation workflow, or business consultation workflow.

4. The requirement processing method based on ontology knowledge-driven and agent-cooperative approaches according to claim 1, characterized in that, The main intelligent agent schedules the corresponding business sub-intelligent agents from the pre-configured business intelligent agent layer according to the current task node of the target workflow, including the following steps: The main intelligent agent selects a business sub-intelligent agent that matches the current business scenario from the pre-configured business intelligent agent layer based on the current task node determined by the target workflow. Construct a context package, the context including: user original input and supplementary information, identified task type and business category, key conclusions and confirmed slots in the session history, relevant knowledge retrieved from the unified knowledge center, and role definition and behavioral constraints. The context packet is passed to the business sub-agent for parsing and execution; The business sub-agent retrieves relevant knowledge from the unified knowledge center using at least one of the following methods: precise query, semantic retrieval, hybrid retrieval, or field retrieval, based on the knowledge query rules of the current task node.

5. The requirement processing method based on ontology knowledge-driven and agent-cooperative approaches according to claim 1, characterized in that, The process of obtaining the node execution result through the main intelligent agent and determining whether the state transition conditions of the current task node are met by combining user feedback includes the following steps: The structured data of the node execution results is obtained and judged in combination with the user's feedback type on the node execution results; the user feedback type includes at least: confirmation, modification, supplementation and continuation; When the user feedback type is "confirmation", the state transition condition is determined to be met; When the user feedback type is "modify", "supplement" or "continue", update or roll back to the relevant node according to preset rules and re-obtain user feedback.

6. The requirement processing method based on ontology knowledge-driven and agent-cooperative approaches according to claim 1, characterized in that, The target workflow includes at least three stages: requirement clarification, requirement design, and requirement generation; the process of summarizing and generating the final standardized requirement processing results includes the following steps: Based on different target task types, assemble the functional chapters, business process chapters, and related system analysis chapters of the requirements document according to differentiated rules; The assembled chapter content is then filled back into the standard requirements template to generate the final standardized requirements processing result.

7. The requirement processing method based on ontology knowledge-driven and agent-cooperative approaches according to claim 3, characterized in that, When the target workflow is a product development optimization workflow, the generation of structured node execution results includes the following steps: Extract the main business process as the analysis framework, and identify the process change nodes in the main business process by comparing them with the transformation points; Output a list of modification points, which shall include at least the object of change, the method of change, and the impact assessment information.

8. The requirement processing method based on ontology knowledge-driven and agent-cooperative approaches according to claim 1, characterized in that, When the target workflow is a third-party system integration workflow, the generation of structured node execution results includes the following steps: Use the built-in standard process to identify the interaction points with third-party systems and output a list of interface points; Match the business semantics in the interface materials with the functional points in the requirement structure library to identify new functional points.

9. The requirement processing method based on ontology knowledge-driven and agent-cooperative approaches according to claim 1, characterized in that, When the target workflow is a new business scenario workflow, the generation of structured node execution results includes the following steps: Based on the requirements structure library, determine the scope of functional levels that need to be fully restored; For missing information items, fill or mark them according to preset rules to generate a function list.

10. A requirement processing system based on ontology knowledge-driven and intelligent agent collaboration, characterized in that, The system for implementing the ontology-knowledge-driven and agent-cooperative requirement processing method according to any one of claims 1-9, the system comprising: The intent recognition module is used to acquire user requests and perform intent recognition on the user requests based on the unified knowledge center to determine the target task type and the business category to which it belongs; The workflow activation module is used to activate a target workflow from a set of multiple scenario-based state machine workflows according to the target task type. Each node of the state machine workflow defines the knowledge query rules and state transition conditions between nodes. Different target task types correspond to different state machine workflows, and each workflow defines differentiated node execution rules and output rules. The sub-agent scheduling module is used to schedule corresponding business sub-agents from the pre-configured business agent layer according to the current task node of the target workflow through the main agent. The business sub-agents call the unified knowledge center to retrieve relevant knowledge according to the knowledge query rules of the current task node, and perform reasoning analysis in combination with the large language model to generate structured node execution results. The state judgment module is used to obtain the execution result of the node through the main intelligent agent and, in combination with user feedback, determine whether the state transition conditions of the current task node are met. The result generation module is used to drive the target workflow to the next node when the state transition conditions are met through the main intelligent agent; after all task nodes of the target workflow have been executed, the final standardized requirement processing result is generated.

Images