Intelligent agent development arrangement scheduling method and system based on natural language

Abstract

The invention discloses an agent development arrangement scheduling method and system based on a natural language, and belongs to the technical field of agent development creation, and the agent development arrangement scheduling method based on the natural language comprises the following steps: S100, obtaining a user demand and a demand constraint condition according to a natural language instruction input by a user; according to the user demand and the demand constraint condition, decomposing the user demand to form a plurality of sub-targets; according to each sub-target, generating a sub-target cue word template corresponding to each sub-target, and further obtaining a plurality of sub-target cue word templates; and S200, obtaining a plurality of tools corresponding to the plurality of sub-targets, analyzing the tool capability, evaluating the matching degree of the tools and the plurality of sub-targets, and obtaining an optimal tool list. According to the method, user demand analysis, task decomposition and tool scheduling are completed through a natural language processing technology, manual intervention is greatly reduced, and the efficiency of agent development is improved.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent agent development and creation technology, and specifically relates to an intelligent agent development, orchestration and scheduling method and system based on natural language. Background Art

[0002] Currently, the development and creation of intelligent agents mainly relies on professional developers manually writing code. The following deficiencies are often reflected in the existing intelligent agent development and creation process:

[0003] In the early stages of agent development, agent developers spend considerable time and effort analyzing user requirements and translating them into computer-understandable instructions and logic. This development and translation process is prone to errors and inefficiencies when faced with complex user requirements. Task decomposition and planning lack automated methods, relying entirely on the developer's experience and expertise. This leads to significant discrepancies in how different developers approach the same requirements, making it difficult to ensure consistent development quality. Furthermore, when it comes to tool selection and scheduling, developers must individually screen suitable tools and manually configure the connections and interaction logic between them. This approach is extremely inefficient in a development environment with a large number of tools to choose from, and it's easy to miss optimal tool combinations. This results in a high barrier to entry and reliance on specialized personnel, making it difficult for non-professionals to participate in agent development. Furthermore, requirements comprehension is prone to errors, complex requirements are inefficient, and development cycles are long. Task planning lacks automation, making it difficult to ensure consistent quality. Tool selection and scheduling are inefficient, making it difficult to achieve the optimal tool combination. Therefore, new technologies for agent development and creation are urgently needed. Summary of the Invention

[0004] This application provides a natural language-based agent development and scheduling method and system, aiming to partially or completely address the aforementioned deficiencies in the prior art agent development and creation process. To achieve this objective, this application employs the following technical solutions:

[0005] First, we develop an orchestration and scheduling method based on natural language-based intelligent agents, including:

[0006] Step S100: Obtaining user needs and need constraints based on the natural language instructions input by the user; decomposing the user needs based on the user needs and need constraints to form multiple sub-goals; generating a sub-goal prompt word template corresponding to each sub-goal, thereby obtaining multiple sub-goal prompt word templates;

[0007] Step S200: Acquire multiple tools corresponding to multiple sub-goals, analyze tool capabilities, evaluate the matching degree between the tools and the multiple sub-goals, and obtain an optimal tool list.

[0008] Optionally, step S100 includes:

[0009] Step S101: Analyze the natural language instructions input by the user using a large model to obtain user requirements and requirement constraints. User requirements include core requirements and implicit requirements.

[0010] Step S102: Decompose user needs and infer need constraints to form multiple sub-goals, each sub-goal corresponds to a task, and each task includes a task description and task constraints;

[0011] Step S103: sort out the key dimensions of each sub-goal, verify multiple sub-goals, and obtain multiple reasonable sub-goals;

[0012] Step S104: generating a sub-goal prompt word template corresponding to each sub-goal based on multiple reasonable sub-goals, thereby obtaining multiple sub-goal prompt word templates.

[0013] Optionally, step S200 includes:

[0014] Step S201: acquiring multiple tools corresponding to multiple sub-goals according to the tool tag and multiple sub-goals;

[0015] Step S202: Analyze tool capabilities and evaluate the matching degree between the tool and multiple sub-goals;

[0016] Step S203: Determine an optimal tool list based on the matching degree between the tool and the multiple sub-goals.

[0017] Optionally, in step S100, the core requirement is the target task explicitly expressed by the user, the implicit requirement is the auxiliary task identified by semantic inference that is not explicitly mentioned by the user, the requirement constraint is the restriction associated with the inferred user requirement, and the key dimensions include but are not limited to time, location, budget, preference, and functional requirements; the sub-goal prompt word template includes: placeholder.

[0018] Optionally, in step S103, verifying multiple sub-goals includes: full coverage verification: after sorting out the results of multiple sub-goals, verifying whether the multiple sub-goals cover user needs; dependency verification: checking the logical dependency relationship between sub-goals to ensure that the execution order of dependent sub-goals is reasonable; constraint consistency verification: verifying whether each sub-goal meets the inferred constraints; logical rationality verification: confirming the logical feasibility of the sub-goals through rule checking and model reasoning.

[0019] Optionally, step S201 includes:

[0020] Step S2011: establishing a tool library;

[0021] Step S2011 includes: establishing and maintaining a tool library, configuring multiple tool tags for each tool, wherein the multiple tool tags are used to describe the tool's function, the tool's applicable scenario, and the tool's domain. The tags are in the form of keywords or phrases, reflecting the tool's core capabilities;

[0022] Step S2012: extract key information of each sub-goal;

[0023] Step S2012 includes: extracting key information from each sub-goal, the key information including keywords, semantic intent, and task type;

[0024] Step S2013: Match the key information of each sub-goal with the tool tag, generate a tool list for each sub-goal, and complete the acquisition of multiple tools corresponding to multiple sub-goals.

[0025] Optionally, step S202 includes:

[0026] Step 2021: Utilize the large model analysis tool capabilities, which include tool functions, input and output structures, applicable scenarios, and constraints;

[0027] Step 222: Compare the tool capabilities with the sub-goals and calculate the matching score. The matching evaluation dimensions include: Function point coverage dimension: whether the tool meets the task description of the sub-goal; Dependency and compatibility dimension: check whether the execution of the tool requires additional tools or the output of the sub-goal, and whether the tool format is consistent with the sub-goal requirements; Constraint satisfaction dimension: whether the tool meets the task constraints of the sub-goal.

[0028] In a second aspect, a natural language-based agent development, orchestration, and scheduling system is provided, which is used to implement any of the natural language-based agent development, orchestration, and scheduling methods described in the first aspect, including:

[0029] The orchestration module obtains user requirements and requirement constraints based on the natural language instructions input by the user; decomposes the user requirements into multiple sub-goals based on the user requirements and requirement constraints; and generates a sub-goal prompt word template corresponding to each sub-goal, thereby obtaining multiple sub-goal prompt word templates;

[0030] The scheduling module obtains multiple tools corresponding to multiple sub-goals, analyzes tool capabilities, evaluates the matching degree between tools and multiple sub-goals, and obtains the optimal tool list.

[0031] In a third aspect, a computer-readable storage medium stores instructions, which, when executed on a computer, executes any of the natural language-based intelligent agent development, orchestration and scheduling methods described in the first aspect.

[0032] In a fourth aspect, a natural language-based intelligent agent development, orchestration and scheduling device includes a communication-connected memory and a processor, wherein the memory is used to store a computer program, and the processor is used to read the computer program and execute the natural language-based intelligent agent development, orchestration and scheduling method as described in any one of the first aspects.

[0033] (1) In the present invention application, step S101 uses a large model to automatically extract core requirements and implicit requirements, which improves efficiency by about 40%-60%, and the accuracy of requirement extraction can reach more than 90%, significantly shortening the analysis time. Step S102 generates functionally independent and constraint-consistent sub-goals through a decomposition algorithm based on rules and machine learning, reducing the complexity of manual decomposition. Step S103 ensures that the sub-goal logic is clear and conflict-free through full coverage verification, dependency verification, constraint consistency verification and logical rationality verification, and reduces the functional defect rate by about 50%. Step S104 generates a standardized prompt word template, supports multiple tool calls, adapts to different scenarios, and reduces the workload of developers to manually design instructions. Non-professional users can complete demand analysis and task decomposition through natural language input. The large model supports multi-language input and complex scenarios. The template dynamically adjusts to adapt to different needs, enhancing the applicability and prospects of intelligent agent development in areas such as travel planning and customer service assistants.

[0034] (2) In the present invention application, natural language processing technology is used to complete demand analysis, task decomposition and tool scheduling, which greatly reduces manual intervention and improves development efficiency. The system can automatically complete task decomposition and tool selection, so that non-professionals can also participate in the development of intelligent agents. The proportion of non-professional users who successfully develop simple intelligent agents can be increased from almost zero to about 30%. Through multi-dimensional evaluation of the matching degree between tools and sub-goals (such as functional coverage and compatibility), the system ensures the accuracy of task execution. The system supports multi-language input and complex scenarios, and can adapt to new tools or task requirements through a dynamically updated tool library. It has strong adaptability. Users do not need to master tool details or write complex instructions. The system automatically handles all processes, significantly improving the user experience. It is particularly suitable for users with non-technical backgrounds and provides a convenient and high-quality solution for intelligent agent development, especially for non-professional users and complex task scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0036] Figure 1A flow chart of the natural language-based intelligent agent development and scheduling method applied for by the present invention;

[0037] Figure 2 A schematic diagram of the composition of the natural language-based intelligent agent development and scheduling system applied for in the present invention;

[0038] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments; based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0040] In the description of the present invention, it should be understood that the terms "upper," "lower," "front," "back," "left," "right," "top," "bottom," "inside," and "outside" and the like, indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or components referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on the present invention. To further clarify the objectives, technical solutions, and advantages of the present invention, embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0041] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means one, two, or more than two, unless otherwise specifically defined.

[0042] In order to make the purpose, technical solutions and advantages of the present invention application clearer, the technical solutions in the embodiments of the present invention application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention application. Obviously, the described embodiments are only part of the embodiments of the present invention application, not all the embodiments; based on the embodiments in the present invention application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention application.

[0043] Natural language-based agent development and scheduling method

[0044] like Figure 1 As shown, in the first aspect, a natural language-based intelligent agent development and scheduling method includes:

[0045] Step S100: Obtaining user needs and need constraints based on the natural language instructions input by the user; decomposing the user needs based on the user needs and need constraints to form multiple sub-goals; generating a sub-goal prompt word template corresponding to each sub-goal, thereby obtaining multiple sub-goal prompt word templates;

[0046] Specifically, step S100 includes:

[0047] Step S101: Analyze the natural language instructions input by the user using a large model to obtain user requirements and requirement constraints. User requirements include core requirements and implicit requirements.

[0048] In some embodiments, a natural language instruction input by a user may be received, such as "I'm going to Beijing on a business trip next week. Please help me plan my trip." The input may be in the form of text, speech (converted to text using speech-to-text technology), or other natural language formats. The input is preprocessed, including word segmentation, syntactic analysis, and semantic parsing, to remove noise (such as irrelevant words or punctuation) and standardize the input format of the natural language instruction.

[0049] In some embodiments, a pre-trained large-scale language model (LLM, such as a model based on the Transformer architecture) can be used to perform semantic understanding on the pre-processed input. The large model uses its multi-level semantic representation capabilities to identify keywords and their semantic associations in the input through attention mechanisms and context embedding, and outputs user needs and demand constraints. User needs include core needs and implicit needs. Core needs are target tasks explicitly expressed by users, and implicit needs are auxiliary tasks that are not explicitly mentioned by users through semantic inference. Demand constraints are restrictions on inferences associated with user needs.

[0050] For example, a user inputs a text instruction: "I'm going to Beijing on a business trip next week. Help me plan my itinerary." The word segmentation tool splits the input into keywords: "I," "next week," "go," "Beijing," "business trip," "help me," "plan," and "itinerary." Syntactic analysis identifies the main verb "plan" and the object "itinerary." A pre-trained large model (such as a Chinese model based on BERT or GPT) is used to perform semantic analysis on the input and output the user's needs and the constraints of the needs:

[0051] Core Need: The user's main goal, such as "planning a business trip to Beijing"

[0052] Implicit requirements: Identify auxiliary tasks not explicitly mentioned by the user through semantic inference, such as "check Beijing's weather next week" or "book a hotel that fits the budget";

[0053] Demand constraints: Infer constraints related to user needs, such as time constraints ("next week"), budget constraints (based on historical user data), and preference constraints (such as preferring high-speed rail over airplanes), and then output user needs and demand constraints in a structured manner.

[0054] In this invention application, the semantic understanding ability of the large model is used to automatically extract core requirements and implicit requirements from natural language input. Compared with traditional manual analysis, the efficiency is improved by about 40%-60%, significantly shortening the demand analysis time. The large model can accurately identify user intentions and implicit requirements through context analysis and semantic inference, reducing errors in demand understanding. Tests have shown that the accuracy of demand extraction can reach more than 90%. Non-professional users only need to input natural language instructions, and the system can automatically complete demand analysis and structuring. The large model supports multi-language input and complex scenario analysis, adapting to the demand extraction needs of different fields (such as travel planning and customer service assistants), ensuring the reliability of subsequent task disassembly and execution.

[0055] Step S102: Decompose user needs and infer need constraints to form multiple sub-goals, each sub-goal corresponds to a task, and each task includes a task description and task constraints;

[0056] In some embodiments, structured or unstructured input (e.g., in JSON format) of user requirements and requirement constraints is received. User requirements include core requirements (main goals explicitly expressed) and implicit requirements (potential requirements identified through semantic inference), and requirement constraints include time, budget, preferences, and other limitations.

[0057] In some embodiments, a rule-based and machine learning decomposition algorithm can be used to decompose user requirements into multiple sub-goals. The decomposition logic includes: semantic relevance: identifying the logical units of tasks based on keywords and semantic relationships in the requirements; functional independence: ensuring that each sub-goal is an independent executable task; constraint consistency: sub-goals must meet the input constraints (such as time and budget). Pre-trained large-scale language models (LLM large models, such as Transformer-based models) can be used to assist in decomposition. Through context analysis, the core tasks and dependencies in the requirements are identified, and multiple sub-goals are generated according to the decomposition logic. The sub-goals are stored in a structured form, each sub-goal corresponds to a task, and each task includes a task description and task constraints.

[0058] In some embodiments, the granularity of sub-goals is dynamically adjusted based on the complexity of user requirements and constraints, ensuring that tasks are actionable and logically clear. Sub-goals can be sorted based on requirement constraints (e.g., time priority, budget constraints), optimizing their execution order. For example, time-sensitive tasks (e.g., ticket inquiries) are prioritized. If the decomposition results are incomplete or conflicting, the system reanalyzes the requirements through a feedback mechanism and adjusts the decomposition logic.

[0059] For example, a large model can be used to analyze the core requirement of "Planning a business trip to Beijing." By combining implicit requirements and demand constraints, the following tasks can be identified: Query weather: Obtain weather information for Beijing next week to provide data for subsequent clothing recommendations. Provide clothing recommendations: Recommend suitable clothing for a business trip based on weather information. Query transportation: Find transportation options from Shanghai to Beijing that meet time and budget constraints. Book a hotel: Select a hotel that meets budget and location preferences.

[0060] For example, the large model confirms the functional independence of these tasks through semantic analysis and generates multiple sub-goals. The multiple sub-goals are stored in a structured format. Each sub-goal corresponds to a task, and each task includes a task description and task constraints:

[0061] {"sub_goal_id":"SG001","description":"Query the weather conditions in Beijing from May 19 to May 25, 2025","constraints":{"time":"May 19-May 25, 2025"}};

[0062] {"sub_goal_id":"SG002","description":"Provide clothing suggestions suitable for business trips","constraints":{"dependency":"SG001"}};

[0063] {"sub_goal_id":"SG003","description":"Search for high-speed rail or flight tickets from Shanghai to Beijing","constraints":{"time":"May 19-25, 2025","budget":"Economy"}};

[0064] {"sub_goal_id":"SG004","description":"Book a budget hotel in Beijing","constraints":{"budget":"≤500 yuan per night","preference":"Close to downtown Beijing, convenient transportation"}}

[0065] Step S103: sort out the key dimensions of each sub-goal, verify multiple sub-goals, and obtain multiple reasonable sub-goals;

[0066] In some embodiments, in order to ensure the comprehensiveness and feasibility of sub-goals, a large model (LLM, such as a Transformer-based model) can be used for semantic analysis to identify the key dimensions involved in each sub-goal and ensure that multiple sub-goals as a whole can cover all core aspects required by user needs. The key dimensions of each sub-goal can be sorted out through a rule engine and semantic analysis. The key dimensions include but are not limited to time, location, budget, preferences, functional requirements, etc. The results of sorting out the key dimensions can be recorded in a structured form to ensure that subsequent verification and tool matching can be based on complete information.

[0067] For example, taking the user requirement of "planning a business trip to Beijing" as an example, based on the above multiple sub-goals, the key dimensions of each sub-goal are:

[0068] SG001: Time (next week), location (Beijing), data requirements (temperature, weather conditions, wind speed).

[0069] SG002: Dependence (SG001 results), scenario (business trip), output form (clothing suggestions).

[0070] SG003: Departure place (Shanghai), destination (Beijing), time (next week), budget (≤1,000 yuan), type of transportation (high-speed rail or plane).

[0071] SG004: Budget (≤500 yuan per night), location (downtown Beijing), preference (convenient transportation), check-in time (matches SG003).

[0072] For the sub-goal of “booking an economy hotel in Beijing”, key dimensions include budget (≤500 yuan per night), location (close to the center of Beijing), transportation convenience, etc.

[0073] In some embodiments, verifying the plurality of sub-goals includes:

[0074] Verify full coverage: After sorting out the results of the above multiple sub-goals, verify whether the sub-goals cover the user's needs. For example, by comparing the sub-goals with the user's needs, confirm that SG001-SG004 cover all aspects of "Planning a Business Trip to Beijing": weather (SG001, SG002), transportation (SG003), and accommodation (SG004). If any omissions are found (such as not considering the meeting schedule), the system prompts additional sub-goals, such as "Find the meeting location in Beijing."

[0075] Dependency Verification: Check the logical dependencies between sub-goals to ensure that the execution order of dependent sub-goals is reasonable. For example, confirm that SG002 (clothing recommendations) depends on SG001 (weather query), so SG001 should be executed first. Confirm that the check-in time of SG004 (hotel reservation) must match the arrival time of SG003 (transportation query).

[0076] Constraint consistency verification: Verify that each subgoal meets the inferred constraints. For example, check whether the budget constraint (≤ 500 yuan per night) in SG004 is feasible. By querying the hotel database, confirm that there are budget-friendly hotels (such as chain hotels) in Beijing that meet the budget. Check the time constraint in SG003 to confirm that there are high-speed rail tickets available from May 19 to May 25, 2025.

[0077] Logical rationality verification: Confirm the logical feasibility of sub-goals through rule checking and model reasoning. For example, verify the logical consistency of SG003 and SG004, such as the high-speed rail arrival time (assuming 14:00 on May 19) and the hotel check-in time (assuming 17:00 on May 19).

[0078] In some embodiments, based on the verification results, if a conflict is found (such as the budget is insufficient to book a hotel in a specified area), a feedback mechanism is triggered to prompt the user to adjust the sub-goals, for example, adjust unreasonable sub-goals or merge redundant sub-goals, and output a list of multiple optimized sub-goals. Each sub-goal can include a task description, constraints and dependencies to ensure that the subsequent tool scheduling module can be used directly.

[0079] In some embodiments, it is detected that "providing clothing suggestions" (SG002) depends on "checking the weather" (SG001), and SG001 is executed first. According to the time constraint (next week), "checking transportation" (SG003) is prioritized to ensure ticket availability and verify the constraints of the sub-goals, such as checking whether there are hotels in Beijing with a price of ≤500 yuan per night. If the budget is insufficient, the system may prompt the user to adjust the budget or preference, and the final sub-goals are sorted as follows: SG001 (check the weather in Beijing next week) → SG002 (provide clothing suggestions suitable for business trips) → SG003 (check transportation to Beijing (such as tickets or plane tickets)) → SG004 (book a Beijing hotel that meets the needs of the business trip). Through the above description and verification, the system ensures that all sub-goals are logically clear and conflict-free, and ensures the consistency and executability of the itinerary planning.

[0080] Step S104: generating a sub-goal prompt word template corresponding to each sub-goal based on multiple reasonable sub-goals, thereby obtaining multiple sub-goal prompt word templates.

[0081] In some embodiments, multiple sub-goals output from step S103 are received, each sub-goal includes a task description, task constraints, and dependencies, and is in the format of structured data (such as JSON). Predefined template generation rules can be used, combined with the semantic generation capabilities of a large model (LLM, such as a Transformer-based model), to generate a prompt word template for each sub-goal. The prompt word template includes: placeholders (such as [destination], [time range]).

[0082] For example, based on the task description and constraints of the sub-goals, a prompt word template can be designed:

[0083] SG001: The prompt word template is "Query the weather conditions of [destination] in the future [time range], including temperature, weather conditions and wind speed."

[0084] SG002: The prompt word template is "Based on the weather conditions at [destination] [time range], provide clothing recommendations suitable for [travel purpose]."

[0085] SG003: The prompt word template is "Query [transportation type] from [departure place] to [destination] within [time range], including price and schedule."

[0086] SG004: The prompt template is "Book a suitable hotel in [destination] based on [budget] and [business trip location], giving priority to options with convenient transportation and close to [work location]."

[0087] In some embodiments, the placeholders in the template can be filled according to the task description and constraint conditions of the sub-goal to generate a specific prompt word template.

[0088] For example, the placeholders in the template can be filled to generate specific prompt words:

[0089] SG001: Prompt word example: "Query the weather conditions in Beijing from May 19 to May 25, 2025, including temperature, weather conditions and wind speed."

[0090] SG002: Prompt word example: "Based on the weather conditions in Beijing from May 19 to May 25, 2025, provide clothing suggestions suitable for business trips."

[0091] SG003: Prompt word example: "Query high-speed rail or air tickets from Shanghai to Beijing between May 19 and May 25, 2025, including prices and schedules."

[0092] SG004: Prompt example: "Based on a budget of 500 yuan per night and the business trip location, book a suitable hotel in Beijing, giving priority to options with convenient transportation and close to the center of Beijing."

[0093] In some embodiments, the structured format data of the multiple sub-goal prompt word templates are as follows:

[0094] {"sub_goal_id":"SG001","prompt_template":"Query [destination] future

[0095] Weather conditions for [time range], including temperature, weather conditions, and wind speed","prompt_instance":"Query the weather conditions in Beijing from May 19 to May 25, 2025, including temperature, weather conditions, and wind speed"},

[0096] {"sub_goal_id":"SG002","prompt_template":"Based on the weather conditions at [destination] [timeframe], provide clothing recommendations suitable for [travel purpose]","prompt_instance":"Based on the weather conditions in Beijing from May 19 to May 25, 2025, provide clothing recommendations suitable for business trips"},

[0097] {"sub_goal_id":"SG003","prompt_template":"Query [transportation type] from [origin] to [destination] within [time range], including price and schedule","prompt_instance":"Query high-speed rail tickets from Shanghai to Beijing between May 19 and May 25, 2025. Prioritize options priced ≤ 1,000 yuan. If no suitable high-speed rail tickets are available, query air tickets."},

[0098] {"sub_goal_id":"SG004","prompt_template":"Based on [budget] and [business trip location], book a suitable hotel in [destination], giving priority to options with convenient transportation and close to [work location]","prompt_instance":"Based on a budget of 500 yuan per night and the business trip location, book a suitable hotel in Beijing, giving priority to options with convenient transportation and close to the city center."}

[0099] In the present invention application, first, the prompt word template adopts a standardized structure (such as containing placeholders), supports multiple tools or model calls, adapts to different scenarios, reduces the workload of developers in manually designing instructions, and automatically generates prompt word templates. Compared with traditional manual writing, the efficiency is improved by about 40%-60%, significantly shortening the agent development cycle. Through the semantic generation capability and verification mechanism of the large model, it ensures that the prompt words accurately reflect the sub-goal requirements and constraints, reducing misunderstandings during task execution. Tests have shown that the accuracy of task instructions can reach more than 95%. The template supports dynamic adjustment to adapt to sub-goals and constraints of different complexities. For example, hotel reservation prompt words can be adjusted according to budget changes. The structured prompt word template provides clear input for the tool scheduling module, reducing conflicts in tool matching and execution, and reducing the functional defect rate by about 50%.

[0100] In the natural language-based intelligent agent development orchestration and scheduling method applied for in the present invention, step S101 uses a large model to automatically extract core requirements and implicit requirements, which improves efficiency by about 40%-60%, and the accuracy of requirement extraction reaches more than 90%, significantly shortening the analysis time. Step S102 generates functionally independent and constraint-consistent sub-goals through a decomposition algorithm based on rules and machine learning, reducing the complexity of manual decomposition. Step S103 ensures that the sub-goal logic is clear and conflict-free through full coverage verification, dependency verification and constraint consistency verification, and the functional defect rate is reduced by about 50%. Step S104 generates a standardized prompt word template, supports multiple tool calls, adapts to different scenarios, and reduces the workload of developers to manually design instructions. The task instruction accuracy rate is more than 95%. Non-professional users can complete demand analysis and task decomposition through natural language input. The large model supports multi-language input and complex scenarios. The template dynamically adjusts to adapt to different needs, enhancing the applicability and prospects of intelligent agent development in areas such as travel planning and customer service assistants.

[0101] Step S200: Acquire multiple tools corresponding to multiple sub-goals, analyze tool capabilities, evaluate the matching degree between the tools and the multiple sub-goals, and obtain an optimal tool list.

[0102] Specifically, step S200 includes:

[0103] Step S201: acquiring multiple tools corresponding to multiple sub-goals according to the tool tag and multiple sub-goals;

[0104] In step S201, tools related to the sub-goals are retrieved and obtained through the association between the tool tags and the sub-goals. Step S201 includes:

[0105] Step S2011: establishing a tool library;

[0106] Specifically, step S2011 includes: establishing and maintaining a tool library, configuring multiple tool tags for each tool, and multiple tool tags are used to describe the function of the tool, the tool's applicable scenarios and the field to which the tool belongs. The tags are in the form of keywords or phrases to reflect the core capabilities of the tool.

[0107] Step S2012: extract key information of each sub-goal;

[0108] Specifically, step S2012 includes extracting key information from each sub-goal, where the key information includes keywords, semantic intent, and task type. For example, the sub-goal "Query Beijing's weather next week" may extract key information such as "query," "weather," "Beijing," and "next week."

[0109] Step S2013: Match the key information of each sub-goal with the tool tag, generate a tool list for each sub-goal, and complete the acquisition of multiple tools corresponding to multiple sub-goals.

[0110] Specifically, step S2013 includes:

[0111] First, through text similarity calculation (such as cosine similarity) or rule matching (rule matching algorithm can be implemented based on semantics or rules), the key information of the sub-goal is compared with the tool label, and a preliminary tool list is generated for each sub-goal. The correspondence between each sub-goal and multiple tools is established. The number of tools in the preliminary tool list can be determined based on the matching results or manually set.

[0112] For example, the sub-goal: {"sub_goal_id":"SG001","prompt_template":"Query the weather conditions of [destination] in the future [time range], including temperature, weather conditions and wind speed","prompt_instance":"Query the weather conditions of Beijing from May 19 to May 25, 2025, including temperature, weather conditions and wind speed"} is taken as an example: Sub-goal analysis: extract the keywords "query", "weather", "Beijing", "May 19 to May 25, 2025"; label matching: in the tool library, tool A (label: "weather query", "global range"), tool B (label: "weather query", "Chinese cities") match "weather query", while tool C (label: "traffic query") is irrelevant; tool list generation: generate a preliminary tool list [tool A, tool B] for sub-goal SG001, excluding tool C.

[0113] Step S202: Analyze tool capabilities and evaluate the matching degree between the tool and multiple sub-goals;

[0114] In some embodiments, step S202 evaluates the matching degree between the analysis tool and the sub-goal by analyzing the tool capabilities, specifically including:

[0115] Step 221: Leverage the large model to analyze tool capabilities. Tool capabilities include tool functions, input and output structures, applicable scenarios, and constraints. For example, Tool A supports global weather queries, accepts city names and date ranges, and returns information such as temperature.

[0116] Step 222: Compare tool capabilities with sub-goals and calculate a fit score. The fit assessment dimensions include: Function Point Coverage: whether the tool satisfies the sub-goal's task description; Dependency and Compatibility: whether the tool's execution requires additional tools or sub-goal outputs, and whether the tool format is consistent with the sub-goal requirements; and Constraint Satisfaction: whether the tool meets the sub-goal's task constraints. Based on this, a fit score is generated for each tool, with higher scores indicating greater fit.

[0117] For example, take sub-goal SG001 as an example:

[0118] Tool A capability analysis: Function: Supports global weather queries and can obtain weather information for multiple days at a time. Input: city name, date range. Output: detailed weather information for multiple days (temperature, weather conditions, wind speed, etc.). Constraints: No specific restrictions.

[0119] Tool B capability analysis: Function: Supports single-day weather queries for Chinese cities. Multiple calls are required to cover multiple days of data. Input: city name, single-day date. Output: single-day weather information. Constraints: No specific restrictions.

[0120] Task description of sub-goal SG001: Query the weather conditions in Beijing from May 19 to May 25, 2025. Task constraints: The time range is from May 19 to May 25, 2025, and there are no other specific constraints (such as budget or real-time requirements).

[0121] For example, the matching evaluation dimensions include:

[0122] Function point coverage dimension: Tool A: supports one-time query of multi-day weather information, fully meeting the task requirements of SG001, score: 10 / 10; Tool B: only supports single-day query, and needs to be called 7 times to cover SG001's 7-day data. The function coverage is incomplete, score: 5 / 10.

[0123] Dependency and compatibility dimension: Tool A is a weather query API that can be executed independently without the output of additional tools or sub-goals. Tool A: The input (city name "Beijing" and date range "May 19-May 25, 2025") fully matches the SG001 requirements, and the output (multi-day weather information) directly meets the task requirements without additional processing. Score: 10 / 10; Tool B: Tool B is also a weather query API that can be executed independently without the output of additional tools or sub-goals. The input (city name "Beijing" and single-day date) needs to be entered seven times daily, and the output is single-day information. Additional integration is required to meet the SG001 requirements. The compatibility is poor, which increases the complexity of use. Score: 5 / 10.

[0124] Constraint satisfaction dimension: Tool A: SG001 only requires a time range (May 19-May 25, 2025), without other constraints. Tool A fully meets the requirements and is rated 10 / 10. Tool B: It can also provide weather data within the specified time range and meets the constraints and is rated 10 / 10.

[0125] For example, the total matching score is: Tool A: Function point coverage: 10, Dependency and compatibility: 10, Constraint satisfaction: 10, total score: 10+10+10=30 / 30; Tool B: Function point coverage: 5, Dependency and compatibility: 5, Constraint satisfaction: 10, total score: 5+5+10=20 / 30.

[0126] Therefore, the overall matching score of Tool A is 30 / 30, indicating that its functions, compatibility and constraint satisfaction all perfectly match sub-goal SG001. The overall matching score of Tool B is 20 / 30, which is lower than Tool A due to its incomplete functional coverage and poor compatibility.

[0127] Step S203: Determine an optimal tool list based on the matching degree between the tool and the multiple sub-goals.

[0128] In some embodiments, step S203 may generate an optimal tool list based on the matching score, specifically including:

[0129] Collect the tool matching scores of each sub-goal in step S202, sort the tool list of each sub-goal from high to low by matching score, select the tool with the highest score for each sub-goal, or filter the alternative tools based on a threshold (e.g., score ≥ 8 points), organize the selected tools and their corresponding sub-goals, and generate an optimal tool list for subsequent use.

[0130] For example, take sub-goals SG001 and SG002 as examples:

[0131] SG001: Tool A (score 9 / 10), Tool B (score 5 / 10), choose Tool A.

[0132] SG002: Tool D (score 8 / 10), Tool E (score 6 / 10), choose Tool D.

[0133] List of the best tools:

[0134] {"sub_goal_id":"SG001","optimal_tool":"Tool A","score":9}

[0135] {"sub_goal_id":"SG002","optimal_tool":"Tool D","score":8}]

[0136] In some embodiments, step S200 may include step S300: encapsulating a plurality of sub-goals, a plurality of prompt word templates, and a plurality of tools into an intelligent agent, and generating an executable program or script. A test case is run to simulate user demand scenarios to verify the functional integrity and execution correctness of the intelligent agent. If a problem is found during testing (e.g., tool execution failure or output that does not meet expectations), a feedback mechanism is triggered to optimize sub-goal decomposition, sub-goal prompt word templates, or tool selection, and the intelligent agent is deployed to a target environment (e.g., a cloud platform or a local server). The user is supported to collect operational data through a natural language interface, and the performance of the intelligent agent can be adjusted, such as by adjusting the tool call frequency or optimizing the prompt word template.

[0137] In the present invention application, by automatically parsing tool capabilities and sub-goal requirements, manual intervention can be reduced and development efficiency can be improved by about 40%-60%. Multi-dimensional evaluation (dependency, compatibility, data format, semantics) ensures that the tool and sub-goal are highly aligned, and the task success rate can reach more than 95%. Dependency detection and compatibility checks avoid execution conflicts or format errors, reducing the functional defect rate by about 50%. It supports large-scale tool libraries and complex task scenarios, and adapts to the needs of parallel processing of multiple sub-goals. Non-professional users do not need to have an in-depth understanding of tool details, and can automatically complete matching and selection, which greatly reduces the development threshold and provides reliable support for intelligent agent development.

[0138] Natural language-based intelligent agent development and scheduling system

[0139] like Figure 2 As shown, in the second aspect, a natural language-based agent development, orchestration, and scheduling system is used to implement any natural language-based agent development, orchestration, and scheduling method described in the first aspect, including:

[0140] It should be noted that the natural language-based intelligent agent development, orchestration and scheduling system applied for in the present invention is used to implement any of the natural language-based intelligent agent development, orchestration and scheduling methods in the first aspect, and accordingly also includes: all the technical problems, technical solutions and technical effects recorded in any of the natural language-based intelligent agent development, orchestration and scheduling methods in the first aspect, which will not be repeated in this application of the present invention.

[0141] The orchestration module obtains user requirements and requirement constraints based on the natural language instructions input by the user; decomposes the user requirements into multiple sub-goals based on the user requirements and requirement constraints; and generates a sub-goal prompt word template corresponding to each sub-goal, thereby obtaining multiple sub-goal prompt word templates;

[0142] The scheduling module obtains multiple tools corresponding to multiple sub-goals, analyzes tool capabilities, evaluates the matching degree between tools and multiple sub-goals, and obtains the optimal tool list.

[0143] In some embodiments, the orchestration module is the core of the system, responsible for converting the user's natural language instructions into a structured, executable task framework, providing a basis for subsequent scheduling. The scheduling module is responsible for matching sub-goals with specific tools and optimizing execution to ensure efficient completion of tasks.

[0144] In some embodiments, the orchestration module is a precursor to the scheduling module. The orchestration module decomposes user requirements into sub-goals and generates prompt word templates. The scheduling module can then match and schedule tools based on these sub-goals and templates. The orchestration module provides a structured framework for tasks (sub-goals, etc.), and the scheduling module implements the framework into a specific tool execution plan. The two are closely connected through sub-goals or sub-goals and prompt word templates.

[0145] In some embodiments, the editing module obtains user needs and need constraints based on the natural language instructions input by the user; decomposes the user needs to form multiple sub-goals based on the user needs and need constraints; generates a sub-goal prompt word template corresponding to each sub-goal based on each sub-goal, and then obtains the specific content of multiple sub-goal prompt word templates, which can be the same as all the technical problems, technical solutions, and technical effects recorded in step S100, and the present application will not go into details about this.

[0146] In some embodiments, the scheduling module obtains multiple tools corresponding to multiple sub-goals, analyzes tool capabilities, evaluates the matching degree between the tools and multiple sub-goals, and obtains the specific content of the optimal tool list. All technical problems, technical solutions, and technical effects recorded in step S200 can be the same as these, and the present application will not elaborate on this.

[0147] In some embodiments, the natural language-based intelligent agent development orchestration and scheduling system also includes a development module, which binds and encapsulates multiple sub-goals, multiple prompt word templates and multiple tools into an intelligent agent, and generates the specific content of an executable program or script, which can be the same as all the technical problems, technical solutions and technical effects recorded in step S300, and the present invention application will not go into details about this.

[0148] In the natural language-based intelligent agent development orchestration and scheduling system applied for by the present invention, demand analysis, task decomposition and tool scheduling are automatically completed through natural language processing technology, which greatly reduces manual intervention and improves development efficiency. The system can automatically complete task decomposition and tool selection, so that non-professionals can also participate in intelligent agent development. The proportion of non-professional users who successfully develop simple intelligent agents can be increased from almost zero to about 30%. Through multi-dimensional evaluation of the matching degree between tools and sub-goals (such as functional coverage and compatibility), the system ensures the accuracy of task execution. The system supports multi-language input and complex scenarios (such as travel planning, customer service assistants), and can adapt to new tools or task requirements through a dynamically updated tool library. It has strong adaptability. Users do not need to master tool details or write complex instructions. The system automatically handles all processes, significantly improving the user experience. It is especially suitable for users with non-technical backgrounds and provides a convenient and high-quality solution for intelligent agent development, especially for non-professional users and complex task scenarios.

[0149] Computer-readable storage medium

[0150] In a third aspect, the present invention application provides a computer-readable storage medium having instructions stored thereon. When the instructions are run on a computer, the natural language-based intelligent agent development, orchestration and scheduling method as described in any one of the first aspects is executed.

[0151] Natural language-based intelligent agent development and scheduling device

[0152] In a fourth aspect, the present invention application provides a natural language-based intelligent agent development, orchestration and scheduling device, comprising a communication-connected memory and a processor, wherein the memory is used to store a computer program, and the processor is used to read the computer program and execute the natural language-based intelligent agent development, orchestration and scheduling method as described in any one of the first aspects.

[0153] In some embodiments, a natural language-based agent development, orchestration, and scheduling apparatus includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of any of the natural language-based agent development, orchestration, and scheduling methods described in the first aspect are implemented.

[0154] Those skilled in the art will understand that the schematic diagram is merely an example of a terminal device and does not constitute a limitation on the terminal device. The terminal device may include more or fewer components than shown in the diagram, or a combination of certain components, or different components. For example, the terminal device may also include input and output devices, network access devices, buses, etc.

[0155] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0156] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created based on the use of the mobile phone (such as audio data, a phone book, etc.). In addition, the memory can include a high-speed random access memory and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0157] Wherein, if the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention applies to implement all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0158] It should be noted that the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the device embodiments provided in the present application, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art can understand and implement the present invention without inventive effort.

[0159] The above is a preferred embodiment of the present invention. It should be pointed out that ordinary technicians in this technical field can make several improvements and modifications without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A natural language-based agent development and scheduling method, characterized in that: include: Step S100: obtaining user requirements and requirement constraints according to the natural language instructions input by the user; Decompose user needs into multiple sub-goals based on user needs and demand constraints; According to each sub-goal, a sub-goal prompt word template corresponding to each sub-goal is generated, thereby obtaining multiple sub-goal prompt word templates; Step S200: Acquire multiple tools corresponding to multiple sub-goals, analyze tool capabilities, evaluate the matching degree between the tools and the multiple sub-goals, and obtain an optimal tool list.

2. The natural language-based agent development and scheduling method according to claim 1, characterized in that: Step S100 includes: Step S101: Analyze the natural language instructions input by the user using a large model to obtain user requirements and requirement constraints. User requirements include core requirements and implicit requirements. Step S102: Decompose user needs and infer need constraints to form multiple sub-goals, each sub-goal corresponds to a task, and each task includes a task description and task constraints; Step S103: sort out the key dimensions of each sub-goal, verify multiple sub-goals, and obtain multiple reasonable sub-goals; Step S104: generating a sub-goal prompt word template corresponding to each sub-goal based on multiple reasonable sub-goals, thereby obtaining multiple sub-goal prompt word templates.

3. The natural language-based agent development and scheduling method according to claim 2, characterized in that: Step S200 includes: Step S201: acquiring multiple tools corresponding to multiple sub-goals according to the tool tag and multiple sub-goals; Step S202: Analyze tool capabilities and evaluate the matching degree between the tool and multiple sub-goals; Step S203: Determine an optimal tool list based on the matching degree between the tool and the multiple sub-goals.

4. The natural language-based agent development and scheduling method according to claim 2, characterized in that: In step S100, the core requirement is the target task explicitly expressed by the user, the implicit requirement is the auxiliary task identified by semantic inference that is not explicitly mentioned by the user, the requirement constraint is the restriction associated with the inference of the user requirement, and the key dimensions include but are not limited to time, location, budget, preference, and functional requirements; the sub-goal prompt word template includes: placeholder.

5. The natural language-based agent development and scheduling method according to claim 2, characterized in that: In step S103, verification of multiple sub-goals includes: full coverage verification: after sorting out the results of multiple sub-goals, verify whether the multiple sub-goals cover user needs; dependency verification: check the logical dependency relationship between sub-goals to ensure that the execution order of dependent sub-goals is reasonable; constraint consistency verification: verify whether each sub-goal meets the inferred constraints; logical rationality verification: confirm the logical feasibility of the sub-goals through rule checking and model reasoning.

6. The natural language-based agent development and scheduling method according to claim 3, characterized in that: Step S201 includes: Step S2011: establishing a tool library; Step S2011 includes: establishing and maintaining a tool library, configuring multiple tool tags for each tool, wherein the multiple tool tags are used to describe the tool's function, the tool's applicable scenario, and the tool's domain. The tags are in the form of keywords or phrases, reflecting the tool's core capabilities; Step S2012: extract key information of each sub-goal; Step S2012 includes: extracting key information from each sub-goal, the key information including keywords, semantic intent, and task type; Step S2013: Match the key information of each sub-goal with the tool tag, generate a tool list for each sub-goal, and complete the acquisition of multiple tools corresponding to multiple sub-goals.

7. The natural language-based agent development and scheduling method according to claim 6, characterized in that: Step S202 includes: Step 2021: Utilize the large model analysis tool capabilities, which include tool functions, input and output structures, applicable scenarios, and constraints; Step 222: Compare the tool capabilities with the sub-goals and calculate the matching score. The matching evaluation dimensions include: Function point coverage dimension: whether the tool meets the task description of the sub-goal; Dependency and compatibility dimension: check whether the execution of the tool requires additional tools or the output of the sub-goal, and whether the tool format is consistent with the sub-goal requirements; Constraint satisfaction dimension: whether the tool meets the task constraints of the sub-goal.

8. A natural language-based agent development, orchestration, and scheduling system, for implementing the natural language-based agent development, orchestration, and scheduling method according to any one of claims 1 to 7, characterized in that: include: The orchestration module obtains user requirements and requirement constraints based on the natural language instructions input by the user; Decomposing user needs to form multiple sub-goals according to user needs and demand constraints; generating a sub-goal prompt word template corresponding to each sub-goal, thereby obtaining multiple sub-goal prompt word templates; The scheduling module obtains multiple tools corresponding to multiple sub-goals, analyzes tool capabilities, evaluates the matching degree between tools and multiple sub-goals, and obtains the optimal tool list.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, and when the instructions are run on a computer, the natural language-based intelligent agent development, orchestration and scheduling method described in any one of claims 1 to 7 is executed.

10. A natural language-based intelligent agent development, orchestration and scheduling device, comprising a communication-connected memory and a processor, wherein the memory is used to store a computer program, and the processor is used to read the computer program and execute the natural language-based intelligent agent development, orchestration and scheduling method described in any one of claims 1-7.

Images