Education task processing method and device based on multi-agent dialogue driving

Abstract

The invention provides an educational task processing method and device based on multi-agent dialogue driving, and the method comprises the steps: obtaining an educational task demand, and analyzing the educational task demand into a plurality of sub-educational tasks; determining a task type corresponding to each sub-education task, and allocating the sub-education tasks to corresponding target task agents according to the task types; establishing an interactive dialogue between the target task agents, executing the corresponding sub-education task through a natural language dialogue between the target task agents to determine a sub-task execution result, and integrating the sub-task execution result into a target task response; and feeding back the target task response to the user. A distributed proxy cooperation mode can be adopted, efficient cooperation is achieved in a dialogue-driven mode, and more efficient and more flexible technical support is provided for task processing in the field of intelligent education.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to a method and device for processing educational tasks based on multi-agent dialogue drive. Background Art

[0002] With the rapid development of artificial intelligence technology, intelligent education systems are becoming an important direction in the field of education. These systems provide personalized, real-time teaching and learning support for teachers and students by combining natural language processing, large language models (LLM) and intelligent agent technology. However, current intelligent education systems usually rely on a single intelligent agent to perform tasks, which is difficult to meet the diverse needs of complex educational tasks. The functions of a single agent are usually predefined and difficult to flexibly adjust according to dynamically changing task requirements; faced with complex educational scenarios involving diverse subtasks such as content generation, knowledge retrieval, and design optimization, a single agent is difficult to provide comprehensive support. Summary of the invention

[0003] The embodiments of the present disclosure at least provide a method and device for processing educational tasks based on multi-agent dialogue-driven, which achieves efficient collaboration through dialogue-driven means by adopting a collaborative mode of distributed agents, thereby providing more efficient and flexible technical support for task processing in the field of intelligent education.

[0004] The present disclosure provides an educational task processing method based on multi-agent dialogue driving, including:

[0005] Obtaining educational task requirements, and parsing the educational task requirements into multiple sub-educational tasks;

[0006] Determine the task type corresponding to each of the sub-education tasks, and assign the sub-education tasks to corresponding target task agents according to the task type;

[0007] Establishing an interactive dialogue between the target task agents, executing the corresponding sub-education tasks through the natural language dialogue between the target task agents to determine the sub-task execution results, and integrating the sub-task execution results into the target task response;

[0008] The target task response is fed back to the user.

[0009] In an optional implementation, the educational task requirement is parsed into a plurality of sub-educational tasks, specifically including:

[0010] Performing semantic understanding on the educational task requirements to determine the semantic content corresponding to the educational task requirements;

[0011] Identify the target educational task included in the semantic content and determine the corresponding task logic, and decompose the semantic content into a plurality of sub-educational tasks according to the task logic;

[0012] The task dependency relationship between the sub-educational tasks is determined according to the task logic, and a corresponding task priority is set for each sub-educational task according to the task dependency relationship.

[0013] In an optional implementation, an interactive dialogue is established between the target task agents, and the corresponding sub-education tasks are executed through the natural language dialogue between the target task agents to determine the sub-task execution results, specifically including:

[0014] Determine the task execution order according to the task priority and the task dependency, and set the agent execution order between the target task agents according to the task execution order;

[0015] Using a unified dialogue interface to establish an interactive dialogue channel between the target task agents;

[0016] For each of the target task agents, during the execution of the sub-education task, according to the execution requirements corresponding to the sub-education task, a target dialogue is established with other target task agents through the interactive dialogue channel;

[0017] Sending demand information corresponding to the execution demand through the target dialogue, and receiving response information returned by other target task agents according to the demand information;

[0018] The execution result of the subtask is determined according to the response information and the sub-educational task.

[0019] In an optional implementation, an interactive dialogue is established between the target task agents, and the corresponding sub-education task is executed through the natural language dialogue between the target task agents to determine the sub-task execution result, which specifically also includes:

[0020] For each of the target task agents, during the execution of the sub-education task, receiving the demand information sent by other target task agents through the target dialogue;

[0021] The response information is generated according to the information content corresponding to the demand information and the context of the target dialogue through a preset response strategy corresponding to the target task agent.

[0022] In an optional implementation manner, the preset response strategy at least includes adopting a large language model reasoning strategy, a function call execution strategy, and a user input strategy;

[0023] The large language model inference strategy is used to generate a natural language response as the response information through a large language model;

[0024] The function call execution strategy is used to generate the response information by running a predefined code or tool;

[0025] The user input strategy is used to request the user to provide input information as the response information.

[0026] In an optional implementation, in the process of performing the corresponding sub-education task and determining the sub-task execution result through the natural language dialogue between the target task agents, the method further includes:

[0027] The execution error information is prompted through natural language, and the data structure of the subtask execution result is defined, and the corresponding termination condition is set for the target dialogue through programming language.

[0028] In an optional implementation manner, after feeding back the target task response to the user, the method further includes:

[0029] Receive feedback and suggestions from users, and analyze the feedback and suggestions to determine corresponding response adjustment requirements;

[0030] The generation logic of the target task agent corresponding to the generation of the subtask execution result is adjusted according to the response adjustment requirement to determine the target optimization response.

[0031] The present disclosure also provides an educational task processing device based on multi-agent dialogue drive, including:

[0032] A requirement parsing module, used to obtain educational task requirements and parse the educational task requirements into multiple sub-educational tasks;

[0033] An agent allocation module, used to determine the task type corresponding to each of the sub-education tasks, and allocate the sub-education tasks to the corresponding target task agents according to the task type;

[0034] A response generation module, used to establish an interactive dialogue between the target task agents, execute the corresponding sub-education tasks through the natural language dialogue between the target task agents to determine the sub-task execution results, and integrate the sub-task execution results into the target task response;

[0035] The response feedback module is used to feedback the target task response to the user.

[0036] In an optional implementation manner, the response generation module is specifically used to:

[0037] Determine the task execution order according to the task priority and the task dependency, and set the agent execution order between the target task agents according to the task execution order;

[0038] Using a unified dialogue interface to establish an interactive dialogue channel between the target task agents;

[0039] For each of the target task agents, during the execution of the sub-education task, according to the execution requirements corresponding to the sub-education task, a target dialogue is established with other target task agents through the interactive dialogue channel;

[0040] Sending demand information corresponding to the execution demand through the target dialogue, and receiving response information returned by other target task agents according to the demand information;

[0041] The execution result of the subtask is determined according to the response information and the sub-educational task.

[0042] An embodiment of the present disclosure also provides an electronic device, including: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the above-mentioned educational task processing method based on multi-agent dialogue drive, or steps in any possible implementation of the above-mentioned educational task processing method based on multi-agent dialogue drive is performed.

[0043] The embodiments of the present disclosure also provide a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program executes the above-mentioned educational task processing method based on multi-agent dialogue drive, or the steps in any possible implementation of the above-mentioned educational task processing method based on multi-agent dialogue drive.

[0044] The disclosed embodiments also provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the above-mentioned educational task processing method based on multi-agent dialogue drive, or the steps in any possible implementation of the above-mentioned educational task processing method based on multi-agent dialogue drive.

[0045] The disclosed embodiment provides a method and device for processing educational tasks based on multi-agent dialogue drive, which obtains educational task requirements and parses the educational task requirements into multiple sub-educational tasks; determines the task type corresponding to each sub-educational task, and assigns the sub-educational task to the corresponding target task agent according to the task type; establishes an interactive dialogue between the target task agents, executes the corresponding sub-educational task through the natural language dialogue between the target task agents, determines the sub-task execution result, and integrates the sub-task execution result into the target task response; and feeds back the target task response to the user. The distributed agent collaboration mode can be adopted to achieve efficient collaboration through dialogue-driven approach, providing more efficient and flexible technical support for task processing in the field of intelligent education.

[0046] In order to make the above-mentioned objectives, features and advantages of the present disclosure more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following is a brief introduction to the drawings required for use in the embodiments. The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present disclosure and are used together with the specification to illustrate the technical solutions of the present disclosure. It should be understood that the following drawings only illustrate certain embodiments of the present disclosure and should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can also be obtained based on these drawings without creative work.

[0048] Figure 1 A flowchart of an educational task processing method based on multi-agent dialogue drive provided by an embodiment of the present disclosure is shown;

[0049] Figure 2 A flowchart of another method for processing educational tasks based on multi-agent dialogue drive provided by an embodiment of the present disclosure is shown;

[0050] Figure 3 A schematic diagram of an educational task processing device based on multi-agent dialogue drive provided by an embodiment of the present disclosure is shown;

[0051] Figure 4 A schematic diagram of an electronic device provided by an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical scheme and advantages of the embodiments of the present disclosure clearer, the technical scheme in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all of the embodiments. The components of the embodiments of the present disclosure generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present disclosure provided in the drawings is not intended to limit the scope of the present disclosure for protection, but merely represents the selected embodiments of the present disclosure. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present disclosure.

[0053] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0054] The term "and / or" herein only describes an association relationship, indicating that three relationships may exist. For example, A and / or B may represent the following three situations: A exists alone, A and B exist at the same time, and B exists alone. In addition, the term "at least one" herein represents any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C may represent including any one or more elements selected from the set consisting of A, B, and C.

[0055] Research has found that with the rapid development of artificial intelligence technology, intelligent education systems are becoming an important direction in the field of education. These systems provide personalized, real-time teaching and learning support for teachers and students by combining natural language processing, large language models, and intelligent agent technology. However, current intelligent education systems usually rely on a single intelligent agent to perform tasks, which is difficult to meet the diverse needs of complex educational tasks. The functions of a single agent are usually predefined and difficult to flexibly adjust according to dynamically changing task requirements; faced with complex educational scenarios involving diverse subtasks such as content generation, knowledge retrieval, and design optimization, a single agent is difficult to provide comprehensive support.

[0056] Based on the above research, the present disclosure provides a method and device for processing educational tasks based on multi-agent dialogue drive, which obtains educational task requirements and parses the educational task requirements into multiple sub-educational tasks; determines the task type corresponding to each sub-educational task, and assigns the sub-educational task to the corresponding target task agent according to the task type; establishes an interactive dialogue between the target task agents, executes the corresponding sub-educational task through the natural language dialogue between the target task agents, determines the sub-task execution result, and integrates the sub-task execution result into the target task response; and feeds back the target task response to the user. The collaborative mode of distributed agents can be adopted to achieve efficient collaboration through dialogue-driven methods, providing more efficient and flexible technical support for task processing in the field of intelligent education.

[0057] To facilitate understanding of this embodiment, first, a method for processing educational tasks based on multi-agent dialogue-driven disclosed in the embodiment of the present disclosure is introduced in detail. The execution subject of the method for processing educational tasks based on multi-agent dialogue-driven provided in the embodiment of the present disclosure is generally a computer device with certain computing capabilities, and the computer device includes, for example: a terminal device or a server or other processing device, and the terminal device can be a user equipment (User Equipment, UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (Personal Digital Assistant, PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementations, the method for processing educational tasks based on multi-agent dialogue-driven can be implemented by a processor calling a computer-readable instruction stored in a memory.

[0058] See also Figure 1 As shown, it is a flowchart of a method for processing educational tasks based on multi-agent dialogue drive provided by an embodiment of the present disclosure, the method comprising steps S101 to S104, wherein:

[0059] S101. Obtaining educational task requirements, and parsing the educational task requirements into multiple sub-educational tasks.

[0060] In specific implementation, educational task requirements usually come from natural language requests from user input (such as teachers, students or education administrators). Users input clear task requirements through the keyboard, such as: "Generate a courseware on photosynthesis.", "Help me retrieve the latest teaching materials related to calculus.", etc. After obtaining the educational task requirements, the obtained educational task requirements are converted into structured data that can be executed by the system, including task objectives, task types and related parameters.

[0061] Here, we can first perform text preprocessing to clean and normalize user input, such as removing redundant punctuation, correcting spelling errors, and using word segmentation technology to divide task requirements into separate words or phrases. We can also use Named Entity Recognition (NER) to extract key entities in the task, such as topics, target operations, etc.

[0062] For example: Input "Generate a courseware about photosynthesis." After parsing, we can get the subject "Photosynthesis", the corresponding operation used is "Generate", and the result target is "Courseware".

[0063] Among them, educational task requirements usually contain multiple logical units, which need to be further parsed into several independently processable sub-educational tasks, and the tasks are logically layered into main task-subtask-subtask details. For example, main task: generate courseware on photosynthesis; subtask 1: generate text explanation of photosynthesis; subtask 2: create the design and layout of the courseware; subtask 3: verify the accuracy of the courseware content.

[0064] Furthermore, according to the analyzed task objectives and parameters, combined with the predefined task type library, subtasks are divided, for example, content generation subtask: generating subject content; knowledge retrieval subtask: finding relevant reference materials; design optimization subtask: completing visual design and layout; feedback integration subtask: integrating the results of each subtask.

[0065] As a possible implementation method, semantic understanding is performed on educational task requirements to determine the semantic content corresponding to the educational task requirements; the target educational tasks included in the semantic content are identified and the corresponding task logic is determined, and the semantic content is decomposed into multiple sub-educational tasks according to the task logic; the task dependency relationship between the sub-educational tasks is determined according to the task logic, and the corresponding task priority is set for each sub-educational task according to the task dependency relationship.

[0066] In the specific implementation, the core elements of the educational task are extracted through named entity recognition: the task goal is the operation that the user wants to complete (such as generation, retrieval, optimization), the task object is the subject content related to the task (such as "photosynthesis", "plant structure"), and the supplementary information is the additional details provided in the user input (such as "suitable for junior high school students", "presented in the form of courseware").

[0067] For example, for the input task: "Generate a courseware on photosynthesis for junior high school students and provide picture design.", the task goal is generation, the task object is photosynthesis, and the supplementary information is junior high school students, courseware, and picture design.

[0068] Here, semantic content analysis is used to identify the target task in the user input and determine the logical structure of the task. A predefined task type library is used to match the task goals and objects in the semantic content: for example, "generate" corresponds to content generation tasks, "retrieve" corresponds to knowledge retrieval tasks, and "optimize" corresponds to design optimization tasks. Task classification models (such as BERT) can also be trained based on classification models to automatically identify task types in complex inputs.

[0069] Furthermore, complex target tasks are decomposed into multiple sub-tasks to form a logic tree of main tasks and sub-tasks, such as main task - generate courseware, sub-task 1 - generate text content of photosynthesis, sub-task 2 - retrieve relevant image materials, sub-task 3 - design courseware layout, and then the tasks are stepped to determine the order of task execution and parallel relationship, for example, sub-task 1 - sub-task 2 in parallel - sub-task 3 integrates the results.

[0070] Here, the semantic content is disassembled according to the task logic, and the semantic content is further refined into multiple sub-educational tasks through the task logic. Disassembly rules may include: disassembly by task module and disassembly by demand dimension. Among them, disassembly by task module can be to separate functional modules according to task objectives, such as content-related-content generation subtasks, data retrieval-related-retrieval subtasks, and design-related-optimization subtasks. Disassembly by demand dimension can be to generate specific subtasks based on supplementary information, for example, input "suitable for junior high school students" corresponding to the supplement "age-appropriate vocabulary and expressions".

[0071] Among them, the determination of task dependencies is the key to ensuring the order of task execution and consistency of results. Some subtasks must rely on the completion results of other tasks to be executed, for example, after the content generation is completed, the courseware layout design can be carried out. However, some subtasks can be completed independently and executed in parallel, for example, generating text content and retrieving image materials can be carried out in parallel.

[0072] Afterwards, the execution priorities of subtasks are assigned according to the dependencies and logic of the tasks to ensure the efficiency and consistency of task execution.

[0073] Here, task parsing refers to the process of extracting semantic information from the natural language task description input by the user and breaking it down into multiple subtasks. Its purpose is to convert high-level task requirements into independent subtasks that can be executed by the system. Semantic understanding extracts the core information of the task, such as verbs, nouns, modifiers, etc., through natural language processing, understands the natural language input by the user, and extracts the intent, object, operation, and additional conditions of the task. Logical disassembly decomposes the task into multiple subtasks based on the task type and complexity, refines the high-level task into multiple independent subtasks, and defines the execution content for each subtask. The execution order setting is to determine the execution order of each subtask according to the dependency and priority of the task after the subtask parsing is completed, so as to optimize the overall execution efficiency and accuracy of the task, determine the dependency between subtasks, and ensure that the tasks are executed in the correct order. The determination of task priority is to assign a priority to each subtask according to the importance, urgency and user needs of the task.

[0074] Among them, task parsing (semantic understanding and logical decomposition) converts the user's natural language task into multiple independent subtasks, focusing on the semantic understanding of the task, task type identification, and task decomposition. The input is the user's natural language task description, and the output is a structured subtask list. The execution order setting (task dependency and priority) determines the execution order of subtasks, optimizes the execution efficiency and accuracy of tasks, focuses on the dependency between tasks, the priority of tasks, and the execution order of tasks. The input is the parsed subtasks and their dependencies, and the output is the execution order and dependency graph of subtasks.

[0075] S102, determining the task type corresponding to each of the sub-educational tasks, and assigning the sub-educational tasks to corresponding target task agents according to the task type.

[0076] In the specific implementation, the determination of task types aims to map subtasks to clear task categories, so that the system can further call the corresponding agents. The types of educational tasks can be classified according to their functions as: content generation tasks, knowledge retrieval tasks, design optimization tasks, feedback verification tasks, and data analysis tasks. Among them, content generation tasks include generating knowledge explanations, course texts, courseware content, etc. Knowledge retrieval tasks include obtaining relevant information or materials from databases or external resources. Design optimization tasks include optimizing the visual, layout, or multimedia content of courseware. Feedback verification tasks include integrating the results of multiple subtasks and providing verification and quality feedback. Data analysis tasks include processing and analyzing learning progress or personalized learning paths.

[0077] Here, the identification of task types can adopt the identification of comprehensive task semantics, combined with context analysis of semantic content, for example, task goal - extract action keywords (such as "generate", "optimize"); task object - extract subject keywords (such as "photosynthesis", "courseware"); supplementary conditions - extract additional requirements (such as "applicable to junior high school students", "need pictures"); user input - "Generate photosynthesis courseware for junior high school students, including experimental pictures."; task goal - generate; task object - photosynthesis; supplementary conditions - junior high school students, courseware, experimental pictures; determine the task type - content generation task, knowledge retrieval task.

[0078] Among them, the target task agents can be divided into the following categories: content generation agents, knowledge retrieval agents, design optimization agents, feedback verification agents, and other agents. Among them, the content generation agent can generate text content for calling the large language model (LLM). The knowledge retrieval agent can retrieve relevant information in the database, document or vector database. The design optimization agent can handle courseware typesetting, image design or multimedia content optimization. The feedback verification agent can integrate subtask results and provide quality feedback. Other agents can perform data analysis, learning path recommendation, etc.

[0079] In specific implementation, a task agent is an entity with a specific role that can send and receive messages to convey information, such as starting or continuing a conversation. It maintains its internal context based on the messages sent and received, and can be configured with a set of capabilities based on demand, which may be provided by LLM, tools, or human input. These agents can take actions based on pre-programmed behavior patterns.

[0080] Here, the capabilities of the task agent directly affect how it processes and responds to messages, so it can have multiple capabilities, for example: Agents supported by LLM can take advantage of multiple capabilities of advanced LLM, such as role-playing, implicit state inference and progression based on dialogue history, providing feedback, adapting to feedback, and programming. These capabilities can be combined in different ways through new prompting techniques to increase the skills and autonomy of the agent. Enhanced LLM reasoning capabilities such as result caching, error handling, message templates, etc. are also provided through the enhanced LLM reasoning layer.

[0081] Meanwhile, task agents allow humans to participate in agent conversations through human agents supported by humans. Depending on the agent configuration, human input can be solicited at certain turns of the conversation. The default task agent allows configuration of different levels and modes of human participation, such as the frequency and conditions for requesting human input, including the option that humans can choose not to provide input.

[0082] Meanwhile, the task agent supported by the tool has the ability to execute the tool through code execution or function execution. For example, the default task agent can execute the code suggested by LLM, or execute the function call suggested by LLM.

[0083] Optionally, depending on the needs of a specific application, each task agent can be configured with a mix of multiple base backend types to exhibit complex behaviors in multi-agent conversations. This allows agents with specialized capabilities and roles to be easily created by reusing or extending built-in agents.

[0084] S103, establishing an interactive dialogue between the target task agents, executing the corresponding sub-education tasks through the natural language dialogue between the target task agents to determine the sub-task execution results, and integrating the sub-task execution results into the target task response.

[0085] In specific implementation, interactive dialogue is a process in which target task agents exchange information in natural language format, collaboratively complete subtasks and integrate results. This mechanism enables agents to achieve dynamic collaboration, result feedback and optimization through dialogue like humans. Dialogue can adapt to task changes and dynamically adjust the task division and execution process between agents. Through the predefined dialogue structure, accurate information transmission is ensured and multiple rounds of dialogue are supported to gradually refine the task results.

[0086] Here, for the process of establishing a dialogue, the dialogue is triggered and the initial task is assigned to the relevant target task agent. Each agent establishes a dialogue context according to the task description. Each agent registers to the dialogue manager to achieve unified dialogue coordination. The dialogue manager is responsible for routing and scheduling messages, such as using the output of one agent as the input of another agent, and adjusting the dialogue content or participating agents according to context changes during task execution.

[0087] Among them, the dialogue between agents is usually divided into the following types: task instruction dialogue, result feedback dialogue, confirmation and verification dialogue. Among them, the task instruction dialogue can be used to pass the instructions and parameters for task execution, for example, the content generation agent requests supporting data from the knowledge retrieval agent. The result feedback dialogue can be used to feedback the results of task execution and request improvements or adjustments. For example, the design optimization agent proposes modification suggestions based on the results of the content generation agent. Confirmation and verification dialogue can be used to check whether the subtask meets the requirements. For example, the feedback integration agent confirms whether the subtask results are consistent.

[0088] Here, the sender agent generates a natural language task instruction based on the task description and sends it to the target agent. For example, the content generation agent requests the knowledge retrieval agent to provide supporting data, and the instruction content is "Please retrieve experimental case pictures related to photosynthesis." After receiving the instruction, the target agent parses the natural language content and performs the corresponding task operation. The target agent returns the task result to the sender agent in a natural language format. For example, the knowledge retrieval agent returns the result: "3 experimental pictures have been found, please confirm whether more data is needed."

[0089] Preferably, if the result does not meet expectations, the agent requests re-execution of the task or refinement of task parameters through additional dialogue.

[0090] Furthermore, through collaborative dialogues between agents, the results of different subtasks are aggregated, verified, and optimized to generate a complete target task response. After completing the task, each agent sends the result to the feedback integration agent. The integration agent collects all subtask results and organizes them according to task logic. The results of each subtask are checked to ensure data consistency and logical correctness. If a conflict is found, a dialogue between agents is triggered to make adjustments.

[0091] Here, multiple rounds of dialogue are used to further optimize the results, such as adjusting the text content, optimizing the layout, etc. The target task response generated in the end contains the results of all the subtasks after integration.

[0092] The following is an example scenario of completing an educational task through interactive dialogue. Specifically, the user request is "Generate a multimedia courseware on photosynthesis, including pictures and experimental cases." Subtask 1 is to generate a text explanation of photosynthesis (content generation agent), subtask 2 is to retrieve experimental case pictures (knowledge retrieval agent), and subtask 3 is to optimize the layout of the courseware (design optimization agent). The main control module assigns tasks and triggers the content generation agent: "Please generate text content about photosynthesis." Among them, the dialogue between the content generation agent and the knowledge retrieval agent includes the following content. Specifically, the content generation agent: "Please retrieve pictures related to photosynthesis experiments." The knowledge retrieval agent: "Three pictures have been found, do you need more?" The result integration and feedback include the dialogue between the knowledge retrieval agent and the design optimization agent. Specifically, the knowledge retrieval agent: "The following are picture materials, please integrate them into the courseware layout." The design optimization agent: "The pictures have been integrated, do you need to further adjust the layout?" The dialogue between the feedback integration agent and all agents includes the following content. Specifically, the feedback integration agent: "Are content generation, picture retrieval, and courseware design all completed?" All agents confirm completion, generate the final target task response and send it to the user.

[0093] In this way, through natural language dialogue, the agent can adjust the task execution process in real time. Multiple rounds of dialogue make the subtask results more accurate and consistent. By adding agent types or adjusting the dialogue logic, different task scenarios can be quickly adapted.

[0094] As a possible implementation, see Figure 2 As shown, it is a flowchart of another method for processing educational tasks based on multi-agent dialogue drive provided by an embodiment of the present disclosure, the method includes steps S1031 to S1035, wherein:

[0095] S1031 . Determine a task execution order according to the task priorities and the task dependencies, and set an agent execution order between the target task agents according to the task execution order.

[0096] S1032: establishing an interactive dialogue channel between the target task agents using a unified dialogue interface.

[0097] S1033. For each of the target task agents, during the execution of the sub-educational task, according to the execution requirements corresponding to the sub-educational task, a target dialogue is established with other target task agents through the interactive dialogue channel.

[0098] S1034: Sending demand information corresponding to the execution demand through the target dialogue, and receiving response information returned by other target task agents according to the demand information.

[0099] S1035. Determine the execution result of the subtask according to the response information and the sub-education task.

[0100] In specific implementations, task priority is determined based on the following factors: task importance, time sensitivity, and user instructions. Among them, task importance can directly affect the final task result. Usually, tasks with high task importance have higher corresponding priorities, such as content generation tasks over design optimization tasks. Time sensitivity emphasizes that tasks that require quick response are executed first. User instructions give priority to urgent tasks or tasks with special requirements marked by users.

[0101] Here, the dependencies between tasks define the execution logic of the tasks. According to the priorities and dependencies, high-priority tasks without pre-dependencies are executed first, independent tasks are allowed to be executed in parallel, and dependent tasks are executed sequentially.

[0102] Here, we analyze the input-output dependencies between agents. For example, the output of the content generation agent directly affects the input of the design optimization agent. We prioritize agents with lower current loads to improve efficiency. If a task requires collaboration among multiple agents, we need to clarify the order of interactions between agents.

[0103] Furthermore, at the beginning of a task, the main control module assigns a unique identifier to each agent and registers it with the dialogue manager. The dialogue manager dynamically establishes a communication channel between agents, and the dialogue channel supports dynamic adjustment according to task requirements. For example, when an agent requests external data, the channel will temporarily connect to the knowledge retrieval agent.

[0104] Here, the agent generates requirement information based on the task description, clarifies the task objectives, data requirements and expected format. After the target agent completes the task, it returns the result in the form of response information. The request agent parses the response information and decides whether to take further action based on the task logic. The agent determines whether the task is completed and whether the result meets the requirements based on the response information. If the response information contains errors or does not meet the conditions, the agent re-requests adjustment through additional dialogues.

[0105] As a possible implementation method, for each target task agent, during the execution of the sub-education task, the demand information sent by other target task agents is received through the target dialogue; according to the information content corresponding to the demand information and the context of the target dialogue, response information is generated through the preset response strategy corresponding to the target task agent.

[0106] In practice, task agents are equipped with unified conversational interfaces, which enable them to perform conversation-centric computational tasks. These interfaces include functions for sending and receiving messages, as well as a generate reply function, which allows task agents to take actions and generate responses based on received messages.

[0107] Here, when a task agent receives a message from another task agent, it automatically triggers the generation of a reply function and transmits the reply back to the message sender, unless the termination condition of the conversation is reached. The task agent has built-in multiple reply functions, including replies based on LLM reasoning, code execution, function calls, or human input. Users can also register custom reply functions to personalize the interaction mode of the task agent, for example, communicate with other task agents before replying to the sender. Under this mechanism, once the reply function is registered and the conversation is initiated, the conversation flow will unfold naturally without the need for an additional control module to manage the conversation flow.

[0108] Among them, the preset response strategy at least includes adopting a large language model inference strategy, a function call execution strategy and a user input strategy; the large language model inference strategy is used to generate a natural language response as response information through a large language model; the function call execution strategy is used to generate response information by running predefined code or tools; the user input strategy is used to request the user to provide input information as response information.

[0109] Here, execution error messages are prompted through natural language, and the data structure of subtask execution results is limited. The corresponding termination conditions are set for the target dialogue through programming language. Use Python code to set the termination conditions, input mode, and execution logic of the tool, such as setting the upper limit of the number of automatic replies. Users can also register the programmed automatic reply function and use Python code to manage the dialogue process.

[0110] Specifically, through the custom reply function, the task agent can maintain the current conversation while initiating a conversation with other task agents based on the content and context of the current message. LLM will decide whether to trigger a specific function based on the real-time status of the conversation. By exchanging messages with other task agents in function calls, LLM can promote dynamic conversations between multi-task agents.

[0111] Optionally, more advanced dynamic group chat features can be supported with the help of built-in components for managing group chat interactions between multiple agents, such as GroupChatManager, which can dynamically select the next speaker and broadcast his answer to all participating task agents.

[0112] In the specific implementation, during the multi-agent collaboration process, tasks may need to be dynamically adjusted due to the following factors: context changes, resource load, and task priority changes. Among them, context changes refer to changes in the external environment or user needs, which require task reallocation or adjustment of agent roles. Resource load refers to the fact that the load of an agent is too high or the response is delayed, and task allocation needs to be adjusted dynamically. Task priority changes refer to the adjustment of task priorities due to changes in user instructions or task dependencies. To this end, the status of the agent is monitored in real time during the task execution process, including task progress, resource load, execution efficiency, etc., to monitor the task execution status in real time, capture task progress information, and record the performance indicators of each agent, such as execution time, error rate, etc.

[0113] Here, each task agent supports a dynamic adjustment implementation mechanism to dynamically adjust task allocation. This process includes task reallocation and role switching functions. For the task reallocation function, when the agent is overloaded or responds late, the task is reallocated to the backup agent. For the role switching function, according to the task progress and context changes, the agent role is switched or a new agent is added to participate in the collaboration.

[0114] Exemplarily, when executing an image retrieval task, if the knowledge retrieval agent cannot respond in time due to excessive load, the task is dynamically assigned to a backup retrieval agent.

[0115] Furthermore, multiple dialogue modes can be supported according to the complexity of the task and the interaction requirements, including multi-round dynamic dialogue and static single-round dialogue. In the multi-round dynamic dialogue mode, during the task execution, agents can interact with each other in multiple rounds to gradually refine the task results. This mode is suitable for complex tasks that require multiple feedback and iterations, such as content generation and design optimization collaboration. In the static single-round dialogue mode, the task is completed in a single dialogue without multiple feedback, which is suitable for simple tasks to be completed in one go, such as data retrieval tasks.

[0116] Here, in order to implement multi-round dynamic dialogues, task agents share task context information through a unified dialogue interface to ensure the continuity of the dialogue; according to the current task status and context, the agents involved in the dialogue and the message flow are dynamically adjusted.

[0117] Exemplarily, multiple rounds of dynamic dialogues are conducted between the content generation agent and the design optimization agent: the content generation agent generates a first draft and sends it to the design optimization agent, the design optimization agent proposes modification suggestions based on the first draft and returns it to the content generation agent for adjustment, and the final content is generated after multiple rounds of interaction.

[0118] Here, for a static single-round dialogue, after receiving the task instruction, the task agent completes the task within a single dialogue and returns the result. After confirming the result of the single-round dialogue, it confirms the completion of the task or triggers a subsequent task.

[0119] Exemplarily, after receiving the request to "retrieve photosynthesis-related images", the knowledge retrieval agent returns the retrieval result once without further interaction.

[0120] S104: Feedback the target task response to the user.

[0121] In a specific implementation, after the target task response is fed back to the user, the method also includes: receiving feedback suggestions from the user, parsing the feedback suggestions to determine corresponding response adjustment requirements; adjusting the generation logic of the target task agent corresponding to the generated subtask execution results according to the response adjustment requirements, and determining the target optimization response.

[0122] Here, in the multi-agent collaboration process, the following abnormal situations may occur: task failure, suboptimal results, and communication interruption. Among them, task failure refers to the failure of the agent to perform the task due to insufficient resources or incorrect input data. Suboptimal results refer to the results generated by the agent that do not fully meet the task requirements and need further optimization. Communication interruption refers to the interruption of the conversation between agents due to network or system failure.

[0123] Among them, the exception handling mechanism includes: real-time error detection and error reporting mechanism. Among them, real-time error detection means that when the agent executes a task, it detects the task status and output results, and identifies errors or suboptimal situations. The error reporting mechanism means that after the agent detects an exception, it reports the error details to the main control module, including the error type, cause and possible solutions.

[0124] As a possible implementation method, for temporary errors, the system automatically triggers task retry and re-calls the agent to execute the task. If the task retry fails multiple times, the system adopts a backup strategy or simplifies the task objectives to ensure partial completion of the task.

[0125] Exemplarily, in a retrieval task, if the knowledge retrieval agent fails to find relevant images for multiple times, the system will prompt the user to modify the retrieval keywords, or switch to an alternative data source for retrieval.

[0126] As another possible implementation, agent collaborative optimization can be used to collaboratively resolve exceptions. Multiple agents can collaborate through dialogue to jointly resolve erroneous tasks. For example, if the content generated by the content generation agent does not meet the format requirements, the design optimization agent will make correction suggestions and regenerate the content. If the result is suboptimal, the agent will optimize the result through multiple rounds of dialogue until it meets the user's expectations.

[0127] For example, if the content format received by the design optimization agent does not meet the courseware typesetting requirements, the agent provides feedback to the content generation agent for format adjustment, and verifies in subsequent conversations whether the adjusted content meets the requirements.

[0128] The disclosed embodiment provides a method for processing educational tasks based on multi-agent dialogue drive, which obtains educational task requirements and parses the educational task requirements into multiple sub-educational tasks; determines the task type corresponding to each sub-educational task, and assigns the sub-educational task to the corresponding target task agent according to the task type; establishes an interactive dialogue between the target task agents, executes the corresponding sub-educational task through the natural language dialogue between the target task agents to determine the sub-task execution result, and integrates the sub-task execution result into the target task response; and feeds back the target task response to the user. The distributed agent collaboration mode can be adopted to achieve efficient collaboration through dialogue-driven approach, providing more efficient and flexible technical support for task processing in the field of intelligent education.

[0129] Those skilled in the art will appreciate that, in the above method of specific implementation, the order in which the steps are written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of the steps should be determined by their functions and possible internal logic.

[0130] Based on the same inventive concept, the embodiment of the present disclosure also provides an educational task processing device based on multi-agent dialogue drive corresponding to the educational task processing method based on multi-agent dialogue drive. Since the principle of solving the problem by the device in the embodiment of the present disclosure is similar to the above-mentioned educational task processing method based on multi-agent dialogue drive in the embodiment of the present disclosure, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0131] See also Figure 3 , Figure 3 Schematic diagram of an educational task processing device based on multi-agent dialogue drive provided by an embodiment of the present disclosure. Figure 3 As shown in , the educational task processing device 300 based on multi-agent dialogue driving provided by the embodiment of the present disclosure includes:

[0132] A requirement parsing module 310 is used to obtain an educational task requirement and parse the educational task requirement into a plurality of sub-educational tasks;

[0133] An agent allocation module 320 is used to determine the task type corresponding to each of the sub-education tasks, and allocate the sub-education tasks to corresponding target task agents according to the task type;

[0134] A response generation module 330 is used to establish an interactive dialogue between the target task agents, execute the corresponding sub-education tasks through the natural language dialogue between the target task agents to determine the sub-task execution results, and integrate the sub-task execution results into a target task response;

[0135] The response feedback module 340 is used to feedback the target task response to the user.

[0136] For descriptions of the processing flow of each module in the device and the interaction flow between each module, reference may be made to the relevant descriptions in the above method embodiment, which will not be described in detail here.

[0137] The disclosed embodiment provides an educational task processing device based on multi-agent dialogue drive, which obtains educational task requirements and parses the educational task requirements into multiple sub-educational tasks; determines the task type corresponding to each sub-educational task, and assigns the sub-educational task to the corresponding target task agent according to the task type; establishes an interactive dialogue between the target task agents, executes the corresponding sub-educational task through the natural language dialogue between the target task agents, determines the sub-task execution result, and integrates the sub-task execution result into the target task response; and feeds back the target task response to the user. The distributed agent collaboration mode can be adopted to achieve efficient collaboration through dialogue-driven approach, providing more efficient and flexible technical support for task processing in the field of intelligent education.

[0138] Corresponds to Figure 1 The educational task processing method based on multi-agent dialogue driving in the present disclosure embodiment also provides an electronic device 400, such as Figure 4 FIG. 4 is a schematic diagram of the structure of an electronic device 400 provided in an embodiment of the present disclosure, including:

[0139] Processor 41, memory 42, and bus 43; memory 42 is used to store execution instructions, including memory 421 and external memory 422; memory 421 here is also called internal memory, which is used to temporarily store operation data in processor 41 and data exchanged with external memory 422 such as hard disk. Processor 41 exchanges data with external memory 422 through memory 421. When the electronic device 400 is running, the processor 41 communicates with the memory 42 through bus 43, so that the processor 41 executes Figure 1 and Figure 2 The steps of the multi-agent dialogue-driven educational task processing method.

[0140] The present disclosure also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the multi-agent dialogue-driven educational task processing method described in the above method embodiment are executed. The storage medium can be a volatile or non-volatile computer-readable storage medium.

[0141] The disclosed embodiments also provide a computer program product, which includes computer instructions. When the computer instructions are executed by a processor, the steps of the educational task processing method based on multi-agent dialogue drive described in the above method embodiment can be executed. For details, please refer to the above method embodiment, which will not be repeated here.

[0142] The computer program product may be implemented in hardware, software or a combination thereof. In one optional embodiment, the computer program product is implemented as a computer storage medium. In another optional embodiment, the computer program product is implemented as a software product, such as a software development kit (SDK).

[0143] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the several embodiments provided in the present disclosure, it should be understood that the disclosed device and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of the device or unit can be electrical, mechanical or other forms.

[0144] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0145] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0146] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present disclosure, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0147] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present disclosure, which are used to illustrate the technical solutions of the present disclosure, rather than to limit them. The protection scope of the present disclosure is not limited thereto. Although the present disclosure is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed in the present disclosure, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims

1. A method for processing educational tasks based on multi-agent dialogue driving, characterized in that: include: Obtaining educational task requirements, and parsing the educational task requirements into multiple sub-educational tasks; Determine the task type corresponding to each of the sub-education tasks, and assign the sub-education tasks to corresponding target task agents according to the task type; Establishing an interactive dialogue between the target task agents, executing the corresponding sub-education tasks through the natural language dialogue between the target task agents to determine the sub-task execution results, and integrating the sub-task execution results into the target task response; The target task response is fed back to the user.

2. The method according to claim 1, characterized in that The educational task requirements are parsed into multiple sub-educational tasks, specifically including: Performing semantic understanding on the educational task requirements to determine the semantic content corresponding to the educational task requirements; Identify the target educational task included in the semantic content and determine the corresponding task logic, and decompose the semantic content into a plurality of sub-educational tasks according to the task logic; The task dependency relationship between the sub-educational tasks is determined according to the task logic, and a corresponding task priority is set for each sub-educational task according to the task dependency relationship.

3. The method according to claim 2, characterized in that Establishing an interactive dialogue between the target task agents, executing the corresponding sub-education task through the natural language dialogue between the target task agents to determine the sub-task execution result, specifically including: Determine the task execution order according to the task priority and the task dependency, and set the agent execution order between the target task agents according to the task execution order; Using a unified dialogue interface to establish an interactive dialogue channel between the target task agents; For each of the target task agents, during the execution of the sub-education task, according to the execution requirements corresponding to the sub-education task, a target dialogue is established with other target task agents through the interactive dialogue channel; Sending demand information corresponding to the execution demand through the target dialogue, and receiving response information returned by other target task agents according to the demand information; Determine the subtask execution result based on the response information and the sub-educational task.

4. The method according to claim 3, characterized in that Establishing an interactive dialogue between the target task agents, executing the corresponding sub-education task through the natural language dialogue between the target task agents to determine the sub-task execution result, specifically also includes: For each of the target task agents, during the execution of the sub-education task, receiving the demand information sent by other target task agents through the target dialogue; The response information is generated according to the information content corresponding to the demand information and the context of the target dialogue through a preset response strategy corresponding to the target task agent.

5. The method according to claim 4, characterized in that: The preset response strategy at least includes adopting a large language model reasoning strategy, a function call execution strategy and a user input strategy; The large language model inference strategy is used to generate a natural language response as the response information through a large language model; The function call execution strategy is used to generate the response information by running a predefined code or tool; The user input strategy is used to request the user to provide input information as the response information.

6. The method according to claim 3, characterized in that In the process of performing the corresponding sub-education task and determining the sub-task execution result through the natural language dialogue between the target task agents, the method further includes: The execution error information is prompted through natural language, and the data structure of the subtask execution result is defined, and the corresponding termination condition is set for the target dialogue through programming language.

7. The method according to claim 1, characterized in that After feeding back the target task response to the user, the method further includes: Receive feedback and suggestions from users, and analyze the feedback and suggestions to determine corresponding response adjustment requirements; The generation logic of the target task agent corresponding to the generation of the subtask execution result is adjusted according to the response adjustment requirement to determine the target optimization response.

8. An educational task processing device based on multi-agent dialogue drive, characterized in that: include: A requirement parsing module, used to obtain educational task requirements and parse the educational task requirements into multiple sub-educational tasks; An agent allocation module, used to determine the task type corresponding to each of the sub-education tasks, and allocate the sub-education tasks to the corresponding target task agents according to the task type; A response generation module, used to establish an interactive dialogue between the target task agents, execute the corresponding sub-education tasks through the natural language dialogue between the target task agents to determine the sub-task execution results, and integrate the sub-task execution results into the target task response; The response feedback module is used to feedback the target task response to the user.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the steps of the educational task processing method based on multi-agent dialogue drive are performed as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of the educational task processing method based on multi-agent dialogue drive as described in any one of claims 1 to 7.

Images