Task processing method and device based on large model and agent

Abstract

This disclosure provides a task processing method, apparatus, and intelligent agent based on a large model, relating to the field of artificial intelligence technology, particularly to AI interaction fields such as AI assistants and AI healthcare. The specific implementation method of the task processing method based on the large model includes: using the large model, based on the requirement information of the task to be processed and the information of multiple candidate intelligent agents, determining the target intelligent agent from multiple candidate intelligent agents to process the task to be processed; wherein the information indicates the task types that each of the multiple candidate intelligent agents can process and the processing status of the tasks that each of the multiple candidate intelligent agents needs to process; and invoking the target intelligent agent to process the task to be processed to obtain processing result information.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, particularly to AI interaction fields such as AI assistants and AI healthcare, and specifically to task processing methods, devices, and intelligent agents based on large models. Background Technology

[0002] In fields such as human-computer interaction, task assistance, and information retrieval, large language models need to perform multiple tasks, including information retrieval and content generation, based on task requirements. As the complexity of application scenarios increases, related technologies struggle to simultaneously meet users' dual demands for task processing efficiency and result accuracy. Summary of the Invention

[0003] This disclosure provides a task processing method, apparatus, and intelligent agent based on a large model.

[0004] According to one aspect of this disclosure, a task processing method based on a large model is provided, comprising: using the large model, determining a target agent for processing the task from multiple candidate agents based on the requirement information of the task to be processed and the information of multiple candidate agents; wherein the information indicates the task type that each of the multiple candidate agents can process and the processing status of the task to be processed by each of the multiple candidate agents; and invoking the target agent to process the task to be processed to obtain processing result information.

[0005] According to another aspect of this disclosure, a task processing apparatus based on a large model is provided, including a determination module and a processing module.

[0006] The determination module is used to utilize a large model to determine the target agent for handling the task from multiple candidate agents based on the requirement information of the task to be processed and the information of multiple candidate agents; wherein, the information indicates the task types that each of the multiple candidate agents can handle and the processing status of the task that each of the multiple candidate agents needs to process.

[0007] The processing module is used to call the target intelligent agent to process the task to be processed and obtain the processing result information.

[0008] According to another aspect of this disclosure, an intelligent agent is provided, comprising: an input module, a processing module, and an output module.

[0009] The input module is used to receive the requirement information of the task to be processed and the information of multiple candidate agents; wherein, the information indicates the type of task that each candidate agent can process and the processing status of the task that each candidate agent needs to process.

[0010] The processing module is used to determine the target task based on the impact information and initial code received by the input module, determine the target large model based on the target task, and obtain the processing result information by calling the target large model to execute the information processing methods described above.

[0011] The output module is used to output the processing results obtained by the processing module.

[0012] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above.

[0013] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform the methods described above.

[0014] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method described above.

[0015] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0016] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0017] Figure 1 This illustration schematically shows an exemplary system architecture for applying large-model-based task processing methods and apparatus according to embodiments of the present disclosure;

[0018] Figure 2 A flowchart illustrating a large-model-based task processing method according to an embodiment of the present disclosure is shown schematically.

[0019] Figure 3 This illustration schematically shows a diagram of determining the moment when a large model participates in an interaction in a multi-person interaction scenario according to an embodiment of the present disclosure;

[0020] Figure 4 This illustration schematically depicts the requirement to determine the task to be processed by analyzing task associations in a multi-person interaction scenario according to an embodiment of the present disclosure;

[0021] Figure 5This illustration schematically shows a diagram of determining a target agent by analyzing the processing pattern of a task to be processed according to an embodiment of the present disclosure;

[0022] Figure 6A This illustration schematically shows a task processing chain determined according to an embodiment of the present disclosure when the task to be processed is related to a historical task and the task requirements have not been updated;

[0023] Figure 6B This illustration schematically shows a task processing chain determined according to an embodiment of the present disclosure when the task to be processed is related to a historical task and the task requirements have been updated.

[0024] Figure 6C This illustration schematically shows a task processing chain determined according to an embodiment of the present disclosure when the task to be processed is unrelated to historical tasks;

[0025] Figure 6D The illustration schematically depicts a task processing chain determined according to embodiments of the present disclosure when there are no data dependencies between multiple subtasks to be processed.

[0026] Figure 7 The diagram illustrates the scheduling relationships between modules in an exemplary system suitable for implementing a large-model-based task processing method according to embodiments of the present disclosure.

[0027] Figure 8 A block diagram of a large-model-based task processing apparatus according to an embodiment of the present disclosure is shown schematically.

[0028] Figure 9 A block diagram of an intelligent agent according to an embodiment of the present disclosure is illustrated schematically; and

[0029] Figure 10 A block diagram of an electronic device suitable for implementing a large-model-based task processing method according to an embodiment of the present disclosure is shown schematically. Detailed Implementation

[0030] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0031] With the development of Large Language Model (LLM) technology, generative artificial intelligence systems trained on large-scale corpora have been widely applied in various scenarios such as question answering, content generation, task assistance, information retrieval, and human-computer interaction. Large language models typically possess strong natural language understanding, reasoning, and generation capabilities, enabling them to generate coherent and context-aware text responses based on user input, and are gradually becoming a core foundational capability of next-generation intelligent interaction systems.

[0032] With the continuous enhancement of large language model capabilities, ChatBot products have become one of the important application forms of generative artificial intelligence. ChatBot products typically refer to intelligent dialogue products that provide users with question-and-answer, assistant services, task processing, or information support based on natural language interaction. These products can be deployed in standalone chat interfaces, office collaboration platforms, customer service scenarios, social scenarios, or business systems to handle user questions, requests, or tasks and return processing results in a dialogue format.

[0033] Furthermore, as application complexity increases, the approach of a single model directly responding to users is gradually evolving towards "agent-based" systems. An agent is typically an intelligent processing unit built to fulfill specific responsibilities, capability boundaries, or task objectives. Based on large language model capabilities, combined with role settings, tool invocation, memory mechanisms, task status, and environmental information, it performs functions such as question answering, analysis, planning, execution, or collaboration. In practical systems, different agents can undertake different types of responsibilities, such as information retrieval, content generation, task scheduling, tool execution, domain-specific question answering, or role-based interaction, thus forming a product form where multiple agents work collaboratively.

[0034] Currently, with the development of multi-agent systems, more and more products are beginning to introduce multiple agents into scenarios such as multi-user interaction, collaborative work, social interaction, and complex task processing, enabling multiple users and multiple agents to participate in communication and task advancement within the same dialogue space. In such scenarios, the system not only needs to understand the user input itself, but also needs to continuously handle complex situations such as the interweaving of information from multiple parties, the evolution of task objectives, role collaboration switching, and dynamic changes in context. Therefore, higher demands are placed on the agent scheduling capability, task management capability, and collaborative decision-making capability.

[0035] However, the relevant examples are usually still dominated by a question-and-answer interaction between a single user and a single model. This interaction paradigm is relatively static and difficult to adapt to complex dialogue scenarios involving multiple participants, multiple roles, and concurrent task evolution. Although some products already support triggering the corresponding agent's response by having the user actively specify a specific agent in a multi-person interaction environment, this type of solution still relies on the user explicitly specifying the target agent. It lacks the system's ability to autonomously understand the multi-person interaction context and dynamically decide the optimal participating agent. Therefore, in complex multi-person interaction scenarios, problems such as inaccurate agent selection, unreasonable response chains, and high user workload are likely to occur.

[0036] Especially in multi-user, multi-agent interactive scenarios, the dialogue topics are often in a continuous evolutionary state. Different user information may have supplementary, corrective, merged, continued, or cross-dependent relationships. Intermediate responses from agents may also have a reverse impact on subsequent task breakdown, collaborative division of labor, and scheduling decisions. In this scenario, if the traditional scheduling method of manual user specification or static rule triggering is still used, the following problems will usually occur: the system has difficulty automatically identifying the most suitable agent to intervene based on the real-time context, resulting in a discrepancy between the responding agent and the task requirements.

[0037] In view of this, this disclosure provides a task processing method based on a large model. Based on the requirement information of the task to be processed, and according to the task types that each candidate agent can handle and the processing status of the tasks each candidate agent needs to process, the most suitable target agent for processing the task is determined from multiple candidate agents. Compared with the traditional scheduling methods in related examples that rely on manual user specification or static rule triggering, this method reduces the deviation between the responding agent and the task requirements, further improving the accuracy of the processing result information. Since the processing status of the tasks each candidate agent needs to process is also considered when determining the target agent, the processing efficiency of the task to be processed is further improved, meeting the user's dual requirements for accuracy and efficiency in interactive scenarios.

[0038] Figure 1 The illustration schematically shows an exemplary system architecture for applying large-model-based task processing methods and apparatus according to embodiments of the present disclosure.

[0039] It is important to note that Figure 1The examples shown are merely examples of system architectures applicable to embodiments of this disclosure, intended to help those skilled in the art understand the technical content of this disclosure, but do not imply that embodiments of this disclosure cannot be used in other devices, systems, environments, or scenarios. For example, in another embodiment, an exemplary system architecture applicable to the large-model-based task processing method and apparatus may include a terminal device, but the terminal device may implement the large-model-based task processing method and apparatus provided by embodiments of this disclosure without interacting with a server.

[0040] like Figure 1 As shown, the system architecture 100 according to this embodiment may include: a terminal device 101, an interaction platform 102, and a candidate intelligent agent 103.

[0041] Terminal device 101 can be various electronic devices with a display screen and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0042] Users can use terminal device 101 to interact with interactive platform 102 via a network to receive or send information, etc. Various communication client applications can be installed on terminal device 101, such as knowledge reading applications, web browser applications, search applications, instant messaging tools, email clients and / or social platform software, etc. (for example only).

[0043] The interactive platform 102 can be configured with a server that can provide various services to transform the information sent by the user into the requirement information for the task to be processed. Figure 1 The progress indicators for each agent can represent the task status of each agent at the current moment. For example, the black area can represent the number of tasks waiting to be processed by each agent. Figure 1 As shown, the progress indicator for agent A2 is blank, indicating that agent A2 currently has no tasks pending and is in an idle state. Therefore, based on the demand information of the tasks pending, agent A2, which matches the demand and is in an idle state, can be matched from candidate agents 103. Agent A2 can then be invoked to process the tasks pending, output task processing results, and return the results to the interaction platform 102. From the interaction platform 102, the results are then returned to the terminal device 101 via the network, thus realizing the interaction between the user and the agent.

[0044] It should be understood that Figure 1 The number of terminal devices 101, interaction platforms 102, and candidate intelligent agents 103 shown in the diagram is merely illustrative. Depending on implementation needs, any number of terminal devices, interaction platforms, and candidate intelligent agents can be included.

[0045] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision, disclosure, and application of any type of information, such as user personal information, comply with the provisions of relevant laws and regulations, necessary confidentiality measures have been taken, and they do not violate public order and good morals.

[0046] In the technical solution disclosed herein, the user's authorization or consent is obtained before acquiring or collecting the user's personal information.

[0047] Figure 2 A flowchart illustrating a large-model-based task processing method according to an embodiment of the present disclosure is shown schematically.

[0048] like Figure 2 As shown, the task processing method 200 may include operations S210 to S220.

[0049] In operation S210, using a large model, based on the requirements of the task to be processed and the information of multiple candidate agents, the target agent for processing the task to be processed is determined from multiple candidate agents.

[0050] In operation S220, the target intelligent agent is invoked to process the task to be processed and the processing result information is obtained.

[0051] The requirement information for the task to be processed can be generated in the interaction scenario between the user and the intelligent agent. For example, it can be determined based on the interactive information input by the user in the interactive scenario, such as: "Please write a development prospect analysis report for a certain industry." Alternatively, it can be determined based on the current information input by the user in the interactive scenario at the current moment, combined with historical interaction information from historical time periods. For example, in an AI (Artificial Intelligence) medical scenario, an AI doctor can interact with the patient in multiple rounds, and after initially determining the patient's condition based on the information from these multiple rounds of interactions, the requirement information for the task to be processed can be determined based on generating recommended treatment plans based on the patient's condition.

[0052] It should be noted that this interaction scenario can include interaction between a single user and multiple agents, interaction between multiple users and multiple agents, or interaction between a single user or multiple users and a single agent.

[0053] The number of target agents can depend on the requirements of the task to be processed. For example, for the requirement of "Please write a report analyzing the development prospects of a certain industry", the target agents should at least include an information retrieval agent and a text generation agent. For example, for the requirement of "What is the weather like today", the target agents could be information query agents.

[0054] The information from multiple candidate agents indicates the types of tasks that each candidate agent can handle and the processing status of the tasks that each candidate agent needs to handle.

[0055] For example, in multi-user and multi-agent interaction scenarios, the interaction information input by multiple users can involve different vertical domains. Therefore, the task types that multiple candidate agents can handle can be matched with the requirements of the task to be processed. These task types can include, but are not limited to, those based on industry domains, such as the medical field or materials field. They can also include, but are not limited to, those based on data modalities, such as images, text, and multimodal data. Furthermore, they can be based on interaction roles, such as experts or designers. Finally, they can be based on any combination of industry domains, data modalities, and interaction roles.

[0056] In scenarios involving multiple users and multiple agents, the interactive information input by users changes continuously with the topic of interaction, the requirements of the task to be processed also evolves, and the context of multiple rounds of interaction is dynamically reconstructed. The target agent dynamically matched based on the requirements of the task to be processed needs to meet the real-time requirements in such dynamic interaction scenarios.

[0057] Therefore, in addition to the task role, when determining the target agent, the processing status of the tasks to be processed by each of the multiple candidate agents can also be considered. This processing status may include, but is not limited to, the number of tasks to be processed by each candidate agent, the stage of task execution, and the number of tasks waiting to be processed.

[0058] The target agent can be the agent whose task type matches the demand information and is in an idle state or whose expected waiting time for the task to be processed is the shortest.

[0059] In this embodiment of the disclosure, a target intelligent agent can be invoked to process a task to be processed and obtain processing result information. For example, the requirement information of the task to be processed may be "help me write an essay about autumn". The target intelligent agent invoked may be a text generation intelligent agent that is in an idle state or has the shortest expected waiting time for the task to be processed. The requirement information of the task to be processed is input into the text generation intelligent agent, and the processing result information obtained may be text information describing autumn in an essay style.

[0060] Compared to the traditional methods in related examples that rely on users to explicitly specify the agent or use a single model for unified responses in multi-person interaction scenarios, this disclosure proposes a method that can automatically match the most suitable target agent for interaction based on real-time demand information. Without requiring users to manually specify the target agent, the method can autonomously determine the target agent to perform the task by combining information content, contextual semantics, the task types that candidate agents can handle, and the processing status of the tasks to be processed by the candidate agents. This significantly improves the smoothness and intelligence of interaction in multi-person interaction scenarios.

[0061] This disclosure provides a task processing method based on a large model. Based on the requirements of the task to be processed, it determines the most suitable target agent from multiple candidate agents based on the task types each agent can handle and the processing status of the tasks each agent needs to process. Compared to traditional scheduling methods in related examples that rely on manual user specification or static rule triggering, this method reduces the deviation between the responding agent and the task requirements, further improving the accuracy of the processing results. Since the processing status of the tasks each candidate agent needs to process is considered when determining the target agent, the processing efficiency of the task to be processed is further improved, satisfying the user's dual requirements for accuracy and efficiency in interactive scenarios.

[0062] In multi-person interaction scenarios, such as AI-powered healthcare, where family members accompany patients during online consultations, interactions may occur between family members and patients, between family members and the AI ​​agent, and between patients and the AI ​​agent. Therefore, in such complex interaction scenarios, accurately defining the moments when the AI ​​agent participates in the interaction based on the dynamic changes in semantics within the interaction context is crucial.

[0063] Figure 3 The illustration shows a schematic diagram of determining the moment when a large model participates in an interaction in a multi-person interaction scenario according to an embodiment of the present disclosure.

[0064] This interactive scenario can include multiple rounds of conversations, where the interaction objects in each round can be the same or different.

[0065] Therefore, embodiments of this disclosure utilize a large model to perform correlation analysis on current information and historical interaction information to obtain interaction intent information.

[0066] like Figure 3 As shown, taking an AI-powered healthcare scenario as an example, historical interaction information may include the following conversation content: User A: "I've had a headache for a while." User B: "His headache gets worse every night, even affecting his sleep." User A: "I've been taking painkillers for a while, but they're not working. What should I do?"

[0067] Current information can refer to user input information received at the current moment, such as user A's input: "I've been taking painkillers for a while, but they're not working. What should I do?" Historical interaction information can refer to multi-turn conversation information received before the current moment. Multi-turn conversations can include multiple users, such as patients and family members, and can also include information output by other agents participating in the multi-turn conversation.

[0068] Large models can be used to perform correlation analysis on current information and historical interaction information to determine interaction intent. This interaction intent indicates the object to be interacted with in response to the current information. During the correlation analysis, the large model can determine the interaction timing of the agent parameters based on the analysis results, such as whether the current information is a response to or supplement to historical interaction content, an opinion expressed on historical interaction content, or simply an expression of emotion.

[0069] For example, when the analysis results indicate that the current information is a response to or supplement to historical interaction content, it can be determined that the intervention of an agent is required.

[0070] For example, User A says, "I've been taking painkillers for a while, but they're not working. What should I do?" By analyzing the context of historical interactions, we can determine that this message is a response to historical interactions and that there is a clear question field. This indicates that the current moment is the moment of interaction between the agent and the parameters.

[0071] Current information can also refer to the information received at the current moment from the agent in response to historical interaction information. A large model can be used to analyze whether the current information requires user supplementation or a user response; if so, agent intervention is unnecessary. For example, the message: "If we describe pain levels on a scale of 1-7, with higher numbers indicating higher severity, what level do you think your current pain level is?" indicates that a user response is required, therefore agent intervention is unnecessary. It can be determined that the object to be interacted with is not an agent, and operation S301 is executed to wait for the next round of interaction information.

[0072] For example, a large model can be used to analyze whether the current information requires assistance from other intelligent agents. If so, the current moment can be determined as the moment for parameter interaction with other intelligent agents. For example, the information "Your current condition suggests that a consultation with an ENT specialist is needed" indicates that "an AI intelligent agent from the ENT department needs to participate in the interaction." It can be determined that the object to be interacted with is an intelligent agent, and operation S302 is executed to call the large model to analyze the current needs.

[0073] In response to the determination that the object to be interacted with is an intelligent agent, the system generates demand information based on current information and historical interaction information using a large model.

[0074] In interactive scenarios involving multiple users and multiple agents sharing common parameters, large models can accurately identify dynamic semantic changes in the interactive scenario by performing correlation analysis on current information and historical interaction information. This further improves the accuracy of when agents participate in the interaction, and compared to relying on users to manually specify agents, it further improves the smoothness of the interaction process and enhances the user experience.

[0075] In interactive scenarios involving multiple users and agents sharing common parameters, users typically express their complete needs by inputting multiple semantically incomplete pieces of information. When a user continuously sends multiple semantically relevant messages, related examples can easily fragment them into multiple independent tasks, leading to redundant scheduling and wasted computational resources.

[0076] Therefore, in response to determining that the object to be interacted with is an intelligent agent, the large model generates demand information based on current information and historical interaction information. This embodiment of the present disclosure may include the following steps: in response to determining that the object to be interacted with is an intelligent agent, the large model is used to analyze the current information and historical interaction information to generate task association information between the task to be processed and the historical tasks; and the large model is used to generate demand information based on the task association information.

[0077] Figure 4 The illustration shows a schematic diagram of determining the needs of tasks to be processed by analyzing task associations in a multi-person interactive scenario according to an embodiment of the present disclosure.

[0078] In this embodiment of the disclosure, current information is associated with the task to be processed. Historical tasks are associated with historical interaction information. Task association information indicates the degree of association between historical tasks and the task to be processed.

[0079] like Figure 4 As shown, historical task requirements and current information corresponding to historical interaction information can be input into the large model to determine whether historical tasks are related to tasks to be processed.

[0080] For example, historical interaction information could include "Please draw a propaganda poster for me." The associated historical task could be "Generate an image for propaganda purposes." The current information can be received during the execution of a historical task, while the historical task is waiting to be executed, or after the historical task has been completed. The current information could include "The image is primarily in a realistic style." The associated pending task could be "Generate an image in a realistic style." The connection between the current task and the historical task can be determined by the context within the interaction content.

[0081] Therefore, by utilizing a large model and updating historical task requirements based on task-related information, current information can be used as supplementary requirements. The generated task requirements could include "generating images for propaganda purposes, primarily in a realistic style." Figure 4 As shown, when the task to be processed is associated with a historical task, the generated requirement information is the requirement associated with the historical task.

[0082] By analyzing the correlation between the task to be processed and historical tasks based on current information and historical interaction information, the aggregation of multiple semantically related task requirements is realized. When the task to be processed is related to historical tasks, the current information is used as a supplement to the requirements of historical tasks. This reduces the probability that related tasks are identified as multiple independent tasks due to excessively fine information granularity, reduces the waste of computing resources caused by repeated task processing, further improves the utilization rate of computing resources, and improves the accuracy and completeness of task requirement identification.

[0083] In some embodiments, generating requirement information using a large model based on task association information and historical interaction information may include the following operations: in response to determining that the task association information indicates that the task to be processed is associated with a historical task, extracting the processing status information of the historical task and the collaboration requirement information of the historical task from the historical interaction information; and generating requirement information based on the processing status information of the historical task and the collaboration requirement information of the historical task using the large model.

[0084] The processing status information of historical tasks can indicate the processing stage of historical tasks, such as: waiting stage, execution stage, or completion stage.

[0085] For example, a historical task could be an image generation task, and historical interaction information could include "Is it not finished yet?" Based on historical interaction information, the processing status of the historical task can be determined as either a waiting stage or an execution stage. Using a large model, based on the processing status information and collaborative requirement information of historical tasks, requirement information is generated. For example, requirement information could be "A certain image generation task needs to be executed as soon as possible," in order to trigger the determination of whether to schedule the historical task to other agents for execution by identifying the current task processing status of the target agent.

[0086] Collaboration requirement information from historical tasks can indicate whether the task requires assistance from other intelligent agents during execution. For example, historical interaction information might include phrases like "Your condition requires a consultation with an ENT specialist." Based on this historical interaction information, it can be determined whether other intelligent agents are needed to assist in task execution. Using a large model, requirement information is generated based on the processing status information and collaboration requirement information of historical tasks. For example, the requirement information might be "An ENT AI agent and a neurology AI agent collaborate to generate a user treatment plan," thus triggering the scheduling of the ENT AI agent to participate in the execution of the historical task.

[0087] In some embodiments, generating requirement information using a large model based on task association information and historical interaction information may include the following operations: in response to determining that the task association information indicates that the task to be processed is not associated with historical tasks, generating requirement information based on current information using a large model.

[0088] Historical interaction information can include "What's the weather like today?". A historical task associated with this information could be "Check the weather." Current information can include "Can I take protein powder while exercising?". A pending task associated with this information could be "Determine if I can take functional foods during a workout." The context within the related interaction content determines that the current task is unrelated to historical tasks.

[0089] Therefore, by utilizing a large model based on task-related information, the current information is used as the requirement information for the tasks to be processed. For example... Figure 4 As shown, when the task to be processed is not related to historical tasks, the generated requirement information is the requirement related to the current information.

[0090] By analyzing the correlation between historical tasks and tasks to be executed, requirement information is generated only when it is determined that there is no correlation between the task to be processed and historical tasks. This reduces the interference of frequent fine-grained information in a short period of time on task requirement identification in complex interaction scenarios.

[0091] According to embodiments of this disclosure, using a large model, based on the requirement information of the task to be processed and the information of multiple candidate agents, determining a target agent for processing the task to be processed from multiple candidate agents may include the following operations: analyzing the requirement information of the task to be processed using the large model to obtain the processing mode of the task to be processed; and determining a target agent that matches the processing mode from multiple candidate agents based on the information of multiple candidate agents.

[0092] Figure 5 The illustration schematically shows a diagram of determining a target agent by analyzing the processing pattern of a task to be processed according to an embodiment of the present disclosure.

[0093] like Figure 5 As shown, the large model is used to analyze the requirements information of the task to be processed, and the processing pattern of the task to be processed is obtained. This processing pattern indicates whether the task to be processed includes multiple sub-tasks that have data dependencies on each other.

[0094] For example, the requirement information for the task to be processed could be "to generate a development analysis report for a certain industry from year XX to year YY". A large model can be used to analyze the requirement information for the task to be processed, breaking it down into multiple sub-requirements using a logical thought process. For example: first, query relevant development data for a certain industry from year XX to year YY; then, perform development trend analysis on the relevant data; and finally, write a development analysis report based on the results of the trend analysis.

[0095] Then, based on these sub-requirements and the processing status of the tasks to be processed by each of the multiple candidate agents, the target agent that can meet each sub-requirement can be determined from the candidate agents, such as: data query agent, data analysis agent, text generation agent, etc.

[0096] Since the data analysis agent analyzes data derived from the query results of the data query agent, and the text generation agent's writing is based on the analysis results of the data analysis results agent, it can be determined that the various sub-tasks to be processed, associated with each sub-requirement, have data dependencies on each other. Therefore, the multiple target agents matching the processing pattern also have data dependencies on each other.

[0097] like Figure 5 As shown, the output of agent A1 is input to agents A2 and A3 respectively, and the output of agent A2 is then returned to agent A1. Then, the output of agent A1 is input to agent A4. Finally, the output of agent A4 and the previous output of agent A3 are input together to agent A1, outputting the processing result information of the task to be processed. This forms a processing chain for the task to be processed, in which multiple agents jointly participate in the execution. This processing chain indicates the processing logic relationship between multiple target agents, such as: cooperation order, task switching conditions, or task switching rules.

[0098] For example, the requirement information for the task to be processed could be "to conduct a discussion with the user on topic XX". During the discussion, different fields and perspectives may be involved, resulting in different viewpoints through multiple rounds of interactive conversations. These viewpoints have weak data dependencies on each other. Therefore, the processing mode obtained by analyzing this type of requirement information can be an exploratory task. The multiple sub-tasks included in this type of task do not form a logically clear processing link. Therefore, it is necessary to exploratoryly determine the intelligent agents participating in the interaction based on the interaction content.

[0099] like Figure 5 As shown, based on the intermediate processing results output by agent A1 determined in the first round of interaction, and combined with the requirement information, agent A to participate in the next round of interaction can be re-determined from the candidate agents. n .

[0100] This exploratory processing mode has no fixed processing chain and can flexibly invoke the intelligent agent at any time according to the dynamic changes in the context during the interaction process.

[0101] Based on demand information analysis, the data dependencies between the multiple subtasks included in the task to be processed are determined, and the processing mode of the task to be processed is determined. This provides an agent scheduling mechanism with different processing modes for complex interaction scenarios. This mechanism simultaneously meets the processing needs of both deterministic and exploratory tasks in the processing chain.

[0102] In some embodiments, the task to be processed includes multiple sub-tasks to be processed; determining a target agent that matches the processing mode from multiple candidate agents based on information from multiple candidate agents may include the following operations: in response to determining that the data dependency between multiple sub-tasks to be processed is greater than a first predetermined threshold, determining a target agent from multiple candidate agents based on information from multiple candidate agents; and determining the processing logic relationship between multiple target agents based on the data dependency relationship between multiple sub-tasks to be processed.

[0103] For example, the data dependencies between multiple subtasks can be determined by breaking down the requirements of the task to be processed into sub-requirements of multiple subtasks and using a large model to perform semantic analysis on the sub-requirements of multiple subtasks.

[0104] Data dependency characterizes the degree to which a subtask depends on the processing results of other subtasks during its own execution. A higher dependency indicates a clearer processing chain for the subtask. The first predetermined threshold can be pre-configured based on actual scenario requirements. When the data dependency between multiple subtasks exceeds the first predetermined threshold, it can be determined that the subtask includes multiple subtasks with data dependencies on each other.

[0105] Therefore, for each sub-task to be processed, the target agent can be determined from multiple candidate agents based on its respective sub-requirements and information from multiple candidate agents. In this scenario, the target agent can include multiple agents. For example, for the task of generating an industry analysis report, the target agents determined should at least include a query agent, an analysis agent, and a text generation agent.

[0106] Then, based on the data dependencies between multiple subtasks to be processed—for example, the analysis task depends on the query results of the query task, and the generation task depends on the analysis results of the analysis task—the processing logic relationships between multiple target agents can be determined. This processing logic relationship can characterize the logical links between each subtask to be processed; therefore, the scheduling order of multiple subtasks to be processed can be determined according to this processing logic relationship. For example, the data query subtask can be executed first, then the data analysis subtask, and finally the text generation subtask.

[0107] For tasks involving multiple subtasks with data dependencies on each other, the scheduling order of the target agents is pre-arranged based on the data dependencies between the subtasks, which further improves the efficiency of cooperation between multiple agents and thus improves the processing efficiency of the task.

[0108] In some embodiments, the task to be processed includes multiple sub-tasks to be processed; the target agent includes a first target agent and a second target agent; determining the target agent that matches the processing mode from the multiple candidate agents based on the information of the multiple candidate agents may include the following operations: in response to determining that the data dependency between the multiple sub-tasks to be processed is less than or equal to a first predetermined threshold, determining the first target agent from the multiple candidate agents based on the information of the multiple candidate agents; invoking the first target agent to process the first sub-task among the multiple sub-tasks to be processed to obtain intermediate processing result information; and determining the second target agent from the multiple candidate agents based on the intermediate processing result information and the requirement information.

[0109] The task type of the first subtask to be processed matches the task type of the first target agent, and the task type of the second subtask to be processed matches the task type of the second target agent.

[0110] For example, the task to be processed can be a topic discussion task. This type of task is an open-ended task, and as the discussion deepens and the content evolves, it drives dynamic changes in the context of the interaction. Therefore, the data dependencies between the multiple subtasks included in this type of task are weak or non-existent.

[0111] In this embodiment of the disclosure, if the data dependency between multiple subtasks to be processed is less than or equal to a first predetermined threshold, the task to be processed is determined to be an open exploratory task.

[0112] Therefore, based on information from multiple candidate agents, a first target agent that matches the requirements can be determined from among the multiple candidate agents.

[0113] For example, the task to be processed could be a discussion task on the development prospects of a certain industry. The first target agent can output opinion information on the development prospects of the industry by searching the development analysis report of the industry. This opinion information can be used as intermediate processing result information.

[0114] Then, based on the intermediate processing results and demand information, a second target agent can be determined from multiple candidate agents. For example, the intermediate processing results might be from an economic perspective, while the second target agent's output could be from an environmental protection perspective.

[0115] In topic discussion scenarios, for open-ended exploration tasks, participants can include not only intelligent agents but also users. Therefore, when a user inputs new viewpoints after the intermediate processing results are output, a second target intelligent agent can be determined based on the intermediate processing results, the user's new viewpoints, and the user's needs. This forms an intelligent agent that makes real-time dynamic decisions based on dynamically updated information in the interactive scenario. It can support complex scenarios involving multiple users and multiple intelligent agents in open, multi-round evolution, and can simultaneously meet the requirements of interaction stability and flexibility in complex scenarios.

[0116] Candidate agents matching the processing pattern can be determined based on demand information combined with the types of tasks that the candidate agents can handle. When there are multiple candidate agents matching the processing pattern, for example, agents capable of performing data query tasks include agent A1 and agent A2.

[0117] At this point, using a large model, based on the demand information of the task to be processed and the information of multiple candidate agents, the target agent for processing the task to be processed is determined from multiple candidate agents. This may also include the following operations: using the large model, generating the expected waiting time required for each of the multiple candidate agents to process the task to be processed based on the processing status information of the tasks to be processed by each of the multiple candidate agents; and determining the target agent from multiple candidate agents that match the processing mode based on the multiple expected waiting times.

[0118] Information on the processing status of the tasks to be processed by each of the multiple candidate agents includes, but is not limited to, the number of tasks to be processed by each candidate agent and the stage of task processing.

[0119] For example, agent A1 needs to process 5 tasks, currently 3 tasks are in the task completion stage, 1 task is in the task execution stage, and 1 task is in the task waiting stage. Agent A2 needs to process 2 tasks, with 1 task in the task completion stage and the other task in the task execution stage.

[0120] Multiple candidate agents can be generated based on multiple processing state information, each with an expected waiting time required to process its assigned task. This expected waiting time can include the sum of the remaining execution time of tasks in the execution phase and the expected processing time of tasks in the waiting phase.

[0121] For example, the expected waiting time for agent A1 to process the task to be processed can be 2 minutes, and the expected waiting time for agent A2 to process the task to be processed can be 20 seconds.

[0122] Then, based on multiple expected waiting times, agent A2 can be selected as the target agent from agents A1 and A2.

[0123] The above operations can be performed when matching a target agent for each subtask to be processed, and will not be elaborated here.

[0124] By analyzing the expected waiting time required for multiple candidate agents to process the tasks to be processed, the target agent dynamically matched based on the demand information of the tasks to be processed can meet the real-time requirements in this dynamic interaction scenario, and further improve the processing efficiency of the tasks to be processed.

[0125] In complex scenarios involving multiple users and agents interacting together, if a certain agent is still processing a historical task and receives new user information, the relevant examples may repeatedly initiate task processes around the old and new information, resulting in task conflicts, state confusion, and redundant responses.

[0126] In view of this, the embodiments of this disclosure may further include the following operations: using a large model to perform correlation analysis on the historical demand information and demand information of historical tasks to obtain demand correlation information; wherein, the historical task is a task that has been assigned to a candidate agent and is in an incomplete state; and determining the target task to be processed for calling the target agent based on the demand correlation information.

[0127] Historical tasks are those assigned to candidate agents and currently in an incomplete state. Requirement association information characterizes the degree of association between historical requirements and the requirement information of tasks to be processed. This degree of association describes the semantic similarity and semantic relevance between the historical requirements and the requirement information of tasks to be processed. Semantic similarity measures the degree of overlap between the requirements of historical requirements and the requirements of tasks to be processed. Semantic relevance measures the degree of correlation between the requirements of historical requirements and the requirements of tasks to be processed.

[0128] For example, the historical requirement could be "Please write an article about autumn in a lyrical style," while the task to be processed could be "Please write a prose piece about autumn." When using a large model to perform correlation analysis on the historical requirement information and the requirement information of the historical task, since lyrical style is the most common writing style for prose, it can be determined that the semantics between the historical requirement information and the requirement of the task to be processed are similar, thus identifying the historical task and the task to be processed as duplicate tasks.

[0129] Based on the demand association information, the target task to be processed by the target agent is determined. When it is determined that the historical task and the task to be processed are duplicate tasks, it is not necessary to create a new processing thread for the task to be processed. The target task is determined to be a historical task, thereby reducing the waste of resources caused by allocating computing resources to duplicate tasks and further improving the utilization rate of computing resources.

[0130] The following is combined Figures 6A-6D The task processing chain, determined based on the requirement correlation information between the tasks to be processed and historical tasks, is described in detail.

[0131] Determining the target task to be processed by the target agent based on the requirement association information may include the following operation: in response to the determination of the association information indicating that the semantic similarity between the historical requirement information and the requirement information is greater than or equal to a second predetermined threshold, determining the historical task as the target task.

[0132] The second predetermined threshold can represent a predetermined threshold for semantic similarity and can be configured according to actual needs.

[0133] Figure 6A The illustration schematically depicts a task processing chain determined according to an embodiment of the present disclosure when the task to be processed is related to a historical task and the task requirements have not been updated.

[0134] like Figure 6A As shown, when the semantic similarity between historical demand information and demand information is greater than or equal to the second predetermined threshold, it indicates that the demand of the task to be processed is related to the historical task and the task demand has not been updated. For example, if the current information is "Is it not finished yet?", a comprehensive analysis of historical interaction information can determine that the demand of the task to be processed is still "Help me draw a picture". Thus, it can be determined that the task to be processed is a duplicate of the historical task.

[0135] Therefore, the historical task can be identified as the target task, and the processing link of the target task is the processing link L1 of the historical task. That is, agent A1 → agent A2 → agent A3 → agent A4.

[0136] In some embodiments, the task to be processed includes multiple sub-tasks to be processed; the target agent includes multiple target sub-agents, each responsible for processing the multiple sub-tasks to be processed. Figure 6AAs shown, the multiple target sub-agents can be: agent A1, agent A2, agent A3, and agent A4.

[0137] According to embodiments of this disclosure, invoking a target intelligent agent to process a task to be processed and obtaining processing result information may include the following operations: in response to determining that the data dependency between multiple sub-tasks to be processed is greater than a first predetermined threshold, determining the invocation order between multiple target intelligent agents based on the data dependency relationship between the multiple sub-tasks to be processed; and invoking multiple target intelligent agents to process their respective associated sub-tasks to be processed according to the invocation order, thereby obtaining processing result information.

[0138] like Figure 6A As shown, the calling order of multiple target sub-agents is as follows: Agent A1 → Agent A2 → Agent A3 → Agent A4.

[0139] Then, in the order of invocation, multiple target sub-agents are invoked to process their respective associated sub-tasks to be processed, and the processing result information is obtained.

[0140] For a task consisting of multiple sub-tasks with high data dependencies, the calling order of multiple target sub-agents is determined based on the data dependencies. Since the calling order conforms to the data dependencies between the sub-tasks, the stability of the processing chain and the collaboration efficiency between the multiple target sub-agents are further improved when multiple target sub-agents are called in order to process their respective related sub-tasks, thereby further improving the task processing efficiency.

[0141] In complex scenarios involving multiple users and multiple agents interacting together, users will input a lot of fragmented information, which may supplement historical needs.

[0142] Figure 6B The illustration schematically depicts a task processing chain determined according to an embodiment of the present disclosure when the task to be processed is related to a historical task and the task requirements have been updated.

[0143] like Figure 6B As shown, the historical requirement could be "Help me generate a development prospect report for industry X", and the current information could be "The report clearly states XX viewpoints". By combining historical interaction information and conducting comprehensive analysis, it can be determined that the requirement of the task to be processed can be "Help me generate a development prospect report for industry X with XX viewpoints as the core".

[0144] According to embodiments of this disclosure, determining the target task to be processed by the target agent based on the demand association information may include the following operations: in response to determining that the semantic similarity indicated by the demand association information is less than a second predetermined threshold and the semantic association between the historical demand information and the demand information is greater than or equal to a third predetermined threshold, correcting the historical task based on the demand information to obtain the target task.

[0145] Since both the historical requirement and the pending task's requirement are related to the generation of a report on the development prospects of industry X, but the pending task's requirement adds "clarifying XX viewpoints" to the historical requirement, it can be calculated that the semantic similarity between the historical requirement and the pending task's requirement is less than a second predetermined threshold, and the semantic correlation between the historical requirement information and the requirement information is greater than or equal to a third predetermined threshold. This indicates that the pending task is related to the historical task and the task requirement has been updated.

[0146] Therefore, historical tasks can be modified based on the needs of the task to be processed. For example, the historical requirement information can be modified to "a report on the development prospects of industry X with XX perspective as the core", thereby obtaining the target task.

[0147] Since the task requirements have been updated, the processing link L1 of the historical task can no longer meet the updated task requirements. Therefore, it is necessary to perform a link update operation on processing link L1 based on the updated task requirements.

[0148] According to embodiments of this disclosure, invoking a target agent to process a task to be processed and obtaining processing result information may further include the following operations: during the processing of the task to be processed, in response to receiving supplementary information associated with the task to be processed, terminating the processing process of the task to be processed; updating the requirement information based on the supplementary information to obtain the requirement information of the updated task; and using a large model, based on the updated requirement information and the information of multiple candidate agents, determining the target agent for processing the updated task from multiple candidate agents.

[0149] like Figure 6B As shown, the updated task requirement information can be "to generate a development prospect report for industry X based on the perspective of XX". Using the large model, based on the updated requirement information and information from multiple candidate agents, the target agent determination operation can be performed. This determination operation is similar to operation S210 described earlier and will not be repeated here.

[0150] The target agents of the updated task can include: agent A2, agent A3, agent A5 and agent A7, and the updated link L' can be agent A3 → agent A2 → agent A5 → agent A7.

[0151] Since the requirements of the pending tasks supplement historical requirements, to avoid conflicts in task execution or confusion in task processing status identification between historical tasks and pending tasks, the processing process of the pending tasks is terminated upon receiving supplementary information associated with them during the processing of the pending tasks. This allows historical tasks that do not meet user needs to release computing resources. The processing chain is dynamically adjusted based on the updated task requirements, enabling dynamic updates to the multi-agent collaboration chain driven by task requirements. This ensures that the multi-agent collaboration process can continuously respond to the real-time evolution of interactive content in complex interactive scenarios, further improving the timeliness of requirement response.

[0152] Figure 6C The illustration schematically depicts a task processing chain determined according to an embodiment of the present disclosure when the task to be processed is unrelated to historical tasks.

[0153] According to embodiments of this disclosure, in response to determining that the semantic similarity of the demand association information is less than a second predetermined threshold and the semantic association is less than a third predetermined threshold, the task to be processed is determined to be the target task.

[0154] like Figure 6C As shown, when the semantic similarity between historical demand information and demand information is less than a second predetermined threshold, it indicates that the demand for the task to be processed is unrelated to the historical task. For example, the current information is "Help me write an essay about autumn," while the historical demand could be "Help me generate a report on the development prospects of industry X." Therefore, the demand for the task to be processed is unrelated to the historical demand. The task to be processed can be determined as a new task, and the operation of determining the target intelligent agent as described in S210 above can be executed to obtain the processing link L2 for processing the task to be processed.

[0155] like Figure 6C As shown, the processing link L2 can be: agent A4 → agent A5.

[0156] This disclosure not only considers the requirements of the task to be processed as represented by the current information, but also combines the correlation between historical tasks and the task to be processed to determine whether the current information has an impact on the execution goal of historical tasks. This realizes dynamic adjustment of the agent scheduling link based on task requirements, further improving task processing efficiency.

[0157] The following is combined Figure 6D The processing mode for exploration tasks is explained in detail.

[0158] Figure 6D The illustration schematically depicts a task processing chain determined according to embodiments of the present disclosure when there are no data dependencies between multiple subtasks to be processed.

[0159] According to embodiments of this disclosure, invoking a target intelligent agent to process a task to be processed and obtain processing result information may include the following operations: in response to determining that the data dependency between multiple sub-tasks to be processed is less than or equal to a first predetermined threshold, invoking a first target intelligent agent to process a first sub-task among the multiple sub-tasks to be processed and obtaining intermediate processing result information; based on the intermediate processing result information and the requirement information, invoking a second target intelligent agent to process a second sub-task among the multiple sub-tasks to be processed and obtaining processing result information.

[0160] The task type of the first subtask to be processed matches the first target agent. For example, if the first subtask is a discussion on a topic in the medical field, the first target agent can be an agent pre-trained based on knowledge in the medical field. The task type of the second subtask to be processed matches the second target agent. For example, if the second subtask is a discussion on a topic in the art field, the second target agent can be an agent pre-trained based on knowledge in the art field.

[0161] like Figure 6D As shown, based on the historical interaction information "recent news YY", a historical task T1 is generated, which calls the intelligent agent A3 to output information and displays it to the user on the terminal device.

[0162] The current information could be "Let's discuss XX." This is combined with historical interaction information for correlation analysis. "XX" could be content mentioned in news articles. The task to be processed could be "Discussion around the topic of XX." This task can include multiple subtasks; for example, each round of interaction can be considered a subtask.

[0163] During the interaction, the first subtask T1, for example, could be participating in the discussion from a technical perspective. This could involve calling upon an agent A1, which is pre-trained based on technical expertise, to execute the first subtask T1 and output the result "From a technical perspective...".

[0164] At this point, the user can intervene and provide feedback on the opinions expressed by agent A1, such as: "Regarding your opinion, I think...".

[0165] In some embodiments, invoking a target agent to process a task to be processed and obtain processing result information may further include the following operations: obtaining feedback information on intermediate processing result information; and, based on the intermediate processing result information, requirement information, and feedback information, invoking a second target agent to process a second subtask among multiple subtasks to be processed and obtain processing result information.

[0166] like Figure 6DAs shown, the second subtask T2 can be generated by combining user feedback information and the content of the previous round of interaction. For example, if the user feedback information involves the field of biological genetics, then the second subtask T2 can be to participate in the discussion from the perspective of biological genetics. The agent A4, which is pre-trained based on professional knowledge in the field of biological genetics, can be called to execute the second subtask T2 and output the result "From the perspective of biological genetics..., Mr. Wang, what do you think?"

[0167] In multi-user, multi-agent interaction scenarios, the timing of user input is random. Therefore, dynamically calling agents to execute subsequent subtasks by combining feedback information on intermediate processing results further enhances the user's sense of participation in the interaction with the agents. This at least solves the problem of reduced user experience caused by ignoring user feedback during agent interaction, and further improves the user experience.

[0168] By asking "What do you think, Mr. Wang?", it can be recognized that after outputting the result, agent A4 has designated Mr. Wang as the next interaction target, and Mr. Wang can then participate in the discussion.

[0169] When Wang inputs the information "Based on YY domain... what do you think, little assistant?", it can be recognized that Wang has designated the next round of interaction object as an intelligent agent. Since this round of conversation is still about "XX topic discussion task", based on the method described above, the large model will associate and combine the tasks to obtain the third sub-task T3, and call the intelligent agent A6, which is pre-trained based on the professional knowledge of YY domain, to execute the third sub-task T3 and output the result "Regarding your point of view, I think...".

[0170] For open and continuously evolving tasks, the system dynamically decides which agents will participate in the interaction based on the latest information output during each round of interaction. This further improves the dynamic changes in complex and ever-changing interaction content and enables real-time scheduling of agents participating in the interaction. This forms a multi-agent collaboration mechanism that combines interaction stability and flexibility, further meeting the dynamic and flexible interaction needs of open tasks.

[0171] Figure 7 The diagram illustrates the scheduling relationships between modules in an exemplary system suitable for implementing a large-model-based task processing method according to embodiments of the present disclosure.

[0172] like Figure 7 As shown, the exemplary system may include an interaction module, a scheduling module, a management module, and an execution module.

[0173] The interaction module is used to interact with the user. The scheduling module receives current information from the interaction module, parses the current information and historical interaction information, and determines whether to trigger an agent interaction. If not triggered, it returns to the interaction module to maintain the current session state. If triggered, it generates a task to be processed.

[0174] The calling module can submit tasks to the management module. The management module can maintain information such as task identifier, task source information, associated context, current execution status, scheduled agent links, and last update time for each task. The execution status can include pending execution, executing, completed, terminated, and expired. By continuously maintaining the task status, the scheduling module can identify tasks that are still executing, avoid repeatedly creating multiple task execution processes around the same requirement, and achieve smooth task replacement and collaboration when the context of the interaction content changes.

[0175] Simultaneously, the management module can also be used for unified registration, tracking, deduplication, updating, and termination management of pending tasks from the scheduling module. When new information appears in the interaction content and triggers a new intelligent agent interaction request, the management module can perform correlation analysis between the new task and historical tasks within the historical time period. If it is determined that the new task is a duplicate task, the scheduling process will not be recreated, and the filtered new task result will be returned to the scheduling module. If it is determined that the new task is a supplement, correction, or requirement update to a historical task, the historical task will be terminated, suspended, or state transitioned to update the task and terminate the old task to process the new task, ensuring that scheduling and response are always based on the latest user needs. If it is determined that the new task is not related to a historical task, the result of creating a new task will be returned to the scheduling module.

[0176] After receiving the judgment result from the management module, the scheduling module issues the currently valid task to the execution module based on the judgment result. The execution module executes the response or task processing, returns the processing result to the interaction module, and simultaneously returns an updated task status to the management module.

[0177] The task processing method based on a large model provided in this disclosure can significantly improve the naturalness of multi-agent interaction and the collaborative efficiency of multi-agents in handling tasks in complex scenarios involving multiple users and multiple agents. Compared to traditional examples that rely on users manually specifying or using fixed rules to trigger agent responses, this disclosure can automatically select the most suitable agent to speak based on real-time dialogue content, contextual semantics, and task status. This reduces the cost of manual scheduling by users, lowers the barrier to entry, and makes the collaborative process of multiple agents in this interactive scenario more natural, smooth, and intelligent.

[0178] Secondly, in real-world interaction scenarios, users often send multiple related messages consecutively or add new requests before the agent has responded. This disclosure addresses the comparison, deduplication, and update management of newly generated pending tasks with historical tasks. It avoids duplicate creation when tasks are repeated and promptly terminates old tasks and switches to new ones when task objectives change. By analyzing the correlation between historical requests and pending task requests, it reduces the probability of the system misjudging these fragmented information as multiple independent tasks, leading to the repeated creation of multiple agent processes, wasting computing resources, and causing chaotic responses. This effectively reduces redundant scheduling, conflicting scheduling, and redundant responses in multi-user, multi-information scenarios, thereby improving system stability and task scheduling accuracy.

[0179] Furthermore, the embodiments disclosed herein balance the needs of tasks with clearly defined processing links with the requirements of open-ended interactive dialogue processing, thereby enhancing adaptability in complex interactive scenarios. On the one hand, for problems with clearly defined processing links, execution can be stably carried out according to the real-time orchestrated calling order between agents; on the other hand, for open-ended, multi-round evolving interactive information, the system can dynamically explore which agent should intervene and participate in the interaction next based on the latest messages. Thus, the system can support both standardized process processing and more flexible, complex, and dynamic collaborative tasks, improving the feasibility and coverage of multi-agent capabilities in real-world interactive scenarios.

[0180] Furthermore, during the execution of the method provided in this embodiment, the user does not need to know in advance the types of tasks that each agent can perform, nor does the user need to frequently manually specify the calling object, thus automatically completing the selection of agents and the advancement of collaboration. For the user, the entire multi-agent collaboration process is more coherent, the responses are more in line with the current context, and the way AI participates in the interaction scenario is closer to the experience of real multi-person collaborative communication, thereby enhancing the user's perception of the product's intelligence level and the immersive experience of dialogue participation.

[0181] Finally, the embodiments disclosed herein also enhance the overall engineering controllability and scalability of the system. By structurally decomposing scheduling decisions, task management, and agent execution processes, it is easier to expand to include new agent roles, new task types, and new scheduling strategies. When the system subsequently introduces more specialized agents or more complex collaborative processes, they can still be smoothly integrated within the existing scheduling framework, reducing system reconstruction costs and enhancing the ability of the multi-agent interaction system to continuously evolve and scale up.

[0182] Figure 8 A block diagram of a large-model-based task processing apparatus according to an embodiment of the present disclosure is shown schematically.

[0183] like Figure 8 As shown, the task processing device 800 based on a large model may include a determination module 810 and a processing module 820.

[0184] The determination module 810 is used to determine the target agent for processing the task from multiple candidate agents by using a large model, based on the requirement information of the task to be processed and the information of multiple candidate agents; wherein, the information indicates the type of task that each of the multiple candidate agents can process and the processing status of the task that each of the multiple candidate agents needs to process.

[0185] The processing module 820 is used to call the target intelligent agent to process the task to be processed and obtain the processing result information.

[0186] According to embodiments of this disclosure, the determination module 810 may include an analysis submodule and a first determination submodule.

[0187] The analysis submodule is used to analyze the demand information of the task to be processed using the large model to obtain the processing mode of the task to be processed; wherein, the processing mode indicates whether the task to be processed includes multiple sub-tasks that have data dependencies on each other.

[0188] The first determination submodule is used to determine the target agent that matches the processing mode from among the multiple candidate agents based on information from multiple candidate agents.

[0189] According to embodiments of this disclosure, the task to be processed includes multiple sub-tasks to be processed; the first determining sub-module may include a first determining unit and a second determining unit.

[0190] The first determining unit is configured to determine a target intelligent agent from multiple candidate intelligent agents based on information from multiple candidate intelligent agents in response to determining that the data dependency between multiple subtasks to be processed is greater than a first predetermined threshold; wherein the target intelligent agent includes multiple agents.

[0191] The second determining unit is used to determine the processing logic relationship between multiple target intelligent agents based on the data dependency relationship between multiple sub-tasks to be processed; wherein the processing logic relationship indicates the scheduling order of multiple target intelligent agents.

[0192] According to embodiments of this disclosure, the task to be processed includes multiple sub-tasks to be processed; the target intelligent agent includes a first target intelligent agent and a second target intelligent agent; the determination submodule may include a third determination unit, an acquisition unit and a fourth determination unit.

[0193] The third determining unit is used to determine the first target agent from multiple candidate agents based on information from multiple candidate agents in response to determining that the data dependency between multiple subtasks to be processed is less than or equal to a first predetermined threshold.

[0194] The obtaining unit is used to call the first target agent to process the first sub-task among multiple sub-tasks to be processed, and obtain intermediate processing result information; wherein, the task type of the first sub-task to be processed is matched with the first target agent.

[0195] The fourth determining unit is used to determine the second target agent from multiple candidate agents based on intermediate processing result information and requirement information; wherein the second target agent matches the task type of the second sub-task to be processed in multiple sub-tasks.

[0196] According to embodiments of this disclosure, the candidate agents that match the processing mode include multiple ones; the determining module may further include a generating submodule and a second determining submodule.

[0197] The generation submodule is used to generate the expected waiting time required for each candidate agent to process its task based on the processing status information of the tasks to be processed by the large model.

[0198] The second determination submodule is used to determine the target agent from multiple candidate agents that match the processing mode based on multiple expected waiting times.

[0199] According to embodiments of this disclosure, the above-described large-model-based task processing apparatus may further include: an analysis module and a task determination module.

[0200] The analysis module is used to perform correlation analysis on historical demand information and demand information of historical tasks using a large model to obtain demand correlation information; among them, historical tasks are tasks that have been assigned to candidate agents and are in an incomplete state.

[0201] The task determination module is used to determine the target task that needs to be processed by the target intelligent agent based on the required association information.

[0202] According to embodiments of this disclosure, the task determination module may include a first task determination submodule, a second task determination submodule, and a third task determination submodule.

[0203] The first task determination submodule is used to determine the historical task as the target task in response to the determination of the semantic similarity between the historical demand information and the demand information being greater than or equal to a second predetermined threshold.

[0204] The second task determination submodule is used to correct historical tasks based on the demand information in response to the determination of demand association information indicating that the semantic similarity is less than a second predetermined threshold and the semantic association between historical demand information and demand information is greater than or equal to a third predetermined threshold, so as to obtain the target task.

[0205] The third task determination submodule is used to determine the task to be processed as the target task in response to the determination of the requirement association information indicating that the semantic similarity is less than the second predetermined threshold and the semantic association is less than the third predetermined threshold.

[0206] According to embodiments of this disclosure, the task to be processed includes multiple sub-tasks to be processed; the target intelligent agent includes multiple target sub-intelligent agents for processing the multiple sub-tasks to be processed; the processing module may include a third determining sub-module and a first invoking sub-module.

[0207] The third determining submodule is used to determine the calling order of multiple target sub-agents based on the data dependency relationship between the multiple sub-tasks to be processed, in response to determining that the data dependency relationship between the multiple sub-tasks to be processed is greater than a first predetermined threshold.

[0208] The first calling submodule is used to call multiple target sub-agents in the order of invocation to process their respective associated sub-tasks and obtain processing result information.

[0209] According to embodiments of this disclosure, the processing module may further include: a termination submodule, an update submodule, and a fourth determination submodule.

[0210] The termination submodule is used to terminate the processing of a pending task in response to receiving supplementary information associated with the pending task during the processing of the pending task.

[0211] The update submodule is used to update the requirement information based on the supplementary information, so as to obtain the updated requirement information of the task.

[0212] The fourth determination submodule is used to utilize the large model to determine the target agent for handling the updated task from multiple candidate agents based on the updated requirement information and the information of multiple candidate agents.

[0213] According to embodiments of this disclosure, the task to be processed includes multiple sub-tasks to be processed; the target intelligent agent includes a first target intelligent agent and a second target intelligent agent; the processing module may include a second invocation sub-module and a third invocation sub-module.

[0214] The second invocation submodule is used to respond to determining that the data dependency between multiple subtasks to be processed is less than or equal to a first predetermined threshold, and to invoke the first target agent to process the first subtask among the multiple subtasks to be processed, and to obtain intermediate processing result information; wherein, the task type of the first subtask to be processed is matched with the first target agent.

[0215] The third invocation submodule is used to invoke the second target agent to process the second subtask among multiple subtasks based on intermediate processing result information and requirement information, and obtain processing result information; wherein, the task type of the second subtask is matched with the second target agent.

[0216] According to embodiments of this disclosure, the processing module may further include: an acquisition submodule and a fourth invocation submodule.

[0217] The Get submodule is used to obtain feedback information regarding intermediate processing results.

[0218] The fourth invocation submodule is used to invoke the second target agent to process the second subtask among multiple subtasks based on intermediate processing result information, requirement information, and feedback information, and obtain processing result information.

[0219] According to embodiments of this disclosure, the task processing apparatus 800 based on a large model may further include: an association analysis module and a generation module.

[0220] The association analysis module is used to perform association analysis on current information and historical interaction information using a large model to obtain interaction intent information; among which, the interaction intent information indicates the object to be interacted with in response to the current information.

[0221] The generation module is used to generate requirement information based on current information and historical interaction information in response to the determination that the object to be interacted with is an intelligent agent.

[0222] According to embodiments of this disclosure, the generation module may include an analysis submodule and a generation submodule.

[0223] The analysis submodule is used to respond to the determination that the object to be interacted with is an intelligent agent, and to use a large model to analyze based on current information and historical interaction information to generate task association information between the task to be processed and historical tasks; wherein, the current information is associated with the task to be processed; and the historical tasks are associated with historical interaction information.

[0224] The generation submodule is used to generate requirement information based on task-related information from the large model.

[0225] According to embodiments of this disclosure, the generation submodule may include an extraction unit and a first generation unit.

[0226] The extraction unit is used to extract the processing status information and collaboration requirement information of the historical task from the historical interaction information in response to the determination of task association information indicating that the task to be processed is associated with the historical task.

[0227] The first generation unit is used to generate requirement information by utilizing the large model and based on the processing status information and collaboration requirement information of historical tasks.

[0228] According to embodiments of this disclosure, the generation submodule may further include a second generation unit, used to generate requirement information based on current information using a large model in response to determining that the task association information indicates that the task to be processed is not associated with historical tasks.

[0229] Figure 9 A block diagram of an intelligent agent according to an embodiment of the present disclosure is shown schematically.

[0230] like Figure 9 As shown, in the embodiments of this disclosure, inspired by the von Neumann architecture in modern computer theory, such as... Figure 9 As shown, the AI ​​agent 900 may include three core modules: an input module 910, an output module 920, and a processing module 930. The processing module 930 may include a control unit 931, a storage unit 932, and a computing unit 933.

[0231] The input module 910 is responsible for receiving or sensing information such as queries, requests, instructions, signals, or data from the outside world (e.g., users or the external environment), and converting it into a format that the AI ​​agent 900 can understand and process. The input module 910 is the primary link for the AI ​​agent 900 to interact with the outside world. It enables the AI ​​agent 900 to efficiently and accurately obtain the necessary "sensory" information from the outside world and respond to this information.

[0232] In the example, the input information received by the input module 910 can be the requirement information of the task to be processed as described above and the information of multiple candidate agents.

[0233] In the example, the processing module 930 is the core support for the AI ​​agent 900's ability to handle complex tasks. The processing module 930 can determine the target task based on the input information received by the input module 910, determine the large model based on the target task, execute the task processing method based on the large model described above by calling the large model, and output the processing result information.

[0234] In the example, the control unit 931 in the processing module 930 will continuously interact with the storage unit 932, the arithmetic unit 933, and / or the output module 920 during operation. However, it should be noted that in the embodiments of this disclosure, the control unit 931 initiates communication with the storage unit 932, the arithmetic unit 933, and / or the output module 920 as a single initiator, and there is no communication coupling between the storage unit 932, the arithmetic unit 933, and the output module 920.

[0235] In the example, the performance of the control unit 931 is closely related to the large model on which the AI ​​agent 900 is based. To fully leverage the capabilities of the large language model, the internal structure of the control unit 931 can be designed to be highly configurable and scalable to handle various types of tasks and requirements in real-world scenarios.

[0236] Storage unit 932 can be responsible for remembering information such as historical dialogues and event streams. Configuration information, target text, and data resources generated in each round can be included in storage unit 932.

[0237] In the example, after receiving a configuration generation request, the AI ​​agent 900 can determine the configuration intent from the initial text using an intent recognition model. The configuration intent can be stored in storage unit 932. The AI ​​agent 900 can retrieve relevant data resources from storage unit 932 and feed them back to control unit 931. Then, control unit 931 can use the returned data resources to obtain configuration data corresponding to the initial text. It can also retrieve relevant text data from storage unit 932 and feed it back to control unit 931. Then, control unit 931 can use the returned text data to obtain the target text and pass the target text and configuration data to output module 920.

[0238] The arithmetic unit 933 can be viewed as a predefined tool library. Renderers and display controls, as mentioned earlier, can be included in the arithmetic unit 933.

[0239] In the example, when the AI ​​agent 900 needs to render multiple output data, it can call the relevant renderer and display controls from the computing unit 933 and feed them back to the control unit 932. Then, the control unit 932 can use the fed-back renderer and display controls to render the first search result and pass it to the output module 920. It's understandable that although large language models have excellent language understanding and generation capabilities, like humans, the tasks they can solve without any tools are very limited. When the AI ​​agent 900 is given the ability to call tools, it can perform tasks such as using a calculator to complete mathematical calculations, using Python to perform data analysis, and using a search engine to complete related tasks.

[0240] In the example, output module 920 can output the processing result information described above.

[0241] The AI ​​agent 900 according to embodiments of this disclosure can simply and effectively improve the level of intelligence, as well as enhance flexibility and versatility.

[0242] According to embodiments of this disclosure, this disclosure also provides an intelligent agent, an electronic device, a readable storage medium, and a computer program product.

[0243] According to an embodiment of the present disclosure, an electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the methods described above.

[0244] According to embodiments of the present disclosure, a non-transitory computer-readable storage medium stores computer instructions, wherein the computer instructions are used to cause a computer to perform the methods described above.

[0245] According to an embodiment of this disclosure, a computer program product includes a computer program that, when executed by a processor, implements the method described above.

[0246] Figure 10 The diagram schematically illustrates an electronic device suitable for implementing a large-model-based task processing method according to embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0247] like Figure 10 As shown, device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in read-only memory (ROM) 1002 or a computer program loaded into random access memory (RAM) 1003 from storage unit 1008. The RAM 1003 may also store various programs and data required for the operation of device 1000. The computing unit 1001, ROM 1002, and RAM 1003 are interconnected via bus 1004. Input / output (I / O) interface 1005 is also connected to bus 1004.

[0248] Multiple components in device 1000 are connected to I / O interface 1005, including: input unit 1006, such as keyboard, mouse, etc.; output unit 1007, such as various types of monitors, speakers, etc.; storage unit 1008, such as disk, optical disk, etc.; and communication unit 1009, such as network card, modem, wireless transceiver, etc. Communication unit 1009 allows device 1000 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0249] The computing unit 1001 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1001 performs the various methods and processes described above, such as large model-based task processing methods. For example, in some embodiments, the large model-based task processing method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1008. In some embodiments, part or all of the computer program may be loaded and / or installed on device 1000 via ROM 1002 and / or communication unit 1009. When the computer program is loaded into RAM 1003 and executed by the computing unit 1001, one or more steps of the large model-based task processing method described above may be performed. Alternatively, in other embodiments, computing unit 1001 may be configured to perform a large-model-based task processing method by any other suitable means (e.g., by means of firmware).

[0250] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0251] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0252] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0253] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0254] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0255] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, distributed system servers, or servers incorporating blockchain technology.

[0256] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0257] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A task processing method based on a large model, comprising: Using a large model, based on the requirements of the task to be processed and information from multiple candidate agents, a target agent is determined from the multiple candidate agents to process the task to be processed; wherein, the information indicates the task types that each of the multiple candidate agents can process and the processing status of the task that each of the multiple candidate agents needs to process; and The target intelligent agent is invoked to process the task to be processed, and the processing result information is obtained.

2. The method of claim 1, wherein, The process of utilizing a large model to determine the target agent for processing the task, based on the requirements of the task to be processed and information from multiple candidate agents, includes: By analyzing the demand information of the task to be processed using a large model, the processing mode of the task to be processed is obtained; wherein, the processing mode indicates whether the task to be processed includes multiple sub-tasks that have data dependencies on each other; and Based on the information of the multiple candidate agents, the target agent that matches the processing mode is determined from the multiple candidate agents.

3. The method of claim 2, wherein, The task to be processed includes multiple sub-tasks to be processed; The step of determining the target agent that matches the processing mode from the multiple candidate agents based on their information includes: In response to determining that the data dependency between the plurality of subtasks to be processed is greater than a first predetermined threshold, the target agent is determined from the plurality of candidate agents based on the information of the plurality of candidate agents; wherein, the target agent includes a plurality of agents; Based on the data dependencies between the multiple subtasks to be processed, the processing logic relationships between the multiple target intelligent agents are determined; wherein, the processing logic relationships indicate the scheduling order of the multiple target intelligent agents.

4. The method of claim 2, wherein, The task to be processed includes multiple sub-tasks to be processed; the target intelligent agent includes a first target intelligent agent and a second target intelligent agent; The step of determining the target agent that matches the processing mode from the multiple candidate agents based on their information includes: In response to determining that the data dependency between the plurality of subtasks to be processed is less than or equal to the first predetermined threshold, a first target agent is determined from the plurality of candidate agents based on the information of the plurality of candidate agents; The first target agent is invoked to process the first subtask among the plurality of subtasks to be processed, and intermediate processing result information is obtained; wherein, the task type of the first subtask to be processed matches that of the first target agent; and Based on the intermediate processing result information and the requirement information, the second target agent is determined from the plurality of candidate agents; wherein the second target agent matches the task type of the second sub-task to be processed in the plurality of sub-tasks.

5. The method of any one of claims 2-4, wherein, The candidate agents that match the processing mode include multiple ones; The step of utilizing a large model to determine the target agent for processing the task based on the requirements of the task to be processed and the information of multiple candidate agents further includes: Using a large model, based on the processing status information of the tasks to be processed by each of the multiple candidate agents, the expected waiting time required for each of the multiple candidate agents to process the task to be processed is generated; and The target agent is determined from a plurality of candidate agents that match the processing mode, based on a plurality of expected waiting times.

6. The method of any one of claims 1-5, wherein, The method further includes: The large model is used to perform correlation analysis on historical task demand information and the demand information itself to obtain demand correlation information; wherein, the historical tasks are tasks that have been assigned to candidate agents and are in an incomplete state; and Based on the aforementioned requirement association information, determine the target task that needs to be processed to invoke the target intelligent agent.

7. The method of claim 6, wherein, The step of determining the target task to be processed by the target agent based on the requirement association information includes: In response to determining that the demand association information indicates that the semantic similarity between the historical demand information and the demand information is greater than or equal to a second predetermined threshold, the historical task is determined to be the target task; In response to determining that the demand association information indicates that the semantic similarity is less than a second predetermined threshold and that the semantic association between the historical demand information and the demand information is greater than or equal to a third predetermined threshold, the historical task is corrected based on the demand information to obtain the target task; and In response to determining that the requirement association information indicates that the semantic similarity is less than the second predetermined threshold and the semantic association is less than the third predetermined threshold, the task to be processed is determined to be the target task.

8. The method of any one of claims 1-7, wherein, The task to be processed includes multiple sub-tasks to be processed; the target intelligent agent includes multiple target sub-intelligent agents, each used to process the multiple sub-tasks to be processed. The process of invoking the target intelligent agent to process the task to be processed and obtaining processing result information includes: In response to determining that the data dependency between the plurality of subtasks to be processed is greater than the first predetermined threshold, the calling order of the plurality of target sub-agents is determined according to the data dependency relationship between the plurality of subtasks to be processed; According to the calling order, multiple target sub-agents are called to process their respective associated sub-tasks to be processed, and the processing result information is obtained.

9. The method of claim 8, wherein, The step of invoking the target intelligent agent to process the task to be processed and obtaining processing result information further includes: During the processing of the pending task, in response to receiving supplementary information associated with the pending task, the processing of the pending task is terminated. The requirement information is updated based on the supplementary information to obtain the updated task requirement information; and Using a large model, based on the updated requirement information and the information of the multiple candidate agents, a target agent for processing the updated task is determined from the multiple candidate agents.

10. The method according to any one of claims 1-7, wherein, The task to be processed includes multiple sub-tasks to be processed; The target intelligent agent includes a first target intelligent agent and a second target intelligent agent; The process of invoking the target intelligent agent to process the task to be processed and obtaining processing result information includes: In response to determining that the data dependency between the plurality of subtasks to be processed is less than or equal to the first predetermined threshold, the first target agent is invoked to process the first subtask among the plurality of subtasks to be processed, and intermediate processing result information is obtained; wherein, the task type of the first subtask to be processed matches the first target agent; and Based on the intermediate processing result information and the requirement information, the second target agent is invoked to process the second sub-task among the multiple sub-tasks to be processed, and the processing result information is obtained; wherein, the task type of the second sub-task to be processed matches the second target agent.

11. The method according to claim 10, wherein, The step of invoking the target intelligent agent to process the task to be processed and obtaining processing result information further includes: Obtain feedback information regarding the intermediate processing results; and Based on the intermediate processing result information, the demand information, and the feedback information, the second target agent is invoked to process the second sub-task among the multiple sub-tasks to be processed, and the processing result information is obtained.

12. The method according to any one of claims 1-10, further comprising: The large model is used to perform correlation analysis on current information and historical interaction information to obtain interaction intent information; wherein, the interaction intent information indicates the object to be interacted with in response to the current information; and In response to determining that the object to be interacted with is an intelligent agent, the demand information is generated using the large model based on the current information and historical interaction information.

13. The method according to claim 12, wherein, The step of responding to determining that the object to be interacted with is an intelligent agent, and generating the demand information based on the current information and historical interaction information using the large model, includes: In response to determining that the object to be interacted with is an agent, the large model is used to analyze the current information and historical interaction information to generate task association information between the task to be processed and the historical tasks; wherein the current information is associated with the task to be processed; the historical tasks are associated with the historical interaction information; and The requirement information is generated using the large model based on the task association information.

14. The method according to claim 13, wherein, The process of generating the requirement information using the large model based on the task association information and the historical interaction information includes: In response to determining that the task association information indicates an association between the task to be processed and the historical task, the processing status information and collaboration requirement information of the historical task are extracted from the historical interaction information; and Using the large model, the requirement information is generated based on the processing status information of the historical tasks and the collaboration requirement information of the historical tasks.

15. The method according to claim 13, wherein, The process of generating the requirement information using the large model based on the task association information and the historical interaction information includes: In response to determining that the task association information indicates that the task to be processed is not associated with the historical task, the requirement information is generated based on the current information using the large model.

16. A task processing device based on a large model, comprising: A determination module is used to utilize a large model to determine, based on the requirements of the task to be processed and information from multiple candidate agents, a target agent for processing the task to be processed; wherein the information indicates the task types that each of the multiple candidate agents can process and the processing status of the task to be processed by each of the multiple candidate agents; and The processing module is used to call the target intelligent agent to process the task to be processed and obtain the processing result information.

17. An intelligent agent, comprising: The input module is used to receive the requirement information of the task to be processed and the information of multiple candidate agents; wherein, the information indicates the task type that each of the multiple candidate agents can process and the processing status of the task that each of the multiple candidate agents needs to process. The processing module is configured to determine a target task based on the influence information received by the input module and the initial code, determine a target large model based on the target task, and execute the method described in any one of claims 1-15 by calling the target large model to obtain the processing result information; The output module is used to output the processing result information obtained by the processing module.

18. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-15.

19. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-15.

20. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-15.

Images