Complex task full-automatic processing method based on multi-agent cooperation and related device

Abstract

The invention provides a complex task full-automatic processing method based on multi-agent cooperation and a related device, and relates to the technical field of artificial intelligence such as large language models, generative models, agents and task intelligent scheduling. The method comprises the following steps: interacting with a target user who puts forward an original task demand to obtain a complete task demand; the complete task demand is split into a plurality of sub-tasks at least comprising a complex sub-task, the complex sub-task refers to a sub-task needing at least two sub-agents to process according to a cooperation process, and a single sub-agent is used for processing a simple sub-task; issuing the complex sub-tasks to corresponding target execution intelligent units, and issuing the simple sub-tasks to corresponding target sub-agents; and summarizing sub-task execution results returned by each target execution intelligent unit and each target sub-agent. According to the method, the execution intelligent unit specially used for processing the complex sub-task is introduced, and the complex sub-task is processed more intensively according to the cooperation process through the at least two sub-agents contained in the execution intelligent unit, so that the task disassembling difficulty is reduced; and a better sub-task processing result can be obtained through the execution intelligent unit integrated by the multiple sub-agents.

Description

Technical Field

[0001] The present disclosure relates to the technical field of data processing, specifically to artificial intelligence technical fields such as large language models, generative models, intelligent agents, and intelligent task scheduling. In particular, it relates to a fully automatic processing method, device, electronic device, computer-readable storage medium, and computer program product for complex tasks based on multi-agent collaboration. Background Art

[0002] With the development of artificial intelligence technology, the application of intelligent agents based on large language models has become increasingly widespread. The independent operation of intelligent agents has been difficult to meet the requirements of complex tasks. Therefore, multi-agent collaboration systems have become an important research direction.

[0003] In a multi-agent system, each intelligent agent can cooperate in division of labor, communicate with each other, and jointly complete more complex tasks, improving the efficiency and intelligence of task completion. Summary of the Invention

[0004] Embodiments of the present disclosure propose a fully automatic processing method, device, electronic device, computer-readable storage medium, and computer program product for complex tasks based on multi-agent collaboration.

[0005] In a first aspect, embodiments of the present disclosure propose a fully automatic processing method for complex tasks based on multi-agent collaboration, including: interacting with a target user who proposes an original task requirement to obtain a complete task requirement; splitting the complete task requirement into multiple sub-tasks each including at least one complex sub-task, where a complex sub-task refers to a sub-task that requires at least two sub-intelligent agents to process according to a collaboration process, and a single sub-intelligent agent is used to process a simple sub-task; sending the complex sub-tasks in each sub-task to corresponding target execution intelligent units, and sending the simple sub-tasks in each sub-task to corresponding target sub-intelligent agents, where the target execution intelligent unit includes at least two sub-intelligent agents for processing complex sub-tasks; summarizing the sub-task execution results returned by each target execution intelligent unit and each target sub-intelligent agent, and presenting the summarized task processing result to the target user.

[0006] Second aspect, embodiments of the present disclosure propose a fully automatic complex task processing device based on multi-agent collaboration, including: a complete task requirement acquisition unit configured to obtain complete task requirements by interacting with a target user who proposes original task requirements; a complete task splitting unit configured to split the complete task requirements into multiple sub-tasks each including at least one complex sub-task; wherein, a complex sub-task refers to a sub-task that requires at least two sub-agents to process according to a collaboration process, and a single sub-agent is used to process a simple sub-task; a sub-task allocation unit configured to send the complex sub-tasks in each sub-task to corresponding target execution intelligent units and send the simple sub-tasks in each sub-task to corresponding target sub-agents; wherein, the target execution intelligent units include at least two sub-agents for processing complex sub-tasks; a sub-task result summarization and presentation unit configured to summarize the sub-task execution results returned by each target execution intelligent unit and each target sub-agent, and present the summarized task processing results to the target user.

[0007] Third aspect, embodiments of the present disclosure provide an electronic device, which 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, and when the instructions are executed by the at least one processor, the at least one processor is enabled to implement the fully automatic complex task processing method based on multi-agent collaboration described in the first aspect.

[0008] Fourth aspect, embodiments of the present disclosure provide a non-transitory computer-readable storage medium storing computer instructions, and the computer instructions are used to enable a computer to implement the fully automatic complex task processing method based on multi-agent collaboration described in the first aspect when executed.

[0009] Fifth aspect, embodiments of the present disclosure provide a computer program product including a computer program, and when the computer program is executed by a processor, it can implement the steps of the fully automatic complex task processing method based on multi-agent collaboration described in the first aspect.

[0010] The fully automatic processing solution for complex tasks based on multi-agent collaboration provided by the present disclosure first interacts with the target user who initiates the original task requirement, thereby determining the complete task requirement through the information obtained during the interaction process, and then splits the complete task requirement into multiple sub-tasks including at least one complex sub-task, where the complex sub-task is used to distinguish simple sub-tasks that can be completed by only a single sub-agent, which refers to sub-tasks that require at least two sub-agents to process according to a collaborative process. Next, the complex sub-task is assigned to a target execution intelligent unit, and the simple sub-task is assigned to a target sub-agent, where the target execution intelligent unit includes multiple sub-agents arranged according to a collaborative process for processing the complex sub-task. Finally, the sub-task execution results returned by each target execution intelligent unit and each target sub-agent are summarized. That is, this solution first attempts to complete some of the missing information in the original task information by interacting with the target user, so as to obtain more accurate complete task requirements, and then introduces an execution intelligence unit dedicated to processing complex sub-tasks to more centrally process the complex sub-tasks through multiple sub-agents arranged according to collaborative processes contained in the execution intelligence unit. This eliminates the need to split the complete task requirements into the most fine-grained simple sub-tasks and reduces the difficulty of task decomposition, while enabling better sub-task processing results to be obtained through the execution intelligence unit integrated with multiple sub-agents.

[0011] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Other features, objects and advantages of the present disclosure will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0013] Figure 1 is an exemplary system architecture in which the present disclosure may be applied;

[0014] Figure 2 A flowchart of a method for fully automatic processing of complex tasks based on multi-agent collaboration provided by an embodiment of the present disclosure;

[0015] Figure 3 A schematic diagram of a branch process for splitting a complete task requirement through different fixed protocol processes provided in an embodiment of the present disclosure;

[0016] Figure 4 A flowchart of a method for ensuring execution of a subtask by completing a question provided in an embodiment of the present disclosure;

[0017] Figure 5-1 and Figure 5-2Schematic flowchart of the full-automatic processing method for complex tasks based on multi-agent collaboration in an application scenario provided by an embodiment of the present disclosure;

[0018] Figure 6 Block diagram of the structure of a full-automatic processing device for complex tasks based on multi-agent collaboration provided by an embodiment of the present disclosure;

[0019] Figure 7 Schematic diagram of the structure of an electronic device suitable for executing the full-automatic processing method for complex tasks based on multi-agent collaboration provided by an embodiment of the present disclosure. Detailed implementation manners

[0020] The following describes exemplary embodiments of the present disclosure with reference to the accompanying drawings. Various details of the embodiments of the present disclosure are included to facilitate understanding, and they should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted below. It should be noted that, without conflict, the embodiments in the present disclosure and the features in the embodiments can be combined with each other.

[0021] In the technical solution of the present disclosure, the processing of collection, storage, use, processing, transmission, provision, and disclosure of the user's personal information involved all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0022] Figure 1 An exemplary system architecture 100 is shown in which embodiments of the full-automatic processing method, device, electronic device, and computer-readable storage medium for complex tasks based on multi-agent collaboration of the present disclosure can be applied.

[0023] As Figure 1 shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0024] Users can use the terminal devices 101, 102, 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various applications for realizing information communication between the two may be installed on the terminal devices 101, 102, 103 and the server 105, such as task processing applications, multi-agent collaboration applications, instant messaging applications, etc. A variety of agents or intelligent task units may be installed or hosted on the server 105 for undertaking tasks of different granularities.

[0025] The terminal devices 101, 102, 103 and the server 105 can be either hardware or software. When the terminal devices 101, 102, 103 are hardware, they can be various electronic devices with a display screen, including but not limited to smartphones, tablets, laptop computers, desktop computers, and so on; when the terminal devices 101, 102, 103 are software, they can be installed in the above-listed electronic devices, and can be implemented as multiple software or software modules, or can be implemented as a single software or software module, and no specific limitation is made here. When the server 105 is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or can be implemented as a single server; when the server is software, it can be implemented as multiple software or software modules, or can be implemented as a single software or software module, and no specific limitation is made here.

[0026] The server 105 can provide various services through various built-in applications. Taking the task processing application that can provide complex task processing services as an example, when the server 105 runs this task processing application, the following effects can be achieved: First, receive the original task requirements put forward by the user using the terminal devices 101, 102, 103 through the network 104, and interact with the user based on the original task requirements to obtain the complete task requirements; then, split the complete task requirements into multiple sub-tasks including at least one complex sub-task. The complex sub-task refers to a sub-task that requires at least two sub-intelligent agents to process according to a collaboration process, and a single sub-intelligent agent is used to process simple sub-tasks; next, send the complex sub-tasks to the corresponding target execution intelligent unit, and send the simple sub-tasks to the corresponding target sub-intelligent agent. The target execution intelligent unit contains at least two sub-intelligent agents for processing complex sub-tasks; finally, summarize the sub-task execution results returned by each target execution intelligent unit and each target sub-intelligent agent, and return the summarized task processing results to the terminal devices 101, 102, 103 through the network 104 for presentation to the user.

[0027] It should be noted that in addition to being able to obtain the original task requirements from the terminal devices 101, 102, 103 through the network 104, the original task requirements can also be pre-stored locally in the server 105 in various ways. Therefore, when the server 105 detects that these data have been stored locally (such as the to-be-processed tasks retained before starting processing), it can choose to directly obtain these data from the local. In this case, the exemplary system architecture 100 may not include the terminal devices 101, 102, 103 and the network 104.

[0028] Since the processing of complex tasks requires a large amount of computing resources and strong computing capabilities, the fully automatic complex task processing method based on multi-agent collaboration provided in the subsequent embodiments of the present disclosure is generally executed by the server 105 with strong computing capabilities and a large amount of computing resources. Correspondingly, the fully automatic complex task processing device based on multi-agent collaboration is generally also set in the server 105. However, it should also be noted that when the terminal devices 101, 102, and 103 also have computing capabilities and computing resources that meet the requirements, the terminal devices 101, 102, and 103 can also complete the above operations that were originally performed by the server 105 through the task processing applications installed on them, and then output the same results as the server 105. Especially in the case where there are multiple terminal devices with different computing capabilities at the same time, when the task processing application determines that the terminal device where it is located has strong computing capabilities and a large amount of remaining computing resources, the terminal device can be allowed to execute the above operations, thereby appropriately reducing the computing pressure on the server 105. Correspondingly, the fully automatic complex task processing device based on multi-agent collaboration can also be set in the terminal devices 101, 102, and 103. In this case, the exemplary system architecture 100 may not include the server 105 and the network 104.

[0029] It should be understood that Figure 1 the numbers of terminal devices, networks, and servers in

[0030] Please refer to Figure 2 , Figure 2 which is a flowchart of a fully automatic complex task processing method based on multi-agent collaboration provided by an embodiment of the present disclosure. The process 200 includes the following steps:

[0031] Step 201: Obtain a complete task requirement by interacting with the target user who proposed the original task requirement;

[0032] The purpose of this step is for the execution entity of the fully automatic complex task processing method based on multi-agent collaboration (such as Figure 1 the server 105 shown or the interaction agent or interaction intelligent unit carried on the server 105) to complete the fuzzy, incomplete, or missing parts that may exist in the original task requirement through interaction with the target user, ensure that the task requirement can accurately reflect the true intention of the user, and provide a reliable basis for subsequent task decomposition and execution. That is, this step wants to achieve the following purposes through interaction: 1) Information completion: Identify and supplement the key information missing in the task requirement; 2) Requirement confirmation: Ensure that the task requirement is consistent with the actual requirement of the user; 3) Task optimization: Optimize the task requirement during the interaction process to make it more in line with the actual execution conditions.

[0033] The interaction behavior mentioned in this step can be specifically carried out in the following manner:

[0034] 1) Ask questions proactively: Based on the analysis of the original task requirements, proactively ask targeted questions to the user; 2) Conduct multi-round conversations: Gradually deepen through multi-round conversations to guide the user to provide more detailed information; 3) Provide example guidance: Provide examples or templates to help the user express their requirements more clearly; 4) Provide feedback and confirmation: Provide real-time feedback to the user on the currently understood task requirements and confirm whether they are accurate.

[0035] An implementation method including but not limited to can be specifically as follows: First, determine the missing items of task information according to the original task requirements proposed by the target user; then, obtain the missing task information through at least one round of questions for supplementing the missing information initiated to the target user; finally, when the missing task information corresponding to all the missing items of task information is obtained, determine the complete task requirements based on all the missing task information and the original task requirements.

[0036] In the intelligent customer service scenario, users usually put forward vague or incomplete requirements, and it is necessary to complete the information through interaction to provide accurate services. For example, when the user enters: "I want to change the ticket", the above-mentioned execution entity can proactively ask: "What is your order number? Which date do you want to change to?" And after the user provides the information, it can further initiate a confirmation question: "You want to change to a flight from Beijing to Shanghai on March 25th, right?" So that the finally completed task requirements are: Change the ticket with the order number 123456 from Beijing to Shanghai on March 25th.

[0037] In the intelligent assistant scenario, users may put forward complex task requirements, and it is necessary to disassemble and complete the information through interaction. For example, when the user enters: "Help me arrange a meeting", the above-mentioned execution entity can proactively ask: "What is the theme of the meeting? Who are the participants? What time do you hope to arrange it?" And after the user provides the information, it can further initiate a confirmation question: "The theme of the meeting is project discussion, the participants include Zhang San, Li Si, and Wang Wu, and the time is 3 pm this Friday, right?" So that the finally completed task requirements are: Arrange a project discussion meeting with the participants Zhang San, Li Si, and Wang Wu at 3 pm this Friday.

[0038] In a decision support system, users may put forward high-level decision-making requirements, and specific parameters and constraints need to be complemented through interaction. For example, when the user enters: "Please help me analyze the market trend", the above-mentioned execution entity can actively ask: "Which market do you want to analyze? What is the time range? Which indicators need to be concerned?" And after the user provides information, it can further initiate a confirmation question: "You want to analyze the Chinese smartphone market, with a time range from 2022 to 2023, and pay attention to sales volume and market share, right?" So that the finally complemented task requirement is: Analyze the sales volume and market share trends of the Chinese smartphone market from 2022 to 2023.

[0039] In an intelligent information service platform, users may put forward vague query requirements, and specific conditions need to be complemented through interaction. For example, when the user enters: "I want to find a nearby restaurant", the above-mentioned execution entity actively asks: "What type of restaurant do you want to find? What is the budget range? What location is needed?" And after the user provides information, it can further initiate a confirmation question: "You want to find a Chinese restaurant, with a budget of less than 100 yuan per person, and the location is in the city center, right?" So that the finally complemented task requirement is: Find a Chinese restaurant within 100 yuan per person in the city center.

[0040] To achieve the above effects, the above-mentioned execution entity should first be able to parse the user's input (i.e., the original task requirement) through natural language understanding (NLU, Natural Language Understanding) technology, identify the key information in the task requirement, and identify the user's core intention, distinguish the "must be satisfied" and "optional" parts in the task requirement, as well as identify the fuzzy or unclear parts in the task requirement, and guide the user to clarify, and detect the logical consistency of the task requirement in real time during the interaction process, and prompt the user to correct further. At the same time, context information should be retained in multi-round conversations to avoid repeated questions or information loss.

[0041] To further improve the interaction effect, the interaction method and content can also be customized according to the user's historical preferences or behavior habits, and combined with the domain knowledge base to provide professional suggestions or options during the interaction process. Even multiple interaction methods (such as voice, text, image, etc.) can be supported to improve the user experience.

[0042] Step 202: Split the complete task requirement into multiple subtasks that include at least one complex subtask;

[0043] Based on step 201, this step aims to enable the above-mentioned execution entity to split the complete task requirements into multiple subtasks (including at least one complex subtask and simple subtasks), and process them respectively by the execution intelligent unit (for processing complex subtasks) and sub-agents (for processing simple subtasks), so as to achieve the efficient execution of the task. Among them, a complex subtask refers to a subtask that requires at least two sub-agents to process according to a collaboration process. That is, by splitting the complete task requirements into multiple subtasks, it is ensured that each subtask can be efficiently processed. At the same time, by introducing the processing mechanism for complex subtasks, the difficulty of task decomposition is reduced and the processing effect is improved.

[0044] Among them, the task decomposition process can first adopt the following recognition mechanism:

[0045] 1) Complex subtask recognition: Identify the parts in the task requirements that require the collaboration of multiple sub-agents. For example, the task requirements contain multiple steps with dependencies, or require the support of multiple professional capabilities.

[0046] 2) Simple subtask recognition: Identify the parts in the task requirements that can be independently processed by a single sub-agent. For example, a single operation or a clear instruction in the task requirements.

[0047] 3) Task dependency analysis: Analyze the dependencies between subtasks to ensure that the decomposed subtasks can be executed in the correct order.

[0048] 4) Task priority division: According to the characteristics of the task requirements, assign priorities to the subtasks to ensure that critical tasks are processed first.

[0049] After identifying the above-mentioned parts, complex subtasks need to be processed by the execution intelligent unit, which contains multiple sub-agents and executes according to the collaboration process:

[0050] 1) Sub-agent division of labor: Each sub-agent is responsible for a specific step in the complex subtask. For example, one sub-agent is responsible for data collection, and another sub-agent is responsible for data analysis.

[0051] 2) Collaboration process: Multiple sub-agents collaborate with each other according to the processing process obtained by dynamic orchestration that conforms to the processing logic to ensure the efficiency and consistency of task processing. For example, after sub-agent A completes the task, it passes the result to sub-agent B, and sub-agent B continues to process based on the result.

[0052] 3) Result integration: The execution intelligent unit integrates the processing results of multiple sub-agents to form the final output of the complex subtask.

[0053] In the intelligent customer service scenario, users may put forward complex requirements involving multiple steps, which need to be processed through task decomposition and collaboration. Taking the user input: "There is a problem with my order. I need a refund and to place a new order" as an example, it can be decomposed into the following subtasks:

[0054] 1. Complex subtask: Process the refund and place a new order (requiring collaboration among multiple sub-intelligent agents), specifically corresponding to sub-intelligent agent A for verifying order information and processing the refund, and sub-intelligent agent B for placing a new order according to the user's requirements;

[0055] 2. Simple subtask: Send a notification of the processing result to the user (processed by a single sub-intelligent agent).

[0056] That is, the overall processing flow in this scenario is: After sub-intelligent agent A completes the refund, it passes the result to sub-intelligent agent B. Sub-intelligent agent B places a new order and notifies the user.

[0057] In the intelligent assistant scenario, users may put forward task requirements involving multiple steps, which need to be processed through task decomposition and collaboration. Taking the user input: "Help me arrange a business trip itinerary, including flight tickets, hotels, and meeting arrangements" as an example, it can be decomposed into the following subtasks:

[0058] 1. Complex subtask: Arrange the business trip itinerary, specifically corresponding to sub-intelligent agent C for querying and booking flight tickets, sub-intelligent agent D for querying and booking hotels, and sub-intelligent agent E for arranging the meeting schedule.

[0059] 2. Simple subtask: Integrate the itinerary information and send it to the user.

[0060] That is, the overall processing flow in this scenario is: After sub-intelligent agents C, D, and E complete their tasks respectively, they pass the results to sub-intelligent agent E. Sub-intelligent agent E integrates the information and notifies the user.

[0061] In the decision support system, users may put forward complex requirements involving multiple analysis steps, which need to be processed through task decomposition and collaboration. Taking the user input: "Please help me analyze the market trend and give investment advice" as an example, it can be decomposed into the following subtasks:

[0062] 1. Complex subtask: Market trend analysis and investment advice generation, specifically corresponding to sub-intelligent agent F for collecting market data, sub-intelligent agent G for analyzing the market trend, and sub-intelligent agent H for generating investment advice.

[0063] 2. Simple subtask: Present the analysis results to the user.

[0064] That is, the overall processing flow in this scenario is: After sub-intelligent agents F, G, and H complete their tasks respectively, they pass the results to sub-intelligent agent H. Sub-intelligent agent D presents the results to the user.

[0065] In an intelligent information service platform, users may put forward complex requirements involving multiple query conditions, which need to be processed through task decomposition and collaboration. Taking the user input: "I want to find a restaurant suitable for family gatherings, which needs to have children's facilities and offer vegetarian options" as an example, it can be decomposed into the following subtasks:

[0066] 1. Complex subtasks: Restaurant query and screening, specifically corresponding to the sub-intelligent agent I for querying restaurants suitable for family gatherings, the sub-intelligent agent J for screening restaurants with children's facilities, and the sub-intelligent agent K for screening restaurants that offer vegetarian options.

[0067] 2. Simple subtask: Present the screening results to the user.

[0068] That is, the overall processing flow in this scenario is: after the sub-intelligent agents I, J, and K complete their tasks respectively, they pass the results to the sub-intelligent agent K, and the sub-intelligent agent K presents the results to the user.

[0069] Furthermore, to improve the effect of task decomposition and processing, it is also possible to dynamically adjust the decomposition method and quantity of subtasks according to the characteristics of task requirements, and intelligently allocate subtasks according to the capabilities and loads of sub-intelligent agents. It is also possible to optimize the collaboration process of complex subtasks by analyzing historical task execution data.

[0070] Step 203: Send each subtask to the corresponding target execution intelligent unit or target sub-intelligent agent respectively;

[0071] Based on step 202, the purpose of this step is for the above-mentioned execution entity to ensure that complex subtasks can be efficiently processed by multiple sub-intelligent agents according to the collaboration process and simple subtasks can be quickly completed by a single sub-intelligent agent by allocating subtasks to the target execution intelligent unit or target sub-intelligent agent. Among them, the target execution intelligent unit contains multiple sub-intelligent agents arranged according to the collaboration process for processing the decomposed complex subtasks. The target execution intelligent unit is an execution intelligent unit for processing complex subtasks split from the complete task requirements. The execution intelligent unit is an intelligent task unit for executing tasks. The intelligent task unit is a general-purpose task completion unit. The intelligent task unit is designed to contain at least two sub-intelligent agents and collaboration process information for characterizing the collaboration process that should be followed between different sub-intelligent agents. The collaboration process information is determined based on the processing logic specifically adapted to the executed complex subtasks. To facilitate the collaboration between different sub-intelligent agents, the intelligent task unit can also be designed to contain: an information storage module and a communication module for realizing information storage and transmission between different sub-intelligent agents. Of course, if there is no specially designed information storage module and communication module, other mechanisms that can achieve similar information interaction and information storage can also be selected, and no specific limitation is made here.

[0072] It can be seen from this that the core objectives of the solution provided in this step are as follows: 1) Task allocation: According to the characteristics of subtasks, complex subtasks are allocated to the execution intelligent unit, and simple subtasks are allocated to sub-agents; 2) Collaborative processing: Through the collaborative process within the execution intelligent unit, ensure that multiple sub-agents can efficiently collaborate to process complex subtasks, and 3) Information transmission: Through the information storage module and communication module, realize information sharing and transmission between sub-agents.

[0073] Among them, the execution intelligent unit is an intelligent task unit dedicated to processing complex subtasks, and its design includes the following core components:

[0074] 1) Multiple sub-agents: Each sub-agent is responsible for a specific step in the complex subtask. For example, sub-agent A is responsible for data collection, and sub-agent B is responsible for data analysis.

[0075] 2) Collaborative process information: Characterize the collaborative process between sub-agents to ensure the orderliness and efficiency of task processing. For example, after sub-agent A completes the task, it passes the result to sub-agent B, and sub-agent B continues to process based on the result.

[0076] 3) Information storage module: Used to store the intermediate results and task status between sub-agents. For example, store the data collected by sub-agent A for use by sub-agent B.

[0077] 4) Communication module: Used to achieve information transmission and synchronization between sub-agents. For example, after sub-agent A completes the task, it notifies sub-agent B to start processing through the communication module.

[0078] As for the allocation of each subtask, it can be flexibly processed according to the characteristics of the subtask:

[0079] 1) Complex subtask allocation: Allocate complex subtasks to the execution intelligent unit, and let the multiple sub-agents inside it process according to the collaborative process dynamically arranged during the processing. For example, allocate the task of "arranging a business trip itinerary" to the execution intelligent unit, and let sub-agents A, B, and C process flight ticket, hotel, and meeting arrangement respectively.

[0080] 2) Simple subtask allocation: Allocate simple subtasks to a single sub-agent to complete independently. For example, allocate the task of "sending a notice" to sub-agent D to complete independently.

[0081] It is also possible to monitor the execution status of the task in real time to ensure that the task can be completed as expected. For example: Monitor the task progress of sub-agents A, B, and C to ensure that the business trip itinerary arrangement is completed on time.

[0082] Step 204: Aggregate the subtask execution results returned by each target execution intelligent unit and each target sub-agent, and present the aggregated task processing result to the target user.

[0083] Based on Step 203, this step aims to have the above-mentioned execution entity aggregate the execution results of each subtask and present the final processing result to the target user, ensuring that the user can obtain complete, accurate, and easily understandable task feedback. That is, it mainly involves result integration (integrating the execution results of each subtask into a complete task processing result), result optimization (optimizing the aggregated result to better meet the user's needs and expectations), and result presentation (presenting the task processing result in a way that is easy for the user to understand to enhance the user experience).

[0084] Specifically, the process of result aggregation and presentation needs to be flexibly handled according to the characteristics of the task:

[0085] 1) Result collection: Collect the execution results of subtasks from each target execution intelligent unit and target sub-agent. For example, collect the results of flight ticket reservation, hotel reservation, and meeting arrangement from sub-agents A, B, and C respectively.

[0086] 2) Result integration: Integrate the collected subtask results into a complete task processing result. For example, integrate the results of flight ticket, hotel, and meeting arrangement into a complete business trip itinerary.

[0087] 3) Result optimization: Optimize the integrated result, such as deduplication, sorting, formatting, etc. For example, arrange the itinerary information in chronological order and format it into a form that is easy for the user to read.

[0088] 4) Result presentation: Present the task processing result in a way that is easy for the user to understand, such as text, table, chart, etc. For example, present the business trip itinerary in a table form and attach detailed time and location information.

[0089] In the intelligent customer service scenario, users may put forward complex requirements involving multiple steps, and a complete solution needs to be provided through result aggregation and presentation. Still taking the user input: "There is a problem with my order. I need a refund and place a new order" as an example, result aggregation: Sub-agent A: The refund process is completed, and the refund amount has been returned to the original payment account; Sub-agent B: The new order placement is completed, and a new order number has been generated.

[0090] Result presentation: "Your refund has been processed, and the refund amount has been returned to the original payment account. A new order has been successfully generated, and the order number is 123456."

[0091] In the intelligent assistant scenario, users may put forward task requirements involving multiple steps, and a complete solution needs to be provided through result aggregation and presentation. Still taking the user input: "Help me arrange a business trip itinerary, including flight tickets, hotel, and meeting arrangements" as an example, the result aggregation is as follows: Sub-agent A: The flight ticket has been booked, the flight number is CA123, and the time is 3:00 p.m. on March 25th; Sub-agent B: The hotel has been booked, the hotel name is XX Hotel, and the check-in time is March 25th; Sub-agent C: The meeting arrangement has been completed, the time is 10:00 a.m. on March 26th, and the location is XX Meeting Room.

[0092] The result presentation is: "Your business trip itinerary has been arranged as follows: Take flight CA123 to the destination at 3:00 p.m. on March 25th and check in at XX Hotel. Attend a meeting in XX Meeting Room at 10:00 a.m. on March 26th."

[0093] In the decision support system, users may put forward complex requirements involving multiple analysis steps, and a complete analysis result needs to be provided through result aggregation and presentation. Still taking the user input: "Please help me analyze the market trend and give investment suggestions" as an example, the result aggregation is as follows: Sub-agent A: The collected market data includes sales volume and market share from 2022 to 2023; Sub-agent B: The analysis result shows that the market presents a growth trend, with an average annual growth rate of 5%; Sub-agent C: The investment suggestion is "It is recommended to increase investment in the XX industry."

[0094] The result presentation is: "The market analysis result shows that from 2022 to 2023, the average annual growth rate of the sales volume and market share of the XX industry is 5%, and it is recommended to increase investment in this industry."

[0095] In the intelligent information service platform, users may put forward complex requirements involving multiple query conditions, and a complete query result needs to be provided through result aggregation and presentation. Still taking the user input: "I want to find a restaurant suitable for family gatherings, which needs to have children's facilities and provide vegetarian options" as an example, the result aggregation is as follows: Sub-agent A: 10 restaurants suitable for family gatherings have been queried; Sub-agent B: 5 restaurants with children's facilities have been screened out; Sub-agent C: 3 restaurants providing vegetarian options have been screened out.

[0096] The result presentation is: "We recommend the following 3 restaurants: 1. XX Restaurant (has children's facilities and provides vegetarian options); 2. YY Restaurant (has children's facilities and provides vegetarian options); 3. ZZ Restaurant (has children's facilities and provides vegetarian options)."

[0097] The fully automatic complex task processing method based on multi-agent collaboration provided by the embodiments of the present disclosure first interacts with the target user who initiates the original task requirement, and determines the complete task requirement through the information obtained in the interaction process. Then, the complete task requirement is split into multiple sub-tasks including at least one complex sub-task. The complex sub-task is used to distinguish from the simple sub-task that can be completed by a single intelligent agent, and refers to the sub-task that requires at least two sub-intelligent agents to be processed according to the collaboration process. Next, the complex sub-tasks are assigned to the target execution intelligent unit, and the simple sub-tasks are assigned to the target sub-intelligent agent. The target execution intelligent unit contains at least two sub-intelligent agents for processing the complex sub-task. Finally, the sub-task execution results returned by each target execution intelligent unit and each target sub-intelligent agent are summarized. That is, this solution first tries to complement the possibly missing part of the original task information by interacting with the target user, so as to obtain a more accurate complete task requirement. And by introducing an execution intelligent unit dedicated to processing complex sub-tasks, the complex sub-task can be more centrally processed by multiple sub-intelligent agents arranged according to the collaboration process included in the execution intelligent unit. It not only does not need to split the complete task requirement into the simplest granular simple sub-tasks to reduce the task splitting difficulty, but also can obtain better sub-task processing results through the execution intelligent unit integrated with multiple sub-intelligent agents.

[0098] Please refer to Figure 3 , Figure 3 FIG. is a schematic diagram of a branch process for splitting a complete task requirement through different fixed protocol processes provided by the embodiments of the present disclosure, showing the following two solutions:

[0099] Solution 1: First, the complete task requirement is sent to a preset planning intelligent unit. The planning intelligent unit is an intelligent task unit for disassembling and executing planning of the received complete task requirement. Then, control the planning intelligent unit to perform task disassembly in sequence through a task understanding sub-intelligent agent and a task splitting sub-intelligent agent according to a preset first collaboration process, and obtain multiple sub-tasks including at least one complex sub-task. The priority of the complex sub-task disassembled by the task splitting sub-intelligent agent is higher than that of the simple sub-task disassembled.

[0100] In this solution, the planning intelligent unit, as an intelligent task unit dedicated to task disassembly and planning, is designed to include the following core components:

[0101] 1) Task understanding sub-intelligent agent: responsible for parsing the core intention and key information of the complete task requirement. For example, identifying the flight tickets, hotels, and meeting arrangements involved in the task of "arranging a business trip itinerary" proposed by the user.

[0102] 2) Task decomposition sub-agent: Responsible for decomposing the complete task requirements into multiple subtasks and preferentially identifying complex subtasks. For example, decompose the task of "arranging a business trip itinerary" into three subtasks: "booking a flight", "booking a hotel", and "arranging a meeting", and identify "arranging a meeting" as a complex subtask.

[0103] 3) The first collaboration process: Defines the order and rules for task understanding and decomposition to ensure the orderliness and efficiency of task decomposition. For example, first parse the task requirements through the task understanding sub-agent, and then perform task decomposition through the task decomposition sub-agent.

[0104] Among them, the task understanding sub-agent parses the core intention and key information of the complete task requirements. For example, identify the time range, market type, and analysis indicators involved in the task of "analyzing market trends" proposed by the user. The task decomposition sub-agent decomposes the complete task requirements into multiple subtasks and preferentially identifies complex subtasks. For example, decompose the task of "analyzing market trends" into three subtasks: "collecting market data", "analyzing market trends", and "generating investment recommendations", and identify "analyzing market trends" as a complex subtask. The task decomposition sub-agent preferentially decomposes complex subtasks to ensure that complex subtasks can be processed first. For example, when decomposing the task of "arranging a business trip itinerary", preferentially identify "arranging a meeting" as a complex subtask.

[0105] Solution 2: First, send the complete task requirements to a preset planning intelligent unit, which is an intelligent task unit used to decompose and execute the plan for the received complete task requirements. Then, control the planning intelligent unit to sequentially perform task decomposition and inspection confirmation through the task understanding sub-agent, the task decomposition sub-agent, and the split result self-checking sub-agent according to the preset second collaboration process, and obtain multiple subtasks including at least one complex subtask. The priority of the complex subtasks decomposed by the task decomposition sub-agent is higher than that of the simple subtasks decomposed, and the split result self-checking sub-agent is used to confirm the correctness and executability of each decomposed subtask.

[0106] Different from Solution 1, the second collaboration process used in Solution 2 further introduces a split result self-checking sub-agent on the basis of the first collaboration process, which can bring the following advantages:

[0107] 1) Improve the correctness of the decomposition result

[0108] Improvement point: Check each decomposed subtask through the split result self-checking sub-agent to ensure its correctness; Advantage: Avoid subsequent execution failures or deviations caused by incorrect task decomposition and improve the reliability of task processing.

[0109] Example: In the task of "Arranging a Business Trip Itinerary", the self-checking sub-agent checks whether the three sub-tasks of "Booking Air Tickets", "Booking Accommodation", and "Arranging Meetings" are complete and conflict-free.

[0110] 2) Ensure the executability of sub-tasks

[0111] Improvement point: By splitting the result self-checking sub-agent, confirm whether each decomposed sub-task is executable; Advantage: Avoid subsequent processing interruptions caused by non-executable sub-tasks and improve the success rate of task processing.

[0112] Example: In the task of "Analyzing Market Trends", the self-checking sub-agent checks whether the three sub-tasks of "Collecting Market Data", "Analyzing Market Trends", and "Generating Investment Recommendations" meet the executable conditions (such as the availability of data sources).

[0113] 3) Reduce the risk of subsequent processing

[0114] Improvement point: Discover and correct problems in the decomposition results in advance through the self-checking process; Advantage: Reduce the risk during the subsequent task execution and improve the efficiency and effectiveness of overall task processing.

[0115] Example: In the task of "Finding a Restaurant", the self-checking sub-agent checks whether there are logical conflicts or non-executable conditions in the three sub-tasks of "Querying Restaurants Suitable for Family Dinners", "Filtering Restaurants with Children's Facilities", and "Filtering Restaurants Offering Vegetarian Options".

[0116] 4) Enhance the robustness of the system

[0117] Improvement point: Enhance the system's ability to handle abnormal situations through the self-checking process; Advantage: Improve the robustness of the system and ensure stable operation under complex task requirements.

[0118] Example: In the task of "Processing Refunds and Reordering", the self-checking sub-agent checks whether there are dependency relationships or execution conflicts between the two sub-tasks of "Processing Refunds" and "Reordering".

[0119] Furthermore, based on the second collaboration process, the intelligence level of the self-checking sub-agent can also be improved through machine learning technology, enabling it to automatically identify and correct more complex decomposition problems. It can even dynamically adjust the self-checking rules according to the characteristics of task requirements to ensure that the self-checking results are more in line with actual needs. Additionally, it can support joint checking of the decomposition results of multiple related tasks to ensure the correct dependency relationships and execution orders between tasks.

[0120] On the basis that the above embodiments have clearly defined how to decompose tasks, it is also possible to further control the planning intelligent unit to sequentially distribute each subtask to the corresponding target execution intelligent unit or target sub-agent through the relevance matching sub-agent and the distribution sub-agent according to a preset third collaboration process. The relevance matching sub-agent is used to determine the target execution intelligent unit or target sub-agent that matches each subtask based on the relevance between the task and the task processing capabilities of the sub-agent. The relevance matching sub-agent associates complex subtasks with the target execution intelligent unit with matching task processing capabilities.

[0121] That is, as the core step of task allocation, this embodiment aims to distribute each subtask to the matching target execution intelligent unit or target sub-agent through the relevance matching sub-agent and the distribution sub-agent according to a preset third collaboration process, ensuring that complex subtasks can be efficiently processed by the execution unit with corresponding capabilities, while simple subtasks can be quickly completed by a single sub-agent. Among them, the complex subtasks are associated with the target execution intelligent unit with corresponding processing capabilities to ensure that they can be processed collaboratively by multiple sub-agents.

[0122] The above-mentioned third collaboration process is the core process of task allocation, and its design includes the following key components:

[0123] 1) Relevance matching sub-agent: Based on the relevance between the task and the sub-agent, match the most suitable execution unit or sub-agent for each subtask. For example, match the "analyze market trends" subtask to the target execution intelligent unit with data analysis capabilities.

[0124] 2) Distribution sub-agent: Distribute the matched subtasks to the corresponding execution unit or sub-agent to ensure that the tasks can be efficiently processed. For example, distribute the "book a flight" subtask to the sub-agent with booking service capabilities.

[0125] 3) Task processing capability library: Store the task processing capability information of each execution unit and sub-agent to support relevance matching. For example: record that a certain execution unit has capabilities such as "data analysis" and "image processing".

[0126] In order to further improve the effect of task matching and distribution, the current loads of each execution unit and sub-agent can be considered during the task matching process to ensure the balance of task allocation. For example, tasks are preferentially allocated to the execution unit with a lighter load to avoid resource overload. The task matching strategy can also be optimized by combining historical task execution data to improve the accuracy of matching. For example, according to the historical success rate, the execution unit with better processing effects is preferentially matched.

[0127] On the basis of the above embodiment, if there is no target execution intelligent unit with task processing capabilities matching the complex subtask, the association matching sub-agent in the planning intelligent unit can also be controlled to return a prompt message of complex subtask association failure to the task splitting sub-agent, so that the task splitting sub-agent can split the originally split complex subtask into multiple simple subtasks according to the prompt message.

[0128] That is, this embodiment is used to handle the abnormal situation when the complex subtask cannot be matched to the appropriate target execution intelligent unit, and by returning the prompt information of the association failure to the task splitting sub-agent, and triggering the re-splitting of the complex subtask, it is ensured that the task can be processed continuously. This process includes exception handling (identifying the abnormal situation that the complex subtask cannot be matched to the target execution intelligent unit), information feedback (returning the prompt information of the association failure to the task splitting sub-agent, providing the basis for re-splitting) and task re-splitting (re-splitting the complex subtask into multiple simple subtasks according to the prompt information to ensure that the task can be processed continuously).

[0129] Specifically, when the relevance matching sub-agent tries to match complex sub-tasks, it finds that there is no target execution intelligent unit with the corresponding task processing capabilities. For example, the complex sub-task "Generate Investment Advice" requires data analysis, risk assessment and prediction capabilities, but there is no execution unit with these capabilities in the current system. And the relevance matching sub-agent generates a prompt message of association failure and returns it to the task splitting sub-agent. The prompt information includes the specific content of the complex sub-task, the reason for the matching failure (such as lack of capabilities), etc. For example, the prompt message is "The complex sub-task 'Generate Investment Advice' failed to match, the reason: lack of data analysis and risk assessment capabilities."

[0130] In the task re-splitting phase, the task splitting sub-agent can split the original complex sub-task into multiple simple sub-tasks according to the prompt information. For example, "generating investment advice" can be split into three simple sub-tasks: "collecting market data", "analyzing market trends" and "assessing investment risks".

[0131] To further improve the effectiveness of task re-splitting, we can first optimize the task splitting rules based on the reasons for the matching failure in the prompt information to ensure that the simple sub-tasks after re-splitting can be processed by the existing sub-agents. For example, for the prompt information of "lack of data analysis and risk assessment capabilities", we will prioritize splitting into simple sub-tasks that can be processed by the existing sub-agents. We can also dynamically adjust the task splitting strategy based on the current resource status of the system to ensure that the simple sub-tasks after re-splitting can be processed efficiently. For example, when system resources are tight, we can split complex sub-tasks into simple sub-tasks with smaller granularity.

[0132] Taking the intelligent customer service scenario as an example, when matching fails for the complex subtask of "processing a refund and re - placing an order", and it is confirmed that the reason for the failure is the lack of an execution task unit with order - processing capabilities, at this time, the complex subtask can be re - split into two simple subtasks of "processing a refund" and "re - placing an order" by the task - splitting sub - agent.

[0133] Based on any of the above embodiments, Figure 4 The following is a flowchart of a method for ensuring the execution of a subtask by completing an interrogation provided in an embodiment of the present disclosure, aiming to describe that in the process where a subtask has entered the execution stage, missing information can be further completed by means of interrogation to better ensure the effective execution of the subtask. Its process 400 includes the following steps:

[0134] Step 401: In response to the target execution intelligent unit or the target sub - agent finding missing necessary information during the execution of the corresponding subtask, control the supplementary interrogation sub - agent in the target execution intelligent unit or the target sub - agent to initiate a completion interrogation for the missing necessary information to the target user;

[0135] Step 402: Control the supplementary interrogation sub - agent in the target execution intelligent unit or the target sub - agent to continue executing the corresponding subtask according to the necessary information supplemented and replied by the target user.

[0136] In this embodiment, steps 401 - 402 are used to handle the abnormal situation where the target execution intelligent unit or the target sub - agent finds missing necessary information during the execution of the subtask. By initiating a completion interrogation to the target user and continuing to execute the subtask according to the user's supplementary reply, it is ensured that the task can be completed completely and accurately. The main key technical points involved are as follows:

[0137] 1) Information completion: Identify the missing necessary information during the execution of the subtask and initiate a completion interrogation to the user; 2) Task continuation: Continue to execute the subtask according to the missing information supplemented by the user to ensure that the task can be completed completely and accurately; 3) User experience optimization: By actively initiating a completion interrogation, avoid task interruption or failure caused by missing information and improve the user experience.

[0138] Among them, the process of information completion can be carried out according to a preset process:

[0139] 1) Missing information identification: The target execution intelligent unit or the target sub - agent finds missing necessary information during the execution of the subtask. For example, when executing the subtask of "booking an air ticket", it is found that the departure date information is missing.

[0140] 2) Initiation of completion interrogation: The supplementary interrogation sub - agent or the target sub - agent initiates a completion interrogation for the missing necessary information to the target user. For example, ask the user "What is your departure date?"

[0141] 3) User supplementary reply: The target user supplements the missing necessary information according to the completed question. For example, the user replies "The departure date is March 25th".

[0142] 4) Task continues to execute: The target execution intelligent unit or target sub-intelligent agent continues to execute the subtask according to the missing information supplemented by the user. For example, according to the departure date "March 25th", continue to execute the subtask of "booking an air ticket".

[0143] To further improve the effect of information completion, context information can also be combined in the completed question to provide more accurate questions. For example, when asking about the departure date, prompt the user "You mentioned that you are available on March 25th before. Do you want to choose this day to depart?" and try to support multi-round completed questions to gradually complete the missing necessary information. For example, ask about the departure date in the first round of questions and ask about the departure time in the second round of questions. Default values or recommended options can also be provided in the completed question to simplify user input. For example, when asking about the departure date, recommend "March 25th" as the default value. Even when it is found that the user has not provided the necessary information or has provided invalid information, exception handling is performed. For example, when the user has not provided the departure date, prompt the user "The departure date is a required field. Please supplement it."

[0144] Taking the intelligent customer service scenario as an example, when executing the subtask of "processing a refund", it is found that the necessary information is missing: the refund amount. So, it can control the supplementary inquiry sub-intelligent agent to ask the user: "What is your refund amount?" Then, get the user's reply: "The refund amount is 500 yuan", so that the target execution intelligent unit continues to execute the "processing a refund" subtask according to the supplementary information.

[0145] Considering the following core challenges that need to be faced in the scenario of completing open and complex tasks: 1) How to flexibly and efficiently understand user needs and accurately generate a task execution plan. Since user input is often vague and abstract, traditional simple queries or keyword matching are difficult to achieve accurate understanding and reasonable decomposition; 2) How to select and schedule appropriate intelligent agents to collaborate to complete complex tasks. Different tasks have diversity, complexity, and uncertainty, requiring the system to accurately identify task requirements and reasonably select and efficiently schedule multiple intelligent agents to work together according to task characteristics; 3) How to effectively display the task execution process and results through interactive interaction, so that users can fully understand and participate in the task execution to improve the quality of task completion and user satisfaction.

[0146] Therefore, based on the inventive concept provided in the above embodiment, this embodiment specifically constructs a unified intelligent task unit as the basic unit of intelligent interaction, realizing the unified scheduling and interaction of intelligent agents with different complexity levels.

[0147] As Figure 5-1 shown, each intelligent task unit consists of a task execution module, a communication interaction module, and a historical storage module. Among them, the task execution module supports multiple implementation methods, including basic script or API (Application Programming Interface) calls, large language model-driven agents, agents based on the combination of large language models and tools, and multi-agent collaborative workflows; the task execution module can receive different types of data input by users (such as text, files, and multimodal content), efficiently complete specific tasks through intelligent tools and collaborative workflows, and output results; the communication interaction module is responsible for unified interaction management with users or other agents, including receiving user input, providing feedback on execution results, and actively obtaining further feedback input from users; the historical storage module records the historical results of each task execution to support the efficient invocation of subsequent tasks and avoid repeated execution.

[0148] As Figure 5-2 shown, based on the above unified intelligent task unit, this embodiment further proposes a task completion system based on multi-agent collaboration. The system includes three core components: an interaction intelligent unit, a planning intelligent unit, and an execution intelligent unit.

[0149] Among them, the interaction intelligent unit, as the main interface between the system and the user, is responsible for natural dialogue interaction with the user. By using semantic understanding technology, it deeply analyzes the needs expressed by the user and accurately converts the user's vague needs into clear task requirement texts. When the user's needs are not clear or complex, the interaction intelligent unit can actively guide the user to provide further clarification input. In addition, the interaction intelligent unit also supports users to provide feedback on previous execution results to dynamically optimize the further execution of tasks.

[0150] Among them, after receiving the clear task requirement text passed by the interaction intelligent unit, the planning intelligent unit will automatically analyze the content and characteristics of the requirement text and dynamically generate a detailed task execution plan, including clarifying the task descriptions of each execution stage and the selection strategy of the corresponding execution intelligent unit. The planning intelligent unit intelligently matches and determines the most suitable execution intelligent unit according to the task requirement characteristics and the capabilities of each agent, and coordinates the execution order of tasks to ensure the efficiency and effect of task execution.

[0151] Among them, the execution intelligent unit is specifically responsible for the actual execution of tasks. It receives the task requirement text clearly transmitted by the planning intelligent unit and the execution output of the previous stage, and calls the corresponding intelligent agents, specialized tools or intelligent collaborative workflows to complete the tasks. When difficulties are encountered during the task execution or additional user confirmation is required, the execution intelligent unit will actively initiate interaction with the user to obtain user feedback for optimizing the execution process. After the task is completed, the execution intelligent unit will feedback the execution result of the task to the planning intelligent unit, so that the planning intelligent unit can integrate the evaluation results, dynamically update the subsequent task planning, or determine whether the task execution meets the user requirements and enter the task summary stage.

[0152] As Figure 5-2 shown, the complete collaboration process of the task completion system based on multi-agent collaboration proposed in this embodiment is specifically as follows:

[0153] 1) The user puts forward task requirements and clarifies the requirements through natural language interaction with the interaction intelligent unit;

[0154] 2) The interaction intelligent unit analyzes and clarifies the task requirements, and converts them into clear task requirement texts to be transmitted to the planning intelligent unit;

[0155] 3) The planning intelligent unit automatically generates a task execution plan according to the task requirement text, including determining the phased execution strategy of the task and selecting the optimal execution intelligent unit;

[0156] 4) The execution intelligent unit specifically executes each stage of the task according to the planning information transmitted by the planning intelligent unit, and calls intelligent agents, tools or collaborative workflows to complete the tasks;

[0157] 5) The execution intelligent unit actively interacts and gives feedback to the user to obtain user confirmation or additional input to ensure task quality and user satisfaction;

[0158] 6) After the completion of a task stage, the execution intelligent unit feeds back the execution result to the planning intelligent unit, and the planning intelligent unit determines whether it is necessary to adjust the subsequent plan or enter the task summary stage;

[0159] 7) After all task stages are executed or the termination conditions are met, the planning intelligent unit aggregates and integrates the overall execution results to form the final output result, which is displayed to the user through the interaction intelligent unit.

[0160] The solution provided in this embodiment has the following inventive points compared with the existing solutions:

[0161] 1) A unified and general intelligent task unit architecture is proposed, realizing the construction of agents with different complexities and types and a unified task scheduling mechanism. This mechanism can efficiently manage the execution and interaction of agents, tools, and workflows, and support the intelligent task unit to actively initiate interactions with users, significantly improving task execution efficiency and user participation.

[0162] 2) A multi-agent collaboration system based on interactive intelligent units, planning intelligent units, and execution intelligent units is constructed, realizing intelligent task decomposition, agent selection, and dynamic automatic scheduling, and completing complex tasks through efficient collaboration. This system can flexibly handle various emergencies and dynamic changes during task execution, ensuring the efficient completion of tasks.

[0163] 3) An interactive visualization display of the task execution process and results is provided. Users can not only intuitively observe the execution situation and achievements of each stage of the task but also participate in the task execution process at any time, helping the system further optimize the task execution strategy through real-time feedback. This interactive task completion mechanism significantly enhances users' understanding and control of the task completion process.

[0164] The present invention is applicable to intelligent customer service, intelligent assistants, decision support systems, and various intelligent information service platforms (for relevant examples, refer to the examples given in the above embodiments), and can greatly improve the automation and intelligent processing level of complex tasks, meeting users' higher-level requirements for interactive experience and task completion effects.

[0165] Further referring to Figure 6 , as an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of a fully automatic complex task processing device based on multi-agent collaboration. This device embodiment corresponds to the Figure 2 shown method embodiment, and this device can be specifically applied to various electronic devices.

[0166] As shown in Figure 6As shown in the figure, the complex task full-automatic processing device 600 based on multi-agent collaboration in this embodiment may include: a complete task requirement acquisition unit 601, a complete task splitting unit 602, a subtask allocation unit 603, and a subtask result summarization and presentation unit 604. Among them, the complete task requirement acquisition unit 601 is configured to obtain a complete task requirement by interacting with the target user who proposes the original task requirement; the complete task splitting unit 602 is configured to split the complete task requirement into multiple subtasks each of which includes at least one complex subtask; where a complex subtask refers to a subtask that requires at least two sub-agents to process according to a collaboration process, and a single sub-agent is used to process a simple subtask; the subtask allocation unit 603 is configured to send the complex subtasks in each subtask to the corresponding target execution intelligent unit and send the simple subtasks in each subtask to the corresponding target sub-agent; where the target execution intelligent unit includes at least two sub-agents for processing complex subtasks; the subtask result summarization and presentation unit 604 is configured to summarize the subtask execution results returned by each target execution intelligent unit and each target sub-agent, and present the summarized task processing result to the target user.

[0167] In this embodiment, in the complex task full-automatic processing device 600 based on multi-agent collaboration: the specific processing of the complete task requirement acquisition unit 601, the complete task splitting unit 602, the subtask allocation unit 603, and the subtask result summarization and presentation unit 604 and the technical effects brought thereby can respectively refer to Figure 2 the relevant descriptions of steps 201-204 in the corresponding embodiment, which will not be elaborated here.

[0168] In some other optional implementation manners of this embodiment, the complete task requirement acquisition unit 601 may be further configured to:

[0169] Determine the missing items of task information according to the original task requirement proposed by the target user;

[0170] Obtain the missing task information through at least one round of asking questions for filling in the missing information initiated to the target user;

[0171] In response to obtaining the missing task information corresponding to all the missing items of task information, determine the complete task requirement based on all the missing task information and the original task requirement.

[0172] In some other alternative implementation manners of this embodiment, the target execution intelligent unit is an execution intelligent unit for processing complex subtasks split from the complete task requirements. The execution intelligent unit is an intelligent task unit for executing tasks. The intelligent task unit is designed to include at least two sub-intelligent agents and collaboration process information for characterizing the collaboration process that should be followed between different sub-intelligent agents. The collaboration process information is determined based on the processing logic specifically adapted to the complex subtasks to be executed.

[0173] In some other alternative implementation manners of this embodiment, the intelligent task unit is further designed to include: an information storage module and a communication module for realizing the transfer of information storage between different sub-intelligent agents.

[0174] In some other alternative implementation manners of this embodiment, the complete task splitting unit 602 is further configured to:

[0175] Send the complete task requirements to a preset planning intelligent unit; wherein, the planning intelligent unit is an intelligent task unit for disassembling and executing the planning of the received complete task requirements;

[0176] Control the planning intelligent unit to perform task disassembly in sequence through a task understanding sub-intelligent agent and a task splitting sub-intelligent agent according to a preset first collaboration process, and obtain multiple sub-tasks including at least one complex sub-task; wherein, the priority of the complex sub-task disassembled by the task splitting sub-intelligent agent is higher than the priority of the simple sub-task disassembled.

[0177] In some other alternative implementation manners of this embodiment, the complete task splitting unit 602 is further configured to:

[0178] Send the complete task requirements to a preset planning intelligent unit; wherein, the planning intelligent unit is an intelligent task unit for disassembling and executing the planning of the received complete task requirements;

[0179] Control the planning intelligent unit to perform task disassembly and inspection confirmation in sequence through a task understanding sub-intelligent agent, a task splitting sub-intelligent agent and a split result self-inspection sub-intelligent agent according to a preset second collaboration process, and obtain multiple sub-tasks including at least one complex sub-task; wherein, the priority of the complex sub-task disassembled by the task splitting sub-intelligent agent is higher than the priority of the simple sub-task disassembled, and the split result self-inspection sub-intelligent agent is used to confirm the correctness and executability of each sub-task disassembled.

[0180] In some other alternative implementation manners of this embodiment, the sub-task allocation unit 603 is further configured to:

[0181] The control and planning intelligent unit distributes the complex subtasks in each subtask to the corresponding target execution intelligent unit and the simple subtasks in each subtask to the corresponding target sub-intelligent agent in sequence according to the preset third collaboration process; wherein, the relevance matching sub-intelligent agent is used to determine the target execution intelligent unit or target sub-intelligent agent respectively matched with each subtask based on the relevance between the task and the task processing capabilities possessed by the sub-intelligent agent, and the relevance matching sub-intelligent agent associates the complex subtasks with the target execution intelligent unit having the matching task processing capabilities.

[0182] In some other alternative implementation manners of this embodiment, the complex task full-automatic processing device 600 based on multi-agent collaboration may further include:

[0183] In response to the non-existence of a target execution intelligent unit having task processing capabilities matching the complex subtasks, control the relevance matching sub-intelligent agent in the control and planning intelligent unit to return a prompt message of failure to associate the complex subtasks to the task splitting sub-intelligent agent, so that the task splitting sub-intelligent agent re-splits the originally split complex subtasks into multiple simple subtasks according to the prompt message.

[0184] In some other alternative implementation manners of this embodiment, the complex task full-automatic processing device 600 based on multi-agent collaboration may further include:

[0185] In response to the discovery of the lack of necessary information by the target execution intelligent unit or target sub-intelligent agent during the execution of the corresponding subtask, control the supplementary inquiry sub-intelligent agent in the target execution intelligent unit or the target sub-intelligent agent to initiate a supplementary question to the target user regarding the missing necessary information;

[0186] Control the supplementary inquiry sub-intelligent agent in the target execution intelligent unit or the target sub-intelligent agent to continue executing the corresponding subtask according to the necessary information supplemented and replied by the target user.

[0187] This embodiment exists as a device embodiment corresponding to the above method embodiment. The complex task full-automatic processing device based on multi-agent collaboration provided in this embodiment first interacts with the target user who initiates the original task requirement, and then determines the complete task requirement through the information obtained in the interaction process. Then, the complete task requirement is split into multiple sub-tasks including at least one complex sub-task. The complex sub-task is used to distinguish from the simple sub-task that can be completed by a single intelligent agent alone, and it refers to a sub-task that requires at least two sub-intelligent agents to process according to the collaboration process. Next, the complex sub-tasks are assigned to the target execution intelligent unit, and the simple sub-tasks are assigned to the target sub-intelligent agents. The target execution intelligent unit contains at least two sub-intelligent agents for processing the complex sub-task. Finally, the sub-task execution results returned by each target execution intelligent unit and each target sub-intelligent agent are summarized. That is, this solution first tries to complement the missing part of the original task information by interacting with the target user, so as to obtain a more accurate complete task requirement. And by introducing an execution intelligent unit dedicated to processing complex sub-tasks, the complex sub-task can be processed more centrally by multiple sub-intelligent agents arranged according to the collaboration process included in the execution intelligent unit. This not only does not need to split the complete task requirement into the finest-grained simple sub-tasks to reduce the difficulty of task decomposition, but also can obtain better sub-task processing results through the execution intelligent unit integrated with multiple sub-intelligent agents.

[0188] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, which 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, and when the instructions are executed by the at least one processor, the at least one processor can implement the complex task full-automatic processing method based on multi-agent collaboration described in any of the above embodiments.

[0189] According to an embodiment of the present disclosure, the present disclosure also provides a readable storage medium, which stores computer instructions, and when the computer instructions are executed, the computer can implement the complex task full-automatic processing method based on multi-agent collaboration described in any of the above embodiments.

[0190] According to an embodiment of the present disclosure, the present disclosure also provides a computer program product, and when the computer program is executed by a processor, it can implement the complex task full-automatic processing method based on multi-agent collaboration described in any of the above embodiments.

[0191] Figure 7FIG. shows a schematic block diagram of an exemplary electronic device 700 that can be used to implement 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 can also represent various forms of mobile devices, such as, personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely exemplary and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0192] As Figure 7 shown, the device 700 includes a computing unit 701 that can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the device 700 can also be stored. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0193] A plurality of components in the device 700 are connected to the I / O interface 705, including: an input unit 706, such as a keyboard, a mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a magnetic disk, an optical disk, etc.; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 709 allows the device 700 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0194] The computing unit 701 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated 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 701 executes the various methods and processes described above, such as the fully automatic processing method for complex tasks based on multi-agent collaboration. For example, in some embodiments, the fully automatic processing method for complex tasks based on multi-agent collaboration can be implemented as a computer software program that is tangibly contained in a machine-readable medium, such as the storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the fully automatic processing method for complex tasks based on multi-agent collaboration described above can be executed. Alternatively, in other embodiments, the computing unit 701 can be configured to execute the fully automatic processing method for complex tasks based on multi-agent collaboration by any other suitable means (e.g., by means of firmware).

[0195] Various embodiments of the systems and techniques described above in this document 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), system-on-chip systems (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor, and can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0196] The program code for implementing the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the program codes are executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program codes can be executed entirely on the machine, partially on the machine, as an independent software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0197] In the context of this disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection 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 include, 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 a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0198] To provide for interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can also be used to provide for 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 acoustic, speech, or tactile input).

[0199] The systems and techniques described herein can be implemented in a computing system including a back-end component (e.g., as a data server), or a computing system including a middleware component (e.g., an application server), or a computing system including a front-end component (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system including any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of a communication network include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0200] A computer system may include a client and a server. The client and the server are generally far from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system to solve the defects of high management difficulty and weak business scalability existing in traditional physical hosts and virtual private server (VPS) services.

[0201] According to the technical solution of the embodiment of the present disclosure, first, interact with the target user who initiates the original task requirement, so as to determine the complete task requirement through the information obtained in the interaction process. Then, split the complete task requirement into multiple sub-tasks including at least one complex sub-task. The complex sub-task is used to distinguish from the simple sub-task that can be completed by only a single sub-intelligent agent, which refers to a sub-task that requires at least two sub-intelligent agents to process according to a collaboration process. Next, allocate the complex sub-task to the target execution intelligent unit and allocate the simple sub-task to the target sub-intelligent agent. The target execution intelligent unit contains at least two sub-intelligent agents for processing the complex sub-task. Finally, summarize the sub-task execution results returned by each target execution intelligent unit and each target sub-intelligent agent. That is, this solution first tries to complement the possibly missing part of the original task information by interacting with the target user, so as to obtain a more accurate complete task requirement. And by introducing an execution intelligent unit dedicated to processing complex sub-tasks, the complex sub-task can be processed more centrally by multiple sub-intelligent agents arranged according to the collaboration process included in the execution intelligent unit. It not only does not need to split the complete task requirement into the finest-grained simple sub-tasks to reduce the task splitting difficulty, but also can obtain better sub-task processing results through the execution intelligent unit integrated with multiple sub-intelligent agents.

[0202] It should be understood that the various forms of processes shown above can be used, with steps reordered, added or deleted. For example, the steps described in the present disclosure can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution disclosed in the present disclosure can be achieved. There is no limitation herein.

[0203] The above specific embodiments do not constitute a limitation to the protection scope of the present 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 the present disclosure shall be included within the protection scope of the present disclosure.

Claims

1. A fully automatic processing method for complex tasks based on multi-agent collaboration, including: Obtaining a complete task requirement by interacting with the target user who puts forward the original task requirement; Splitting the complete task requirement into multiple subtasks each containing at least one complex subtask; wherein, the complex subtask refers to a subtask that requires at least two sub-agents to be processed according to a collaboration process, and a single sub-agent is used to process simple subtasks; Sending the complex subtasks in each of the subtasks to the corresponding target execution intelligent unit, and sending the simple subtasks in each of the subtasks to the corresponding target sub-agent; wherein, the target execution intelligent unit contains at least two sub-agents for processing the complex subtasks; Summarizing the subtask execution results returned by each of the target execution intelligent units and each of the target sub-agents, and presenting the summarized task processing result to the target user.

2. The method according to claim 1, wherein The obtaining a complete task requirement by interacting with the target user who puts forward the original task requirement includes: Determining the missing items of task information according to the original task requirement put forward by the target user; Obtaining the missing task information through at least one round of asking questions for complementing the missing information initiated to the target user; In response to obtaining the missing task information corresponding to all the missing items of task information, determining the complete task requirement based on all the missing task information and the original task requirement.

3. The method according to claim 1, wherein, The target execution intelligent unit is an execution intelligent unit for processing the complex subtasks split from the complete task requirement. The execution intelligent unit is an intelligent task unit for executing tasks. The intelligent task unit is designed to include at least two sub-agents and collaboration process information for characterizing the collaboration process that should be followed between different sub-agents. The collaboration process information is determined based on the processing logic specifically adapted to the executed complex subtasks.

4. The method according to claim 3, wherein The intelligent task unit is also designed to include: an information storage module and a communication module for realizing the storage and transfer of information between different sub-agents.

5. The method according to claim 3 or 4, wherein, The splitting the complete task requirement into multiple subtasks each containing at least one complex subtask includes: Sending the complete task requirement to a preset planning intelligent unit; wherein, the planning intelligent unit is an intelligent task unit for disassembling and executing the plan for the received complete task requirement; Controlling the planning intelligent unit to disassemble the task in sequence through a task understanding sub-agent and a task splitting sub-agent according to a preset first collaboration process, and obtaining multiple subtasks split out and containing at least one complex subtask; wherein, the priority of the complex subtasks disassembled by the task splitting sub-agent is higher than the priority of the simple subtasks disassembled.

6. The method according to claim 3 or 4, wherein The splitting the complete task requirement into multiple subtasks each containing at least one complex subtask includes: Sending the complete task requirement to a preset planning intelligent unit; wherein, the planning intelligent unit is an intelligent task unit for disassembling and executing the plan for the received complete task requirement; Control the planning intelligent unit to perform task decomposition and inspection confirmation in sequence through a task understanding sub-intelligent unit, a task splitting sub-agent, and a splitting result self-inspection sub-intelligent unit according to a preset second collaboration process, so as to obtain multiple sub-tasks including at least one complex sub-task after splitting; wherein, the priority of the complex sub-task decomposed by the task splitting sub-agent is higher than that of the simple sub-task decomposed, and the splitting result self-inspection sub-intelligent unit is used to confirm the correctness and executability of each sub-task after splitting.

7. The method according to claim 5, wherein The step of sending the complex sub-tasks in each of the sub-tasks to the corresponding target execution intelligent unit and sending the simple sub-tasks in each of the sub-tasks to the corresponding target sub-agent includes: Control the planning intelligent unit to send the complex sub-tasks in each of the sub-tasks to the corresponding target execution intelligent unit and send the simple sub-tasks in each of the sub-tasks to the corresponding target sub-agent in sequence through a relevance matching sub-intelligent unit and a sending sub-intelligent unit according to a preset third collaboration process; wherein, the relevance matching sub-intelligent unit is used to determine the target execution intelligent unit or target sub-agent respectively matched with each sub-task based on the relevance between the task and the task processing capabilities possessed by the sub-agent, and the relevance matching sub-intelligent unit associates the complex sub-task with the target execution intelligent unit having the matching task processing capabilities.

8. The method according to claim 7, further comprising: In response to the non-existence of a target execution intelligent unit having task processing capabilities matching the complex sub-task, control the relevance matching sub-intelligent unit in the planning intelligent unit to return a prompt message indicating the failure of associating the complex sub-task to the task splitting sub-agent, so that the task splitting sub-agent re-splits the originally split complex sub-task into multiple simple sub-tasks according to the prompt message.

9. The method according to any one of claims 1-4, further comprising: In response to the discovery of the lack of necessary information during the execution of the corresponding sub-task by the target execution intelligent unit or the target sub-agent, control the supplementary inquiry sub-intelligent unit in the target execution intelligent unit or the target sub-agent to initiate a supplementary question regarding the lack of necessary information to the target user; Control the supplementary inquiry sub-intelligent unit in the target execution intelligent unit or the target sub-agent to continue executing the corresponding sub-task according to the necessary information supplemented and replied by the target user.

10. A complex task full-automatic processing device based on multi-agent collaboration, comprising: A complete task requirement acquisition unit configured to obtain a complete task requirement by interacting with a target user who proposes the original task requirement; A complete task splitting unit configured to split the complete task requirement into multiple sub-tasks including at least one complex sub-task; wherein, the complex sub-task refers to a sub-task that requires at least two sub-agents to process according to a collaboration process, and a single sub-agent is used to process simple sub-tasks; A subtask allocation unit, configured to send complex subtasks among the respective subtasks to corresponding target execution intelligent units, and send simple subtasks among the respective subtasks to corresponding target sub-agents; wherein, at least two sub-agents for processing the complex subtasks are included in the target execution intelligent unit; A subtask result summarization and presentation unit, configured to summarize the subtask execution results returned by the respective target execution intelligent units and the respective target sub-agents, and present the summarized task processing result to the target user.

11. An electronic device, 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, and the instructions are executed by the at least one processor so that the at least one processor can execute the fully automatic complex task processing method based on multi-agent collaboration according to any one of claims 1-9.

12. A non-transitory computer-readable storage medium storing computer instructions, the computer instructions being used to cause the computer to execute the fully automatic complex task processing method based on multi-agent collaboration according to any one of claims 1-9.

13. A computer program product, comprising a computer program, the computer program, when executed by a processor, implements the steps of the fully automatic complex task processing method based on multi-agent collaboration according to any one of claims 1-9.

Images