Multi-agent cooperation-based explicit result acquisition class task processing method and related device

Abstract

The invention provides a multi-agent cooperation-based explicit result acquisition class task processing method and a related device, and relates to the technical field of artificial intelligence such as generative large language models and agents. The method comprises the steps of determining a to-be-solved target task according to natural language input of a user; the target task is decomposed into a plurality of sub-target tasks at least including clear result acquisition class tasks, each sub-target task is correspondingly issued to each sub-target agent at least including clear result acquisition agents, and different sub-agents are used for processing different types of sub-tasks; controlling each sub-target agent to output a sub-processing result corresponding to the subordinate sub-target task in combination with the personalized preference of the user, wherein the sub-processing result comprises a clear result which is output by the clear result acquisition agent and corresponds to the subordinate clear result acquisition class task; and summarizing the sub-processing results to obtain a target processing result corresponding to the target task. By applying the scheme, a better comprehensive task processing effect can be brought under the condition of lower comprehensive cost.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, specifically to the field of artificial intelligence technology such as generative large language models and intelligent agents, and more particularly to a method, device, electronic device, computer-readable storage medium and computer program product for processing tasks of obtaining clear results based on multi-agent collaboration. Background Art

[0002] With the rapid development and iteration of generative large language models, they have a better understanding of user input requirements and the ability to give corresponding results.

[0003] In order to make the output of the generative large language model more in line with specific needs, an intelligent agent was constructed by using the generative large language model as the base model and combining it with pre-set role parameters.

[0004] How to use intelligent agents to solve complex problems raised by users is still a technical problem that needs to be urgently solved by technical personnel in this field. Summary of the invention

[0005] The embodiments of the present disclosure propose a method, device, electronic device, computer-readable storage medium and computer program product for processing clear result acquisition tasks based on multi-agent collaboration.

[0006] In the first aspect, an embodiment of the present disclosure proposes a method for processing clear result acquisition tasks based on multi-agent collaboration, including: determining a target task to be solved based on a user's natural language input; decomposing the target task into a plurality of sub-target tasks that at least include a clear result acquisition task, and issuing each sub-target task to each sub-target agent that includes at least a clear result acquisition agent; wherein different sub-agents are used to process different types of sub-tasks; controlling each sub-target agent to output a sub-processing result corresponding to the sub-target task to which it belongs in combination with the user's personalized preferences; wherein the sub-processing result includes a clear result output by the clear result acquisition agent that corresponds to the clear result acquisition task to which it belongs; and summarizing each sub-processing result to obtain a target processing result corresponding to the target task.

[0007] In the second aspect, the embodiments of the present disclosure propose a device for processing clear result acquisition tasks based on multi-agent collaboration, including: a target task determination unit, configured to determine the target task to be solved according to the user's natural language input; a task decomposition and corresponding issuance unit, configured to decompose the target task into multiple sub-target tasks including at least a clear result acquisition task, and issue each sub-target task to each sub-target agent including at least a clear result acquisition agent; wherein different sub-agents are used to process different types of sub-tasks; a sub-target agent control output unit, configured to control each sub-target agent to output sub-processing results corresponding to the sub-target task to which it belongs in combination with the user's personalized preferences; wherein the sub-processing results include clear results output by the clear result acquisition agent and corresponding to the clear result acquisition task to which it belongs; a sub-processing result aggregation unit, configured to aggregate each sub-processing result to obtain a target processing result corresponding to the target task.

[0008] In a third aspect, an embodiment of the present disclosure provides 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 when the at least one processor executes, it is able to implement the method for processing tasks based on multi-agent collaboration for obtaining clear results as described in the first aspect.

[0009] In a fourth aspect, an embodiment of the present disclosure provides a non-transitory computer-readable storage medium storing computer instructions, which are used to enable a computer to implement the method for processing tasks such as obtaining clear results based on multi-agent collaboration as described in the first aspect when executed.

[0010] In a fifth aspect, an embodiment of the present disclosure provides a computer program product comprising a computer program, which, when executed by a processor, can implement the various steps of the method for processing tasks based on multi-agent collaboration for obtaining clear results as described in the first aspect.

[0011] The clear result acquisition task processing solution based on multi-agent collaboration provided by the present disclosure is that the main agent determines the target task to be solved according to the user's natural language input, and then decomposes the target task into multiple sub-target tasks including at least the clear result acquisition task, and then sends each sub-target task to each sub-target agent including at least the clear result acquisition agent. Then, each sub-target agent will output the sub-processing result corresponding to the sub-target task under the arrangement of the main agent in combination with the user's personalized preference. Finally, the main agent summarizes the sub-processing results to obtain the target processing result corresponding to the target task.

[0012] That is, the present disclosure adopts an agent cluster formed by a pre-built main agent and multiple sub-agents to handle the task requirements proposed by users, wherein the main agent is responsible for understanding the user's task requirements and decomposing the overall task requirements into multiple sub-target tasks that can be executed by different sub-agents, and each sub-agent processes the sub-task that matches itself according to the mobilization of the main agent. That is, through the collaboration between the main agent and each sub-agent, different components of complex tasks can be handled in their respective roles, and because different sub-agents are pre-built to be dedicated to handling different types of tasks, the scheme of the main agent and multiple sub-agents used in the present disclosure to collaboratively handle task requirements has better processing effects on a single type of task than using a single, all-powerful agent, and the relatively small-scale sub-agents are also convenient for flexible addition and modification of corresponding functions. This scheme can bring better comprehensive task processing effects at a lower overall cost.

[0013] 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

[0014] 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:

[0015] Figure 1 is an exemplary system architecture in which the present disclosure may be applied;

[0016] Figure 2 A flowchart of a method for processing a task of obtaining a clear result based on multi-agent collaboration provided in an embodiment of the present disclosure;

[0017] Figure 3 A flowchart of a method for determining a target task according to natural language input provided by an embodiment of the present disclosure;

[0018] Figure 4 A flowchart of a method for decomposing a target task into multiple sub-target tasks including a clear result acquisition task according to task elements provided by an embodiment of the present disclosure;

[0019] Figure 5 A flowchart of a method for creating a sub-agent provided in an embodiment of the present disclosure;

[0020] Figure 6 A flowchart of a method for controlling each sub-goal agent to perform a corresponding sub-goal task according to a determined execution order provided by an embodiment of the present disclosure;

[0021] Figure 7 A two-branch schematic diagram of controlling a sub-target agent to output a corresponding sub-processing result provided by an embodiment of the present disclosure;

[0022] Figure 8-1 to Figure 8-10 All of them are example diagrams of a method for processing a task of obtaining a clear result based on multi-agent collaboration in an application scenario provided by an embodiment of the present disclosure;

[0023] Fig. 9 A structural block diagram of a device for processing tasks of obtaining explicit results based on multi-agent collaboration provided by an embodiment of the present disclosure;

[0024] Fig.10 A schematic diagram of the structure of an electronic device suitable for executing a method for processing tasks of obtaining clear results based on multi-agent collaboration, provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0025] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description. It should be noted that the embodiments in the present disclosure and the features in the embodiments may be combined with each other without conflict.

[0026] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0027] Figure 1 An exemplary system architecture 100 is shown to which embodiments of the method, apparatus, electronic device, and computer-readable storage medium for processing explicit result-acquisition tasks based on multi-agent collaboration of the present disclosure can be applied.

[0028] like Figure 1 As 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 optical fiber cables, etc.

[0029] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Terminal devices 101, 102, 103 and server 105 can be installed with various applications for realizing information communication between the two, such as complex task processing applications, browser applications, instant messaging applications, etc.

[0030] Terminal devices 101, 102, 103 and server 105 can be hardware or software. When terminal devices 101, 102, 103 are hardware, they can be various electronic devices with display screens, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, etc.; when terminal devices 101, 102, 103 are software, they can be installed in the electronic devices listed above, which can be implemented as multiple software or software modules, or as a single software or software module, and no specific limitation is made here. When server 105 is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server; when the server is software, it can be implemented as multiple software or software modules, or as a single software or software module, and no specific limitation is made here.

[0031] The server 105 can provide various services through various built-in applications. Taking a complex task processing application that can provide a one-click processing service for complex tasks as an example, the server 105 can achieve the following effects when running the complex task processing application: first, receive the natural language input transmitted by the user through the terminal devices 101, 102, and 103 through the network 104, and then determine the target task to be solved based on the natural language input; then, decompose the target task into a plurality of sub-target tasks that at least include a clear result acquisition task, and issue each sub-target task to each sub-target agent that at least includes a clear result acquisition agent, and different sub-agents are used to process different types of subtasks; next, control each sub-target agent to output a sub-processing result corresponding to the corresponding sub-target task in combination with the user's personalized preferences, and the sub-processing result includes a clear result corresponding to the clear result acquisition task output by the clear result acquisition agent; finally, summarize each sub-processing result to obtain a target processing result corresponding to the target task.

[0032] Furthermore, the server 105 may also transmit the target processing result back to the terminal devices 101 , 102 , 103 via the network 104 , so that the terminal devices 101 , 102 , 103 display the received target processing result to the user.

[0033] It should be noted that, in addition to being obtained from the terminal devices 101, 102, 103 through the network 104, the natural language input can also be pre-stored in the server 105 in various ways. Therefore, when the server 105 detects that the data has been stored locally (for example, starting to process the tasks to be processed before), it can choose to directly obtain the data locally. In this case, the exemplary system architecture 100 may also not include the terminal devices 101, 102, 103 and the network 104.

[0034] Since processing complex tasks requires more computing resources and stronger computing power, the methods for processing tasks based on multi-agent collaboration to obtain clear results provided in the subsequent embodiments of the present disclosure are generally executed by the server 105 with stronger computing power and more computing resources, and accordingly, the processing device for processing tasks based on multi-agent collaboration to obtain clear results is generally also set in the server 105. However, it should also be pointed out that when the terminal devices 101, 102, and 103 also have computing power and computing resources that meet the requirements, the terminal devices 101, 102, and 103 can also complete the various operations assigned to the server 105 by the complex task processing applications installed thereon, and then output the same results as the server 105. Especially when there are multiple terminal devices with different computing capabilities, but the terminal device where the complex task processing application is located has strong computing capabilities and more remaining computing resources, the terminal device can be allowed to perform the above-mentioned operations, thereby appropriately reducing the computing pressure of the server 105. Correspondingly, the task processing device for obtaining clear results 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 also not include the server 105 and the network 104.

[0035] It should be noted that the main agent and each sub-target agent can be installed on the server 105 at the same time, and each sub-target agent called and controlled by the main agent can also be installed on other servers or terminal devices different from the server 105, which is not specifically limited here.

[0036] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0037] Please refer to Figure 2 , Figure 2 A flowchart of a method for processing a task of obtaining a clear result based on multi-agent collaboration provided by an embodiment of the present disclosure, wherein process 200 includes the following steps:

[0038] Step 201: Determine a target task to be solved according to a natural language input of a user;

[0039] This step aims to obtain the clear result of the task processing method based on multi-agent collaboration by the execution subject (e.g. Figure 1 The server 105 carrying the main agent shown in the figure fully understands and analyzes the received natural language input of the user (for example, through Figure 1 The network 104 shown receives the natural language input by the user through the terminal devices 101, 102, 103 used, and then determines the target task that the user wants to solve, that is, the target task corresponds to a task requirement.

[0040] Specifically, this step is usually a comprehensive process involving multiple steps, involving complex intelligent reasoning links such as language understanding, context analysis, and task modeling. For example, receiving input, preprocessing and normalization, semantic understanding and intent recognition, task modeling, and reasoning.

[0041] Among them, the main intelligent agent as the execution subject first receives the user's natural language input, then preprocesses and normalizes the received natural language input to ensure that it can be effectively understood and processed, and then understands the user's intention reflected by the input information obtained after preprocessing and normalization based on the context, thereby determining the target task.

[0042] Among them, the input form of natural language may be voice, text, image, video or other forms of language data. For the convenience of processing, conversion technology can also be used to convert non-text natural language input into natural language text that is easy to identify and process; considering that natural language input has multiple forms of request information (such as direct requests, indirect requests and multi-step requests), in order to effectively understand the user's actual task requirements, pre-processing and normalization processing including word segmentation and annotation, noise removal, spelling correction and synonym processing can also be used; in the intent recognition stage, keyword extraction, entity recognition and contextual information can be used to accurately identify the intention that the user wants to express through natural language input; in order to convert the identified intention into a specific task, task modeling operations including task abstraction (mapping the user's needs to a task framework that the system can recognize) and target task definition can also be used to finally obtain the target task corresponding to the natural language input.

[0043] Step 202: decompose the target task into multiple sub-target tasks including at least a clear result acquisition task, and issue each sub-target task to each sub-target agent including at least a clear result acquisition agent;

[0044] Based on step 201, this step aims to decompose the target task into multiple sub-target tasks by the execution subject, and issue each sub-target task to each sub-target agent. Among them, the multiple sub-target tasks decomposed in this step at least include clear result acquisition tasks, and correspondingly, each sub-target agent should at least include a clear result acquisition agent dedicated to processing clear result acquisition tasks.

[0045] Among them, different sub-agents are used to process different types of sub-tasks, that is, sub-agents dedicated to a single type of task are created in advance according to different task types. There are various ways to divide task types, such as according to task processing steps, task processing methods, task complexity, data types involved in the task, etc., which are not specifically limited here, that is, different sub-agents are used to process different atomic tasks (that is, the smallest unit of task, the "smallest" here is a relative concept, not an absolute concept, and it is the smallest task unit that can be decomposed at present).

[0046] Among them, clear result acquisition tasks refer to tasks with clear results, which can be unique or non-unique, but the results should be clear and will not be doubted, such as four arithmetic operations tasks, fixed parameter query tasks, matching tasks with fixed corresponding relationships, etc. Taking the task of "finding a travel plan from place A to place B" as an example, it can be seen that: this task clearly states that the departure place of the travel plan to be obtained is place A and the destination is place B. Then, in the absence of the necessary information to determine the travel plan, multiple feasible travel plans can be clearly determined. Of course, the number of travel plans will also decrease with the increase of task restrictions, such as adding task restrictions such as "departure on the afternoon of X", "fastest travel plan", "travel plan that takes no more than Y hours", and "travel plan without transfer and direct access", but it will not affect the clear results. In other words, the clear result acquisition task is a type of task that only needs to determine the clear existing results that the user expects to obtain and feedback them to him. Therefore, the execution focus of this type of task is to obtain clear existing results that meet the task requirements.

[0047] It should be noted that this embodiment requires that in this step, the multiple sub-target tasks decomposed from the target task should include at least a clear result acquisition task. In addition, there is no limitation on the task types of other decomposed sub-target tasks, which can be any type of task. Correspondingly, it is also necessary to use a matching sub-target intelligent agent with the corresponding type of task processing capability.

[0048] Step 203: Control each sub-goal agent to output the sub-processing results corresponding to the sub-goal task in combination with the user's personalized preferences;

[0049] On the basis of step 202, this step aims to control each sub-target agent to output the sub-processing results corresponding to the sub-target task respectively in combination with the user's personalized preferences by the above-mentioned execution subject. Among them, the multiple sub-processing results at least include the clear results corresponding to the clear result acquisition type task output by the clear result acquisition agent, and the remaining sub-processing results can also be the results obtained by other agents when processing the corresponding type of tasks. The sub-processing result can be obtained by relying on other sub-processing results, or it can be determined by itself according to the task description information of the sub-target task without relying on other sub-processing results.

[0050] That is, each sub-goal agent, under the call or arrangement of the main agent, outputs a sub-processing result corresponding to the issued sub-goal task in combination with the user's personalized preferences. Specifically, the main agent can simultaneously inform the execution order or execution trigger signal when issuing each sub-goal task to each sub-goal agent, so that each sub-goal agent executes the corresponding sub-goal task in accordance with the informed execution order or executes the corresponding sub-goal task when judging that the current state meets the requirements of the execution trigger signal, that is, each sub-goal agent actively executes the corresponding sub-goal task at the appropriate time; the main agent can also actively call each sub-goal agent to execute the corresponding sub-goal task as required when determining that different sub-goal agents meet the execution time, that is, each sub-goal agent only passively executes the corresponding sub-goal task according to the call instruction.

[0051] Both the active execution mechanism and the passive execution mechanism mentioned above belong to different collaborative forms of the main intelligent agent and multiple sub-intelligent agents to jointly process tasks. The specific choice can be based on the actual needs of the actual application scenario, and no specific limitation is made here.

[0052] Step 204: Summarize the sub-processing results to obtain the target processing result corresponding to the target task.

[0053] On the basis of step 203, this step aims at allowing the above-mentioned execution subject to finally obtain the target processing result corresponding to the target task by summarizing each sub-processing result.

[0054] Specifically, when summarizing the sub-processing results, one can fully refer to the decomposition method of decomposing multiple sub-target tasks from the target task, and adopt a variety of processing methods including deduplication, adjustment of expressions, adjustment of sentence patterns and enrichment of expressions, so that the final target processing results can not only meet the user's task requirements, but also highlight the key requirement result information as much as possible and improve information recognition.

[0055] The method for processing tasks of explicit result acquisition based on multi-agent collaboration provided by the embodiment of the present disclosure adopts an agent cluster formed by a pre-built main agent and multiple sub-agents to process the task requirements proposed by the user, wherein the main agent is responsible for understanding the task requirements of the user and decomposing the overall task requirements into multiple sub-target tasks that can be executed by different sub-agents, and each sub-agent processes the sub-tasks that match itself according to the mobilization of the main agent. That is, through the collaboration between the main agent and each sub-agent, different components of complex tasks can be processed in their respective roles, and since different sub-agents are pre-built to be dedicated to processing different types of tasks, the scheme of the main agent and multiple sub-agents used in the present disclosure to collaboratively process task requirements has better processing effects on a single type of task than using a single, all-powerful agent, and the relatively small-scale sub-agents are also convenient for flexibly adding and changing corresponding functions, and the scheme can bring better comprehensive task processing effects at a lower comprehensive cost.

[0056] To deepen your understanding of how to determine the target tasks, please refer to Figure 3 , Figure 3 A flowchart of a method for determining a target task according to natural language input provided by an embodiment of the present disclosure, wherein process 300 includes the following steps:

[0057] Step 301: Convert the user's natural language input into natural language text;

[0058] This step aims to convert the natural language input in various forms into natural language text that is easy to recognize and output by the above-mentioned execution subject, that is, to uniformly convert various forms into text form. For example, speech-to-text technology can be used to convert natural speech signals in speech form into natural language text.

[0059] Step 302: performing intent recognition on the natural language text to obtain an intent recognition result including a clear result acquisition intent;

[0060] On the basis of step 301, this step aims to perform intent recognition on the natural language text by the above-mentioned execution subject to obtain an intent recognition result including the explicit result acquisition intent. That is, the intent recognition result at least includes the explicit result acquisition intent, and may also include other intents, which are not specifically limited here.

[0061] An implementation method including but not limited to may include the following specific steps:

[0062] First, the execution subject performs semantic understanding on the natural language text to obtain a semantic understanding result; then, the execution subject determines a preliminary intention (or suspected intention) based on the semantic understanding result. In order to improve the accuracy of intention determination, the execution subject may also use the personalized preferences of the user to perform personalized modification on the preliminary intention to obtain an intention recognition result including a clear result acquisition intention.

[0063] That is, this implementation method first determines the preliminary intention that directly corresponds to the natural language text by using semantic understanding, and then uses the user's personalized preferences to personalize the preliminary intention, such as screening, excluding, and supplementing implicit or missing information, so that the final intention recognition result is more in line with the user's actual expectations.

[0064] In addition to this implementation method, you can also use the various technologies involved in the multiple processing links mentioned in step 201 to combine them with actual application scenarios to construct other implementation methods to obtain a determined intent recognition result, which will not be listed here one by one.

[0065] Step 303: Determine the target task to be solved according to the intention recognition result.

[0066] Based on step 302, this step aims to determine the target task to be solved by the execution subject according to the intention recognition result. The target task includes a task for obtaining a clear existing result.

[0067] Specifically, in order to convert the identified intent into a specific task, task modeling operations including task abstraction (mapping the user's needs into a task framework recognizable by the system) and target task definition can also be used to ultimately obtain the target task corresponding to the natural language input.

[0068] This embodiment provides a more specific implementation method for determining the target task based on natural language input through steps 301 to 303, which includes the key processing steps of conversion of expression form, intent recognition, and determination of the target task based on the intent recognition result, so as to provide an implementation plan with high feasibility and more accurate determination of the target task as much as possible.

[0069] To deepen your understanding of how to decompose the target task into multiple sub-target tasks, see Figure 4 , Figure 4 A flowchart of a method for decomposing a target task into multiple sub-target tasks including a clear result acquisition task according to task elements is provided in an embodiment of the present disclosure, wherein the process 400 includes the following steps:

[0070] Step 401: determining a plurality of task elements constituting a target task;

[0071] This step aims to enable the above-mentioned execution subject to determine multiple task elements that constitute the target task, and different task elements correspond to different types of tasks.

[0072] Among them, task elements refer to the basic units or components that constitute the target task. Each target task may be composed of several independent elements, which are the key steps, subtasks or resources necessary to achieve the target task. Each task element may correspond to an independent task, and they work together to promote the completion of the target task.

[0073] Assuming the target task is "develop a new software product", the target task can be decomposed into multiple task elements, each of which corresponds to a different type of task:

[0074] 1) Demand analysis (management task): determine user needs and market needs;

[0075] 2) System design (technical tasks): architecture design, database design, interface design;

[0076] 3) Coding development (technical tasks): programmers write codes and implement functions;

[0077] 4) Testing and quality control (technical and management tasks): unit testing, integration testing, quality assurance;

[0078] 5) Marketing and promotion (management and communication tasks): formulate marketing plans and promote products.

[0079] Step 402: Decomposing a clear result acquisition task for acquiring a clear result from the target task according to a single task element specifically for determining a clear existing result matching the question raised by the user;

[0080] This step aims to decompose the clear result acquisition tasks for obtaining clear results from the target tasks by the above-mentioned execution subject. Specifically, it relies on a single task element for determining a clear existing result that matches the question raised by the user.

[0081] Step 403: Decompose the target task into multiple sub-target tasks each containing only a single task element.

[0082] This step aims to decompose the target task into multiple sub-target tasks containing only a single task element, by referring to the method of decomposing by task elements in step 402 by the above-mentioned execution subject to obtain clear result acquisition tasks, that is, the sub-target tasks at least include clear result acquisition tasks.

[0083] This embodiment provides a task decomposition method through steps 401 to 403, which is based on the task elements that constitute the target task and decomposes each task element into sub-target tasks. Combined with the pre-definition of the task elements (for example, according to the method of constructing different sub-agents), the task decomposition can be flexibly completed according to actual conditions.

[0084] Furthermore, considering that even the same task element may be attached with additional conditions, it is necessary to decompose it into sub-target tasks carrying the additional conditions when decomposing the task, so that when the sub-target tasks are subsequently issued to the sub-target agents, the appropriate sub-agent can be selected as the sub-target agent based on the additional conditions.

[0085] One implementation method including but not limited to may be: in response to the existence of additional conditions corresponding to the task elements, decompose the target task into multiple sub-target tasks containing only a single task element and the corresponding additional conditions, and the additional conditions may include: a specified task processing method and / or a specified result presentation form. Among them, different task processing methods are abstracted from different processing logics for the same type of tasks (for example, the same type of tasks may have multiple different processing logics, but they can all get the correct answer in the end); the result presentation form may include: at least one of text, table, image, video, and interactive card.

[0086] For a deeper understanding of how to pre-create different sub-agents, see Figure 5 , Figure 5 A flowchart of a method for creating a sub-agent provided by an embodiment of the present disclosure, wherein the process 500 includes the following steps:

[0087] Step 501: Determine a set of task types for creating a sub-agent;

[0088] Step 502: For each type of task in the task type set, a sub-agent having the task processing capability of the corresponding type of task is constructed.

[0089] This embodiment provides a solution for building a sub-agent with task processing capabilities of the corresponding type of task for each type of task determined through steps 501 and 502. Further, the number of sub-agents corresponding to each type of task may be only one or multiple, and the multiple sub-agents may be different or completely the same. Whether the distinction is needed depends on whether the type of task usually contains further additional conditions.

[0090] Specifically, for a first target type task with at least two task processing methods, a sub-agent corresponding to each task processing method can be constructed; for a second target type task with at least two result presentation forms, a sub-agent corresponding to each result presentation form can be constructed.

[0091] To deepen your understanding of how multiple sub-goal agents execute in an orderly manner to ensure that the desired sub-processing results are finally obtained, please refer to Figure 6 , Figure 6 A flowchart of a method for controlling each sub-goal agent to perform a corresponding sub-goal task according to a determined execution order is provided in an embodiment of the present disclosure, wherein the process 600 includes the following steps:

[0092] Step 601: Determine the execution order between different sub-goal tasks with dependency relationships as serial;

[0093] Step 602: determining the execution order between different sub-goal tasks without dependency relationship as parallel;

[0094] That is, step 601 - step 602 determines whether there is a dependency relationship between different sub-goal tasks, thereby determining the serial or parallel execution order.

[0095] Step 603: Control each sub-goal agent executed in a serial manner to sequentially output the sub-processing results corresponding to the corresponding sub-goal tasks in combination with personalized preferences;

[0096] Among them, the sub-processing results output by the previous sub-goal agent executed in a serial manner will serve as another part of the input of the subsequent sub-goal agent.

[0097] Step 604: Control each sub-goal agent executed in parallel to output the sub-processing results corresponding to the corresponding sub-goal task in combination with personalized preferences.

[0098] Step 603 and step 604 provide an implementation scheme in which the execution subject controls each sub-goal agent to execute in the serial or parallel execution order determined above.

[0099] Based on any of the above embodiments, considering that the user may modify the content of part of the output sub-processing results or issue new restriction information at any time during the entire process of each sub-target agent performing the corresponding sub-target task and outputting the sub-processing results, please refer to Figure 7 , Figure 7 A two-branch schematic diagram of controlling a sub-target agent to output a corresponding sub-processing result provided by an embodiment of the present disclosure, wherein the process 700 includes the following steps:

[0100] Step 701: In response to the instruction input box being in the selected state, controlling the currently executed sub-goal agent to pause outputting the corresponding sub-processing result;

[0101] The instruction input box is used for the user to input instructions, and the instruction input box is in an unselected state during the process of the sub-target agent outputting the corresponding sub-processing result. That is, in the schemes of the embodiments provided in the disclosure, once the instruction input box is in a selected state, it may represent that the user needs to input some new instructions and needs to interrupt the original sub-processing result output process.

[0102] Step 702: In response to no new instruction being generated in the process of the instruction input box being restored from the selected state to the unselected state, controlling the currently executed sub-goal agent to continue outputting the corresponding sub-processing result;

[0103] This step corresponds to a branch situation in which no new instruction is generated during the process of the instruction input box being restored from the selected state to the unselected state, that is, the user does not enter a new instruction, so the sub-processing result output result originally interrupted by the selection of the instruction input box will continue.

[0104] Step 703: in response to a new instruction being generated in the process of the instruction input box being restored from the selected state to the unselected state, extracting correction information from the new instruction;

[0105] This step corresponds to another branch situation in which a new instruction is generated during the process of the instruction input box being restored from a selected state to an unselected state. In this case, the above-mentioned execution subject is required to extract correction information from the new instruction, and the correction information includes corrections to the original information and additions of new information.

[0106] Step 704: Determine the sub-target agent affected by the correction information;

[0107] Step 705: Control the affected sub-target agent to re-output the corresponding sub-processing result in combination with the correction information.

[0108] Steps 704 and 705 are that the execution subject first determines the sub-target agent affected by the correction information, and then controls the affected sub-target agent to re-output the corresponding sub-processing result in combination with the correction information. Specifically, if all sub-target agents are affected, the execution can also be restarted from the process of outputting the sub-processing result of the first executed sub-target agent based on the correction information.

[0109] On the basis of any of the above embodiments, in order to deepen the understanding of how to aggregate the sub-processing results to obtain the target processing result, a specific implementation method is also provided, including the following steps:

[0110] First, according to the decomposition method of decomposing each sub-target task from the target task, each sub-processing result is spliced ​​in reverse to obtain a reverse splicing result; then, the repeated and redundant parts in the reverse splicing result are removed and the expression style of the reverse splicing result is adjusted to obtain the target processing result corresponding to the target task. Furthermore, the expression style of the reverse splicing result can be adjusted to the expression style required by the personalized preference, or the expression style can be adjusted to other required expression styles.

[0111] To deepen understanding, the present disclosure also provides a specific implementation scheme that attempts to eliminate the defects of the prior art and overcome the problems of the prior art in combination with the actual defects of the prior art in specific application scenarios:

[0112] The existing way to meet needs is mainly through search engines, using multi-layer system funnels to recall, sort, and mix heterogeneous results of relevant web pages. Each strategy funnel sorts web pages based on basic relevance, user feedback behavior, authority and other information and truncates the output to the next layer. The disadvantage of this system is that it can only match relevance from the content level through relevance, and cannot understand and solve problems for users from the task level. In addition, web pages can only use fixed content to meet user needs. When users express personalized needs and need to be further satisfied through multiple rounds, high-quality and coherent satisfaction effects cannot be achieved. Again, the entire matching process is not explainable to users, and users can only perform the final information screening.

[0113] Related assistant products based on big models usually try to solve this problem through an all-round assistant based on big models. However, it is difficult for a single big model to achieve high-quality satisfaction in different open fields and tasks. In professional fields and special scenarios, it is difficult for a single assistant to establish trust among users and achieve the expected satisfaction.

[0114] That is, the relevant technologies have the following difficulties in completing tasks in open fields:

[0115] 1) How to efficiently understand and decompose the needs expressed by users into key steps;

[0116] 2) How to find the best way to meet each key step;

[0117] 3) How to output the execution process and integration results in a user-friendly and complete manner.

[0118] To solve the above difficulties, this embodiment proposes a solution to end-to-end task completion based on multi-agent collaboration. The features of this solution are as follows:

[0119] Before, searches were all single-point, with one query (query word, query statement, search word, search statement) completing one requirement. Each requirement required the user to think and define it by themselves, which was too costly for the user. For a complex task, the user often did not know how to break down the task and conceive multiple queries for searching. The paradigm proposed in this embodiment is intended to help users complete tasks thoroughly. The task completion paradigm can help users complete the problems that need to be solved or the tasks to be completed in one step. Compared with the single query satisfaction of the search, the task satisfaction paradigm deeply understands the user's complete task and breaks it down to complete it, which is more in line with the user's needs.

[0120] This embodiment is based on the collaboration between multiple agents to achieve the decomposition and multi-step satisfaction of user needs, and finally automatically integrates them into a complete solution that can meet the entire task. This solution can complete the following functions:

[0121] 1) Understand the key steps of the requirements, build a main agent, understand the intention of the user's expressed requirements and break them down into key steps that can be completed by multiple sub-task agents;

[0122] 2) Subtask agent generation: Based on the understanding of key steps and agent capabilities, the main agent is responsible for generating multiple subtask agent candidates required to complete the current task;

[0123] 3) Multi-agent collaborative task completion: multiple subtask agents complete the entire task through multi-agent scheduling collaboration according to the requirements of the task and key steps, as well as user preference information input, and generate step-by-step results of task completion;

[0124] 4) Complete the integrated output of the plan. The main agent integrates the task decomposition process and the results of collaboration among multiple subtask agents, selects the best presentation form, and presents the final complete process and results.

[0125] That is, this embodiment constructs a main agent to schedule multiple subtask agents to collaborate, and completes the entire task through collaboration between the main agent and multiple subtask agents, and collaboration between multiple subtask agents. Compared with search or intelligent assistant products, it can be completed using agents with better effects in subdivided fields, and the effect of satisfaction is better;

[0126] By providing a scheduling and distribution mechanism based on an end-to-end generative large model, the large model has stronger expressive power, can overcome the problems of insufficient effect consistency and global optimality of the traditional multi-layer sorting mechanism, and can achieve task-oriented global optimal combination scheduling optimization. From the perspective of model capability, it can achieve end-to-end alignment training of the model from the perspective of agent understanding and task understanding.

[0127] At the same time, since the generative completion of the entire task is personalized, everything from task decomposition to sub-task completion is generated based on user input and user preferences. Different solution decompositions, sub-task agent completion solutions and presentation effects are generated for different users to improve user satisfaction.

[0128] The implementation block diagram of this solution is as follows Figure 8-1 As shown in the figure, the input of the system includes: user input requirements and a set of distributable agents (including the basic settings of these agents, mounted plug-ins, workflows, etc.), and the output is multi-level generated content to complete the task. The entire system is built based on multi-agent collaboration, and mainly includes two types of agents: 1) Main Agent: It performs overall understanding, decomposition and concatenation of tasks and gives the final result. 2) Task Agent: The main agent dispatches subtask agents to collaborate to complete the tasks issued by the main agent.

[0129] The solution provided in this embodiment can be widely applied to various information-satisfying applications or independent products. The following takes the scenario where the BaiX application is used as an example to illustrate the specific application form of this embodiment:

[0130] 1) While performing traditional searches, users can switch to the invented satisfying mode with just one click (such as the “AI button” on the bottom bar in the figure below).

[0131] The user enters "How to arrange a 2-day tour in Harbin" (see Figure 8-2 ), display the results of web search and intelligent answer, switch to the page of "Harbin 2-Day Tour - Intelligent Private Customization" through the AI ​​button, and display the results generated for user personalization, as shown below Figure 8-3 The task completion page displayed includes personalized requirements such as "departing from Beijing", "single white-collar worker", "special forces travel", and "high cost performance".

[0132] On the task completion page, the work results of multiple task completion agents are displayed, including the host agent travel planner, and sub-task agents "Harbin Travel Assistant", "What to Wear in Harbin at This Time", "Harbin Attractions Connection Master" and "Harbin Hotel Recommendation Assistant".

[0133] The results of each task completion agent are displayed through a structured, personalized, and user-friendly generated UI, and users can download and share the results, such as Figure 8-4 , Figure 8-5 and Figure 8-6 shown.

[0134] The system fully personalizes the solutions presented to different users, such as Figure 8-7 and Figure 8-8Shows the effect of “Harbin 2-day tour” for elderly users.

[0135] In the revisit form, when the user returns to the BaiX application, even if he does not enter a new query in the search bar, he can return to the task list that can display the history by clicking the entry button. At this time, he can still click to enter the corresponding result page, such as Figure 8-9 and Figure 8-10 Other applications may also have the same independent entrance (not shown separately).

[0136] It can be more specifically seen from the above examples that this embodiment has the following improvements and technical effects compared to the prior art:

[0137] 1) Task satisfaction paradigm: the user's request is no longer regarded as an information retrieval, but upgraded to a complete multi-agent task completion satisfaction model

[0138] The current search engines can retrieve information by matching requests with relevance, and the intelligent assistants can decompose a task into multiple steps and search for them step by step. However, decomposing tasks and satisfying users' complete needs through multi-agent collaboration is a new user needs satisfaction model. The decomposition and multi-agent capability alignment training is proposed, which can achieve high consistency between task decomposition and task completion, greatly improving the task completion effect.

[0139] 2) Multi-agent candidate generation based on multi-agent collaboration

[0140] Based on the results of task decomposition, the target collaborative agent set is generated end-to-end through the large model. This is a brand-new approach that is different from the traditional method of sorting agents through a retrieval recall sorting system. The task of multi-agent collaboration meets the need to combine multiple agents when the task is not completed. The large model is required to have a deep understanding of the task decomposition and the ability of each agent to carefully depict the boundary of the problem, and finally obtain the optimal agent set under the current goal through combinatorial optimization. The difficulties here include:

[0141] Difficulty 1: In-depth understanding of the capabilities of intelligent agents, understanding the boundaries of the capabilities of intelligent agents and the tasks that intelligent agents are good at, and carefully depicting the subtle differences between capabilities, such as the different creative styles of painting agents and the subtle differences in capabilities in photo editing.

[0142] Difficulty 2: Understanding the matching relationship between tasks and intelligent agent capabilities is not a semantic similarity match, but a deep match in task completion capabilities. It requires a deep characterization and understanding of the tasks, task types and task boundaries. Matching the open set of intelligent agent capabilities and task requirements is a great challenge.

[0143] Difficulty 3: Given a multi-agent collaborative task, obtaining the optimal combination is a combinatorial optimization problem that requires combinatorial optimization.

[0144] 3) The task completion process and results are generated end-to-end by the agent. Currently, it is difficult for search engines to display the retrieval process to users, while intelligent assistants display thinking results and single-step execution results, and the display UI is basically a pre-set style. Thanks to the task decomposition, combination and multi-subtask completion mode of multi-agent collaboration, this embodiment can display the results in an end-to-end generated manner, including the steps, style and results of each step of the display content. And the results of each step and the subtask results completed by each subtask agent can be further interacted. This experience is a comprehensive innovation.

[0145] Further references Fig. 9 As an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of a device for processing tasks based on multi-agent collaboration to obtain clear results. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0146] like Fig. 9 As shown, the clear result acquisition task processing device 900 based on multi-agent collaboration of this embodiment may include: a target task determination unit 901, a task decomposition and corresponding distribution unit 902, a sub-target agent control output unit 903, and a sub-processing result summary unit 904. The target task determination unit 901 is configured to determine the target task to be solved according to the natural language input of the user; the task decomposition and corresponding distribution unit 902 is configured to decompose the target task into a plurality of sub-target tasks including at least a clear result acquisition task, and distribute each sub-target task to each sub-target agent including at least a clear result acquisition agent; wherein different sub-agents are used to process different types of sub-tasks; the sub-target agent control output unit 903 is configured to control each sub-target agent to output a sub-processing result corresponding to the sub-target task respectively in combination with the personalized preference of the user; wherein the sub-processing result includes a clear result corresponding to the clear result acquisition task output by the clear result acquisition agent; and the sub-processing result summary unit 904 is configured to summarize each sub-processing result to obtain a target processing result corresponding to the target task.

[0147] In this embodiment, in the multi-agent collaborative clear result acquisition task processing device 900, the specific processing of the target task determination unit 901, the task decomposition and corresponding distribution unit 902, the sub-target agent control output unit 903, and the sub-processing result summary unit 904 and the technical effects thereof can be referred to respectively. Figure 2The relevant descriptions of steps 201 - 204 in the corresponding embodiment are not repeated here.

[0148] In some optional implementations of this embodiment, the target task determination unit 901 may include:

[0149] a conversion subunit configured to convert a user's natural language input into a natural language text;

[0150] An intention recognition subunit is configured to perform intention recognition on the natural language text and obtain an intention recognition result including a clear result acquisition intention;

[0151] The target task determination subunit is configured to determine the target task to be solved according to the intention recognition result; wherein the target task includes a task for obtaining a clearly existing result.

[0152] In some optional implementations of this embodiment, the intention identification subunit may be further configured to:

[0153] Perform semantic understanding on natural language text to obtain semantic understanding results;

[0154] The preliminary intention is determined based on the semantic understanding results, and the preliminary intention is personalized and modified using the user's personalized preferences to obtain the intent recognition result that includes the clear result acquisition intention.

[0155] In some optional implementations of this embodiment, the task decomposition and corresponding issuing unit 902 may include a task decomposition subunit configured to decompose the target task into a plurality of sub-target tasks, and the task decomposition subunit may include:

[0156] A task element determination module is configured to determine a plurality of task elements constituting a target task; wherein different task elements correspond to different types of tasks;

[0157] A clear result acquisition task decomposition module is configured to decompose clear result acquisition tasks for obtaining clear results from the target task according to a single task element specifically for determining a clear existing result matching the question raised by the user;

[0158] The sub-goal task decomposition module is configured to decompose the target task into multiple sub-goal tasks each containing only a single task element; wherein the sub-goal tasks at least include tasks for obtaining clear results.

[0159] In some optional implementations of this embodiment, the sub-goal task decomposition module may be further configured as follows:

[0160] In response to the existence of additional conditions corresponding to the task elements, the target task is decomposed into multiple sub-target tasks containing only a single task element and the corresponding additional conditions; wherein the additional conditions include: a specified task processing method and / or a specified result presentation form.

[0161] In some optional implementations of this embodiment, the multi-agent collaboration-based explicit result acquisition task processing device 900 may further include: a sub-agent construction unit configured to construct different sub-agents, and the sub-agent construction unit may include:

[0162] A task type set determination subunit configured to determine a task type set for creating a sub-agent;

[0163] The construction sub-unit is configured to construct a sub-agent with task processing capability of the corresponding type of task for each type of task in the task type set.

[0164] In some optional implementations of this embodiment, the construction subunit may be further configured as follows:

[0165] In response to the first target type task having at least two task processing methods, a sub-agent corresponding to each task processing method is constructed for the first target type task; wherein the different task processing methods are abstracted from different processing logics for the same type of tasks.

[0166] In some optional implementations of this embodiment, the construction subunit may be further configured as follows:

[0167] In response to the second target type task having at least two result presentation forms, a sub-agent corresponding to each result presentation form is constructed for the second target type task; wherein the result presentation form includes: at least one of: text, table, image, video, and interactive card.

[0168] In some optional implementations of this embodiment, the multi-agent collaboration-based clear result acquisition task processing device 900 may further include:

[0169] A serial order determination unit is configured to determine the execution order between different sub-goal tasks with dependency relationships as serial;

[0170] A parallel order determination unit is configured to determine the execution order between different sub-goal tasks without dependency relationship as parallel;

[0171] Correspondingly, the sub-goal agent control output unit 903 can be further configured as follows:

[0172] Controlling each sub-goal agent executed in a serial manner to sequentially output sub-processing results corresponding to the corresponding sub-goal tasks in combination with personalized preferences; wherein the sub-processing results output by the previous sub-goal agent executed in a serial manner will serve as another part of the input of the subsequent sub-goal agent;

[0173] Each sub-goal agent that is controlled to be executed in parallel outputs the sub-processing results corresponding to the sub-goal task to which it belongs in combination with personalized preferences.

[0174] In some optional implementations of this embodiment, the multi-agent collaboration-based clear result acquisition task processing device 900 may further include:

[0175] The pause output control unit is configured to control the currently executed sub-goal agent to pause outputting the corresponding sub-processing result in response to the instruction input box being in a selected state; wherein the instruction input box is used for the user to input instructions, and the instruction input box is in an unselected state during the process of the sub-goal agent outputting the corresponding sub-processing result;

[0176] The continue output control unit is configured to control the currently executed sub-target agent to continue outputting the corresponding sub-processing result in response to the fact that no new instruction is generated during the process of the instruction input box being restored from the selected state to the unselected state.

[0177] In some optional implementations of this embodiment, the multi-agent collaboration-based clear result acquisition task processing device 900 may further include:

[0178] a correction information extraction unit, configured to extract correction information from a new instruction in response to a new instruction being generated in the process of the instruction input box being restored from a selected state to an unselected state;

[0179] An affected sub-goal agent determination unit, configured to determine the sub-goal agent affected by the correction information;

[0180] The re-output control unit is configured to control the affected sub-target agent to re-output the corresponding sub-processing result in combination with the correction information.

[0181] In some optional implementations of this embodiment, the sub-processing result aggregation unit 904 may include:

[0182] The reverse splicing subunit is configured to splice each sub-processing result in a reverse manner according to the decomposition method of decomposing each sub-target task from the target task to obtain a reverse splicing result;

[0183] The optimization processing subunit is configured to remove repeated redundant parts in the reverse splicing result, adjust the expression style of the reverse splicing result, and obtain the target processing result corresponding to the target task.

[0184] In some optional implementations of this embodiment, the optimization processing subunit may include a presentation style adjustment module configured to adjust the presentation style of the reverse splicing result. The presentation style adjustment module may be further configured to:

[0185] Adjust the presentation style of the reverse splicing results to the presentation style required by your personalized preferences.

[0186] This embodiment exists as a device embodiment corresponding to the above method embodiment. The clear result acquisition task processing device based on multi-agent collaboration provided by this embodiment adopts an agent cluster formed by a pre-built main agent and multiple sub-agents to process the task requirements proposed by the user, wherein the main agent is responsible for understanding the user's task requirements and decomposing the overall task requirements into multiple sub-target tasks that can be executed by different sub-agents, and each sub-agent processes the sub-tasks that match itself according to the mobilization of the main agent. That is, through the collaboration between the main agent and each sub-agent, different components of the complex task can be processed in their respective roles, and because different sub-agents are pre-built to be dedicated to processing different types of tasks, the scheme of the main agent and multiple sub-agents cooperating to process task requirements adopted by the present disclosure has better processing effects on a single type of task than using a single, all-powerful agent, and the relatively small-scale sub-agents are also convenient for flexibly adding and changing corresponding functions. This scheme can bring better comprehensive task processing effects at a lower comprehensive cost.

[0187] 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 the instructions are executed by the at least one processor so that the at least one processor can implement the method for processing tasks based on multi-agent collaboration for obtaining clear results as described in any of the above embodiments when executing.

[0188] According to an embodiment of the present disclosure, the present disclosure also provides a readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to implement the method for processing tasks such as obtaining clear results based on multi-agent collaboration described in any of the above embodiments when executed.

[0189] According to an embodiment of the present disclosure, the present disclosure also provides a computer program product, which, when executed by a processor, can implement the method for processing tasks of obtaining clear results based on multi-agent collaboration described in any of the above embodiments.

[0190] Fig.10A schematic block diagram of an example electronic device 1000 that can be used to implement an embodiment of the present disclosure is shown. 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 processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0191] like Fig.10 As shown, the device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1002 or a computer program loaded from a storage unit 1008 into a random access memory (RAM) 1003. In the RAM 1003, various programs and data required for the operation of the device 1000 can also be stored. The computing unit 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0192] A number of components in the device 1000 are connected to the I / O interface 1005, including: an input unit 1006, such as a keyboard, a mouse, etc.; an output unit 1007, such as various types of displays, speakers, etc.; a storage unit 1008, such as a disk, an optical disk, etc.; and a communication unit 1009, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1009 allows the device 1000 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0193] The computing unit 1001 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1001 performs the various methods and processes described above, such as a clear result acquisition class task processing method based on multi-agent collaboration. For example, in some embodiments, a clear result acquisition class task processing method based on multi-agent collaboration may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 1008. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 1000 via the ROM 1002 and / or the communication unit 1009. When the computer program is loaded into the RAM 1003 and executed by the computing unit 1001, one or more steps of the clear result acquisition class task processing method based on multi-agent collaboration described above may be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured in any other appropriate manner (e.g., by means of firmware) to execute a clear result acquisition type task processing method based on multi-agent collaboration.

[0194] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations 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 that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0195] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0196] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, 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 disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0197] To provide 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 pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the 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 input, voice input, or tactile input).

[0198] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0199] A computer system may include a client and a server. The client and the server are generally remote 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 corresponding 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 cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and virtual private servers (VPS) services.

[0200] The technical solution of the disclosed embodiment adopts an agent cluster formed by a pre-built main agent and multiple sub-agents to process the task requirements proposed by the user, wherein the main agent is responsible for understanding the user's task requirements and decomposing the overall task requirements into multiple sub-target tasks that can be executed by different sub-agents, and each sub-agent processes the sub-task that matches itself according to the mobilization of the main agent. That is, through the collaboration between the main agent and each sub-agent, different components of complex tasks can be processed in their respective roles, and because different sub-agents are pre-built to be dedicated to processing different types of tasks, the scheme of the main agent and multiple sub-agents used in the present disclosure to collaboratively process task requirements has better processing effects on a single type of task than using a single, all-powerful agent, and the relatively small-scale sub-agents are also convenient for flexibly adding and changing corresponding functions. This scheme can bring better comprehensive task processing effects at a lower overall cost.

[0201] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0202] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for processing tasks of explicit result acquisition based on multi-agent collaboration, applied to the main agent, comprising: Determine the target task to be solved based on the user's natural language input; Decomposing the target task into a plurality of sub-target tasks including at least a task of obtaining a clear result, and issuing each of the sub-target tasks to each sub-target agent including at least a clear result obtaining agent; wherein different sub-agents are used to process different types of sub-tasks; Controlling each of the sub-goal agents to output sub-processing results corresponding to the corresponding sub-goal tasks in combination with the personalized preferences of the user; wherein the sub-processing results include the explicit results corresponding to the corresponding explicit result acquisition tasks output by the explicit result acquisition agent; The sub-processing results are summarized to obtain a target processing result corresponding to the target task.

2. The method according to claim 1, wherein: The step of determining the target task to be solved according to the user's natural language input includes: Converting the user's natural language input into natural language text; Performing intent recognition on the natural language text to obtain an intent recognition result including a clear result acquisition intent; Determine the target task to be solved according to the intention recognition result; wherein the target task includes a task for obtaining a clear existing result.

3. The method according to claim 2, wherein: The performing intent recognition on the natural language text to obtain an intent recognition result includes: Performing semantic understanding on the natural language text to obtain a semantic understanding result; A preliminary intention is determined according to the semantic understanding result, and the preliminary intention is personalized modified using the personalized preference of the user to obtain an intention recognition result including a clear result acquisition intention.

4. The method according to claim 1, wherein: Decomposing the target task into a plurality of sub-target tasks comprises: Determine a plurality of task elements constituting the target task; wherein different task elements correspond to different types of tasks; Decomposing a clear result acquisition task for acquiring a clear result from the target task according to a single task element specifically for determining a clear existing result matching the question raised by the user; The target task is decomposed into a plurality of sub-target tasks each including only a single task element; wherein the sub-target tasks at least include the clear result acquisition type task.

5. The method according to claim 4, wherein: Decomposing the target task into a plurality of sub-target tasks each containing only a single task element comprises: In response to the existence of additional conditions corresponding to the task elements, the target task is decomposed into multiple sub-target tasks each containing only a single task element and the corresponding additional conditions; wherein the additional conditions include: a specified task processing method and / or a specified result presentation form.

6. The method according to claim 1, wherein: The process of constructing different sub-agents includes: Determining a set of task types for creating sub-agents; For each type of task in the task type set, a sub-agent having the task processing capability of the corresponding type of task is constructed.

7. The method according to claim 6, wherein: For each type of task in the task type set, a sub-agent having the task processing capability of the corresponding type of task is constructed, including: In response to the first target type task having at least two task processing methods, a sub-agent corresponding to each task processing method is constructed for the first target type task; wherein different task processing methods are abstracted from different processing logics for the same type of tasks.

8. The method according to claim 6, wherein: For each type of task in the task type set, a sub-agent having the task processing capability of the corresponding type of task is constructed, including: In response to the second target type task having at least two result presentation forms, a sub-agent corresponding to each result presentation form is constructed for the second target type task; wherein the result presentation form includes: at least one of: text type, table type, image type, video type, and interactive card type.

9. The method according to claim 1, further comprising: Determine the execution order between different sub-goal tasks with dependencies as serial; Determine the execution order between different sub-goal tasks that do not have dependencies as parallel; Correspondingly, the controlling each of the sub-target agents to output the sub-processing results corresponding to the corresponding sub-target tasks respectively in combination with the personalized preferences of the user includes: Controlling each sub-goal agent executed in the serial manner to sequentially output sub-processing results corresponding to the corresponding sub-goal tasks in combination with the personalized preferences; wherein the sub-processing results output by the previous sub-goal agent executed in the serial manner will serve as another part of the input of the subsequent sub-goal agent; Each sub-goal agent executed in the parallel manner is controlled to output a sub-processing result corresponding to the corresponding sub-goal task in combination with the personalized preference.

10. The method according to claim 9, further comprising: In response to the instruction input box being in a selected state, controlling the currently executed sub-goal agent to pause outputting the corresponding sub-processing result; wherein the instruction input box is used for the user to input instructions, and the instruction input box is in an unselected state during the process of the sub-goal agent outputting the corresponding sub-processing result; In response to no new instruction being generated in the process of the instruction input box being restored from the selected state to the unselected state, the currently executed sub-goal agent is controlled to continue outputting the corresponding sub-processing result.

11. The method according to claim 10, further comprising: In response to a new instruction being generated in the process of the instruction input box being restored from the selected state to the unselected state, extracting correction information from the new instruction; Determining the sub-target agent affected by the correction information; Control the affected sub-target agent to re-output the corresponding sub-processing result in combination with the correction information.

12. The method according to any one of claims 1 to 11, wherein: The summarizing of the sub-processing results to obtain a target processing result corresponding to the target task includes: According to the decomposition method of decomposing the target task into the sub-target tasks, the sub-processing results are spliced ​​in a reverse manner to obtain a reverse splicing result; Repetitive and redundant parts in the reverse splicing result are removed, and the expression style of the reverse splicing result is adjusted to obtain a target processing result corresponding to the target task.

13. The method according to claim 12, wherein: The adjusting the expression style of the reverse splicing result includes: The expression style of the reverse splicing result is adjusted to the expression style required by the personalized preference.

14. A device for processing tasks of obtaining clear results based on multi-agent collaboration, applied to a main agent, comprising: A target task determination unit, configured to determine a target task to be solved according to a natural language input of a user; The task decomposition and corresponding distribution unit is configured to decompose the target task into a plurality of sub-target tasks including at least a task of obtaining a clear result, and distribute each of the sub-target tasks to each sub-target agent including at least a clear result obtaining agent; wherein different sub-agents are used to process different types of sub-tasks; The sub-goal agent control output unit is configured to control each of the sub-goal agents to output the sub-processing results corresponding to the corresponding sub-goal tasks in combination with the personalized preferences of the user; wherein the sub-processing results include the explicit results corresponding to the explicit result acquisition tasks output by the explicit result acquisition agent; The sub-processing result summarizing unit is configured to summarize the sub-processing results to obtain a target processing result corresponding to the target task.

15. The device according to claim 14, wherein: The target task determination unit comprises: a conversion subunit, configured to convert the user's natural language input into a natural language text; an intention recognition subunit configured to perform intention recognition on the natural language text to obtain an intention recognition result including a clear result acquisition intention; The target task determination subunit is configured to determine the target task to be solved according to the intention recognition result; wherein the target task includes a task for obtaining a clearly existing result.

16. The device according to claim 15, wherein: The intention recognition subunit is further configured to: Performing semantic understanding on the natural language text to obtain a semantic understanding result; A preliminary intention is determined according to the semantic understanding result, and the preliminary intention is personalized modified using the personalized preference of the user to obtain an intention recognition result including a clear result acquisition intention.

17. The device according to claim 14, wherein: The task decomposition and corresponding issuing unit includes a task decomposition subunit configured to decompose the target task into a plurality of sub-target tasks, and the task decomposition subunit includes: A task element determination module is configured to determine a plurality of task elements constituting the target task; wherein different task elements correspond to different types of tasks; A clear result acquisition task decomposition module is configured to decompose clear result acquisition tasks for obtaining clear results from the target task according to a single task element specifically for determining a clear existing result matching the question raised by the user; The sub-goal task decomposition module is configured to decompose the target task into multiple sub-goal tasks that only contain a single task element; wherein the sub-goal tasks at least include the clear result acquisition type task.

18. The device according to claim 17, wherein: The sub-goal task decomposition module is further configured as follows: In response to the existence of additional conditions corresponding to the task elements, the target task is decomposed into multiple sub-target tasks each containing only a single task element and the corresponding additional conditions; wherein the additional conditions include: a specified task processing method and / or a specified result presentation form.

19. The apparatus according to claim 14, further comprising: A sub-agent construction unit configured to construct different sub-agents, the sub-agent construction unit comprising: A task type set determination subunit configured to determine a task type set for creating a sub-agent; The construction subunit is configured to construct a sub-agent with task processing capability of the corresponding type of task for each type of task in the task type set.

20. The device according to claim 19, wherein The construction subunit is further configured to: In response to the first target type task having at least two task processing methods, a sub-agent corresponding to each task processing method is constructed for the first target type task; wherein different task processing methods are abstracted from different processing logics for the same type of tasks.

21. The device according to claim 19, wherein The construction subunit is further configured to: In response to the second target type task having at least two result presentation forms, a sub-agent corresponding to each result presentation form is constructed for the second target type task; wherein the result presentation form includes: at least one of: text type, table type, image type, video type, and interactive card type.

22. The apparatus of claim 14, further comprising: A serial order determination unit is configured to determine the execution order between different sub-goal tasks with dependency relationships as serial; A parallel order determination unit is configured to determine the execution order between different sub-goal tasks without dependency relationship as parallel; Correspondingly, the sub-goal agent control output unit is further configured as follows: Controlling each sub-goal agent executed in the serial manner to sequentially output sub-processing results corresponding to the corresponding sub-goal tasks in combination with the personalized preferences; wherein the sub-processing results output by the previous sub-goal agent executed in the serial manner will serve as another part of the input of the subsequent sub-goal agent; Each sub-goal agent executed in the parallel manner is controlled to output a sub-processing result corresponding to the corresponding sub-goal task in combination with the personalized preference.

23. The apparatus according to claim 22, further comprising: The pause output control unit is configured to control the currently executed sub-goal agent to pause outputting the corresponding sub-processing result in response to the instruction input box being in a selected state; wherein the instruction input box is used for the user to input instructions, and the instruction input box is in an unselected state during the process of the sub-goal agent outputting the corresponding sub-processing result; The continue output control unit is configured to control the currently executed sub-target agent to continue outputting the corresponding sub-processing result in response to the instruction input box not generating new instructions during the process of restoring from the selected state to the unselected state.

24. The apparatus according to claim 23, further comprising: a correction information extraction unit, configured to extract correction information from a new instruction in response to the instruction input box generating a new instruction in the process of restoring from the selected state to the unselected state; An affected sub-target agent determining unit, configured to determine the sub-target agent affected by the correction information; The re-output control unit is configured to control the affected sub-target agent to re-output the corresponding sub-processing result in combination with the correction information.

25. The device according to any one of claims 14 to 24, wherein: The sub-processing result summary unit includes: The reverse splicing subunit is configured to splice each of the sub-processing results in a reverse manner according to the decomposition method of decomposing each of the sub-target tasks from the target task to obtain a reverse splicing result; The optimization processing subunit is configured to remove repeated redundant parts in the reverse splicing result, adjust the expression style of the reverse splicing result, and obtain a target processing result corresponding to the target task.

26. The device according to claim 25, wherein The optimization processing subunit includes a presentation style adjustment module configured to adjust the presentation style of the reverse splicing result, and the presentation style adjustment module is further configured to: The expression style of the reverse splicing result is adjusted to the expression style required by the personalized preference.

27. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method for processing clear result acquisition tasks based on multi-agent collaboration as described in any one of claims 1-13.

28. A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the method for processing tasks of obtaining clear results based on multi-agent collaboration as described in any one of claims 1-13.

29. A computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the method for processing tasks of explicit result acquisition based on multi-agent collaboration according to any one of claims 1 to 13.

Images