Standard operation program scene-oriented multi-agent collaboration method supporting human-computer interaction

Abstract

The invention discloses a standard operation program scene-oriented multi-agent cooperation method supporting human-computer interaction, and the method comprises the steps: recognizing the intention of a user through a large model, determining an SOP task, dynamically matching a preset converted workflow workflow, automatically arranging, scheduling and driving the cooperative execution of a single agent in the workflow, and achieving the cooperative execution of the single agent in the workflow. The sharing of the structured information among the single agents is realized through the shared environment, and the preset single agent with the autonomous decision-making capability can efficiently obtain the structured information output by other agents from the shared environment in the execution process, so that the key execution steps of the SOP task are strictly followed; the controllability and the accuracy of the SOP execution process can be ensured; meanwhile, each single agent can utilize a large model to autonomously decide and call an external tool during execution, and human feedback optimization is utilized to improve an execution result, so that key execution links also have enough controllability while the single agents keep autonomy.

Description

Technical Field

[0001] The present application relates to the technical field of multi-agent collaborative architecture, and in particular to a multi-agent collaborative method that supports human-computer interaction in standard operating procedure scenarios. Background Art

[0002] At present, the architecture of multi-agent collaboration is mainly divided into: (1) Leader mode: a leader agent and collaborative agents with clear division of labor are defined, with only one agent acting as the leader. Depending on the architecture, the collaborative agents may only communicate with the leader agent or the leader may be elected through a shared dialogue among all agents; (2) Non-leader mode: only collaborative agents are defined, all collaborative agents have the same role and are part of the corresponding group chat discussion on the relevant task; communication between agents takes place in a shared environment, and each agent can see all messages from other agents; agents can complete certain tasks or call tools autonomously as needed, without the need for the leader agent to assign work.

[0003] For the leader mode, it is usually required that all communications between collaborative agents must go through the leader agent. On the one hand, it makes the leader agent more complex in managing communications, and there will be higher performance loss when there is more communication. On the other hand, it limits the autonomy of collaborative agents, and they cannot interact with other agents on demand, which is not conducive to expanding the capabilities of collaborative agents.

[0004] In the non-leader mode, agents spend most of their time issuing orders to each other and spend the rest of their time sharing information or requesting guidance. This dialogue and discussion mode may lead to unnecessary chats between agents, but will not promote the achievement of team goals and consumes more large-model computing resources, which is not economical in practical applications.

[0005] That is, the existing technology has the following technical problems: the common multi-agent collaborative architecture does not consider the characteristics of SOP-type tasks, which have clear and definite execution steps and processes, and have high requirements for the execution effect of key links. It does not fully consider the integration of human-computer interaction into the execution process, and does not take into account the autonomy and controllability of multi-agent collaboration. At the same time, it has high performance loss and is easy to consume more large model computing resources. Therefore, in response to the above problems, a multi-agent collaborative method that supports human-computer interaction for standard operating procedures (SOP) scenarios is proposed. Summary of the invention

[0006] In this embodiment, a multi-agent collaboration method that supports human-computer interaction for standard operating procedure scenarios is provided to solve the problem that the ordinary multi-agent collaboration architecture in the prior art has high performance loss and easily consumes more large model computing resources.

[0007] According to one aspect of the present application, a multi-agent collaboration method supporting human-computer interaction in a standard operating procedure scenario is provided, and the multi-agent collaboration method supporting human-computer interaction in a standard operating procedure scenario comprises the following steps:

[0008] (1) Implementing a human-computer interaction session manager based on a general lightweight network communication component;

[0009] (2) Setting up an intent-recognition routing agent;

[0010] (3) Setting up multiple single agents with autonomous decision-making capabilities;

[0011] (4) Establish a multi-agent shared environment;

[0012] (5) Send SOP tasks to the agent through a visual user client, and determine the SOP tasks to be executed through human-computer interaction;

[0013] (6) The workflow of the preset SOP task is searched according to the SOP, and the workflow executor loads the found workflow and automatically schedules and drives each single agent in the workflow to execute until the entire workflow is executed.

[0014] Furthermore, in the step (1), a human-computer interaction session manager is implemented based on a general lightweight network communication component, and the human-computer interaction session manager supports asynchronous communication with a visualized human user client.

[0015] Furthermore, the human-computer interaction session manager can establish a session channel with session-level isolation, and the session channel functionally supports accepting connection requests from human user clients and automatically establishing a session channel, and different sessions are isolated from each other.

[0016] Further, the conversation channel may support waiting for and receiving information sent by a human user client with which a conversation relationship has been established;

[0017] The session channel functionally supports sending information to a human user client with which a session relationship has been established.

[0018] Furthermore, in step (2), the intent recognition routing agent can extract the instruction intent input by the human user, and obtain the necessary structured information slots required to execute the intent through continuous interaction with humans.

[0019] Furthermore, in the step (3), a plurality of single agents with autonomous decision-making capabilities are set up, and the function of the single agents is to use the large model and autonomously with the help of external tools to execute the subtasks after the SOP task is disassembled.

[0020] Furthermore, the single agent can obtain the required structured information from the environment and publish potentially valuable information to the environment in a structured form;

[0021] The single agents collaborate using structured data.

[0022] Furthermore, the single agent can conduct human-computer interaction with a human user client through a human-computer interaction session manager, output the results of execution using the large model to the human user, and the human user can provide feedback on the execution results and participate in the self-reflection of the agent to improve the execution process.

[0023] Furthermore, in step (4), a multi-agent shared environment is established, and the establishment of the shared environment includes the following steps:

[0024] a. Different agents in a multi-agent system exchange information and collaborate through a shared environment during operation;

[0025] b. A single agent dynamically obtains structured information published by other agents through a shared environment and uses it as part of the input data;

[0026] c. Publish the output structured data with potential value to the environment for other intelligent agents to obtain and use.

[0027] Furthermore, in step (5), the human user issues SOP tasks to the multi-agents through a visual user client, and determines the SOP tasks to be executed through human-computer interaction;

[0028] a. The intent recognition routing agent first extracts the instruction intent input by the human user and obtains the necessary structured information slots required to execute the intent;

[0029] b. When the matching preset SOP task intent cannot be identified, the intent recognition process is looped, and feedback is given to the user to prompt that the input instruction intent is unclear. At the same time, the intent entity with the highest matching degree is recommended based on the user input for the user to confirm;

[0030] c. The intention recognition process is executed cyclically until the user's SOP task intention and the necessary structured information slots required to execute the SOP task are confirmed;

[0031] d. Confirm the user's SOP task intention;

[0032] e. Pass the matched SOP tasks and the slot information required to execute the SOP to the multi-agent Workflow executor through routing;

[0033] f. Trigger multiple agents to automatically execute SOP tasks.

[0034] Furthermore, in step (6), after determining the SOP task to be performed by the human user, the following steps are performed:

[0035] a. Find the preset workflow of the SOP task according to the SOP. The workflow is to convert the SOP into a directed acyclic graph to express the pipeline of the SOP task execution according to the specific execution steps of the SOP;

[0036] b. Enter the execution phase of SOP;

[0037] c. The workflow execution controller loads the workflow of the entire SOP and executes each single agent in a concurrent manner according to the topological sorting method until the entire workflow execution is completed.

[0038] Through the above-mentioned technical scheme of the present application, a set of simple, efficient, economical, and more controllable multi-agent collaboration scheme for the execution process and result accuracy of the SOP task scenario is constructed in view of the characteristics of the clear execution process and high accuracy requirements of the SOP task scenario. The present application converts the SOP task into a workflow, supports the autonomous decision-making of a single agent to call external tools, and introduces human-computer interaction links in key steps, and optimizes the self-reflection of the agent using human feedback to improve the execution results, so that the execution process of the multi-agent takes into account both autonomy and controllability. Furthermore, the experience is improved through the conversational task assignment method, and data exchange and collaboration are achieved by subscribing or publishing data from a shared environment in the form of structured data between multiple agents, without consuming large model computing power for information extraction, making it more economical, efficient and accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 The figure is a schematic diagram of the overall process of an embodiment of the present application. DETAILED DESCRIPTION

[0041] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0042] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0043] In the present application, the terms "upper", "lower", "left", "right", "front", "back", "top", "bottom", "inner", "outer", "middle", "vertical", "horizontal", "lateral", "longitudinal" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings. These terms are mainly used to better describe the present application and its embodiments, and are not used to limit the indicated devices, elements or components to have a specific orientation, or to be constructed and operated in a specific orientation.

[0044] In addition, some of the above terms may be used to express other meanings in addition to indicating orientation or positional relationship. For example, the term "on" may also be used to express a certain dependency or connection relationship in some cases. For those of ordinary skill in the art, the specific meanings of these terms in this application can be understood according to specific circumstances.

[0045] In addition, the terms "installed", "set", "provided with", "connected", "connected", and "socketed" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection, or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be an internal connection between two devices, elements, or components. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0046] See also Figure 1As shown, a multi-agent collaboration method supporting human-computer interaction for a standard operating procedure scenario includes the following steps:

[0047] (1) Implementing a human-computer interaction session manager based on a general lightweight network communication component;

[0048] (2) Setting up an intent-recognition routing agent;

[0049] (3) Setting up multiple single agents with autonomous decision-making capabilities;

[0050] (4) Establish a multi-agent shared environment;

[0051] (5) Send SOP tasks to the agent through a visual user client, and determine the SOP tasks to be executed through human-computer interaction;

[0052] (6) The workflow of the preset SOP task is searched according to the SOP, and the workflow executor loads the found workflow and automatically schedules and drives each single agent in the workflow to execute until the entire workflow is executed.

[0053] In the step (1), a human-computer interaction session manager is implemented based on a general lightweight network communication component, and the human-computer interaction session manager supports asynchronous communication with a visualized human user client;

[0054] The human-computer interaction session manager can establish a session-level isolated session channel, which functionally supports accepting a connection request from a human user client and automatically establishing a session channel, and different sessions are isolated from each other;

[0055] The session channel can support waiting for and receiving information sent by a human user client with which a session relationship has been established;

[0056] The session channel functionally supports sending information to a human user client with which a session relationship has been established.

[0057] In step (2), the intent recognition routing agent can extract the instruction intent input by the human user, and obtain the necessary structured information slots required to execute the intent through continuous interaction with humans.

[0058] In the step (3), a plurality of single agents with autonomous decision-making capabilities are set up, and the function of the single agent is to use the large model and autonomously use external tools to perform the subtasks after the SOP task is disassembled;

[0059] The single agent can obtain the required structured information from the environment and publish potentially valuable information to the environment in a structured form;

[0060] The single agent uses structured data for collaboration;

[0061] The single agent can perform human-computer interaction with a human user client through a human-computer interaction session manager, output the results of execution using the large model to the human user, and the human user can provide feedback on the execution results and participate in the self-reflection of the agent to improve the execution process.

[0062] In the step (4), a multi-agent shared environment is established, and the establishment of the shared environment includes the following steps:

[0063] a. Different agents in a multi-agent system exchange information and collaborate through a shared environment during operation;

[0064] b. A single agent dynamically obtains structured information published by other agents through a shared environment and uses it as part of the input data;

[0065] c. Publish the output structured data with potential value to the environment for other intelligent agents to obtain and use.

[0066] Through the above technical solution, the intelligent agent uses structured data to subscribe to the data it needs from the shared environment or publish the data that other intelligent agents may need to the shared environment, avoiding the process of each intelligent agent performing complex processing again from the original information generated by many intelligent agents to extract the required key data. This greatly reduces the consumption of large model tokens and also avoids the problem of incorrect extraction results caused by the hallucination problem when using the large model to extract structured data. Overall, this multi-agent collaboration method is more economical and efficient, and the accuracy of the information transmitted is very high.

[0067] In the step (5), the human user issues the SOP task to the multi-agent through a visual user client, and determines the SOP task to be executed through human-computer interaction;

[0068] a. The intent recognition routing agent first extracts the instruction intent input by the human user and obtains the necessary structured information slots required to execute the intent;

[0069] b. When the matching preset SOP task intent cannot be identified, the intent recognition process is looped, and feedback is given to the user to prompt that the input instruction intent is unclear. At the same time, the intent entity with the highest matching degree is recommended based on the user input for the user to confirm;

[0070] c. The intention recognition process is executed cyclically until the user's SOP task intention and the necessary structured information slots required to execute the SOP task are confirmed;

[0071] d. Confirm the user's SOP task intention;

[0072] e. Pass the matched SOP tasks and the slot information required to execute the SOP to the multi-agent Workflow executor through routing;

[0073] f. Trigger multiple agents to automatically execute SOP tasks.

[0074] In step (6), after determining the SOP task to be performed by the human user, the following steps are performed:

[0075] a. Find the preset workflow of the SOP task according to the SOP. The workflow is to convert the SOP into a directed acyclic graph to express the pipeline of the SOP task execution according to the specific execution steps of the SOP;

[0076] b. Enter the execution phase of SOP;

[0077] c. The workflow execution controller loads the workflow of the entire SOP and executes each single agent in a concurrent manner according to the topological sorting method until the entire workflow execution is completed.

[0078] Through this technical solution, a simple, efficient, economical, and more controllable multi-agent collaboration solution is constructed in view of the characteristics of SOP task scenarios with clear execution processes and high accuracy requirements. This application converts SOP tasks into workflows, supports single-agent autonomous decision-making to call external tools, and introduces human-computer interaction links in key steps, using human feedback to optimize the self-reflection of the agent to improve the execution results, so that the execution process of multiple agents takes into account both autonomy and controllability. Furthermore, a better experience is achieved through a conversational task assignment method, and data exchange and collaboration is achieved by subscribing or publishing data from a shared environment in the form of structured data between multiple agents, without consuming large model computing power for information extraction, making it more economical, efficient and accurate.

[0079] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A multi-agent collaboration method supporting human-computer interaction for standard operating procedure scenarios, characterized by: The multi-agent collaboration method supporting human-computer interaction for standard operating procedure scenarios comprises the following steps: (1) Implementing a human-computer interaction session manager based on a general lightweight network communication component; (2) Setting up an intent-recognition routing agent; (3) Setting up multiple single agents with autonomous decision-making capabilities; (4) Establish a multi-agent shared environment; (5) Send SOP tasks to the agent through a visual user client, and determine the SOP tasks to be executed through human-computer interaction; (6) The workflow of the preset SOP task is searched according to the SOP, and the workflow executor loads the found workflow and automatically schedules and drives each single agent in the workflow to execute until the entire workflow is executed.

2. According to claim 1, a multi-agent collaboration method supporting human-computer interaction for standard operating procedure scenarios, characterized in that: In the step (1), a human-computer interaction session manager is implemented based on a general lightweight network communication component, and the human-computer interaction session manager supports asynchronous communication with a visualized human user client.

3. The multi-agent collaboration method supporting human-computer interaction for standard operating procedure scenarios according to claim 2, characterized in that: The human-computer interaction session manager can establish a session channel with session-level isolation. The session channel functionally supports accepting connection requests from human user clients and automatically establishing a session channel, and different sessions are isolated from each other.

4. The multi-agent collaboration method supporting human-computer interaction for standard operating procedure scenarios according to claim 2, characterized in that: The session channel can support waiting for and receiving information sent by a human user client with which a session relationship has been established; The session channel functionally supports sending information to a human user client with which a session relationship has been established.

5. The multi-agent collaboration method supporting human-computer interaction for standard operating procedure scenarios according to claim 1, characterized in that: In step (2), the intent recognition routing agent can extract the instruction intent input by the human user, and obtain the necessary structured information slots required to execute the intent through continuous interaction with humans.

6. The multi-agent collaboration method supporting human-computer interaction for standard operating procedure scenarios according to claim 1, characterized in that: In the step (3), a plurality of single intelligent agents with autonomous decision-making capabilities are set up, and the function of the single intelligent agents is to use the large model and autonomously with the help of external tools to execute the subtasks after the SOP task is disassembled.

7. The multi-agent collaboration method supporting human-computer interaction for standard operating procedure scenarios according to claim 6, characterized in that: The single agent can obtain the required structured information from the environment and publish potentially valuable information to the environment in a structured form; the single agent uses structured data for collaboration; the single agent can conduct human-computer interaction with a human user client through a human-computer interaction session manager, output the results of execution using a large model to the human user, and the human user provides feedback on the execution results and participates in the self-reflection of the agent to improve the execution process.

8. The multi-agent collaboration method supporting human-computer interaction for standard operating procedure scenarios according to claim 1, characterized in that: In the step (4), a multi-agent shared environment is established, and the establishment of the shared environment includes the following steps: a. Different agents in a multi-agent system exchange information and collaborate through a shared environment during operation; b. A single agent dynamically obtains structured information published by other agents through a shared environment and uses it as part of the input data; c. Publish the output structured data with potential value to the environment for other intelligent agents to obtain and use.

9. The multi-agent collaboration method supporting human-computer interaction for standard operating procedure scenarios according to claim 1, characterized in that: In the step (5), the human user issues the SOP task to the multi-agent through a visual user client, and determines the SOP task to be executed through human-computer interaction; a. The intent recognition routing agent first extracts the instruction intent input by the human user and obtains the necessary structured information slots required to execute the intent; b. When the matching preset SOP task intent cannot be identified, the intent recognition process is looped, and feedback is given to the user to prompt that the input instruction intent is unclear. At the same time, the intent entity with the highest matching degree is recommended based on the user's input for the user to confirm; c. The intention recognition process is executed cyclically until the user's SOP task intention and the necessary structured information slots required to execute the SOP task are confirmed; d. Confirm the user's SOP task intention; e. Pass the matched SOP tasks and the slot information required to execute the SOP to the multi-agent Workflow executor through routing; f. Trigger multiple agents to automatically execute SOP tasks.

10. The multi-agent collaboration method supporting human-computer interaction for standard operating procedure scenarios according to claim 1, characterized in that: In step (6), after determining the SOP task to be performed by the human user, the following steps are performed: a. Find the preset workflow of the SOP task according to the SOP. The workflow is to convert the SOP into a directed acyclic graph to express the pipeline of the SOP task execution according to the specific execution steps of the SOP; b. Enter the execution phase of SOP; c. The workflow execution controller loads the workflow of the entire SOP and executes each single agent in a concurrent manner according to the topological sorting method until the entire workflow execution is completed.

Images