A multi-agent and artificial collaboration method, system, device and medium

Abstract

The present application relates to the technical field of multi-agent and artificial collaboration, and particularly relates to a multi-agent and artificial collaboration method, system, device and medium. The method comprises: performing intent recognition and confidence evaluation on a user message based on a pre-constructed sub-agent registry; determining a processing path of the user message according to the results of the intent recognition and the confidence evaluation, wherein the processing path comprises a sub-agent processing path, a knowledge base query path and a human handling path; when it is determined to enter the human handling path, generating a structured human handling instruction containing a human handling identifier and a complete context summary, and accessing an artificial conversation according to the structured human handling instruction; after the artificial conversation is accessed, continuously monitoring subsequent dialogue messages in the artificial conversation, and generating auxiliary information for the subsequent dialogue messages in real time. A unified central scheduling layer is constructed, so that complex and cross-domain user messages can be accurately routed to a specialized sub-agent processing path.

Description

Technical Field

[0001] This invention relates to the field of multi-agent and human collaboration technology, and in particular to a method, system, device and medium for multi-agent and human collaboration. Background Technology

[0002] The following are some common technical solutions used in current service scenarios: First, traditional AI customer service: This primarily involves directly processing user messages through an LLM (Large Model), with the LLM then providing a response. Second, traditional multi-agent systems: These systems integrate multiple business agent entry points within a single interface. Each agent typically has an independent entry point and can handle different tasks. Third, conventional AI-to-human conversion solutions: This is currently the most common approach. Upon receiving a user message, it is directly forwarded to the LLM for processing, followed by a response. If the LLM cannot process the message, it guides the user to send a keyword for transferring to a human agent; the system captures this keyword and then transfers the message to a human. Fourth, human customer service: This is the traditional customer service system implementation, where incoming user messages are directly transferred to a human agent for processing.

[0003] In summary, the existing technology typically suffers from the following technical defects: Traditional AI customer service struggles to handle complex, cross-domain, and multi-round intents, often resulting in misanswers or blunt responses. Traditional multi-agent systems lack a unified central control, making it difficult for agents to connect on demand; agents typically have independent entry points and can only complete a single task. In conventional AI-to-human transfer solutions, the AI ​​completely leaves the system after the transfer, creating a communication gap between AI and human customer service. Human agents, upon joining, still need to understand the context from scratch, further exacerbating the communication gap. Human customer service lacks real-time, intelligent, and directly applicable auxiliary suggestions, frequently handling repetitive conversations with high average processing times. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a method, system, device and medium for multi-agent and human collaboration, so as to solve the technical defects of traditional AI customer service in that it is difficult to handle complex cross-domain issues, business intelligent agents are difficult to access on demand, and multiple intelligent agents are difficult to manage in a unified manner. It can achieve seamless switching between human and AI, and there is no message gap when human access is available. Moreover, human can adopt or label AI opinions to supervise and fine-tune the model.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a method for multi-agent and artificial collaboration, comprising: Receive user messages; Intent recognition and confidence assessment of user messages are performed based on a pre-built sub-agent registry. Based on the results of intent recognition and confidence assessment, the processing path of the user message is determined. The processing path includes the sub-agent processing path, the knowledge base query path, and the path to manual processing. When it is determined that the process will proceed to the human intervention stage, a structured human intervention instruction containing a human intervention identifier and a complete context summary is generated, and the human intervention session is accessed based on the structured human intervention instruction. After a live agent session is initiated, the system continuously monitors subsequent conversation messages and generates supplementary information for those messages in real time, which is then pushed to the live agent's interface.

[0006] Based on the above technical solution, the present invention can be further improved as follows.

[0007] Furthermore, the sub-agent registry includes the identity identifiers and capability descriptions of registered sub-agents.

[0008] Furthermore, the pre-built sub-agent registry supports dynamic updates at runtime, and the update process includes: Receive a sub-agent registration request, the registration request including the identity identifier, capability description, invocation identifier, and invocation parameter description of the sub-agent to be registered; Based on the registration request, the sub-agent to be registered is dynamically added to the sub-agent registry.

[0009] Furthermore, determining the processing path for user messages based on the results of intent recognition and confidence assessment includes: The user intent is determined based on the intent recognition results, and then compared with the capability descriptions of registered sub-intelligent agents in the sub-intelligent agent registry. If the comparison is successful and the confidence assessment result is greater than the preset threshold, then the target sub-agent is determined to be matched and the sub-agent processing path is entered. If the comparison fails or the confidence assessment result is less than the preset threshold, it is determined that no target sub-agent has been matched, and the query path of the knowledge base is entered. If no matching content is found through the knowledge base query path, or if the user message contains preset emergency handling content, then the process will proceed to the manual handling path.

[0010] Furthermore, in the sub-agent processing path and the knowledge base query path, the processing result or knowledge retrieval result of the target sub-agent is used as a new message layer, and intent recognition and response processing are performed again to obtain the final output result.

[0011] Furthermore, the structured-to-manual instructions include: Conversation identifiers and emotion tags; The conversation identifier, emotion tag, human intervention identifier, and complete context summary are sent as parameters to the agent's seat corresponding to the human conversation.

[0012] Furthermore, the auxiliary information includes suggested response text, user core demands and emotion extraction results, compliance risk warnings, similar case recommendations, and policy and regulatory provisions; The human operator interface receives operation instructions for auxiliary information, including acceptance instructions, modification instructions, or rejection instructions. Record the operation behavior data corresponding to the operation instructions, and use the operation behavior data as supervision data to train the model that generates auxiliary information.

[0013] Secondly, the present invention provides a multi-agent and artificial collaborative system, comprising: The message receiving module is used to receive user messages; The central agent is used to perform intent recognition and confidence assessment on user messages based on a pre-built sub-agent registry, which includes the identity identifiers and capability descriptions of registered sub-agents; it is used to determine the processing path of user messages based on the results of intent recognition and confidence assessment, which includes a sub-agent processing path, a knowledge base query path, and a path to manual processing. The central output routing processing module is used to generate a structured human-to-human transfer instruction containing a human-to-human transfer identifier and a complete context summary when the path to human-to-human transfer is determined, and to access the human-to-human conversation based on the structured human-to-human transfer instruction; after accessing the human-to-human conversation, it is used to continuously monitor subsequent dialogue messages in the human-to-human conversation, and generate auxiliary information for subsequent dialogue messages in real time, and push the auxiliary information to the human agent's operation interface.

[0014] Thirdly, the present invention provides a computer device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of a multi-agent and human-assisted collaborative method.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the multi-agent and human-assisted collaborative method. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a multi-agent and human-assisted collaborative method. Figure 2 This is a schematic diagram of a multi-agent and human collaborative system. Figure 3 This is a schematic diagram illustrating the specific interaction process of a multi-agent and human collaboration method. Figure 4 This is a schematic diagram of a human agent interface for a multi-agent and human collaboration method. Figure 5 This is a schematic diagram of the electronic device of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be described below with reference to the accompanying drawings. In the description of the present invention, unless otherwise stated, " / " indicates that the objects before and after are in an "or" relationship. For example, A / B can represent A or B. The "or" in the present invention is merely a description of the relationship between the related objects, indicating that three relationships can exist. For example, A or B can represent: A alone, A and B simultaneously, and B alone. A and B can be singular or plural. Furthermore, in the description of the present invention, unless otherwise stated, "multiple" refers to two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items.

[0018] Furthermore, to facilitate a clear description of the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" are not necessarily different.

[0019] In this embodiment of the invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in this embodiment of the invention should not be construed as superior or more advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner for ease of understanding.

[0020] Example 1, such as Figure 1 As shown in the figure, an embodiment of the present invention provides a method for multi-agent and artificial collaboration, including: S1: Receive user messages; S2: Perform intent recognition and confidence assessment on user messages based on a pre-built sub-agent registry; S3: Based on the results of intent recognition and confidence assessment, determine the processing path of the user message, which includes the sub-agent processing path, the knowledge base query path, and the path to manual processing. The knowledge base is used for knowledge retrieval using user message keywords. In this application embodiment, it uses vector databases or RAG retrieval techniques, or it can be a shared database query for company business.

[0021] S4: When it is determined that the process will enter the human intervention path, a structured human intervention instruction containing a human intervention identifier and a complete context summary will be generated, and the human intervention session will be accessed according to the structured human intervention instruction. S5: After a human agent session is initiated, continuously monitor subsequent dialogue messages in the human agent session and generate auxiliary information for subsequent dialogue messages in real time, and push the auxiliary information to the human agent's operation interface.

[0022] It should be noted that, addressing the shortcomings of existing technologies where a single large model struggles to handle complex cross-domain intentions and where multiple agents lack a unified central control and have isolated entry points, a pre-built sub-agent registry and intent recognition and confidence assessment mechanism are introduced through S2 and S3. This mechanism dynamically compares the intent of user messages with the capability descriptions in the registry to determine the processing path. This technical solution constructs a unified central scheduling layer, breaking down the barriers of traditional independent entry points for multiple agents. This allows complex, cross-domain user messages to be accurately routed to specialized sub-agent processing paths, solving the technical problem of business agents being difficult to access and manage on demand.

[0023] Furthermore, addressing the shortcomings of existing technologies that rely on simple keyword capture for AI-assisted human intervention and require humans to understand the context from scratch after the AI ​​leaves the interface, a structured transfer mechanism is proposed in S4. When the path to human intervention is determined, the session is not directly terminated; instead, a structured human intervention instruction containing a human intervention identifier and a complete context summary is generated. This technical solution structurally encapsulates key information such as the session state and core requests processed by the AI ​​and transmits it with the instruction, eliminating the message gap between AI and humans at the technical level and achieving seamless switching and contextual integration when humans access the interface.

[0024] Furthermore, addressing the shortcomings of existing technologies such as the lack of real-time assistance suggestions from human customer service and the ineffectiveness of AI-driven transfers in conventional solutions, S5 breaks the existing paradigm of AI leaving the field once a human is connected. After a human conversation begins, it continuously monitors subsequent dialogue messages and generates auxiliary information in real time, pushing it to the human agent's interface, achieving continuous on-loop and deep collaboration of AI. This technical solution directly provides real-time, operable intelligent assistance to human customer service to reduce repetitive work. Through the interface-based push of auxiliary information, it provides the necessary interactive foundation for human agents to perform tagging operations, establishes a high-quality monitoring data collection link, and achieves the data closed-loop goal of monitoring and fine-tuning the model.

[0025] Example 2 is the second embodiment of the present invention.

[0026] First, it provides a multi-agent and human collaborative system, such as Figure 2 As shown, it includes: Message receiving module, sub-agent management module, central agent, central output routing processing module; The message receiving module is used to receive user messages; The sub-agent management module is used to receive dynamic registration requests from sub-agents, build and maintain a sub-agent registry in real time, and the registry includes the identity identifier, capability description, call identifier and call parameter description of the registered sub-agents. The central agent is used to perform intent recognition and confidence assessment on user messages based on a pre-built sub-agent registry, which includes the identity identifiers and capability descriptions of registered sub-agents; it is used to determine the processing path of user messages based on the results of intent recognition and confidence assessment, which includes a sub-agent processing path, a knowledge base query path, and a path to manual processing. The central output routing processing module is used to generate a structured human-to-human transfer instruction containing a human-to-human transfer identifier and a complete context summary when the path to human-to-human transfer is determined, and to access the human-to-human conversation based on the structured human-to-human transfer instruction; after accessing the human-to-human conversation, it is used to continuously monitor subsequent dialogue messages in the human-to-human conversation, and generate auxiliary information for subsequent dialogue messages in real time, and push the auxiliary information to the human agent's operation interface.

[0027] In this embodiment, the system, through the coordinated operation of various functional modules, aims to implement the technical solution of the aforementioned method embodiment. Its purpose is to accurately process complex cross-domain intentions through unified central scheduling, achieve seamless human-machine switching and context transmission using structured instructions, and provide continuous on-loop real-time assistance after human intervention, thereby constructing an efficient human-machine deep collaborative closed loop. The specific implementation is as follows: First, at the system architecture level, this system adopts a star-shaped collaborative architecture, comprising a central intelligent agent as the control core and a collection of sub-intelligent agents specializing in different domain business. Each sub-intelligent agent maintains its structured capability description (such as a JSON description). To achieve flexible expansion, each sub-intelligent agent supports dynamic access (i.e., registering new sub-intelligent agents at runtime via API without restarting the system).

[0028] Based on a dynamic access mechanism, the system constructs and maintains a sub-agent registry (containing a unique agent ID and a structured capability description in JSON format) in real time through a sub-agent management module. This registry supports dynamic updates at runtime. The update process includes: receiving a registration request from a sub-agent (this request includes, but is not limited to, fields such as the sub-agent's identity identifier ENTRY_ID, capability description ENTRY_DESC, call identifier ENTRY_KEY, call parameter description ENTRY_ARGS, and output parameter description ENTRY_OUTPUT); and dynamically adding the sub-agent to be registered to the sub-agent registry according to the registration request.

[0029] Based on the above system architecture, such as Figure 3 As shown, when a user initiates an interaction, the system's message processing flow is as follows: The message entry point into the system is the message receiving module, which executes step S1 to receive user messages. After the message enters, it queries the user's current session state and historical session context based on the user ID, and then passes the user message, session state, and session context to the central intelligent agent for processing.

[0030] It should be noted that, based on the Large Language Model (LLM) as the central decision-making agent, after receiving user messages, it processes all user messages and executes steps S2 and S3 (implemented through the Large Language Model prompt word engineering). The specific implementation method is as follows: It should be noted that, based on the results of intent recognition and confidence assessment, the processing path for user messages is determined, including: determining the user intent based on the intent recognition results, comparing the user intent with the capability descriptions of registered sub-intelligent agents in the sub-intelligent agent registry; if the comparison is successful and the confidence assessment result is greater than a preset threshold, then a target sub-intelligent agent is determined to be matched, and the process enters the sub-intelligent agent processing path; if the comparison fails or the confidence assessment result is less than the preset threshold, then a target sub-intelligent agent is determined not to be matched, and the process enters the knowledge base query path; if no matching content is found through the knowledge base query path, or if the user message contains preset emergency processing content, then the process enters the manual processing path. The specific implementation method is as follows: Determine the intent of the input message and, in conjunction with the sub-agent registry information, determine if there is a sub-agent capable of processing the input message. If the confidence level is greater than the threshold and matches the capability of a certain sub-agent, output the sub-agent's processing request and the complete context (CONTEXT) as parameters.

[0031] If the identification fails to match the user message with the sub-agent's capabilities or the confidence assessment result is less than a preset threshold, a knowledge base query request will be output.

[0032] If the large model receives a knowledge base query request or a response from a sub-agent, it organizes the content and outputs it in a form that is easy for the user to understand. If the large model fails to match the sub-agent and the knowledge base cannot retrieve the content, or if the user inputs a pre-set emergency handling text, it outputs a manual transfer instruction and outputs the complete context and conversation summary as parameters.

[0033] It should be noted that after the central agent completes the intent analysis, it generates a structured routing control instruction. This instruction contains at least two types of fields: a route identifier field and a route parameter field. The central agent can distribute the task to the corresponding processing link by parsing the route identifier field; the route parameter field carries the execution data that matches the path.

[0034] In this embodiment, the specific field format of the above-mentioned structured routing control instruction can be implemented using the following mapping relationship: The routing identifier field may include, but is not limited to, the following specific identifier values: a first identifier for indicating a knowledge base query (such as KB_QUERY), a second identifier for indicating a transfer to human (such as TO_AGENT), a third identifier for indicating a sub-agent call (such as ENTERY_CALL), a fourth identifier for indicating a direct reply (such as ANWSER), and a fifth identifier for indicating assistance in a human session (such as SESSION_HELP).

[0035] The specific content of the routing parameter field is dynamically generated based on the value of the routing identifier. For example, when the routing identifier is a sub-agent call, the routing parameters are automatically generated from the corresponding call parameters in the sub-agent's registry. When the routing identifier is a manual or knowledge base query, the routing parameters are filled with the corresponding fixed parameters (such as session ID, context summary, etc.).

[0036] To enable those skilled in the art to better understand the specific form of the above-mentioned structured routing control instructions, the following example, taking the determination of the entry path into the sub-agent processing path, provides a sample of the output data (JSON format) of the central agent, including: { "ruote":"ENTRY_CALL", "args": { "ENTRY_KEY":"order_agent_001", "ENTRY_ARGS": { "userinput":"I would like to subscribe to an ice cream combo." }, "SESSION_ID":"00001", CONTEXT: [ { "role":"user", "text":"Hello" }, { "role":"agent", "text":"Hello, how can I help you?" }, { "role":"user", "text":"I'd like to order an ice cream combo." } ], } }

[0037] It should be noted that, as shown in the data example above, when the value of the route identifier field (route) is the sub-agent call identifier, the route parameter field carries the target sub-agent's call key (ENTRY_KEY), the specific user input (ENTRY_ARGS), and a context array (CONTEXT) containing historical dialogue role information. By parsing this route identifier, the central agent can accurately route the above parameters to the corresponding sub-agent.

[0038] Based on the routing decision results output by the central intelligent agent, the central output routing processing module enters the specific execution and coordination stage, and the specific implementation method is as follows: It should be noted that in the sub-agent processing path and the knowledge base query path, the processing result or knowledge retrieval result of the target sub-agent is used as a new message layer for further intent recognition and response processing to obtain the final output result. The specific implementation method is as follows: When performing a knowledge base query (vector database or RAG), the knowledge base query result is used as a new message layer and then used as a parameter to re-initiate the central agent processing request.

[0039] When a sub-agent is invoked (ENTERY_CALL), the sub-agent management module is called to initiate a sub-agent processing request. The session context, session summary, user message, and call parameters in the sub-agent description are used as input parameters. Then, the processing result of the sub-agent is used as a new message layer as a parameter to initiate the central agent processing request again.

[0040] When the routing decision determines that the process should proceed to human intervention, the transfer procedure in step S4 is executed. The parameters output by the central agent (including session ID, complete session context, session summary, and emotion tag) are compiled and sent to the human agent for processing.

[0041] After a human session is initiated, proceed to step S5. For example... Figure 4 As shown, after connecting to a human agent, the conversation processing interface still displays auxiliary information from the large model, indicating that humans and AI work together deeply to provide services.

[0042] It should be noted that after the connection is transferred to a human agent, the central AI agent and related sub-agents do not exit and continue to monitor subsequent conversations in real time. Various AI-assisted suggestions are generated and displayed in real time on the right side of the human agent's interface, including but not limited to: multiple suggested response texts, extraction of the user's core needs / emotions, compliance risk warnings, similar case recommendations, and citations of policies and regulations. Human customer service representatives can accept, modify, or reject instructions with a single click. Each modification, acceptance, or rejection of AI suggestions by the human agent is recorded and used as high-quality supervisory data for subsequent model fine-tuning.

[0043] It should be noted that the embodiments of this application solve several technical defects existing in the conventional technology, and have the following beneficial effects compared with the prior art: First, a unified central scheduling layer is built through a pre-built sub-agent registry and a dynamic registration mechanism (new sub-agents are registered at runtime via API without requiring a system restart), breaking down the barriers of independent entry points and single business operations inherent in traditional multi-agent systems. This mechanism enables the system to accurately route complex, cross-domain user intents to specialized sub-agents, while also supporting flexible access and unified management of business agents, reducing system maintenance costs and enhancing adaptability to business changes.

[0044] Secondly, addressing the drawback of traditional AI-to-human referral solutions where AI departures necessitate repeated contextual understanding by human agents, our technical solution generates structured instructions containing a human referral identifier and a complete context summary. This structured encapsulates key information such as the session state and core requests processed by the AI ​​and transmits it to the human agent. When a human agent connects, there's no need to re-question historical information, achieving seamless switching and contextual integration, significantly improving the efficiency and user experience of the human referral service.

[0045] Third, by continuously monitoring conversations during human interaction and generating auxiliary information in real time (such as suggested response text, user sentiment extraction, and compliance risk warnings), real-time and actionable intelligent support is provided to human agents, significantly reducing the processing time for repetitive conversations. Simultaneously, the human agents' "acceptance / modification / rejection" actions on the auxiliary information are recorded as supervisory data, used to continuously fine-tune the model that generates the auxiliary information, forming a positive cycle and improving the accuracy and adaptability of AI assistance.

[0046] Fourth, in the sub-agent processing path and the knowledge base query path, the sub-agent results or knowledge retrieval results are treated as a new message layer for further intent recognition and response processing. This mechanism ensures that the final output conforms to user understanding habits, improves the accuracy and naturalness of the response, and enhances user satisfaction with the service.

[0047] Example 3: In some embodiments, the multi-agent and human-cooperative system of the present invention can be implemented by a combination of hardware and software. As an example, the multi-agent and human-cooperative system of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the multi-agent and human-cooperative method of the present invention. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0048] The modules described in the embodiments of this invention can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.

[0049] An electronic device according to an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned multi-agent and artificial collaboration methods. That is, an electronic device according to an embodiment of the present invention may include, but is not limited to: a processor and a memory; the memory is used to store the computer program; the processor is used to execute a multi-agent and artificial collaboration method shown in any embodiment of the present invention by calling the computer program.

[0050] In one alternative embodiment, an electronic device is provided, such as Figure 5 As shown, Figure 5The illustrated electronic device 4000 includes a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present invention.

[0051] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0052] Bus 4002 may include a path for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 4002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus 4002 is represented by only one thick line, but this does not mean that there is only one bus or one type of bus.

[0053] The memory 4003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0054] The memory 4003 stores application code (computer program) for executing the present invention, and its execution is controlled by the processor 4001. The processor 4001 executes the application code stored in the memory 4003 to implement the content shown in the foregoing method embodiments.

[0055] Among them, electronic devices can also be terminal devices, which can be any device that can install applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.

[0056] It should be noted that, Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0057] An embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the above-mentioned multi-agent and human-assisted collaborative methods.

[0058] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0059] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the aforementioned multi-agent and human-assisted collaborative method.

[0060] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0061] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0062] The computer-readable storage medium provided in this invention can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EEPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0063] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the method shown in the above embodiments.

[0064] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0065] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0066] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, this invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0067] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for multi-agent collaboration with artificial intelligence, characterized in that, include: Receive user messages; Intent recognition and confidence assessment of user messages are performed based on a pre-built sub-agent registry. Based on the results of intent recognition and confidence assessment, the processing path of the user message is determined. The processing path includes the sub-agent processing path, the knowledge base query path, and the path to manual processing. When it is determined that the process will proceed to the human intervention stage, a structured human intervention instruction containing a human intervention identifier and a complete context summary is generated, and the human intervention session is accessed based on the structured human intervention instruction. After a live agent session is initiated, the system continuously monitors subsequent conversation messages and generates supplementary information for those messages in real time, which is then pushed to the live agent's interface.

2. The multi-agent and artificial collaboration method according to claim 1, characterized in that, in, The sub-agent registry includes the identity identifiers and capability descriptions of registered sub-agents.

3. The multi-agent and artificial collaboration method according to claim 2, characterized in that, The pre-built sub-agent registry supports dynamic updates at runtime, and the update process includes: Receive a sub-agent registration request, the registration request including the identity identifier, capability description, invocation identifier, and invocation parameter description of the sub-agent to be registered; Based on the registration request, the sub-agent to be registered is dynamically added to the sub-agent registry.

4. The multi-agent and artificial collaboration method according to claim 1, characterized in that, The step of determining the processing path of user messages based on the results of intent recognition and confidence assessment includes: The user intent is determined based on the intent recognition results, and then compared with the capability descriptions of registered sub-intelligent agents in the sub-intelligent agent registry. If the comparison is successful and the confidence assessment result is greater than the preset threshold, then the target sub-agent is determined to be matched and the sub-agent processing path is entered. If the comparison fails or the confidence assessment result is less than the preset threshold, it is determined that no target sub-agent has been matched, and the query path of the knowledge base is entered. If no matching content is found through the knowledge base query path, or if the user message contains preset emergency handling content, then the process will proceed to the manual handling path.

5. The multi-agent and artificial collaboration method according to claim 4, characterized in that, In the sub-agent processing path and the knowledge base query path, the processing result or knowledge retrieval result of the target sub-agent is used as a new message layer, and intent recognition and response processing are performed again to obtain the final output result.

6. The multi-agent and artificial collaboration method according to claim 1, characterized in that, The structured instructions converted to human input include: conversation identifiers and emotion tags; The session identifier, emotion tag, human intervention identifier, and complete context summary are sent as parameters to the human agent corresponding to the human session.

7. The multi-agent and artificial collaboration method according to claim 1, characterized in that, The auxiliary information includes suggested response text, user core demands and emotion extraction results, compliance risk warnings, similar case recommendations, and policy and regulatory provisions. The human operator interface receives operation instructions for auxiliary information, including acceptance instructions, modification instructions, or rejection instructions. Record the operation behavior data corresponding to the operation instructions, and use the operation behavior data as supervision data to train the model that generates auxiliary information.

8. A multi-agent and artificial collaborative system, employing the method described in any one of claims 1-7, characterized in that, include: The message receiving module is used to receive user messages; The central agent is used to perform intent recognition and confidence assessment on user messages based on a pre-built sub-agent registry, which includes the identity identifiers and capability descriptions of registered sub-agents; it is used to determine the processing path of user messages based on the results of intent recognition and confidence assessment, which includes a sub-agent processing path, a knowledge base query path, and a path to manual processing. The central output routing processing module is used to generate a structured human intervention instruction containing a human intervention identifier and a complete context summary when it is determined that the path to human intervention is to be entered; and to access the human intervention session according to the structured human intervention instruction. This is used to continuously monitor subsequent dialogue messages in a human agent's session after the session has started, and to generate auxiliary information for those messages in real time, which is then pushed to the human agent's interface.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images