Multi-agent collaborative interaction method and device, storage medium and electronic equipment

Abstract

The invention provides a multi-agent collaborative interaction method and device, a computer storage medium and electronic equipment, and relates to the technical field of artificial intelligence. The method comprises the following steps: carrying out relevancy judgment on received to-be-processed information and multiple pieces of interactive context information; in response to the condition that the relevancy between the to-be-processed information and the target interaction context information is greater than a relevancy threshold value, labeling a target theme identifier which is the same as the target interaction context information for the to-be-processed information; and sending a first task processing request for the to-be-processed information to a second agent in the plurality of agents, so that a large model of the second agent obtains target interaction context information according to a target theme identifier contained in the first task processing request, and processes the to-be-processed information based on the target interaction context information. According to the method, the processing information amount of the large model can be reduced, the system performance pressure is relieved, and part of interaction information is prevented from being discarded, so that the output quality of the large model and the intelligent agent scheduling performance are improved.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to a multi-agent collaborative interaction method, a multi-agent collaborative interaction device, a computer storage medium, and an electronic device. Background Art

[0002] With the rapid development of artificial intelligence (AI), large AI models (abbreviated as "large models") have demonstrated promising results in handling complex tasks. Consequently, agent-based AI has made significant progress in practical applications, gradually evolving towards intelligent agents based on multiple large models. For example, multi-agent collaborative systems based on large language models (LLMs) enable multiple agents to collaboratively solve various complex tasks, achieving perception, learning, reasoning, and coordinated action.

[0003] Currently, in multi-agent collaborative interactions, a unified memory approach is often used. This means that the context information provided to the LLM during each interaction includes all the memory data. Alternatively, within the context window of the larger model, only the most recent memory data within the context window is provided.

[0004] However, as the amount of interactive information increases, the first solution significantly increases the amount of information to be processed and the performance pressure on the LLM model. For the second solution, when the data volume exceeds the context window size of the large model, some interactive information is easily discarded, seriously affecting the model's output quality and agent scheduling performance. Summary of the Invention

[0005] The present disclosure provides a multi-agent collaborative interaction method, a multi-agent collaborative interaction device, a computer storage medium and an electronic device, which can reduce the amount of information processed by a large model, thereby alleviating the system performance pressure during the multi-agent collaborative interaction process, and avoiding the discarding of some interaction information, thereby improving the output quality of the large model and the agent scheduling performance.

[0006] In a first aspect, an embodiment of the present disclosure provides a collaborative interaction method for multiple agents, which is applied to a first agent among multiple agents, and the method includes: receiving information to be processed, and judging the relevance of the information to be processed with multiple pre-stored interaction context information based on a large model of the first agent; wherein each interaction context information in the multiple interaction context information corresponds to a first topic identifier; in response to the relevance of the information to be processed with the target interaction context information in the multiple interaction context information being greater than a relevance threshold, marking the information to be processed with a target topic identifier that is the same as the target interaction context information; and sending a first task processing request for the information to be processed to a second agent among the multiple agents, so that the large model of the second agent obtains the target interaction context information based on the target topic identifier contained in the first task processing request, and processes the information to be processed based on the target interaction context information.

[0007] In the second aspect, an embodiment of the present disclosure provides a multi-agent collaborative interaction device, which is applied to a first agent among multiple agents, and the device includes: a relevance judgment module, which is used to receive information to be processed, and judge the relevance of the information to be processed with multiple pre-stored interaction context information based on the big model of the first agent; wherein each interaction context information in the multiple interaction context information corresponds to a first topic identifier; a topic identifier labeling module, which is used to label the information to be processed with a target topic identifier that is the same as the target interaction context information in response to the relevance between the information to be processed and the target interaction context information in the multiple interaction context information being greater than a relevance threshold; a task processing module, which is used to send a first task processing request for the information to be processed to a second agent among the multiple agents, so that the big model of the second agent obtains the target interaction context information based on the target topic identifier contained in the first task processing request, and processes the information to be processed based on the target interaction context information.

[0008] In a third aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which implements the above-mentioned multi-agent collaborative interaction method when executed by a processor.

[0009] In a fourth aspect, an embodiment of the present disclosure provides an electronic device, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the above-mentioned multi-agent collaborative interaction method by executing the executable instructions.

[0010] In a fifth aspect, an embodiment of the present disclosure provides a computer program product, including a computer program, which is executed by a processor to implement the above multi-agent collaborative interaction method.

[0011] The technical solution disclosed in this disclosure has the following beneficial effects:

[0012] The above-mentioned multi-agent collaborative interaction method receives information to be processed, and judges the relevance of the information to be processed with multiple pre-stored interaction context information based on the big model of the first agent; wherein each interaction context information in the multiple interaction context information corresponds to a first topic identifier; in response to the relevance of the information to be processed and the target interaction context information in the multiple interaction context information being greater than a relevance threshold, the target topic identifier that is the same as the target interaction context information is marked on the information to be processed; a first task processing request for the information to be processed is sent to the second agent among the multiple agents, so that the big model of the second agent obtains the target interaction context information based on the target topic identifier contained in the first task processing request, and processes the information to be processed based on the target interaction context information.

[0013] On the one hand, by adding a corresponding topic identifier to each interactive context information for distinction, it is convenient for the first agent to compare the correlation between the information to be processed and the multiple pre-stored interactive context information, so that only the target interactive context information corresponding to the information to be processed with a similarity greater than the correlation threshold needs to be sent to the second agent that processes the above-mentioned task to be processed, thereby avoiding the technical problem that the related technical solution needs to send all interactive context information to the large model of the second agent for processing, resulting in a large amount of model processing information, which in turn increases the large model context processing burden and system pressure, affects the large model processing efficiency, and thus affects the user experience, thereby achieving the technical effect of reducing the large model context processing burden and system pressure, and improving the large model processing efficiency. On the other hand, this method distinguishes each interactive context information by a topic identifier, realizes efficient management of existing interactive context information, avoids the problem of information discarded due to large model window limitations in related technical solutions, thereby ensuring the output quality of the large model and the scheduling performance of the agent.

[0014] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification, serve to explain the principles of the present disclosure. Obviously, the drawings described below are only some embodiments of the present disclosure, and it is possible for a person skilled in the art to derive other drawings based on these drawings without inventive effort.

[0016] Figure 1 A schematic diagram schematically illustrates a multi-agent collaborative interaction method of a related technical solution in this exemplary embodiment;

[0017] Figure 2 A schematic diagram schematically illustrates an architecture of a multi-agent collaborative interaction system in this exemplary embodiment;

[0018] Figure 3 A flowchart schematically illustrates a multi-agent collaborative interaction method in this exemplary embodiment;

[0019] Figure 4 A schematic diagram schematically illustrates a complete multi-agent collaborative interaction process in this exemplary embodiment;

[0020] Figure 5 A schematic diagram schematically illustrates another complete multi-agent collaborative interaction process in this exemplary embodiment;

[0021] Figure 6 Schematically shows a schematic structural diagram of a multi-agent collaborative interaction device in this exemplary embodiment;

[0022] Figure 7 The following schematically shows a structural diagram of an electronic device in this exemplary embodiment. DETAILED DESCRIPTION

[0023] The exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, the exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete and the concepts of the exemplary embodiments will be fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or that other methods, components, devices, steps, etc. may be employed. In other cases, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0024] In addition, the accompanying drawings are merely schematic illustrations of the present disclosure and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0025] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all steps. For example, some steps may be decomposed, while some steps may be combined or partially combined, so the actual execution order may change according to actual circumstances.

[0026] First, some nouns or terms that appear in the description of the embodiments of the present application are subject to the following interpretations:

[0027] Large models / AI large models: These refer to a class of AI models with a large number of parameters, constructed using artificial neural networks. They are typically pre-trained on massive amounts of data through self-supervised or semi-supervised learning, and their performance and capabilities are further optimized through methods such as instruction fine-tuning and human alignment. Large models are characterized by large numbers of parameters, extensive training data, and extensive computing resources. They are capable of solving general tasks, following human instructions, and performing complex reasoning. Major categories of large models include large language models, large vision models, large multimodal models, and large basic science models.

[0028] Large Language Model (LLM): A deep learning-based artificial intelligence system that can understand, generate, and process human natural language by training massive amounts of text data (typically containing billions to trillions of parameters). It has the ability to generate coherent, context-relevant text in open domains.

[0029] Intelligent agent system: A computer system based on artificial intelligence (AI) technology, in which the AI ​​agent relies on a large model as the core and combines tools to call the knowledge base to complete complex tasks.

[0030] Agent: An entity with autonomous decision-making capabilities that can achieve goals by perceiving the environment, understanding task instructions (including natural language), executing actions, and collaborating with other agents.

[0031] With the rapid development of artificial intelligence technology, large artificial intelligence models (abbreviated as "large models") have demonstrated good results in handling complex tasks. Corresponding agent-based artificial intelligence has made significant progress in practical applications. In particular, single-agent systems are often unable to effectively cope with changing environments in complex interactive scenarios. Therefore, they are gradually developing in the direction of collaborative interaction between agents of multiple large models. Multi-agent collaborative systems based on large models enable multiple agents to collaboratively solve various complex tasks to achieve perception, learning, reasoning, and collaborative action. For example, in the field of natural language processing, large language models (LLMs) can be applied to business acceptance, customer service, and other fields to improve customer service perception. At the same time, in the above application fields, due to the complexity of the business, multiple agents are often required to interact collaboratively to provide corresponding customer service.

[0032] Currently, in multi-agent collaborative interactions, two approaches typically employ a unified memory approach, whereby the context information provided to the macro model during each multi-agent interaction includes all the memory data. Alternatively, within the constraints of the macro model's context window, only the most recent memory data within the context window is provided.

[0033] Figure 1 The present disclosure refers to the schematic diagram of the multi-agent collaborative interaction method shown in the first related technical solution above, and Figure 1 As shown, the diagram includes multiple agents, namely a main agent and N collaborative agents. When the main agent receives input information to be processed (for example, business requirement information input by a customer), the main agent obtains all stored interaction context information from the memory module, namely message1 to messageN, and provides it to the main agent or at least one collaborative agent's macro model, so that the macro model generates a model output result corresponding to the information to be processed based on message1 to messageN, and sends it to the customer for review.

[0034] However, in the first method mentioned above, the main agent sends all interaction context information to the large model. For complex businesses, this contains a huge amount of data and a large amount of invalid information. This not only greatly increases the amount of information to be processed by the large model and the performance pressure of the multi-agent system, but also causes a waste of system resources and reduces the task / business processing efficiency of the large model. Continuing to take the application fields of business acceptance and customer service as an example, in the current marketing, acceptance and service processing context, due to the increasing complexity of customers' pending business, there are also discontinuities in the business processing process. The amount of content information data of the relevant business processing interactions may be very large. When multi-agent collaboration is used to handle business, the amount of information to be processed by the LLM model and the performance pressure of the multi-agent system are greatly increased.

[0035] In the second method mentioned above, when the amount of memory data exceeds the size of the context window of the LLM model, some information will be discarded, which can seriously affect the quality of the output results of the LLM model and the performance of multi-agent coordinated scheduling.

[0036] Taking the above problems into consideration, the exemplary embodiment of the present disclosure proposes a collaborative interaction method for multiple agents. On the one hand, the method distinguishes each interaction context information by adding a corresponding topic identifier, which facilitates the subsequent first agent to compare the correlation between the information to be processed and the multiple interaction context information stored in advance, so that only the target interaction context information corresponding to the information to be processed whose similarity is greater than the correlation threshold needs to be sent to the second agent that processes the above-mentioned task to be processed, thereby avoiding the technical problem that all interaction context information needs to be sent to the large model of the second agent for processing in the related technical solution, resulting in a large amount of model processing information, which in turn increases the large model context processing burden and system pressure, affects the large model processing efficiency, and thus affects the user experience. The technical effect of reducing the large model context processing burden and system pressure and improving the large model processing efficiency is achieved. On the other hand, the method distinguishes each interaction context information by a topic identifier, realizes efficient management of existing interaction context information, avoids the problem of information discarded due to large model window limitations in the related technical solution, and thus ensures the output quality of the large model and the scheduling performance of the agent.

[0037] The embodiments of the present disclosure provide a method and device for collaborative interaction of multiple intelligent agents, which can be applied to Figure 2 In the system architecture of the exemplary application environment shown.

[0038] like Figure 2As shown, system architecture 200 may include one or more of terminal devices 201, 202, 203, and 204, a network 205, and a server 206. Network 205 is a medium for providing communication links between terminal devices 201, 202, 203, and 204 and server 206. Network 205 may include various connection types, such as wired or wireless communication links or fiber optic cables. Terminal devices 201, 202, 203, and 204 may be, for example, but are not limited to, smartphones, personal digital assistants (PDAs), laptops, servers, desktop computers, or any other computing devices with network capabilities.

[0039] It should be understood that Figure 2 The number of terminal devices, networks, and servers in the example is merely illustrative. Any number of terminal devices, networks, and servers may be provided as needed. For example, server 206 may be an independent physical server, or a server cluster or distributed system consisting of multiple physical servers. It may also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0040] The multi-agent collaborative interaction method provided by the embodiment of the present disclosure can be executed on the server 206, and accordingly, the multi-agent collaborative interaction device is generally set in the server 206. The multi-agent collaborative interaction method provided by the embodiment of the present disclosure can also be executed in the terminal device, and accordingly, the multi-agent collaborative interaction device can also be set in the terminal device. The multi-agent collaborative interaction method provided by the embodiment of the present disclosure can also be partially executed in the server 206 and partially executed in the terminal device, and accordingly, some modules of the multi-agent collaborative interaction device can be set in the server 206, and some modules can be set in the terminal device.

[0041] In an optional embodiment, for example, based on any one or more of the terminal devices 201, 202, 203, or 204 running an application capable of supporting multi-agent collaborative interaction, such as an application providing business acceptance or customer service, the server 206 can provide corresponding multi-agent collaborative interaction services for the aforementioned application. A user can enter information to be processed on the user interaction interface of the terminal device. The server 206 receives the information to be processed and, based on the macro model of the first agent, determines the relevance of the information to be processed with multiple pre-stored interaction context information. Each of the multiple interaction context information corresponds to a first topic identifier. In response to the relevance of the information to be processed with target interaction context information in the multiple interaction context information being greater than a relevance threshold, the information to be processed is labeled with the same target topic identifier as the target interaction context information. A first task processing request for the information to be processed is sent to a second agent among the multiple agents, so that the macro model of the second agent obtains the target interaction context information based on the target topic identifier contained in the first task processing request and processes the information to be processed based on the target interaction context information.

[0042] After understanding the above system structure, the method provided by the exemplary embodiment of the present application is described below in combination with the application scenario described above and with reference to the accompanying drawings. It should be noted that the above application scenario is only shown to facilitate understanding of the spirit and principles of the present application, and the implementation of the present application is not limited in this respect. Figure 3 A flowchart schematically illustrates a method for collaborative interaction among multiple agents in this exemplary embodiment, see Figure 3 As shown, the specific implementation process of the method includes the following steps S301 to S303:

[0043] Step S301: Receive information to be processed, and determine the relevance of the information to be processed with a plurality of pre-stored interaction context information based on a large model of the first agent; wherein each of the plurality of interaction context information corresponds to a first topic identifier.

[0044] Step S302: In response to the relevance between the information to be processed and the target interaction context information among the multiple interaction context information being greater than a relevance threshold, the information to be processed is marked with the same target topic identifier as the target interaction context information.

[0045] Step S303: Send a first task processing request for the information to be processed to the second agent among the multiple agents, so that the large model of the second agent obtains the target interaction context information based on the target topic identifier contained in the first task processing request, and processes the information to be processed based on the target interaction context information.

[0046] exist Figure 3 In the provided technical solution, on the one hand, by adding a corresponding topic identifier to each interactive context information for distinction, it is convenient for the first agent to compare the correlation between the information to be processed and the multiple pre-stored interactive context information, so that only the target interactive context information corresponding to the information to be processed whose similarity is greater than the correlation threshold needs to be sent to the second agent that processes the above-mentioned task to be processed, thereby avoiding the technical problem that all interactive context information needs to be sent to the large model of the second agent for processing in the related technical solution, resulting in a large amount of model processing information, which in turn increases the burden of large model context processing and system pressure, affects the efficiency of large model processing, and affects the user experience, thereby achieving the technical effect of reducing the burden of large model context processing and system pressure, and improving the efficiency of large model processing. On the other hand, this method distinguishes each interactive context information by a topic identifier, realizes efficient management of existing interactive context information, avoids the problem of information discarded due to large model window limitations in the related technical solution, thereby ensuring the output quality of the large model and the scheduling performance of the agent.

[0047] The following will be combined with specific embodiments to Figure 3 The specific implementation of each step in the embodiment shown is described in detail:

[0048] In step S301, information to be processed is received, and the relevance between the information to be processed and a plurality of pre-stored interaction context information is determined based on the large model of the first agent; wherein each of the plurality of interaction context information corresponds to a first topic identifier.

[0049] In a master-slave multi-agent system, at least one master agent and at least one collaborative agent are included. The first agent can be a master agent among multiple agents, which can implement task scheduling. Accordingly, the master agent receives the information to be processed and performs task scheduling to assign it to the LLM model of the corresponding collaborative agent for processing. Alternatively, the first agent can also be any one of the multiple agents. Accordingly, any agent in the multi-agent system receives the information to be processed and receives the information to be processed based on the macro model of the above-mentioned agent.

[0050] In another multi-agent system, it only includes at least one main agent and 0 collaborative agents. For example, for tasks to be processed with low task complexity (task complexity is less than the complexity threshold), the main agent can directly process the above-mentioned information to be processed.

[0051] Each of the multiple agents corresponds to a large model. The large model is determined based on the technical field and needs of the current application. For example, the large model can be a large language model, a large visual model, a large multimodal model, or a large basic science model. The embodiments of this disclosure do not impose any specific restrictions on this. For example, in the application fields of business acceptance and customer service, the large model in this case is a large language model.

[0052] For example, when the large model of the first agent among multiple agents receives information to be processed, the relevance of the processed information and multiple pre-stored interaction context information can be determined.

[0053] The pre-stored interaction context information is generated by the first agent after receiving and processing the pending information in the past. For example, a user inputs information, a response is generated based on the large model, and the user continues to interact based on the response, forming the interaction context information.

[0054] Exemplarily, multiple interaction context information can be pre-stored in the memory module, which provides interaction context information for multi-agent collaboration and provides interaction context information related to the topic identifier provided by the main agent to the relevant parties of each multi-agent collaboration.

[0055] The memory module manages both short-term memory (primarily including interaction context information that can be incorporated into the context window of the current large model) and long-term memory (interaction context information that needs to be incorporated into the storage of external modules). Each piece of interaction context information recorded in the memory module corresponds to a topic identifier.

[0056] For example, the N pieces of interaction context information contained in the memory module can be represented by message1 through messageN. When the first agent's large model receives a message to be processed, it can evaluate the relevance of the message to be processed with the N pieces of interaction context information in the memory module. Furthermore, each piece of interaction context information is assigned a unique topic identifier to distinguish it from other pieces of interaction context information. For example, the topic identifiers corresponding to message1, message2, ..., messageN can be topic1, topic2, ..., topicn, respectively.

[0057] Accordingly, the memory module may store multiple pieces of interaction context information in advance and store the mapping relationship in the following manner:

[0058] Topic 1: message1;

[0059] Topic 2: message2;

[0060] …

[0061] Topic n: messageN.

[0062] It is understandable that other storage methods may also be used, and it is sufficient to express a one-to-one correspondence between the topic identifier and the interaction context information. The embodiments of the present disclosure do not impose any special restrictions on this.

[0063] The subject identifier in the above embodiment is an identifier mainly used to mark the category of related information, and its specific types include but are not limited to:

[0064] 1) Random character string (characters here include all character types supported by the computer system, including but not limited to English letters, numbers, Chinese characters, etc.).

[0065] 2) Strings or numbers with a specific structure, such as hyphenated, hierarchical strings, such as:

[0066] - "Business Consulting-Prices";

[0067] - "CRM-Business Processing-Broadband Business-New Installation".

[0068] When performing the step of receiving the information to be processed in S301, the information to be processed may be obtained based on a combination of one or more of the following embodiments:

[0069] In an optional embodiment, the first information to be processed may be obtained as input based on a large model of the first agent.

[0070] And / or, the large model of the first agent determines a task to be executed in a task execution plan according to a pre-established task execution plan, and obtains a second pending message. The task execution plan includes multiple tasks to be executed in a preset execution order, and the task to be executed is one of the multiple tasks.

[0071] And / or, the large model of the first agent receives feedback information sent by other agents among the multiple agents except the first agent to obtain third information to be processed.

[0072] The first pending message is a user-input request message, i.e., the first agent invokes a request message from a user outside the agent system. The second pending message is the next task plan information required by the first agent to execute according to a pre-defined task execution plan. The third pending message is feedback information received by the first agent from other agents in the multi-agent system.

[0073] For example, for the second pending message, the task execution plan typically includes at least one task to be executed. In this embodiment, the task execution plan includes multiple tasks to be executed in a preset execution order, such as Task 1 -> Task 2 -> Task 3. Assuming that after Task 1 is completed in collaborative agent a, the main agent b will start the execution task for the next task plan information, i.e., Task 2, in the execution order. At this time, the second pending message received by the large model of the first agent is the Task 2 plan information.

[0074] Exemplarily, the information to be processed received based on the large model of the first intelligent agent may be one of the first information to be processed, the second message to be processed, and the third message to be processed, or a combination of any of the above multiple messages to be processed. For example, the information to be processed received may be the first information to be processed and the second message to be processed.

[0075] It should be explained that, for the second message to be processed, the pre-formulated task execution plan can be a planned task pre-formulated by the first intelligent agent (for example, the main intelligent agent), or it can be a planned task pre-formulated by other intelligent agents (for example, collaborative intelligent agents) and sent to the first intelligent agent. The embodiments of the present disclosure do not impose any special restrictions on this.

[0076] Continuing with reference to step S301, after the large model based on the first agent receives the information to be processed, the relevance of the information to be processed and the multiple pre-stored interaction context information can be determined.

[0077] For example, after the first agent receives the information to be processed, it can determine the relevance of the information to be processed with a plurality of pre-stored interaction context information based on the large model of the first agent.

[0078] In an optional embodiment of the present disclosure, multiple pre-stored initial interaction context information are obtained, and the multiple initial interaction context information are grouped according to application scenario categories to obtain multiple initial information groups; based on the application scenario category of the information to be processed, a target information group is determined from the multiple initial information groups; based on the semantic analysis algorithm pre-configured in the large model of the first intelligent agent, semantic analysis is performed on each interaction context information in the information to be processed and the target information group to obtain a semantic analysis result; based on the semantic analysis result, a relevance judgment is made on the information to be processed and each interaction context information in the target information group.

[0079] The application scenario category is the scenario category information involved. For example, the application scenario category can be determined by performing keyword extraction, semantic analysis, or other methods on the information to be processed. For example, if the user inputs the information to be processed as "What's the weather like today?", the application scenario category of the information to be processed can be determined to be weather.

[0080] Furthermore, the Large Language Model (LLM) collects a vast amount (tens of billions) of text from the internet. By reading and analyzing the text, the LLM can identify the usage and meaning of words and sentences, as well as the relationships between them, thereby continuously improving language accuracy and semantic understanding. The pre-configured semantic analysis algorithm in the large model can be the current semantic analysis process of the large language model or a semantic analysis method developed in the future. The embodiments of this disclosure do not impose any special restrictions on this.

[0081] For example, to reduce computational complexity and improve the real-time processing of pending information by large models, embodiments of the present disclosure first group multiple initial interaction context information based on the application scenario categories they relate to, thereby generating multiple initial information groups. It is understood that a single piece of initial interaction context information can relate to one or more application scenario categories, and embodiments of the present disclosure do not impose any particular limitations on this.

[0082] After obtaining multiple initial information groups, the application scenario categories of the information to be processed can be compared with the application scenario categories corresponding to the multiple initial information groups to determine the initial information group with the same or similar application scenario category as the information to be processed as the target information group. Then, using the semantic analysis algorithm pre-configured in the large model of the first agent, a semantic analysis is performed on the interaction context information in the information to be processed and the target information group. Based on the semantic analysis results, the relevance of the interaction context information in the information to be processed and the target information group is determined.

[0083] In addition to the above embodiments, keywords between the information to be processed and multiple interactive context information may be determined, and relevance may be determined by keyword proportion or other methods. The embodiments of the present disclosure do not impose any special restrictions on this.

[0084] In step S302 , in response to the relevance between the information to be processed and the target interaction context information among the plurality of interaction context information being greater than a relevance threshold, the information to be processed is marked with the same target topic identifier as the target interaction context information.

[0085] Among them, the correlation threshold is a reference value used to detect the high / low correlation between the information to be processed and each interactive context information in multiple interactive context information. It can be a fixed value pre-configured by R&D personnel according to actual needs, or it can be a dynamically adjusted value determined based on an artificial intelligence algorithm or other methods. The embodiments of the present disclosure do not impose any special restrictions on this.

[0086] For example, when the correlation between the pending information and the target interaction context information exceeds a correlation threshold, the pending information can be labeled with the same target topic identifier as the target interaction context information. For example, if the memory module contains N interaction context information, message1 through messageN, and the correlation between the pending information and message1 exceeds the correlation threshold, the target topic identifier added to the pending information is topic1.

[0087] The following will explain in detail the process of adding subject identifiers for different types of information to be processed.

[0088] In an optional embodiment of the present disclosure, when the information to be processed received by the large model based on the first intelligent agent includes at least the third information to be processed, the feedback information sent by other intelligent agents also includes the third topic identifier and the interaction context information corresponding to the third topic identifier, so that the large model of the second intelligent agent processes the information to be processed based on the interaction context information corresponding to the third topic identifier.

[0089] The other agents are any one or more agents other than the first agent in the multi-agent system. Taking the first agent as the main agent as an example, the other agents are any one or more collaborative agents.

[0090] For example, when the information to be processed received by the large model of the first intelligent agent includes the third information to be processed, and the feedback information sent by other intelligent agents carries the subject identifier of the task to be processed (i.e., the third subject identifier), there is no need to re-label the subject identifier of the task to be processed (i.e., the third information to be processed), and the existing third subject identifier can be directly continued to be used.

[0091] Through this embodiment, there is no need to re-label the tasks to be processed with existing subject identifiers, thereby simplifying the processing flow of the large model, reducing redundant steps, and further improving the task processing efficiency of the large model.

[0092] Furthermore, when the information to be processed received by the large model of the first intelligent agent comes from multiple sources, in order to ensure the business processing logic and avoid confusion problems in multiple business processing processes, it can be processed based on the hierarchical structure relationship.

[0093] In an optional embodiment of the present disclosure, in the above step S301, the information to be processed received based on the big model of the first intelligent agent includes the first information to be processed and the second information to be processed, and the hierarchical structure relationship between the first information to be processed and the second information to be processed is determined; according to the hierarchical structure relationship, the first information to be processed and the second information to be processed are respectively assigned to the fourth intelligent agent matching the number of levels, so that the big model of the fourth intelligent agent processes the first information to be processed and the second information to be processed.

[0094] The hierarchical structure relationship includes at least the number of levels of the first information to be processed and the second information to be processed. For example, if the first information to be processed is at the first level and the second information to be processed is at the second level, it can be expressed as follows: the subject identifier of the first information to be processed - the subject identifier of the second processing message.

[0095] Exemplarily, the first information to be processed and the second information to be processed are stored in a hierarchical structure, so that the subject identifier and interaction context information determined for the first information to be processed can be isolated and separated from the subject identifier and interaction context information determined for the second information to be processed. After determining the hierarchical structure relationship between the first information to be processed and the second information to be processed, the first information to be processed is assigned to the intelligent agent corresponding to its level number for processing according to the level number of the first information to be processed; correspondingly, the second information to be processed is assigned to the intelligent agent corresponding to its level number for processing according to the level number of the second information to be processed.

[0096] For example, assume that the first agent receives pending message 1 at the first level, pending messages 2 and 3 at the second level, and pending messages 4 and 5 at the third level. Message 1 can be assigned to the collaborative agent 1 corresponding to the first level for processing, messages 2 and 3 can be assigned to the collaborative agent 2 corresponding to the second level for processing, and messages 4 and 5 can be assigned to the collaborative agent 3 corresponding to the third level for processing.

[0097] In step S303, a first task processing request for the information to be processed is sent to a second agent among multiple agents, so that the large model of the second agent obtains the target interaction context information based on the target topic identifier included in the first task processing request, and processes the information to be processed based on the target interaction context information.

[0098] The second agent can be the same agent as the first agent in a multi-agent system, or it can be a different agent from the first agent in a multi-agent system. For example, the first agent is the master agent, and the second agent is the collaborative agent. The second agent is used to represent the agent that performs the processing of the aforementioned pending task.

[0099] For example, after executing S302 to label the information to be processed with the same target topic identifier as the target interaction context information, the first agent can send a first task processing request for the information to be processed to the second agent among the multiple agents, so that the large model of the second agent can obtain the target interaction context information corresponding to the target topic identifier from the memory module according to the target topic identifier contained in the first task processing request, and thus process the information to be processed according to the target interaction context information.

[0100] Through this embodiment, it is only necessary to send the target interaction context information with a high relevance to the information to be processed from multiple pre-stored interaction context information to the second intelligent agent that processes the task, and this process only requires memory and subsequent retrieval through the corresponding topic identifier, which greatly reduces the amount of data processed by the large model and improves the retrieval efficiency and large model processing efficiency.

[0101] Furthermore, in another optional embodiment, there is a situation where the relevance between the information to be processed and the target interaction context information in the multiple interaction context information is not greater than the relevance threshold. This can be achieved according to the following embodiment:

[0102] In another optional embodiment of the present disclosure, in response to the relevance of the information to be processed to multiple interactive context information being no greater than a relevance threshold, a second topic identifier is marked for the information to be processed; and a second task processing request for the information to be processed is sent to a third agent among the multiple agents, so that the large model of the third agent processes the information to be processed according to a preset process.

[0103] The second topic identifier is different from the first topic identifiers corresponding to the multiple pieces of interaction context information. Specifically, when the relevance between the information to be processed and the multiple pieces of interaction context information is not greater than a relevance threshold, that is, when the current memory module does not store any interaction context information related to the information to be processed, a new topic identifier may be assigned to the information to distinguish it from other received and processed information.

[0104] Among them, the preset process represents the normal process of the large model for processing information to be processed. For example, the large model first pre-processes the input information to be processed, such as text cleaning, word segmentation (breaking sentences into words or phrases), and converting it into a format that the large model can understand. Then, based on the Transformer neural network architecture, it analyzes the input text to understand its semantics and context. The large model searches for information related to the input in its huge training data set, and then generates a response based on the above information. The generated response will undergo post-processing such as adjusting the language style and proofreading grammatical errors to improve its accuracy, coherence and naturalness. The processed response information is sent back to the graphical user interface of the terminal device through the server for display.

[0105] For example, if the relevance between the information to be processed and multiple pieces of interaction context information is no greater than a relevance threshold, a new topic identifier (i.e., a second topic identifier) ​​that does not overlap with existing topic identifiers is generated and labeled for the information to be processed. Simultaneously, a third agent among the multiple agents sends a second task processing request for the information to be processed, causing the third agent's large model to process the information to be processed with the second topic identifier according to a preset process.

[0106] After the information to be processed whose relevance to multiple interaction context information is not greater than the relevance threshold is processed based on the above embodiment, a mapping relationship between the interaction context information generated for the information to be processed and the second topic identifier is constructed, and the mapping relationship between the interaction context information generated for the information to be processed and the second topic identifier is stored.

[0107] For example, its topic ID (i.e., the second topic identifier) ​​can be used as the classification ID of the interaction context information of the relevant conversation to store the corresponding relationship in the memory module. At the same time, the interaction context information can be memorized and retrieved based on this ID, so that in the subsequent conversation with the relevant collaborative intelligent agent, the second topic identifier and the corresponding interaction context information can be directly queried from the memory module.

[0108] The following will refer to Figure 4 、 Figure 5 The entire multi-agent collaborative interaction process of the multi-agent collaborative interaction method according to the exemplary embodiment of the present disclosure is described in detail.

[0109] In one embodiment, among the plurality of interaction context information pre-stored in the memory module, there is target interaction context information whose relevance to the information to be processed is higher than a relevance threshold.

[0110] Figure 4 Schematic diagram showing a complete multi-agent collaborative interaction process in this exemplary embodiment; Figure 4 As shown, the multi-agent system includes a main agent and multiple collaborative agents. When the main agent's large model receives information to be processed, it determines the relevance of the information to be processed with multiple interactive context information pre-stored in the memory module; wherein each of the multiple interactive context information in the memory module corresponds to a first topic identifier; that is, referring to Figure 4 , the memory module contains N interaction context information, namely message1 to messageN, and the N interaction context information corresponds to a topic identifier respectively, and is stored in the memory module according to the corresponding relationship. For example, the memory module stores:

[0111] Topic 1: message1

[0112] Topic 2: message2

[0113] …

[0114] Topic n: messageN

[0115] When the correlation between Message1 in the information to be processed and message1 in the memory module is greater than the correlation threshold, and the correlation between Message2 in the information to be processed and message2 in the memory module is greater than the correlation threshold, the subject identifier of Message1 in the message to be processed is marked as the same subject identifier as message1, that is, subject 1; at the same time, the subject identifier of Message2 in the message to be processed is marked as the same subject identifier as message2, that is, subject 2.

[0116] Message1 in the message to be processed is sent to collaborative agent 1, so that collaborative agent 1 obtains message1 corresponding to topic 1 from the memory module and processes Message1 in the message to be processed based on message1.

[0117] Message2 in the message to be processed is sent to the collaborative agent N, so that the collaborative agent N obtains message2 corresponding to topic 2 from the memory module and processes Message2 in the message to be processed based on message2.

[0118] After collaborative agent 1 completes processing of Message 1 in the pending message, it sends the processing result to the master agent, and the processing result carries the subject identifier of Topic 1. It should be noted that during the above-mentioned collaborative agent 1 processing of Message 1 in the pending message, the master agent and collaborative agent 1 may need to interact multiple times, and the subject identifier (i.e., Topic 1) needs to be carried in each interaction.

[0119] Similarly, after collaborative agent N completes processing of Message 2 in the pending message, it sends the processing result to the master agent, and the processing result carries the subject identifier of Subject 2. It should be noted that during the above-mentioned collaborative agent N's processing of Message 2 in the pending message, the master agent and collaborative agent N may need to interact multiple times, and the subject identifier (i.e., Subject 2) needs to be carried in each interaction.

[0120] Finally, the main intelligent agent determines whether the target task for the information to be processed is completed, and merges the processing results for Message1 and Message2 and sends them to the terminal device to display the final processing results to the user through the terminal device.

[0121] Furthermore, in another embodiment, the target interaction context information having a relevance to the information to be processed higher than a relevance threshold does not exist in the plurality of interaction context information pre-stored in the memory module.

[0122] Figure 5 Schematic diagram showing another complete multi-agent collaborative interaction process in this exemplary embodiment; Figure 5 As shown, the multi-agent system includes a main agent, multiple collaborative agents, and preset agents. When the main agent's large model receives information to be processed, it determines the relevance of the information to be processed with multiple interactive context information pre-stored in the memory module; wherein each of the multiple interactive context information in the memory module corresponds to a first topic identifier; that is, referring to Figure 5 , the memory module contains N interaction context information, namely message1 to messageN, and the N interaction context information corresponds to a topic identifier respectively, and is stored in the memory module according to the corresponding relationship. For example, the memory module stores:

[0123] Topic 1: message1

[0124] Topic 2: message2

[0125] …

[0126] Topic n: messageN

[0127] When the correlation between the information to be processed Message and the multiple interactive context information message1 to messageN pre-stored in the memory module is not greater than (that is, less than or equal to) the correlation threshold, the main intelligent agent labels the information to be processed Message with an existing topic identifier different from the existing topic identifier stored in the memory module, for example, labeling it as topic n+1.

[0128] Then, the topic n+1 and the message to be processed are assigned to the default agent (i.e., the third agent) for processing. If the message to be processed is successfully processed by the default agent, the topic n+1 and the interaction context message N+1 for the message to be processed are memorized and stored in the memory module according to the mapping relationship for subsequent retrieval.

[0129] Finally, the main intelligent agent determines whether the target task for the information to be processed is completed, and sends it to the terminal device to display the final processing results to the user through the terminal device.

[0130] In order to implement the above-mentioned multi-agent collaborative interaction method, a multi-agent collaborative interaction device is provided in one embodiment of the present disclosure. Figure 6 The schematic diagram of the architecture of a multi-agent collaborative interaction device is shown schematically.

[0131] The multi-agent collaborative interaction device 600 is applied to the first service agent of the first distributed network, and the multi-agent collaborative interaction device 600 includes: a relevance judgment module 601 , a topic identification marking module 602 and a task processing module 603 .

[0132] The above-mentioned relevance judgment module 601 is used to receive the information to be processed, and to judge the relevance of the information to be processed with multiple pre-stored interaction context information based on the big model of the first intelligent agent; wherein each interaction context information in the multiple interaction context information corresponds to a first topic identifier; the topic identifier labeling module 602 is used to label the information to be processed with the same target topic identifier as the target interaction context information in response to the relevance between the information to be processed and the target interaction context information in the multiple interaction context information being greater than the relevance threshold; the task processing module 603 is used to send a first task processing request for the information to be processed to the second intelligent agent among the multiple intelligent agents, so that the big model of the second intelligent agent obtains the target interaction context information based on the target topic identifier contained in the first task processing request, and processes the information to be processed based on the target interaction context information.

[0133] In an optional embodiment of the present disclosure, the device further includes a topic identifier labeling module, which is used to label the information to be processed with a second topic identifier in response to the relevance of the information to be processed to multiple interactive context information being not greater than a relevance threshold; wherein the second topic identifier is different from the first topic identifiers corresponding to the multiple interactive context information; the task processing module 603 is also used to send a second task processing request for the information to be processed to a third intelligent agent among the multiple intelligent agents, so that the large model of the third intelligent agent processes the information to be processed according to a preset process.

[0134] In an optional embodiment of the present disclosure, the device further includes a mapping relationship construction module, which is used to construct a mapping relationship between the interactive context information generated for the information to be processed and the second topic identifier, and store the mapping relationship between the interactive context information generated for the information to be processed and the second topic identifier.

[0135] In an optional embodiment of the present disclosure, the apparatus further comprises an information receiving module, the information receiving module being configured to obtain input first information to be processed based on the macro model of the first agent;

[0136] and / or, the large model of the first agent determines, based on a pre-established task execution plan, a task to be executed in the task execution plan, and obtains second information to be processed, wherein the task execution plan includes a plurality of tasks to be executed in a preset execution order, and the task to be executed is one of the plurality of tasks;

[0137] And / or, the large model of the first agent receives feedback information sent by other agents among the multiple agents except the first agent to obtain third information to be processed.

[0138] In an optional embodiment of the present disclosure, the information to be processed received by the big model based on the first intelligent agent includes at least the third information to be processed, and the feedback information sent by other intelligent agents includes the third topic identifier and the interaction context information corresponding to the third topic identifier, so that the big model of the second intelligent agent processes the information to be processed based on the interaction context information corresponding to the third topic identifier.

[0139] In an optional embodiment of the present disclosure, the information to be processed received by the large model based on the first intelligent agent includes first information to be processed and second information to be processed. The device also includes a hierarchical structure determination module, which is used to determine the hierarchical structure relationship between the first information to be processed and the second information to be processed, and the hierarchical structure relationship at least includes the number of levels of the first information to be processed and the second information to be processed; the task processing module 603 is also used to allocate the first information to be processed and the second information to be processed to a fourth intelligent agent matching the number of levels according to the hierarchical structure relationship, so that the large model of the fourth intelligent agent processes the first information to be processed and the second information to be processed.

[0140] In an optional embodiment of the present disclosure, the relevance judgment module 601 is specifically used to obtain multiple pre-stored initial interaction context information, and group the multiple initial interaction context information according to application scenario categories to obtain multiple initial information groups; determine the target information group from the multiple initial information groups based on the application scenario category of the information to be processed; perform semantic analysis on each interaction context information in the information to be processed and the target information group based on the semantic analysis algorithm pre-configured in the large model of the first intelligent agent to obtain a semantic analysis result; and perform relevance judgment on each interaction context information in the information to be processed and the target information group based on the semantic analysis result.

[0141] The multi-agent collaborative interaction device 600 provided in the embodiment of the present disclosure can execute the technical solution of the multi-agent collaborative interaction method in any of the above-mentioned embodiments. Its implementation principle and beneficial effects are similar to the implementation principle and beneficial effects of the multi-agent collaborative interaction method. Please refer to the implementation principle and beneficial effects of the multi-agent collaborative interaction method, and no further details will be given here.

[0142] In exemplary embodiments of the present disclosure, a computer-readable storage medium is also provided, on which is stored a program product capable of implementing the methods described above. In some possible implementations, various aspects of the present invention may also be implemented in the form of a program product comprising program code that, when executed on a terminal device, causes the terminal device to execute the steps according to various exemplary embodiments of the present invention described in the "Exemplary Methods" section above.

[0143] According to an embodiment of the present invention, a program product for implementing the above-mentioned method can be a portable compact disc read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto. In this document, a readable storage medium can be any tangible medium containing or storing a program, and the program can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0144] The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable 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 disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0145] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0146] The program code contained on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, radio frequency (RF), etc., or any suitable combination of the foregoing.

[0147] The program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, and the like, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a stand-alone software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0148] In an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided.

[0149] Those skilled in the art will appreciate that various aspects of the present invention may be implemented as systems, methods, or program products. Therefore, various aspects of the present invention may be implemented in the following forms: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, which may be collectively referred to herein as "circuits," "modules," or "systems."

[0150] Refer to the following Figure 7 An electronic device 700 according to this embodiment of the present invention will be described. Figure 7 The electronic device 700 shown is merely an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0151] like Figure 7 As shown, electronic device 700 is implemented as a general-purpose computing device. Components of electronic device 700 may include, but are not limited to, the aforementioned at least one processing unit 710, the aforementioned at least one storage unit 720, a bus 730 connecting various system components (including storage unit 720 and processing unit 710), and a display unit 740.

[0152] The storage unit stores program codes, which can be executed by the processing unit 710, so that the processing unit 710 performs the steps according to various exemplary embodiments of the present invention described in the "Exemplary Method" section above. For example, the processing unit 710 can perform the following steps: Figure 3 Steps 301 to 303 shown in FIG.

[0153] The storage unit 720 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 7201 and / or a cache memory unit 7202 , and may further include a read-only memory unit (ROM) 7203 .

[0154] The storage unit 720 may also include a program / utility 7204 having a set (at least one) of program modules 7205, such program modules 7205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0155] Bus 730 may represent one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0156] The electronic device 700 can also communicate with one or more external devices 1000 (e.g., a keyboard, a pointing device, a Bluetooth device, etc.), one or more devices that enable a user to interact with the electronic device 700, and / or any device that enables the electronic device 700 to communicate with one or more other computing devices (e.g., a router, a modem, etc.). This communication can occur via an input / output (I / O) interface 750. Furthermore, the electronic device 700 can communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network such as the Internet) via a network adapter 760. As shown, the network adapter 760 communicates with other modules of the electronic device 700 via a bus 730. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the electronic device 700, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, Redundant Arrays of Independent Disks (RAID) systems, tape drives, and data backup storage systems.

[0157] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiments of the present disclosure.

[0158] Furthermore, the above-described figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention and are not intended to be limiting. It is readily understood that the processes illustrated in the above-described figures do not indicate or limit the temporal order of these processes. Furthermore, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0159] It should be noted that although several modules or units of the device for action execution are mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0160] Other embodiments of the present disclosure will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow from the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the claims.

[0161] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A multi-agent collaborative interaction method, characterized in that: Applied to a first agent of a plurality of agents, the method comprises: receiving information to be processed, and determining relevance between the information to be processed and a plurality of pre-stored interaction context information based on the macro model of the first agent; wherein each of the plurality of interaction context information corresponds to a first topic identifier; In response to a relevance between the information to be processed and target interaction context information among the plurality of interaction context information being greater than a relevance threshold, marking the information to be processed with a target topic identifier that is the same as that of the target interaction context information; A first task processing request for the information to be processed is sent to a second agent among the multiple agents, so that the large model of the second agent obtains the target interaction context information based on the target topic identifier included in the first task processing request, and processes the information to be processed based on the target interaction context information.

2. The method according to claim 1, characterized in that The method further comprises: In response to the relevance between the information to be processed and the plurality of interaction context information being not greater than the relevance threshold, marking the information to be processed with a second topic identifier; wherein the second topic identifier is different from the first topic identifiers corresponding to the plurality of interaction context information respectively; A second task processing request for the information to be processed is sent to a third agent among the multiple agents, so that the large model of the third agent processes the information to be processed according to a preset process.

3. The method according to claim 2, characterized in that After the large model of the third agent processes the information to be processed according to a preset process, the method further includes: A mapping relationship between the interaction context information generated for the information to be processed and the second topic identifier is constructed, and the mapping relationship between the interaction context information generated for the information to be processed and the second topic identifier is stored.

4. The method according to claim 1, wherein The receiving of information to be processed includes: Based on the large model of the first intelligent agent, obtaining input first information to be processed; and / or, the large model of the first agent determines, based on a pre-established task execution plan, a task to be executed in the task execution plan, to obtain second information to be processed, wherein the task execution plan includes a plurality of tasks to be executed in a preset execution order, and the task to be executed is one of the plurality of tasks; And / or, the large model of the first agent receives feedback information sent by other agents among the multiple agents except the first agent to obtain third information to be processed.

5. The method according to claim 4, characterized in that The information to be processed received by the large model based on the first intelligent agent includes at least the third information to be processed, and the feedback information sent by the other intelligent agents includes the third topic identifier and the interaction context information corresponding to the third topic identifier, so that the large model of the second intelligent agent processes the information to be processed based on the interaction context information corresponding to the third topic identifier.

6. The method according to claim 4, characterized in that The information to be processed received based on the large model of the first agent includes the first information to be processed and the second information to be processed, and the method further includes: Determining a hierarchical structure relationship between the first information to be processed and the second information to be processed, wherein the hierarchical structure relationship includes at least the number of levels of the first information to be processed and the second information to be processed; According to the hierarchical structure relationship, the first information to be processed and the second information to be processed are respectively assigned to a fourth agent matching the corresponding number of levels, so that the large model of the fourth agent processes the first information to be processed and the second information to be processed.

7. The method according to claim 1, characterized in that The determining of the relevance between the information to be processed and a plurality of pre-stored interaction context information includes: Acquire multiple pre-stored initial interaction context information, and group the multiple initial interaction context information according to application scenario categories to obtain multiple initial information groups; determining a target information group from the multiple initial information groups according to an application scenario category of the information to be processed; Based on the semantic analysis algorithm pre-configured in the large model of the first agent, semantic analysis is performed on each interaction context information in the information to be processed and the target information group to obtain a semantic analysis result; According to the semantic analysis result, a relevance judgment is performed between the information to be processed and each interactive context information in the target information group.

8. A multi-agent collaborative interaction device, characterized in that: Applied to a first agent of a plurality of agents, the apparatus comprises: a relevance determination module, configured to receive information to be processed and, based on the macro model of the first agent, determine the relevance of the information to be processed with a plurality of pre-stored interaction context information; wherein each of the plurality of interaction context information corresponds to a first topic identifier; a topic identifier tagging module configured to tag the information to be processed with the same target topic identifier as the target interaction context information in response to a relevance between the information to be processed and target interaction context information among the plurality of interaction context information being greater than a relevance threshold; A task processing module is used to send a first task processing request for the information to be processed to a second agent among the multiple agents, so that the large model of the second agent obtains the target interaction context information based on the target topic identifier included in the first task processing request, and processes the information to be processed based on the target interaction context information.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the multi-agent collaborative interaction method according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: include: processor; as well as a memory for storing executable instructions of the processor; Wherein, the processor is configured to execute the multi-agent collaborative interaction method described in any one of claims 1 to 7 by executing the executable instructions.

Images