Multi-agent cooperative processing method, storage medium and program product

Abstract

Embodiments of the present application provide a multi-agent collaborative processing method, a storage medium and a program product, which first acquires first information input by a user and accurately determines a first agent for core processing, then filters one or more second agents for auxiliary processing based on a preset agent dependency graph, and then formulates a scientific and efficient task execution strategy and collaboratively calls the two types of agents to carry out processing work, and finally outputs second information, which can effectively break the limitations of single-agent serial processing, realize multi-agent division of labor and cooperation, and parallelly promote the task, greatly improve the user information processing efficiency, speed up the message reply speed, and at the same time, rely on the collaborative cooperation between agents to improve the accuracy and comprehensiveness of the processing result, and provide users with a more optimal and convenient service experience.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a multi-agent collaborative processing method, storage medium, and program product. Background Technology

[0002] With the rapid development of artificial intelligence technology, intelligent customer service systems have been widely applied in various fields such as telecommunications, finance, and e-commerce.

[0003] In related technologies, intelligent customer service systems adopt a single-agent approach, where the execution logic is based on preset business rules, and each functional module responds sequentially according to predetermined triggering conditions.

[0004] However, single-agent responses are inefficient and slow. Summary of the Invention

[0005] The multi-agent collaborative processing method, storage medium, and program product provided in this application are used to improve the response efficiency of intelligent customer service systems.

[0006] In a first aspect, embodiments of this application provide a multi-agent cooperative processing method, including:

[0007] Get the first information input by the user;

[0008] Based on the first information, a first intelligent agent is determined, and the first intelligent agent is used to process the first information;

[0009] Based on the first intelligent agent, traverse the preset intelligent agent dependency graph to determine one or more second intelligent agents, which are used to assist the first intelligent agent in processing the first information;

[0010] Based on the agent dependency graph, determine the task execution strategy;

[0011] Based on the task execution strategy, the first and second intelligent agents are invoked to process the first information and obtain the second information.

[0012] Output the second piece of information.

[0013] Secondly, embodiments of this application provide a multi-agent collaborative processing device, comprising:

[0014] The acquisition module is used to acquire the initial information input by the user.

[0015] The processing module is also used to determine a first intelligent agent based on the first information, and the first intelligent agent is used to process the first information;

[0016] The processing module is also used to traverse a preset agent dependency graph based on the first agent to determine one or more second agents, which are used to assist the first agent in processing the first information.

[0017] The processing module is also used to determine the task execution strategy based on the agent dependency graph;

[0018] The processing module is also used to call the first and second intelligent agents to process the first information and obtain the second information based on the task execution strategy;

[0019] The output module is used to output the second information.

[0020] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0021] The memory stores instructions that the computer executes;

[0022] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0023] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0024] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0025] The multi-agent collaborative processing method, storage medium, and program product provided in this application first obtain the first information input by the user and accurately determine the first intelligent agent for core processing. Then, based on a preset intelligent agent dependency graph, one or more second intelligent agents are selected to assist in processing. Subsequently, a scientific and efficient task execution strategy is formulated and the two types of intelligent agents are coordinated to carry out processing work. Finally, the second information is output. This can effectively break the limitations of single-agent serial processing, realize multi-agent division of labor and cooperation, and advance tasks in parallel. It can significantly improve the efficiency of user information processing, speed up message reply, and improve the accuracy and comprehensiveness of processing results by relying on the cooperation between intelligent agents, thus providing users with a better and more convenient service experience. Attached Figure Description

[0026] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0027] Figure 1 A flowchart illustrating the multi-agent cooperative processing method provided in the embodiments of this application. Figure 1 ;

[0028] Figure 2 A flowchart illustrating the multi-agent cooperative processing method provided in the embodiments of this application. Figure 2 ;

[0029] Figure 3 This is a schematic diagram of the structure of the multi-agent collaborative processing device provided in the embodiments of this application;

[0030] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0031] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0032] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0033] To address the technical issues of rigid execution logic, sequential responses from functional modules, low overall processing efficiency, and slow message response in the aforementioned single-agent solutions, the following technical concept is proposed: A multi-agent collaborative processing system is constructed. First, user input information is acquired and accurately matched to the core processing agent (the first agent). Then, based on a pre-defined agent dependency graph, associated agents (the second agents) that can assist the core agent in efficiently completing tasks are automatically identified and selected. Subsequently, a reasonable task execution strategy is formulated based on the dependencies between agents. By collaboratively calling the core agent and the auxiliary agents to advance information processing in parallel, accurate processing results are quickly generated and output, thereby overcoming the limitations of single-agent processing and improving overall response efficiency and service quality.

[0034] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0035] Figure 1 A flowchart illustrating the multi-agent cooperative processing method provided in the embodiments of this application. Figure 1The methods described above in the embodiments of this application can be applied to any electronic device. For example... Figure 1 As shown, the method includes:

[0036] S101, Obtain the first information input by the user.

[0037] Among them, the first information refers to the original information actively input by the user based on their own needs. It is the starting point and core basis for multi-agent collaborative processing. Its form may include text, voice, images, etc., and its content covers various business-related information such as user inquiries, requests, and instructions.

[0038] Specifically, through preset information receiving interfaces (such as chat input boxes, voice input modules, and form submission portals for intelligent customer service), the system receives raw information actively input by users in real time, and performs preliminary format verification (such as determining whether the information is complete and conforms to the system's receiving specifications) and format conversion (such as converting voice information to text and recognizing and extracting text content from image information) to ensure that the intelligent agent can effectively process the information in the future.

[0039] S102. Based on the first information, determine the first intelligent agent, which is used to process the first information.

[0040] Among them, the first intelligent agent refers to the core processing intelligent agent selected based on the first information input by the user. It is the intelligent agent responsible for directly responding to the user's core needs and leading the information processing process. It has exclusive business capabilities to handle such core needs (such as call charge inquiry, bill analysis, problem complaint handling, etc.).

[0041] Specifically, the first information obtained is semantically understood and the requirements are broken down (e.g., core demand keywords and business domain tags are extracted from the information through natural language processing technology). Then, based on the preset intelligent agent matching rules (e.g., the demand for "call bill inquiry" is matched with the "call bill inquiry intelligent agent", and the demand for "complaint handling" is matched with the "complaint work order intelligent agent"), the core intelligent agents that can directly handle the core needs of users are selected from the intelligent agent resource pool.

[0042] S103. Based on the first intelligent agent, traverse the preset intelligent agent dependency graph to determine one or more second intelligent agents, which are used to assist the first intelligent agent in processing the first information.

[0043] The agent dependency graph refers to a pre-defined topology graph used to record the business relationships (including auxiliary relationships, collaborative relationships, pre- and post-requirement relationships, etc.) between agents. Nodes in the graph represent agents, and edges represent dependencies between agents (e.g., the "call charge details query agent" depends on the "user authentication agent" and the "bill data retrieval agent"). The second agent refers to an agent selected from the agent dependency graph based on the business processing needs of the first agent. It assists the first agent in completing core processing tasks, does not dominate the processing flow, but provides necessary support (such as authentication, data retrieval, and result verification).

[0044] Optionally, the second intelligent agent includes functional sub-agents or business sub-agents. Functional sub-agents are responsible for executing specific, atomic functional operations and are highly reusable. These agents include, but are not limited to, authentication sub-agents, network diagnostic sub-agents, information query sub-agents, payment processing sub-agents, and SMS sending sub-agents. Each functional agent has clearly defined input / output specifications, receives standardized input parameters, and returns standardized execution results. Business sub-agents are responsible for handling complete business processes, typically involving multiple steps and user interactions. These agents also employ a star topology, containing one master business agent and multiple business sub-agents. The master business agent is responsible for the state management and flow control of the business process, while the business sub-agents are responsible for executing specific steps. Business sub-agents can reuse functional sub-agents, avoiding redundant implementation. Reuse is achieved through standardized interfaces. Each functional sub-agent defines clearly defined input parameter and output result specifications. Business sub-agents only need to construct input parameters according to the specifications and call the interface, without needing to concern themselves with the internal implementation details of the functional agents. This loosely coupled design allows functional sub-agents to be upgraded and optimized independently without affecting the business sub-agents that call them.

[0045] Specifically, starting with the first agent, according to the preset traversal rules (such as depth-first traversal and breadth-first traversal), all agent nodes in the agent dependency graph that have auxiliary dependency associations with the first agent are retrieved. Then, combined with the specific processing requirements of the first information (such as whether authentication is required or whether specific data needs to be retrieved), one or more auxiliary agents required for the current task are selected, namely the second agent.

[0046] S104. Determine the task execution strategy based on the agent dependency graph.

[0047] Among them, the task execution strategy refers to the specific task allocation scheme, execution order rules (serial, parallel or pipeline) and exception handling mechanism formulated based on the relationship between each agent (first agent, second agent) in the agent dependency graph, which clarifies the task responsibilities, execution timing and coordination method of each agent.

[0048] Specifically, based on the dependency relationship graph of the first agent and the second agent, and the dependency logic between each second agent (such as whether there are preconditions: billing data can only be retrieved after identity verification), the overall processing task is decomposed into sub-tasks that can be executed by each agent. The execution order of each sub-task is clarified (e.g., serial: identity verification → data retrieval → core query; parallel: some non-dependent auxiliary tasks are executed synchronously), the task allocation object (which agent executes which sub-task), and the exception handling rules (e.g., retry mechanism when identity verification fails, feedback mechanism when data retrieval fails). Finally, a task execution strategy that can directly guide the collaborative work of each agent is formed.

[0049] S105. Based on the task execution strategy, call the first intelligent agent and the second intelligent agent to process the first information and obtain the second information.

[0050] The second information refers to the final processing result information (such as query results, solutions, feedback content, etc.) generated after the first and second intelligent agents collaboratively process the first information according to the task execution strategy. It is a precise response to the user's original needs.

[0051] Specifically, according to the order and allocation rules determined by the task execution strategy, the pre-set intelligent agent scheduling module sequentially sends task execution instructions (including first information-related data, sub-task requirements, etc.) to each second and first intelligent agent, triggering each intelligent agent to execute the corresponding sub-task. After receiving the instructions, each intelligent agent independently completes its own sub-task (such as the identity verification intelligent agent verifying user identity, and the data retrieval intelligent agent retrieving billing data), and feeds back the sub-task execution results to the system scheduling module. The system scheduling module then synchronizes the feedback results to subsequent intelligent agents (such as synchronizing the identity verification pass result to the billing data retrieval intelligent agent). Finally, the first intelligent agent, based on the auxiliary results of each second intelligent agent, completes the core processing task (such as parsing billing data and extracting call charge deduction items), and finally integrates all processing results to generate the second information.

[0052] S106, Output the second information.

[0053] Specifically, the generated second information is formatted (e.g., converting structured billing data into easily understandable natural language, and segmenting long text results for formatting). Then, based on the channel through which the user inputs the first information, the corresponding output interface is selected (e.g., APP chat window, speech synthesis module, SMS sending interface, etc.) to provide the optimized second information back to the user. At the same time, the system records the output log (including output time, output content, user reception status, etc.) for subsequent tracking and optimization.

[0054] The multi-agent collaborative processing method provided in this application first obtains the first information input by the user and accurately determines the first intelligent agent for core processing. Then, based on a preset intelligent agent dependency graph, it traverses and filters out one or more second intelligent agents for auxiliary processing. Subsequently, it formulates a scientific and efficient task execution strategy and coordinates the two types of intelligent agents to carry out processing work, and finally outputs the second information. This method can effectively break the limitations of single-agent serial processing, realize multi-agent division of labor and cooperation, and advance tasks in parallel, greatly improve the efficiency of user information processing, speed up message reply, and improve the accuracy and comprehensiveness of processing results by relying on the cooperation between intelligent agents, thus providing users with a better and more convenient service experience.

[0055] Figure 2 A flowchart illustrating the multi-agent cooperative processing method provided in the embodiments of this application. Figure 2 .like Figure 2 As shown, the method includes:

[0056] S201. Obtain the first information input by the user.

[0057] S202. Based on the first information, determine the first intelligent agent, which is used to process the first information.

[0058] In one possible implementation, based on a preset intent recognition model, the user intent in the first information is identified to obtain the third information; key features in the first information are extracted to obtain the fourth information; user information related to the user is obtained to obtain the fifth information; a global state diagram is constructed based on the third, fourth, and fifth information, and the global state diagram is used to record the state information of the interaction with the user; a first intelligent agent is determined based on the global state diagram and a preset business rule base.

[0059] The intent recognition model is a pre-defined algorithm used to parse user input and identify core user needs, outputting the specific type of user request (e.g., query, complaint, processing). The third information, the user's core intent obtained after parsing the first information through the intent recognition model, is one of the core bases for determining the first intelligent agent. Key features refer to core data related to business processing in the first information (e.g., phone number, time range, business type, keywords), and the fourth information is the extracted set of these key features. The fifth information refers to pre-defined information related to the currently interacting user (e.g., user identity information, historical interaction records, business processing permissions, package information, etc.). The global state graph records the topology of all state information during user-system interaction, including user intent, key features, user information, historical interaction nodes, etc., supporting intelligent agent matching and subsequent process advancement. The business rule base refers to a pre-defined set of business processing specifications, recording intelligent agent matching rules corresponding to different scenarios, intents, and user information (e.g., "VIP user call charge query matched with exclusive high-end query intelligent agent").

[0060] Specifically, the first piece of information is input into a preset intent recognition model. The model then identifies the user's core intent, generating the third piece of information to clarify what the user wants to do. The first piece of information is then decomposed to extract key data necessary for business processing (such as phone number, time, and business keywords), resulting in the fourth piece of information, which provides data support for intelligent agent matching. Various information related to the current user (historical interactions, identity permissions, etc.) is obtained from the system's user database to form the fifth piece of information, taking into account the user's personalized scenarios. The third, fourth, and fifth pieces of information are integrated to construct a global state diagram, systematically and visually recording the core elements of user interaction. Combining the complete information in the global state diagram with the preset matching rules in the business rule base, the first intelligent agent that best suits the current scenario and user needs is finally determined.

[0061] It should be noted that the information obtained in the embodiments of this application is either user-authorized information or publicly available information.

[0062] For example, the first information is input into the intent recognition model. The model identifies the user's core intent as "querying phone bill deduction details" and generates the third information "User intent: phone bill deduction details query". The mobile phone number "138XXXX1234", the time "last month", and the business keyword "phone bill deduction items" are extracted from the first information to obtain the fourth information. The user information is retrieved from the system's user database and it is found that the user is a VIP user of the customer service system (who can enjoy exclusive customer service), has phone bill query records for the past 3 months, and the current package is "5G Enjoy Package", forming the fifth information. The third (query intent), fourth (mobile phone number / time / keyword), and fifth (VIP / history / package) information are integrated to construct a global state graph and record the core state of the current interaction. Based on the global state graph and the business rule base (rule: "VIP user + phone bill query intent → match VIP exclusive phone bill query intelligent agent"), the first intelligent agent is finally determined to be "VIP exclusive phone bill details query intelligent agent", rather than the query intelligent agent of ordinary users.

[0063] In one possible implementation, based on the first agent, a preset agent dependency graph is traversed to obtain the agent dependency chain of the first agent. The agent dependency chain is used to indicate the agents that the first agent depends on in the agent dependency graph, and the dependency chain formed by the agents they depend on. Based on the global state graph, a first identifier of each agent in the agent dependency chain is determined. The first identifier is used to indicate whether the output result of the agent based on the first information is stored in the global state graph, and the output result is within a preset valid time interval. Based on the first identifier, the agent dependency chain is deredundant to obtain one or more second agents.

[0064] The agent dependency chain refers to a sequence of agents that have direct or indirect dependencies on the first agent, based on the agent dependency graph. It clearly presents all auxiliary agents and their hierarchical relationships required for the first agent to perform its task (e.g., A→B→C, where A is the first agent, B depends on A, and C depends on B). The first identifier is used to mark the output state of each agent in the agent dependency chain. Specifically, it indicates whether the output result of the agent for the current first information has been stored in the global state graph, and whether the result is within a preset valid time interval (e.g., valid within 1 hour). The identifier result is usually "valid (stored and within the validity period)" or "invalid (not stored / expired)".

[0065] Specifically, the first step is to traverse the agent dependency graph according to a preset traversal rule (such as depth-first search) with the first agent as the starting node, and identify all direct / indirect dependent agents required by the first agent to perform the task, forming an agent dependency chain. The second step is to determine the first identifier of each agent in the dependency chain by combining the historical interaction data recorded in the global state graph, and determine whether its output result for the current first information is valid. The third step is to perform redundancy removal on the dependency chain based on the first identifier of each agent, remove agents with valid identifiers (no need to repeat the execution), and the remaining agents with invalid identifiers (need to be re-executed) are the second agents.

[0066] For example, starting with the first agent, traversing the agent dependency graph, we can identify its complete dependency chain: VIP exclusive call charge details query agent → user authentication agent → bill data retrieval agent → user package information verification agent (Note: user package permissions must be verified before retrieving bill data, and the verification depends on successful authentication). That is, the dependency chain contains 3 auxiliary agents. Combining the global state graph, we can query the output status of each agent: a. User authentication agent: The user did not authenticate in this interaction, and there is no output result of this agent for the current first information in the global state graph. The first identifier is invalid; b. Bill data retrieval agent: There is no historical retrieval record, the output result is not stored, and the first identifier is invalid; c. User plan information verification agent: The user just queried the plan information 10 minutes ago, and the global state graph stores the output result of this agent (the plan is a 5G unlimited plan, and the permissions are normal), and it is within the 1-hour valid time interval, so the first identifier is valid; the user plan information verification agent with the first identifier being valid is removed from the dependency chain, and the remaining 2 agents with invalid identifiers (user authentication agent and billing data retrieval agent) are finally determined to be the second agent. Only these two agents are called to assist the first agent in its work, avoiding duplicate verification of plan information.

[0067] S203. Based on the first intelligent agent, traverse the preset intelligent agent dependency graph to determine one or more second intelligent agents, which are used to assist the first intelligent agent in processing the first information.

[0068] S204. Determine the task execution strategy based on the agent dependency graph.

[0069] S205. Based on the task execution strategy, call the first intelligent agent and the second intelligent agent to process the first information and obtain the second information.

[0070] In one possible implementation, sixth information is extracted from the first information, and the sixth information is used to indicate the user's emotional state; the sixth information is input into a preset user emotion scoring model to obtain the user's emotion score; when the emotion score is greater than or equal to a first threshold, a third intelligent agent is determined, and the third intelligent agent is used to soothe the user's negative emotions; based on the third intelligent agent, the second information is refined.

[0071] The sixth piece of information refers to the core information extracted from the first piece of information that indicates the user's emotional state (such as emotional keywords, tone features, punctuation features, etc., like "angry," "why was money deducted again," "!!!," etc.). The user emotion scoring model is a pre-defined algorithm model used to quantify the user's emotional state. After inputting the sixth piece of information, it outputs an emotion score from 0 to 100 (the higher the score, the stronger the negative emotion). The first threshold is a pre-defined critical value for the emotion score (e.g., 80 points), used to determine whether the user has strong negative emotions. When the score is greater than or equal to the threshold, the emotion soothing process needs to be initiated. The third intelligent agent refers to an intelligent agent specifically designed to soothe the user's negative emotions, possessing the ability to refine responses with emotional adaptation (e.g., optimizing a harsh response into a gentle, apologetic one).

[0072] Specifically, the first step involves deconstructing the user's initial input information into emotional features, extracting a sixth piece of information (keywords, tone, punctuation, etc.) that reflects the user's emotional state. The second step involves inputting this sixth piece of information into a pre-defined user emotion scoring model, which calculates the user's emotion score to quantify the intensity of their emotion. The third step compares the emotion score with a pre-defined first threshold. If the score is greater than or equal to the first threshold (indicating strong negative emotion), the emotion-soothing process is initiated, and a third-party intelligent agent is invoked. If the score is less than the first threshold (indicating stable emotion), the soothing process is not initiated. The fourth step, if the third-party intelligent agent is invoked, refines the generated second piece of information based on the user's negative emotion type (e.g., anger, dissatisfaction, anxiety), optimizing the tone and wording of the response to better suit the soothing needs.

[0073] S206. Based on a preset natural language model, convert the second information into text information or voice information.

[0074] Specifically, the system invokes a pre-defined natural language model and, based on the channel through which the user inputs the first information (such as text input or voice input) and the system's pre-defined output priority, converts the second information (which may be structured data) generated by multi-agent collaborative processing into text or voice information that is easily accepted by the user (e.g., user voice input is preferentially converted into voice information, and user text input is preferentially converted into text information).

[0075] S207. Display text information to the user or play voice information to the user.

[0076] Specifically, depending on the type of information converted, the corresponding output method is selected: if it is text information, the text information is displayed to the user through user interaction channels (such as chat windows or SMS interfaces); if it is voice information, the voice information is played to the user through audio output devices (such as mobile phone speakers or headphones).

[0077] In one possible implementation, a seventh piece of information is obtained, which is information fed back by the user based on the second piece of information; the global state graph is updated based on the second and seventh pieces of information.

[0078] The seventh message refers to the feedback from the user after receiving the second message output by the system. The form may include text reply, voice reply, operation feedback (such as clicking the "Satisfied" or "Dissatisfied" button), etc. The content may cover the recognition of the result, questioning, further requests, etc. (such as "Understood" or "Still don't understand why the data fee is 20 yuan").

[0079] Specifically, the first step is to "acquire" the seventh information based on the second information in real time through user interaction channels (such as chat windows, voice receivers, and feedback buttons) to ensure that the user's attitude towards the processing result and subsequent demands are fully captured. The second step is to associate and integrate the acquired seventh information with the previously output second information (such as clarifying which second information the seventh information is in response to), and then "update" the global state graph based on this association, adding user feedback nodes and updating the user interaction state (such as updating from "output processing result" to "user approves the result" or "user has further demands").

[0080] By constructing a global state graph through intent recognition, key feature extraction, and user information acquisition, and combining it with a business rule base to determine the first intelligent agent, the limitations of matching the core intelligent agent in a single dimension are overcome. This effectively improves the accuracy of the first intelligent agent's adaptation to user needs and avoids problems such as low processing efficiency or distorted results caused by intelligent agent matching deviations. It lays a precise core subject foundation for subsequent multi-agent collaborative processing.

[0081] By analyzing the dependency chain of the first agent and combining it with the global state diagram to determine the validity of each agent's output and perform redundancy removal, the system can accurately select the auxiliary agents that are truly needed for the current task, avoid repeatedly calling agents that already have valid output results, significantly reduce system resource waste, improve the overall efficiency of multi-agent collaborative processing, and ensure that the calling of auxiliary agents is more targeted.

[0082] Before outputting the second piece of information, the system extracts the user's emotional state and scores the emotion. For users with strong negative emotions, a third-party intelligent agent is invoked to refine the response. This achieves synergy between business processing and emotional soothing, effectively avoiding the problem of harsh responses exacerbating negative emotions. It makes the service response more humanized, significantly improves the user interaction experience and satisfaction, and balances business processing efficiency with users' emotional needs.

[0083] The second information is converted into text or voice information through a natural language model and output according to the user interaction channel, breaking the limitation of a single output format. It can meet the information receiving habits of different users (such as those who prefer text reading or those who prefer voice listening), ensuring that users can conveniently and quickly obtain the processing results, and improving the flexibility of service output and the ease of use for users.

[0084] The global state graph is updated based on user feedback on the second information, enabling dynamic iteration of user interaction status. This allows the global state graph to record the entire user interaction process and the latest requests in real time and completely, providing accurate status support for possible secondary interactions (such as further user inquiries or questions). This avoids repeatedly asking users for information, greatly improves the processing efficiency of secondary interactions, and ensures the continuity and consistency of services.

[0085] Figure 3 This is a schematic diagram of the structure of the multi-agent cooperative processing device provided in the embodiments of this application, as shown below. Figure 3 As shown, the multi-agent collaborative processing device 30 provided in this embodiment includes an acquisition module 301, a processing module 302, and an output module 303.

[0086] The acquisition module 301 is used to acquire the first information input by the user;

[0087] The processing module 302 is also used to determine a first intelligent agent based on the first information, wherein the first intelligent agent is used to process the first information;

[0088] The processing module 302 is also used to traverse a preset agent dependency graph based on the first agent to determine one or more second agents, the second agents being used to assist the first agent in processing the first information;

[0089] Processing module 302 is also used to determine task execution strategy based on agent dependency graph;

[0090] The processing module 302 is also used to call the first intelligent agent and the second intelligent agent to process the first information and obtain the second information based on the task execution strategy;

[0091] Output module 303 is used to output the second information.

[0092] In one possible implementation, the processing module 302 is specifically used for:

[0093] Based on a pre-defined intent recognition model, the user intent in the first information is identified to obtain the third information;

[0094] Extract the key features from the first information to obtain the fourth information;

[0095] Obtain user information related to the user to obtain the fifth piece of information;

[0096] Based on the third, fourth, and fifth information, a global state diagram is constructed. The global state diagram is used to record the state information of user interaction.

[0097] Based on the global state diagram and the preset business rule base, the first intelligent agent is determined.

[0098] In one possible implementation, the processing module 302 is specifically used for:

[0099] Based on the first agent, the preset agent dependency graph is traversed to obtain the agent dependency chain of the first agent. The agent dependency chain is used to indicate the agent that the first agent in the agent dependency graph depends on, and the dependency chain formed by the agent that it depends on.

[0100] Based on the global state diagram, the first identifier of each agent in the agent dependency chain is determined. The first identifier is used to indicate whether the output result of the agent based on the first information is stored in the global state diagram, and the output result is within a preset valid time interval.

[0101] Based on the first identifier, the dependency chain of the agent is deredundant to obtain one or more second agents.

[0102] In one possible implementation, the processing module 302 is further configured to:

[0103] Extract the sixth information from the first information; the sixth information is used to indicate the user's emotional state.

[0104] The sixth piece of information is input into the preset user emotion rating model to obtain the user's emotion rating;

[0105] When the emotion score is greater than or equal to the first threshold, a third agent is identified, which is used to soothe the user's negative emotions.

[0106] The second piece of information is refined based on the third intelligent agent.

[0107] In one possible implementation, the output module 303 is specifically used for:

[0108] Based on a pre-defined natural language model, the second information is converted into text or speech information.

[0109] Display text information to the user, or play voice information to the user.

[0110] In one possible implementation, the processing module 302 is further configured to:

[0111] Obtain the seventh piece of information, which is the information provided by the user based on the second piece of information;

[0112] The global state graph is updated based on the second and seventh pieces of information.

[0113] The multi-agent collaborative processing device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0114] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4 As shown, the electronic device 40 provided in this embodiment includes at least one processor 401 and a memory 402. Optionally, the device 40 further includes a communication component 403. The processor 401, memory 402, and communication component 403 are connected via a bus.

[0115] In a specific implementation, at least one processor 401 executes computer execution instructions stored in memory 402, causing at least one processor 401 to perform the above-described method.

[0116] The specific implementation process of processor 401 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0117] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0118] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0119] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0120] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0121] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0122] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0123] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in a device. The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed can be indirect coupling or communication connection through some interface, device, or unit, and can be electrical, mechanical, or other forms.

[0124] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0125] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0126] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0127] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware associated with program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks. Finally, it should be noted that those skilled in the art, upon considering the specification and practicing the invention disclosed herein, will readily conceive of other embodiments of the invention. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A multi-agent cooperative processing method, characterized in that, include: Get the first information input by the user; Based on the first information, a first intelligent agent is determined, and the first intelligent agent is used to process the first information; Based on the first agent, a preset agent dependency graph is traversed to determine one or more second agents, which are used to assist the first agent in processing the first information. Based on the agent dependency graph, the task execution strategy is determined; Based on the task execution strategy, the first intelligent agent and the second intelligent agent are invoked to process the first information to obtain the second information; Output the second piece of information.

2. The method according to claim 1, characterized in that, The step of determining the first intelligent agent based on the first information includes: Based on a preset intent recognition model, the user intent in the first information is recognized to obtain the third information; Extract the key features from the first information to obtain the fourth information; Obtain user information related to the user to obtain the fifth piece of information; Based on the third information, the fourth information, and the fifth information, a global state diagram is constructed, which is used to record the state information of the interaction with the user. Based on the global state diagram and the preset business rule base, the first intelligent agent is determined.

3. The method according to claim 2, characterized in that, The step of determining one or more second agents by traversing a preset agent dependency graph based on the first agent includes: Based on the first agent, a preset agent dependency graph is traversed to obtain the agent dependency chain of the first agent. The agent dependency chain is used to instruct the first agent in the agent dependency graph to execute the agent it depends on, and the dependency chain formed by the agent it depends on. Based on the global state diagram, a first identifier is determined for each agent in the agent dependency chain. The first identifier is used to indicate whether the global state diagram stores the output result of the agent based on the first information, and the output result is within a preset valid time interval. Based on the first identifier, the dependency chain of the agent is deredundant to obtain one or more second agents.

4. The method according to claim 1, characterized in that, Before outputting the second information, the method further includes: The sixth information is extracted from the first information, and the sixth information is used to indicate the user's emotional state; The sixth piece of information is input into a preset user emotion rating model to obtain the user's emotion rating; When the emotion score is greater than or equal to the first threshold, a third agent is identified, which is used to soothe the user's negative emotions. Based on the third intelligent agent, the second information is refined.

5. The method according to claim 1, characterized in that, The output of the second information includes: Based on a preset natural language model, the second information is converted into text information or voice information; The text information is displayed to the user, or the voice information is played to the user.

6. The method according to claim 2, characterized in that, Also includes: Obtain the seventh piece of information, which is the information fed back by the user based on the second piece of information; The global state graph is updated based on the second information and the seventh information.

7. A multi-agent collaborative processing device, characterized in that, include: The acquisition module is used to acquire the initial information input by the user. The processing module is further configured to determine a first intelligent agent based on the first information, wherein the first intelligent agent is configured to process the first information; The processing module is further configured to, based on the first intelligent agent, traverse a preset intelligent agent dependency graph to determine one or more second intelligent agents, wherein the second intelligent agents are used to assist the first intelligent agent in processing the first information; The processing module is also used to determine the task execution strategy based on the agent dependency graph; The processing module is further configured to, based on the task execution strategy, invoke the first intelligent agent and the second intelligent agent to process the first information and obtain the second information; The output module is used to output the second information.

8. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-6.

Images